Advanced Calculus 13

Lecture Notes for Advanced Calculus
James S. Cook
Liberty University
Department of Mathematics
Fall 2013
2
introduction and motivations for these notes

There are many excellent texts on portions of this subject. However, the particular path I choose
this semester is not quite in line with any particular text. I required the text on Advanced Cal-
culus by Edwards because it contains all the major theorems that traditionally are covered in an
Advanced Calculus course.
Linear algebra is not a prerequisite for this course. However, I will use linear algebra. Matrices,
linear transformations and vector spaces are necessary ingredients for a proper discussion of ad-
vanced calculus. I believe an interested student can easily assimilate the needed tools as we go so I
am not terribly worried if you have not had linear algebra previously. I will make a point to include
some baby1 linear exercises to make sure everyone who is working at this course keeps up with the
story that unfolds.
Real analysis is also not a prerequisite for this course. However, I will present some proofs which
properly fall in the category of real analysis. Some of these proofs involve a sophistication that
is beyond the usual day-to-day business of the course. I include these thoughts for the sake of
completeness. If I am to test them it’s probably more on the question of were you paying attention
as opposed to can you reconstruct the monster from scratch.
Certainly my main intent in this course is that you learn calculus more deeply. Yes we’ll learn
some new calculations, but I also hope that what we cover also gives you deeper insight into your
previous experience with calculus. Towards that end, I am including a section or two on series and
sequences and a discussion of pointwise verses uniform convergence. These discussions set-up the
technique of exchanging derivatives and integrals which is a powerful technique seldom discussed
in current calculus courses.
Doing the homework is doing the course. I cannot overemphasize the importance of thinking
through the homework. I would be happy if you left this course with a working knowledge of:
X set-theoretic mapping langauge, fibers and images and how to picture relationships diagra-
matically.
X continuity in view of the metric topology in n-space.
X the concept and application of the derivative and differential of a mapping.
X continuous differentiability
X inverse function theorem
X implicit function theorem
X tangent space and normal space via gradients or derivatives of parametrizations

1
if you view this as an insult then you haven’t met the right babies yet. Baby exercises are cute.
3
X extrema for multivariate functions, critical points and the Lagrange multiplier method
X multivariate Taylor series.
X quadratic forms
X critical point analysis for multivariate functions
X dual space and the dual basis.
X multilinear algebra.
X metric dualities and Hodge duality.
X the work and flux form mappings for R3 .
X basic manifold theory (don0 t let me get too deep, please...)2
X vector fields as derivations.
X differential forms and the exterior derivative
X integration of forms
X generalized Stokes’s Theorem.
X push-fowards and pull-backs
X how differential forms and submanifolds naturally geometrize differential equations
X elementary differential geometry of curves and surfaces via the method of moving frames
X basic variational calculus (how to calculate the Euler-Lagrange equations for a given La-
grangian)
When I say working knowledge what I intend is that you have some sense of the problem and at
least know where to start looking for a solution. Some of the topics above take a much longer time
to understand deeply. I cover them to spark your interest and seed your intuition if all goes well.
Before we begin, I should warn you that I assume quite a few things from the reader. These notes
are intended for someone who has already grappled with the problem of constructing proofs. I
assume you know the difference between ⇒ and ⇔. I assume the phrase ”iff” is known to you.
I assume you are ready and willing to do a proof by induction, strong or weak. I assume you
know what R, C, Q, N and Z denote. I assume you know what a subset of a set is. I assume you
know how to prove two sets are equal. I assume you are familar with basic set operations such
as union and intersection (although we don’t use those much). More importantly, I assume you
have started to appreciate that mathematics is more than just calculations. Calculations without
context, without theory, are doomed to failure. At a minimum theory and proper mathematics
4
allows you to communicate analytical concepts to other like-educated individuals.
Some of the most seemingly basic objects in mathematics are insidiously complex. We’ve been
taught they’re simple since our childhood, but as adults, mathematically-minded adults, we find
the actual definitions of such objects as R or C are rather involved. I will not attempt to provide
foundational arguments to build numbers from basic set theory. I believe it is possible, I think
it’s well-thought-out mathematics, but we take the existence of the real numbers as an axiom for
these notes. We assume that R exists and that the real numbers possess all their usual properties.
In fact, I assume R, C, Q, N and Z all exist complete with their standard properties. In short, I
assume we have numbers to work with. We leave the rigorization of numbers to a different course.
I have avoided use of Einstein’s implicit summation notation in the majority of these notes. This has
introduced some clutter in calculations, but I hope the student finds the added detail helpful. Nat-
urally if one goes on to study tensor calculations in physics then no such luxury is granted. In view
of this, I left the more physicsy notation in the discussion of electromagnetism via differential forms.
This is the third time I have prepared an official offering of Advanced Calculus. The first offering
was given to about 10 students, half engineering, half math, it was deliberately given with a com-
putational focus. The second offering was intended for an audience of about 6 math students, all
bailed except 1 and the course modified into a more serious, theoretically-focused introduction to
manifolds (Spencer 2011). I have taught it off and on as an indpendent study to several students,
Bobbi Beller, Jin Li.
This semester I hope to go further into the exposition of differential forms than I have previously.
In past attempts, too much time was devoted to developing constructions in basic manifold theory
we didn’t really need. So, this time, I take a somewhat formal approach to manifolds. We’ll see
how differential forms allow great insight into the shape of surfaces and the geometrization of dif-
ferential equations. Finally, at the end of the course I again spend several lectures on the calculus
of variations.
note on editing: ran a little short on time this summer, sorry but only pages 1-224 ok
for printing at moment. The remaining 225 and beyond are only about 80% finished.
I will let you know once those are fixed. Thanks!
James Cook, August 18, 2013.

Contents
1 sets, functions and euclidean space 11

1.1 set theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
1.2 functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
1.3 vectors and geometry for n-dimensional space . . . . . . . . . . . . . . . . . . . . . . 18
1.3.1 vector algebra for three dimensions . . . . . . . . . . . . . . . . . . . . . . . . 23
1.3.2 compact notations for vector arithmetic . . . . . . . . . . . . . . . . . . . . . 24
2 linear algebra 27
2.1 vector spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
2.2 matrix calculation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
2.3 linear transformations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
2.3.1 a gallery of linear transformations . . . . . . . . . . . . . . . . . . . . . . . . 39
2.3.2 standard matrices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
2.3.3 coordinates and isomorphism . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
3 topology and limits 51

3.1 elementary topology and limits . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
3.2 normed vector spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
3.2.1 open balls, limits and sequences in normed vector spaces . . . . . . . . . . . . 63
3.3 intuitive theorems of calculus . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
3.3.1 theorems for single-variable calculus . . . . . . . . . . . . . . . . . . . . . . . 69
3.3.2 theorems for multivariate calculus . . . . . . . . . . . . . . . . . . . . . . . . 74
4 differentiation 75
4.1 the Frechet differential . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76
4.2 partial derivatives and the Jacobian matrix . . . . . . . . . . . . . . . . . . . . . . . 82
4.2.1 directional derivatives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
4.3 derivatives of sum and scalar product . . . . . . . . . . . . . . . . . . . . . . . . . . 88
4.4 a gallery of explicit derivatives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89
4.5 chain rule . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
4.6 common product rules . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96
4.6.1 scalar-vector product rule . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96
5
6 CONTENTS
4.6.2 calculus of paths in R3 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98

4.6.3 calculus of matrix-valued functions of a real variable . . . . . . . . . . . . . . 100
4.6.4 calculus of complex-valued functions of a real variable . . . . . . . . . . . . . 101
4.7 continuous differentiability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102
4.8 on why partial derivatives commute . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
4.9 complex analysis in a nutshell . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110
4.9.1 harmonic functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115
5 inverse and implicit function theorems 117

5.1 inverse function theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117
5.2 implicit function theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 122
5.3 implicit differentiation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 132
5.3.1 computational techniques for partial differentiation with side conditions . . . 135
5.4 the constant rank theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 138
6 two views of manifolds in Rn 143

6.1 definition of level set . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 144
6.2 tangents and normals to a level set . . . . . . . . . . . . . . . . . . . . . . . . . . . . 145
6.3 tangent and normal space from patches . . . . . . . . . . . . . . . . . . . . . . . . . 151
6.4 summary of tangent and normal spaces . . . . . . . . . . . . . . . . . . . . . . . . . 152
6.5 method of Lagrange mulitpliers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 153
7 critical point analysis for several variables 161

7.1 multivariate power series . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 161
7.1.1 taylor’s polynomial for one-variable . . . . . . . . . . . . . . . . . . . . . . . . 161
7.1.2 taylor’s multinomial for two-variables . . . . . . . . . . . . . . . . . . . . . . 163
7.1.3 taylor’s multinomial for many-variables . . . . . . . . . . . . . . . . . . . . . 166
7.2 a brief introduction to the theory of quadratic forms . . . . . . . . . . . . . . . . . . 169
7.2.1 diagonalizing forms via eigenvectors . . . . . . . . . . . . . . . . . . . . . . . 172
7.3 second derivative test in many-variables . . . . . . . . . . . . . . . . . . . . . . . . . 179
8 convergence and estimation 185

8.1 sequences of functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 185
8.2 power series . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 185
8.3 matrix exponential . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 185
8.4 uniform convergence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 185
8.5 differentiating under the integral . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 185
8.6 contraction mappings and Newton’s method . . . . . . . . . . . . . . . . . . . . . . . 185
8.7 multivariate mean value theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 185
8.8 proof of the inverse function theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . 185
CONTENTS 7
9 multilinear algebra 187

9.1 dual space . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 187
9.2 multilinearity and the tensor product . . . . . . . . . . . . . . . . . . . . . . . . . . . 189
9.2.1 bilinear maps . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 190
9.2.2 trilinear maps . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 194
9.2.3 multilinear maps . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 197
9.3 wedge product . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 200
9.3.1 wedge product of dual basis generates basis for ΛV . . . . . . . . . . . . . . . 200
9.3.2 the exterior algebra . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 203
9.3.3 connecting vectors and forms in R3 . . . . . . . . . . . . . . . . . . . . . . . . 208
9.4 bilinear forms and geometry, metric duality . . . . . . . . . . . . . . . . . . . . . . . 210
9.4.1 metric geometry . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 210
9.4.2 metric duality for tensors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 212
9.4.3 inner products and induced norm . . . . . . . . . . . . . . . . . . . . . . . . . 215
9.5 hodge duality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 216
9.5.1 hodge duality in euclidean space R3 . . . . . . . . . . . . . . . . . . . . . . . 216
9.5.2 hodge duality in minkowski space R4 . . . . . . . . . . . . . . . . . . . . . . . 218
9.6 coordinate change . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 222
9.6.1 coordinate change for T20 (V ) . . . . . . . . . . . . . . . . . . . . . . . . . . . 224
10 calculus with differential forms 225

10.1 an informal introduction to manifolds . . . . . . . . . . . . . . . . . . . . . . . . . . 227
10.2 vectors as derivations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 231
10.2.1 concerning the geometry of derivations . . . . . . . . . . . . . . . . . . . . . . 237
10.3 differential for manifolds, the push-forward . . . . . . . . . . . . . . . . . . . . . . . 237
10.3.1 intuition for the push-forward . . . . . . . . . . . . . . . . . . . . . . . . . . . 239
10.3.2 a pragmatic formula for the push-forward . . . . . . . . . . . . . . . . . . . . 239
10.4 cotangent space . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 242
10.5 differential forms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 243
10.6 the exterior derivative . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 246
10.6.1 exterior derivatives on R3 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 248
10.6.2 coordinate independence of exterior derivative . . . . . . . . . . . . . . . . . . 249
10.7 the pull-back . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 250
10.7.1 intuitive computation of pull-backs . . . . . . . . . . . . . . . . . . . . . . . . 252
10.7.2 implicit function theorem in view of forms . . . . . . . . . . . . . . . . . . . . 255
10.8 integration of forms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 258
10.8.1 integration of k-form on Rk . . . . . . . . . . . . . . . . . . . . . . . . . . . . 258
10.8.2 orientations and submanifolds . . . . . . . . . . . . . . . . . . . . . . . . . . . 259
10.9 Generalized Stokes Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 267
10.10Poincare lemma . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 270
10.10.1 exact forms are closed . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 270
10.10.2 potentials for closed forms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 271
8 CONTENTS
10.11introduction to geometric differential equations . . . . . . . . . . . . . . . . . . . . . 278

10.11.1 exact differential equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . 278
10.11.2 differential equations via forms . . . . . . . . . . . . . . . . . . . . . . . . . . 279
11 Geometry by frames and forms 281

11.1 basic terminology and vector fields . . . . . . . . . . . . . . . . . . . . . . . . . . . . 282
11.2 connection forms on R3 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 285
11.2.1 Frenet Serret equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 289
11.3 structure equations for frame field on R3 . . . . . . . . . . . . . . . . . . . . . . . . . 290
11.4 surfaces in R3 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 292
11.5 isometry . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 296
12 Electromagnetism in differential form 297

12.1 differential forms in Minkowski space . . . . . . . . . . . . . . . . . . . . . . . . . . . 297
12.2 exterior derivatives of charge forms, field tensors, and their duals . . . . . . . . . . . 302
12.3 coderivatives and comparing to Griffith’s relativitic E & M . . . . . . . . . . . . . . 304
12.4 Maxwell’s equations are relativistically covariant . . . . . . . . . . . . . . . . . . . . 305
12.5 Electrostatics in Five dimensions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 307
13 introduction to variational calculus 313

13.1 history . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 313
13.2 the variational problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 315
13.3 variational derivative . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 316
13.4 Euler-Lagrange examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 317
13.4.1 shortest distance between two points in plane . . . . . . . . . . . . . . . . . . 318
13.4.2 surface of revolution with minimal area . . . . . . . . . . . . . . . . . . . . . 319
13.4.3 Braichistochrone . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 320
13.5 Euler-Lagrange equations for several dependent variables . . . . . . . . . . . . . . . . 322
13.5.1 free particle Lagrangian . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 323
13.5.2 geodesics in R3 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 323
13.6 the Euclidean metric . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 324
13.7 geodesics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 326
13.7.1 geodesic on cylinder . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 326
13.7.2 geodesic on sphere . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 327
13.8 Lagrangian mechanics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 327
13.8.1 basic equations of classical mechanics summarized . . . . . . . . . . . . . . . 327
13.8.2 kinetic and potential energy, formulating the Lagrangian . . . . . . . . . . . . 328
13.8.3 easy physics examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 330
14 leftover manifold theory 331

14.1 manifold defined . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 332
14.1.1 embedded manifolds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 335
14.1.2 diffeomorphism . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 337
CONTENTS 9
14.2 tangent space . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 338

14.2.1 equivalence classes of curves . . . . . . . . . . . . . . . . . . . . . . . . . . . . 339
14.2.2 contravariant vectors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 340
14.2.3 dictionary between formalisms . . . . . . . . . . . . . . . . . . . . . . . . . . 341
14.3 Gauss map to a sphere . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 342
14.4 tensors at a point . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 344
14.5 smoothness of differential forms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 345
14.6 tensor fields . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 346
14.7 metric tensor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 348
14.7.1 classical metric notation in Rm . . . . . . . . . . . . . . . . . . . . . . . . . . 348
14.7.2 metric tensor on a smooth manifold . . . . . . . . . . . . . . . . . . . . . . . 349
14.8 on boundaries and submanifolds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 352
10 CONTENTS
Chapter 1
sets, functions and euclidean space
In this chapter we settle some basic terminology about sets and functions.
1.1 set theory

Let us denote sets by capital letters in as much as is possible. Often the lower-case letter of the
same symbol will denote an element; a ∈ A is to mean that the object a is in the set A. We can
abbreviate a1 ∈ A and a2 ∈ A by simply writing a1 , a2 ∈ A, this is a standard notation. The union
of two sets A and B is denoted A ∪ B = {x|x ∈ A or x ∈ B}. The intersection of two sets is
denoted A ∩ B = {x|x ∈ A and x ∈ B}. If a set S has no elements then we say S is the empty set
and denote this by writing S = ∅. It sometimes convenient to use unions or intersections of several
sets: [
Uα = {x | there exists α ∈ Λ with x ∈ Uα }
α∈Λ
\
Uα = {x | for all α ∈ Λ we have x ∈ Uα }
α∈Λ
we say Λ is the index set in the definitions above. If Λ is a finite set then the union/intersection
is said to be a finite union/interection. If Λ is a countable set then the union/intersection is said
to be a countable union/interection1 . Suppose A and B are both sets then we say A is a subset
of B and write A ⊆ B iff a ∈ A implies a ∈ B for all a ∈ A. If A ⊆ B then we also say B is a
superset of A. If A ⊆ B then we say A ⊂ B iff A 6= B and A 6= ∅. Recall, for sets A, B we define
A = B iff a ∈ A implies a ∈ B for all a ∈ A and conversely b ∈ B implies b ∈ A for all b ∈ B. This
is equivalent to insisting A = B iff A ⊆ B and B ⊆ A. The difference of two sets A and B is
denoted A − B and is defined by A − B = {a ∈ A | such that a ∈ / B}2 .
1
recall the term countable simply means there exists a bijection to the natural numbers. The cardinality of such
a set is said to be ℵo
2
other texts somtimes use A − B = A \ B
11
12 CHAPTER 1. SETS, FUNCTIONS AND EUCLIDEAN SPACE
Real numbers can be constructed from set theory and about a semester of mathematics. We will
accept the following as axioms3
Definition 1.1.1. real numbers

The set of real numbers is denoted R and is defined by the following axioms:
(A1) addition commutes; a + b = b + a for all a, b ∈ R.
(A2) addition is associative; (a + b) + c = a + (b + c) for all a, b, c ∈ R.
(A3) zero is additive identity; a + 0 = 0 + a = a for all a ∈ R.
(A4) additive inverses; for each a ∈ R there exists −a ∈ R and a + (−a) = 0.
(A5) multiplication commutes; ab = ba for all a, b ∈ R.
(A6) multiplication is associative; (ab)c = a(bc) for all a, b, c ∈ R.
(A7) one is multiplicative identity; a1 = a for all a ∈ R.
(A8) multiplicative inverses for nonzero elements;

for each a 6= 0 ∈ R there exists a1 ∈ R and a a1 = 1.
(A9) distributive properties; a(b + c) = ab + ac and (a + b)c = ac + bc for all a, b, c ∈ R.
(A10) totally ordered field; for a, b ∈ R:
(i) antisymmetry; if a ≤ b and b ≤ a then a = b.

(ii) transitivity; if a ≤ b and b ≤ c then a ≤ c.
(iii) totality; a ≤ b or b ≤ a
(A11) least upper bound property: every nonempty subset of R that has an upper bound,
has a least upper bound. This makes the real numbers complete.
Modulo A11 and some math jargon this should all be old news. An upper bound for a set S ⊆ R
is a number M ∈ R such that M > s for all s ∈ S. Similarly a lower bound on S is a number
m ∈ R such that m < s for all s ∈ S. If a set S is bounded above and below then the set is said
to be bounded. For example, the open set (a, b) is bounded above by b and it is bounded below
by a. In contrast, rays such as (0, ∞) are not bounded above. Closed intervals contain their least
upper bound and greatest lower bound. The bounds for an open interval are outside the set.
3
an axiom is a basic belief which cannot be further reduced in the conversation at hand. If you’d like to see a
construction of the real numbers from other math, see Ramanujan and Thomas’ Intermediate Analysis which has
the construction both from the so-called Dedekind cut technique and the Cauchy-class construction. Also, I’ve been
informed, Terry Tao’s Analysis I text has a very readable exposition of the construction from the Cauchy viewpoint.
1.1. SET THEORY 13
We often make use of the following standard sets:
natural numbers (positive integers); N = {1, 2, 3, . . . }.
natural numbers up to the number n; Nn = {1, 2, 3, . . . , n − 1, n}.
integers; Z = {. . . , −2, −1, 0, 1, 2, . . . }. Note, Z>0 = N.
non-negative integers; Z≥0 = {0, 1, 2, . . . } = N ∪ {0}.
negative integers; Z<0 = {−1, −2, −3, . . . } = −N.
rational numbers; Q = { pq | p, q ∈ Z, q 6= 0}.
irrational numbers; J = {x ∈ R | x ∈
/ Q}.
open interval from a to b; (a, b) = {x|a < x < b}.
half-open interval; (a, b] = {x | a < x ≤ b} or [a, b) = {x | a ≤ x < b}.
closed interval; [a, b] = {x | a ≤ x ≤ b}.
The final, and for us the most important, construction in set-theory is called the Cartesian product.
Let A, B, C be sets, we define:
A × B = {(a, b) | a ∈ A and b ∈ B}
By a slight abuse of notation4 we also define:
A × B × C = {(a, b, c) | a ∈ A and b ∈ B and c ∈ C}
In the case the sets comprising the cartesian product are the same we use an exponential notation
for the construction:
A2 = A × A, A3 = A × A × A
We can extend to finitely many sets. Suppose Ai is a set for i = 1, 2, . . . n then we denote the
Cartesian product by
A1 × A2 × · · · An = ×ni=1 Ai
and define ~x ∈ ×ni=1 Ai iff ~x = (a1 , a2 , . . . , an ) where ai ∈ Ai for each i = 1, 2, . . . n. An element ~x
as above is often called an n-tuple.
We define R2 = {(x, y) | x, y ∈ R}. I refer to R2 as ”R-two” in conversational mathematics. Like-

wise, ”R-three” is defined by R3 = {(x, y, z) | x, y, z ∈ R}. We are ultimately interested in studying
”R-n” where Rn = {(x1 , x2 , . . . , xn ) | xi ∈ R for i = 1, 2, . . . , n}. In this course if we consider Rm
4
technically A × (B × C) 6= (A × B) × C since objects of the form (a, (b, c)) are not the same as ((a, b), c), we
ignore these distinctions and map both of these to the triple (a, b, c) without ambiguity in what follows
it is assumed from the context that m ∈ N.
In terms of cartesian products you can imagine the x-axis as the number line then if we paste
another numberline at each x value the union of all such lines constucts the plane; this is the
picture behind R2 = R × R. Another interesting cartesian product is the unit-square; [0, 1]2 =
[0, 1] × [0, 1] = {(x, y) | 0 ≤ x ≤ 1, 0 ≤ y ≤ 1}. Sometimes a rectangle in the plane with it’s edges
included can be written as [x1 , x2 ] × [y1 , y2 ]. If we want to remove the edges use (x1 , x2 ) × (y1 , y2 ).
Moving to three dimensions we can construct the unit-cube as [0, 1]3 . A generic rectangu-
lar solid can sometimes be represented as [x1 , x2 ] × [y1 , y2 ] × [z1 , z2 ] or if we delete the edges:
(x1 , x2 ) × (y1 , y2 ) × (z1 , z2 ).
1.2 functions
Suppose A and B are sets, we say f : A → B is a function if for each a ∈ A the function f
assigns a single element f (a) ∈ B. Moreover, if f : A → B is a function we say it is a B-valued
function of an A-variable and we say A = dom(f ) whereas B = codomain(f ). For example,
if f : R2 → [0, 1] then f is real-valued function of R2 . On the other hand, if f : C → R2 then
we’d say f is a vector-valued function of a complex variable. The term mapping will be used
interchangeably with function in these notes5 . Suppose f : U → V and U ⊆ S and V ⊆ T then we
may consisely express the same data via the notation f : U ⊆ S → V ⊆ T .
Sometimes we can take two given functions and construct a new function.
1. if f : U → V and g : V → W then g ◦ f : U → W is the composite of g with f .
2. if f, g : U → V and V is a set with an operation of addition then we define f ± g : U → V

pointwise by the natural assignment (f ± g)(x) = f (x) ± g(x) for each x ∈ U . We say that
f ± g is the sum(+) or difference(−) of f and g.
3. if f : U → V and c ∈ S where there is an operation of scalar multiplication by S on V then

cf : U → V is defined pointwise by (cf )(x) = cf (x) for each x ∈ U . We say that cf is scalar
multiple of f by c.
Usually we have in mind S = R or S = C and often the addition is just that of vectors, however
the definitions (2.) and (3.) apply equally well to matrix-valued functions or operators which is
another term for function-valued functions. For example, in the first semester of calculus we study
d/dx which is a function of functions; d/dx takes an input of f and gives the output df /dx. If we
5
in my first set of advanced calculus notes (2010) I used the term function to mean the codomain was real numbers
whereas mapping implied a codomain of vectors. I was following Edwards as he makes this convention in his text. I
am not adopting that terminology any longer, I think it’s better to use the term function as we did in Math 200 or
250. A function is an abstract construction which allows for a vast array of codomains.
1.2. FUNCTIONS 15
write L = 3d/dx we have a new operator defined by (3d/dx)[f ] = 3df /dx for each function f in
the domain of d/dx.
Definition 1.2.1.
Suppose f : U → V . We define the image of U1 under f as follows:
f (U1 ) = { y ∈ V | there exists x ∈ U1 with f (x) = y}.
The range of f is f (U ). The inverse image of V1 under f is defined as follows:
f −1 (V1 ) = { x ∈ U | f (x) ∈ V1 }.
The inverse image of a single point in the codomain is called a fiber. Suppose f : U → V .
We say f is surjective or onto V1 iff there exists U1 ⊆ U such that f (U1 ) = V1 . If a function
is onto its codomain then the function is surjective. If f (x1 ) = f (x2 ) implies x1 = x2
for all x1 , x2 ∈ U1 ⊆ U then we say f is injective on U1 or 1 − 1 on U1 . If a function
is injective on its domain then we say the function is injective. If a function is both
injective and surjective then the function is called a bijection or a 1-1 correspondance.
Example 1.2.2. Suppose f : R2 → R and f (x, y) = x for each (x, y) ∈ R2 . The function is not
injective since f (1, 2) = 1 and f (1, 3) = 1 and yet (1, 2) 6= (1, 3). Notice that the fibers of f are
simply vertical lines:
f −1 (xo ) = {(x, y) ∈ dom(f ) | f (x, y) = xo } = {(xo , y) | y ∈ R} = {xo } × R
√
Example 1.2.3. Suppose f : R → R and f (x) = x2 + 1 for each x ∈ R. This function is not
surjective because 0 ∈ / f (R). In contrast, if we construct g : R → [1, ∞) with g(x) = f (x) for each
x ∈ R then can argue that g is surjective. Neither f nor g is injective, the fiber of xo is {−xo , xo }
for each xo 6= 0. At all points except zero these maps are said to be two-to-one. This is an
abbreviation of the observation that two points in the domain map to the same point in the range.
Definition 1.2.4.
Suppose f : U ⊆ Rp → V ⊆ Rn and suppose further that for each x ∈ U ,
f (x) = (f1 (x), f2 (x), . . . , fn (x)).
Then we say that f = (f1 , f2 , . . . , fn ) and for each j ∈ Np the functions fj : U ⊆ Rp → R are
called the component functions of f . Furthermore, we define the projection πj : Rn → R
to be the map πj (x) = x · ej for each j = 1, 2, . . . n. This allows us to express each of the
component functions as a composition fj = πj ◦ f .
Example 1.2.5. Suppose f : R3 → R2 and f (x, y, z) = (x2 + y 2 , z) for each (x, y, z) ∈ R3 . Identify
that f1 (x, y, z) = x2 + y 2 whereas f2 (x, y, z) = z. You can easily see that range(f ) = [0, ∞] × R.
Suppose R2 ∈ [0, ∞) and zo ∈ R then
f −1 ({(R2 , zo )}) = S1 (R) × {zo }
where S1 (R) denotes a circle of radius R. This result is a simple consequence of the observation
that f (x, y, z) = (R2 , zo ) implies x2 + y 2 = R2 and z = zo .
Example 1.2.6. Let a, b, c ∈ R be particular constants. Suppose f : R3 → R and f (x, y, z) =

ax + by + cz for each (x, y, z) ∈ R3 . Here there is just one component function so we could say that
f = f1 but we don’t usually bother to make such an observation. If at least one of the constants
a,
b, c is
nonzero then the fibers of this map are planes in three dimensional space with normal
a, b, c .
f −1 ({d}) = {(x, y, z) ∈ R3 | ax + by + cz = d}
If a = b = c = 0 then the fiber of f is simply all of R3 and the range(f ) = {0}.
The definition below explains how to put together functions with a common domain. The codomain
of the new function is the cartesian product of the old codomains.
Definition 1.2.7.
Let f : U1 ⊆ Rn → V1 ⊆ Rp and g : U1 ⊆ Rn → V2 ⊆ Rq be a mappings then (f, g) is a
mapping from U1 to V1 × V2 defined by (f, g)(x) = (f (x), g(x)) for all x ∈ U1 .
There’s more than meets the eye in the definition above. Let me expand it a bit here:
(f, g)(x) = (f1 (x), f2 (x), . . . , fp (x), g1 (x), g2 (x), . . . , gq (x)) where x = (x1 , x2 , . . . , xn )
You might notice that Edwards uses π for the identity mapping whereas I use Id. His notation is
quite reasonable given that the identity is the cartesian product of all the projection maps:
π = (π1 , π2 , . . . , πn )
I’ve had courses where we simply used the coordinate notation itself for projections, in that nota-
tion have formulas such as x(a, b, c) = a, xj (a) = aj and xj (ei ) = δji .
Another way to modify a given function is to adjust the domain of a given mapping by restriction
and extension.
Definition 1.2.8.
Let f : U ⊆ Rn → V ⊆ Rm be a mapping. If R ⊂ U then we define the restriction of f
to R to be the mapping f |R : R → V where f |R (x) = f (x) for all x ∈ R. If U ⊆ S and
V ⊂ T then we say a mapping g : S → T is an extension of f iff g|dom(f ) = f .
When I say g|dom(f ) = f this means that these functions have matching domains and they agree at
each point in that domain; g|dom(f ) (x) = f (x) for all x ∈ dom(f ). Once a particular subset is chosen
the restriction to that subset is a unique function. Of course there are usually many susbets of
dom(f ) so you can imagine many different restictions of a given function. The concept of extension
is more vague, once you pick the enlarged domain and codomain it is not even necessarily the case
that another extension to that same pair of sets will be the same mapping. To obtain uniqueness
1.2. FUNCTIONS 17
for extensions one needs to add more stucture. This is one reason that complex variables are
interesting, there are cases where the structure of the complex theory forces the extension of a
complex-valued function of a complex variable to be unique. This is very surprising. Similarly a
linear transformation is uniquely defined by its values on a basis, it extends uniquely from that
finite set of vectors to the infinite number of points in the vector space. This is very restrictive on
the possible ways we can construct linear mappings. Maybe you can find some other examples of
extensions as you collect your own mathematical storybook.
Definition 1.2.9.
Let f : U ⊆ Rn → V ⊆ Rm be a mapping, if there exists a mapping g : f (U ) → U such that
f ◦ g = Idf (U ) and g ◦ f = IdU then g is the inverse mapping of f and we denote g = f −1 .
If a mapping is injective then it can be shown that the inverse mapping is well defined. We define
f −1 (y) = x iff f (x) = y and the value x must be a single value if the function is one-one. When a
function is not one-one then there may be more than one point which maps to a particular point
in the range.
Notice that the inverse image of a set is well-defined even if there is no inverse mapping. Moreover,
it can be shown that the fibers of a mapping are disjoint and their union covers the domain of the
mapping:
[
f (y) 6= f (z) ⇒ f −1 {y} ∩ f −1 {z} = ∅ f −1 {y} = dom(f ).
y ∈ range(f )
This means that the fibers of a mapping partition the domain.
Example 1.2.10. Consider f (x, y) = x2 + y 2 this describes a mapping from R2 to R. Observe that
f −1 {R2 } = {x2 + y 2 = R2 | (x, y) ∈ R2 }. In words, the nonempty fibers of f are concentric circles
about the origin and the origin itself.
Technically, the emptyset is always a fiber. It is the fiber over points in the codomain which are
not found in the range. In the example above, f −1 (−∞, 0) = ∅.
Definition 1.2.11.
Let f : U ⊆ Rn → V ⊆ Rm be a mapping. A cross section of the fiber partiition is a
subset S ⊆ U for which S ∩ f −1 {v} contains a single element for every v ∈ f (U ).
How do we construct a cross section for a particular mapping? For particular examples the details
of the formula for the mapping usually suggests some obvious choice. However, in general if you
accept the axiom of choice then you can be comforted in the existence of a cross section even in
the case that there are infinitely many fibers for the mapping.
Example 1.2.12. An easy cross-section for f (x, y) = x2 + y 2 is given by any ray eminating from
the origin. Notice that, if ab 6= 0 then S = {t(a, b) | t ∈ [0, ∞)} interects the a circle of radius
R2 = t2 (a2 + b2 ) at the point (ta, tb)
Proposition 1.2.13.
Let f : U ⊆ Rn → V ⊆ Rm be a mapping. The restriction of f to a cross section S

of U is an injective function. The mapping f˜ : U → f (U ) is a surjection. The mapping
f˜|S : S → f (U ) is a bijection.
The proposition above tells us that we can take any mapping and cut down the domain and/or
codomain to reduce the function to an injection, surjection or bijection. If you look for it you’ll see
this result behind the scenes in other courses. For example, in linear algebra if we throw out the
kernel of a linear mapping then we get an injection. The idea of a local inverse is also important
to the study of calculus.
Example 1.2.14. Continuing with our example, f : R2 → R with f (x, y) = x2 + y 2 is neither

surjective or injective. However, just to make a choice, S = {(t, 0) | t ∈ [0, ∞)} then clearly
f˜ : S → [0, ∞) defined by f˜(x, y) = f (x, y) for all (x, y) ∈ S is a bijection.
Definition 1.2.15.
Let f : U ⊆ Rn → V ⊆ Rm be a mapping then we say a mapping g is a local inverse of f
iff there exits S ⊆ U such that g = (f |S )−1 .
Usually we can find local inverses for functions in calculus. For example, f (x) = sin(x) is not 1-1
therefore it is not invertible. However, it does have a local inverse g(y) = sin−1 (y). If we were
−1
more pedantic we wouldn’t write sin−1 (y). Instead we would write g(y) = sin |[ −π , π ] (y) since
2 2
the inverse sine is actually just a local inverse. To construct a local inverse for some mapping we
must locate some subset of the domain upon which the mapping is injective. Then relative to that
subset we can reverse the mapping. The inverse mapping theorem (which we’ll study mid-course)
will tell us more about the existence of local inverses for a given mapping.
1.3 vectors and geometry for n-dimensional space

Definition 1.3.1.
Let n ∈ N, we define Rn = {(x1 , x2 , . . . , xn ) | xj ∈ R f or j = 1, 2, . . . , n}. If v ∈ Rn
then we say v is an n-vector. The numbers in the vector are called the components;
v = (v1 , v2 , . . . , vn ) has j-th component vj .
Notice, a consequence of the definition above and the construction of the Cartesian product6 is that
two vectors v and w are equal iff vj = wj for all j ∈ Nn . Equality of two vectors is only true if all
components are found to match. It is therefore logical to define addition and scalar multiplication
in terms of the components of vectors as follows:
6
see my Math 200 notes or ask me if interested, it’s not entirely trivial
1.3. VECTORS AND GEOMETRY FOR N -DIMENSIONAL SPACE 19
Definition 1.3.2.
Define functions + : Rn × Rn → Rn and · : R × Rn → Rn by the following rules: for each
v, w ∈ Rn and c ∈ R:
(1.) (v + w)j = vj + wj (2.) (cv)j = cvj
for all j ∈ {1, 2, . . . , n}. The operation + is called vector addition and it takes two
vectors v, w ∈ Rn and produces another vector v + w ∈ Rn . The operation · is called scalar
multiplication and it takes a number c ∈ R and a vector v ∈ Rn and produces another
vector c · v ∈ Rn . Often we simply denote c · v by juxtaposition cv.
If you are a gifted at visualization then perhaps you can add three-dimensional vectors in your
mind. If you’re mind is really unhinged maybe you can even add 4 or 5 dimensional vectors. The
beauty of the definition above is that we have no need of pictures. Instead, algebra will do just
fine. That said, let’s draw a few pictures.
Notice these pictures go to show how you can break-down vectors into component vectors which
point in the direction of the coordinate axis. Vectors of length one which point in the coordinate
directions make up what is called the standard basis7 It is convenient to define special notation
for the standard basis. First I define a useful shorthand,
Definition 1.3.3.
(
1 ,i = j
The symbol δij = is called the Kronecker delta.
0 , i 6= j
For example, δ22 = 1 while δ12 = 0.
7
the term ”basis” is carefully developed in the linear algebra course. In a nutshell we need two things: (1.) the
basis has to be big enough that we can add togther the basis elements to make any thing in the set (2.) the basis is
minimal so no single element in the basis can be formed by adding togther other basis elements
Definition 1.3.4.
Let ei ∈ Rn×1 be defined by (ei )j = δij . The size of the vector ei is determined by context.
We call ei the i-th standard basis vector.
Example 1.3.5. Let me expand on what I mean by ”context” in the definition above:
In R we have e1 = (1) = 1 (by convention we drop the brackets in this case)
In R2 we have e1 = (1, 0) and e2 = (0, 1).
In R3 we have e1 = (1, 0, 0) and e2 = (0, 1, 0) and e3 = (0, 0, 1).
In R4 we have e1 = (1, 0, 0, 0) and e2 = (0, 1, 0, 0) and e3 = (0, 0, 1, 0) and e4 = (0, 0, 0, 1).
A real linear combination of {v1 , v2 , · · · , vn } is simply a finite weighted-sum of the objects from
the set; c1 v1 + c2 v2 + · · · ck vk where c1 , c2 , · · · ck ∈ R. If we take coefficients c1 , c2 , · · · ck ∈ C then
is is said to be a complex linear combination. I invite the reader to verify that every vector in
Rn is a linear combination of e1 , e2 , . . . , en 8 . It is not difficult to prove the following properties for
vector addition and scalar multiplication: for all x, y, z ∈ Rn and a, b ∈ R,
(i.) x + y = y + x, (ii.) (x + y) + z = x + (y + z)
(iii.) x + 0 = x, (iv.) x − x = 0
(v.) 1x = x, (vi.) (ab)x = a(bx),
(vii.) a(x + y) = ax + ay, (viii.) (a + b)x = ax + bx
(ix.) x + y ∈ Rn (x.) cx ∈ Rn
These properties of Rn are abstracted in linear algebra to form the definition of an abstract vector
space. Naturally Rn is a vector space, in fact it is the quintessial model for all other vector spaces.
Fortunately Rn also has a dot-product. The dot-product is a mapping from Rn × Rn to R. We take
in a pair of vectors and output a real number.
Definition 1.3.6. Let x, y ∈ Rn we define x · y ∈ R by
x · y = x1 y1 + x2 y2 + · · · + xn yn .
Example 1.3.7. Let v = (1, 2, 3, 4, 5) and w = (6, 7, 8, 9, 10)
v · w = 6 + 14 + 24 + 36 + 50 = 130
Example 1.3.8. Suppose we are given a vector v ∈ Rn . We can select the j-th component by
taking the dot-product of v with ej . Observe that ei · ej = δij and consider,
Xn Xn n
X
v · ej = vi ei · ej = vi ei · ej = vi δij = v1 δ1j + · · · + vj δjj + · · · + δnj vn = vj .
i=1 i=1 i=1
The dot-product with ej has given us the length of the vector v in the j-th direction.
8
the calculation is given explicitly in my linear notes
The length or norm of a vector and the angle between two vectors are induced from the dot-product:
Definition 1.3.9.
√
The length or norm of x ∈ Rn is a real number which is defined by ||x|| = x · x.
Furthermore, n
−1
x·y let x, y be nonzero vectors in R we define the angle θ between x and y by
cos ||x|| ||y|| . R together with these defintions of length and angle forms a Euclidean
Geometry.
Technically, before we make this definition we should make sure that the formulas given above even
make sense. I have not shown that x · x is nonnegative and how do we know that argument of
the inverse cosine is within its domain of [−1, 1]? I now state the propositions which justify the
preceding definition.(proofs of the propositions below are found in my linear algebra notes)
Proposition 1.3.10.
Suppose x, y, z ∈ Rn and c ∈ R then
1. x · y = y · x
2. x · (y + z) = x · y + x · z
3. c(x · y) = (cx) · y = x · (cy)
4. x · x ≥ 0 and x · x = 0 iff x = 0
The formula cos−1 ||x||x·y||y|| is harder to justify. The inequality that we need for it to be reasonable

is ||x||x·y||y|| ≤ 1, otherwise we would not have a number in the dom(cos−1 ) = range(cos) = [−1, 1].

An equivalent inequality is |x · y| ≤ ||x|| ||y|| which is known as the Cauchy-Schwarz inequality.
Proposition 1.3.11.
If x, y ∈ Rn then |x · y| ≤ ||x|| ||y||

Example 1.3.12. Let v = [1, 2, 3, 4, 5]T and w = [6, 7, 8, 9, 10]T find the angle between these vectors
and calculate the unit vectors in the same directions as v and w. Recall that, v · w = 6 + 14 + 24 +
36 + 50 = 130. Furthermore,
p √ √
||v|| = 12 + 22 + 32 + 42 + 52 = 1 + 4 + 9 + 16 + 25 = 55
p √ √
||w|| = 62 + 72 + 82 + 92 + 102 = 36 + 49 + 64 + 81 + 100 = 330
We find unit vectors via the standard trick, you just take the given vector and multiply it by the
reciprocal of its length. This is called normalizing the vector,
v̂ = √1 [1, 2, 3, 4, 5]T ŵ = √ 1 [6, 7, 8, 9, 10]T

55 330
The angle is calculated from the definition of angle,

−1 130
θ = cos √ √ = 15.21o
55 330
It’s good we have this definition, 5-dimensional protractors are very expensive.
Proposition 1.3.13.
Let x, y ∈ Rn and suppose c ∈ R then
1. ||cx|| = |c| ||x||
2. ||x + y|| ≤ ||x|| + ||y|| (triangle inequality)
3. ||x|| ≥ 0
4. ||x|| = 0 iff x = 0
The four properties above make Rn paired with || · || : Rn × Rn → R a normed linear space.
We’ll see how differentiation can be defined given this structure. It turns out that we can define a
reasonable concept of differentiation for other normed linear spaces. In this course we’ll study how
to differentiate functions to and from Rn , matrix-valued functions and complex-valued functions of
a real variable. Finally, if time permits, we’ll study differentiation of functions of functions which
is the central task of variational calculus. In each case the underlying linear structure along
with the norm is used to define the limits which are necessary to set-up the derivatives. The focus
of this course is the process and use of derivatives and integrals so I have not given proofs of the
linear algebraic propositions in this chapter. The proofs and a deeper view of the meaning of these
propositions is given at length in Math 321. If you haven’t had linear then you’ll just have to trust
me on these propositions9
9
or you could just read the linear notes if curious
Definition 1.3.14.
The distance between a ∈ Rn and b ∈ Rn is defined to be d(a, b) ≡ ||b − a||.
If we draw a picture this definition is very natural. Here we are thinking of the points a, b as vectors
from the origin then b − a is the vector which points from a to b (this is algebraically clear since
a + (b − a) = b). Then the distance between the points is the length of the vector that points from
one point to the other. If you plug in two dimensional vectors you should recognize the distance
formula from middle school math:
p
d((a1 , a2 ), (b1 , b2 )) = (b1 − a1 )2 + (b2 − a2 )2
Proposition 1.3.15.
Let d : Rn × Rn → R be the distance function then
1. d(x, y) = d(y, x)
2. d(x, y) ≥ 0
3. d(x, x) = 0 iff x = 0
4. d(x, y) + d(y, z) ≥ d(x, z)
In real analysis one studies a set paired with a distance function. Abstractly speaking such a pair
is called a metric space. A vector space with a norm is called a normed linear space. Because
we can always induce a distance function from the norm via the formula d(a, b) = ||b − a|| every
normed linear space is a metric space. The converse fails. Metric spaces need not be vector spaces,
a metric space could just be formed from some subset of a vector space or something more exotic10 .
The absolute value function on R defines distance function d(a, b) = |b − a|. In your real analysis
course you will study the structure of the metric space (R, | · | : R × R → R) in great depth. I
include these comments here to draw your attention to the connection between this course and the
real analysis course. I primarily use the norm in what follows, but it should be noted that many
things could be written in terms of the distance function.
1.3.1 vector algebra for three dimensions

Every nonzero vector can be written as a unit vector scalar multiplied by its magnitude.
v ∈ V n such that v 6= 0 ⇒ v = ||v||v̂ where v̂ = 1

||v|| v.
You should recall that we can write any vector in V 3 as

v =< a, b, c >= a < 1, 0, 0 > +b < 0, 1, 0 > +c < 0, 0, 1 >= aî + bĵ + ck̂
10
there are many texts to read on metric spaces, one nice treatment is Rosenlicht’s Introduction to Analysis, it’s a
good read
where we defined the î =< 1, 0, 0 >, ĵ =< 0, 1, 0 >, k̂ =< 0, 0, 1 >. You can easily verify that
distinct Cartesian unit-vectors are orthogonal. Sometimes we need to produce a vector which is
orthogonal to a given pair of vectors, it turns out the cross-product is one of two ways to do that
in V 3 . We will see much later that this is special to three dimensions.
Definition 1.3.16.
If A =< A1 , A2 , A3 > and B =< B1 , B2 , B3 > are vectors in V 3 then the cross-product
of A and B is a vector A × B which is defined by:
~×B
A ~ =< A2 B3 − A3 B2 , A3 B1 − A1 B3 , A1 B2 − A2 B1 > .
~ ×B
The magnitude of A ~ can be shown to satisfy ||A
~ × B||
~ = ||A||
~ ||B||
~ sin(θ) and the direction can
be deduced by right-hand-rule. The right hand rule for the unit vectors yields:
î × ĵ = k̂, k̂ × î = ĵ, ĵ × k̂ = î
If I wish to discuss both the point and the vector to which it corresponds we may use the notation
P = (a1 , a2 , . . . , an ) ←→ P~ =< a1 , a2 , . . . , an >
With this notation we can easily define directed line-segments as the vector which points from one
point to another, also the distance bewtween points is simply the length of the vector which points
from one point to the other:
Definition 1.3.17.
−−→ ~ − P~ . This vector is
Let P, Q ∈ Rn . The directed line segment from P to Q is P Q = Q
drawn from tail Q to the tip P where we denote the direction by drawing an arrowhead.
−−→
The distance between P and Q is d(P, Q) = || P Q ||.
1.3.2 compact notations for vector arithmetic

I prefer the following notations over the hat-notation of the preceding section because this notation
generalizes nicely to n-dimensions.
e1 =< 1, 0, 0 > e2 =< 0, 1, 0 > e3 =< 0, 0, 1 > .
Likewise the Kronecker delta and the Levi-Civita symbol are at times very convenient for abstract
calculation:

(
1
 (i, j, k) ∈ {(1, 2, 3), (3, 1, 2), (2, 3, 1)}
1 i=j
δij = ijk = −1 (i, j, k) ∈ {(3, 2, 1), (2, 1, 3), (1, 3, 2)}
0 i 6= j 
0 if any index repeats

An equivalent definition for the Levi-Civita symbol is simply that 123 = 1 and it is antisymmetric
with respect to the interchange of any pair of indices;
ijk = jki = kij = −kji = −jik = −ikj .
Now let us restate some earlier results in terms of the Einstein repeated index conventions11 , let
~ B
A, ~ ∈ V n and c ∈ R then
~ = Ak ek
A standard basis expansion
ei · ej = δij orthonormal basis
~ + B)
(A ~ i=A ~i + B
~i vector addition
~ ~ ~
(A − B)i = Ai − B ~i vector subtraction
~ i = cA
(cA) ~i scalar multiplication
~·B
A ~ = Ak Bk dot product
~
(A × B) ~ k = ijk Ai Bj cross product.
All but the last of the above are readily generalized to dimensions other than three by simply
increasing the number of components. However, the cross product is special to three dimensions.
I can’t emphasize enough that the formulas given above for the dot and cross products can be
utilized to yield great efficiency in abstract calculations.
Example 1.3.18. To prove the linearity of the cross-product in the second argument:
~ × (B
(A ~ + C))
~ k = ijk Ai (B
~ + C)
~ j
= ijk Ai (Bj + Cj )
= ijk Ai Bj + ijk Ai Cj
= (A~ × B)
~ k + (A ~ × C)
~ k.
If you look at my grad, curl and div section in Chapter 7 of my Calculus III notes you’ll see how
this notation allows very elegant proofs of the basic identities of differential vector calculus.
11
I don’t use the Einstein convention generally until the end of these notes, if in doubt, ask.
Chapter 2
linear algebra
Our goal in the first section of this chapter is to gain conceptual clarity on the meaning of the
central terms from linear algebra. This is a birds-eye view of linear, my selection of topics here is
centered around the goal of helping you to see the linear algebra in calculus. Once you see it then
you can use the theory of linear algebra to help organize your thinking. Our ultimate goal is that
organizational principle. Our goal here is not to learn all of linear algebra, rather we wish to use it
as a tool for the right jobs as they arise this semester.
In the second section we summarize the tools of matrix computation. We will use matrix addition,
multiplication and throughout this course. Inverse matrices and the noncommuative nature of
matrix multiplication are illustrated. It is assumed that the reader has some previous experience
in matrix computation, at least in highschool you should have spent some time.
In the third section the concept of a linear transformation is formalized. The formula for any
linear transformation from Rm to Rm can be expressed as a matrix multiplication. We study
this standard matrix in enough depth as to understand it’s application in for differentiation. A
number of examples to visualize the role of a linear transformation are offered for breadth. Finally,
isomorphisms and coordinate maps are discussed.
In the fourth section we define norms for vector spaces. We study how the norm allows us to define
limits for an abstract vector space. This is important since it allows us to quantify continuity for
abstract linear transformations as well as ultimately to define differentiation on a normed vector
space in the chapter that follows.
One important thing to remember, we do not typically use notation to denote domain. Sometimes
x is a vector. Sometimes x is the first coordinate function. Therefore, if we just say ”x” then it
is ambiguous. We must state domains and context to engage in meaningful math here. Notation
which encodes domain is convenient, but our interests are too varied to allow such abbreviated
language.
27
28 CHAPTER 2. LINEAR ALGEBRA
2.1 vector spaces

Suppose we have a set V paired with an addition and scalar multiplication such that for all x, y, z ∈
V and a, b, c ∈ R:
(i.) x + y = y + x, (ii.) (x + y) + z = x + (y + z)
(iii.) x + 0 = x, (iv.) x − x = 0
(v.) 1x = x, (vi.) (ab)x = a(bx),
(vii.) a(x + y) = ax + ay, (viii.) (a + b)x = ax + bx
(ix.) x + y ∈ Rn (x.) cx ∈ Rn
then we say that V is a vector space over R. To be a bit more precise, by (iii.) I mean to say
that there exist some element 0 ∈ V such that x + 0 = x for each x ∈ V . Also, (iv.) should be
understood to say that for each x ∈ V there exists another element −x ∈ V such that x + (−x) = 0.
Example 2.1.1. Rn is a vector space with respect to the standard vector addition and scalar
multiplication.
Example 2.1.2. C = {a + ib | a, b ∈ R} is a vector space where the usual complex number addition
provides the vector addition and multiplication by a real number c(a + ib) = ca + i(cb) clearly defines
a scalar multiplication.
Example 2.1.3. The set of all m × n matrices is a vector space with respect to the usual matrix
addition and scalar multiplication. We will elaborate on the details in an upcoming section.
Example 2.1.4. Suppose F is the set of all functions from a set S to a vector space V then F is
naturally a vector space with respect to the function addition and multiplication by a scalar. Both
of those operations are well-defined on the values of the function since we assumed the codomain of
each function in F is the vector space V .
There are many subspaces of function space which provide interesting examples of vector spaces.
For example, the set of continuous functions:
Example 2.1.5. Let C 0 (Rn , Rm ) be a set of continuous functions from Rn to Rm then C 0 (Rn , Rm )
is a vector space with respect to function addition and the usual multiplication. This fact relies on
the sum and scalar multiple of continuous functions is once more continuous.
Definition 2.1.6.
We say a subset S of a vector space V is linearly independent (LI) iff for scalars
c1 , c2 , . . . , ck ,
c1 v1 + c2 v2 + · · · ck vk = 0 ⇒ c1 = c2 = · · · = 0
for each finite subset {v1 , v2 , . . . , vk } of S.
In the case that S is finite it suffices to show the implication for a linear combination of all the
vectors in the set. Notice that if any vector in the set S can be written as a linear combination of
the other vectors in S then that makes S fail the test for linear independence. Moreover, if a set S
is not linearly independent then we say S is linearly dependent.
2.1. VECTOR SPACES 29
Example 2.1.7. The standard basis of Rn is denoted {e1 , e2 , . . . , en }. We can show linear inde-
pendence easily via the dot-product: suppose that c1 e1 + c2 e2 + · · · cn en = 0 and take the dot-product
of both sides with ej to obtain
(c1 e1 + c2 e2 + · · · cn en ) · ej = 0 · ej ⇒ (c1 e1 · ej + c2 e2 · ej + · · · cn en · ej = 0 ⇒ cj (1) = 0
but, j was arbitrary hence it follows that c1 = c2 = · · · = cn = 0 which establishes the linear
independence of the standard basis.
Example 2.1.8. Consider S = {1, i} ⊂ C. We can argue S is LI as follows: suppose c1 (1)+c2 (i) =
0. Thus c1 +ic2 = 0 for some real numbers c1 , c2 . Recall that a basic property of complex numbers is
that if z1 = z2 then Re(z1 ) = Re(z2 ) and Im(z1 ) = Im(z2 ) where zj = Re(zj )+iIm(zj ). Therefore,
the complex equation c1 + ic2 = 0 yields two real equations c1 = 0 and c2 = 0.
Example 2.1.9. Let C 0 (R) be the vector space of all continuous functions from R to R. Suppose
S is the set of monic1 monomials S = {1, t, t2 , t3 , . . . }. This is an infinite set. We can argue LI
as follows: suppose c1 tp1 + c2 tp2 + · · · + ck tpk = 0. For convenience relable the powers p1 , p2 , . . . , pk
by pi1 , pi2 , . . . , pik such that 1 < pi1 < pi2 < · · · < pik . This notation just shuffles the terms in the
finite sum around so that the first term has the lowest order: consider
ci1 tpi1 + ci1 tpi2 + · · · + cik tpik = 0 ?
If pi1 = 0 then evaluate ? at t = 0 to obtain ci1 = 0. If pi1 > 0 then differentiate ? pi1 times and
denote this new equation by Dpi1 ?. Evaluate Dpi1 ? at t = 0 to find
pi1 (pi1 − 1) · · · 3 · 2 · 1ci1 = 0
hence ci1 = 0. Since we set-up pi1 < pi2 it follows that after pi1 -differentiations the second summand
is still nontrivial in Dpi1 ?. However, we can continue differentiating ? until we reach Dpi2 ? and
then constant term is pi2 !ci2 so evaluation will show ci2 = 0. We continue in this fashion until we
have shown that cij = 0 for j = 1, 2, . . . k. It follows that S is a linearly independent set.
We spend considerable effort in linear algebra to understand LI from as many angles as possible.
One equivalent formulation of LI is the ability to equate coefficients. In other words, a set of objects
is LI iff whenever we have an equation with thos objects we can equate coefficients. In calculus
when we equate coefficients we implicitly assume that the functions in question are LI. Generally
speaking two functions are LI if their graphs have distinct shapes which cannot be related by a
simple vertical stretch.
Example 2.1.10. Consider S = {2t , 3(1/2)−t } as a subset the vector space C 0 (R). To show linear
dependence we observe that c1 2t + c2 3(1/2)−t = 0 yields (c1 + 3c2 )2t = 0. Hence c1 + 3c2 = 0 which
means nontrivial solutions exist. Take c2 = 1 then c1 = −3. Of course the heart of the matter is
that 3(1/2)−t = 3(2t ) so the second function is just a scalar multiple of the first function.
1
monic means that the leading coefficient is 1.
If you’ve taken differential equations then you should recognize the concept of LI from your study
of solution sets to differential equations. Given an n-th order linear differential equation we always
have a goal of calculating n-LI solutions. In that context LI is important because it helps us
make sure we do not count the same solution twice. The general solution is formed from a linear
combination of the LI solution set. Of course this is not a course in differential equations, I include
this comment to make connections to the other course. One last example on LI should suffice to
make certain you at least have a good idea of the concept:
Example 2.1.11. Consider R3 as a vector space and consider the set S = {~v , î, ĵ, k̂} where we
could also denote î = e1 , ĵ = e2 , k̂ = e3 but I’m aiming to make your mind connect with your
calculus III background. This set is clearly linearly dependent since we can write any vector ~v as
a linear combination of the standard unit-vectors: moreover, we can use dot-products to select the
x, y and z components as follows:
~v = (~v · î)î + (~v · ĵ)ĵ + (~v · k̂)k̂
Linear independence helps us quantify a type of redundancy for vectors in a given set. The next
definition is equally important and it is sort of the other side of the coin; spanning is a criteria
which helps us insure a set of vectors will cover a vector space without missing anything.
Definition 2.1.12.
We say a subset S of a vector space V is a spanning set for V iff for each v ∈ V there
exist scalars c1 , c2 , . . . , ck and vectors v1 , v2 , . . . , vk ∈ V such that v = c1 v1 + c2 v2 + · · · ck vk .
We denote Span{v1 , v2 , . . . , vk } = {c1 v1 + c2 v2 + · · · ck vk | c1 , c2 , . . . , ck ∈ R}.
If S ⊂ V and V is a vector space then it is immediately obvious that Span(S) ⊆ V . If S is a
spanning set then it is obvious that V ⊆ Span(S). It follows that when S is a spanning set for V
we have Span(S) = V .
Example 2.1.13. It is easy to show that if v ∈ Rn then v = v1 e1 + v2 e2 + · · · + vn en . It follows
that Rn = Span{ei }ni=1 .
Example 2.1.14. Let 1, i ∈ C where i2 = −1. Clearly C = Span{1, i}.
Example 2.1.15. Let P be the set of polynomials. Since the sum of any two polynomials and
the scalar multiple of any polynomial is once more a polynomial we find P is a vector space with
respect to function addition and multiplication of a function by a scalar. We can argue that the set
of monic monomials {1, t, t2 , . . . } a spanning set for P . Why? Because if f (t) ∈ P then that means
there are scalars a0 , a1 , . . . , an such that f (x) = a0 + a1 t + a2 t2 + · · · + an tn
Definition 2.1.16.
We say a subset β of a vector space V is a basis for V iff β is a linearly independent
spanning set for V . If β is a finite set then V is said to be finite dimensional and the
number of vectors in β is called the dimension of V . That is, if β = {v1 , v2 , . . . , vn } is a
basis for V then dim(V ) = n. If no finite basis exists for V then V is said to be infinite
dimensional.
2.2. MATRIX CALCULATION 31
The careful reader will question why this concept of dimension is well-defined. Why can we not
have bases of differing dimension for a given vector space? I leave this question for linear algebra,
the theorem which asserts the uniqueness of dimension is one of the deeper theorems in the course.
However, like most everything in linear, at some level it just boils down to solving some particular
set of equations. You might tell Dr. Sprano it’s just algebra. In any event, it is common practice
to use the term dimension in courses where linear algebra is not understood. For example, R2 is a
two-dimensional space. Or we’ll say that R3 is a three-dimensional space. This terminology agrees
with the general observation of the next example.
Example 2.1.17. The standard basis {ei }ni=1 for Rn is a basis for Rn and dim(Rn ) = n. This
result holds for all n ∈ N. The line is one-dimensional, the plane is two-dimensional, three-space
is three-dimensional etc...
Example 2.1.18. The set {1, i} is a basis for C. It follows that dim(C) = 2. We say that the
complex numbers form a two-dimensional real vector space.
Example 2.1.19. The set of polynomials is clearly infinite dimensional. Contradiction shows this
without much effort. Suppose P had a finite basis β. Choose the polynomial of largest degree (say
k) in β. Notice that f (t) = tk+1 is a polynomial and yet clearly f (t) ∈
/ Span(β) hence β is not a
spanning set. But this contradicts the assumption β is a basis. Hence, by contradiction, no finite
basis exists and we conclude the set of polynomials is infinite dimensional.
There is a more general use of the term dimension which is beyond the context of linear algebra.
For example, in calculus II or III you may have heard that a circle is one-dimensional or a surface
is two-dimensional. Well, circles and surfaces are not usually vector spaces so the terminology is
not taken from linear algebra. In fact, that use of the term dimension stems from manifold theory.
I hope to discuss manifolds later in this course.
2.2 matrix calculation

An m × n matrix is an array of numbers with m-rows and n-columns. We define R m×n to be the
set of all m × n matrices. The set of all n-dimensional column vectors is R n×1 = Rn2 . The set of
all n-dimensional row vectors is R1×n . A given matrix A ∈ R m×n has mn-components Aij . Notice
that the components are numbers; Aij ∈ R for all i, j such that 1 ≤ i ≤ m and 1 ≤ j ≤ n. We
should not write A = Aij because it is nonesense, however A = [Aij ] is quite fine.
Suppose A ∈ R m×n , note for 1 ≤ j ≤ n we have colj (A) ∈ Rm×1 whereas for 1 ≤ i ≤ m we find
rowi (A) ∈ R1×n . In other words, an m×n matrix has n columns of length m and n rows of length m.
2
We will use the convention that points in Rn are column vectors. However, we will use the somewhat subtle
notation (x1 , x2 , . . . xn ) = [x1 , x2 , . . . xn ]T . This helps me write Rn rather than R n×1 and I don’t have to pepper
transposes all over the place. If you’ve read my linear algebra notes you’ll appreciate the wisdom of our convention.
Two matrices A and B are equal iff Aij = Bij for all i, j. Given matrices A, B with components
Aij , Bij and constant c ∈ R we define
(A + B)ij = Aij + Bij (cA)ij = cAij , for all i, j.
The zero matrix in R m×n is denoted 0 and defined by 0ij = 0 for all i, j. The additive inverse
of A ∈ R m×n is the matrix −A such that A + (−A) = 0. The components of the additive inverse
matrix are given by (−A)ij = −Aij for all i, j. Likewise, if A ∈ R m×n and B ∈ R n×p then the
product AB ∈ R m×p is defined by3 :
n
X
(AB)ij = Aik Bkj
k=1
for each 1 ≤ i ≤ m and 1 ≤ j ≤ p. In the case m = p = 1 the indices i, j are omitted in the equation
since the matrix product is simply a number which needs no index. The identity matrix n×n
( in R
1 i=j
is the n × n square matrix I whose components are the Kronecker delta; Iij = δij = .
0 i 6= j

1 0
The notation In is sometimes used. For example, I2 = . If the size of the identity matrix
0 1
needs emphasis otherwise the size of the matrix I is to be understood from the context.
Let A ∈ R n×n . If there exists B ∈ R n×n such that AB = I and BA = I then we say that A
is invertible and A−1 = B. Invertible matrices are also called nonsingular. If a matrix has no
inverse then it is called a noninvertible or singular matrix.
Let A ∈ R m×n then AT ∈ R n×m is called the transpose of A and is defined by (AT )ji = Aij
for all 1 ≤ i ≤ m and 1 ≤ j ≤ n. It is sometimes useful to know that (AB)T = B T AT and
(AT )−1 = (A−1 )T . It is also true that (AB)−1 = B −1 A−1 . Furthermore, note dot-product of
v, w ∈ V n is given by v · w = v T w.
The ij-th standard basis matrix for R m×n is denoted Eij for 1 ≤ i ≤ m and 1 ≤ j ≤ n. The
matrix Eij is zero in all entries except for the (i, j)-th slot where it has a 1. In other words, we
define (Eij )kl = δik δjl . I invite the reader to show that the term basis is justified in this context4 .
Given this basis we see that the vector space R m×n has dim(R m×n ) = mn.
3
this product is defined so the matrix of the composite of a linear transformation is the product of the matrices
of the composed transformations. This is illustrated later in this section and is proved in my linear algebra notes.
4
the theorem stated below contains the needed results and then some, you can find the proof is given in my linear
algebra notes. It would be wise to just work it out in the 2 × 2 case as a warm-up if you are interested
Theorem 2.2.1.
If A ∈ R m×n and v ∈ R then

n
X m X
X n
(i.) v = vn en (ii.) A = Aij Eij
i=1 i=1 j=1
(iii.) [ei T A] = rowi (A) (iv.) [Aei ] = coli (A)

(v.) Eij Ekl = δjk Eil (vi.) Eij = ei ej T (vii.) ei T ej = ei · ej = δij .
You can look in my linear algebra notes for the details of the theorem. I’ll just expand one point
here: Let A ∈ R m×n then
 
A11 A12 · · · A1n
 A21 A22 · · · A2n 
A = .
 
.. .. 
 .. . ··· . 
Am1 Am2 · · · Amn
     
1 0 ··· 0 0 1 ··· 0 0 0 ··· 0
 0 0 ··· 0   0 0 ··· 0   0 0 ··· 0 
= A11   + A12   + · · · + Amn 
     
.. .. .. .. .. .. 
 . . ··· 0   . . ··· 0   . . ··· 0 
0 0 ··· 0 0 0 ··· 0 0 0 ··· 1
= A11 E11 + A12 E12 + · · · + Amn Emn .
The calculation above follows from repeated mn-applications of the definition of matrix addition
and another mn-applications of the definition of scalar multiplication of a matrix.
Example 2.2.2. Suppose A = [ 14 25 36 ]. We see that A has 2 rows and 3 columns thus A ∈ R2×3 .
Moreover, A11 = 1, A12 = 2, A13 = 3, A21 = 4, A22 = 5, and A23 = 6. It’s not usually possible to
find a formula for a generic element in the matrix, but this matrix satisfies Aij = 3(i − 1) + j for
all i, j 5 . The columns of A are,

1 2 3
col1 (A) = , col2 (A) = , col3 (A) = .
4 5 6
The rows of A are

row1 (A) = 1 2 3 , row2 (A) = 4 5 6
5
In the statement ”for all i, j” it is to be understood that those indices range over their allowed values. In the
preceding example 1 ≤ i ≤ 2 and 1 ≤ j ≤ 3.
h1 4i
Example 2.2.3. Suppose A = [ 14 25 36 ] then AT = 2 5 . Notice that
36
row1 (A) = col1 (AT ), row2 (A) = col2 (AT )
and
col1 (A) = row1 (AT ), col2 (A) = row2 (AT ), col3 (A) = row3 (AT )
Notice (AT )ij = Aji = 3(j − 1) + i for all i, j; at the level of index calculations we just switch the
indices to create the transpose.
Example 2.2.4. Let A = [ 13 24 ] and B = [ 57 68 ]. We calculate

1 2 5 6 6 8
A+B = + = .
3 4 7 8 10 12

1 2 5 6 −4 −4
A−B = − = .
3 4 7 8 −4 −4
Now multiply A by the scalar 5,

1 2 5 10
5A = 5 =
3 4 15 20
Example 2.2.6. Let A, B ∈ R m×n be defined by Aij = 3i + 5j and Bij = i2 for all i, j. Then we
can calculate (A + B)ij = 3i + 5j + i2 for all i, j.
Example 2.2.7. Solve the following matrix equation,

x y −1 −2 0 0 x−1 y−2
0= + ⇒ =
z w −3 −4 0 0 z−3 w−4
The definition of matrix equality means this single matrix equation reduces to 4 scalar equations:
0 = x − 1, 0 = y − 2, 0 = z − 3, 0 = w − 4. The solution is x = 1, y = 2, z = 3, w = 4.
The definition of matrix multiplication ((AB)ij = nk=1 Aik Bkj ) is very nice for general proofs, but
P
pragmatically I usually think of matrix multiplication in terms of dot-products. It turns out we can
view the matrix product as a collection of dot-products: suppose A ∈ R m×n and B ∈ R n×p then
 
row1 (A) · col1 (B) row1 (A) · col2 (B) ··· row1 (A) · colp (B)
 row2 (A) · col1 (B) row2 (A) · col2 (B) ··· row2 (A) · colp (B) 
AB = 
 
.. .. .. 
 . . ··· . 
rowm (A) · col1 (B) rowm (A) · col2 (B) · · · rowm (A) · colp (B)
Let me explain how this works. The formula above claims (AB)ij = rowi (A) · colj (B) for all i, j.
Recall that (rowi (A))k = Aik and (colj (B))k = Bkj thus
n
X n
X
(AB)ij = Aik Bkj = (rowi (A))k (colj (B))k
k=1 k=1
Hence, using definition of the dot-product, (AB)ij = rowi (A) · colj (B). This argument holds for
all i, j therefore the dot-product formula for matrix multiplication is valid.
Example 2.2.8. The product of a 3 × 2 and 2 × 3 is a 3 × 3
  
[1, 0][4, 7]T [1, 0][5, 8]T [1, 0][6, 9]T
  
1 0 4 5 6
4 5 6
 0 1  =  [0, 1][4, 7]T [0, 1][5, 8]T [0, 1][6, 9]T  =  7 8 9 
 
7 8 9
0 0 [0, 0][4, 7]T [0, 0][5, 8]T [0, 0][6, 9]T 0 0 0
Example 2.2.9. The product of a 3 × 1 and 1 × 3 is a 3 × 3
     
1 4 · 1 5 · 1 6 · 1 4 5 6
 2  4 5 6 =  4 · 2 5 · 2 6 · 2  =  8 10 12 
3 4·3 5·3 6·3 12 15 18

1 2 5 6
AB =
3 4 7 8
" #
[1, 2][5, 7]T [1, 2][6, 8]T
=
[3, 4][5, 7]T [3, 4][6, 8]T

5 + 14 6 + 16
=
15 + 28 18 + 32

19 22
=
43 50
Notice the product of square matrices is square. For numbers a, b ∈ R it we know the product of a
and b is commutative (ab = ba). Let’s calculate the product of A and B in the opposite order,

5 6 1 2
BA =
7 8 3 4
" #
[5, 6][1, 3]T [5, 6][2, 4]T
=
[7, 8][1, 3]T [7, 8][2, 4]T

5 + 18 10 + 24
=
7 + 24 14 + 32

23 34
=
31 46
Clearly AB 6= BA thus matrix multiplication is noncommutative or nonabelian.
When we say that matrix multiplication is noncommuative that indicates that the product of two
matrices does not generally commute. However, there are special matrices which commute with
other matrices.

Example 2.2.11. Let I = [ 10 01 ] and A = ac db . We calculate

1 0 a b a b
IA = =
0 1 c d c d
Likewise calculate,
a b 1 0 a b
AI = =
c d 0 1 c d
Since the matrix A was arbitrary we conclude that IA = AI for all A ∈ R2×2 .
Example 2.2.12. Consider A, v, w from Example ??.

5 6 11
v+w = + =
7 8 15
Using the above we calculate,

1 2 11 11 + 30 41
A(v + w) = = = .
3 4 15 33 + 60 93
In constrast, we can add Av and Aw,

19 22 41
Av + Aw = + = .
43 50 93
Behold, A(v + w) = Av + Aw for this example. It turns out this is true in general.
I collect all my favorite properties for matrix multiplication in the theorem below. To summarize,
matrix math works as you would expect with the exception that matrix multiplication is not
commutative. We must be careful about the order of letters in matrix expressions.
Theorem 2.2.13.
If A, B, C ∈ R m×n , X, Y ∈ R n×p , Z∈R p×q and c1 , c2 ∈ R then
1. (A + B) + C = A + (B + C),
2. (AX)Z = A(XZ),
3. A + B = B + A,
4. c1 (A + B) = c1 A + c2 B,
5. (c1 + c2 )A = c1 A + c2 A,
6. (c1 c2 )A = c1 (c2 A),
7. (c1 A)X = c1 (AX) = A(c1 X) = (AX)c1 ,
8. 1A = A,
9. Im A = A = AIn ,
10. A(X + Y ) = AX + AY ,
11. A(c1 X + c2 Y ) = c1 AX + c2 AY ,
12. (A + B)X = AX + BX,
Proof: I will prove a couple of these primarily to give you a chance to test your understanding
of the notation. Nearly all of these properties are proved by breaking the statement down to
components then appealing to a property of real numbers. Just a reminder, we assume that it is
known that R is an ordered field. Multiplication of real numbers is commutative, associative and
distributes across addition of real numbers. Likewise, addition of real numbers is commutative,
associative and obeys familar distributive laws when combined with addition.
Proof of (1.): assume A, B, C are given as in the statement of the Theorem. Observe that
((A + B) + C)ij = (A + B)ij + Cij defn. of matrix add.
= (Aij + Bij ) + Cij defn. of matrix add.
= Aij + (Bij + Cij ) assoc. of real numbers
= Aij + (B + C)ij defn. of matrix add.
= (A + (B + C))ij defn. of matrix add.
for all i, j. Therefore (A + B) + C = A + (B + C).
Proof of (6.): assume c1 , c2 , A are given as in the statement of the Theorem. Observe that
((c1 c2 )A)ij = (c1 c2 )Aij defn. scalar multiplication.
= c1 (c2 Aij ) assoc. of real numbers
= (c1 (c2 A))ij defn. scalar multiplication.
for all i, j. Therefore (c1 c2 )A = c1 (c2 A).

Proof of (10.): assume A, X, Y are given as in the statement of the Theorem. Observe that
P
((A(X + Y ))ij = Pk Aik (X + Y )kj defn. matrix multiplication,
= Pk Aik (Xkj + Ykj ) defn. matrix addition,
= Pk (Aik Xkj +P Aik Ykj ) dist. of real numbers,
= k Aik Xkj + k Aik Ykj ) prop. of finite sum,
= (AX)ij + (AY )ij defn. matrix multiplication(× 2),
= (AX + AY )ij defn. matrix addition,
for all i, j. Therefore A(X + Y ) = AX + AY .
The proofs of the other items are similar, we consider the i, j-th component of the identity and then
apply the definition of the appropriate matrix operation’s definition. This reduces the problem to
a statement about real numbers so we can use the properties of real numbers at the level of
components. Then we reverse the steps. Since the calculation works for arbitrary i, j it follows the
the matrix equation holds true.
2.3 linear transformations

Definition 2.3.1.
Let V and W be vector spaces over R. If a mapping L : V → W satisfies
1. L(x + y) = L(x) + L(y) for all x, y ∈ V ; this is called additivity.
2. L(cx) = cL(x) for all x ∈ V and c ∈ R; this is called homogeneity.
then we say L is a linear transformation. If V = W then we may say that L is a linear

transformation on V .
Proposition 2.3.2.
If A ∈ R m×n and L : Rn → Rm is defined by L(x) = Ax for each x ∈ Rn then L is a linear

transformation.
Proof: Let A ∈ R m×n and define L : Rn → Rm by L(x) = Ax for each x ∈ Rn . Let x, y ∈ Rn and
c ∈ R,
L(x + y) = A(x + y) = Ax + Ay = L(x) + L(y)
and
L(cx) = A(cx) = cAx = cL(x)
thus L is a linear transformation.
Obviously this gives us a nice way to construct examples. The following proposition is really at the
heart of all the geometry in this section.
2.3. LINEAR TRANSFORMATIONS 39
Proposition 2.3.3.
Let L = {p + tv | t ∈ [0, 1], p, v ∈ Rn with v 6= 0} define a line segment from p to p + v in
Rn . If T : Rn → Rm is a linear transformation then T (L) is a either a line-segment from
T (p) to T (p + v) or a point.
Proof: suppose T and L are as in the proposition. Let y ∈ T (L) then by definition there exists
x ∈ L such that T (x) = y. But this implies there exists t ∈ [0, 1] such that x = p + tv so
T (p + tv) = y. Notice that
y = T (p + tv) = T (p) + T (tv) = T (p) + tT (v).
which implies y ∈ {T (p) + sT (v) | s ∈ [0, 1]} = L2 . Therefore, T (L) ⊆ L2 . Conversely, suppose
z ∈ L2 then z = T (p) + sT (v) for some s ∈ [0, 1] but this yields by linearity of T that z = T (p + sv)
hence z ∈ T (L). Since we have that T (L) ⊆ L2 and L2 ⊆ T (L) it follows that T (L) = L2 . Note
that L2 is a line-segment provided that T (v) 6= 0, however if T (v) = 0 then L2 = {T (p)} and the
proposition follows.
2.3.1 a gallery of linear transformations

My choice of mapping the unit square has no particular signficance in the examples below. I
merely wanted to keep it simple and draw your eye to the distinction between the examples.
In each example we’ll map the four corners of the square to see where the transformation takes
the unit-square. Those corners are simply (0, 0), (1, 0), (1, 1), (0, 1) as we traverse the square in a
counter-clockwise direction.

k 0
Example 2.3.4. Let A = for some k > 0. Define L(v) = Av for all v ∈ R2 . In particular
0 k
this means,
k 0 x kx
L(x, y) = A(x, y) = = .
0 k y ky
We find L(0, 0) = (0, 0), L(1, 0) = (k, 0), L(1, 1) = (k, k), L(0, 1) = (0, k). This mapping is called
a dilation.

−1 0
Example 2.3.5. Let A = . Define L(v) = Av for all v ∈ R2 . In particular this means,
0 −1

−1 0 x −x
L(x, y) = A(x, y) = = .
0 −1 y −y
We find L(0, 0) = (0, 0), L(1, 0) = (−1, 0), L(1, 1) = (−1, −1), L(0, 1) = (0, −1). This mapping is
called an inversion.

1 2
3 4

1 2 x x + 2y
L(x, y) = A(x, y) = = .
3 4 y 3x + 4y
We find L(0, 0) = (0, 0), L(1, 0) = (1, 3), L(1, 1) = (3, 7), L(0, 1) = (2, 4). This mapping shall
remain nameless, it is doubtless a combination of the other named mappings.

1 −1
√1
Example 2.3.7. Let A = . Define L(v) = Av for all v ∈ R2 . In particular this
1 2 1
means,

1 1 −1 x 1 x−y
L(x, y) = A(x, y) = √ = √ .
2 1 1 y 2 x+y
We find L(0, 0) = (0, 0), L(1, 0) = √1 (1, 1), L(1, 1) = √1 (0, 2), L(0, 1) = √1 (−1, 1). This mapping
2 2 2
is a rotation by π/4 radians.

1 −1
1 1

1 −1 x x−y
L(x, y) = A(x, y) = = .
1 1 y x+y
We find L(0, 0) = (0, 0), L(1, 0) = (1,√1), L(1, 1) = (0, 2), L(0, 1) = (−1, 1). This mapping is a
rotation followed by a dilation by k = 2.

cos(θ) − sin(θ)
Example 2.3.9. Let A = . Define L(v) = Av for all v ∈ R2 . In particular
sin(θ) cos(θ)
this means,

cos(θ) − sin(θ) x x cos(θ) − y sin(θ)
L(x, y) = A(x, y) = = .
sin(θ) cos(θ) y x sin(θ) + y cos(θ)
We find L(0, 0) = (0, 0), L(1, 0) = (cos(θ), sin(θ)), L(1, 1) = (cos(θ)−sin(θ), cos(θ)+sin(θ)) L(0, 1) =
(sin(θ), cos(θ)). This mapping is a rotation by θ in the counter-clockwise direction. Of course you
could have derived the matrix A from the picture below.

1 0
0 1

1 0 x x
L(x, y) = A(x, y) = = .
0 1 y y
We find L(0, 0) = (0, 0), L(1, 0) = (1, 0), L(1, 1) = (1, 1), L(0, 1) = (0, 1). This mapping is a
rotation by zero radians, or you could say it is a dilation by a factor of 1, ... usually we call this
the identity mapping because the image is identical to the preimage.

1 0
Example 2.3.11. Let A1 = . Define P1 (v) = A1 v for all v ∈ R2 . In particular this
0 0
means,
1 0 x x
P1 (x, y) = A1 (x, y) = = .
0 0 y 0
We find P1 (0, 0) = (0, 0), P1 (1, 0) = (1, 0), P1 (1, 1) = (1, 0), P1 (0, 1) = (0, 0). This mapping is a
projection
onto the first coordinate.
0 0
Let A2 = . Define L(v) = A2 v for all v ∈ R2 . In particular this means,
0 1

0 0 x 0
P2 (x, y) = A2 (x, y) = = .
0 1 y y
We find P2 (0, 0) = (0, 0), P2 (1, 0) = (0, 0), P2 (1, 1) = (0, 1), P2 (0, 1) = (0, 1). This mapping is
projection onto the second coordinate.
We can picture both of these mappings at once:

1 1
1 1

1 1 x x+y
L(x, y) = A(x, y) = = .
1 1 y x+y
We find L(0, 0) = (0, 0), L(1, 0) = (1, 1), L(1, 1) = (2, 2), L(0, 1) = (1, 1). This mapping is not a
projection, but it does collapse the square to a line-segment.
Remark 2.3.13.
The examples here have focused on linear transformations from R2 to R2 . It turns out that
higher dimensional mappings can largely be understood in terms of the geometric operations
we’ve seen in this section.
 
0 0
Example 2.3.14. Let A =  1 0 . Define L(v) = Av for all v ∈ R2 . In particular this means,
0 1
   
0 0 0
x
L(x, y) = A(x, y) = 1 0
  = x .

y
0 1 y
We find L(0, 0) = (0, 0, 0), L(1, 0) = (0, 1, 0), L(1, 1) = (0, 1, 1), L(0, 1) = (0, 0, 1). This mapping
moves the xy-plane to the yz-plane. In particular, the horizontal unit square gets mapped to vertical
unit square; L([0, 1] × [0, 1]) = {0} × [0, 1] × [0, 1]. This mapping certainly is not surjective because
no point with x 6= 0 is covered in the range.

1 1 0
1 1 1
means,  
x
1 1 0   x+y
L(x, y, z) = A(x, y, z) = y = .
1 1 1 x+y+z
z
Let’s study how L maps the unit cube. We have 23 = 8 corners on the unit cube,
L(0, 0, 0) = (0, 0), L(1, 0, 0) = (1, 1), L(1, 1, 0) = (2, 2), L(0, 1, 0) = (1, 1)
L(0, 0, 1) = (0, 1), L(1, 0, 1) = (1, 2), L(1, 1, 1) = (2, 3), L(0, 1, 1) = (1, 2).
This mapping squished the unit cube to a shape in the plane which contains the points (0, 0), (0, 1),
(1, 1), (1, 2), (2, 2), (2, 3). Face by face analysis of the mapping reveals the image is a parallelogram.
This mapping is certainly not injective since two different points get mapped to the same point. In
particular, I have color-coded the mapping of top and base faces as they map to line segments. The
vertical faces map to one of the two parallelograms that comprise the image.
I have used terms like ”vertical” or ”horizontal” in the standard manner we associate such terms
with three dimensional geometry. Visualization and terminology for higher-dimensional examples is
not as obvious. However, with a little imagination we can still draw pictures to capture important
aspects of mappings.

1 0 0 0
1 0 0 0
means,  
x
1 0 0 0  y  x
L(x, y, z, t) = A(x, y, z, t) =   = .
1 0 0 0  z  x
t
Let’s study how L maps the unit hypercube [0, 1]4 ⊂ R4 . We have 24 = 16 corners on the unit
hypercube, note L(1, a, b, c) = (1, 1) whereas L(0, a, b, c) = (0, 0) for all a, b, c ∈ [0, 1]. Therefore,
the unit hypercube is squished to a line-segment from (0, 0) to (1, 1). This mapping is neither
surjective nor injective. In the picture below the vertical axis represents the y, z, t-directions.
√
Example 2.3.17. Suppose f (t, s) = ( t, s2 + t) note that f (1, 1) = (1, 2) and f (4, 4) = (2, 20).
Note that (4, 4) = 4(1, 1) thus we should see f (4, 4) = f (4(1, 1)) = 4f (1, 1) but that fails to be true
so f is not a linear transformation.
Example 2.3.18. Let L(x, y) = x2 + y 2 define a mapping from R2 to R. This is not a linear
transformation since
L(c(x, y)) = L(cx, cy) = (cx)2 + (cy)2 = c2 (x2 + y 2 ) = c2 L(x, y).
We say L is a nonlinear transformation.
Example 2.3.19. Suppose L : R → R is defined by L(x) = mx + b for some constants m, b ∈ R.

Is this a linear transformation on R? Observe:
L(0) = m(0) + b = b
thus L is not a linear transformation if b 6= 0. On the other hand, if b = 0 then L is a linear

transformation.
A mapping on Rn which has the form T (x) = x + b is called a translation. If we have a mapping of
the form F (x) = Ax + b for some A ∈ R n×n and b ∈ R then we say F is an affine tranformation
on Rn . Technically, in general, the line y = mx + b is the graph of an affine function on R. I invite
the reader to prove that affine transformations also map line-segments to line-segments (or points).
2.3.2 standard matrices

Definition 2.3.20.
Let L : Rn → Rm be a linear transformation, the matrix A ∈ R m×n such that L(x) = Ax
for all x ∈ Rn is called the standard matrix of L. We denote this by [L] = A or more
compactly, [LA ] = A, we say that LA is the linear transformation induced by A. Moreover,
the components of the matrix A are found from Aji = (L(ei )))j .
Example 2.3.21. Given that L([x, y, z]T ) = [x+2y, 3y+4z, 5x+6z]T for [x, y, z]T ∈ R3 find the the
standard matrix of L. We wish to find a 3×3 matrix such that L(v) = Av for all v = [x, y, z]T ∈ R3 .
Write L(v) then collect terms with each coordinate in the domain,
         
x x + 2y 1 2 0
L   y  = 3y + 4z = x 0 + y 3 + z 4 
      
z 5x + 6z 5 0 6
It’s not hard to see that,

      
x 1 2 0 x 1 2 0
L  y  =  0 3 4   y  ⇒ A = [L] =  0 3 4 
z 5 0 6 z 5 0 6
Notice that the columns in A are just as you’d expect from the proof of theorem ??. [L] =
[L(e1 )|L(e2 )|L(e3 )]. In future examples I will exploit this observation to save writing.
Example 2.3.22. Suppose that L((t, x, y, z)) = (t + x + y + z, z − x, 0, 3t − z), find [L].

 
L(e1 ) = L((1, 0, 0, 0)) = (1, 0, 0, 3) 1 1 1 1
L(e2 ) = L((0, 1, 0, 0)) = (1, −1, 0, 0)  0 −1 0 1 
⇒ [L] =   0 0 0 0 .

L(e3 ) = L((0, 0, 1, 0)) = (1, 0, 0, 0)
L(e4 ) = L((0, 0, 0, 1)) = (1, 1, 0, −1) 3 0 0 −1
I invite the reader to check my answer here and see that L(v) = [L]v for all v ∈ R4 as claimed.
Proposition 2.3.23.
Suppose T : Rn → Rm and S : Rn → Rm are linear transformations then S + T and cT are

linear transformations and
(1.) [T + S] = [T ] + [S], (2.) [T − S] = [T ] − [S], (3.) [cT ] = c[T ].
In words, the standard matrix of the sum, difference or scalar multiple of linear transfor-
mations is the sum, difference or scalar multiple of the standard matrices of the respsective
linear transformations.
Example 2.3.24. Suppose T (x, y) = (x + y, x − y) and S(x, y) = (2x, 3y). It’s easy to see that

1 1 2 0 3 1
[T ] = and [S] = ⇒ [T + S] = [T ] + [S] =
1 −1 0 3 1 2

3 1 x 3x + y
Therefore, (T + S)(x, y) = = = (3x + y, x + 2y). Naturally this is the
1 2 y x + 2y
same formula that we would obtain through direct addition of the formulas of T and S.
Proposition 2.3.25.
L1 : Rm → Rn and L2 : Rn → Rp are linear transformations then L2 ◦ L1 : Rm → Rp is a

linear transformation with matrix [L2 ◦ L1 ] such that
n
X
[L2 ◦ L1 ]ij = [L2 ]ik [L1 ]kj
k=1
for all i = 1, 2, . . . p and j = 1, 2 . . . , m.
Example 2.3.26. Let T : R 2×1 →R 2×1 be defined by
T ([x, y]T ) = [x + y, 2x − y]T

for all [x, y]T ∈ R 2×1 . Also let S : R 2×1 →R 3×1 be defined by
S([x, y]T ) = [x, x, 3x + 4y]T
for all [x, y]T ∈ R 2×1 . We calculate the composite as follows:
(S ◦ T )([x, y]T ) = S(T ([x, y]T ))

= S([x + y, 2x − y]T )
= [x + y, x + y, 3(x + y) + 4(2x − y)]T
= [x + y, x + y, 11x − y]T
Notice we can write the formula above as a matrix multiplication,

   
1 1 1 1
x
(S ◦ T )([x, y]T ) =  1 1  ⇒ [S ◦ T ] =  1 1 .
y
11 −1 11 −1
Notice that the standard matrices of S and T are:

 
1 0
1 1
[S] = 1 0  [T ] =
2 −1
3 4
It’s easy to see that [S ◦ T ] = [S][T ] (as we should expect since these are linear operators)
Notice that T ◦ S is not even defined since the dimensions of the codomain of S do not match
the domain of T . Likewise, the matrix product [T ][S] is not defined since there is a dimension
mismatch; (2 × 2)(3 × 2) is not a well-defined product of matrices.
2.3.3 coordinates and isomorphism

Let V be a finite dimensional vector space with basis β = {v1 , v2 , . . . vn }. The coordinate map
Φβ : V → Rn is defined by
Φβ (x1 v1 + x2 v2 + · · · + xn vn ) = x1 e1 + x2 e2 + · · · + xn en
for all v = x1 v1 + x2 v2 + · · · + xn vn ∈ V . Sometimes we have to adjust the numbering a bit for

double-indices. For example:
Example 2.3.27. Let Φ : R m×n → Rmn be defined by

X
Φ( Aij Eij ) = (A11 , . . . , A1n , A21 , . . . , A2n , . . . , Am1 , . . . , Amn )
i,j
This map simply takes the entries in the matrix and strings them out to a vector of length mn.
Example 2.3.28. Let Ψ : C → R2 be defined by Ψ(x + iy) = (x, y). This is the coordinate map for
the basis {1, i}.
Matrix multiplication is for vectors in Rn . Direct matrix multiplication of an abstract vector makes
no sense (how would you multiply a polynomial and a matrix?), however, since we can use the
coordinate map to change the abstract vector to a vector in Rn . The diagram below illustrates the
idea for a linear transformation T from an abstract vector space V with basis β to another abstract
vector space W with basis β̄:
V
T / W
O
Φ−1
β
Φβ̄

Rn / Rn
L[T ]
β,β̄
Let’s walk through the formula [T ]β,β̄ x = Φβ̄ (T (Φ−1

β (x))): we begin on the RHS with a column
vector x, then Φβ lifts the column vector up to the abstract vector Φ−1
−1
β (x) in V . Next we operate
−1
by T which moves us over to the vector T (Φβ (x)) which is in W . Finally the coordinate map Φβ̄
pushes the abstract vector in W back to a column vector Φβ̄ (T (Φ−1β (x))) which is in R
m×1 . The
same journey is accomplished by just multiplying x by the m × n matrix [T ]β,β̄ .
Example 2.3.29. Let β = {1, x, x2 } be the basis for P2 and consider the derivative mapping
D : P2 → P2 . Find the matrix of D assuming that P2 has coordinates with respect to β on both
copies of P2 . Define and observe
Φ(xn ) = en+1 whereas Φ−1 (en ) = xn−1
for n = 0, 1, 2. Recall D(ax2 + bx + c) = 2ax + bx.
col1 ([D]β,β ) = Φβ (D(Φ−1

β (e1 ))) = Φβ (D(1)) = Φβ (0) = 0
col2 ([D]β,β ) = Φβ (D(Φ−1
β (e2 ))) = Φβ (D(x)) = Φβ (1) = e1
col3 ([D]β,β ) = Φβ (D(Φ−1 2
β (e3 ))) = Φβ (D(x )) = Φβ (2x) = 2e2
Therefore we find,  
0 1 0
[D]β,β =  0 0 2 .
0 0 0
Calculate D3 . Is this surprising?
A one-one correspondence is a map which is 1-1 and onto. If we can find such a mapping between
two sets then it shows those sets have the same cardnality. Cardnality is a crude idea of size, it
turns out that all finite dimensional vector spaces over R have the same cardnality. On the other
hand, not all vector spaces have the same dimension. Isomorphisms help us discern if two vector
spaces have the same dimension.
Definition 2.3.30.
Let V, W be vector spaces then Φ : V → W is an isomorphism if it is a 1-1 and onto
mapping which is also a linear transformation. If there is an isomorphism between vector
spaces V and W then we say those vector spaces are isomorphic and we denote this by
V u W.
Other authors sometimes denote isomorphism by equality. But, I’ll avoid that custom as I am
reserving = to denote set equality. Details of the first two examples below can be found in my
linear algebra notes.
Example 2.3.31. Let V = R3 and W = P2 . Define a mapping Φ : P2 → R3 by
Φ(ax2 + bx + c) = (a, b, c)
for all ax2 + bx + c ∈ P2 . As vector spaces, R3 and polynomials of upto quadratic order are the
same.
Example 2.3.32. Let S2 be the set of 2 × 2 symmetric matrices. Let Ψ : R3 → S2 be defined by

x y
Ψ(x, y, z) = .
y z
Example 2.3.33. Let L(Rn , Rm ) denote the set of all linear transformations from Rn to Rm .
L(Rn , Rm ) forms a vector space under function addition and scalar multiplication. There is a
natural isomorphism to m × n matrices. Define Ψ : L(Rn , Rm ) → R m×n by Ψ(T ) = [T ] for all
linear transformations T ∈ L(Rn , Rm ). In other words, linear transformations and matrices are
the same as vector spaces.
The quantification of ”same” is a large theme in modern mathematics. In fact, the term iso-
morphism as we use it here is more accurately phrased vector space isomorphism. The are other
kinds of isomorphisms which preserve other interesting stuctures like Group, Ring or Lie Algebra
isomorphism. But, I think we’ve said more than enough for this course.
Chapter 3
topology and limits
Topology is the study of continuity and open sets in general. A topology on a given set of points
is a collection of all sets which are defined to be open. Of course, the subject of topology is more
interesting than just picking which sets are open, that’s just the start of it. Topology also looks to
find criteria with which to classify spaces as topologically distinct. Two topologies are equivalent
if there exists a homeomorphism from one space to the other. A homeomorphism is bijection
which preserves the topological structure. Just as an isomorphism was a bijection which preserved
linear structure in the last Chapter. The abstract study of topology is a nontrivial task which we
do not undertake here, I merely make these comments for some context.
Our study of topology is very introductory. We merely need topology as a langauge to describe sets
in which are theorems and calculations typically hold. Moreover, the topologies we consider are all
metric topologies. This means the definition of our open sets is based on some natural concept
of distance. In the context of Rn it is simply Euclidean distance function. For a vector space with
norm the distance is naturally induced from the norm. Typically, the theorems which interest us
here can be shown true in the context of an abstract vector space with norm. Therefore, we give
some proofs in that context. However, to be kind to the student we begin with a discussion of
topology and limits in Euclidean space before we abstract to the arena of normed vector spaces.
Continuity and limits in Euclidean spaces present new difficulties in two or more dimensions. Limits
for finite dimensional vector spaces with a norm are not much different. In some sense, it’s just the
Euclidean case with some extra notational baggage. In particular, spaces of matrices provide an
interesting class of examples which take us a bit beyond the context of Euclidean space.
Occasionally we need to quote a nontrivial topological result. For example, the fact that the real
numbers are complete is an important base fact. However, discussing the justifcation of that fact
is outside the scope (and interest) of this course. That said, we conclude this chapter by collecting
a few basic theorems of a topological nature which we will not prove. I also recap all the interesting
proofs from calculus I which we at times use in our study of advanced calculus. These are included
for your edification and convenience, it is unlikely we devote much lecture time to the calculus I
proofs.
51
52 CHAPTER 3. TOPOLOGY AND LIMITS
3.1 elementary topology and limits

In this section we describe the metric topology for Rn . In the study of functions of one real variable
we often need to refer to open or closed intervals. The definition that follows generalizes those
concepts to n-dimensions.
Definition 3.1.1.
An open ball of radius centered at a ∈ Rn is the subset all points in Rn which are less
than units from a, we denote this open ball by
B (a) = {x ∈ Rn | ||x − a|| < }
The closed ball of radius centered at a ∈ Rn is likewise defined
B (a) = {x ∈ Rn | ||x − a|| ≤ }
Notice that in the n = 1 case we observe an open ball is an open interval: let a ∈ R,
B (a) = {x ∈ R | ||x − a|| < } = {x ∈ R | |x − a| < } = (a − , a + )
In the n = 2 case we observe that an open ball is an open disk: let (a, b) ∈ R2 ,
p
B ((a, b)) = (x, y) ∈ R2 | || (x, y) − (a, b) || < = (x, y) ∈ R2 | (x − a)2 + (y − b)2 <

For n = 3 an open-ball is a sphere without the outer shell. In contrast, a closed ball in n = 3 is a
solid sphere which includes the outer shell of the sphere.
Definition 3.1.2.
Let D ⊆ Rn . We say y ∈ D is an interior point of D iff there exists some open ball
centered at y which is completely contained in D. We say y ∈ Rn is a limit point of D iff
every open ball centered at y contains points in D − {y}. We say y ∈ Rn is a boundary
point of D iff every open ball centered at y contains points not in D and other points which
are in D − {y}. We say y ∈ D is an isolated point of D if there exist open balls about
y which do not contain other points in D. The set of all interior points of D is called the
interior of D. Likewise the set of all boundary points for D is denoted ∂D. The closure
of D is defined to be D = D ∪ {y ∈ Rn | y a limit point}
All the terms are aptly named. The term ”limit point” is given because those points are the ones for
which it is natural to define a limit. The picture below illustrates an interior point y1 , a boundary
point y2 and an isolated poin y3 . Also, the dotted-line around the hole indicates that edge is not
part of the set. However, the closure of D would be formed by connecting those dots to give D a
solid edge.
3.1. ELEMENTARY TOPOLOGY AND LIMITS 53
The dotted circle around y1 is meant to illustrate an open ball which is centered on y1 and it
contained within D, the existence of this open ball is what proves y1 is an interior point.
Definition 3.1.3.
Let A ⊆ Rn is an open set iff for each x ∈ A there exists > 0 such that x ∈ B (x) and
B (x) ⊆ A. Let B ⊆ Rn is an closed set iff its complement Rn − B = {x ∈ Rn | x ∈ / B}
is an open set.
Notice that R − [a, b] = (∞, a) ∪ (b, ∞). It is not hard to prove that open intervals are open hence
we find that a closed interval is a closed set. Likewise it is not hard to prove that open balls are
open sets and closed balls are closed sets. In fact, it can be shown that a closed set contains all its
limit points, that is A ⊆ Rn is closed iff A = A.
Limits of vector-valued functions of several real variables are defined below:
Definition 3.1.4.
Let f : U ⊆ Rn → V ⊆ Rm be a mapping. We say that f has limit b ∈ Rm at limit point a

of U iff for each > 0 there exists a δ > 0 such that x ∈ Rn with 0 < ||x − a|| < δ implies
||f (x) − b|| < . In such a case we can denote the above by stating that
lim f (x) = b.
x→a
In calculus I the limit of a function is defined in terms of deleted open intervals centered about the
limit point. We just defined the limit of a mapping in terms of deleted open balls centered at the
limit point. The term ”deleted” refers to the fact that we assume 0 < ||x − a|| which means we
do not consider x = a in the limiting process. In other words, the limit of a mapping considers
values close to the limit point but not necessarily the limit point itself. The case that the function
is defined at the limit point is special, when the limit and the mapping agree then we say the
mapping is continuous at that point.
1
Example 3.1.5. Let F : Rn − {a} → Rn be defined by F (x) = ||x−a|| (x − a). In this case, certainly
a is a limit point of F but geometrically it is clear that limx→a F (x) does not exist. Notice for n = 1,
the discontinuity of F at a can be understood by seeing that left and right limits exist, but are not
||x−a||
equal. On the other hand, G(x) = ||x−a|| (x − a) clearly has limx→a G(x) = 0 and we could classify
the discontinuity of G at x = a as removeable. Clearly G̃(x) = x − a is a continuous extension of
G to all of Rn
Multivariate limits are much trickier than single-variate limits because there are infinitely many
ways to approach a limit point. In the single-variate case we can only approach from the left
x → a− or from the right x → a+ . However, even in R2 there are infinitely many lines on which
we can approach a limit point. But, perhaps, even more insidiously, there are infinitely many
parabolas, cubics, exponentials etc... which intersect the limit point. It turns out that the method
of approach matters. It is possible for the function to limit to the same value by all linear paths
and yet parabolic paths yield different values.
( 2
2x y
4 2 (x, y) 6= (0, 0)
Example 3.1.6. Suppose f (x, y) = x +y . Notice that we can calculate the limit
0 (x, y) = (0, 0)
for (a, b) 6= (0, 0) with ease:
2a2 b
lim = 4 .
(x,y)→(a,b) a + b2
However, if we consider the limit at (0, 0) it is indeterminant since we have an expression of type
0/0. Other calculation is required. Consider the path ~r(t) = (t, mt) then clearly this is continuous
at t = 0 and ~r(0) = (0, 0); in-fact, this is just the parametric equation of a line y = mx. Consider,
for m 6= 0,
2mt4 2mt2 2m(0)
lim f (~r(t)) = lim 4 = lim = = 0.
t→0 t→0 t + m2 t2 t→0 t2 + m2 0 + m2
If ~r(t) = (t, 0) then for t 6= 0 we have f (~r(t)) = f (t, 0) = 0 thus the limit of the function restricted
to any linear path is just zero. The three pictures on the right illustrate how differing linear paths
yield the same limits. The red lines are the x, y axes.
What about parabolic paths? Those are easily constructed via ~r2 (t) = (t, kt2 ) again ~r2 (0) = (0, 0)
and limt→0 ~r2 (t) = (0, 0). Calculate, for k 6= 0,
2kt4 2k 2k
lim f (~r2 (t)) = lim 4 2 4
= lim 2
= .
t→0 t→0 t + k t t→0 1 + k 1 + k2
Clearly if we choose differing values for k we obtain different values for the limit hence the limit of
f does not exist as (x, y) → (0, 0). Here’s the graph of this function, maybe you can see the problem
at the origin. The red plane is vertical through the origin. The three pictures on the right illustrate
how differing parabolic paths yield differing limits. The red lines are the x, y axes.
See the limit of the blue path would be negative whereas the yellow path would give a positive limit.
Definition 3.1.7.
Let f : U ⊆ Rn → V ⊆ Rm be a function. If a ∈ U is a limit point of f then we say that f
is continuous at a iff
lim f (x) = f (a)
x→a
If a ∈ U is an isolated point then we also say that f is continous at a. The mapping f is

continous on S iff it is continous at each point in S. The mapping f is continuous iff
it is continuous on its domain.
Notice that in the m = n = 1 case we recover the definition of continuous functions from calc. I.
Another funny consequence of this definition is that sequences are by default continuous functions
since each point in their domain is isolated.
Proposition 3.1.8.
Let f : U ⊆ Rn → V ⊆ Rm be a mapping with component functions f1 , f2 , . . . , fm hence

f = (f1 , f2 , . . . , fm ). If a ∈ U is a limit point of f then
lim f (x) = b ⇔ lim fj (x) = bj for each j = 1, 2, . . . , m.

x→a x→a
.
Proof: (⇒) Suppose limx→a f (x) = b. Then for each > 0 choose δ > 0 such that 0 < ||x − a|| < δ
implies ||f (x) − b|| < . This choice of δ suffices for our purposes as:
v
um
q uX
|fj (x) − bj | = (fj (x) − bj )2 ≤ t (fj (x) − bj )2 = ||f (x) − b|| < .
j=1
Hence we have shown that limx→a fj (x) = bj for all j = 1, 2, . . . m.
(⇐) Suppose limx→a fj (x) = bj for all j = 1, 2, . . . m. Let > 0. Note that /m > 0 and therefore
p
by the given limits we can choose δj > 0 such that 0 < ||x − a|| < δ implies ||fj (x) − bj || < /m.
Choose δ = min{δ1 , δ2 , . . . , δm } clearly δ > 0. Moreoever, notice 0 < ||x − a|| < δ ≤ δj hence
requiring 0 < ||x − a|| < δ automatically induces 0 < ||x − a|| < δj for all j. Suppose that x ∈ Rn
and 0 < ||x − a|| < δ it follows that
v
m
um m p m
X uX X X
||f (x) − b|| = || (fj (x) − bj )ej || = t |fj (x) − bj |2 < ( /m)2 < /m = .
j=1 j=1 j=1 j=1
Therefore, limx→a f (x) = b and the proposition follows.
The beauty of the previous proposition is that it means we can analyze the limit of a vector-valued
function by analyzing the limits of the component functions. However, this does not remove the
fundamental difficulty of analyzing the multivariate limits of the component functions. It just
means we can tackel the problem one-component at time. This is a relief, it would be annoying if
the range was as intertwined as the domain in this analysis.
Proposition 3.1.9.
Suppose that f : U ⊆ Rn → V ⊆ Rm is a vector-valued function with component functions
f1 , f2 , . . . , fm . Let a ∈ U be a limit point of f then f is continous at a iff fj is continuous
at a for j = 1, 2, . . . , m. Moreover, f is continuous on S iff all the component functions of f
are continuous on S. Finally, a vector-valued function f is continous iff all of its component
functions are continuous. .
Proposition 3.1.10.
The projection functions are continuous. The identity mapping is continuous.
Proof: Let > 0 and choose δ = . If x ∈ Rn such that 0 < ||x − a|| < δ then it follows that
||x−a|| < .. Therefore, limx→a x = a which means that limx→a Id(x) = Id(a) for all a ∈ Rn . Hence
Id is continuous on Rn which means Id is continuous. Since the projection functions are component
functions of the identity mapping it follows that the projection functions are also continuous (using
the previous proposition).
Definition 3.1.11.
The sum and product are functions from R2 to R defined by
s(x, y) = x + y p(x, y) = xy
Proposition 3.1.12.
The sum and product functions are continuous.
Preparing for the proof: Let the limit point be (a, b). Consider what we wish to show: given a
point (x, y) such that 0 < ||(x, y) − (a, b)|| < δ we wish to show that
|s(x, y) − (a + b)| < or for the product |p(x, y) − (ab)| <
follow for appropriate choices of δ. Think about the sum for a moment,
|s(x, y) − (a + b)| = |x + y − a − b| ≤ |x − a| + |y − b|
I just used the triangle inequality for the absolute value of real numbers. We see that if we could
somehow get control of |x − a| and |y − b| then we’d be getting closer to the prize. We have control
of 0 < ||(x, y) − (a, b)|| < δ notice this reduces to
p
||(x − a, y − b)|| < δ ⇒ (x − a)2 + (y − b)2 < δ
it is clear that (x − a)2 < δ 2 since if it was otherwise the inequality above would be violated as
adding a nonegative quantity (y − b)2 only increases the radicand resulting in the squareroot to be
larger than δ. Hence we may assume (x − a)2 < δ 2 and since δ > 0 it follows |x − a| < δ . Likewise,
|y − b| < δ . Thus |s(x, y) − (a + b)| = |x + y − a − b| < |x − a| + |y − b| < 2δ. We see for the sum
proof we can choose δ = /2 and it will work out nicely.
Proof: Let > 0 and let (a, b) ∈ R2 . Choose δ = /2 and suppose (x, y) ∈ R2 such that
||(x, y) − (a, b)|| < δ. Observe that
||(x, y) − (a, b)|| < δ ⇒ ||(x − a, y − b)||2 < δ 2 ⇒ |x − a|2 + |y − b|2 < δ 2 .
It follows |x − a| < δ and |y − b| < δ. Thus
|s(x, y) − (a + b)| = |x + y − a − b| ≤ |x − a| + |y − b| < δ + δ = 2δ = .
Therefore, lim(x,y)→(a,b) s(x, y) = a + b. and it follows that the sum function if continuous at (a, b).
But, (a, b) is an arbitrary point thus s is continuous on R2 hence the sum function is continuous. .
Preparing for the proof of continuity of the product function: I’ll continue to use the same
notation as above. We need to study |p(x, y) − (ab)| = |xy − ab| < . Consider that
|xy − ab| = |xy − ya + ya − ab| = |y(x − a) + a(y − b)| ≤ |y||x − a| + |a||y − b|
We know that |x−a| < δ and |y−b| < δ. There is one less obvious factor to bound in the expression.
What should we do about |y|?. I leave it to the reader to show that:
|y − b| < δ ⇒ |y| < |b| + δ

Now put it all together and hopefully we’ll be able to ”solve” for .
|xy − ab| =≤ |y||x − a| + |a||y − b| < (|b| + δ)δ + |a|δ = δ 2 + δ(|a| + |b|) ” = ”
I put solve in quotes because we have considerably more freedom in our quest for finding δ. We
could just as well1 find δ which makes the ” = ” become an <. That said let’s pursue equality,
p
2 −|a| − |b| ± (|a| + |b|)2 + 4
δ + δ(|a| + |b|) − = 0 δ=
2
1
Minh Nguyen showed me a much shorter argument involving well-known inequalities for the geometric and
arithmetic means, this argument can be improved!
p p
Since , |a|, |b| > 0 it follows that (|a| + |b|)2 + 4 < (|a| + |b|)2 √
= |a|+|b| hence the (+) solution
−|a|−|b|+ (|a|+|b|)2 +4
to the quadratic equation yields a positive δ namely: δ = 2 .
Proof: Let > 0 and let (a, b) ∈ R2 . By the calculations that prepared for the proof we know that
the following quantity is positive, hence choose
p
−|a| − |b| + (|a| + |b|)2 + 4
δ= > 0.
2
Note that2 ,
|xy − ab| = |xy − ya + ya − ab| = |y(x − a) + a(y − b)| algebra

≤ |y||x − a| + |a||y − b| triangle inequality
< (|b| + δ)δ + |a|δ by the boxed lemmas
= δ 2 + δ(|a| + |b|) algebra
=
where we know that last step follows due to the steps leading to the boxed equation in the proof
preparation. Therefore, lim(x,y)→(a,b) p(x, y) = ab. and it follows that the product function if con-
tinuous at (a, b). But, (a, b) is an arbitrary point thus p is continuous on R2 hence the product
function is continuous. .
Proposition 3.1.13.
Let f : V ⊆ Rp → Rm and g : U ⊆ Rn → Rp be mappings. Suppose that

limx→a g(x) = b and suppose that f is continuous at b then

lim (f ◦ g)(x) = f lim g(x) .
x→a x→a
The proof is on pages 46-47 of C.H. Edwards Advanced Calculus text. I will provide a proof in the
setting of normed spaces later in this chapter. Notice that the proposition above immediately gives
us the important result below:
Proposition 3.1.14.
Let f and g be mappings such that f ◦ g is well-defined. The composite function f ◦ g is

continuous for points a ∈ dom(f ◦ g) such that the following two conditions hold:
1. g is continuous at a
2. f is continuous at g(a).
2
my notation is that when we stack inequalities the inequality in a particular line refers only to the immediate
vertical successor.
I make use of the earlier proposition that a mapping is continuous iff its component functions are
continuous throughout the examples that follow. For example, I know (Id, Id) is continuous since
Id was previously proved continuous.

Example 3.1.15. Note that if f = p ◦ (Id, Id) then f (x) = p ◦ (Id, Id) (x) = p (Id, Id)(x) =
p(x, x) = x2 . Therefore, the quadratic function f (x) = x2 is continuous on R as it is the composite
of continuous functions.
Example 3.1.16. Note that if f = p ◦ (p ◦ (Id, Id), Id) then f (x) = p(x2 , x) = x3 . Therefore, the
cubic function f (x) = x3 is continuous on R as it is the composite of continuous functions.
Example 3.1.17. The power function is inductively defined by x1 = x and xn = xxn−1 for all
n ∈ N. We can prove f (x) = xn is continous by induction on n. We proved the n = 1 case
previously. Assume inductively that f (x) = xn−1 is continuous. Notice that
xn = xxn−1 = xf (x) = p(x, f (x)) = (p ◦ (Id, f ))(x).
Therefore, using the induction hypothesis, we see that g(x) = xn is the composite of continuous
functions thus it is continuous. We conclude that f (x) = xn is continuous for all n ∈ N.
We can play similar games with the sum function to prove that sums of power functions are
continuous. In your homework you will prove constant functions are continuous. Putting all of
these things together gives us the well-known result that polynomials are continuous on R.
Proposition 3.1.18.
Let a be a limit point of mappings f, g : U ⊆ Rn → V ⊆ R and suppose c ∈ R. If
limx→a f (x) = b1 ∈ R and limx→a g(x) = b2 ∈ R then
1. limx→a (f (x) + g(x)) = limx→a f (x) + limx→a g(x).

2. limx→a (f (x)g(x)) = limx→a f (x) limx→a g(x) .
3. limx→a (cf (x)) = c limx→a f (x).
Moreover, if f, g are continuous then f + g, f g and cf are continuous.

Proof: Edwards proves (1.) carefully on pg. 48. I’ll do (2.) here: we are given that If limx→a f (x) =
b1 ∈ R and limx→a g(x) = b2 ∈ R thus by Proposition 3.1.8 we find limx→a (f, g)(x) = (b1 , b2 ).
Consider then,

limx→a (f (x)g(x)) = limx→a p(f, g) defn. of product function
= p limx→a (f, g) since p is continuous
= p(b1 , b2 ) by Proposition 3.1.8.
= b1 b2 definition of product function
= limx→a f (x) limx→a g(x) .
I leave it to the reader to show limx→a c = c and hence item (3.) follows from (2.). .
The proposition that follows does follow immediately from the proposition above, however I give a
proof that again illustrates the idea we used in the examples. Reinterpreting a given function as a
composite of more basic functions is a useful theoretical and calculational technique.
Proposition 3.1.19.
Assume f, g : U ⊆ Rn → V ⊆ R are continuous functions at a ∈ U and suppose c ∈ R.
1. f + g is continuous at a.
2. f g is continuous at a
3. cf is continuous at a.
Moreover, if f, g are continuous then f + g, f g and cf are continuous.
Proof: Observe that (f + g)(x) = (s ◦ (f, g))(x) and (f g)(x) = (p ◦ (f, g))(x). We’re given that
f, g are continuous at a and we know s, p are continuous on all of R2 thus the composite functions
s ◦ (f, g) and p ◦ (f, g) are continuous at a and the proof of items (1.) and (2.) is complete. To
prove (3.) I refer the reader to their homework where it was shown that h(x) = c for all x ∈ U is a
continuous function. We then find (3.) follows from (2.) by setting g = h (function multiplication
commutes for real-valued functions). .
We can use induction arguments to extend these results to arbitrarily many products and sums of
power functions.To prove continuity of algebraic functions we’d need to do some more work with
quotient and root functions. I’ll stop here for the moment, perhaps I’ll ask you to prove a few more
fundamentals from calculus I. I haven’t delved into the definition of exponential or log functions
not to mention sine or cosine3 We will assume that the basic functions of calculus are continuous
on the interior of their respective domains. Basically if the formula for a function can be evaluated
at the limit point then the function is continuous.
It’s not hard to see that the comments above extend to functions of several variables and map-
pings. If the formula for a mapping is comprised of finite sums and products of power func-
tions then we can prove such a mapping is continuous using the techniques developed in this
section. If we have a mapping with a more complicated formula built from elementary func-
tions then that mapping will be continuous provided its component functions have formulas which
are sensibly calculated at the limit point. In other words, if you are willing to believe me that
√
sin(x), cos(x), ex , ln(x), cosh(x), sinh(x), x, x1n , . . . are continuous on the interior of their domains
3
sine, cosine and exponentials are all nicely defined in terms of power series arguments, if time permits we may
sketch the development of these basic functions when we discuss series calculation
then it’s not hard to prove:

r
1
√
q
x x+ yz )
2 x
f (x, y, z) = sin(x) + e + cosh(x ) + y + e , cosh(xyz), xe
is a continuous mapping at points where the radicands of the square root functions are nonnegative.
It wouldn’t be very fun to write explicitly but it is clear that this mapping is the Cartesian product
of functions which are the sum, product and composite of continuous functions.
Definition 3.1.20.
A polynomial in n-variables has the form:
∞
X
f (x1 , x2 , . . . , xn ) = ci1 ,i2 ,...,in xi11 xi22 · · · xink
i1 ,i2 ,...,ik =0
where only finitely many coefficients ci1 ,i2 ,...,in 6= 0. We denote the set of multinomials in
n-variables as R[x1 , x2 , . . . , xn ].
Polynomials are R[x]. Polynomials in two variables are R[x, y], for example,
f (x, y) = ax + by deg(f ) = 1, linear function

f (x, y) = ax + by + c deg(f ) = 1, affine function
f (x, y) = ax2 + bxy + cy 2 deg(f)=2, quadratic form
f (x, y) = ax2 + bxy + cy 2 + dx + ey + g deg(f)=2
If all the terms in the polynomial have the same number of variables then it is said to be homo-
geneous. In the list above only the linear function and the quadratic form were homogeneous.
3.2 normed vector spaces

Much of linear algebra is done without any regard to the length of vectors. In fact, if one begins
linear algebra with both linear structure and the concept of vector length then many proofs can
be made rather intuitive. If you wish to read linear algebra from such a viewpoint then you might
find a copy of Gilbert Strang’s linear algebra text as it makes use of geometry from the outset.
Definition 3.2.1.
Suppose V is a vector space. If || · || : V × V → R is a function such that for all x, y ∈ V

and c ∈ R:
1. ||cx|| = |c| ||x||
2. ||x + y|| ≤ ||x|| + ||y|| (triangle inequality)
3. ||x|| ≥ 0
4. ||x|| = 0 iff x = 0
then we say (V, || · ||) is a normed vector space. When there is no danger of ambiguity we
also say that V is a normed vector space.
Notice that we did not assume V was finite-dimensional in the definition above. Our current focus
is on finite-dimensional cases.
√
Example 3.2.2. Rn can be given the Euclidean norm which is defined by ||x|| = x · x for each
x ∈ Rn .
Example 3.2.3. Rn can also be given the 1-norm which is defined by ||x||1 = |x1 | + |x2 | + · · · + |xn |
for each x ∈ Rn .
We use the Euclidean norm by default.
Example 3.2.4.
√ Consider C as a two dimensional real vector space. Let a + ib ∈ C and define
||a + ib|| = a + b2 . This is a norm for C.
2
Example 3.2.5. Let A ∈ R m×n . For each A = [Aij ] we define

v
um X
n
q uX
||A|| = A211 + A212 + · · · + A2mn = t A2ij .
i=1 j=1
This is the Frobenius norm for matrices.
Each of the norms above allows us to define a distance function and hence open sets and limits for
functions as we discuss next.
3.2. NORMED VECTOR SPACES 63
3.2.1 open balls, limits and sequences in normed vector spaces

In what follows here I give a brief overview of how we define open sets and limits in a normed
vector space. The answer is simply to replace Euclidean vector length with the appropriate norm.
Definition 3.2.6. Open sets and limit points for normed space V
An open ball in (V, || · ||V ) is defined
B (xo ) = {y ∈ V | ||y − xo ||V < }.
We define the deleted open ball by removing the center from the open ball
B (xo )o = B (xo ) − {xo } = {y ∈ V | 0 < ||y − xo ||V < }.
We say xo is a limit point of a function f iff there exists a deleted open ball which is
contained in the dom(f ). We say U ⊆ V is an open set iff for each u ∈ U there exists an
open ball B (u) ⊆ U .
Limits and continuous functions are also defined in the same way as in Rn .
Definition 3.2.7. Limits and continuity in normed spaces.
If f : V → W is a function from normed space (V, || · ||V ) to normed vector space (W, || · ||W )
then we say limx→xo f (x) = L iff for each > 0 there exists δ > 0 such that for all x ∈ V
subject to 0 < ||x − xo ||V < δ it follows ||f (x) − f (xo )||W < . If limx→xo f (x) = f (xo ) then
we say that f is a continuous function at xo .
Let (V, || · ||V ) be a normed vector space, a function from N to V is a called a sequence. Limits
of sequences play an important role in analysis in normed linear spaces. The real analysis course
makes great use of sequences to tackle questions which are more difficult with only − δ arguments.
In fact, we can reformulate limits in terms of sequences and subsequences. Perhaps one interesting
feature of abstract topological spaces is the appearance of spaces in which sequential convergence is
insufficient to capture the concept of limits. In general, one needs nets and filters. I digress. More
important to our context, the criteria of completeness. Let us settle a few definitions to make
the words meaningful.
Definition 3.2.8.
Suppose {an } is a sequence then we say limn→∞ an = L ∈ V iff for each > 0 there exists
M ∈ N such that ||an − L||V < for all n ∈ N with n > M . If limn→∞ an = L ∈ V then we
say {an } is a convergent sequence.
We spent some effort attempting to understand the definition above and its application to the
problem of infinite summations in calculus II. It is less likely you have thought much about the
following:
Definition 3.2.9.
We say {an } is a Cauchy sequence iff for each > 0 there exists M ∈ N such that
||am − an ||V < for all m, n ∈ N with m, n > M .
In other words, a sequence is Cauchy if the terms in the sequence get arbitarily close as we go
sufficiently far out in the list. Many concepts we cover in calculus II are made clear with proofs
built around the concept of a Cauchy sequence. The interesting thing about Cauchy is that for some
spaces of numbers we can have a sequence which converges but is not Cauchy. For example, if you
think about the rational numbers Q we can construct a sequence of truncated decimal expansions
of π:
{an } = {3, 3.1, 3.14, 3.141, 3.1415 . . . }
note that an ∈ Q for all n ∈ N and yet the an → π ∈
/ Q. When spaces are missing their limit points
they are in some sense incomplete.
Definition 3.2.10.
If every Cauchy sequence in a metric space converges to a point within the space then we
say the metric space is complete. If a normed vector space V is complete then we say V
is a Banach space.
A metric space need not be a vector space. In fact, we can take any open set of a normed vector
space and construct a metric space. Metric spaces require less structure.
Fortunately all the main examples of this course are built on the real numbers which are complete,
this induces completeness for C, Rn and R m×n . The proof that R, C, Rn and R m×n are Banach
spaces follow from arguments similar to those given in the example below.
Example 3.2.11. Claim: R complete implies R2 is complete.
Proof: suppose (xn , yn ) is a Cauchy sequence in R2 . Therefore, for each > 0 there exists N ∈ N
such that m, n ∈ N with N < m < n implies ||(xm , ym ) − (xn , yn )|| < . Consider that:
p
||(xm , ym ) − (xn , yn )|| = (xm − xn )2 + (ym − yn )2
p
Therefore, as |xm − xn | = (xm − xn )2 , it is clear that:
|xm − xn | ≤ ||(xm , ym ) − (xn , yn )||
But, this proves that {xn } is a Cauchy sequence of real numbers since for each > 0 we can choose
N > 0 such that N < m < n implies |xm − xn | < . The same holds true for the sequence {yn }.
By completeness of R we have xn → x and yn → y as n → ∞. We propose that (xn , yn ) → (x, y).
Let > 0 once more and choose Nx > 0 such that n > Nx implies |xn − x| < /2 and Ny > 0 such
that n > Ny implies |yn − y| < /2. Let N = max(Nx , Ny ) and suppose n > N :
||(xn , yn ) − (x, y)|| = ||xn − x, 0) + (0, yn − y)|| ≤ |xn − x| + |yn − y| < /2 + /2 = .
The key point here is that components of a Cauchy sequence form Cauchy sequences in R. That
will also be true for sets of matrices and complex numbers. Moreover, this proof is almost identical
to that which I gave to prove the limit of a sum was the sum of the limits. It is not an accident
that the structure of sequential limits and continuum limits are so closely paired. However, I leave
further analysis of that point to analysis. At the risk of killing that which is already dead4
Proposition 3.2.12.
Let V, W be normed vector spaces. Let a be a limit point of mappings f, g : U ⊆ V → W

and suppose c ∈ R. If limx→a f (x) = b1 ∈ W and limx→a g(x) = b2 ∈ W then
1. limx→a (f (x) + g(x)) = limx→a f (x) + limx→a g(x).
2. limx→a (cf (x)) = c limx→a f (x).
Moreover, if f, g are continuous then f + g and cf are continuous.

Proof: Let > 0 and suppose limx→a f (x) = b1 ∈ W and limx→a g(x) = b2 ∈ W . Choose δ1 , δ2 > 0
such that 0 < ||x−a|| < δ1 implies ||f (x)−b1 || < /2 and 0 < ||x−a|| < δ2 implies ||g(x)−b2 || < /2.
Choose δ = min(δ1 , δ2 ) and suppose 0 < ||x − a|| < δ ≤ δ1 , δ2 hence
||(f + g)(x) − (b1 + b2 )|| = ||f (x) − b1 + g(x) − b2 || ≤ ||f (x) − b1 || + ||g(x) − b2 || < /2 + /2 = .
Item (2.) follows. To prove (2.) note that if c = 0 the result is clearly true so suppose c 6= 0.
Suppose > 0 and choose δ > 0 such that ||f (x) − b1 || < /|c|. Note that if 0 < ||x − a|| < δ then
||(cf )(x) − cb1 || = ||c(f (x) − b1 )|| = |c|||f (x) − b1 || < |c|/|c| = .
The claims about continuity follow immediately from the limit properties .
Perhaps you recognize these arguments from calculus I. The logic used to prove the basic limit
theorems on R is essentially identical.
Proposition 3.2.13.
Suppose V1 , V2 , V3 are normed vector spaces with norms || · ||1 , || · ||2 , || · ||3 respective. Let
f : dom(f ) ⊆ V2 → V3 and g : dom(g) ⊆ V1 → V2 be mappings. Suppose that
limx→xo g(x) = yo and suppose that f is continuous at yo then

lim (f ◦ g)(x) = f lim g(x) .
x→xo x→xo
Proof: Let > 0 and choose β > 0 such that 0 < ||y − b||2 < β implies ||f (y) − f (yo )||3 < . We
can choose such a β since Since f is continuous at yo thus it follows that limy→yo f (y) = f (yo ).
Next choose δ > 0 such that 0 < ||x − xo ||1 < δ implies ||g(x) − yo ||2 < β. We can choose such
4
turns out this is quite interesting television and cinema for what it’s worth
a δ because we are given that limx→xo g(x) = yo . Suppose 0 < ||x − xo ||1 < δ and let y = g(x)
note ||g(x) − yo ||2 < β yields ||y − yo ||2 < β and consequently ||f (y) − f (yo )||3 < . Therefore, 0 <
||x−xo ||1 < δ implies ||f (g(x))−f (yo )||3 < . It follows that limx→xo (f (g(x)) = f (limx→xo g(x)).
The squeeze theorem relies heavily on the order properties of R. Generally a normed vector space
has no natural ordering. For example, is 1 > i or is 1 < i in C ? That said, we can state a squeeze
theorem for functions whose domain reside in a normed vector space. This is a generalization of
what we learned in calculus I. That said, the proof offered below is very similar to the typical proof
which is not given in calculus I5
Proposition 3.2.14. squeeze theorem.
Suppose f : dom(f ) ⊆ V → R, g : dom(g) ⊆ V → R, h : dom(h) ⊆ V → R where V is a

normed vector space with norm || · ||. Let f (x) ≤ g(x) ≤ h(x) for all x on some δ > 0 ball
of6 a ∈ V then we find that the limits at xo follow the same ordering,
lim f (x) ≤ lim g(x) ≤ lim h(x).

x→a x→a x→a
Moreover, if limx→xo f (x) = limx→xo h(x) = L ∈ R then limx→xo f (x) = L.

Proof: Suppose f (x) ≤ g(x) for all x ∈ Bδ1 (a)o for some δ1 > 0 and also suppose limx→a f (x) =
Lf ∈ R and limx→a g(x) = Lg ∈ R. We wish to prove that Lf ≤ Lg . Suppose otherwise towards a
contradiction. That is, suppose Lf > Lg . Note that limx→a [g(x) − f (x)] = Lg − Lf by the linearity
of the limit. It follows that for = 21 (Lf − Lg ) > 0 there exists δ2 > 0 such that x ∈ Bδ2 (a)o implies
|g(x) − f (x) − (Lg − Lf )| < = 12 (Lf − Lg ). Expanding this inequality we have
1 1
− (Lf − Lg ) < g(x) − f (x) − (Lg − Lf ) < (Lf − Lg )
2 2
adding Lg − Lf yields,
3 1
− (Lf − Lg ) < g(x) − f (x) < − (Lf − Lg ) < 0.
2 2
Thus, f (x) > g(x) for all x ∈ Bδ2 (a)o . But, f (x) ≤ g(x) for all x ∈ Bδ1 (a)o so we find a contradic-
tion for each x ∈ Bδ (a) where δ = min(δ1 , δ2 ). Hence Lf ≤ Lg . The same proof can be applied to
g and h thus the first part of the theorem follows.
Next, we suppose that limx→a f (x) = limx→a h(x) = L ∈ R and f (x) ≤ g(x) ≤ h(x) for all
x ∈ Bδ1 (a) for some δ1 > 0. We seek to show that limx→a f (x) = L. Let > 0 and choose δ2 > 0
such that |f (x) − L| < and |h(x) − L| < for all x ∈ Bδ (a)o . We are free to choose such a
δ2 > 0 because the limits of f and h are given at x = a. Choose δ = min(δ1 , δ2 ) and note that if
x ∈ Bδ (a)o then
f (x) ≤ g(x) ≤ h(x)
5
this is lifted word for word from my calculus I notes, however here the meaning of open ball is considerably more
general and the linearity of the limit which is referenced is the one proven earlier in this section
hence,
f (x) − L ≤ g(x) − L ≤ h(x) − L
but |f (x) − L| < and |h(x) − L| < imply − < f (x) − L and h(x) − L < thus
− < f (x) − L ≤ g(x) − L ≤ h(x) − L < .
Therefore, for each > 0 there exists δ > 0 such that x ∈ Bδ (a)o implies |g(x) − L| < so
limx→a g(x) = L.
Our typical use of the theorem above applies to equations of norms from a normed vector space.
The norm takes us from V to R so the theorem above is essential to analyze interesting limits. We
shall make use of it in the next chapter.
Proposition 3.2.15. norm is continuous with respect to itself.
Suppose V has norm || · || then f : V → R defined by f (x) = ||x|| is continuous.
Proof: Suppose a ∈ V and let > 0. Choose δ = and consider x ∈ V such that 0 < ||x − a|| < δ.
Observe ||x|| = ||x − a + a|| ≤ ||x − a|| + ||a|| = δ + ||a|| and hence
|f (x) − f (a)| = |||x|| − ||a||| < |δ + ||a|| − ||a||| = |δ| = .
Thus f (x) → f (a) as x → a and as a ∈ V was arbitrary the proposition follows .
Perhaps the most interesting feature of an abstract vector space is the loss of a canonical basis in
general. One might worry that this ambiguity spoils the result of Proposition 3.1.8, however, this
is not the case. Existence of limits for a set of component functions with respect to a particular
basis implies existence for all others7
Theorem 3.2.16.
Let V, W be normed vector spaces and suppose W has basis β = {wj }m

j=1 . Let a ∈ V then
m
X
lim F (x) = B = B j wj ⇔ lim Fj (x) = Bj for all j = 1, 2, . . . m.
x→a x→a
j=1
Proof: Throughout this proof we should keep in mind the basis vector wj 6= 0 implies ||wj || 6= 0
hence we may form quotients by the length of any basis vector if the need arises.
7
component functions with respect to different bases are related by multiplication by a constant nonsingular
matrix, so this result is not too surprising.
Pm
Suppose limx→a F (x) = B = j=1 Bj wj . Also, let Fj : V P → R denote the j-th component
m
function of F with respect to basis β; that is suppose F = j=1 Fj wj . Consider that by the
triangle inequality:

(Fj (x) − Bj )wj ≤ ||(F1 (x) − B1 )w1 + · · · + (Fm (x) − Bm )wm || = ||F (x) − B||
Therefore, using the above in the last step, observe:
||wj || 1 1
Fj (x) − Bj = |Fj (x) − Bj | ≤ (Fj (x) − Bj )wj ≤ ||F (x) − B||.
||wj || ||wj || ||wj ||
Therefore, |Fj (x) − Bj | ≤ ||w1j || ||F (x) − B||. Let > 0 and choose δ > 0 such that ||F (x) − B|| <
||wj ||. If x ∈ V such that 0 < ||x − a|| < δ then note,
1
|Fj (x) − Bj | = ≤ ||F (x) − B|| < .
||wj ||
Thus limx→a Fj (x) = Bj as x → a for all j ∈ Nn .

Conversely, suppose limx→a Fj (x) = Bj as x → a for all j ∈ Nn . Let > 0 and choose δj > 0 such
that 0 < ||x − a|| < δj implies ||Fj (x) − Bj || < ||wj||m . We are free to choose such δj by the given
limits as clearly ||wj||m > 0 for each j. Choose δ = min(δj | j ∈ Nm } and suppose x ∈ V such that
0 < ||x − a|| < δ. Using properties ||x + y|| ≤ ||x|| + ||y|| and ||cx|| = |c|||x|| multiple times yield:
m m m m
X X X X
||F (x) − B|| = || (Fj (x) − Bj )wj || ≤ |Fj (x) − Bj |||wj || < ||wj || = = .
||wj ||m m
j=1 j=1 j=1 j=1
Therefore, limx→a F (x) = B and this completes the proof .
Remark 3.2.17.
There are other topologies possible for Rn . For example, one can prove that
||v||1 = |v1 | + |v2 | + · · · + |vn |
gives a norm on Rn and the theorems we proved transfer over almost without change by
just trading || · || for || · ||1 . The unit ”ball” becomes a diamond for the 1-norm. There are
many other norms which can be constructed, infinitely many it turns out. However, it has
been shown that the topology of all these different norms is equivalent. This means that
open sets generated from different norms will be the same class of sets. For example, if
you can fit an open disk around every point in a set then it’s clear you can just as well fit
an open diamond and vice-versa. One of the things that makes infinite dimensional linear
algebra more fun is the fact that the topology generated by distinct norms need not be
equivalent for infinite dimensions. There is a difference between the open sets generated by
the Euclidean norm verses those generated by the 1-norm. Incidentally, my thesis work is
mostly built over the 1-norm. It makes the supernumbers happy.
3.3. INTUITIVE THEOREMS OF CALCULUS 69
3.3 intuitive theorems of calculus

In this section we review some of the important theorems of calculus which justify analytical
arguments via derivatives and continuity. Most of these you should have seen in Calculus I, but
perhaps the proof was not given. I give the proofs here for your mathematical edification. I
doubt we’ll cover them in lecture. That said, the end of this section presents the Extreme Value
Theorem’s generalization to multivariate, real-valued functions and we will probably discuss that.
One interesting thing to think about, which theorems in the next subsection generalize to functions
whose domain is a Banach space and range if R?
3.3.1 theorems for single-variable calculus

Proposition 3.3.1.
Let f be continuous at c such that f (c) 6= 0 then there exists δ > 0 such that either f (x) > 0
or f (x) < 0 for all x ∈ (c − δ, c + δ).
Proof: we are given that limx→c f (x) = f (a) 6= 0.
1.) Assume that f (a) > 0. Choose = f (a) 2 and use existence of the limit limx→c f (x) = f (a) to
select δ > 0 such that 0 < |x − c| < δ implies |f (x) − f (a)| < f (a) f (a) f (a)
2 hence − 2 < f (x) − f (a) < 2 .
Adding f (a) across the inequality yields 0 < f (a)
2 < f (x) <
3f (a)
2 .
2.) If f (a) < 0 then we can choose = − f (a)

2 > 0 and select δ > 0 such that 0 < |x − c| < δ implies
|f (x) − f (a)| < − 2 hence 2 < f (x) − f (a) < − f (a)
f (a) f (a)
2 . It follows that
3f (a)
2 < f (x) < f (a)
2 < 0.
The proposition follows.
Bolzano understood there was a gap in the arguments of the founders of calculus. Often, theorems
like those stated in this section would merely be claimed without proof. The work of Bolzano and
others like him ultimately gave rise to the careful rigorous study of the real numbers and more
generally the study of real analysis 8
Proposition 3.3.1 is clearly extended to sets which have boundary points. If we know a function
is continuous on [a, b) and f (a) 6= 0 then we can find δ > 0 such that f ([a, a + δ)) > 0. ( This is
needed in the proof below in the special case that c = a and a similar comment applies to c = b.)
8
the Bolzano-Weierstrauss theorem is one of the central theorems of real analysis, in 1817 Bolzano used it to
prove the IVT. It states every bounded sequence contains a convergent subsequence. Sequences can also be used to
formulate limits and continuity. Sequential convergence is dealt with properly in Math 431 at LU.
Theorem 3.3.2. Bolzano’s theorem

Let f be continuous on [a.b] such that f (a)f (b) < 0 then there exists c ∈ (a, b) such that
f (c) = 0.
Proof: suppose f (a) < f (b) then f (a)f (b) < 0 implies f (a) < 0 and f (b) > 0. We can use axiom
A11 for the heart of this proof. Our goal is to find a nonempty subset S ⊆ R which has an upper
bound. Axiom A11 will then provides the existence of the least upper bound. We should like to
construct a set which has the property desired in this theorem. Define S = {x ∈ [a, b] | f (x) < 0}.
Notice that a ∈ S since f (a) < 0 thus S 6= ∅. Moreover, it is clear that x ≤ b for all x ∈ S thus S
is bounded above. Axiom A11 states that there exists a least upper bound c ∈ S. To say c is the
least upper bound means that any other upperbound of S is larger than c.
We now seek to show that f (c) = 0. Consider that there exist three possibilities:
1. if f (c) < 0 then the continuous function f has f (c) 6= 0 so by prop. 3.3.1 there exists δ > 0
such that x ∈ (c − δ, c + δ) ∩ [a, b] implies f (x) < 0. However, this implies there is a value
x ∈ [c, c + δ) such that f (x) < 0 and x > c which means x is in S and is larger than the
upper bound c. Therefore, c is not an upper bound of S. Obviously this is a contradiction
therefore f (c) ≮ 0.
2. if f (c) > 0 then the continuous function f has f (c) 6= 0 so by prop. 3.3.1 there exists δ > 0
such that x ∈ (c − δ, c + δ) ∩ [a, b] implies f (x) > 0. However, this implies that all values
x ∈ (c − δ, c] have f (x) > 0 and thus x ∈
/ S which means x = c − δ/2 < c is an upper bound
of S which is smaller than the least upper bound c. Therefore, c is not the least upper bound
of S. Obviously this is a contradiction therefore f (c) ≯ 0.
3. if f (c) = 0 then no contradiction is found. The theorem follows.
My proof here essentially follows Apostol’s argument, however I suspect this argument in one form
or another can be found in many serious calculus texts. With Bolzano’s theorem settled we can
prove the IVT without much difficulty.
Proposition 3.3.3. Intermediate Value Theorem .
Suppose that f is continuous on an interval [a, b] with f (a) 6= f (b) and let N be a number
such that N is between f (a) and f (b) then there exists c ∈ (a, b) such that f (c) = N .
Proof: let N be as described above and define g(x) = f (x) − N . Note that g is clearly continuous.
Suppose that f (a) < f (b) then we must have f (a) < N < f (b) which gives f (a)−N ≤ 0 ≤ f (b)−N
hence g(a) < 0 < g(b). Applying Bolzano’s theorem to g gives c ∈ (a, b) such that g(c) = 0. But,
g(c) = f (c) − N = 0 therefore f (c) = N . If f (a) > f (b) then a similar argument applies. .
Proposition 3.3.4. Extreme value theorem.
Suppose that f is a function which is continuous on [a, b] then f attains its absolute maxi-
mum f (c) on [a, b] and its absolute minimum f (d) on [a, b] for some c, d ∈ [a, b].
It’s easy to see why the requirement of continuity is essential. If the function had a vertical
asymptote on [a, b] then the function gets arbitrarily large or negative so there is no biggest or
most negative value the function takes on the closed interval. Of course, if we had a vertical
asymptote then the function is not continuous at the asymptote. The proof of this theorem is
technical and beyond the scope of this course. See Apostol pages 150-151 for a nice proof.
Definition 3.3.5. critical numbers.
We say c ∈ R is a critical number of a function f if either f 0 (c) = 0 or f 0 (c) does not

exist. If c ∈ dom(f ) is a critical number then (c, f (c)) is a critical point of f .
Notice that a critical number need not be in the domain of a given function. For example, f (x) =
1/x has f 0 (x) = −1/x2 and thus c = 0 is a critical numbers as f 0 (0) does not exist in R. Clearly
0∈/ dom(f ) either. It is usually the case that a vertical asymptote of the function will likewise be
a vertical asymptote of the derivative function.
Proposition 3.3.6. Fermat’s theorem.
If f has a local extreme value of f (c) and f 0 (c) exists then f 0 (c) = 0.
Proof: suppose f (c) is a local maximum. Then there exists δ1 > 0 such that f (c + h) ≤ f (c) for
all h ∈ Bδ1 (0). Furthermore, since f 0 (c) ∈ R we have limh→0 f (c+h)−f
h
(c)
= f 0 (c) ∈ R. If h > 0 and
h ∈ Bδ1 (0) then f (c + h) − f (c) ≤ 0 hence,
f (c + h) − f (c)
≤0
h
f (c+h)−f (c)
Using the squeeze theorem we find f 0 (c) = limh→0+ h ≤ limh→0 (0) = 0. Likewise, if h < 0
and h ∈ Bδ1 (0) then f (c + h) − f (c) ≤ 0 hence,
f (c + h) − f (c)
≥0
h
Using the squeeze theorem we find f 0 (c) = limh→0− f (c+h)−f h

(c)
≥ limh→0 (0) = 0. Consequently,
f 0 (c) ≤ 0 and f 0 (c) ≥ 0 therefore f 0 (c) = 0. The proof in the case that f (c) is a local minimum is
similar.
Remember, if f 0 (c) does not exist then c is a critical point by definition. Therefore, if f (c) is a
local extrema then c must be a critical point for one of two general reasons:
1. f 0 (c) exists so Fermat’s theorem proves f 0 (c) = 0 so c is a critical point.

2. f 0 (c) does not exist so by definition c is a critical point.

Sometimes Fermat’s Theorem is simply stated as ”local extrema happen at critical points”.
The converse of this Theorem is not true. We can have a critical number c such that f (c) is not a
local maximum or minimum. For example, f (x) = x3 has critical number c = 0 yet f (0) = 0 which
is neither a local max. nor min. value of f (x) = x3 . It turns out that (0, 0) is actually an inflection
point as we’ll discuss soon. Another example of a critical point which yields something funny is a
constant function; if g(x) = k then g 0 (x) = 0 for each and every x ∈ dom(g). Technically, y = k is
both the minimum and maximum value of g. Constant functions are a sort of exceptional case in
this game we are playing.
Proposition 3.3.7. Rolle’s theorem.
Suppose that f is a function such that
1. f is continuous on [a, b],
2. f is differentiable on (a, b),
3. f (a) = f (b).
Then there exists c ∈ (a, b) such that f 0 (c) = 0.

Proof: If f (x) = k for all x ∈ [a, b] then every point is a critical point and the theorem is triv-
ially satisfied. Suppose f is not constant, apply the Extreme Value Theorem to show there exists
c, d ∈ [a, b] such that f (c) ≥ f (x) for all x ∈ [a, b] and f (d) ≤ f (x) for all x ∈ [a, b]. Since
f (xo ) 6= f (a) for at least one xo ∈ (a, b) it follows that f (xo ) > f (a) or f (xo ) < f (a). If xo ∈ (a, b)
and f (xo ) > f (a) then f (a) is not the absolute maximum therefore we deduce c ∈ (a, b) is the ab-
solute maximum. Likewise, if xo ∈ (a, b) and f (xo ) < f (a) then f (a) is not the absolute minimum
therefore we deduce d ∈ (a, b) is the absolute maximum. In all cases there is an absolute extremum
in the open set (a, b) hence there exists a critical point in the interior of the set. Moreover, since
f is differentiable on (a, b) it follows that either f 0 (c) = 0 or f 0 (d) = 0 and Rolle’s theorem follows.
Proposition 3.3.8. Mean Value Theorem (MVT).
Suppose that f is a function such that
1. f is continuous on [a, b],
2. f is differentiable on (a, b),

f (b)−f (a)
Then there exists c ∈ (a, b) such that f 0 (c) = b−a . That is, there exists c ∈ (a, b) such
that f (b) − f (a) = f 0 (c)(b − a).
Proof:(essentially borrowed from Stewart’s Calculus, this proof is common to a host of calculus
texts). The equation of the secant line to y = f (x) on the interval [a, b] is y = s(x) where s(x) is
defined via the standard formula for a line going from (a, f (a) to (b, f (b))
f (b) − f (a)
s(x) = f (a) + (x − a).
b−a
The Mean Value Theorem proposes that there is some point on the interval [a, b] such that the
slope of the tangent line is equal to the slope of the secant line y = s(x). Consider a new function
defined to be the difference of the secant line and the given function, call it h:
f (b) − f (a)
h(x) = f (x) − s(x) = f (x) − f (a) − (x − a)
b−a
Observe that h(a) = h(b) = 0 and h is clearly continuous on [a, b] because f is continuous and
besides that the function is constructed from a sum of a polynomial with f . Additionally, it is clear
that h is differentiable on (a, b) since polynomials are differentiable everywhere and f was assumed
to be differentiable on (a, b). Thus Rolle’s Theorem applies to h so there exists a c ∈ (a, b) such
that h0 (c) = 0 which yields
f (b) − f (a) f (b) − f (a)

h0 (c) = f 0 (c) − =0 =⇒ f 0 (c) = .
b−a b−a
Proposition 3.3.9. sign of the derivative function f 0 indicates strict increase or decrease of f .
Suppose that f is a function and J is a connected subset of dom(f )
1. if f 0 (x) > 0 for all x ∈ J then f is strictly increasing on J
2. if f 0 (x) < 0 for all x ∈ J then f is strictly decreasing on J.
Proof: suppose f 0 (x) > 0 for all x ∈ J. Let [a, b] ⊆ J and note f is continuous on [a, b] since it
is given to be differentiable on a superset of [a, b]. The MVT applied to f with respect to [a, b]
implies there exists c ∈ [a, b] such that
f (b) − f (a)
= f 0 (c).
b−a
Notice that f (b) − f (a) = (b − a)f 0 (c) but b − a > 0 and f 0 (c) > 0 hence f (b) − f (a) > 0. Therefore,
for each pair a, b ∈ J with a < b we find f (a) < f (b) which means f is strictly increasing on J.
Likewise, if f 0 (c) < 0 then almost the same argument applies to show a < b implies f (a) > f (b).
3.3.2 theorems for multivariate calculus

To begin, we need to define9 compact sets in Rn .
Definition 3.3.10.
We say C is a bounded subset of Rn if there exists an open ball which contains C. A
subset C ⊆ Rn is compact iff it is both closed and bounded.
In the one-dimensional case, clearly closed intervals [a, b] are compact sets. In higher dimensions
it’s not hard to see that spheres, cubes, disks, ellipses, ellipsoids and more generally finite shapes
with solid edges are compact. In particular, if there is some maximum distance between points
then it is clear we can place a slightly larger ball around the set so it’s bounded. Also, if the edges
are solid then the set should be closed.
Definition 3.3.11.
Let S ⊂ R. We say u ∈ R is an upper bound of S if u ≥ s for all s ∈ S. We say l ∈ R is an
lower bound of S if l ≤ s for all s ∈ S. Define the least upper bound to be sup(S) ∈ R
for which sup(S) ≤ u for all upper bounds u of S. Similarly, define the greatest lower
bound to be inf (S) ∈ R for which inf (S) ≥ l for all lower bounds l of S.
The completeness of the real numbers can be expressed by the statement that every set which is
bounded above has a suprememum sup(S). This is a clever way10 to capture the idea that every
Cauchy sequence of real numbers converges to a real number.
Finally we reach the punchline of this section. The following theorem generalizes the Extreme
Value Theorem for R to the n-dimensional case.
Theorem 3.3.12.
Let C be a compact subset of Rn and suppose f : C → R is a continuous function. Then
sup(f (C)), inf (f (C)) ∈ R. The values of f attain a minimum and maximum over C.
Even in the one-dimensional case the proof of this theorem is non-trivial. Take a look at Apostol
pages 150-151. Note that pages 49-55 of C.H. Edwards’ Advanced Calculus of Several Variables
discuss compactness in greater depth and list several theorems which are related to the Extreme
Value Theorem given here. I think my comments here suffice for our purposes.
9
the real definition of compactness says that C is compact if every open cover of C admits a finite subcover then
this definition is a theorem which can be proved in the context of Rn . See the Heine Borel theorem and related
discussion in a good real analysis text
10
some mathematicians construct the real numbers by simply adjoining these limit points to the rational numbers.
Chapter 4
differentiation
Our goal in this chapter is to describe differentiation for functions to and from normed linear spaces.
It turns out this is actually quite simple given the background of the preceding chapter. The dif-
ferential at a point is a linear transformation which best approximates the change in a function at
a particular point. We can quantify ”best” by a limiting process which is naturally defined in view
of the fact there is a norm on the spaces we consider.
The most important example is of course the case f : Rn → Rm . In this context it is natural to write
the differential as a matrix multiplication. The matrix of the differential is what Edwards calls the
derivative. Partial derivatives are also defined in terms of directional derivatives. The directional
derivative is sometimes defined where the differential fails to exist. We will discuss how the criteria
of continuous differentiability allows us to build the differential from the directional derivatives.
We’ll see how the Cauchy-Riemann equations of complex analysis are really just an algebraic result
if we already have the theorem for continuously differentiability. We will see how this general con-
cept of differentiation recovers all the derivatives you’ve seen previously in calculus and much more.
On the other hand, I postpone implicit differentiation for a future chapter where we have the
existence theorems for implicit and inverse functions. I also postpone discussion of the geometry
of the differential. In short, existence of the differential and the tangent space are essentially two
sides of the same problem. In fact, the approach of this chapter is radically different than my first
set of notes on advanced calculus. In those notes I followed Edwards a bit more and built up to
the definition of the differential on the basis of the directional derivative and geometry. I don’t
think students appreciate geometry or directional differentiation well enough to make that approach
successful. Consquently, I begin with the unjustified definition of the derivative and then spend the
rest of the chapter working out precise implications and examples that flow from the defintition.
I essentially ignore the question of motivating the defintiion here. If you want motivation, think
backward with this chapter or perhaps read Edwards or my old notes.
75
76 CHAPTER 4. DIFFERENTIATION
4.1 the Frechet differential

The definition1 below says that 4F = F (a + h) − F (a) ≈ dFa (h) when h is close to zero.
Definition 4.1.1.
Let (V, || · ||V ) and (W, || · ||W ) be normed vector spaces. Suppose that U is open and
F : U ⊆ V → W is a function the we say that F is differentiable at a ∈ U iff there exists
a linear mapping L : V → W such that

F (a + h) − F (a) − L(h)
lim = 0.
h→0 ||h||V
In such a case we call the linear mapping L the differential at a and we denote L = dFa .
In the case V = Rm and W = Rn are given the standard euclidean norms, the matrix of
the differential is called the derivative of F at a and we denote [dFa ] = F 0 (a) ∈ R m×n
which means that dFa (v) = F 0 (a)v for all v ∈ Rn .
Notice this definition gives an equation which implicitly defines dFa . For the moment the only way
we have to calculate dFa is educated guessing. We simply use brute-force calculation to suggest
a guess for L which forces the Frechet quotient to vanish. In the next section we’ll discover a
systematic calculational method for functions on euclidean spaces. The purpose of this section is
to understand the definition of the differential and to connect it to some previous concepts of basic
calculus. Efficient calculational schemes are discussed later in this chapter.
Example 4.1.2. Suppose T : V → W is a linear transformation of normed vector spaces V and W .

I propose L = T . In other words, I think we can show the best linear approximation to the change
in a linear function is simply the function itself. Clearly L is linear since T is linear. Consider the
difference quotient:
T (a + h) − T (a) − L(h) T (a) + T (h) − T (a) − T (h) 0

= = .
||h||V ||h||V ||h||V
Note h 6= 0 implies ||h||V 6= 0 by the definition of the norm. Hence the limit of the difference quotient
vanishes since it is identically zero for every nonzero value of h. We conclude that dTa = T .
Example 4.1.3. Let T : V → W where V and W are normed vector spaces and define T (v) = wo
for all v ∈ V . I claim the differential is the zero transformation. Linearity of L(v) = 0 is trivially
verified. Consider the difference quotient:
T (a + h) − T (a) − L(h) wo − wo − 0 0
= = .
||h||V ||h||V ||h||V
Using the arguments to the preceding example, we find dTa = 0.

1
Some authors might put a norm in the numerator of the quotient. That is an equivalent condition since a function
g : V → W has limh→0 g(h) = 0 iff limh→0 ||g(h)||W = 0
4.1. THE FRECHET DIFFERENTIAL 77
Typically the difference quotient is not identically zero. The pair of examples above are very special
cases. Now that we’ve seen a sample pair of abstract examples now we turn to the question of how
this general definition recovers the concept of differentiation we studied in introductory calculus.
Example 4.1.4. Suppose f : dom(f ) ⊆ R → R is differentiable at x. It follows that there exists a
linear function dfx : R → R such that2
f (x + h) − f (x) − dfx (h)
lim = 0.
h→0 |h|
Note that
f (x + h) − f (x) − dfx (h) f (x + h) − f (x) − dfx (h)
lim =0 ⇔ lim = 0.
h→0 |h| h→0± |h|
In the left limit h → 0− we have h < 0 hence |h| = −h. On the other hand, in the right limit h → 0+
we have h > 0 hence |h| = h. Thus, differentiability suggests that limh→0± f (x+h)−f±h
(x)−dfx (h)
= 0.
But we can pull the minus out of the left limit to obtain limh→0− f (x+h)−fh(x)−dfx (h) = 0. Therefore,
after an algebra step, we find:

f (x + h) − f (x) dfx (h)
lim − = 0.
h→0 h h
Linearity of dfx : R → R implies there exists m ∈ R1×1 = R such that dfx (h) = mh. Observe that
dfx (h) mh
lim= lim = m.
h→0 h h→0 h
It is a simple exercise to show that if lim(A − B) = 0 and lim(B) exists then lim(A) exists and
lim(A) = lim(B). Identify A = f (x+h)−f
h
(x)
and B = dfxh(h) . Therefore,
f (x + h) − f (x)
m = lim .
h→0 h
Consequently, we find the 1 × 1 matrix m of the differential is precisely f 0 (x) as we defined it via
a difference quotient in first semester calculus. In summary, we find dfx (h) = f 0 (x)h . In other
words, if a function is differentiable in the sense we defined at the beginning of this chapter then it
is differentiable in the terminology we used in calculus I. Moreover, the derivative at x is precisely
the matrix of the differential.
Remark 4.1.5.
Incidentally, I should mention that dfx is the differential of f at the point x. The differential
of f would be the mapping x 7→ dfx . Technically, the differential df is a function from R to
the set of linear transformations on R. You can contrast this view with that of first semester
calculus. There we say the mapping x 7→ f 0 (x) defines the derivative f 0 as a function from
R to R. This simplification in perspective is only possible because calculus in one-dimension
is so special. More on this later. This distinction is especially important to understand if
you begin to look at questions of higher derivatives.
2
√
unless we state otherwise, Rn is assumed to have the euclidean norm, in this case ||x||R = x2 = |x|.
Example 4.1.6. Suppose F : R2 → R3 is defined by F (x, y) = (xy, x2 , x + 3y) for all (x, y) ∈ R2 .
Consider the difference function 4F at (x, y):
4F = F ((x, y) + (h, k)) − F (x, y) = F (x + h, y + k) − F (x, y)
Calculate,
4F = (x + h)(y + k), (x + h)2 , x + h + 3(y + k) − xy, x2 , x + 3y

Simplify by cancelling terms which cancel with F (x, y):
4F = xk + hy + hk, 2xh + h2 , h + 3k)

Identify the linear part of 4F as a good candidate for the differential. I claim that:

L(h, k) = xk + hy, 2xh, h + 3k .
is the differential for f at (x,y). Observe first that we can write

 
y x
h
L(h, k) =  2x 0  .
k
1 3
therefore L : R2 → R3 is manifestly linear. Use the algebra above to simplify the difference quotient
below:
(hk, h2 , 0)

4F − L(h, k)
lim = lim
(h,k)→(0,0) ||(h, k)|| (h,k)→(0,0) ||(h, k)||
√
Note ||(h, k)|| = h2 + k 2 therefore we fact the task of showing that √h21+k2 (hk, h2 , 0) → (0, 0, 0)
as (h, k) → (0, 0). Notice that:
p
||(hk, h2 , 0)|| = |h| h2 + k 2
Therefore, as (h, k) → 0 we find

√ 1 2

h2 + k 2 (hk, h , 0) = |h| → 0.

However, if ||v|| → 0 it follows v → 0 so we derive the desired limit. Therefore,

 
y x
h
df(x,y) (h, k) =  2x 0  .
k
1 3
Computation of less trivial multivariate limits is an art we’d like to avoid if possible. It turns out
that we can actually avoid these calculations by computing partial derivatives. However, we still
need a certain multivariate limit to exist for the partial derivative functions so in some sense it’s
unavoidable. The limits are there whether we like to calculate them or not.
Calculation of the differential simplifies considerably when the domain is one-dimensional. We

already worked out the case of f : R → R in Example 4.1.4 and the following pair of examples work
out the concrete case of F : R → C and then the general case F : R → V for an arbitrary finite
dimensional normed linear space V .
Example 4.1.7. Suppose F (t) = U (t)+iV (t) for all t ∈ dom(f ) and both U and V are differentiable
functions on dom(F ). By the arguments given in Example 4.1.4 it suffices to find L : R → C such
that
F (t + h) − F (t) − L(h)
lim = 0.
h→0 h
I propose that on the basis of analogy to Example 4.1.4 we ought to have dFt (h) = (U 0 (t) + iV 0 (t))h.
Let L(h) = (U 0 (t) + iV 0 (t))h. Observe that, using properties of C:
L(h1 + ch2 ) = (U 0 (t) + iV 0 (t))(h1 + ch2 )

= (U 0 (t) + iV 0 (t))h1 + c(U 0 (t) + iV 0 (t))h2
= L(h1 ) + cL(h2 ).
for all h1 , h2 ∈ R and c ∈ R. Hence L : R → C is linear. Moreover,

F (t+h)−F (t)−L(h) 1 0 0
h = h U (t + h) + iV (t + h) − U (t) + iV (t) − (U (t) + iV (t))h

= h1 U (t + h) − U (t) − U 0 (t)h + i h1 V (t + h) − V (t) − V 0 (t)h
Consider the problem of calculating limh→0 F (t+h)−Fh (t)−L(h) . We observe that a complex function
converges to zero iff the real and imaginary parts of the function separately converge to zero (this
is covered by Theorem 3.2.16). By differentiability of U and V we find again using Example 4.1.4

1 0 1 0
lim U (t + h) − U (t) − U (t)h = 0 lim V (t + h) − V (t) − V (t)h = 0.
h→0 h h→0 h
Therefore, dFt (h) = (U 0 (t) + iV 0 (t))h. Note that the quantity U 0 (t) + iV 0 (t) is not a real matrix
in this case. To write the derivative in terms of a real matrix multiplication we need to construct
some further notation which makes use of the isomorphism between C and R2 . Actually, it’s pretty
easy if you agree that a + ib = (a, b) then dFt (h) = (U 0 (t), V 0 (t))h so the matrix of the differential
is (U 0 (t), V 0 (t)) ∈ R1×2 which makes since as F : C ≈ R2 → R.
Example 4.1.8. Suppose V is a normed vector space with basis β = {f1 , f2 , . . . , fn }. Futhermore,
let G : I ⊆ R → V be defined by
Xn
G(t) = Gi (t)fi
i=1
wherePGi : I → R is differentiable on I for Pi = 1, 2, . . . , n. Recall Theorem 3.2.16 revealed that
T = nj=1 Tj fj : R → V then limt→0 T (t) = nj=1 lj fj iff limt→0 Tj (t) = lj for all j = 1, 2, . . . , n.
In words, the limit of a vector-valued function can be parsed into a vector of limits. With this
in mind consider (again we can trade |h| for h as we explained in-depth in Example 4.1.4) the
Pn dGi
G(t+h)−G(t)−h i=1 f
dt i
difference quotient limh→0 h , factoring out the basis yields:
Pn n
+ h) − Gi (t) − h dG Gi (t + h) − Gi (t) − h dG
X
i=1 [Gi (t dt ]fi
i i
dt
lim = lim fi = 0
h→0 h h→0 h
i=1
where the zero above follows from the supposed differentiability of each component function. It
follows that:
n
X dGi
dGt (h) = h fi
dt
i=1
The example above encompasses a number of cases at once:

1. V = R, functions on R, f : R → R
2. V = Rn , space curves in R, ~r : R → Rn
3. V = C, complex-valued functions of a real variable, f = u + iv : R → C
4. V = R m×n , matrix-valued functions of a real variable, F : R → R m×n .
In short, when we differentiate a function which has a real domain then we can define the derivative
of such a function by component-wise differentiation. It gets more interesting when the domain has
several independent variables as Examples 4.1.6 and 4.1.9 illustrate.
Example 4.1.9. Suppose F : R n×n →R n×n is defined by F (X) = X 2 . Notice
4F = F (X + H) − F (X) = (X + H)(X + H) − X 2 = XH + HX + H 2
I propose that F is differentiable at X and L(H) = XH + HX. Let’s check linearity,
L(H1 + cH2 ) = X(H1 + cH2 ) + (H1 + cH2 )X = XH1 + H1 X + c(XH2 + H2 X)
Hence L : R n×n → R n×n is a linear transformation. By construction of L the linear terms in the
numerator cancel leaving just the quadratic term,
F (X + H) − F (X) − L(H) H2
lim = lim .
H→0 ||H|| H→0 ||H||
2
It suffices to show that limH→0 ||H ||
||H|| = 0 since lim(||g||) = 0 iff lim(g) = 0 in a normed vector
space. Fortunately the normed vector space R n×n is actually a Banach algebra. A vector space
with a multiplication operation is called an algebra. In the current context the multiplication is
simply matrix multiplication. A Banach algebra is a normed vector space with a multiplication that
satisfies ||XY || ≤ ||X|| ||Y ||. Thanks to this inequality3 we can calculate our limit via the squeeze
2 || ||H 2 ||
theorem. Observe 0 ≤ ||H ||H|| ≤ ||H||. As H → 0 it follows ||H|| → 0 hence limH→0 ||H|| = 0. We
find dFX (H) = XH + HX.
Generally constructing the matrix for a function f : V → W where V, W 6= R involves a fair

number of relatively ad-hoc conventions because the constructions necessarily involving choosing
coordinates. The situation is similar in linear algebra. Writing abstract linear transformations in
terms of matrix multiplication takes a little thinking. If you look back you’ll notice that I did not
bother to try to write a matrix for the differential in Examples 4.1.2 or 4.1.3.
Remark 4.1.10.
I have deliberately defined the derivative in slightly more generality than we need for this
course. It’s probably not much trouble to continue to develop the theory of differentiation
for a normed vector space, however I will for the most part stop here modulo an example
here or there.
If you understand many of the theorems that follow from here on out for Rn then it
is a simple matter to transfer arguments to the setting of a Banach space by using
an appropriate isomorphism. Traditionally this type of course only covers continuous
differentiability, inverse and implicit function theorems in the context of mappings from
Rn to Rm .
For the reader interested in generalizing these results to the context of an abstract normed
vector space feel free to discuss it with me sometime. Also, if you want to read a master
on these topics you could look at the text by Shlomo Sternberg on Advanced Calculus.
He develops many things for normed spaces. Or, take a look at Dieudonne’s Modern
Analysis which pays special attention to reaping infinite dimensional results from our finite-
dimensional arguments. Both of those texts would be good to read to follow-up my course
with something deeper.
3
it does take a bit of effort to prove this inequality holds for the matrix norm, I omit it since it would be distracting
here
4.2 partial derivatives and the Jacobian matrix

In the preceding section we calculated the differential at a point via educated guessing. We should
like to find better method to derive differentials. It turns out that we can systematically calculate
the differential from partial derivatives of the component functions. However, certain topological
conditions are required for us to properly paste together the partial derivatives of the component
functions. The story is given near the end of the Chapter in Section 4.7. There we describe how
the criteria of continuous differentiability achieves this goal. The difference between what we do
here and calculus III is that the derivative is no longer a number or a vector, instead as we shall see
soon, it is a matrix. Even this is a convenience, the better view is that the differential is primary.
Thus the better concept for the derivative is that of linear transformations. Dieudonne said it best:
this is the introduction to his chapter on differentiation in Modern Analysis Chapter VIII.
The subject matter of this Chapter is nothing else but the elementary theorems of
Calculus, which however are presented in a way which will probably be new to most
students. That presentation, which throughout adheres strictly to our general ”geomet-
ric” outlook on Analysis, aims at keeping as close as possible to the fundamental idea
of Calculus, namely the ”local” approximation of functions by linear functions. In
the classical teaching of Calculus, the idea is immediately obscured by the
accidental fact that, on a one-dimensional vector space, there is a one-to-
one correspondence between linear forms and numbers, and therefore the
derivative at a point is defined as a number instead of a linear form. This
slavish subservience to the shibboleth4 of numerical interpretation at any
cost becomes much worse when dealing with functions of several variables...
Dieudonne’s then spends the next half page continuing this thought with explicit examples of how
this custom of our calculus presentation injures the conceptual generalization. If you want to see
differentiation written for mathematicians, that is the place to look. He proves many results for
infinite dimensions because, well, why not?
4.2.1 directional derivatives

The directional derivative of a mapping F at a point a ∈ dom(F ) along v is defined to be the
derivative of the curve γ(t) = F (a + tv). In other words, the directional derivative gives you the
instantaneous vector-rate of change in the mapping F at the point a along v.
4
from wikipedia: is a word, sound, or custom that a person unfamiliar with its significance may not pronounce
or perform correctly relative to those who are familiar with it. It is used to identify foreigners or those who do not
belong to a particular class or group of people. It also refers to features of language, and particularly to a word or
phrase whose pronunciation identifies a speaker as belonging to a particular group.
4.2. PARTIAL DERIVATIVES AND THE JACOBIAN MATRIX 83
Definition 4.2.1.
Let F : dom(F ) ⊆ Rn → Rm and suppose the limit below exists for a ∈ dom(F ) and v ∈ Rn
then we define the directional derivative of F at a along v to be Dv F (a) ∈ Rm where
F (a + hv) − F (a)
Dv F (a) = lim
h→0 h
One great contrast we should pause to note is that the definition of the directional derivative
is explicit whereas the definition of the differential was implicit. The picture below might help
motivate the definition we just offered.
In the case that m = 1 then F : dom(F ) ⊆ Rn → R and the directional derivative gives the
instantaneous rate of change of the function F at the point a along v. You probably insisted that
||v|| = 1 in calculus III but we make no such demand here. We define the directional derivative for
mappings and vectors of non-unit length.
Proposition 4.2.2.
If F : U ⊆ Rn → Rm is differentiable at a ∈ U then the directional derivative Dv F (a) exists

for each v ∈ Rn and Dv F (a) = dFa (v).
Proof: Suppose a ∈ U such that dFa is well-defined then we are given that
F (a + h) − F (a) − dFa (h)
lim = 0.
h→0 ||h||
This is a limit in Rn , when it exists it follows that the limits that approach the origin along
particular paths also exist and are zero. In particular we can consider the path t 7→ tv for v 6= 0
and t > 0, we find
F (a + tv) − F (a) − dFa (tv) 1 F (a + tv) − F (a) − tdFa (v)

lim = lim = 0.
tv→0, t>0 ||tv|| ||v|| t→0+ |t|
Hence, as |t| = t for t > 0 we find
F (a + tv) − F (a) tdFa (v)

lim = lim = dFa (v).
t→0+ t t→0 t
Likewise we can consider the path t 7→ tv for v 6= 0 and t < 0
F (a + tv) − F (a) − dFa (tv) 1 F (a + tv) − F (a) − tdFa (v)

lim = lim = 0.
tv→0, t<0 ||tv|| ||v|| t→0 − |t|
Note |t| = −t thus the limit above yields
F (a + tv) − F (a) tdFa (v) F (a + tv) − F (a)

lim = lim ⇒ lim = dFa (v).
t→0− −t t→0− −t t→0− t
Therefore,
F (a + tv) − F (a)
lim = dFa (v)
t→0 t
and we conclude that Dv F (a) = dFa (v) for all v ∈ Rn since the v = 0 case follows trivially.
Let’s think about the problem we face. We want to find a nice formula for the differential. We
now know that if it exists then the directional derivatives allow us to calculate the values of the
differential in particular directions. The natural thing to do is to calculate the standard matrix
for the differential using the preceding proposition. Recall that if L : Rn → Rm then the standard
matrix was simply
[L] = [L(e1 )|L(e2 )| · · · |L(en )]
and thus the action of L is expressed nicely as a matrix multiplication; L(v) = [L]v. Similarly,
dfa : Rn → Rm is linear transformation and thus dfa (v) = [dfa ]v where
[dfa ] = [dfa (e1 )|dfa (e2 )| · · · |dfa (en )].
Moreover, by the preceding proposition we can calculate dfa (ej ) = Dej f (a) for j = 1, 2, . . . , n.
Clearly the directional derivatives in the coordinate directions are of great importance. For this
reason we make the following definition:
Definition 4.2.3. Partial derivatives are directional derivatives in coordinate directions.

Suppose that F : U ⊆ Rn → Rm is a mapping the we say that F is has partial derivative
∂F
∂xi (a) at a ∈ U iff the directional derivative in the ei direction exists at a. In this case we
denote,
∂F
(a) = Dei F (a).
∂xi
∂F
Also, the notation Dei F (a) = Di F (a) or ∂i F = ∂x i
is convenient. We construct the partial
n m
derivative function ∂i F : V ⊆ R → R as the function defined pointwise for each v ∈ V
where ∂i F (v) exists.
Let’s expand this definition a bit. Note that if F = (F1 , F2 , . . . , Fm ) then

F (a + hei ) − F (a) Fj (a + hei ) − Fj (a)
Dei F (a) = lim ⇒ [Dei F (a)] · ej = lim
h→0 h h→0 h
for each j = 1, 2, . . . m. But then the limit of the component function Fj is precisely the directional
derivative at a along ei hence we find the result
∂F ∂Fj
· ej = in other words, ∂i F = (∂i F1 , ∂i F2 , . . . , ∂i Fm ).
∂xi ∂xi
In the particular case f : R2 → R the partial derivatives with respect to x and y at (xo , yo ) are
related to the graph z = f (x, y) as illustrated below:
Similar pictures can be imagined for partial derivatives of more variables, even for vector-valued
maps, but direct visualization is not possible (at least for me).
The proposition below shows how the differential of a m-vector-valued function of n-real variables
is connected to a matrix of partial derivatives.
Proposition 4.2.4.
If F : U ⊆ Rn → Rm is differentiable at a ∈ U then the differential dFa has derivative
matrix F 0 (a) and it has components which are expressed in terms of partial derivatives of
the component functions:
[dFa ]ij = ∂j Fi
for 1 ≤ i ≤ m and 1 ≤ j ≤ n.
Perhaps it is helpful to expand the derivative matrix explicitly for future reference:
 
∂1 F1 (a) ∂2 F1 (a) · · · ∂n F1 (a)
 ∂1 F2 (a) ∂2 F2 (a) · · · ∂n F2 (a) 
F 0 (a) = 
 
.. .. .. .. 
 . . . . 
∂1 Fm (a) ∂2 Fm (a) · · · ∂n Fm (a)
Let’s write the operation of the differential for a differentiable mapping at some point a ∈ R in
terms of the explicit matrix multiplication by F 0 (a). Let v = (v1 , v2 , . . . vn ) ∈ Rn ,
  
∂1 F1 (a) ∂2 F1 (a) · · · ∂n F1 (a) v1
 ∂1 F2 (a) ∂2 F2 (a) · · · ∂n F2 (a)   v2 
dFa (v) = F 0 (a)v = 
  
.. .. .. ..   .. 
 . . . .  . 
∂1 Fm (a) ∂2 Fm (a) · · · ∂n Fm (a) vn
You may recall the notation from calculus III at this point, omitting the a-dependence,
T
∇Fj = grad(Fj ) = ∂1 Fj , ∂2 Fj , · · · , ∂n Fj
So if the derivative exists we can write it in terms of a stack of gradient vectors of the component
functions: (I used a transpose to write the stack side-ways),
T
F 0 = ∇F1 |∇F2 | · · · |∇Fm

Finally, just to collect everything together,
(∇F1 )T
   
∂1 F1 ∂2 F1 · · · ∂n F1
 ∂1 F2 ∂2 F2 · · · ∂n F2   (∇F2 )T 
F0 =  . = ∂ F | ∂ F | · · · | ∂ F =
   
.. .. .. 1 2 n ..
 ..
 
. . .   . 
∂1 Fm ∂2 Fm · · · ∂n Fm (∇Fm )T
Example 4.2.5. Recall that in Example 4.1.6 we showed that F : R2 → R3 defined by F (x, y) =
(xy, x2 , x + 3y) for all (x, y) ∈ R2 was differentiable. In fact we calculated that
 
y x
h
dF(x,y) (h, k) =  2x 0  .
k
1 3
If you recall from calculus III the mechanics of partial differentiation it’s simple to see that
 
y
∂F ∂
= (xy, x2 , x + 3y) = (y, 2x, 1) =  2x 
∂x ∂x
1
 
x
∂F ∂ 2
= (xy, x , x + 3y) = (x, 0, 3) = 0 

∂y ∂y
3
Thus [dF ] = [∂x F |∂y F ] (as we expect given the derivations in this section!)
Directional derivatives and partial derivatives are of secondary importance in this course. They are
merely the substructure of what is truly of interest: the differential. That said, it is useful to know
how to construct directional derivatives via partial derivative formulas. In fact, in careless calculus
texts it sometimes presented as the definition.
Proposition 4.2.6.
If F : U ⊆ Rn → Rm is differentiable at a ∈ U then the directional derivative Dv F (a) can
be expressed as a sum of partial derivative maps for each v =< v1 , v2 , . . . , vn >∈ Rn :
n
X
Dv F (a) = vj ∂j F (a)
j=1
Proof: since F is differentiable at a the differential dFa exists and Dv F (a) = dFa (v) for all v ∈ Rn .
Use linearity of the differential to calculate that
Dv F (a) = dFa (v1 e1 + · · · + vn en ) = v1 dFa (e1 ) + · · · + vn dFa (en ).
Note dFa (ej ) = Dej F (a) = ∂j F (a) and the prop. follows.
Example 4.2.7. Suppose f : R3 → R then ∇f = [∂x f, ∂y f, ∂z f ]T and we can write the directional
derivative in terms of
Dv f = [∂x f, ∂y f, ∂z f ]T v = ∇f · v
if we insist that ||v|| = 1 then we recover the standard directional derivative we discuss in calculus
III. Naturally the ||∇f (a)|| yields the maximum value for the directional derivative at a if we limit
the inputs to vectors of unit-length. If we did not limit the vectors to unit length then the directional
derivative at a can become arbitrarily large as Dv f (a) is proportional to the magnitude of v. Since
our primary motivation in calculus III was describing rates of change along certain directions for
some multivariate function it made sense to specialize the directional derivative to vectors of unit-
length. The definition used in these notes better serves the theoretical discussion.5
In Section 4.4 we give many explicit examples.
5
If you read my calculus III notes you’ll find a derivation of how the directional derivative in Stewart’s calculus
arises from the general definition of the derivative as a linear mapping. Look up page 305g.
4.3 derivatives of sum and scalar product

Suppose F1 : U ⊆ Rn → Rm and F2 : U ⊆ Rn → Rm . Furthermore, suppose both of these are
differentiable at a ∈ Rn . It follows that (dF1 )a = L1 and (dF2 )a = L2 are linear operators from Rn
to Rm which approximate the change in F1 and F2 near a, in particular,
F1 (a + h) − F1 (a) − L1 (h) F2 (a + h) − F2 (a) − L2 (h)

lim =0 lim =0
h→0 ||h|| h→0 ||h||
To prove that H = F1 + F2 is differentiable at a ∈ Rn we need to find a differential at a for H.

Naturally, we expect dHa = d(F1 + F2 )a = (dF1 )a + (dF2 )a . Let L = (dF1 )a + (dF2 )a and consider,
H(a+h)−H(a)−L(h) F1 (a+h)+F2 (a+h)−F1 (a)−F2 (a)−L1 (h)−L2 (h)
lim ||h|| = lim ||h||
h→0 h→0
F1 (a+h)−F1 (a)−L1 (h)
= lim ||h|| + lim F2 (a+h)−F 2 (a)−L2 (h)
||h||
h→0 h→0
=0+0
=0
Note that breaking up the limit was legal because we knew the subsequent limits existed and
were zero by the assumption of differentiability of F1 and F2 at a. Finally, since L = L1 + L2 we
know L is a linear transformation since the sum of linear transformations is a linear transformation.
Moreover, the matrix of L is the sum of the matrices for L1 and L2 . Let c ∈ R and suppose G = cF1
then we can also show that dGa = d(cF1 )a = c(dF1 )a , the calculation is very similar except we just
pull the constant c out of the limit. I’ll let you write it out. Collecting our observations:
Proposition 4.3.1.
Suppose F1 : U ⊆ Rn → Rm and F2 : U ⊆ Rn → Rm are differentiable at a ∈ U then

F1 + F2 is differentiable at a and
d(F1 + F2 )a = (dF1 )a + (dF2 )a or (F1 + F2 )0 (a) = F10 (a) + F20 (a)
Likewise, if c ∈ R then
d(cF1 )a = c(dF1 )a or (cF1 )0 (a) = c(F10 (a))
These results suggest that the differential of a function is a new object which has a vector space
structure. Of course, from a calculational view, these also say that the Jacobian matrix of a sum
or scalar product is simply the sum or scalar product of the Jacobian matrices.
4.4. A GALLERY OF EXPLICIT DERIVATIVES 89
4.4 a gallery of explicit derivatives

Our goal here is simply to exhibit the Jacobian matrix and partial derivatives for a few mappings.
At the base of all these calculations is the observation that partial differentiation is just ordinary
differentiation where we treat all the independent variable not being differentiated as constants.
The criteria of independence is important. We’ll study the case where variables are not independent
in a later section (see implicit differentiation).
Example 4.4.1. Let f (t) = (t, t2 , t3 ) then f 0 (t) = (1, 2t, 3t2 ). In this case we have
 
1
f 0 (t) = [dft ] =  2t 
3t2
Example 4.4.2. Let f (~x, ~y ) = ~x · ~y be a mapping from R3 × R3 → R. I’ll denote the coordinates
in the domain by (x1 , x2 , x3 , y1 , y2 , y3 ) thus f (~x, ~y ) = x1 y1 + x2 y2 + x3 y3 . Calculate,
[df(~x,~y) ] = ∇f (~x, ~y )T = [y1 , y2 , y3 , x1 , x2 , x3 ]
Example 4.4.3. Let f (~x, ~y ) = ~x · ~y be a mapping fromP Rn × Rn → R. I’ll denote the coordinates
in the domain by (x1 , . . . , xn , y1 , . . . , yn ) thus f (~x, ~y ) = ni=1 xi yi . Calculate,
n
X n
X n
X
∂ ∂xi
xj xi yi = xj yi = δij yi = yj
i=1 i=1 i=1
Likewise,
n
X n
X n
X
∂
yj x i yi = xi ∂y i
yj = xi δij = xj
i=1 i=1 i=1
Therefore, noting that ∇f = (∂x1 f, . . . , ∂xn f, ∂y1 f, . . . , ∂yn f ),
[df(~x,~y) ]T = (∇f )(~x, ~y ) = ~y × ~x = (y1 , . . . , yn , x1 , . . . , xn )
Example 4.4.4. Suppose F (x, y, z) = (xyz, y, z) we calculate,

∂F ∂F ∂F
∂x = (yz, 0, 0) ∂y = (xz, 1, 0) ∂z = (xy, 0, 1)
Remember these are actually column vectors in my sneaky notation; (v1 , . . . , vn ) = [v1 , . . . , vn ]T .
This means the derivative or Jacobian matrix of F at (x, y, z) is
 
yz xz xy
F 0 (x, y, z) = [dF(x,y,z) ] =  0 1 0 
0 0 1
Example 4.4.5. Suppose F (x, y, z) = (x2 + z 2 , yz) we calculate,

∂F ∂F ∂F
∂x = (2x, 0) ∂y = (0, z) ∂z = (2z, y)
The derivative is a 2 × 3 matrix in this example,

0 2x 0 2z
F (x, y, z) = [dF(x,y,z) ] =
0 z y
Example 4.4.6. Suppose F (x, y) = (x2 + y 2 , xy, x + y) we calculate,

∂F ∂F
∂x = (2x, y, 1) ∂y = (2y, x, 1)

 
2x 2y
F 0 (x, y) = [dF(x,y) ] =  y x 
1 1
Example 4.4.7. Suppose P (x, v, m) = (Po , P1 ) = ( 12 mv 2 + 12 kx2 , mv) for some constant k. Let’s
calculate the derivative via gradients this time,
∇Po = (∂Po /∂x, ∂Po /∂v, ∂Po /∂m) = (kx, mv, 12 v 2 )
∇P1 = (∂P1 /∂x, ∂P1 /∂v, ∂P1 /∂m) = (0, m, v)

Therefore,
1 2

0 kx mv 2v
P (x, v, m) =
0 m v
Example 4.4.8. Let F (r, θ) = (r cos θ, r sin θ). We calculate,
∂r F = (cos θ, sin θ) and ∂θ F = (−r sin θ, r cos θ)
Hence,
cos θ −r sin θ
F 0 (r, θ) =
sin θ r cos θ
p
Example 4.4.9. Let G(x, y) = ( x2 + y 2 , tan−1 (y/x)). We calculate,
∂x G = √ x
, −y ∂y G = √ 2y 2 , x2 +y
x

2 x2 +y 2
and 2
x2 +y x +y
Hence,
"
√ x √ y #
y
x
p
0 x2 +y 2 x2 +y 2

G (x, y) = = r r using r = x2 + y 2
−y x −y x
x2 +y 2 x2 +y 2 r2 r2
4.4. A GALLERY OF EXPLICIT DERIVATIVES 91
p
Example 4.4.10. Let F (x, y) = (x, y,R2 − x2 − y 2 ) for a constant R. We calculate,

−y
p
−x
∇ R − x − y = √ 2 2 2, √ 2 2 2
2 2 2
R −x −y R −x −y
Also, ∇x = (1, 0) and ∇y = (0, 1) thus

 
1 0
F 0 (x, y) =  0 1
 
√ −y

√ −x
R2 −x2 −y 2 R2 −x2 −y 2
p
Example 4.4.11. Let F (x, y, z) = (x, y, z, R2 − x2 − y 2 − z 2 ) for a constant R. We calculate,

−y
p
−x −z
∇ R −x −y −z = √
2 2 2 2 , √ , √
R2 −x2 −y 2 −z 2 R2 −x2 −y 2 −z 2 R2 −x2 −y 2 −z 2
Also, ∇x = (1, 0, 0), ∇y = (0, 1, 0) and ∇z = (0, 0, 1) thus

 
1 0 0
 0 1 0 
F 0 (x, y, z) = 
 
0 0 1 
−y
 
√ −x √ √ −z
R2 −x2 −y 2 −z 2 R2 −x2 −y 2 −z 2 R2 −x2 −y 2 −z 2
Example 4.4.12. Let f (x, y, z) = (x + y, y + z, x + z, xyz). You can calculate,

 
1 1 0
 0 1 1 
[df(x,y,z) ] = 
 1 0

1 
yz xz xy
Example 4.4.13. Let f (x, y, z) = xyz. You can calculate,

[df(x,y,z) ] = yz xz xy
Example 4.4.14. Let f (x, y, z) = (xyz, 1 − x − y). You can calculate,

yz xz xy
[df(x,y,z) ] =
−1 −1 0
Example 4.4.15. Let f : R3 × R3 be defined by f (x) = x × v for a fixed vector v 6= 0. We denote

x = (x1 , x2 , x3 ) and calculate,
∂ ∂ X X ∂xi X X
(x × v) = ijk xi vj ek = ijk vj e k = ijk δia vj ek = ajk vj ek
∂xa ∂xa ∂xa
i,j,k i,j,k i,j,k j,k
It follows,
∂ X
(x × v) = 1jk vj ek = v2 e3 − v3 e2 = (0, −v3 , v2 )
∂x1
j,k
∂ X
(x × v) = 2jk vj ek = v3 e1 − v1 e3 = (v3 , 0, −v1 )
∂x2
j,k
∂ X
(x × v) = 3jk vj ek = v1 e2 − v2 e1 = (−v2 , v1 , 0)
∂x3
j,k
Thus the Jacobian is simply,

 
0 v3 −v2
[df(x,y) ] =  −v3 0 −v1 
v2 v1 0
In fact, dfp (h) = f (h) = h × v for each p ∈ R3 . The given mapping is linear so the differential of
the mapping is precisely the mapping itself (we could short-cut much of this calculation and simply
quote Example 4.1.2 where we proved dT = T for linear T ).
Example 4.4.16. Let f (x, y) = (x, y, 1 − x − y). You can calculate,

 
1 0
[df(x,y,z) ] =  0 1 
−1 −1
Example 4.4.17. Let X(u, v) = (x, y, z) where x, y, z denote functions of u, v and I prefer to omit
the explicit depedendence to reduce clutter in the equations to follow.
∂X ∂X
= Xu = (xu , yu , zu ) and = Xv = (xv , yv , zv )
∂u ∂v
Then the Jacobian is the 3 × 2 matrix

 
xu xv
dX(u,v) =  yu yv 
zu zv
Remark 4.4.18.
I return to these examples in the next chapter and we’ll explore the geometric content of
these formulas as they support the application of certain theorems. More on that later, for
the remainder of this chapter we continue to focus on properties of differentiation.
4.5. CHAIN RULE 93
4.5 chain rule

The proof in Edwards is on 77-78. I’ll give a heuristic proof here which captures the essence of the
argument. The simplicity of this rule continues to amaze me.
Proposition 4.5.1.
If F : U ⊆ Rn → Rp is differentiable at a and G : V ⊆ Rp → Rm is differentiable at
F (a) ∈ V then G ◦ F is differentiable at a and
d(G ◦ F )a = (dG)F (a) ◦ dFa or, in matrix notation, (G ◦ F )0 (a) = G0 (F (a))F 0 (a)
Proof Sketch: (please forgive my lazy formatting, if only the summer was a year)
In calculus III you may have learned how to calculate partial derivatives in terms of tree-diagrams
and intermediate variable etc... We now have a way of understanding those rules and all the
other chain rules in terms of one over-arching calculation: matrix multiplication of the constituent
Jacobians in the composite function. Of course once we have this rule for the composite of two
functions we can generalize to n-functions by a simple induction argument. For example, for three
suitably defined mappings F, G, H,
(F ◦ G ◦ H)0 (a) = F 0 (G(H(a)))G0 (H(a))H 0 (a)
Example 4.5.2. .
Example 4.5.3. .
Example 4.5.4. .
4.5. CHAIN RULE 95
Example 4.5.5. .
Example 4.5.6. .
4.6 common product rules

What sort of product can we expect to find among mappings? Remember two mappings have
vector outputs and there is no way to multiply vectors in general. Of course, in the case we have
two mappings that have equal-dimensional outputs we could take their dot-product. There is a
~ B
product rule for that case: if A, ~ : Rn → Rm then
~ · B)
∂j (A ~ = (∂j A)
~ · B)
~ +A
~ · (∂j B)
~
Or in the special case of m = 3 we could even take their cross-product and there is another product
rule in that case:
~ × B)
∂j (A ~ ×B
~ = (∂j A) ~ +A ~
~ × (∂j B)
What other case can we ”multiply” vectors? One very important case is R2 = C where is is
customary to use the notation (x, y) = x + iy and f = u + iv. If our range is complex numbers
then we again have a product rule: if f : Rn → C and g : Rn → C then
∂j (f g) = (∂j f )g + f (∂j g)
I have relegated the proof of most of these product rules to the end of this section. One other
object worth differentiating is a matrix-valued function of Rn . If we define the partial derivative
of a matrix to be the matrix of partial derivatives then partial differentiation will respect the sum
and product of matrices (we may return to this in depth if need be later on):
∂j (A + B) = ∂j B + ∂j B ∂j (AB) = (∂j A)B + A(∂j B)
Moral of this story? If you have a pair mappings whose ranges allow some sort of product then it
is entirely likely that there is a corresponding product rule 6 .
4.6.1 scalar-vector product rule

There is one product rule which we can state for arbitrary mappings, note that we can always
sensibly multiply a mapping by a function. Suppose then that G : U ⊆ Rn → Rm and f : U ⊆
Rn → R are differentiable at a ∈ U . It follows that there exist linear transformations LG : Rn → Rm
and Lf : Rn → R where
G(a + h) − G(a) − LG (h) f (a + h) − f (a) − Lf (h)
lim =0 lim =0
h→0 ||h|| h→0 h
Since G(a + h) ≈ G(a) + LG (h) and f (a + h) ≈ f (a) + Lf (h) we expect
f G(a + h) ≈ (f (a) + Lf (h))(G(a) + LG (h))
≈ (f G)(a) + G(a)Lf (h) + f (a)LG (h) + Lf (h)LG (h)
| {z } | {z }
linear in h 2nd order in h
6
In my research I consider functions on supernumbers, these also can be multiplied. Naturally there is a product
rule for super functions, the catch is that super numbers z, w do not necessarily commute. However, if they’re
homogeneneous zw = (−1)w z wz. Because of this the super product rule is ∂M (f g) = (∂M f )g + (−1)f M f (∂M g)
4.6. COMMON PRODUCT RULES 97
Thus we propose: L(h) = G(a)Lf (h) + f (a)LG (h) is the best linear approximation of f G.
(f G)(a + h) − (f G)(a) − L(h)

lim =
h→0 ||h||
f (a + h)G(a + h) − f (a)G(a) − G(a)Lf (h) − f (a)LG (h)
= lim
h→0 ||h||
f (a + h)G(a + h) − f (a)G(a) − G(a)Lf (h) − f (a)LG (h)
= lim +
h→0 ||h||
f (a)G(a + h) − G(a + h)f (a)
+ lim
h→0 ||h||
f (a + h)G(a) − G(a)f (a + h)
+ lim
h→0 ||h||
f (a)G(a) − G(a)f (a)
+ lim
h→0 ||h||

= lim f (a) + G(a)+
h→0 ||h|| ||h||

G(a + h) − G(a)
+ f (a + h) − f (a)
||h||

= f (a) lim + lim G(a)
h→0 ||h|| h→0 ||h||
=0
Where we have made use of the differentiability and the consequent continuity of both f and G at
a. Furthermore, note
L(h + ck) = G(a)Lf (h + ck) + f (a)LG (h + ck)

= G(a)(Lf (h) + cLf (k)) + f (a)(LG (h) + cLG (k))
= G(a)Lf (h) + f (a)(LG (h) + c(G(a)Lf (k) + f (a)LG (k))
= L(h) + cL(k)
for all h, k ∈ Rn and c ∈ R hence L = G(a)Lf + f (a)LG is a linear transformation. We have proved
(most of) the following proposition:
Proposition 4.6.1.
If G : U ⊆ Rn → Rm and f : U ⊆ Rn → R are differentiable at a ∈ U then f G is

differentiable at a and
d(f G)a = (df )a G(a) + f (a)dGa (f G)0 (a) = f 0 (a)G(a) + f (a)G0 (a)
The argument above covers the ordinary product rule and a host of other less common rules. Note
again that G(a) and G0 (a) are vectors.
4.6.2 calculus of paths in R3

A path is a mapping from R to Rm . We use such mappings to model position, velocity and
acceleration of particles in the case m = 3. Some of these things were proved in previous sections
of this chapter but I intend for this section to be self-contained so that you can read it without
digging through the rest of this chapter.
Proposition 4.6.2.
If F, G : U ⊆ R → Rm are differentiable vector-valued functions and φ : U ⊆ R → R is a
differentiable real-valued function then for each t ∈ U ,
1. (F + G)0 (t) = F 0 (t) + G0 (t).
2. (cF )0 (t) = cF 0 (t).
3. (φF )0 (t) = φ0 (t)F (t) + φ(t)F 0 (t).
4. (F · G)0 (t) = F 0 (t) · G(t) + F (t) · G0 (t).
5. provided m = 3, (F × G)0 (t) = F 0 (t) × G(t) + F (t) × G0 (t).
6. provided φ(U ) ⊂ dom(F 0 ), (F ◦ φ)0 (t) = φ0 (t)F (φ(t)).
We have to insist that m = 3 for the statement with cross-products since we only have a standard
cross-product in R3 . We prepare for the proof of the proposition with a useful lemma. Notice this
lemma tells us how to actually calculate the derivative of paths in examples. The derivative of
component functions is nothing more than calculus I and one of our goals is to reduce things to
those sort of calculations whenever possible.
Lemma 4.6.3.
If F : U ⊆ R → Rm is differentiable vector-valued function then for all t ∈ U ,
F 0 (t) = (F10 (t), F20 (t), . . . , Fm

0
(t))
We are given that the following vector limit exists and is equal to F 0 (t),
F (t + h) − F (t)
F 0 (t) = lim
h→0 h
then by Proposition 3.1.8 the limit of a vector is related to the limits of its components as follows:
Fj (t + h) − Fj (t)
F 0 (t) · ej = lim .
h→0 h
Thus (F 0 (t))j = Fj0 (t) and the lemma follows7 . O
7
this notation I first saw in a text by Marsden, it means the proof is partially completed but you should read on
to finish the proof
P P
Proof of proposition: We use the notation F P = Fj ej = (F1 , . . . , Fm ) and G = i Gi ei =
(G1 , . . . , Gm ) throughout the proofs below. The is understood to range over 1, 2, . . . m. Begin
with (1.),
[(F + G)0 ]j = d
dt [(F + G)j ] using the lemma
d
= dt [Fj + Gj ] using def. (F + G)j = Fj + Gj
= d d
dt [Fj ] + dt [Gj ] by calculus I, (f + g)0 = f 0 + g 0 .
= [F 0 + G0 ]j def. of vector addition for F 0 and G0
Hence (F × G)0 = F 0 × G + F × G0 .The proofs of 2,3,5 and 6 are similar. I’ll prove (5.),
[(F × G)0 ]k = d
dt [(F × G)k ] using the lemma
X
d
= dt [ ijk Fi Gj ] using def. F × G
X
= d
ijk dt [Fi Gj ] repeatedly using, (f + g)0 = f 0 + g 0
dG
X
= ijk [ dF j
dt Gj + Fi dt ]
i
repeatedly using, (f g)0 = f 0 g + f g 0
dG
X X X
= ijk dFdt Gj
i
ijk Fi dtj ] property of finite sum
= ( dF
dt × G)k + (F ×
dG
dt )k ) def. of cross product
dF dG

= dt ×G+F × dt k def. of vector addition
Notice that the calculus step really just involves calculus I applied to the components. The ordinary
product rule was the crucial factor to prove the product rule for cross-products. We’ll see the same
for the dot product of mappings. Prove (4.)
X
(F · G)0 (t) = dt
d
[ Fk G k ] using def. F · G
X
= d
dt [Fk Gk ] repeatedly using, (f + g)0 = f 0 + g 0
X
= [ dF dGk
dt Gk + Fk dt ]
k
repeatedly using, (f g)0 = f 0 g + f g 0
dF dG
= dt ·G+F · dt . def. of dot product
The proof of (3.) follows from applying the product rule to each component of φ(t)F (t). The proof
of (2.) follow from (3.) in the case that φ(t) = c so φ0 (t) = 0. Finally the proof of (6.) follows from
applying the chain-rule to each component.
4.6.3 calculus of matrix-valued functions of a real variable

Definition 4.6.4.
A matrix-valued function of a real variable is a function from I ⊆ R to R m×n . Suppose

A : I ⊆ R → R m×n is such that Aij : I ⊆ R → R is differentiable for each i, j then we
define dAij
dA
dt = dt
0 0
R R
which can also be denoted (A )ij = Aij . We likewise define Adt = [ Aij dt] for A with
integrable components. Definite integrals and higher derivatives are also defined component-
wise.
2t 3t2

Example 4.6.5. Suppose A(t) = . I’ll calculate a few items just to illustrate the
4t3 5t4
definition above. calculate; to differentiate a matrix we differentiate each component one at a time:

0 2 6t 00 0 6 0 2 0
A (t) = A (t) = A (0) =
12t2 20t3 24t 60t2 0 0
Integrate by integrating each component:

 
2 2 t3 2

2 t
t2 + c1 t3 + c2
Z Z
0 0  4 8
A(t)dt = A(t)dt =  =

t4 + c3 t5 + c4 0 2 2  16 32
t4 t5
0 0
Proposition 4.6.6.
Suppose A, B are matrix-valued functions of a real variable, f is a function of a real variable,

c is a constant, and C is a constant matrix then
1. (AB)0 = A0 B + AB 0 (product rule for matrices)
2. (AC)0 = A0 C
3. (CA)0 = CA0
4. (f A)0 = f 0 A + f A0
5. (cA)0 = cA0
6. (A + B)0 = A0 + B 0
where each of the functions is evaluated at the same time t and I assume that the functions
and matrices are differentiable at that value of t and of course the matrices A, B, C are such
that the multiplications are well-defined.
Proof: Suppose A(t) ∈ R m×n and B(t) ∈ R n×p consider,
(AB)0 ij d
= dt ((AB)ij ) defn. derivative of matrix
d P
= dt ( k Aik Bkj ) defn. of matrix multiplication
P d
= k dt (Aik Bkj ) linearity of derivative
P dAik dB
= k dt Bkj + Aik dtkj ordinary product rules
dB
= k dAdtik Bkj + k Aik dtkj
P P
algebra
= (A0 B)ij + (AB 0 )ij defn. of matrix multiplication
= (A0 B + AB 0 )ij defn. matrix addition
this proves (1.) as i, j were arbitrary in the calculation above. The proof of (2.) and (3.) follow
quickly from (1.) since C constant means C 0 = 0. Proof of (4.) is similar to (1.):
(f A)0 ij d
= dt ((f A)ij ) defn. derivative of matrix
d
= dt (f Aij ) defn. of scalar multiplication
df dA
= dt Aij + f dtij ordinary product rule
df
= ( dt A + f dA
dt )ij defn. matrix addition
= ( df
dt A + f dA
dt )ij defn. scalar multiplication.
The proof of (5.) follows from taking f (t) = c which has f 0 = 0. I leave the proof of (6.) as an
exercise for the reader. .
To summarize: the calculus of matrices is the same as the calculus of functions with the small
qualifier that we must respect the rules of matrix algebra. The noncommutativity of matrix mul-
tiplication is the main distinguishing feature.
4.6.4 calculus of complex-valued functions of a real variable

Differentiation of functions from R to C is defined by splitting a given function into its real and
imaginary parts then we just differentiate with respect to the real variable one component at a
time. For example:
d 2t d d
(e cos(t) + ie2t sin(t)) = (e2t cos(t)) + i (e2t sin(t))
dt dt dt
= (2e2t cos(t) − e2t sin(t)) + i(2e2t sin(t) + e2t cos(t)) (4.1)
2t
= e (2 + i)(cos(t) + i sin(t))
= (2 + i)e(2+i)t
d λt
where I have made use of the identity8 ex+iy = ex (cos(y) + i sin(y)). We just saw that dt e = λeλt
which seems obvious enough until you appreciate that we just proved it for λ = 2 + i.
8
or definition, depending on how you choose to set-up the complex exponential, I take this as the definition in
calculus II
4.7 continuous differentiability

We have noted that differentiablility on some set U implies all sorts of nice formulas in terms of
the partial derivatives. Curiously the converse is not quite so simple. It is possible for the partial
derivatives to exist on some set and yet the mapping may fail to be differentiable. We need an extra
topological condition on the partial derivatives if we are to avoid certain pathological9 examples.
Example 4.7.1. I found this example in Hubbard’s advanced calculus text(see Ex. 1.9.4, pg. 123).
It is a source of endless odd examples, notation and bizarre quotes. Let f (x) = 0 and
x 1
f (x) = + x2 sin
2 x
for all x 6= 0. I can be shown that the derivative f 0 (0) = 1/2. Moreover, we can show that f 0 (x)
exists for all x 6= 0, we can calculate:
1 1 1
f 0 (x) = + 2x sin − cos
2 x x
Notice that dom(f 0 ) = R. Note then that the tangent line at (0, 0) is y = x/2.
You might be tempted to say then that this function is increasing at a rate of 1/2 for x near zero.
But this claim would be false since you can see that f 0 (x) oscillates wildly without end near zero.
We have a tangent line at (0, 0) with positive slope for a function which is not increasing at (0, 0)
(recall that increasing is a concept we must define in a open interval to be careful). This sort of
thing cannot happen if the derivative is continuous near the point in question.
The one-dimensional case is really quite special, even though we had discontinuity of the derivative
we still had a well-defined tangent line to the point. However, many interesting theorems in calculus
of one-variable require the function to be continuously differentiable near the point of interest. For
example, to apply the 2nd-derivative test we need to find a point where the first derivative is zero
and the second derivative exists. We cannot hope to compute f 00 (xo ) unless f 0 is continuous at xo .
The next example is sick.
9
”pathological” as in, ”your clothes are so pathological, where’d you get them?”
4.7. CONTINUOUS DIFFERENTIABILITY 103
Example 4.7.2. Let us define f (0, 0) = 0 and
x2 y
f (x, y) =
x2 + y 2
for all (x, y) 6= (0, 0) in R2 . It can be shown that f is continuous at (0, 0). Moreover, since
f (x, 0) = f (0, y) = 0 for all x and all y it follows that f vanishes identically along the coordinate
axis. Thus the rate of change in the e1 or e2 directions is zero. We can calculate that
∂f 2xy 3 ∂f x4 − x2 y 2
= 2 and = 2
∂x (x + y 2 )2 ∂y (x + y 2 )2
If you examine the plot of z = f (x, y) you can see why the tangent plane does not exist at (0, 0, 0).
Notice the sides of the box in the picture are parallel to the x and y axes so the path considered
below would fall on a diagonal slice of these boxes10 . Consider the path to the origin t 7→ (t, t) gives
fx (t, t) = 2t4 /(t2 + t2 )2 = 1/2 hence fx (x, y) → 1/2 along the path t 7→ (t, t), but fx (0, 0) = 0 hence
the partial derivative fx is not continuous at (0, 0). In this example, the discontinuity of the partial
derivatives makes the tangent plane fail to exist.
One might be tempted to suppose that if a function is continuous at a given point and if all
the possible directional derivatives exist then differentiability should follow. It turns out this is
not sufficient since continuity of the function does not imply some continuity along the partial
derivatives. For example:
Example 4.7.3. Let us define f : R2 → R by f (x, y) = 0 for y = 6 x2 and f (x, x2 ) = x. I invite the
reader to verify that this function is continuous at the origin. Moreover, consider the directional
derivatives at (0, 0). We calculate, if v = ha, bi
f (0 + hv) − f (0) f (ah, bh) 0

Dv f (0, 0) = lim = lim = lim = 0.
h→0 h h→0 h h→0 h
10
the argument to follow stands alone, you don’t need to understand the picture to understand the math here, but
it’s nice if you do
To see why f (ah, bh) = 0, consider the intersection of ~r(h) = (ha, hb) and y = x2 the intersection
is found at hb = (ha)2 hence, noting h = 0 is not of interest in the limit, b = ha2 . If a = 0
then clearly (ah, bh) only falls on y = x2 at (0, 0). If a 6= 0 then the solution h = b/a2 gives
f (ha, hb) = ha a nontrivial value. However, as h → 0 we eventually reach values close enough
to (0, 0) that f (ah, bh) = 0. Hence we find all directional derivatives exist and are zero at (0, 0).
Let’s examine the graph of this example to see how this happened. The pictures below graph the
xy-plane as red and the nontrivial values of f as a blue curve. The union of these forms the graph
z = f (x, y).
Clearly, f is continuous at (0, 0) as I invited you to prove. Moreover, clearly z = f (x, y) cannot be
well-approximated by a tangent plane at (0, 0, 0). If we capture the xy-plane then we lose the blue
curve of the graph. On the other hand, if we use a tilted plane then we lose the xy-plane part of
the graph.
The moral of the story in the last two examples is simply that derivatives at a point, or even all
directional derivatives at a point do not necessarily tell you much about the function near the point.
This much is clear: something else is required if the differential is to have meaning which extends
beyond one point in a nice way. Therefore, we consider the following:
Definition 4.7.4.
A mapping F : U ⊆ Rn → Rm is continuously differentiable at a ∈ U iff the partial
derivative mappings Dj F exist on an open set containing a and are continuous at a.
Technically, the term continuously differentiable would seem to indicate that the mapping
x → dFx is a continuous mapping at x = a. If dFx is the differential at x then surely continuous
differentiability ought to indicate that the linear transformations dFx (thinking of x as varying) are
glued together in some continuous fashion. That is correct, however, it is beyond our techniques
at the present to discuss continuity of an operator-valued function. Further techniques must be
developed to properly address the continuity in question. That said, once we do those things11
11
see C.H. Edwards pages 172-180 and Proposition 2.4 in particular if you are so impatient as to not wait for me
4.7. CONTINUOUS DIFFERENTIABILITY 105
then we’ll find that this much more abstract idea of continuity indicates that the partial derivatives
of the Jacobian are continuous at x = a. Therefore, the definition above is not at odds with the
natural definition12
The import of the theorem below is that we can build the tangent plane from the Jacobian matrix
provided the partial derivatives exist near the point of tangency and are continuous at the point
of tangency. This is a very nice result because the concept of the linear mapping is quite abstract
but partial differentiation of a given mapping is often easy. The proof that follows here is found in
many texts, in particular see C.H. Edwards Advanced Calculus of Several Variables on pages 72-73.
Theorem 4.7.5.
If F : Rn → R is continuously differentiable at a then F is differentiable at a
Proof: Consider a+h sufficiently close to a that all the partial derivatives of F exist. Furthermore,
consider going from a to a+h by traversing a hyper-parallel-piped travelling n-perpendicular paths:
a → a + h1 e1 → a + h1 e1 + h2 e2 → · · · a + h1 e1 + · · · + hn en = a + h.
|{z} | {z } | {z } | {z }
po p1 p2 pn
Pj
Let us denote pj = a + bj where clearly bj ranges from bo = 0 to bn = h and bj = i=1 hi ei . Notice
that the difference between pj and pj−1 is given by:
j
X j−1
X
pj − pj−1 = a + hi ei − a − hi ei = hj ej
i=1 i=1
Consider then the following identity,
F (a + h) − F (a) = F (pn ) − F (pn−1 ) + F (pn−1 ) − F (pn−2 ) + · · · + F (p1 ) − F (po )
This is to say the change in F from po = a to pn = a + h can be expressed as a sum of the changes
along the n-steps. Furthermore, if we consider the difference F (pj ) − F (pj−1 ) you can see that only
the j-th component of the argument of F changes. Since the j-th partial derivative exists on the
interval for hj considered by construction we can apply the mean value theorem to locate cj such
that:
hj ∂j F (pj−1 + cj ej ) = F (pj ) − F (pj−1 )
Therefore, using the mean value theorem for each interval, we select c1 , . . . cn with:
n
X
F (a + h) − F (a) = hj ∂j F (pj−1 + cj ej )
j=1
12
Obviously, the reason this is given as the definition of continuously differentiable here and in most textbooks is
to avoid the discussion about continuity of operators.
It follows we should propose L to satisfy the definition of Frechet differentation as follows:

n
X
L(h) = hj ∂j F (a)
j=1
It is clear that L is linear (in fact, perhaps you recognize this as L(h) = (∇F )(a) • h). Let us
prepare to study the Frechet quotient,
n
X n
X
F (a + h) − F (a) − L(h) = hj ∂j F (pj−1 + cj ej ) − hj ∂j F (a)
j=1 j=1
Xn

= hj ∂j F (pj−1 + cj ej ) − ∂j F (a)
| {z }
j=1
gj (h)
Observe that pj−1 + cj ej → a as h → 0. Thus, gj (h) → 0 by the continuity of the partial derivatives
at x = a. Finally, consider the Frechet quotient:
P
F (a + h) − F (a) − L(h) j hj gj (h)
X hj
lim = lim = lim gj (h)
h→0 ||h|| h→0 ||h|| h→0 ||h||
j

hj
Notice |hj | ≤ ||h|| hence ||h|| ≤ 1 and

hj
0 ≤ gj (h) ≤ |gj (h)|
||h||
Apply the squeeze theorem to deduce each term in the sum ? limits to zero. Consquently, L(h)
satisfies the Frechet quotient and we have shown that F is differentiable
Pnat x = a and the differen-
tial is expressed in terms of partial derivatives as expected; dFx (h) = j=1 hj ∂j F (a) .
Given the result above it is a simple matter to extend the proof to F : Rn → Rm .
Theorem 4.7.6.
If F : Rn → Rm is continuously differentiable at a then F is differentiable at a
Proof: If F is continuously differentiable at a then clearly each component function F j : Rn → R

is continuously differentiable at a. Thus, by Theorem 4.7.5 we have F j differentiable at a hence
F j (a + h) − F j (a) − dFaj (h) F (a + h) − F (a) − dFa (h)

lim = 0 for all j ∈ Nm ⇒ lim =0
h→0 ||h|| h→0 ||h||
by Proposition 3.1.8. This proves F is differentiable at a .

4.8. ON WHY PARTIAL DERIVATIVES COMMUTE 107
4.8 on why partial derivatives commute

The goal of this section is to prove the partial derivatives commute for nice functions. Of course
some of the results we discuss on the way to that goal are interesting in their own right as well.
Definition 4.8.1.
We say U ⊆ Rn is path connected iff any two points in U can be connected by a path
which is contained within the set.
For example, Rn is connected since given any two points a, b ∈ Rn we can construct the path
φ(t) = a + t(b − a) from a to b and naturally the path is within the set. You can easily verify that
open and closed balls are also path connected. Even a donut is path connected. However, a pair
donuts is not path connected unless it’s one of those artsy figure-8 deals.
Proposition 4.8.2.
If U is a connected open subset of Rn then a differentiable mapping F : U → Rm is constant
iff F 0 (u) = 0 for all u ∈ U .
Proposition 4.8.3.
If U is a connected open subset of Rn and the differentiable mappings F, G : U → R such
that F 0 (x) = G0 (x) for all x ∈ U then there exists a constant vector c ∈ Rm such that
F (x) = G(x) + c for all x ∈ U .
There is no mean value theorem for mappings since counter-examples exist. For example, Exercise
1.12 on page 63 shows the mean value theorem fails for the helix. In particular, you can find
average velocity vector over a particular time interval such that the velocity vector never matches
the average velocity over that time period. Fortunately, if we restrict our attention to mappings
with one-dimensional codomains we still have a nice theorem:
Proposition 4.8.4. (Mean Value Theorem)
Suppose that f : U → R is a differentiable function and U is an open set. Furthermore,
suppose U contains the line segment from a to b in U ;
La,b = {a + t(b − a) | t ∈ [0, 1]} ⊂ U.
It follows that there exists some point c ∈ La,b such that f (b) − f (a) = f 0 (c)(b − a).
Definition 4.8.5. (higher derivatives)

We define nested directional derivatives in the natural way:
Dh f (x + tk) − Dh f (x)
Dk Dh f (x) = Dk (Dh f (x)) = lim
t→0 t
Furthermore, the second difference is defined by
∆2 fa (h, k) = f (a + h + k) − f (a + h) − f (a + k) + f (a)
4.8. ON WHY PARTIAL DERIVATIVES COMMUTE 109
This is Lemma 3.5 on page 86 of Edwards.

Proposition 4.8.6.
Suppose U us an open set and f : U → R which is differentiable on U with likewise
differentiable directional derivative function on U . Suppose that a, a + h, a + k, a + h + k
are all in U then there exist α, β ∈ (0, 1) such that ∆2 fa (h, k) = Dk Dh f (a + αh + βk).
The proof is rather neat. The α and β stem from two applications of the MVT, once for the
function then once for its directional derivative.
Proposition 4.8.7.
Let U be an open subset of Rn . If f : U → R is a function with continuous first and second
partial derivatives on U then for all i, j = 1, 2, . . . , n we have Di Dj f = Dj Di f on U ;
∂2f ∂2f
= .
∂xi ∂xj ∂xj ∂xi
Pull out h, k using homogeneity of Dcv f = cDv f and take limit (h, k) → 0 to drop the h, k inside
the Df (· · · ) terms. This is possible thanks to the continuity of the partial derivatives near a.
4.9 complex analysis in a nutshell

Differentiation with respect to a real variable can be reduced to the slogan that we differentiate
componentwise. Differentiation with respect to a complex variable requires additional structure.
They key distinguishing ingredient is complex linearity:
Definition 4.9.1.
If we have some function T : C → C such that
(1.) T (v + w) = T (v) + T (w) for all v, w ∈ C (2.) T (cv) = cT (v) for all c, v ∈ C
then we would say that T is complex-linear.

Condition (1.) is additivity whereas condition (2.) is homogeneity. Note that complex linearity
implies real linearity however the converse is not true.
Example 4.9.2. Suppose T (z) = z̄ where if z = x + iy for x, y ∈ R then z̄ = x − iy is the complex
conjugate of z. Consider for c = a + ib where a, b ∈ R,
T (cz) = T ((a + ib)(x + iy))
= T (ax − by + i(ay + bx))
= ax − by − i(ay + bx)
= (a − ib)(x − iy)
= c̄ T (z)
hence this map is not complex linear. On the other hand, if we study mutiplication by just a ∈ R,
T (az) = T (a(x + iy)) = T (ax + iay) = ax − iay = a(x − iy) = aT (x + iy) = aT (z)
thus T is homogeneous with respect to real-number multiplication and it is also additive hence T is
real linear.
Suppose that Lis a linear 2 2
mapping from R to R . It is known from linear algebra that there exists
a b
a matrix A = such that L(v) = Av for all v ∈ R2 . In this section we use the notation
c d
a + ib = (a, b) and
(a, b) ∗ (c, d) = (a + ib)(c + id) = ac − bd + i(ad + bc) = (ac − bd, ad + bc).
This construction is due to Gauss in the early nineteenth century, the idea is to use two component
vectors to construct complex numbers. 13
There are other ways to construct complex numbers .
a b x
Notice that L(x + iy) = = (ax + by, cx + dy) = ax + by + i(cx + dy) defines a real
c d y
linear mapping on C for any choice of the real constants a, b, c, d. In contrast, complex linearity
puts strict conditions on these constants:
13
the same is true for real numbers, you can construct them in more than one way, however all constructions agree
on the basic properties and as such it is the properties of real or complex numbers which truly defined them. That
said, we choose Gauss’ representation for convenience.
4.9. COMPLEX ANALYSIS IN A NUTSHELL 111
Theorem 4.9.3.
The linear mapping L(v) = Av is complex linear iff the matrix A will have the special form
below:
a b
−b a
To be clear, we mean to identify R2 with C as before. Thus the condition of complex
homogeneity reads L((a, b) ∗ (x, y)) = (a, b) ∗ L(x, y)
Proof: assume L is complex linear. Define the matrix of L as before:

a b x
L(x, y) =
c d y
This yields,
L(x + iy) = ax + by + i(cx + dy)
We can gain conditions on the matrix by examining the special points 1 = (1, 0) and i = (0, 1)
L(1, 0) = (a, c) L(0, 1) = (b, d)
Note that (c1 , c2 ) ∗ (1, 0) = (c1 , c2 ) hence L((c1 + ic2 )1) = (c1 + ic2 )L(1) yields
(ac1 + bc2 ) + i(cc1 + dc2 ) = (c1 + ic2 )(a + ic) = c1 a − c2 c + i(c1 c + c2 a)
We find two equations by equating the real and imaginary parts:
ac1 + bc2 = c1 a − c2 c cc1 + dc2 = c1 c + c2 a
Therefore, bc2 = −c2 c and dc2 = c2 a for all (c1 , c2 ) ∈ C. Suppose c1 = 0 and c2 = 1. We find
b = −c and d = a. We leave the converse proof to the reader. The proposition follows.
In analogy with the real case we define f 0 (z) as follows:

Definition 4.9.4.
Suppose f : dom(f ) ⊆ C → C and z ∈ dom(f ) then we define f 0 (z) by the limit below:
f (z + h) − f (z)
f 0 (z) = lim .
h→0 h
The derivative function f 0 is defined pointwise for all such z ∈ dom(f ) that the limit above
exists.
f 0 (z)h
Note that f 0 (z) = limh→0 h hence
f 0 (z)h f (z + h) − f (z) f (z + h) − f (z) − f 0 (z)h
lim = lim ⇒ lim =0
h→0 h h→0 h h→0 h
Note that the limit above simply says that L(v) = f 0 (z)v gives the is the best complex-linear
approximation of ∆f = f (z + h) − f (z).
Proposition 4.9.5.
If f is a complex differentiable at zo then linearization L(h) = f 0 (zo )h is a complex linear

mapping.
Proof: let c, h ∈ C and note L(ch) = f 0 (zo )(ch) = cf 0 (zo )h = cL(h).
It turns out that complex differentiability automatically induces real differentiability:

Proposition 4.9.6.
If f is a complex differentiable at zo then f is (real) differentiable at zo with L(h) = f 0 (zo )h.

f (z+h)−f (z)−f 0 (z)h
Proof: note that limh→0 h = 0 implies
f (z + h) − f (z) − f 0 (z)h
lim =0
h→0 |h|
but then |h| = ||h|| and we know L(h) = f 0 (zo )h is real-linear hence L is the best linear approxi-
mation to ∆f at zo and the proposition follows.
Let’s summarize what we’ve learned: if f : dom(f ) → C is complex differentiable at zo and

f = u + iv then,
1. L(h) = f 0 (zo )h is complex linear.
2. L(h) = f 0 (zo )h is the best real linear approximation to f viewed as a mapping on R2 .

The Jacobian matrix for f = (u, v) has the form

ux (po ) uy (po )
Jf (po ) =
vx (po ) vy (po )
Theorem 4.9.3 applies to Jf (po ) since L is a complex linear mapping. Therefore we find the Cauchy
Riemann equations: ux = vy and uy = −vx . We have proved the following theorem:
Theorem 4.9.7.
If f = u + iv is a complex function which is complex-differentiable at zo then the partial
derivatives of u and v exist at zo and satisfy the Cauchy-Riemann equations at zo
∂u ∂v ∂u ∂v
= =− .
∂x ∂y ∂y ∂x
Example 4.9.8. Let f (z) = ez where the definition of the complex exponential function is given
by the following, for each x, y ∈ R and z = x + iy
f (x + iy) = ex+iy = ex (cos(y) + i sin(y)) = ex cos(y) + iex sin(y)

Identify for f = u + iv we have u(x, y) = ex cos(y) and v(x, y) = ex sin(y). Calculate:

∂u ∂ x ∂u ∂ x
e cos(y) = ex cos(y) e cos(y) = −ex sin(y),

= & =
∂x ∂x ∂y ∂y
∂v ∂ x ∂v ∂ x
e sin(y) = ex sin(y) e sin(y) = ex cos(y).

= & =
∂x ∂x ∂x ∂y
∂u ∂v ∂u ∂v
Thus f satisfies the CR-equations ∂x = ∂y and ∂y = − ∂x . The complex exponential function is
complex differentiable.
The converse of Theorem 4.9.7 is not true in general. It is possible to have functions u, v : U ⊆
R2 → R that satisfy the CR-equations at zo ∈ U and yet f = u+iv fails to be complex differentiable
at zo .
(
0 if xy 6= 0
Example 4.9.9. Counter-example to converse of Theorem 4.9.7. Suppose f (x+iy) = .
1 if xy = 0
Clearly f has constant value 1 on the coordinate axes thus along the x-axes we can calculate the
partial derivatives for u and v and they are both zero.
Likewise, along the y-axis we find uy and vy exist and are zero. At the origin we find ux , uy , vx , vy
all exist and are zero. Therefore, the Cauchy-Riemann equations hold true at the origin. However,
this function is not even continuous at the origin, thus it is not real differentiable!
The example above equally well serves as an example for a point where a function has partial
derivatives which exist at all orders and yet the differential fails to exist. It’s not a problem of
complex variables in my opinion, it’s a problem of advanced calculus. The key concept to reverse
the theorem is continuous differentiability.
Theorem 4.9.10.
If u, v, ux , uy , vx , vy are continuous functions in some open disk of zo and ux (zo ) = vy (zo )
and uy (zo ) = −vx (zo ) then f = u + iv is complex differentiable at zo .
Proof: we are given that a function f : D (zo ) ⊂ R2 → R2 is continuous with continuous partial
derivatives of its component functions u and v. Therefore, by Theorem 4.7.6 we know f is (real)
differentiable at zo . Therefore, we have a best linear approximation to the change in f near zo
which can be induced via multiplication of the Jacobian matrix:

ux (zo ) uy (zo ) v1
L(v1 , v2 ) = .
vx (zo ) vy (zo ) v2
Note then that the given CR-equations show the matrix of L has the form

a b
[L] =
−b a
where a = ux (zo ) and b = vx (zo ). Consequently we find L is complex linear and it follows that f
is complex differentiable at zo since we have a complex linear map L such that
f (z + h) − f (z) − L(h)
lim =0
h→0 ||h||
note that the limit with h in the denominator is equivalent to the limit above which followed directly
from the (real) differentiability at zo . (the following is not needed for the proof of the theorem, but
perhaps it is interesting anyway) Moreover, we can write

ux uy h1
L(h1 , h2 ) =
−uy ux h2

ux h1 + uy h2
=
−uy h1 + ux h2
= ux h1 + uy h2 + i(−uy h1 + ux h2 )
= (ux − iuy )(h1 + ih2 )
Therefore we find f 0 (zo ) = ux − iuy gives L(h) = f 0 (zo )z.
In the preceding section we found necessary and sufficient conditions for the component functions
u, v to construct an complex differentiable function f = u + iv. The definition that follows is the
next logical step: we say a function is analytic14 at zo if it is complex differentiable at each point
in some open disk about zo .
Definition 4.9.11.
Let f = u + iv be a complex function. If there exists > 0 such that f is complex
differentiable for each z ∈ D (zo ) then we say that f is analytic at zo . If f is analytic for
each zo ∈ U then we say f is analytic on U . If f is not analytic at zo then we say that zo
is a singular point. Singular points may be outside the domain of the function. If f is
analytic on the entire complex plane then we say f is entire. Analytic functions are
also called holomorphic functions
If you look in my complex variables notes you can find proof of the following theorem (well, partial
proof perhaps, but this result is shown in every good complex variables text)
14
you may recall that a function on R was analyic at xo if its Talyor series at xo converged to the function in some
neighborhood of xo . This terminology is consistent but I leave the details for your complex analysis course
Theorem 4.9.12.
If f : R → R ⊂ C is a function and f˜ : C → C is an extension of f which is analytic then

f˜ is unique. In particular, if there is an analytic extension of sine, cosine, hyperbolic sine
or hyperbolic cosine then those extensions are unique.
This means if we demand analyticity then we actually had no freedom in our choice of the exponen-
tial. If we find a complex function which matches the exponential function on a line-segment ( in
particular a closed interval in R viewed as a subset of C is a line-segment ) then there is just one
complex function which agrees with the real exponential and is complex differentiable everywhere.
f (x) = ex extends uniquely to f˜(z) = eRe(z) (cos(Im(z)) + i sin(Im(z))).
Note f˜(x + 0i) = ex (cos(0) + i sin(0)) = ex thus f˜|R = f . Naturally, analyiticity is a desireable
property for the complex-extension of known functions so this concept of analytic continuation is
very nice. Existence aside, we should first construct sine, cosine etc... then we have to check they
are both analytic and also that they actually agree with the real sine or cosine etc... If a function
on R has vertical asymptotes, points of discontinuity or points where it is not smooth then the
story is more complicated.
4.9.1 harmonic functions

We’ve discussed in some depth how to determine if a given function f = u + iv is in fact analytic.
In this section we study another angle on the story. We learn that the component functions
u, v of an analytic function f = u + iv are harmonic conjugates and they satisfy the phyically
significant Laplace’s equation ∇2 φ = 0 where ∇2 = ∂ 2 /∂x2 + ∂ 2 /∂y 2 . In addition we’ll learn
that if we have one solution of Laplace’s equation then we can consider it to be the ”u” of some
yet undetermined analytic function f = u + iv. The remaining function v is then constructed
through some integration guided by the CR-equations. The construction is similar to the problem
of construction of a potential function for a given conservative force in calculus III.
Proposition 4.9.13.
If f = u + iv is analytic on some domain D ⊆ C then u and v are solutions of Laplace’s

equation φxx + φyy = 0 on D.
Proof: since f = u + iv is analytic we know the CR-equations hold true; ux = vy and uy = −vx .
Moreover, f is continuously differentiable so we may commute partial derivatives by a theorem
from multivariate calculus. Consider
uxx + uyy = (ux )x + (uy )y = (vy )x + (−vx )y = vyx − vxy = 0
Likewise,
vxx + vyy = (vx )x + (vy )y = (−uy )x + (ux )y = −uyx + uxy = 0
Of course these relations hold for all points inside D and the proposition follows.
Example 4.9.14. Note f (z) = z 2 is analytic with u = x2 − y 2 and v = 2xy. We calculate,
uxx = 2, uyy = −2 ⇒ uxx + uyy = 0
Note vxx = vyy = 0 so v is also a solution to Laplace’s equation.

Now let’s see if we can reverse this idea.
Example 4.9.15. Let u(x, y) = x + c1 notice that u solves Laplace’s equation. We seek to find a
harmonic conjugate of u. We need to find v such that,
∂v ∂u ∂v ∂u
= =1 =− =0
∂y ∂x ∂x ∂y
Integrate these equations to deduce v(x, y) = y + c2 for some constant c2 ∈ R. We thus construct
an analytic function f (x, y) = x + c1 + i(y + c2 ) = x + iy + c1 + ic2 . This is just f (z) = z + c for
c = c1 + ic2 .
Example 4.9.16. Suppose u(x, y) = ex cos(y). Note that uxx = u whereas uyy = −u hence
uxx + uyy = 0. We seek to find v such that
∂v ∂u ∂v ∂u
= = ex cos(y) =− = ex sin(y)
∂y ∂x ∂x ∂y
Integrating vy = ex cos(y) with respect to y and vx = ex sin(y) with respect to x yields v(x, y) =
ex sin(y). We thus construct an analytic function f (x, y) = ex cos(y) + iex sin(y). Of course we
should recognize the function we just constructed, it’s just the complex exponential f (z) = ez .
Notice we cannot just construct an analytic function from any given function of two variables. We
have to start with a solution to Laplace’s equation. This condition is rather restrictive. There
is much more to say about harmonic functions, especially where applications are concerned. My
goal here was just to give another perspective on analytic functions. Geometrically one thing we
could see without further work at this point is that for an analytic function f = u + iv the families
of level curves u(x, y) = c1 and v(x, y) = c2 are orthogonal. Note grad(u) =< ux , uy > and
grad(v) =< vx , vy > have
grad(u) · grad(v) = ux vx + uy vy = −ux uy + uy ux = 0
This means the normal lines to the level curves for u and v are orthogonal. Hence the level curves
of u and v are orthogonal. Another way to twist this, if you want to obtain orthogonal families of
curves then analytic functions provide an easy way to create examples.
Remark 4.9.17.
This section covers a few lectures of the complex analysis course. I include it here in part
to make connections. I always encourage students to understand math outside the comfort
zone of isolated course components. Whenever we can understand material from several
courses as part of a larger framework it is a step in the right direction.
Chapter 5
inverse and implicit function theorems
It is tempting to give a complete and rigourous proof of these theorems at the outset, but I will
resist the temptation in lecture. I’m actually more interested that the student understand what the
theorem claims before I show the real proof. I will sketch the proof and show many applications.
A nearly complete proof is found in Edwards where he uses an iterative approximation technique
founded on the contraction mapping principle, we will go through that a bit later in the course. I
probably will not have typed notes on that material this semester, but Edward’s is fairly readable
and I think we’ll profit from working through those sections. That said, we develop an intuition for
just what these theorems are all about to start. That is the point of this chapter: to grasp what
the linear algebra of the Jacobian suggests about the local behaviour of functions and equations.
5.1 inverse function theorem

Consider the problem of finding a local inverse for f : dom(f ) ⊆ R → R. If we are given a point
p ∈ dom(f ) such that there exists an open interval I containing p with f |I a one-one function then
we can reasonably construct an inverse function by the simple rule f −1 (y) = x iff f (x) = y for
x ∈ I and y ∈ f (I). A sufficient condition to insure the existence of a local inverse is that the
derivative function is either strictly positive or strictly negative on some neighborhood of p. If we
are give a continuously differentiable function at p then it has a derivative which is continuous on
some neighborhood of p. For such a function if f 0 (p) 6= 0 then there exists some interval centered at
p for which the derivative is strictly positive or negative. It follows that such a function is strictly
monotonic and is hence one-one thus there is a local inverse at p. We should all learn in calculus
I that the derivative informs us about the local invertibility of a function. Natural question to ask
for us here: does this extend to higher dimensions? If so, how?
The arguments I just made are supported by theorems that are developed in calculus I. Let me shift
gears a bit and give a direct calculational explaination based on the linearization approximation.
117
118 CHAPTER 5. INVERSE AND IMPLICIT FUNCTION THEOREMS
If x ≈ p then f (x) ≈ f (p) + f 0 (p)(x − p). To find the formula for the inverse we solve y = f (x) for
x:
1
y ≈ f (p) + f 0 (p)(x − p) ⇒ x ≈ p + 0

y − f (p)
f (p)
1
Therefore, f −1 (y) ≈ p +

y − f (p) for y near f (p).
f 0 (p)
Example 5.1.1. Just to help you believe me, consider f (x) = 3x − 2 then f 0 (x) = 3 for all x.
Suppose we want to find the inverse function near p = 2 then the discussion preceding this example
suggests,
1
f −1 (y) = 2 + (y − 4).
3
I invite the reader to check that f (f (y)) = y and f −1 (f (x)) = x for all x, y ∈ R.
−1
In the example above we found a global inverse exactly, but this is thanks to the linearity of the
function in the example. Generally, inverting the linearization just gives the first approximation to
the inverse.
Consider F : dom(F ) ⊆ Rn → Rn . If F is differentiable at p ∈ Rn then we can write F (x) ≈

F (p) + F 0 (p)(x − p) for x ≈ p. Set y = F (x) and solve for x via matrix algebra. This time we need
to assume F 0 (p) is an invertible matrix in order to isolate x,
y ≈ F (p) + F 0 (p)(x − p) ⇒ x ≈ p + (F 0 (p))−1 y − f (p)

Therefore, F −1 (y) ≈ p + (F 0 (p))−1 y − f (p) for y near F (p). Apparently the condition to find a

local inverse for a mapping on Rn is that the derivative matrix is nonsingular1 in some neighbor-
hood of the point. Experience has taught us from the one-dimensional case that we must insist the
derivative is continuous near the point in order to maintain the validity of the approximation.
Recall from calculus II that as we attempt to approximate a function with a power series it takes
an infinite series of power functions to recapture the formula exactly. Well, something similar is
true here. However, the method of approximation is through an iterative approximation procedure
which is built off the idea of Newton’s method. The product of this iteration is a nested sequence
of composite functions. To prove the theorem below one must actually provide proof the recur-
sively generated sequence of functions converges. See pages 160-187 of Edwards for an in-depth
exposition of the iterative approximation procedure. Then see pages 404-411 of Edwards for some
material on uniform convergence2 The main analytical tool which is used to prove the convergence
is called the contraction mapping principle. The proof of the principle is relatively easy to
follow and interestingly the main non-trivial step is an application of the geometric series. For
1
nonsingular matrices are also called invertible matrices and a convenient test is that A is invertible iff det(A) 6= 0.
2
actually that later chapter is part of why I chose Edwards’ text, he makes a point of proving things in Rn in such
a way that the proof naturally generalizes to function space. This is done by arguing with properties rather than
formulas. The properties offen extend to infinite dimensions whereas the formulas usually do not.
5.1. INVERSE FUNCTION THEOREM 119
the student of analysis this is an important topic which you should spend considerable time really
trying to absorb as deeply as possible. The contraction mapping is at the base of a number of
interesting and nontrivial theorems. Read Rosenlicht’s Introduction to Analysis for a broader and
better organized exposition of this analysis. In contrast, Edwards’ uses analysis as a tool to obtain
results for advanced calculus but his central goal is not a broad or well-framed treatment of analysis.
Consequently, if analysis is your interest then you really need to read something else in parallel to
get a better ideas about sequences of functions and uniform convergence. I have some notes from
a series of conversations with a student about Rosenlicht, I’ll post those for the interested student.
These notes focus on the part of the material I require for this course. This is Theorem 3.3 on page
185 of Edwards’ text:
Theorem 5.1.2. ( inverse function theorem )
Suppose F : Rn → Rn is continuously differentiable in an open set W containing a and the

derivative matrix F 0 (a) is invertible. Then F is locally invertible at a. This means that
there exists an open set U ⊆ W containing a and V a open set containing b = F (a) and
a one-one, continuously differentiable mapping G : V → W such that G(F (x)) = x for all
x ∈ U and F (G(y)) = y for all y ∈ V . Moreover, the local inverse G can be obtained as the
limit of the sequence of successive approximations defined by
Go (y) = a and Gn+1 (y) = Gn (y) − [F 0 (a)]−1 [F (Gn (y)) − y] for all y ∈ V .
The qualifier local is important to note. If we seek a global inverse then other ideas are needed.
If the function is everywhere injective then logically F (x) = y defines F −1 (y) = x and F −1 so
constructed in single-valued by virtue of the injectivity of F . However, for differentiable mappings,
one might wonder how can the criteria of global injectivity be tested via the differential. Even in
the one-dimensional case a vanishing derivative does not indicate a lack of injectivity; f (x) = x3
√
has f −1 (y) = 3 y and yet f 0 (0) = 0 (therefore f 0 (0) is not invertible). One the other hand, we’ll see
in the examples that follow that even if the derivative is invertible over a set it is possible for the
values of the mapping to double-up and once that happens we cannot find a single-valued inverse
function3
Remark 5.1.3. James R. Munkres’ Analysis on Manifolds good for a different proof.
Another good place to read the inverse function theorem is in James R. Munkres Analysis
on Manifolds. That text is careful and has rather complete arguments which are not entirely
the same as the ones given in Edwards. Munkres’ text does not use the contraction mapping
principle, instead the arguments are more topological in nature.
3
there are scientists and engineers who work with multiply-valued functions with great success, however, as a point
of style if nothing else, we try to use functions in math.
To give some idea of what I mean by topological let be give an example of such an argument.
Suppose F : Rn → Rn is continuously differentiable and F 0 (p) is invertible. Here’s a sketch of the
argument that F 0 (x) is invertible for all x near p as follows:
1. the function g : Rn → R defined by g(x) = det(F 0 (x)) is formed by a multinomial in the
component functions of F 0 (x). This function is clearly continuous since we are given that the
partial derivatives of the component functions of F are all continuous.
2. note we are given F 0 (p) is invertible and hence det(F 0 (p)) 6= 0 thus the continuous function g
is nonzero at p. It follows there is some open set U containing p for which 0 ∈ / g(U )
3. we have det(F 0 (x)) 6= 0 for all x ∈ U hence F 0 (x) is invertible on U .

I would argue this is a topological argument because the key idea here is the continuity of g.
Topology is the study of continuity in general.
Remark 5.1.4. James J. Callahan’s Advanced Calculus: a Geometric View, good reading.
James J. Callahan’s Advanced Calculus: a Geometric View has great merit in both visual-
ization and well-thought use of linear algebraic techniques. In addition, many student will
enjoy his staggered proofs where he first shows the proof for a simple low dimensional case
and then proceeds to the general case.
Example 5.1.5. Suppose F (x, y) = sin(y) + 1, sin(x) + 2 for (x, y) ∈ R2 . Clearly F is contin-

uously differentiable as all its component functions have continuous partial derivatives. Observe,

0 0 cos(y)
F (x, y) = [ ∂x F | ∂y F ] =
cos(x) 0
Hence F 0 (x, y) is invertible at points (x, y) such that det(F 0 (x, y)) = − cos(x) cos(y) 6= 0. This
means we may not be able to find local inverses at points (x, y) with x = 21 (2n + 1)π or y =
1 0
2 (2m + 1)π for some m, n ∈ Z. Points where F (x, y) are singular are points where one or both
of sin(y) and sin(x) reach extreme values thus the points where the Jacobian matrix are singular
are in fact points where we cannot find a local inverse. Why? Because the function is clearly not
1-1 on any set which contains the points of singularity for dF . Continuing, recall from precalculus
that sine has a standard inverse on [−π/2, π/2]. Suppose (x, y) ∈ [−π/2, π/2]2 and seek to solve
F (x, y) = (a, b) for (x, y):
y = sin−1 (a − 1)

sin(y) + 1 a sin(y) + 1 = a
F (x, y) = = ⇒ ⇒
sin(x) + 2 b sin(x) + 2 = b x = sin−1 (b − 2)
It follows that F −1 (a, b) = sin−1 (b − 2), sin−1 (a − 1) for (a, b) ∈ [0, 2] × [1, 3] where you should

note F ([−π/2, π/2]2 ) = [0, 2] × [1, 3]. We’ve found a local inverse for F on the region [−π/2, π/2]2 .
In other words, we just found a global inverse for the restriction of F to [−π/2, π/2]2 . Technically
we ought not write F −1 , to be more precise we should write:
(F |[−π/2,π/2]2 )−1 (a, b) = sin−1 (b − 2), sin−1 (a − 1) .

5.1. INVERSE FUNCTION THEOREM 121
It is customary to avoid such detail in many contexts. Inverse functions for sine, cosine, tangent
etc... are good examples of this slight of langauge.
A coordinate system on Rn is an invertible mapping of Rn to Rn . However, in practice the term

coordinate system is used with less rigor. Often a coordinate system has various degeneracies. For
example, in polar coordinates you could say θ = π/4 or θ = 9π/4 or generally θ = 2πk + π/4 for
any k ∈ Z. Let’s examine polar coordinates in view of the inverse function theorem.

Example 5.1.6. Let T (r, θ) = r cos(θ), r sin(θ) for (r, θ) ∈ [0, ∞) × (−π/2, π/2). Clearly
T is continuously differentiable as all its component functions have continuous partial derivatives.
To find the inverse we seek to solve T (r, θ) = (x, y) for (r, θ). Hence, consider x = r cos(θ) and
y = r sin(θ). Note that
x2 + y 2 = r2 cos2 (θ) + r2 sin2 (θ) = r2 (cos2 (θ) + sin2 (θ)) = r2
and
y r sin(θ)
= = tan(θ).
x r cos(θ)
p
It follows that r = x2 + y 2 and θ = tan−1 (y/x) for (x, y) ∈ (0, ∞) × R. We find
p
−1 2 2 −1
T (x, y) = x + y , tan (y/x) .
Let’s see how the derivative fits with our results. Calcuate,

0 cos(θ) −r sin(θ)
T (r, θ) = [ ∂r T | ∂θ T ] =
sin(θ) r cos(θ)
note that det(T 0 (r, θ)) = r hence we the inverse function theorem provides the existence of a local
inverse around any point except the origin. Notice the derivative does not detect the defect in the
angular coordinate. Challenge, find the inverse function for T (r, θ) = r cos(θ), r sin(θ) with
dom(T ) = [0, ∞) × (π/2, 3π/2). Or, find the inverse for polar coordinates in a neighborhood of
(0, −1).
Example 5.1.7. Suppose T : R3 → R3 is defined by T (x, y, z) = (ax, by, cz) for constants a, b, c ∈
R where abc 6= 0. Clearly T is continuously differentiable as all its component functions have
continuous partial derivatives. We calculate T 0 (x, y, z) = [∂x T |∂y T |∂z T ] = [ae1 |be2 |ce3 ]. Thus
det(T 0 (x, y, z)) = abc 6= 0 for all (x, y, z) ∈ R3 hence this function is locally invertible everywhere.
Moreover, we calculate the inverse mapping by solving T (x, y, z) = (u, v, w) for (x, y, z):
(ax, by, cz) = (u, v, w) ⇒ (x, y, z) = (u/a, v/b, w/c) ⇒ T −1 (u, v, w) = (u/a, v/b, w/c).
Example 5.1.8. Suppose F : Rn → Rn is defined by F (x) = Ax+b for some matrix A ∈ R n×n and
vector b ∈ Rn . Under what conditions is such a function invertible ?. Since the formula for
this function gives each component function as a polynomial in the n-variables we can conclude the
function is continuously differentiable. You can calculate that F 0 (x) = A. It follows that a sufficient
condition for local inversion is det(A) 6= 0. It turns out that this is also a necessary condition as
det(A) = 0 implies the matrix A has nontrivial solutions for Av = 0. We say v ∈ N ull(A) iff
Av = 0. Note if v ∈ N ull(A) then F (v) = Av + b = b. This is not a problem when det(A) 6= 0
for in that case the null space is contains just zero; N ull(A) = {0}. However, when det(A) = 0 we
learn in linear algebra that N ull(A) contains infinitely many vectors so F is far from injective. For
example, suppose N ull(A) = span{e1 } then you can show that F (a1 , a2 , . . . , an ) = F (x, a2 , . . . , an )
for all x ∈ R. Hence any point will have other points nearby which output the same value under F .
Suppose det(A) 6= 0, to calculate the inverse mapping formula we should solve F (x) = y for x,
y = Ax + b ⇒ x = A−1 (y − b) ⇒ F −1 (y) = A−1 (y − b).
Remark 5.1.9. inverse function theorem holds for higher derivatives.
In Munkres the inverse function theorem is given for r-times differentiable functions. In
short, a C r function with invertible differential at a point has a C r inverse function local
to the point. Edwards also has arguments for r > 1, see page 202 and arguments and
surrounding arguments.
5.2 implicit function theorem

Consider the problem of solving x2 + y 2 = 1 for y as a function of x.
p
x2 + y 2 = 1 ⇒ y 2 = 1 − x2 ⇒ y = ± 1 − x2 .
A function cannot have two outputs for a single input, when we write ± in the expression above
it simply indicates our ignorance as to which is chosen. Once further information is given then we
may be able to choose a + or a −. For example:
√
1. if x2 + y 2 = 1 and we want to solve for y near (0, 1) then y = 1 − x2 is the correct choice
since y > 0 at the point of interest.
√
2. if x2 + y 2 = 1 and we want to solve for y near (0, −1) then y = − 1 − x2 is the correct choice
since y < 0 at the point of interest.
3. if x2 + y 2 = 1 and we want to solve for y near (1, 0) then it’s impossible to find a single
function which reproduces x2 + y 2 = 1 on an open disk centered at (1, 0).
What is the defect of case (3.) ? The trouble is that no matter how close we zoom in to the point
there are always two y-values for each given x-value. Geometrically, this suggests either we have a
discontinuity, a kink, or a vertical tangent in the graph. The given problem has a vertical tangent
and hopefully you can picture this with ease since its just the unit-circle. In calculus I we studied
5.2. IMPLICIT FUNCTION THEOREM 123
implicit differentiation, our starting point was to assume y = y(x) and then we differentiated
equations to work out implicit formulas for dy/dx. Take the unit-circle and differentiate both sides,
dy dy x
x2 + y 2 = 1 ⇒ 2x + 2y =0 ⇒ =− .
dx dx y
dy
Note dx is not defined for y = 0. It’s no accident that those two points (−1, 0) and (1, 0) are
precisely the points at which we cannot solve for y as a function of x. Apparently, the singularity
in the derivative indicates where we may have trouble solving an equation for one variable as a
function of the remaining variable.
We wish to study this problem in general. Given n-equations in (m+n)-unknowns when can we solve
for the last n-variables as functions of the first m-variables ? Given a continuously differentiable
mapping G = (G1 , G2 , . . . , Gn ) : Rm × Rn → Rn study the level set: (here k1 , k2 , . . . , kn are
constants)
G1 (x1 , . . . , xm , y1 , . . . , yn ) = k1
G2 (x1 , . . . , xm , y1 , . . . , yn ) = k2
..
.
Gn (x1 , . . . , xm , y1 , . . . , yn ) = kn
We wish to locally solve for y1 , . . . , yn as functions of x1 , . . . xm . That is, find a mapping h : Rm →

Rn such that G(x, y) = k iff y = h(x) near some point (a, b) ∈ Rm × Rn such that G(a, b) = k. In
this section we use the notation x = (x1 , x2 , . . . xm ) and y = (y1 , y2 , . . . , yn ).
Before we turn to the general problem let’s analyze the unit-circle problem in this notation. We
are given G(x, y) = x2 + y 2 and we wish to find f (x) such that y = f (x) solves G(x, y) = 1.
Differentiate with respect to x and use the chain-rule:
∂G dx ∂G dy
+ =0
∂x dx ∂y dx
We find that dy/dx = −Gx /Gy = −x/y. Given this analysis we should suspect that if we are
given some level curve G(x, y) = k then we may be able to solve for y as a function of x near p
if G(p) = k and Gy (p) 6= 0. This suspicion is valid and it is one of the many consequences of the
implicit function theorem.
We again turn to the linearization approximation. Suppose G(x, y) = k where x ∈ Rm and y ∈ Rn

and suppose G : Rm × Rn → Rn is continuously differentiable. Suppose (a, b) ∈ Rm × Rn has
G(a, b) = k. Replace G with its linearization based at (a, b):
G(x, y) ≈ k + G0 (a, b)(x − a, y − b)

here we have the matrix multiplication of the n × (m + n) matrix G0 (a, b) with the (m + n) × 1
column vector (x − a, y − b) to yield an n-component column vector. It is convenient to define
partial derivatives with respect to a whole vector of variables,
∂G1 ∂G1 ∂G1 ∂G1
···
   
∂x1 ··· ∂xm ∂y1 ∂yn
∂G  .. .. ∂G  .. ..
= =
 
∂x . . 
∂y . . 
∂Gn ∂Gn ∂Gn ∂Gn
∂x1 ··· ∂xm ∂y1 ··· ∂yn
In this notation we can write the n × (m + n) matrix G0 (a, b) as the concatenation of the n × m
matrix ∂G ∂G
∂x (a, b) and the n × n matrix ∂y (a, b)

0 ∂G ∂G
G (a, b) = (a, b)
(a, b)
∂x ∂y
Therefore, for points close to (a, b) we have:
∂G ∂G
G(x, y) ≈ k + (a, b)(x − a) + (a, b)(y − b)
∂x ∂y
The nonlinear problem G(x, y) = k has been (locally) replaced by the linear problem of solving
what follows for y:
∂G ∂G
k≈k+ (a, b)(x − a) + (a, b)(y − b) (5.1)
∂x ∂y
Suppose the square matrix ∂G ∂y (a, b) is invertible at (a, b) then we find the following approximation
for the implicit solution of G(x, y) = k for y as a function of x:
−1
∂G ∂G
y =b− (a, b) (a, b)(x − a) .
∂y ∂x
Of course this is not a formal proof, but it does suggest that det ∂G

∂y (a, b) 6= 0 is a necessary
condition for solving for the y variables.
As before suppose G : Rm × Rn → Rn . Suppose we have a continuously differentiable function

h : Rm → Rn such that h(a) = b and G(x, h(x)) = k. We seek to find the derivative of h in terms
of the derivative of G. This is a generalization of the implicit differentiation calculation we perform
in calculus I. I’m including this to help you understand the notation a bit more before I state the
implicit function theorem. Differentiate with respect to xl for l ∈ Nm :
Xm n n
∂ ∂G ∂xi X ∂G ∂hj ∂G X ∂G ∂hj
G(x, h(x)) = + = + =0
∂xl ∂xi ∂xl ∂yj ∂xl ∂xl ∂yj ∂xl
i=1 j=1 j=1
∂xi
we made use of the identity ∂x k
= δik to squash the sum of i to the single nontrivial term and the
∂
zero on the r.h.s follows from the fact that ∂x l
(k) = 0. Concatenate these derivatives from k = 1
up to k = m:
n n n
∂G X ∂G ∂hj ∂G X ∂G ∂hj ∂G X ∂G ∂hj
+ + · · · + = [0|0| · · · |0]
∂x1 ∂yj ∂x1 ∂x2 ∂yj ∂x2 ∂xm ∂yj ∂xm
j=1 j=1 j=1
Properties of matrix addition allow us to parse the expression above as follows:

n
X n n
∂G ∂G ∂G ∂G ∂hj X ∂G ∂hj X ∂G ∂hj
· · · + · · · = [0|0| · · · |0]
∂x1 ∂x2 ∂xm ∂yj ∂x1 ∂yj ∂x2 ∂yj ∂xm
j=1 j=1 j=1
But, this reduces to

∂G ∂G ∂h ∂G ∂h ∂G ∂h m×n
+ ···
=0∈R
∂x ∂y ∂x1 ∂y ∂x2 ∂y ∂xm
The concatenation property of matrix multiplication states [Ab1 |Ab2 | · · · |Abm ] = A[b1 |b2 | · · · |bm ]
we use this to write the expression once more,
∂G −1 ∂G

∂G ∂G ∂h ∂h ∂h ∂G ∂G ∂h ∂h
+ ···
=0 ⇒ + =0 ⇒ =−
∂x ∂y ∂x1 ∂x2 ∂xm ∂x ∂y ∂x ∂x ∂y ∂x
∂G
where in the last implication we made use of the assumption that ∂y is invertible.
Theorem 5.2.1. (Theorem 3.4 in Edwards’s Text see pg 190)
Let G : dom(G) ⊆ Rm × Rn → Rn be continuously differentiable in a open ball about the

point (a, b) where G(a, b) = k (a constant vector in Rn ). If the matrix ∂G
∂y (a, b) is invertible
then there exists an open ball U containing a in Rm and an open ball W containing (a, b)
in Rm × Rn and a continuously differentiable mapping h : U → Rn such that G(x, y) = k
iff y = h(x) for all (x, y) ∈ W . Moreover, the mapping h is the limit of the sequence of
successive approximations defined inductively below
−1
ho (x) = b, hn+1 = hn (x) − [ ∂G
∂y (a, b)] G(x, hn (x)) for all x ∈ U .
We will not attempt a proof of the last sentence for the same reasons we did not pursue the details
in the inverse function theorem. However, we have already derived the first step in the iteration in
our study of the linearization solution.
Proof: Let G : dom(G) ⊆ Rm × Rn → Rn be continuously differentiable in a open ball B about

the point (a, b) where G(a, b) = k (k ∈ Rn a constant). Furthermore, assume the matrix ∂G ∂y (a, b)
is invertible. We seek to use the inverse function theorem to prove the implicit function theorem.
Towards that end consider F : Rm × Rn → Rm × Rn defined by F (x, y) = (x, G(x, y)). To begin,
observe that F is continuously differentiable in the open ball B which is centered at (a, b) since
G and x have continuous partials of their components in B. Next, calculate the derivative of
F = (x, G),
0 ∂x x ∂y x Im 0m×n
F (x, y) = [∂x F |∂y F ] = =
∂x G ∂y G ∂x G ∂y G
The determinant of the matrix above is the product of the deteminant of the blocks Im and
∂y G; det(F 0 (x, y) = det(Im )det(∂y G) = ∂y G. We are given that ∂G ∂y (a, b) is invertible and hence
det( ∂y (a, b)) 6= 0 thus det(F 0 (x, y) 6= 0 and we find F 0 (a, b) is invertible. Consequently, the inverse
∂G
function theorem applies to the function F at (a, b). Therefore, there exists F −1 : V ⊆ Rm × Rn →
U ⊆ Rm × Rn such that F −1 is continuously differentiable. Note (a, b) ∈ U and V contains the
point F (a, b) = (a, G(a, b)) = (a, k).
Our goal is to find the implicit solution of G(x, y) = k. We know that
F −1 (F (x, y)) = (x, y) and F (F −1 (u, v)) = (u, v)
for all (x, y) ∈ U and (u, v) ∈ V . As usual to find the formula for the inverse we can solve
F (x, y) = (u, v) for (x, y) this means we wish to solve (x, G(x, y)) = (u, v) hence x = u. The
formula for v is more elusive, but we know it exists by the inverse function theorem. Let’s say
y = H(u, v) where H : V → Rn and thus F −1 (u, v) = (u, H(u, v)). Consider then,
(u, v) = F (F −1 (u, v) = F (u, H(u, v)) = (u, G(u, H(u, v))
Let v = k thus (u, k) = (u, G(u, H(u, k)) for all (u, v) ∈ V . Finally, define h(u) = H(u, k) for
all (u, k) ∈ V and note that k = G(u, h(u)). In particular, (a, k) ∈ V and at that point we find
h(a) = H(a, k) = b by construction. It follows that y = h(x) provides a continuously differentiable
solution of G(x, y) = k near (a, b).
Uniqueness of the solution follows from the uniqueness for the limit of the sequence of functions
described in Edwards’ text on page 192. However, other arguments for uniqueness can be offered,
independent of the iterative method, for instance: see page 75 of Munkres Analysis on Manifolds.
Remark 5.2.2. notation and the implementation of the implicit function theorem.
We assumed the variables y were to be written as functions of x variables to make explicit

a local solution to the equation G(x, y) = k. This ordering of the variables is convenient to
argue the proof, however the real theorem is far more general. We can select any subset of n
input variables to make up the ”y” so long as ∂G∂y is invertible. I will use this generalization
of the formal theorem in the applications that follow. Moreover, the notations x and y are
unlikely to maintain the same interpretation as in the previous pages. Finally, we will for
convenience make use of the notation y = y(x) to express the existence of a function f such
that y = f (x) when appropriate. Also, z = z(x, y) means there is some function h for which
z = h(x, y). If this notation confuses then invent names for the functions in your problem.
Example 5.2.3. Suppose G(x, y, z) = x2 + y 2 + z 2 . Suppose we are given a point (a, b, c) such
that G(a, b, c) = R2 for a constant R. Problem: For which variable can we solve? What, if
any, influence does the given point have on our answer? Solution: to begin, we have one
equation and three unknowns so we should expect to find one of the variables as functions of the
remaining two variables. The implicit function theorem applies as G is continuously differentiable.
1. if we wish to solve z = z(x, y) then we need Gz (a, b, c) = 2c 6= 0.
2. if we wish to solve y = y(x, z) then we need Gy (a, b, c) = 2b 6= 0.
3. if we wish to solve x = x(y, z) then we need Gx (a, b, c) = 2a 6= 0.
The point has no local solution for z if it is a point on the intersection of the xy-plane and the
sphere G(x, y, z) = R2 . Likewise, we cannot solve for y = y(x, z) on the y = 0 slice of the sphere
and we cannot solve for x = x(y, z) on the x = 0 slice of the sphere.
Notice, algebra verifies the conclusions we reached via the implicit function theorem:
p p p
z = ± R 2 − x2 − y 2 y = ± R 2 − x2 − z 2 x = ± R2 − y 2 − z 2
When we are at zero for one of the coordinates then we cannot choose + or − since we need both on
an open ball intersected with the sphere centered at such a point4 . Remember, when I talk about
local solutions I mean solutions which exist over the intersection of the solution set and an open
ball in the ambient space (R3 in this context). The preceding example is the natural extension of
the unit-circle example to R3 . A similar result is available for the n-sphere in Rn . I hope you get
the point of the example, if we have one equation then if we wish to solve for a particular variable in
terms of the remaining variables then all we need is continuous differentiability of the level function
and a nonzero partial derivative at the point where we wish to find the solution. Now, the implicit
function theorem doesn’t find the solution for us, but it does provide the existence. In the section
on implicit differentiation, existence is really all we need since focus our attention on rates of change
rather than actually solutions to the level set equation.
Example 5.2.4. Consider the equation exy + z 3 − xyz = 2. Can we solve this equation for
z = z(x, y) near (0, 0, 1)? Let G(x, y, z) = exy + z 3 − xyz and note G(0, 0, 1) = e0 + 1 + 0 = 2 hence
(0, 0, 1) is a point on the solution set G(x, y, z) = 2. Note G is clearly continuously differentiable
and
Gz (x, y, z) = 3z 2 − xy ⇒ Gz (0, 0, 1) = 3 6= 0
therefore, there exists a continuously differentiable function h : dom(h) ⊆ R2 → R which solves
G(x, y, h(x, y)) = 2 for (x, y) near (0, 0) and h(0, 0) = 1.
I’ll not attempt an explicit solution for the last example.

4
if you consider G(x, y, z) = R2 as a space then the open sets on the space are taken to be the intersection with
the space and open balls in R3 . This is called the subspace topology in topology courses.
Example 5.2.5. Let (x, y, z) ∈ S iff x + y + z = 2 and y + z = 1. Problem: For which

variable(s) can we solve? Solution: define G(x, y, z) = (x + y + z, y + z) we wish to study
G(x, y, z) = (2, 1). Notice the solution set is not empty since G(1, 0, 1) = (1 + 0 + 1, 0 + 1) = (2, 1)
Moreover, G is continuously differentiable. In this case we have two equations and three unknowns
so we expect two variables can be written in terms of the remaining free variable. Let’s examine
the derivative of G:
0 1 1 1
G (x, y, z) =
0 1 1
Suppose we wish to solve x = x(z) and y = y(z) then we should check invertiblility of5

∂G 1 1
= .
∂(x, y) 0 1
The matrix above is invertible hence the implicit function theorem applies and we can solve for x
and y as functions of z. On the other hand, if we tried to solve for y = y(x) and z = z(x) then
we’ll get no help from the implicit function theorem as the matrix

∂G 1 1
= .
∂(y, z) 1 1
is not invertible. Geometrically, we can understand these results from noting that G(x, y, z) = (2, 1)
is the intersection of the plane x + y + z = 2 and y + z = 1. Substituting y + z = 1 into x + y + z = 2
yields x + 1 = 2 hence x = 1 on the line of intersection. We can hardly use x as a free variable for
the solution when the problem fixes x from the outset.
The method I just used to analyze the equations in the preceding example was a bit adhoc. In
linear algebra we do much better for systems of linear equations. A procedure called Gaussian
elimination naturally reduces a system of equations to a form in which it is manifestly obvious how
to eliminate redundant variables in terms of a minimal set of basic free variables. The ”y” of the
implicit function proof discussions plays the role of the so-called pivotal variables whereas the
”x” plays the role of the remaining free variables. These variables are generally intermingled in
the list of total variables so to reproduce the pattern assumed for the implicit function theorem
we would need to relabel variables from the outset of a calculation. In the following example, I
show how reordering the variables allows us to solve for various pairs. In short, put the dependent
variable first and the independent variables second so the Gaussian elimination shows the solution
with minimal effort. Here’s how:
5 ∂(x,y)
this notation should not be confused with ∂(u,v) which is used to denote a particular determinant associated
with coordinate change of integrals, or pull-back of a differential form as explained on page 100 of H.M Edward’s
Advanced Calculus: A differential Forms Approach, we should discuss it in a later chapter.
Example 5.2.6. Consider G(x, y, u, v) = (3x + 2y − u, 2x + y − v) = (−1, 3). We have two

equations with four variables. Let’s investigate which pairs of variables can be taken as independent
or dependent variables. The most efficient method to dispatch these questions is probably Gaussian
elimination. I leave it to the reader to verify that:

3 2 −1 0 −1 1 0 1 −2 7
rref =
2 1 0 −1 3 0 1 −2 3 −11
We can immediately read from the result above that x, y can be taken to depend on u, v via the
formulas:
x = −u + 2v + 7, y = 2u − 3v − 11
On the other hand, if we order the variables (u, v, x, y) then Gaussian elimination gives:

−1 0 3 2 −1 1 0 −3 −2 1
rref =
0 −1 2 1 3 0 1 −2 −1 −3
Therefore, we find u(x, y) and v(x, y) as follows:
u = 3x + 2y + 1, v = 2x + y − 3.
To solve for x, u as functions of y, v consider:

3 −1 2 0 −1 1 0 1/2 −1/2 3/2
rref =
2 0 1 −1 3 0 1 −1/2 −3/2 11/2
From which we can read,
x = −y/2 + v/2 + 3/2, u = y/2 + 3v/2 + 11/2.
I could solve the problem below in the efficient style above, but I will instead follow the method in
which we discussed in the paragraphs surrounding Equation 5.1. In contrast to the general case,
because the problem is linear the solution of Equation 5.1 is also a solution of the actual problem.
Example 5.2.7. Solve the following system of equations near (1,2,3,4,5).

   
x + y + z + 2a + 2b 24
G(x, y, z, a, b) =  x + 0 + 2z + 2a + 3b  =  30 
3x + 2y + z + 3a + 4b 42
Differentiate to find the Jacobian:

 
1 1 1 2 2
G0 (x, y, z, a, b) =  1 0 2 2 3 
3 2 1 3 4
Let us solve G(x, y, z, a, b) = (24, 30, 42) for x(a, b), y(a, b), z(a, b) by the method of Equation 5.1.
I’ll omit the point-dependence of the Jacobian since it clearly has none.
   
24 x−1
∂G ∂G a−4
G(x, y, z, a, b) = 30 +
   y−2 + 
∂(x, y, z) ∂(a, b) b − 5
42 z−3
Let me make the notational chimera above explicit:
      
24 1 1 1 x−1 2 2
a−4
G(x, y, z, a, b) = 30 + 1 0 2
     y−2 + 2 3
  
b−5
42 3 2 1 z−3 3 4
To solve G(x, y, z, a, b) = (24, 30, 42) for (x, y, z) we may use the expression above. After a little
calculation one finds:
 −1  
1 1 1 −4 1 2
1
 1 0 2  =  5 −2 −1 
3
3 2 1 2 1 −1
The constant term cancels and we find:
    
x−1 −4 1 2 2 2
 y − 2  = − 1  5 −2 −1   2 3  a − 4
3 b−5
z−3 2 1 −1 3 4
Multiplying the matrices gives:
       
x−1 0 3 0 −1 5−b
1
 y−2 =−  3 0  a − 4 a − 4
=  −1 0  = 4−a 
3 b−5 b−5
z−3 3 3 −1 −1 9−a−b
Therefore,
x = 6 − b, y = 6 − a, z = 12 − a − b.
Is it possible to solve for any triple of the variables x, y, z, a, b for the given system? In fact,
no. Let me explain by linear algebra. We can calculate: the augmented coefficient matrix for
G(x, y, z, a, b) = (24, 30, 42) Gaussian eliminates as follows:
   
1 1 1 2 2 24 1 0 0 0 1 6
rref  1 0 2 2 3 30  =  0 1 0 1 0 6  .
3 2 1 3 4 42 0 0 1 1 1 12
First, note this is consistent with the answer we derived above. Second, examine the columns of
rref [G0 ]. You can ignore the 6-th column in the interest of this thought extending to nonlinear
systems. The question of the suitability of a triple amounts to the invertibility of the submatrix of
G0 which corresponds to the triple. Examine:
   
1 1 2 1 1 2
∂G ∂G
=  0 2 2 , = 1 2 3 
∂(y, z, a) ∂(x, z, b)
2 1 3 3 1 4
both of these are clearly singular since the third column is the sum of the first two columns. Alter-
natively, you can calculate the determinant of each of the matrices above is zero. In contrast,
 
1 2 2
∂G
= 2 2 2 
∂(z, a, b)
1 3 4
is non-singular. How to I know there is no linear dependence? Well, we could calculate the de-
terminant is 1(8 − 6) − 2(8 − 2) + 2(6 − 2) = −2 6= 0. Or, we could examine the row reduction
above. The column correspondance property6 states that linear dependences amongst columns of a
matrix are preserved under row reduction. This means we can easily deduce dependence (if there
is any) from the reduced matrix. Observe that column 4 is clearly the sum of columns 2 and 3.
Likewise, column 5 is the sum of columns 1 and 3. On the other hand, columns 3, 4, 5 admit no
linear dependence. In general, more calculation would be required to ”see” the independence of the
far right columns. One reorders the columns and performs a new reduction to ascertain dependence.
No such calculation is needed here since the problem is not that complicated.
I find calculating the determinant of sub-Jacobian matrices is the simplest way for most students
to quickly understand. I’ll showcase this method in a series of examples attached to a later section.
I have made use of some matrix theory in this section. If you didn’t learn it in linear (or haven’t
taken linear yet) it’s worth learning. These are nice tools to keep for later problems in life.
Remark 5.2.8. independent constraints
Gaussian elimination on a system of linear equations

mayproduce
a rowof zeros. For
1 1 0 1 1 0
example, x + y = 0 and 2x + 2y = 0 gives rref = . The reason
2 2 0 0 0 0
for this is quite obvious: the equations consider are not indpendent. In fact the second
equation is a scalar multiple of the first. Generally, if there is some linear-dependence
in a set of equations then we can expect this will happen. Although, if the equations
are inhomogenous the last column might not be trivial because the system could be
inconsistent (for example x + y = 1 and 2x + 2y = 5).
Consider G : Rn × Rp → Rn . As we linearize G = k we arrive at a homogeneous system

which can be written briefly as G0~r = 0 (think about Equation 5.1 with the k’s cancelled).
We should define G(~r) = k is a system of n independent equations at ~ro iff G(~ro ) = k
and rref[G0 (~ro )] has zero row. In other terminology, we could say the system of (possibly
nonlinear) equations G(~r) = k is built from n-independent equations near ~ro iff the Jacobian
matrix has full-rank at ~ro . If this full-rank condition is met then we can solve for n of the
variables in terms of the remaining p variables. In general there will be many choices of
how to do this, and some choices will be forbidden as we have seen in the examples already.
6
I like to call it the CCP in my linear notes
5.3 implicit differentiation

Enough theory, let’s calculate. In this section I apply previous theoretical constructions to specific
problems. I also introduce standard notation for ”constrained” partial differentiation which is
also sometimes called ”partial differentiation with a side condition”. The typical problem is the
following: given equations:
G1 (x1 , . . . , xm , y1 , . . . , yn ) = k1
G2 (x1 , . . . , xm , y1 , . . . , yn ) = k2
..
.
Gn (x1 , . . . , xm , y1 , . . . , yn ) = kn
calculate partial derivative of dependent variables with respect to independent variables. Contin-
uing with the notation of the implicit function discussion we’ll assume that y will be dependent
on x. I want to recast some of our arguments via differentials7 . Take the total differential of each
equation above,
dG1 (x1 , . . . , xm , y1 , . . . , yn ) = 0
dG2 (x1 , . . . , xm , y1 , . . . , yn ) = 0
..
.
dGn (x1 , . . . , xm , y1 , . . . , yn ) = 0
Hence,
∂x1 G1 dx1 + · · · + ∂xm G1 dxm + ∂y1 G1 dy1 + · · · + ∂yn G1 dyn = 0

∂x1 G2 dx1 + · · · + ∂xm G2 dxm + ∂y1 G2 dy1 + · · · + ∂yn G2 dyn = 0
..
.
∂x1 Gn dx1 + · · · + ∂xm Gn dxm + ∂y1 Gn dy1 + · · · + ∂yn Gn dyn = 0
Notice, this can be nicely written in column vector notation as:
∂x1 Gdx1 + · · · + ∂xm Gdxm + ∂y1 Gdy1 + · · · + ∂yn Gdyn = 0
Or, in matrix notation:

   
dx1 dy1
[∂x1 G| · · · |∂xm G]  ...  + [∂y1 G| · · · |∂yn G]  ...  = 0
   
dxm dyn
7
in contrast, In the previous section we mostly used derivative notation
5.3. IMPLICIT DIFFERENTIATION 133
Finally, solve for dy, we assume [∂y1 G| · · · |∂yn G]−1 exists,

   
dy1 dx1
 ..  −1  . 
 .  = −[∂y1 G| · · · |∂yn G] [∂x1 G| · · · |∂xm G]  .. 
dyn dxm
∂yi
Given all of this we can calculate ∂x j
by simply reading the coeffient dxj in the i-th row. I will
make this idea quite explicit in the examples that follow.
Example 5.3.1. Let’s return to a common calculus III problem. Suppose F (x, y, z) = k for some
constant k. Find partial derivatives of x, y or z with repsect to the remaining variables.
Solution: I’ll use the method of differentials once more:
dF = Fx dx + Fy dy + Fz dz = 0
We can solve for dx, dy or dz provided Fx , Fy or Fz is nonzero respective and these differential
expressions reveal various partial derivatives of interest:
Fy Fz ∂x Fy ∂x Fz
dx = − dy − dz ⇒ =− & =−
Fx Fx ∂y Fx ∂z Fx
Fx Fz ∂y Fx ∂y Fz
dy = − dx − dz ⇒ =− & =−
Fy Fy ∂x Fy ∂z Fy
Fx Fy ∂z Fx ∂z Fy
dz = − dx − dy ⇒ =− & =−
Fz Fz ∂x Fz ∂y Fz
In each case above, the implicit function theorem allows us to solve for one variable in terms of the
remaining two. If the partial derivative of F in the denominator are zero then the implicit function
theorem does not apply and other thoughts are required. Often calculus text give the following as a
homework problem:
∂x ∂y ∂z Fy Fz Fx
=− = −1.
∂y ∂z ∂x Fx Fy Fz
In the equation above we have x appear as a dependent variable on y, z and also as an independent
variable for the dependent variable z. These mixed expressions are actually of interest to engineering
and physics. The less mbiguous notation below helps better handle such expressions:

∂x ∂y ∂z
= −1.
∂y z ∂z x ∂x y
In each part of the expression we have clearly denoted which variables are taken to depend on the
others and in turn what sort of partial derivative we mean to indicate. Partial derivatives are not
taken alone, they must be done in concert with an understanding of the totality of the indpendent
variables for the problem. We hold all the remaining indpendent variables fixed as we take a partial
derivative.
The explicit independent variable notation is more important for problems where we can choose
more than one set of indpendent variables for a given dependent variables. In the example that
follows we study w =w(x, y) but we could just
as well consider w ∂w
= w(x, z). Generally it will not
be the case that ∂w ∂w

∂x y is the same as ∂x z . In calculation of ∂x y we hold y constant as we
∂w

vary x whereas in ∂x z we hold z constant as we vary x. There is no reason these ought to be the
same8 .
Example 5.3.2. Suppose x+y+z+w = 3 and x2 −2xyz+w3 = 5. Calculate partial derivatives
of z and w with respect to the independent variables x, y. Solution: we begin by calculation
of the differentials of both equations:
dx + dy + dz + dw = 0
(2x − 2yz)dx − 2xzdy − 2xydz + 3w2 dw = 0
We can solve for (dz, dw). In this calculation we can treat the differentials as formal variables.
dz + dw = −dx − dy
−2xydz + 3w2 dw = −(2x − 2yz)dx + 2xzdy
I find matrix notation is often helpful,

1 1 dz −dx − dy
=
−2xy 3w2 dw −(2x − 2yz)dx + 2xzdy
Use Kramer’s rule, multiplication by inverse, substitution, adding/subtracting equations etc... what-
ever technique of solving linear equations you prefer. Our goal is to solve for dz and dw in terms
of dx and dy. I’ll use Kramer’s rule this time:

−dx − dy 1
det
−(2x − 2yz)dx + 2xzdy 3w2 3w2 (−dx − dy) + (2x − 2yz)dx − 2xzdy
dz = =
3w2 + 2xy

1 1
det
−2xy 3w2
Collecting terms,
−3w2 + 2x − 2yz −3w2 − 2xz

dz = dx + dy
3w2 + 2xy 3w2 + 2xy
From the expression above we can read various implicit derivatives,
−3w2 + 2x − 2yz −3w2 − 2xz

∂z ∂z
= & =
∂x y 3w2 + 2xy ∂y x 3w2 + 2xy
The notation above indicates that z is understood to be a function of independent variables x, y.

∂z
∂x y means we take the derivative of z with respect to x while holding y fixed. The appearance
8
a good exercise would be to do the example over but instead aim to calculate partial derivatives for y, w with
respect to independent variables x, z
of the dependent variable w can be removed by using the equations G(x, y, z, w) = (3, 5). Similar
ambiguities exist for implicit differentiation in calculus I. Apply Kramer’s rule once more to solve
for dw:

1 −dx − dy
det
−2xy −(2x − 2yz)dx + 2xzdy −(2x − 2yz)dx + 2xzdy − 2xy(dx + dy)
dw = =
3w2 + 2xy

1 1
det 2
−2xy 3w
Collecting terms,
−2x + 2yz − 2xy 2xzdy − 2xydy
dw = dx + dy
3w2 + 2xy 3w2 + 2xy
We can read the following from the differential above:

∂w −2x + 2yz − 2xy ∂w 2xzdy − 2xydy
= & =
∂x y 3w2 + 2xy ∂y x 3w2 + 2xy
You should ask: where did we use the implicit function theorem in the preceding example? Notice
our underlying hope is that we can solve for z = z(x, y)
and w = w(x, y). The implicit function the-
∂G 1 1
orem states this is possible precisely when ∂(z,w) = is non singular. Interestingly
−2xy 3w2
this is the same matrix we must consider to isolate dz and dw. The calculations of the example
1 1
are only meaningful if the det 6= 0. In such a case the implicit function theorem
−2xy 3w2
applies and it is reasonable to suppose z, w can be written as functions of x, y.
5.3.1 computational techniques for partial differentiation with side conditions

In this section I show you how I teach this to calculus III. In other words, we set-aside the explicit
mention of the implicit function theorem and work out some typical calculations. If one desires
rigor then the answer is found from the implicit function theorems careful application, that is how
to justify what follows. These notes are taken from my calculus III notes, but I thought it wise
to include them here since most calculus texts do not bother to show these calculations (which
is sad since they actually matter to the application of multivariate analysis to many real world
applications) To begin, we define9 the total differential.
Definition 5.3.3.
∂f ∂f ∂f
If f = f (x1 , x2 , . . . , xn ) then df = ∂x1 dx1 + ∂x2 dx2 + ··· + ∂xn dxn .
9
I invite the reader to verify the notation ”defined” in this section is in fact totally sympatico with our previous
definitions
Example 5.3.4. Suppose E = pv + t2 then dE = vdp + pdv + 2tdt. In this example the dependent
variable is E whereas the independent variables are p, v and t.
Example 5.3.5. Problem: what are ∂F/∂x and ∂F/∂y if we know that F = F (x, y) and
dF = (x2 + y)dx − cos(xy)dy.
Solution: if F = F (x, y) then the total differential has the form dF = Fx dx + Fy dy. We simply
compare the general form to the given dF = (x2 + y)dx − cos(xy)dy to obtain:
∂F ∂F
= x2 + y, = − cos(xy).
∂x ∂y
Example 5.3.6. Suppose w = xyz then dw = yzdx + xzdy + xydz. On the other hand, we can
solve for z = z(x, y, w)
w w w 1
z= ⇒ dz = − 2
dx − 2 dy + dw. ?
xy x y xy xy
If we solve dw = yzdx + xzdy + xydz directly for dz we obtain:
z z 1
dz = − dx − dy + dw ? ?.
x y xy
w xyz z w xyz
Are ? and ?? consistent? Well, yes. Note x2 y
= x2 y
= x and xy 2
= xy 2
= yz .
Which variables are independent/dependent in the example above? It depends. In this initial
portion of the example we treated x, y, z as independent whereas w was dependent. But, in the
last half we treated x, y, w as independent and z was the dependent variable. Consider this, if I
∂z
ask you what the value of ∂x is in the example above then this question is ambiguous!
∂z ∂z −z
=0 verses =
∂x
| {z } | {z x}
∂x
z indpendent of x z depends on x
Obviously this sort of ambiguity is rather unpleasant. A natural solution to this trouble is simply
to write a bit more when variables are used in multiple contexts. In particular,

∂z ∂z −z
=0 is different than = .
∂x y,z
∂x y,w
x
| {z } | {z }
means x,y,z independent means x,y,w independent
The key concept is that all the other independent variables are held fixed as an indpendent variable
∂z
is partial differentiated. Holding y, z fixed as x varies means z does not change hence ∂x y,z
= 0.
On the other hand, if we hold y, w fixed as x varies then the change in z need not be trivial;
∂z −z
∂x y,w = x . Let me expand on how this notation interfaces with the total differential.
Definition 5.3.7.
If w, x, y, z are variables then

∂w ∂w ∂w
dw = dx + dy + dz.
∂x y,z ∂y x,z ∂z x,y
Alternatively,
∂x ∂x ∂x
dx = dw + dy + dz.
∂w y,z ∂y w,z ∂z w,y
The larger idea here is that we can identify partial derivatives from the coefficients in equations of
differentials. I’d say a differential equation but you might get the wrong idea... Incidentally, there
is a whole theory of solving differential equations by clever use of differentials, I have books if you
are interested.
Example 5.3.8. Suppose w = x + y + z and x + y = wz then calculate ∂w ∂w

∂x y and ∂x z . Notice we

must choose dependent and independent variables to make sense of partial derivatives in question.
1. suppose w, z both depend on x, y. Calculate,

∂w ∂ ∂x ∂y ∂z ∂z
= (x + y + z) = + + =1+0+ ?
∂x y ∂x y ∂x y ∂x y ∂x y ∂x y
To calculate further we need to eliminate w by substituting w = x + y + z into x + y = wz;

thus x + y = (x + y + z)z hence dx + dy = (dx + dy + dz)z + (x + y + z)dz
(2z + x + y)dz = (1 − z)dx + (1 − z)dy ? ?
Therefore,

1−z 1−z ∂z ∂z ∂z 1−z
dz = dx + dy = dx + dy ⇒ = .
2z + x + y 2z + x + y ∂x y ∂y x ∂x y 2z + x + y
Returning to ? we derive

∂w 1−z
=1+ .
∂x y 2z + x + y
2. suppose w, y both depend on x, z. Calculate,

∂w ∂ ∂x ∂y ∂z ∂y
= (x + y + z) = + + =1+ +0
∂x z ∂x z ∂x z ∂x z ∂x z ∂x z
To complete this calculation we need to eliminate w as before, using ??,

∂y
(1 − z)dy = (1 − z)dx − (2z + x + y)dz ⇒ = 1.
∂x z
Therefore,

∂w
= 2.
∂x z
I hope you can begin to see how the game is played. Basically the example above generalizes the
idea of implicit differentiation to several equations of many variables. This is actually a pretty
important type of calculation for engineering. The study of thermodynamics is full of variables
which are intermittently used as either dependent or independent variables. The so-called equation
of state can be given in terms of about a dozen distinct sets of state variables.
Example 5.3.9. The ideal gas law states that for a fixed number of particles n the pressure P ,
volume V and temperature T are related by P V = nRT where R is a constant. Calculate,

∂P ∂ nRT nRT
= =− 2 ,
∂V T ∂V V
T V

∂V ∂ nRT nR
= = ,
∂T P ∂T P T P

∂T ∂ P V V
= = .
∂P V ∂P nR T nR
∂P ∂V ∂T

You might expect that ∂V T ∂T P ∂P V
= 1. Is it true?

∂P ∂V ∂T nRT nR V −nRT
=− 2 · · = = −1.
∂V T ∂T P ∂P V
V P nR PV
This is an example where naive cancellation of partials fails.

F
Example 5.3.10. Suppose F (x, y) = 0 then dF = Fx dx+Fy dy = 0 and it follows that dx = − Fxy dy
F ∂y
or dy = − FFxy dx. Hence, ∂x
∂y = − Fxy and ∂x = − FFxy . Therefore,
∂x ∂y Fy Fx
= · =1
∂y ∂x Fx Fy
for (x, y) such that Fx 6= 0 and Fy 6= 0. The condition Fx 6= 0 suggests we can solve for y = y(x)
whereas the condition Fy 6= 0 suggests we can solve for x = x(y).
5.4 the constant rank theorem

The implicit function theorem required we work with independent constraints. However, one does
not always have that luxury. There is a theorem which deals with the slightly more general case.
The base idea is that if the Jacobian has rank k then it locally injects a k-dimensional image into
the codomain. If we are using a map as a parametrization then the rank k condition suggests the
mapping does parametrize a k-fold, at least locally. On the other hand, if we are using the map to
5.4. THE CONSTANT RANK THEOREM 139
define a space as a level set then F : Rn → Rp has F −1 (C) as a (n − k)-fold. Previously, we would
have insisted k = p. I’ve run out of time for 2013 notes so sadly I have no reference for this claim.
However, the troubing Section 10.7.2 quotes the Theorem we desire in a somewhat unfortunate
language for our current purposes. In any event, theorems aside, I think the red comments are
worth some discussion.
Remark 5.4.1.
I have put remarks about the rank of the derivative in red for the examples below.
Example 5.4.2. Let f (t) = (t, t2 , t3 ) then f 0 (t) = (1, 2t, 3t2 ). In this case we have
 
1
f 0 (t) = [dft ] =  2t 
3t2
The Jacobian here is a single column vector. It has rank 1 provided the vector is nonzero. We
see that f 0 (t) 6= (0, 0, 0) for all t ∈ R. This corresponds to the fact that this space curve has a
well-defined tangent line for each point on the path.
Example 5.4.3. Let f (~x, ~y ) = ~x · ~y be a mapping from R3 × R3 → R. I’ll denote the coordinates
in the domain by (x1 , x2 , x3 , y1 , y2 , y3 ) thus f (~x, ~y ) = x1 y1 + x2 y2 + x3 y3 . Calculate,
[df(~x,~y) ] = ∇f (~x, ~y )T = [y1 , y2 , y3 , x1 , x2 , x3 ]
The Jacobian here is a single row vector. It has rank 6 provided all entries of the input vectors are
nonzero.
Example 5.4.4. Let f (~x, ~y ) = ~x · ~y be a mapping fromP Rn × Rn → R. I’ll denote the coordinates
in the domain by (x1 , . . . , xn , y1 , . . . , yn ) thus f (~x, ~y ) = ni=1 xi yi . Calculate,
n
X Xn n
X
∂ ∂xi
xj xi yi = xj iy = δij yi = yj
i=1 i=1 i=1
Likewise,
n
X n
X n
X
∂
yj x i yi = xi ∂y
yj
i
= xi δij = xj
i=1 i=1 i=1
Therefore, noting that ∇f = (∂x1 f, . . . , ∂xn f, ∂y1 f, . . . , ∂yn f ),
[df(~x,~y) ]T = (∇f )(~x, ~y ) = ~y × ~x = (y1 , . . . , yn , x1 , . . . , xn )
The Jacobian here is a single row vector. It has rank 2n provided all entries of the input vectors
are nonzero.
Example 5.4.5. Suppose F (x, y, z) = (xyz, y, z) we calculate,

∂F ∂F ∂F
∂x = (yz, 0, 0) ∂y = (xz, 1, 0) ∂z = (xy, 0, 1)
Remember these are actually column vectors in my sneaky notation; (v1 , . . . , vn ) = [v1 , . . . , vn ]T .
This means the derivative or Jacobian matrix of F at (x, y, z) is
 
yz xz xy
F 0 (x, y, z) = [dF(x,y,z) ] =  0 1 0 
0 0 1
Note, rank(F 0 (x, y, z)) = 3 for all (x, y, z) ∈ R3 such that y, z 6= 0. There are a variety of ways to
see that claim, one way is to observe det[F 0 (x, y, z)] = yz and this determinant is nonzero so long
as neither y nor z is zero. In linear algebra we learn that a square matrix is invertible iff it has
nonzero determinant iff it has linearly indpendent column vectors.
Example 5.4.6. Suppose F (x, y, z) = (x2 + z 2 , yz) we calculate,

∂F ∂F ∂F
∂x = (2x, 0) ∂y = (0, z) ∂z = (2z, y)

0 2x 0 2z
F (x, y, z) = [dF(x,y,z) ] =
0 z y
The maximum rank for F 0 is 2 at a particular point (x, y, z) because there are at most two linearly
independent vectors in R2 . You can consider the three square submatrices to analyze the rank for
a given point. If any one of these is nonzero then the rank (dimension of the column space) is two.

2x 0 2x 2z 0 2z
M1 = M2 = M3 =
0 z 0 y z y
We’ll need either det(M1 ) = 2xz 6= 0 or det(M2 ) = 2xy 6= 0 or det(M3 ) = −2z 2 6= 0. I believe
the only point where all three of these fail to be true simulataneously is when x = y = z = 0. This
mapping has maximal rank at all points except the origin.
Example 5.4.7. Suppose F (x, y) = (x2 + y 2 , xy, x + y) we calculate,

∂F ∂F
∂x = (2x, y, 1) ∂y = (2y, x, 1)

 
2x 2y
F 0 (x, y) = [dF(x,y) ] =  y x 
1 1
5.4. THE CONSTANT RANK THEOREM 141
The maximum rank is again 2, this time because we only have two columns. The rank will be two
if the columns are not linearly dependent. We can analyze the question of rank a number of ways
but I find determinants of submatrices a comforting tool in these sort of questions. If the columns
are linearly dependent then all three sub-square-matrices of F 0 will be zero. Conversely, if even one
of them is nonvanishing then it follows the columns must be linearly independent. The submatrices
for this problem are:

2x 2y 2x 2y y x
M1 = M2 = M3 =
y x 1 1 1 1
You can see det(M1 ) = 2(x2 − y 2 ), det(M2 ) = 2(x − y) and det(M3 ) = y − x. Apparently we have
rank(F 0 (x, y, z)) = 2 for all (x, y) ∈ R2 with y 6= x. In retrospect this is not surprising.
p
Example 5.4.8. Let F (x, y) = (x, y, R2 − x2 − y 2 ) for a constant R. We calculate,

, √ −y
p
∇ R2 − x2 − y 2 = √ −x
R2 −x2 −y 2 R2 −x2 −y 2
Also, ∇x = (1, 0) and ∇y = (0, 1) thus

 
1 0
F 0 (x, y) =  0 1
 
√ −y

√ −x
R2 −x2 −y 2 R2 −x2 −y 2
2 2 2
p that we need R − x − y > 0 for the
This matrix clearly has rank 2 where is is well-defined. Note
2 2 2
derivative to exist. Moreover, we could define G(y, z) = ( R − y − z , y, z) and calculate,
 
1 0
−y −z
G0 (y, z) =  √R2 −y2 −z 2 √R2 −y2 −z 2  .
 
0 1
Observe that G0 (y, z) exists when R2 − y 2 − z 2 > 0. Geometrically, F parametrizes the sphere
above the equator at z = 0 whereas G parametrizes the right-half of the sphere with x > 0. These
parametrizations overlap in the first octant where both x and z are positive. In particular, dom(F 0 )∩
dom(G0 ) = {(x, y) ∈ R2 | x, y > 0 and x2 + y 2 < R2 }
p
Example 5.4.9. Let F (x, y, z) = (x, y, z, R2 − x2 − y 2 − z 2 ) for a constant R. We calculate,

−y
p
−x −z
∇ R 2 − x2 − y 2 − z 2 = √ , √ , √
R2 −x2 −y 2 −z 2 R2 −x2 −y 2 −z 2 R2 −x2 −y 2 −z 2
Also, ∇x = (1, 0, 0), ∇y = (0, 1, 0) and ∇z = (0, 0, 1) thus

 
1 0 0
 0 1 0 
F 0 (x, y, z) = 
 
0 0 1 
−y
 
√ −x √ √ −z
R2 −x2 −y 2 −z 2 R2 −x2 −y 2 −z 2 R2 −x2 −y 2 −z 2
This matrix clearly has rank 3 where is is well-defined. Note that we need R2 −x2 −y 2 −z 2 > 0 for the
derivative to exist. This mapping gives us a parametrization of the 3-sphere x2 + y 2 + z 2 + w2 = R2
for w > 0. (drawing this is a little trickier)
Example 5.4.10. Let f (x, y, z) = (x + y, y + z, x + z, xyz). You can calculate,

 
1 1 0
 0 1 1 
[df(x,y,z) ] = 
 1 0

1 
yz xz xy
This matrix clearly has rank 3 and is well-defined for all of R3 .
Example 5.4.11. Let f (x, y, z) = xyz. You can calculate,

[df(x,y,z) ] = yz xz xy
This matrix fails to have rank 3 if x, y or z are zero. In other words, f 0 (x, y, z) has rank 3 in
R3 provided we are at a point which is not on some coordinate plane. (the coordinate planes are
x = 0, y = 0 and z = 0 for the yz, zx and xy coordinate planes respective)
Example 5.4.12. Let f (x, y, z) = (xyz, 1 − x − y). You can calculate,

yz xz xy
[df(x,y,z) ] =
−1 −1 0
This matrix has rank 3 if either xy 6= 0 or (x − y)z 6= 0. In contrast to the preceding example, the
derivative does have rank 3 on certain points of the coordinate planes. For example, f 0 (1, 1, 0) and
f 0 (0, 1, 1) both give rank(f 0 ) = 3.
Example 5.4.13. Let X(u, v) = (x, y, z) where x, y, z denote functions of u, v and I prefer to omit
the explicit depedendence to reduce clutter in the equations to follow.
∂X ∂X
= Xu = (xu , yu , zu ) and = Xv = (xv , yv , zv )
∂u ∂v
Then the Jacobian is the 3 × 2 matrix
 
x u x v
dX(u,v) =  yu yv 
zu zv

The matrix dX(u,v) has rank 2 if at least one of the determinants below is nonzero,

xu xv xu xv yu yv
det det det
yu yv zu zv zu z v
Chapter 6
two views of manifolds in Rn
In this chapter we describe spaces inside Rn which are k-dimensional 1 . Technically, to make this
precise we would need to study manifolds with boundary. Careful discussion of manifolds with
boundary in euclidean space can be found in Munkres Analysis on Manifolds. In the interest of
focusing on examples, I’ll be a bit fuzzy about the defintion of a k-dimensional subspace S of
euclidean space. This much we can say: there are two ways to envision the geometry of S:
(1.) Parametrically: provide a patch R such that R : U ⊆ Rk → S ⊆ Rn . Here U is called the

parameter space and R−1 is called a coordinate chart. The cannonical example:
R(x1 , . . . xk ) = (x1 , . . . xk , 0, . . . , 0).
(2.) Implicitly: provide a level function G : Rk × Rp → Rp such that S = G−1 {c} = S. This
viewpoint casts S as points in x ∈ Rk × Rp for which G(x) = k. The cannonical example:
G(x1 , . . . , xk+p ) = (xk+1 , . . . , xk+p ) = (0, . . . , 0).
The cannonical examples of (1.) and (2.) are both the x1 . . . xk -coordinate plane embedded in Rn .
Just to take it down a notch. If n = 3 then we could look at the xy-plane in either view as follows:
(1.) R(x, y) = (x, y, 0) (2.) G(x, y, z) = z = 0.
Which viewpoint should we adopt? What is the dimension of a given space S? How should we
find tangent space to S? How should we find the normal space to S? These are the questions we
set-out to answer in this chapter.
Orthogonal complements help us to understand how all of this fits together. This is possible since we
deal with embedded manifolds for which the euclidean dot-product of Rn is available to sort out the
geometry. Finally, we use this geometry and a few simple lemmas to justify the method of Lagrange
multipliers. Lagrange’s technique paired with the theory of multivariate Taylor polynomials form
1
I’ll try to stick with this notation for this chapter, n ≥ k and n = p + k
143
144 CHAPTER 6. TWO VIEWS OF MANIFOLDS IN RN
the basis for analyzing extrema for multivariate functions. In this chapter we deal with the question
of extrema on the edges of a set. The second half of the story is found in the next chapter where
we deal with the interior points via the theory of quadratic forms applied to the second-order
approximation to a function of several variables.
6.1 definition of level set

A level set is the solution set of some equation or system of equations. We confine our interest to
level sets of Rn . For example, the set of all (x, y) that satisfy
G(x, y) = c
is called a level curve in R2 . Often we can use k to label the curve. You should also recall level
surfaces in R3 are defined by an equation of the form
G(x, y, z) = c.
The set of all (x1 , x2 , x3 , x4 ) ∈ R4 which solve G(x1 , x2 , x3 , x4 ) = c is a level volume in R4 . We

can obtain lower dimensional objects by simultaneously imposing several equations at once. For
example, suppose G1 (x, y, z) = z = 1 and G2 (x, y, z) = x2 + y 2 + z 2 = 5, points (x, y, z) which solve
both of these equations are on the intersection of the plane z = 1 and the sphere x2 + y 2 + z 2 = 5.
Let G = (G1 , G2 ), note that G(x, y, z) = (1, 5) describes a circle in R3 . More generally:
Definition 6.1.1.
Suppose G : dom(G) ⊆ Rk × Rp → Rp . Let c be a vector of constants in Rp and suppose

S = {x ∈ Rk × Rp | G(x) = c} is non-empty and G is continuously differentiable on an
open set containing S. We say S is an k-dimensional level set iff G0 (x) has p linearly
independent rows at each x ∈ S.
The condition of linear independence of the rows is give to eliminate possible redundancy in the
system of equations. In the case that p = 1 the criteria reduces to G0 (x) 6= 0 over the level set
of dimension n − 1. Intuitively we think of each equation in G(x) = c as removing one of the
dimensions of the ambient space Rn = Rk × Rp . It is worthwhile to cite a useful result from linear
algebra at this point:
Proposition 6.1.2.
Let A ∈ R m×n . The number of linearly independent columns in A is the same as the
number of linearly independent rows in A. This invariant of A is called the rank of A.
Given the wisdom of linear algebra we see that we should require a k-dimensional level set S =
G−1 (c) to have a level function G : Rn → Rp whose derivative is of rank n − k = p over all of S.
We can either analyze linear independence of columns or rows.
6.2. TANGENTS AND NORMALS TO A LEVEL SET 145
Example 6.1.3. Consider G(x, y, z) = x2 + y 2 − z 2 and suppose S = G−1 {0}. Calculate,

G0 (x, y, z) = [2x, 2y, −2z]
Notice that (0, 0, 0) ∈ S and G0 (0, 0, 0) = [0, 0, 0] hence G0 is not rank one at the origin. At all
other points in S we have G0 (x, y, z) 6= 0 which means this is almost a 3 − 1 = 2-dimensional
level set. However, almost is not good enough in math. Under our definition the cone S is not a
2-dimensional level set since it fails to meet the full-rank criteria at the point of the cone.
A p-dimensional level set is an example of a p-dimensional manifold. The example above with the
origin included is a manifold paired with a singular point, such spaces are known as orbifolds.
The study of orbifolds has attracted considerable effort in recent years as the singularities of such
orbifolds can be used to do physics in string theory. I digress. Let us examine another level set:
Example 6.1.4. Let G(x, y, z) = (x, y) and define S = G−1 (a, b) for some fixed pair of constants
a, b ∈ R. We calculate that G0 (x, y, z) = I2 ∈ R2×2 . We clearly have rank two at all points in S
hence S is a 3 − 2 = 1-dimensional level set. Perhaps you realize S is the vertical line which passes
through (a, b, 0) in the xy-plane.
6.2 tangents and normals to a level set

There are many ways to define a tangent space for some subset of Rn . One natural definition is
that the tangent space to p ∈ S is simply the set of all tangent vectors to curves on S which pass
through the point p. In this section we study the geometry of curves on a level-set. We’ll see how
the tangent space is naturally a vector space in the particular context of level-sets in Rn .
Throughout this section we assume that S is a k-dimensional level set defined by G : Rk × Rp → Rp

where G−1 (c) = S. This means that we can apply the implicit function theorem to S and for
any given point p = (px , py ) ∈ S where px ∈ Rk and py ∈ Rp . There exists a local continuously
differentiable solution h : U ⊆ Rk → Rp such that h(px ) = py and for all x ∈ U we have G(x, h(x)) =
c. We can view G(x, y) = c for x near p as the graph of y = h(x) for x ∈ U . With the set-up above
in mind, suppose that γ : R → U ⊆ S. If we write γ = (γx , γy ) then it follows γ = (γx , h ◦ γx ) over
the subset U × h(U ) of S. More explicitly, for all t ∈ R such that γ(t) ∈ U × h(U ) we have
γ(t) = (γx (t), h(γx (t))).
Therefore, if γ(0) = p then γ(0) = (px , h(px )). Differentiate, use the chain-rule in the second factor
to obtain:
γ 0 (t) = (γx0 (t), h0 (γx (t))γx0 (t)).
We find that the tangent vector to p ∈ S of γ has a rather special form which was forced on us by
the implicit function theorem:
γ 0 (0) = (γx0 (0), h0 (px )γx0 (0)).
Or to cut through the notation a bit, if γ 0 (0) = v = (vx , vy ) then v = (vx , h0 (px )vx ). The second
component of the vector is not free of the first, it essentially redundant. This makes us suspect
that the tangent space to S at p is k-dimensional.
Theorem 6.2.1.
Let G : Rk × Rp → Rp be a level-mappping which defines a k-dimensional level set S

by G−1 (c) = S. Suppose γ1 , γ2 : R → S are differentiable curves with γ10 (0) = v1 and
γ20 (0) = v2 then there exists a differentiable curve γ : R → S such that γ 0 (0) = v1 + v2 and
γ(0) = p. Moreover, there exists a differentiable curve β : R → S such that β 0 (0) = cv1 and
β(0) = p.
Proof: It is convenient to define a map which gives a local parametrization of S at p. Since
we have a description of S locally as a graph y = h(x) (near p) it is simple to construct the
parameterization. Define Φ : U ⊆ Rk → S by Φ(x) = (x, h(x)). Clearly Φ(U ) = U × h(U ) and
there is an inverse mapping Φ−1 (x, y) = x is well-defined since y = h(x) for each (x, y) ∈ U × h(U ).
Let w ∈ Rk and observe that
ψ(t) = Φ(Φ−1 (p) + tw) = Φ(px + tw) = (px + tw, h(px + tw))
is a curve from R to U ⊆ S such that ψ(0) = (px , h(px )) = (px , py ) = p and using the chain rule on
the final form of ψ(t):
ψ 0 (0) = (w, h0 (px )w).
The construction above shows that any vector of the form (vx , h0 (px )vx ) is the tangent vector of a
particular differentiable curve in the level set (differentiability of ψ follows from the differentiability
of h and the other maps which we used to construct ψ). In particular we can apply this to the
case w = v1x + v2x and we find γ(t) = Φ(Φ−1 (p) + t(v1x + v2x )) has γ 0 (0) = v1 + v2 and γ(0) = p.
Likewise, apply the construction to the case w = cv1x to write β(t) = Φ(Φ−1 (p) + t(cv1x )) with
β 0 (0) = cv1 and β(0) = p.
The idea of the proof is encapsulated in the picture below. This idea of mapping lines in a flat
domain to obtain standard curves in a curved domain is an idea which plays over and over as you
study manifold theory. The particular redundancy of the x and y sub-vectors is special to the
discussion level-sets, however anytime we have a local parametrization we’ll be able to construct
curves with tangents of our choosing by essentially the same construction. In fact, there are infinitely
many curves which produce a particular tangent vector in the tangent space of a manifold.
picture.
Theorem 6.2.1 shows that the definition given below is logical. In particular, it is not at all obvious
that the sum of two tangent vectors ought to again be a tangent vector. However, that is just what
the Theorem 6.2.1 told us for level-sets2 .
2
technically, there is another logical gap which I currently ignore. I wonder if you can find it.
Definition 6.2.2.
Suppose S is a k-dimensional level-set defined by S = G−1 {c} for G : Rk × Rp → Rp . We

define the tangent space at p ∈ S to be the set of pairs:
Tp S = {(p, v) | there exists differentiable γ : R → S and γ(0) = p where v = γ 0 (0)}
Moreover, we define (i.) addition and (ii.) scalar multiplication of vectors by the rules
(i.) (p, v1 ) + (p, v2 ) = (p, v1 + v2 ) (ii.) c(p, v1 ) = (p, cv1 )
for all (p, v1 ), (p, v2 ) ∈ Tp S and c ∈ R.

When I picture Tp S in my mind I think of vectors pointing out from the base-point p. To make
an explicit connection between the pairs of the above definition and the classical geometric form
of the tangent space we simply take the image of Tp S under the mapping Ψ(x, y) = x + y thus
Ψ(Tp S) = {p + v | (p, v) ∈ Tp S}. I often picture Tp S as ψ(Tp S)3
We could set out to calculate tangent spaces in view of the definition above, but we are actually
interested in more than just the tangent space for a level-set. In particular. we want a concrete
description of all the vectors which are not in the tangent space.
Definition 6.2.3.
Suppose S is a k-dimensional level-set defined by S = G−1 {c} for G : Rk × Rp → Rp and

Tp S is the tangent space at p. Note that Tp S ≤ Vp where Vp = {p} × Rk × Rp is given the
natural vector space structure which we already exhibited on the subspace Tp S. We define
the inner product on Vp as follows: for all (p, v), (p, w) ∈ Vp ,
(p, v) · (p, w) = v · w.
The length of a vector (p, v) is naturally defined by ||(p, v)|| = ||v||. Moreover, we say two
vectors (p, v), (p, w) ∈ Vp are orthogonal iff v · w = 0. Given a set of vectors R ⊆ Vp we
define the orthogonal complement by
R⊥ = {(p, v) ∈ Vp | (p, v) · (p, r) for all (p, r) ∈ R}.
Suppose W1 , W2 ⊆ Vp then we say W1 is orthogonal to W2 iff w1 · w2 = 0 for all w1 ∈ W1

and w2 ∈ W2 . We denote orthogonality by writing W1 ⊥ W2 . If every v ∈ Vp can be written
as v = w1 + w2 for a pair of w1 ∈ W1 and w2 ∈ W2 where W1 ⊥ W2 then we say that Vp is
the direct sum of W1 and W2 which is denoted by Vp = W1 ⊕ W2 .
There is much more to say about orthogonality, however, our focus is not in that vein. We just
need the langauge to properly define the normal space. The calculation below is probably the most
3
In truth, as you continue to study manifold theory you’ll find at least three seemingly distinct objects which are
all called ”tangent vectors”; equivalence classes of curves, derivations, contravariant tensors.
important calculation to understand for a level-set. Suppose we have a curve γ : R → S where

S = G−1 (c) is a k-dimensional level-set in Rk × Rp . Observe that for all t ∈ R,
G(γ(t)) = c ⇒ G0 (γ(t))γ 0 (t) = 0.
In particular, suppose for t = 0 we have γ(0) = p and v = γ 0 (0) which makes (p, v) ∈ Tp S with
G0 (p)v = 0.
Recall G : Rk × Rp → Rp has an p × n derivative matrix where the j-th row is the gradient vector
of the j-th component function. The equation G0 (p)v = 0 gives us p-independent equations as
we examine it componentwise. In particular, it reveals that (p, v) is orthogonal to ∇Gj (p) for
j = 1, 2, . . . , p. We have derived the following theorem:
Theorem 6.2.4.
Let G : Rk × Rp → Rp be a level-mappping which defines a k-dimensional level set S by

G−1 (c) = S. The gradient vectors ∇Gj (p) are perpendicular to the tangent space at p; for
each j ∈ Np
(p, ∇(Gj (p))T ) ∈ (Tp S)⊥ .
It’s time to do some counting. Observe that the mapping φ : Rk → Tp S defined by φ(v) = (p, v)
is an isomorphism of vector spaces hence dim(Tp S) = k. But, by the same isomorphism we can
see that Vp = φ(Rk × Rp ) hence dim(Vp ) = p + k. In linear algebra we learn that if we have a
k-dimensional subspace W of an n-dimensional vector space V then the orthogonal complement
W ⊥ is a subspace of V with codimension k. The term codimension is used to indicate a loss
of dimension from the ambient space, in particular dim(W ⊥ ) = n − k. We should note that the
direct sum of W and W ⊥ covers the whole space; W ⊕ W ⊥ = V . In the case of the tangent space,
the codimension of Tp S ≤ Vp is found to be p + k − k = p. Thus dim(Tp S)⊥ = p. Any basis for
this space must consist of p linearly independent vectors which are all orthogonal to the tangent
space. Naturally, the subset of vectors {(p, (∇Gj (p))T )pj=1 forms just such a basis since it is given
to be linearly independent by the rank(G0 (p)) = p condition. It follows that:
(Tp S)⊥ ≈ Row(G0 (p))
where equality can be obtained by the slightly tedious equation (Tp S)⊥ = φ(Col(G0 (p)T )) . That
equation simply does the following:
1. transpose G0 (p) to swap rows to columns
2. construct column space by taking span of columns in G0 (p)T
3. adjoin p to make pairs of vectors which live in Vp

many wiser authors wouldn’t bother. The comments above are primarily about notation. Certainly
hiding these details would make this section prettier, however, would it make it better? Finally, I
once more refer the reader to linear algebra where we learn that (Row(A))⊥ = N ull(AT ). Let me
walk you through the proof: let A ∈ R m×n . Observe v ∈ N ull(AT ) iff AT v = 0 for v ∈ Rm iff
v T A = 0 iff v T colj (A) = 0 for j = 1, 2, . . . , n iff v · colj (A) = 0 for j = 1, 2, . . . , n iff v ∈ Col(A)⊥ .
Another useful identity for the ”perp” is that (A⊥ )⊥ = A. With those two gems in mind consider
that:
(Tp S)⊥ ≈ Row(G0 (p)) ⇒ Tp S ≈ Row(G0 (p))⊥ = N ull(G0 (p)T )
Let me once more replace ≈ by a more tedious, but explicit, procedure:
Tp S = φ(N ull(G0 (p)T ))
Theorem 6.2.5.
Let G : Rk × Rp → Rp be a level-mappping which defines a k-dimensional level set S by

G−1 (c) = S. The tangent space Tp S and the normal space Np S at p ∈ S are given by
Tp S = {p} × N ull(G0 (p)T ) & Np S = {p} × Col(G0 (p)T ).
Moreover, Vp = Tp S ⊕ Np S. Every vector can be uniquely written as the sum of a tangent

vector and a normal vector.
The fact that there are only tangents and normals is the key to the method of Lagrange multipliers.
It forces two seemingly distinct objects to be in the same direction as one another.
Example 6.2.6. Let g : R4 → R be defined by g(x, y, z, t) = t+x2 +y 2 −2z 2 note that g(x, y, z, t) = 0
gives a three dimensional subset of R4 , let’s call it M . Notice ∇g =< 2x, 2y, −4z, 1 > is nonzero
everywhere. Let’s focus on the point (2, 2, 1, 0) note that g(2, 2, 1, 0) = 0 thus the point is on M .
The tangent plane at (2, 2, 1, 0) is formed from the union of all tangent vectors to g = 0 at the
point (2, 2, 1, 0). To find the equation of the tangent plane we suppose γ : R → M is a curve with
γ 0 6= 0 and γ(0) = (2, 2, 1, 0). By assumption g(γ(s)) = 0 since γ(s) ∈ M for all s ∈ R. Define
γ 0 (0) =< a, b, c, d >, we find a condition from the chain-rule applied to g ◦ γ = 0 at s = 0,
d
g ◦ γ(s) = ∇g (γ(s)) · γ 0 (s) = 0

⇒ ∇g(2, 2, 1, 0) · < a, b, c, d >= 0
ds
⇒ < 4, 4, −4, 1 > · < a, b, c, d >= 0
⇒ 4a + 4b − 4c + d = 0
Thus the equation of the tangent plane is 4(x − 2) + 4(y − 2) − 4(z − 1) + t = 0. In invite the
reader to find a vector in the tangent plane and check it is orthogonal to ∇g(2, 2, 1, 0). However,
this should not be surprising, the condition the chain rule just gave us is just the statement that
< a, b, c, d >∈ N ull(∇g(2, 2, 1, 0)T ) and that is precisely the set of vector orthogonal to ∇g(2, 2, 1, 0).
Example 6.2.7. Let G : R4 → R2 be defined by G(x, y, z, t) = (z + x2 + y 2 − 2, z + y 2 + t2 − 2). In

this case G(x, y, z, t) = (0, 0) gives a two-dimensional manifold in R4 let’s call it M . Notice that
G1 = 0 gives z + x2 + y 2 = 2 and G2 = 0 gives z + y 2 + t2 = 2 thus G = 0 gives the intersection of
both of these three dimensional manifolds in R4 (no I can’t ”see” it either). Note,
∇G1 =< 2x, 2y, 1, 0 > ∇G2 =< 0, 2y, 1, 2t >
It turns out that the inverse mapping theorem says G = 0 describes a manifold of dimension 2 if
the gradient vectors above form a linearly independent set of vectors. For the example considered
here the gradient vectors are linearly dependent at the origin since ∇G1 (0) = ∇G2 (0) = (0, 0, 1, 0).
In fact, these gradient vectors are colinear along along the plane x = t = 0 since ∇G1 (0, y, z, 0) =
∇G2 (0, y, z, 0) =< 0, 2y, 1, 0 >. We again seek to contrast the tangent plane and its normal at
some particular point. Choose (1, 1, 0, 1) which is in M since G(1, 1, 0, 1) = (0 + 1 + 1 − 2, 0 +
1 + 1 − 2) = (0, 0). Suppose that γ : R → M is a path in M which has γ(0) = (1, 1, 0, 1) whereas
γ 0 (0) =< a, b, c, d >. Note that ∇G1 (1, 1, 0, 1) =< 2, 2, 1, 0 > and ∇G2 (1, 1, 0, 1) =< 0, 2, 1, 1 >.
Applying the chain rule to both G1 and G2 yields:
(G1 ◦ γ)0 (0) = ∇G1 (γ(0))· < a, b, c, d >= 0 ⇒ < 2, 2, 1, 0 > · < a, b, c, d >= 0
0
(G2 γ) (0) = ∇G2 (γ(0))· < a, b, c, d >= 0
◦ ⇒ < 0, 2, 1, 1 > · < a, b, c, d >= 0
This is two equations and four unknowns, we can solve it and write the vector in terms of two free
variables correspondant to the fact the tangent space is two-dimensional. Perhaps it’s easier to use
matrix techiques to organize the calculation:
 
a
2 2 1 0   b = 0

0 2 1 1  c  0
d

2 2 1 0 1 0 0 −1/2
We calculate, rref = . It’s natural to chose c, d as free vari-
0 2 1 1 0 1 1/2 1/2
ables then we can read that a = d/2 and b = −c/2 − d/2 hence
c
< a, b, c, d >=< d/2, −c/2 − d/2, c, d >= 2 < 0, −1, 2, 0 > + d2 < 1, −1, 0, 2 >
We can see a basis for the tangent space. In fact, I can give parametric equations for the tangent
space as follows:
X(u, v) = (1, 1, 0, 1) + u < 0, −1, 2, 0 > +v < 1, −1, 0, 2 >
Not surprisingly the basis vectors of the tangent space are perpendicular to the gradient vectors
∇G1 (1, 1, 0, 1) =< 2, 2, 1, 0 > and ∇G2 (1, 1, 0, 1) =< 0, 2, 1, 1 > which span the normal plane
Np to the tangent plane Tp at p = (1, 1, 0, 1). We find that Tp is orthogonal to Np . In summary
Tp⊥ = Np and Tp ⊕ Np = R4 . This is just a fancy way of saying that the normal and the tangent
plane only intersect at zero and they together span the entire ambient space.
6.3. TANGENT AND NORMAL SPACE FROM PATCHES 151
6.3 tangent and normal space from patches

I use the term parametrization in courses more basic than this, however, perhaps the term patch
would be better. It’s certainly easier to say and in our current context has the same meaning. I
suppose the term parametrization is used in a bit less technical sense, so it fits calculus III better.
In any event, we should make a definition of patched k-dimensional surface for the sake of concrete
discussion in this section.
Definition 6.3.1.
Suppose R : dom(R) ⊆ Rk → S ⊆ Rn . We say S is an k-dimensional patch iff R0 (t) has
rank k for each t ∈ dom(R). We also call S a k-dimensional parametrized subspace of Rn .
The condition R0 (t) is just a slick way to say that the k-tangent vectors to S obtained by par-
tial differentiation with respect to t1 , . . . , tk are linearly independent at t = (t1 , . . . , tk ). I spent
considerable effort justifying the formulae for the level-set case. I believe what follows should be
intuitively clear given our previous efforts. Or, if that leaves you unsatisfied then read on to the
examples. It’s really not that complicated. This theorem is dual to Theorem 6.2.5.
Theorem 6.3.2.
Suppose R : dom(R) ⊆ Rk → S ⊆ Rn defines a k-dimensional patch of S. The tangent
space Tp S and the normal space at p = R(t) ∈ S are given by
Tp S = {p} × Col(R0 (t)) & Np S = {p} × N ull(R0 (t)T ).
Moreover, Vp = Tp S ⊕ Np S. Every vector can be uniquely written as the sum of a tangent

vector and a normal vector.
Once again, the vector space structure of Tp S and Np S is given by the addition of vectors based
at p. Let us begin with a reasonably simple example.
Example 6.3.3. Let R : R2 → R3 with R(x, y) = (x, y, xy) define S ⊂ R3 . We calculate,
 
1 0
R0 (x, y) = [∂x R|∂y R] =  0 1 
y x
If p = (a, b, ab) ∈ S then Tp S = {(a, b, ab)} × span{(1, 0, b), (0, 1, a)}. The normal space is found
from N ull(R0 (a, b)T ). A short calculation shows that

1 0 b
N ull = span{(−b, −a, 1)}
0 1 a
As a quick check, note (1, 0, b) • (−b, −a, 1) = 0 and (0, 1, a) • (−b, −a, 1) = 0. We conclude, for
p = (a, b, ab) the normal space is simply:
Np S = {(a, b, ab)} × span{(−b, −a, 1)}.

In the previous example, we could rightly call Tp S the tangent plane at p and Np S the normal line
through p. Moreover, we could have used three-dimensional vector analysis to find the normal line
from the cross-product. However, that will not be possible in what follows:
Example 6.3.4. Let R : R2 → R4 with R(s, t) = (s2 , t2 , t, s) define S ⊂ R4 . We calculate,

 
2s 0
 0 2t 
R0 (s, t) = [∂s R|∂t R] = 
 0 1 

1 0
If p = (1, 9, 3, 1) ∈ S then Tp S = {(1, 9, 3, 1)} × span{(2, 0, 0, 1), (0, 6, 3, 0)}. The normal space is
found from N ull(R0 (1, 3)T ). A short calculation shows that

2 0 0 1
N ull = span{(−1, 0, 0, 2), (0, −3, 6, 0)}
0 6 3 0
We conclude, for p = (1, 9, 3, 1) the normal space is simply:
Np S = {(1, 9, 3, 1)} × span{(−1, 0, 0, 2), (0, −3, 6, 0)}.
6.4 summary of tangent and normal spaces

Let me briefly draw together what we did thus far in this chapter: the notation below given in I is
also used in II. and III.
(I.) a set S has dimension k if
(a) {∂1 R(t), . . . , ∂k R(t)} is pointwise linearly independent at each t ∈ U where R : U → S

is a patch.
(b) rank(F 0 (x)) = p for all x ∈ S̃ where S̃ is open and contains S = F −1 {c} for continu-
ously differentiable F : Rk × Rp → Rp
(II.) the tangent space at xo for the k-dimensional set S is found from:
(a) attaching the span of the vectors {∂1 R(to ), . . . , ∂k R(to )} to xo = R(to ) ∈ S.
(b) attaching the Row(F 0 (xo ))⊥ to xo ∈ S.
(III.) the normal space to a k-dimensional set S (embedded in Rn ) is found from:
(a) attaching {∂1 R(to ), . . . , ∂k R(to )}⊥ to xo = R(to ).

(b) attaching Row(F 0 (xo )) to xo ∈ S.
6.5. METHOD OF LAGRANGE MULITPLIERS 153
6.5 method of Lagrange mulitpliers

Let us begin with a statement of the problem we wish to solve.
Problem: given an objective function f : Rn → R and continuously differentiable

constraint function G : Rn → Rp , find extreme values for the objective function
f relative to the constraint G(x) = c.
Note that G(x) = c is a vector notation for p-scalar equations. If we suppose rank(G0 (x)) = p
then the constraint surface G(x) = c will form an (n − p)-dimensional level set. Let us make that
supposition throughout the remainder of this section.
In order to solve a problem it is sometimes helpful to find necessary conditions by assuming an

answer exists. Let us do that here. Suppose xo maps to the local extrema of f (xo ) on S = G−1 {c}.
This means there exists an open ball around xo for which f (xo ) is either an upper or lower bound
of all the values of f over the ball intersected with S. One clear implication of this data is that
if we take any continuously differentiable curve on S which passes through xo , say γ : R → Rn
with γ(0) = xo and G(γ(t)) = c for all t, then the composite f ◦ γ is a function on R which takes
an extreme value at t = 0. Fermat’s theorem from calculus I applies and as f ◦ γ is differentiable
near t = 0 we find (f ◦ γ)0 (0) = 0 is a necessary condition. But, this means we have two necessary
conditions on γ:
1. G(γ(t)) = c
2. (f ◦ γ)0 (0) = 0
Let us expand a bit on both of these conditions:
1. G0 (xo )γ 0 (0) = 0
2. f 0 (xo )γ 0 (0) = 0
The first of these conditions places γ 0 (0) ∈ Txo S but then the second condition says that f 0 (xo ) =
(∇f )(xo )T is orthogonal to γ 0 (0) hence (∇f )(xo )T ∈ Nxo . Now, recall from the last section that
the gradient vectors of the component functions to G span the normal space, this means any vector
in Nxo can be written as a linear combination of the gradient vectors. In particular, this means
there exist constants λ1 , λ2 , . . . , λp such that
(∇f )(xo )T = λ1 (∇G1 )(xo )T + λ2 (∇G2 )(xo )T + · · · + λp (∇Gp )(xo )T
We may summarize the method of Lagrange multipliers as follows:

1. choose n-variables which aptly describe your problem.
2. identify your objective function and write all constraints as level surfaces.
3. solve ∇f = λ1 ∇G1 + λ2 ∇G2 + · · · + λp ∇Gp subject to the constraint G(x) = c.
4. test the validity of your proposed extremal points.
The obvious gap in the method is the supposition that an extrema exists for the restriction f |S .
Well examine a few examples before I reveal a sufficient condition. We’ll also see how absence of
that sufficient condition does allow the method to fail.
Example 6.5.1. Suppose we wish to find maximum and minimum distance to the origin for points
on the curve x2 − y 2 = 1. In this case we can use the distance-squared function as our objective
f (x, y) = x2 + y 2 and the single constraint function is g(x, y) = x2 − y 2 . Observe that ∇f =<
2x, 2y > whereas ∇g =< 2x, −2y >. We seek solutions of ∇f = λ∇g which gives us < 2x, 2y >=
λ < 2x, −2y >. Hence 2x = 2λx and 2y = −2λy. We must solve these equations subject to the
condition x2 − y 2 = 1. Observe that x = 0 is not a solution since 0 − y 2 = 1 has no real solution.
On the other hand, y = 0 does fit the constraint and x2 − 0 = 1 has solutions x = ±1. Consider
then
2x = 2λx and 2y = −2λy ⇒ x(1 − λ) = 0 and y(1 + λ) = 0
Since x 6= 0 on the constraint curve it follows that 1 − λ = 0 hence λ = 1 and we learn that
y(1 + 1) = 0 hence y = 0. Consequently, (1, 0 and (−1, 0) are the two point where we expect to find
extreme-values of f . In this case, the method of Lagrange multipliers served it’s purpose, as you
can see in the graph. Below the green curves are level curves of the objective function whereas the
particular red curve is the given constraint curve.
The picture below is a screen-shot of the Java applet created by David Lippman and Konrad
Polthier to explore 2D and 3D graphs. Especially nice is the feature of adding vector fields to given
objects, many other plotters require much more effort for similar visualization. See more at the
website: http://dlippman.imathas.com/g1/GrapherLaunch.html.
Note how the gradient vectors to the objective function and constraint function line-up nicely at
those points.
In the previous example, we actually got lucky. There are examples of this sort where we could get
false maxima due to the nature of the constraint function.
Example 6.5.2. Suppose we wish to find the points on the unit circle g(x, y) = x2 + y 2 = 1 which
give extreme values for the objective function f (x, y) = x2 − y 2 . Apply the method of Lagrange
multipliers and seek solutions to ∇f = λ∇g:
< 2x, −2y >= λ < 2x, 2y >
We must solve 2x = 2xλ which is better cast as (1 − λ)x = 0 and −2y = 2λy which is nicely written
as (1 + λ)y = 0. On the basis of these equations alone we have several options:
1. if λ = 1 then (1 + 1)y = 0 hence y = 0
2. if λ = −1 then (1 − (1))x = 0 hence x = 0

But, we also must fit the constraint x2 + y 2 = 1 hence we find four solutions:
1. if λ = 1 then y = 0 thus x2 = 1 ⇒ x = ±1 ⇒ (±1, 0)
2. if λ = −1 then x = 0 thus y 2 = 1 ⇒ y = ±1 ⇒ (0, ±1)

We test the objective function at these points to ascertain which type of extrema we’ve located:
f (0, ±1) = 02 − (±1)2 = −1 & f (±1, 0) = (±1)2 − 02 = 1
When constrained to the unit circle we find the objective function attains a maximum value of 1 at
the points (1, 0) and (−1, 0) and a minimum value of −1 at (0, 1) and (0, −1). Let’s illustrate the
answers as well as a few non-answers to get perspective. Below the green curves are level curves of
the objective function whereas the particular red curve is the given constraint curve.
The success of the last example was no accident. The fact that the constraint curve was a circle
which is a closed and bounded subset of R2 means that is is a compact subset of R2 . A well-known
theorem of analysis states that any real-valued continuous function on a compact domain attains
both maximum and minimum values. The objective function is continuous and the domain is
compact hence the theorem applies and the method of Lagrange multipliers succeeds. In contrast,
the constraint curve of the preceding example was a hyperbola which is not compact. We have
no assurance of the existence of any extrema. Indeed, we only found minima but no maxima in
Example 6.5.1.
The generality of the method of Lagrange multipliers is naturally limited to smooth constraint
curves and smooth objective functions. We must insist the gradient vectors exist at all points of
inquiry. Otherwise, the method breaks down. If we had a constraint curve which has sharp corners
then the method of Lagrange breaks down at those corners. In addition, if there are points of dis-
continuity in the constraint then the method need not apply. This is not terribly surprising, even in
calculus I the main attack to analyze extrema of function on R assumed continuity, differentiability
and sometimes twice differentiability. Points of discontinuity require special attention in whatever
context you meet them.
At this point it is doubtless the case that some of you are, to misquote an ex-student of mine, ”not-
impressed”. Perhaps the following examples better illustrate the dangers of non-compact constraint
curves.
Example 6.5.3. Suppose we wish to find extrema of f (x, y) = x when constrained to xy = 1.

Identify g(x, y) = xy = 1 and apply the method of Lagrange multipliers and seek solutions to
∇f = λ∇g:
< 1, 0 >= λ < y, x > ⇒ 1 = λy and 0 = λx
If λ = 0 then 1 = λy is impossible to solve hence λ 6= 0 and we find x = 0. But, if x = 0 then
xy = 1 is not solvable. Therefore, we find no solutions. Well, I suppose we have succeeded here
in a way. We just learned there is no extreme value of x on the hyperbola xy = 1. Below the
green curves are level curves of the objective function whereas the particular red curve is the given
constraint curve.
Example 6.5.4. Suppose we wish to find extrema of f (x, y) = x when constrained to x2 − y 2 = 1.

Identify g(x, y) = x2 − y 2 = 1 and apply the method of Lagrange multipliers and seek solutions to
∇f = λ∇g:
< 1, 0 >= λ < 2x, −2y > ⇒ 1 = 2λx and 0 = −2λy
If λ = 0 then 1 = 2λx is impossible to solve hence λ 6= 0 and we find y = 0. If y = 0 and x2 −y 2 = 1
then we must solve x2 = 1 whence x = ±1. We are tempted to conclude that:
1. the objective function f (x, y) = x attains a maximum on x2 −y 2 = 1 at (1, 0) since f (1, 0) = 1
2. the objective function f (x, y) = x attains a minimum on x2 − y 2 = 1 at (−1, 0) since f (1, 0) =
−1
√ 2 √
But, both conclusions
√ are false. Note
√ √ 2 −√ 12 = 1 hence (± 2, 1) are points on the constraint
curve and f ( 2, 1) = 2 and f (− 2, 1) = − 2. The error of the method of Lagrange multipliers
in this context is the supposition that there exists extrema to find, in this case there are no such
points. It is possible for the gradient vectors to line-up at points where there are no extrema. Below
the green curves are level curves of the objective function whereas the particular red curve is the
given constraint curve.
Incidentally, if you want additional discussion of Lagrange multipliers for two-dimensional problems
one very nice source I certainly profitted from was the YouTube video by Edward Frenkel of Berkley.
See his website http://math.berkeley.edu/ frenkel/ for links.
Example 6.5.5. Find points on the circle x2 +y 2 = 1 which are closest to the parabola y 2 = 2(4−x).
Example 6.5.6. Find points on x2 + y 2 + z 2 = 1 and plane x + y + z = 3 which are closest.
Notice that on page 116 of Edwards he derives this as a mere special case of the fascinatingly general
Example 10 of that section.
Chapter 7
critical point analysis for several

variables
In the typical calculus sequence you learn the first and second derivative tests in calculus I. Then
in calculus II you learn about power series and Taylor’s Theorem. Finally, in calculus III, in many
popular texts, you learn an essentially ad-hoc procedure for judging the nature of critical points
as minimum, maximum or saddle. These topics are easily seen as disconnected events. In this
chapter, we connect them. We learn that the geometry of quadratic forms is ellegantly revealed by
eigenvectors and more than that this geometry is precisely what elucidates the proper classifications
of critical points of multivariate functions with real values.
7.1 multivariate power series

We set aside the issue of convergence for now. We will suppose the series discussed in this section
exist on and converge on some domain, but we do not seek to treat that topic here. Our focus is
computational. How should we calculate the Taylor series for f (x, y) at (a, b)? Or, what about
f (x) at xo ∈ Rn ?.
7.1.1 taylor’s polynomial for one-variable

If f : U ⊆ R → R is analytic at xo ∈ U then we can write
∞
X f (n) (xo )
1
f (x) = f (xo ) + f 0 (xo )(x − xo ) + f 00 (xo )(x − xo )2 + · · · = (x − xo )n
2 n!
n=0
d
We could write this in terms of the operator D = dt and the evaluation of t = xo
∞
X
1 n n
f (x) = (x − t) D f (t) =
n! t=xo
n=0
161
162 CHAPTER 7. CRITICAL POINT ANALYSIS FOR SEVERAL VARIABLES
I remind the reader that a function is called entire if it is analytic on all of R, for example ex , cos(x)
and sin(x) are all entire. In particular, you should know that:
∞
1 X 1
e x = 1 + x + x2 + · · · = xn
2 n!
n=0
X (−1)n ∞
1 1
cos(x) = 1 − x2 + x4 · · · = x2n
2 4! (2n)!
n=0
∞
1 3 1 X (−1)n 2n+1
sin(x) = x − x + x5 · · · = x
3! 5! (2n + 1)!
n=0
Since ex = cosh(x) + sinh(x) it also follows that

∞
1 1 X 1
cosh(x) = 1 + x2 + x4 · · · = x2n
2 4! (2n)!
n=0
∞
1 3 1 X 1
sinh(x) = x + x + x5 · · · = x2n+1
3! 5! (2n + 1)!
n=0
The geometric series is often useful, for a, r ∈ R with |r| < 1 it is known
∞
X a
a + ar + ar2 + · · · = arn =
1−r
n=0
This generates a whole host of examples, for instance:

1
= 1 − x2 + x4 − x6 + · · ·
1 + x2
1
= 1 + x3 + x6 + x9 + · · ·
1 − x3
x3
= x3 (1 + 2x + (2x)2 + · · · ) = x3 + 2x4 + 4x5 + · · ·
1 − 2x
Moreover, the term-by-term integration and differentiation theorems yield additional results in
conjuction with the geometric series:
∞ ∞
(−1)n 2n+1
Z Z X
−1 dx X 1 1
tan (x) = = (−1)n x2n dx = x = x − x3 + x5 + · · ·
1 + x2 2n + 1 3 5
n=0 n=0
∞ ∞
−1 −1 n+1
Z Z Z X
d X
ln(1 − x) = ln(1 − x)dx = dx = − xn dx = x
dx 1−x n+1
n=0 n=0
7.1. MULTIVARIATE POWER SERIES 163
Of course, these are just the basic building blocks. We also can twist things and make the student
use algebra,
1
ex+2 = ex e2 = e2 (1 + x + x2 + · · · )
2
or trigonmetric identities,
sin(x) = sin(x − 2 + 2) = sin(x − 2) cos(2) + cos(x − 2) sin(2)

∞ ∞
X (−1)n 2n+1
X (−1)n
⇒ sin(x) = cos(2) (x − 2) + sin(2) (x − 2)2n .
(2n + 1)! (2n)!
n=0 n=0
Feel free to peruse my most recent calculus II materials to see a host of similarly sneaky calculations.
7.1.2 taylor’s multinomial for two-variables

Suppose we wish to find the taylor polynomial centered at (0, 0) for f (x, y) = ex sin(y). It is a
simple as this:

1 2 1 3 1 2 1
f (x, y) = 1 + x + x + · · · y − y + · · · = y + xy + x y − y3 + · · ·
2 6 2 6
the resulting expression is called a multinomial since it is a polynomial in multiple variables. If

all functions f (x, y) could be written as f (x, y) = F (x)G(y) then multiplication of series known
from calculus II would often suffice. However, many functions do not possess this very special
form. For example, how should we expand f (x, y) = cos(xy) about (0, 0)?. We need to derive the
two-dimensional Taylor’s theorem.
We already know Taylor’s theorem for functions on R,

1 1
g(x) = g(a) + g 0 (a)(x − a) + g 00 (a)(x − a)2 + · · · + g (k) (a)(x − a)k + Rk
2 k!
and... If the remainder term vanishes as k → ∞ then the function g is represented by the Taylor
series given above and we write:
∞
X 1 (k)
g(x) = g (a)(x − a)k .
k!
k=0
Consider the function of two variables f : U ⊆ R2 → R which is smooth with smooth partial
derivatives of all orders. Furthermore, let (a, b) ∈ U and construct a line through (a, b) with
direction vector (h1 , h2 ) as usual:
φ(t) = (a, b) + t(h1 , h2 ) = (a + th1 , b + th2 )
for t ∈ R. Note φ(0) = (a, b) and φ0 (t) = (h1 , h2 ) = φ0 (0). Construct g = f ◦ φ : R → R and
choose dom(g) such that φ(t) ∈ U for t ∈ dom(g). This function g is a real-valued function of a
real variable and we will be able to apply Taylor’s theorem from calculus II on g. However, to
differentiate g we’ll need tools from calculus III to sort out the derivatives. In particular, as we
differentiate g, note we use the chain rule for functions of several variables:
g 0 (t) = (f ◦ φ)0 (t) = f 0 (φ(t))φ0 (t)

= ∇f (φ(t)) · (h1 , h2 )
= h1 fx (a + th1 , b + th2 ) + h2 fy (a + th1 , b + th2 )
Note g 0 (0) = h1 fx (a, b) + h2 fy (a, b). Differentiate again (I omit (φ(t)) dependence in the last steps),
g 00 (t) = h1 fx0 (a + th1 , b + th2 ) + h2 fy0 (a + th1 , b + th2 )

= h1 ∇fx (φ(t)) · (h1 , h2 ) + h2 ∇fy (φ(t)) · (h1 , h2 )
= h21 fxx + h1 h2 fyx + h2 h1 fxy + h22 fyy
= h21 fxx + 2h1 h2 fxy + h22 fyy
Thus, making explicit the point dependence, g 00 (0) = h21 fxx (a, b) + 2h1 h2 fxy (a, b) + h22 fyy (a, b). We
may construct the Taylor series for g up to quadratic terms:
1
g(0 + t) = g(0) + tg 0 (0) + g 00 (0) + · · ·
2
t2 2
h1 fxx (a, b) + 2h1 h2 fxy (a, b) + h22 fyy (a, b) + · · ·

= f (a, b) + t[h1 fx (a, b) + h2 fy (a, b)] +
2
Note that g(t) = f (a + th1 , b + th2 ) hence g(1) = f (a + h1 , b + h2 ) and consequently,
f (a + h1 , b + h2 ) = f (a, b) + h1 fx (a, b) + h2 fy (a, b)+

1 2 2
+ h1 fxx (a, b) + 2h1 h2 fxy (a, b) + h2 fyy (a, b) + · · ·
2
Omitting point dependence on the 2nd derivatives,

1
h21 fxx + 2h1 h2 fxy + h22 fyy + · · ·

f (a + h1 , b + h2 ) = f (a, b) + h1 fx (a, b) + h2 fy (a, b) + 2
Sometimes we’d rather have an expansion about (x, y). To obtain that formula simply substitute
x − a = h1 and y − b = h2 . Note that the point (a, b) is fixed in this discussion so the derivatives
are not modified in this substitution,
f (x, y) = f (a, b) + (x − a)fx (a, b) + (y − b)fy (a, b)+

1 2 2
+ (x − a) fxx (a, b) + 2(x − a)(y − b)fxy (a, b) + (y − b) fyy (a, b) + · · ·
2
At this point we ought to recognize the first three terms give the tangent plane to z = f (z, y) at
(a, b, f (a, b)). The higher order terms are nonlinear corrections to the linearization, these quadratic
terms form a quadratic form. If we computed third, fourth or higher order terms we will find that,
using a = a1 and b = a2 as well as x = x1 and y = x2 ,
∞ X
2 X
2 2
X X 1 ∂ (n) f (a1 , a2 )
f (x, y) = ··· (xi − ai1 )(xi2 − ai2 ) · · · (xin − ain )
n! ∂xi1 ∂xi2 · · · ∂xin 1
n=0 i1 =0 i2 =0 in =0
Example 7.1.1. Expand f (x, y) = cos(xy) about (0, 0). We calculate derivatives,
fx = −y sin(xy) fy = −x sin(xy)
fxx = −y 2 cos(xy) fxy = − sin(xy) − xy cos(xy) fyy = −x2 cos(xy)

fxxx = y 3 sin(xy) fxxy = −y cos(xy) − y cos(xy) + xy 2 sin(xy)
fxyy = −x cos(xy) − x cos(xy) + x2 y sin(xy) fyyy = x3 sin(xy)
Next, evaluate at x = 0 and y = 0 to find f (x, y) = 1 + · · · to third order in x, y about (0, 0). We
can understand why these derivatives are all zero by approaching the expansion a different route:
simply expand cosine directly in the variable (xy),
1 1 1 1
f (x, y) = 1 − (xy)2 + (xy)4 + · · · = 1 − x2 y 2 + x4 y 4 + · · · .
2 4! 2 4!
Apparently the given function only has nontrivial derivatives at (0, 0) at orders 0, 4, 8, .... We can
deduce that fxxxxy (0, 0) = 0 without furthter calculation.
This is actually a very interesting function, I think it defies our analysis in the later portion of this
chapter. The second order part of the expansion reveals nothing about the nature of the critical
point (0, 0). Of course, any student of trigonometry should recognize that f (0, 0) = 1 is likely
a local maximum, it’s certainly not a local minimum. The graph reveals that f (0, 0) is a local
maxium for f restricted to certain rays from the origin whereas it is constant on several special
directions (the coordinate axes).
And, if you were wondering, yes, we could also derive this from subsitution of u = xy into the
standard expansion for cos(u) = 1 − 12 u2 + 4!1 u4 + · · · . Often such subsitutions are the quickest way
to generate interesting examples.
7.1.3 taylor’s multinomial for many-variables

Suppose f : dom(f ) ⊆ Rn → R is a function of n-variables and we seek to derive the Taylor series
centered at a = (a1 , a2 , . . . , an ). Once more consider the composition of f with a line in dom(f ).
In particular, let φ : R → Rn be defined by φ(t) = a + th where h = (h1 , h2 , . . . , hn ) gives the
direction of the line and clearly φ0 (t) = h. Let g : dom(g) ⊆ R → R be defined by g(t) = f (φ(t))
for all t ∈ R such that φ(t) ∈ dom(f ). P Differentiate, use the multivariate chain rule, recall here
that ∇ = e1 ∂x∂ 1 + e2 ∂x∂ 2 + · · · + en ∂x∂n = ni=1 ei ∂i ,
n
X
g 0 (t) = ∇f (φ(t)) · φ0 (t) = ∇f (φ(t)) · h = hi (∂i f )(φ(t))
i=1
If we omit the explicit dependence on φ(t) then we find the simple formula g 0 (t) = ni=1 hi ∂i f .
P
Differentiate a second time,
n Xn Xn
00 d X d
hi ∇∂i f (φ(t)) · φ0 (t)

g (t) = hi ∂i f (φ(t)) = hi ∂i f (φ(t)) =
dt dt
i=1 i=1 i=1
Omitting the φ(t) dependence and once more using φ0 (t) = h we find
n
X
00
g (t) = hi ∇∂i f · h
i=1
Pn
Recall that ∇ = j=1 ej ∂j and expand the expression above,
n
X n
X n X
X n
00
g (t) = hi ej ∂j ∂i f ·h= hi hj ∂j ∂i f
i=1 j=1 i=1 j=1
where we should remember ∂j ∂i f depends on φ(t). It should be clear that if we continue and take
k-derivatives then we will obtain:
n X
X n n
X
g (k) (t) = ··· hi1 hi2 · · · hik ∂i1 ∂i2 · · · ∂ik f
i1 =1 i2 =1 ik =1
More explicitly,
n X
X n n
X
(k)
g (t) = ··· hi1 hi2 · · · hik (∂i1 ∂i2 · · · ∂ik f )(φ(t))
i1 =1 i2 =1 ik =1
Hence, by Taylor’s theorem, provided we are sufficiently close to t = 0 as to bound the remainder1
∞ n n n
X 1 XX X
g(t) = ··· hi1 hi2 · · · hik (∂i1 ∂i2 · · · ∂ik f )(φ(t)) tk
k!
k=0 i1 =1 i2 =1 ik =1
1
there exist smooth examples for which no neighborhood is small enough, the bump function in one-variable has
higher-dimensional analogues, we focus our attention to functions for which it is possible for the series below to
converge
1
Recall that g(t) = f (φ(t)) = f (a + th). Put2 t = 1 and bring in the k! to derive
∞ X
n X
n n
X X 1
f (a + h) = ··· ∂i1 ∂i2 · · · ∂ik f (a) hi1 hi2 · · · hik .
k!
k=0 i1 =1 i2 =1 ik =1
Naturally, we sometimes prefer to write the series expansion about a as an expresssion in x = a + h.

With this substitution we have h = x − a and hij = (x − a)ij = xij − aij thus
∞ X
n X
n n
X X 1
f (x) = ··· ∂i1 ∂i2 · · · ∂ik f (a) (xi1 − ai1 )(xi2 − ai2 ) · · · (xik − aik ).
k!
k=0 i1 =1 i2 =1 ik =1
Example 7.1.2. Suppose f : R3 → R let’s unravel the Taylor series centered at (0, 0, 0) from the
general formula boxed above. Utilize the notation x = x1 , y = x2 and z = x3 in this example.
∞ X
3 X
3 3
X X 1
f (x) = ··· ∂i1 ∂i2 · · · ∂ik f (0) xi1 xi2 · · · xik .
k!
k=0 i1 =1 i2 =1 ik =1
The terms to order 2 are as follows:
f (x) = f (0)
+ fx (0)x + fy (0)y + fz (0)z
+ 12 fxx (0)x2 + fyy (0)y 2 + fzz (0)z 2 +

+fxy (0)xy + fxz (0)xz + fyz (0)yz + fyx (0)yx + fzx (0)zx + fzy (0)zy + ···
Partial derivatives commute for smooth functions hence,
f (x) = f (0)
+ fx (0)x + fy (0)y + fz (0)z
+ 21 fxx (0)x2 + fyy (0)y 2 + fzz (0)z 2 + 2fxy (0)xy + 2fxz (0)xz + 2fyz (0)yz

1
+ 3! fxxx (0)x3 + fyyy (0)y 3 + fzzz (0)z 3 + 3fxxy (0)x2 y + 3fxxz (0)x2 z

+3fyyz (0)y 2 z + 3fxyy (0)xy 2 + 3fxzz (0)xz 2 + 3fyzz (0)yz 2 + 6fxyz (0)xyz + ···
Example 7.1.3. Suppose f (x, y, z) = exyz . Find a quadratic approximation to f near (0, 1, 2).
Observe:
fx = yzexyz fy = xzexyz fz = xyexyz
fxx = (yz)2 exyz fyy = (xz)2 exyz fzz = (xy)2 exyz
fxy = zexyz + xyz 2 exyz fyz = xexyz + x2 yzexyz fxz = yexyz + xy 2 zexyz
2
if t = 1 is not in the domain of g then we should rescale the vector h so that t = 1 places φ(1) in dom(f ), if f is
smooth on some neighborhood of a then this is possible
Evaluating at x = 0, y = 1 and z = 2,
fx (0, 1, 2) = 2 fy (0, 1, 2) = 0 fz (0, 1, 2) = 0
fxx (0, 1, 2) = 4 fyy (0, 1, 2) = 0 fzz (0, 1, 2) = 0

fxy (0, 1, 2) = 2 fyz (0, 1, 2) = 0 fxz (0, 1, 2) = 1
Hence, as f (0, 1, 2) = e0 = 1 we find
f (x, y, z) = 1 + 2x + 2x2 + 2x(y − 1) + 2x(z − 2) + · · ·
Another way to calculate this expansion is to make use of the adding zero trick,
1 2
f (x, y, z) = ex(y−1+1)(z−2+2) = 1 + x(y − 1 + 1)(z − 2 + 2) + x(y − 1 + 1)(z − 2 + 2) + · · ·
2
Keeping only terms with two or less of x, (y − 1) and (z − 2) variables,
1
f (x, y, z) = 1 + 2x + x(y − 1)(2) + x(1)(z − 2) + x2 (1)2 (2)2 + · · ·
2
Which simplifies once more to f (x, y, z) = 1 + 2x + 2x(y − 1) + x(z − 2) + 2x2 + · · · .
7.2. A BRIEF INTRODUCTION TO THE THEORY OF QUADRATIC FORMS 169
7.2 a brief introduction to the theory of quadratic forms

Definition 7.2.1.
Generally, a quadratic form Q is a function Q : Rn → R whose formula can be written
T A~
Q(~x) = ~x x forall ~x ∈ Rn where A ∈ R n×n such that AT = A. In particular, if ~x = (x, y)
a b
and A = then
b c
Q(~x) = ~xT A~x = ax2 + bxy + byx + cy 2 = ax2 + 2bxy + y 2 .
The n = 3 case is similar,denote A = [Aij ] and ~x = (x, y, z) so that
Q(~x) = ~xT A~x = A11 x2 + 2A12 xy + 2A13 xz + A22 y 2 + 2A23 yz + A33 z 2 .
Generally, if [Aij ] ∈ R n×n and ~x = [xi ]T then the associated quadratic form is
X n
X X
Q(~x) = ~xT A~x = Aij xi xj = Aii x2i + 2Aij xi xj .
i,j i=1 i<j
In case you wondering, yes you could write a given quadratic form with a different matrix which
is not symmetric, but we will find it convenient to insist that our matrix is symmetric since that
choice is always possible for a given quadratic form.
It is at times useful to use the dot-product to express a given quadratic form:
~xT A~x = ~x · (A~x) = (A~x) · ~x = ~xT AT ~x
Some texts actually use the middle equality above to define a symmetric matrix.
Example 7.2.2.
2 1 x
2x2 + 2xy + 2y 2 =

x y
1 2 y
Example 7.2.3.
  
2 1 3/2 x
2x2 + 2xy + 3xz − 2y 2 − z 2 = x y z  1 −2 0   y 

3/2 0 −1 z
Proposition 7.2.4.
The values of a quadratic form on Rn − {0} is completely determined by it’s values on

the (n − 1)-sphere Sn−1 = {~x ∈ Rn | ||~x|| = 1}. In particular, Q(~x) = ||~x||2 Q(x̂) where
x̂ = ||~x1|| ~x.
Proof: Let Q(~x) = ~xT A~x. Notice that we can write any nonzero vector as the product of its
magnitude ||x|| and its direction x̂ = ||~x1|| ~x,
Q(~x) = Q(||~x||x̂) = (||~x||x̂)T A||~x||x̂ = ||~x||2 x̂T Ax̂ = ||x||2 Q(x̂).

Therefore Q(~x) is simply proportional to Q(x̂) with proportionality constant ||~x||2 .
The proposition above is very interesting. It says that if we know how Q works on unit-vectors then
we can extrapolate its action on the remainder of Rn . If f : S → R then we could say f (S) > 0
iff f (s) > 0 for all s ∈ S. Likewise, f (S) < 0 iff f (s) < 0 for all s ∈ S. The proposition below
follows from the proposition above since ||~x||2 ranges over all nonzero positive real numbers in the
equations above.
Proposition 7.2.5.
If Q is a quadratic form on Rn and we denote Rn∗ = Rn − {0}
1.(negative definite) Q(Rn∗ ) < 0 iff Q(Sn−1 ) < 0
2.(positive definite) Q(Rn∗ ) > 0 iff Q(Sn−1 ) > 0
3.(non-definite) Q(Rn∗ ) = R − {0} iff Q(Sn−1 ) has both positive and negative values.
Before I get too carried away with the theory let’s look at a couple examples.
Example 7.2.6. Consider the quadric form Q(x, y) = x2 + y 2 . You can check for yourself that
z = Q(x, y) is a cone and Q has positive outputs for all inputs except (0, 0). Notice that Q(v) = ||v||2
so it is clear that Q(S1 ) = 1. We find agreement with the preceding proposition. Next, √ think about
2 2
the application of Q(x, y) to level curves; x + y = k is simply a circle of radius k or just the
origin. Here’s a graph of z = Q(x, y):
Notice that Q(0, 0) = 0 is the

absolute minimum for Q. Finally, let’s take a moment to write
1 0 x
Q(x, y) = [x, y] in this case the matrix is diagonal and we note that the e-values are
0 1 y
λ1 = λ2 = 1.
Example 7.2.7. Consider the quadric form Q(x, y) = x2 − 2y 2 . You can check for yourself
that z = Q(x, y) is a hyperboloid and Q has non-definite outputs since sometimes the x2 term
dominates whereas other points have −2y 2 as the dominent term. Notice that Q(1, 0) = 1 whereas
Q(0, 1) = −2 hence we find Q(S1 ) contains both positive and negative values and consequently we
find agreement with the preceding proposition. Next, think about the application of Q(x, y) to level
curves; x2 − 2y 2 = k yields either hyperbolas which open vertically (k > 0) or horizontally (k < 0)
or a pair of lines y = ± x2 in the k = 0 case. Here’s a graph of z = Q(x, y):

1 0 x
The origin is a saddle point. Finally, let’s take a moment to write Q(x, y) = [x, y]
0 −2 y
in this case the matrix is diagonal and we note that the e-values are λ1 = 1 and λ2 = −2.
Example 7.2.8. Consider the quadric form Q(x, y) = 3x2 . You can check for yourself that z =
Q(x, y) is parabola-shaped trough along the y-axis. In this case Q has positive outputs for all inputs
except (0, y), we would call this form positive semi-definite. A short calculation reveals that
Q(S1 ) = [0, 3] thus we again find agreement with the preceding proposition (case 3). Next, p think
about the application of Q(x, y) to level curves; 3x2 = k is a pair of vertical lines: x = ± k/3 or
just the y-axis. Here’s a graph of z = Q(x, y):

3 0 x
Finally, let’s take a moment to write Q(x, y) = [x, y] in this case the matrix is
0 0 y
diagonal and we note that the e-values are λ1 = 3 and λ2 = 0.
Example 7.2.9. Consider the quadric form Q(x, y, z) = x2 +2y 2 +3z 2 . Think about the application
of Q(x, y, z) to level surfaces; x2 + 2y 2 + 3z 2 = k is an ellipsoid. I can’t graph a function of three
variables, however, we can look at level surfaces of the function. I use Mathematica to plot several
below:
 

1 0 0 x
Finally, let’s take a moment to write Q(x, y, z) = [x, y, z]  0 2 0   y  in this case the matrix
0 0 3 z
is diagonal and we note that the e-values are λ1 = 1 and λ2 = 2 and λ3 = 3.
7.2.1 diagonalizing forms via eigenvectors

The examples given thus far are the simplest cases. We don’t really need linear algebra to un-
derstand them. In contrast, e-vectors and e-values will prove a useful tool to unravel the later
examples3
Definition 7.2.10.
Let A ∈ R n×n . If v ∈ R n×1 is nonzero and Av = λv for some λ ∈ C then we say v is an

eigenvector with eigenvalue λ of the matrix A.
Proposition 7.2.11.
Let A ∈ R n×n then λ is an eigenvalue of A iff det(A − λI) = 0. We say P (λ) = det(A − λI)
the characteristic polynomial and det(A − λI) = 0 is the characteristic equation.
Proof: Suppose λ is an eigenvalue of A then there exists a nonzero vector v such that Av = λv
which is equivalent to Av − λv = 0 which is precisely (A − λI)v = 0. Notice that (A − λI)0 = 0
3
this is the one place in this course where we need eigenvalues and eigenvector calculations, I include these to
illustrate the structure of quadratic forms in general, however, as linear algebra is not a prerequisite you may find some
things in this section mysterious. The homework and study guide will elaborate on what is required this semester
thus the matrix (A − λI) is singular as the equation (A − λI)x = 0 has more than one solution.
Consequently det(A − λI) = 0.
Conversely, suppose det(A − λI) = 0. It follows that (A − λI) is singular. Clearly the system
(A − λI)x = 0 is consistent as x = 0 is a solution hence we know there are infinitely many solu-
tions. In particular there exists at least one vector v 6= 0 such that (A − λI)v = 0 which means the
vector v satisfies Av = λv. Thus v is an eigenvector with eigenvalue λ for A.

3 1
Example 7.2.12. Let A = find the e-values and e-vectors of A.
3 1

3−λ 1
det(A − λI) = det = (3 − λ)(1 − λ) − 3 = λ2 − 4λ = λ(λ − 4) = 0
3 1−λ
We find λ1 = 0 and λ2 = 4. Now find the e-vector with e-value λ1 = 0, let u1 = [u, v]T denote the
e-vector we wish to find. Calculate,

3 1 u 3u + v 0
(A − 0I)u1 = = =
3 1 v 3u + v 0
Obviously the equations above are redundant and we have infinitely

many
solutions
of the form
u 1
3u + v = 0 which means v = −3u so we can write, u1 = =u . In applications we
−3u −3
often make a choice to select a particular e-vector. Most modern graphing calculators can calcu-
late e-vectors. It is customary for the e-vectors to be chosen to have length one. That is a useful
choice forcertain
applications as we will later discuss. If you use a calculator it would likely give
1 √
u1 = √110 although the 10 would likely be approximated unless your calculator is smart.
−3
Continuing we wish to find eigenvectors u2 = [u, v]T such that (A − 4I)u2 = 0. Notice that u, v
are disposable variables in this context, I do not mean to connect the formulas from the λ = 0 case
with the case considered now.

−1 1 u −u + v 0
(A − 4I)u1 = = =
3 −3 v 3u − 3v 0
Againthe equations
are
redundant and we have infinitely many solutions of the form v = u. Hence,
u 1
u2 = =u is an eigenvector for any u ∈ R such that u 6= 0.
u 1
Theorem 7.2.13.
A matrix A ∈ R n×n is symmetric iff there exists an orthonormal eigenbasis for A.

There is a geometric proof of this theorem in Edwards4 (see Theorem 8.6 pgs 146-147) . I prove half
of this theorem in my linear algebra notes by a non-geometric argument (full proof is in Appendix C
of Insel,Spence and Friedberg). It might be very interesting to understand the connection between
the geometric verse algebraic arguments. We’ll content ourselves with an example here:
 
0 0 0
Example 7.2.14. Let A =  0 1 2 . Observe that det(A − λI) = −λ(λ + 1)(λ − 3) thus λ1 =
0 2 1
0, λ2 = −1, λ3 = 3. We can calculate orthonormal e-vectors of v1 = [1, 0, 0]T , v2 = √12 [0, 1, −1]T
and v3 = √1 [0, 1, 1]T . I invite the reader to check the validity of the following equation:
2
   
1 0 0 1 0 0
 
0 0 0 0 0 0
 0 √1 −1
√  0 √1 √1
 0 1 2   = 0 −1 0 
   
 2 2 2 2
√1 √1 −1 √1
0 2 2
0 2 1 0 √
2 2
0 0 3
Its really neat that

 to find the 
inverse
 of a matrix
 of orthonormal e-vectors we need only take the
1 0 0 1 0 0
 
1 0 0
1 −1 1 1
transpose; note  0 √2 √2   0 √2 √2  =  0 1 0 .
  
1 1 −1 1
0 √2 √2 0 √2 √2 0 0 1
Proposition 7.2.15.
If Q is a quadratic form on Rn with matrix A and e-values λ1 , λ2 , . . . , λn with orthonormal

e-vectors v1 , v2 , . . . , vn then
Q(vi ) = λi 2
for i = 1, 2, . . . , n. Moreover, if P = [v1 |v2 | · · · |vn ] then
Q(~x) = (P T ~x)T P T AP P T ~x = λ1 y12 + λ2 y22 + · · · + λn yn2
where we defined ~y = P T ~x.

Let me restate the proposition above in simple terms: we can transform a given quadratic form to
a diagonal form by finding orthonormalized e-vectors and performing the appropriate coordinate
transformation. Since P is formed from orthonormal e-vectors we know that P will be either a
rotation or reflection. This proposition says we can remove ”cross-terms” by transforming the
quadratic forms with an appropriate rotation.
Example 7.2.16. Consider the quadric form Q(x, y) = 2x2 + 2xy + 2y 2 . It’s not immediately
obvious (to me) what the level curves Q(x, y) = k look like. We’ll make
use of the preceding
2 1 x
proposition to understand those graphs. Notice Q(x, y) = [x, y] . Denote the matrix
1 2 y
4
think about it, there is a 1-1 correspondance between symmetric matrices and quadratic forms
of the form by A and calculate the e-values/vectors:

2−λ 1
det(A − λI) = det = (λ − 2)2 − 1 = λ2 − 4λ + 3 = (λ − 1)(λ − 3) = 0
1 2−λ
Therefore, the e-values are λ1 = 1 and λ2 = 3.

1 1 u 0 1 1
(A − I)~u1 = = ⇒ ~u1 = √
1 1 v 0 2 −1
I just solved u + v = 0 to give v = −u choose u = 1 then normalize to get the vector above. Next,

−1 1 u 0 1 1
(A − 3I)~u2 = = ⇒ ~u2 = √
1 −1 v 0 2 1
I just solved u − v = 0 to give v = u choose u = 1 then normalize to get the vector above. Let
P = [~u1 |~u2 ] and introduce new coordinates ~y = [x̄, ȳ]T defined by ~y = P T ~x. Note these can be
inverted by multiplication by P to give ~x = P ~y . Observe that
x = 21 (x̄ + ȳ) x̄ = 21 (x − y)

1 1 1
P = ⇒ or
2 −1 1 y = 21 (−x̄ + ȳ) ȳ = 12 (x + y)
The proposition preceding this example shows that substitution of the formulas above into Q yield5 :
Q̃(x̄, ȳ) = x̄2 + 3ȳ 2
It is clear that in the barred coordinate system the level curve Q(x, y) = k is an ellipse. If we draw
the barred coordinate system superposed over the xy-coordinate system then you’ll see that the graph
of Q(x, y) = 2x2 + 2xy + 2y 2 = k is an ellipse rotated by 45 degrees. Or, if you like, we can plot
z = Q(x, y):
5
technically Q̃(x̄, ȳ) is Q(x(x̄, ȳ), y(x̄, ȳ))
Example 7.2.17. Consider the quadric form Q(x, y) = x2 + 2xy + y 2 . It’s not immediately obvious
(to me) what the level curves Q(x, y) = k look like.
We’ll make use of the preceding proposition to
1 1 x
understand those graphs. Notice Q(x, y) = [x, y] . Denote the matrix of the form by
1 1 y
A and calculate the e-values/vectors:

1−λ 1
det(A − λI) = det = (λ − 1)2 − 1 = λ2 − 2λ = λ(λ − 2) = 0
1 1−λ
Therefore, the e-values are λ1 = 0 and λ2 = 2.

1 1 u 0 1 1
(A − 0)~u1 = = ⇒ ~u1 = √
1 1 v 0 2 −1

−1 1 u 0 1 1
(A − 2I)~u2 = = ⇒ ~u2 = √
1 −1 v 0 2 1
x = 21 (x̄ + ȳ) x̄ = 21 (x − y)

1 1 1
P = ⇒ 1 or
2 −1 1 y = 2 (−x̄ + ȳ) ȳ = 12 (x + y)
The proposition preceding this example shows that substitution of the formulas above into Q yield:
Q̃(x̄, ȳ) = 2ȳ 2
It is clear that in the barred coordinate system the level curve Q(x, y) = k is a pair of paralell
lines. If we draw the barred coordinate system superposed over the xy-coordinate system then you’ll
see that the graph of Q(x, y) = x2 + 2xy + y 2 = k is a line with slope −1. Indeed, with a little
algebraic√insight we could 2
√ have anticipated this result since Q(x, y) = (x+y) so Q(x, y) = k implies
x + y = k thus y = k − x. Here’s a plot which again verifies what we’ve already found:
Example 7.2.18. Consider the quadric form Q(x, y) = 4xy. It’s not immediately obvious (to
me) what the level curves Q(x, y) = k look like. We’llmake
use of the preceding proposition to
0 2 x
understand those graphs. Notice Q(x, y) = [x, y] . Denote the matrix of the form by
0 2 y
A and calculate the e-values/vectors:

−λ 2
det(A − λI) = det = λ2 − 4 = (λ + 2)(λ − 2) = 0
2 −λ
Therefore, the e-values are λ1 = −2 and λ2 = 2.

2 2 u 0 1 1
(A + 2I)~u1 = = ⇒ ~u1 = √
2 2 v 0 2 −1

−2 2 u 0 1 1
(A − 2I)~u2 = = ⇒ ~u2 = √
2 −2 v 0 2 1
x = 21 (x̄ + ȳ) x̄ = 21 (x − y)

1 1 1
P = ⇒ 1 or
2 −1 1 y = 2 (−x̄ + ȳ) ȳ = 12 (x + y)
Q̃(x̄, ȳ) = −2x̄2 + 2ȳ 2
It is clear that in the barred coordinate system the level curve Q(x, y) = k is a hyperbola. If we
draw the barred coordinate system superposed over the xy-coordinate system then you’ll see that
the graph of Q(x, y) = 4xy = k is a hyperbola rotated by 45 degrees. The graph z = 4xy is thus a
hyperbolic paraboloid:
The fascinating thing about the mathematics here is that if you don’t want to graph z = Q(x, y),
but you do want to know the general shape then you can determine which type of quadraic surface
you’re dealing with by simply calculating the eigenvalues of the form.
Remark 7.2.19.
I made the preceding triple of examples all involved the same rotation. This is purely for my
lecturing convenience. In practice the rotation could be by all sorts of angles. In addition,
you might notice that a different ordering of the e-values would result in a redefinition of
the barred coordinates. 6
We ought to do at least one 3-dimensional example.
Example 7.2.20. Consider the quadric form Q defined below:

  
6 −2 0 x
Q(x, y, z) = [x, y, z] −2
 6 0   y 
0 0 5 z
Denote the matrix of the form by A and calculate the e-values/vectors:

 
6 − λ −2 0
det(A − λI) = det  −2 6 − λ 0 
0 0 5−λ
= [(λ − 6)2 − 4](5 − λ)
= (5 − λ)[λ2 − 12λ + 32](5 − λ)
= (λ − 4)(λ − 8)(5 − λ)
Therefore, the e-values are λ1 = 4, λ2 = 8 and λ3 = 5. After some calculation we find the following
orthonormal e-vectors for A:
     
1 1 0
1   1 
~u1 = √ 1 ~u2 = √ −1  ~u3 =  0 
2 0 2 0 1
Let P = [~u1 |~u2 |~u3 ] and introduce new coordinates ~y = [x̄, ȳ, z̄]T defined by ~y = P T ~x. Note these
can be inverted by multiplication by P to give ~x = P ~y . Observe that
x = 12 (x̄ + ȳ) = 12 (x − y)
 
1 1 0 x̄
1 
P =√ −1 1 √0  ⇒ y = 12 (−x̄ + ȳ) or ȳ = 12 (x + y)
2 0 0 2 z = z̄ z̄ = z
Q̃(x̄, ȳ, z̄) = 4x̄2 + 8ȳ 2 + 5z̄ 2

7.3. SECOND DERIVATIVE TEST IN MANY-VARIABLES 179
It is clear that in the barred coordinate system the level surface Q(x, y, z) = k is an ellipsoid. If we
draw the barred coordinate system superposed over the xyz-coordinate system then you’ll see that
the graph of Q(x, y, z) = k is an ellipsoid rotated by 45 degrees around the z − axis. Plotted below
are a few representative ellipsoids:
In summary, the behaviour of a quadratic form Q(x) = xT Ax is governed by it’s set of eigenvalues7
{λ1 , λ2 , . . . , λk }. Moreover, the form can be written as Q(y) = λ1 y12 + λ2 y22 + · · · + λk yk2 by choosing
the coordinate system which is built from the orthonormal eigenbasis of col(A). In this coordinate
system the shape of the level-sets of Q becomes manifest from the signs of the e-values. )
Remark 7.2.21.
If you would like to read more about conic sections or quadric surfaces and their connection
to e-values/vectors I reccommend sections 9.6 and 9.7 of Anton’s linear algebra text. I
have yet to add examples on how to include translations in the analysis. It’s not much
more trouble but I decided it would just be an unecessary complication this semester.
Also, section 7.1,7.2 and 7.3 in Lay’s linear algebra text show a bit more about how to
use this math to solve concrete applied problems. You might also take a look in Gilbert
Strang’s linear algebra text, his discussion of tests for positive-definite matrices is much
more complete than I will give here.
7.3 second derivative test in many-variables

There is a connection between the shape of level curves Q(x1 , x2 , . . . , xn ) = k and the graph xn+1 =
f (x1 , x2 , . . . , xn ) of f . I’ll discuss n = 2 but these comments equally well apply to w = f (x, y, z) or
higher dimensional examples. Consider a critical point (a, b) for f (x, y) then the Taylor expansion
about (a, b) has the form
f (a + h, b + k) = f (a, b) + Q(h, k)
where Q(h, k) = 21 h2 fxx (a, b) + hkfxy (a, b) + 12 h2 fyy (a, b) = [h, k][Q](h, k). Since [Q]T = [Q] we can
find orthonormal e-vectors ~u1 , ~u2 for [Q] with e-values λ1 and λ2 respective. Using U = [~u1 |~u2 ] we
7
this set is called the spectrum of the matrix
can introduce rotated coordinates (h̄, k̄) = U (h, k). These will give
Q(h̄, k̄) = λ1 h̄2 + λ2 k̄ 2
Clearly if λ1 > 0 and λ2 > 0 then f (a, b) yields the local minimum whereas if λ1 < 0 and λ2 < 0
then f (a, b) yields the local maximum. Edwards discusses these matters on pgs. 148-153. In short,
supposing f ≈ f (p) + Q, if all the e-values of Q are positive then f has a local minimum of f (p)
at p whereas if all the e-values of Q are negative then f reaches a local maximum of f (p) at p.
Otherwise Q has both positive and negative e-values and we say Q is non-definite and the function
has a saddle point. If all the e-values of Q are positive then Q is said to be positive-definite
whereas if all the e-values of Q are negative then Q is said to be negative-definite. Edwards
gives a few nice tests for ascertaining if a matrix is positive definite without explicit computation
of e-values. Finally, if one of the e-values is zero then the graph will be like a trough.
Example 7.3.1. Suppose f (x, y) = exp(−x2 − y 2 + 2y − 1) expand f about the point (0, 1):
f (x, y) = exp(−x2 )exp(−y 2 + 2y − 1) = exp(−x2 )exp(−(y − 1)2 )
expanding,
f (x, y) = (1 − x2 + · · · )(1 − (y − 1)2 + · · · ) = 1 − x2 − (y − 1)2 + · · ·
Recenter about the point (0, 1) by setting x = h and y = 1 + k so
f (h, 1 + k) = 1 − h2 − k 2 + · · ·
If (h, k) is near (0, 0) then the dominant terms are simply those we’ve written above hence the graph
is like that of a quadraic surface with a pair of negative e-values. It follows that f (0, 1) is a local
maximum. In fact, it happens to be a global maximum for this function.
Example 7.3.2. Suppose f (x, y) = 4−(x−1)2 +(y−2)2 +Aexp(−(x−1)2 −(y−2)2 )+2B(x−1)(y−2)

for some constants A, B. Analyze what values for A, B will make (1, 2) a local maximum, minimum
or neither. Expanding about (1, 2) we set x = 1 + h and y = 2 + k in order to see clearly the local
behaviour of f at (1, 2),
f (1 + h, 2 + k) = 4 − h2 − k 2 + Aexp(−h2 − k 2 ) + 2Bhk
= 4 − h2 − k 2 + A(1 − h2 − k 2 ) + 2Bhk · · ·
= 4 + A − (A + 1)h2 + 2Bhk − (A + 1)k 2 + · · ·
There is no nonzero linear term in the expansion at (1, 2) which indicates that f (1, 2) = 4 + A
may be a local extremum. In this case the quadratic terms are nontrivial which means the graph of
this function is well-approximated by a quadraic surface near (1, 2). The quadratic form Q(h, k) =
−(A + 1)h2 + 2Bhk − (A + 1)k 2 has matrix

−(A + 1) B
[Q] = .
B −(A + 1)2
The characteristic equation for Q is

−(A + 1) − λ B
det([Q] − λI) = det = (λ + A + 1)2 − B 2 = 0
B −(A + 1)2 − λ
We find solutions λ1 = −A − 1 + B and λ2 = −A − 1 − B. The possibilities break down as follows:

1. if λ1 , λ2 > 0 then f (1, 2) is local minimum.
2. if λ1 , λ2 < 0 then f (1, 2) is local maximum.
3. if just one of λ1 , λ2 is zero then f is constant along one direction and min/max along another
so technically it is a local extremum.
4. if λ1 λ2 < 0 then f (1, 2) is not a local etremum, however it is a saddle point.

In particular, the following choices for A, B will match the choices above
1. Let A = −3 and B = 1 so λ1 = 3 and λ2 = 1;
2. Let A = 3 and B = 1 so λ1 = −3 and λ2 = −5
3. Let A = −3 and B = −2 so λ1 = 0 and λ2 = 4
4. Let A = 1 and B = 3 so λ1 = 1 and λ2 = −5

Here are the graphs of the cases above, note the analysis for case 3 is more subtle for Taylor
approximations as opposed to simple quadraic surfaces. In this example, case 3 was also a local
minimum. In contrast, in Example 7.2.17 the graph was like a trough. The behaviour of f away
from the critical point includes higher order terms whose influence turns the trough into a local
minimum.
Example 7.3.3. Suppose f (x, y) = sin(x) cos(y) to find the Taylor series centered at (0, 0) we can
simply multiply the one-dimensional result sin(x) = x − 3!1 x3 + 5!1 x5 + · · · and cos(y) = 1 − 2!1 y 2 +
1 4
4! y + · · · as follows:
f (x, y) = (x − 3!1 x3 + 5!1 x5 + · · · )(1 − 2!1 y 2 + 4!1 y 4 + · · · )

= x − 21 xy 2 + 24
1
xy 4 − 16 x3 − 12
1 3 2
x y + ···
= x + ···
The origin (0, 0) is a critical point since fx (0, 0) = 0 and fy (0, 0) = 0, however, this particular
critical point escapes the analysis via the quadratic form term since Q = 0 in the Taylor series
for this function at (0, 0). This is analogous to the inconclusive case of the 2nd derivative test in
calculus III.
Example 7.3.4. Suppose f (x, y, z) = xyz. Calculate the multivariate Taylor expansion about the
point (1, 2, 3). I’ll actually calculate this one via differentiation, I have used tricks and/or calculus
II results to shortcut any differentiation in the previous examples. Calculate first derivatives
fx = yz fy = xz fz = xy,
and second derivatives,

fxx = 0 fxy = z fxz = y
fyx = z fyy = 0 fyz = x
fzx = y fzy = x fzz = 0,
and the nonzero third derivatives,
fxyz = fyzx = fzxy = fzyx = fyxz = fxzy = 1.
It follows,
f (a + h, b + k, c + l) =
= f (a, b, c) + fx (a, b, c)h + fy (a, b, c)k + fz (a, b, c)l +
1
2 ( fxx hh + fxy hk + fxz hl + fyx kh + fyy kk + fyz kl + fzx lh + fzy lk + fzz ll ) + · · ·
Of course certain terms can be combined since fxy = fyx etc... for smooth functions (we assume
smooth in this section, moreover the given function here is clearly smooth). In total,
1 1
f (1 + h, 2 + k, 3 + l) = 6 + 6h + 3k + 2l + 3hk + 2hl + 3kh + kl + 2lh + lk + (6)hkl
2 3!
Of course, we could also obtain this from simple algebra:
f (1 + h, 2 + k, 3 + l) = (1 + h)(2 + k)(3 + l) = 6 + 6h + 3k + l + 3hk + 2hl + kl + hkl.

Example 7.3.5. Find the extreme values of w = x + y subject to the condition x2 + y 2 + z 2 ≤ 1.

Chapter 8
convergence and estimation
8.1 sequences of functions
8.2 power series
8.3 matrix exponential
8.4 uniform convergence
8.5 differentiating under the integral
8.6 contraction mappings and Newton’s method
8.7 multivariate mean value theorem
8.8 proof of the inverse function theorem
185
186 CHAPTER 8. CONVERGENCE AND ESTIMATION
Chapter 9
multilinear algebra
The principle aim of this chapter is to introduce how to calculate with ⊗ and ∧. We take a very
concrete approach where the tensor and wedge product are understood in terms of multilinear map-
pings for which they form a basis. That said, there is a univerisal algebraic approach to construct
the tensor and wedge products, I encourage the reader to study Dummit and Foote’s Abstract
Algebra Part III on Modules and Vector Spaces where these constructions are explained in a much
broader algebraic context.
Beyond the basic definitions, we also study how wedge products capture determinants and give a
natural language to ask certain questions of linear dependence. We also study metrics with partic-
ular attention to four dimensional Minkowski space with signature (−1, 1, 1, 1). Hodge duality is
detailed for three dimensional Euclidean space and four dimensional Minkowski space. However,
there is sufficient detail that one ought to be able to extrapolate to euclidean spaces of other di-
mension. Moreover, the Hodge duality is reduced to a few tables for computational convenience. I
encourage the reader to see David Bleeker’s text for a broader discussion of Hodge duality with a
physical bent.
When I lecture this material I’ll probably just give examples to drive home the computational
aspects. Also, it should be noted this Chapter can be studied without delving deeply into Sections
9.4 and 9.6. However, Chapter 12 requires some understanding of both of those sections.
9.1 dual space

Definition 9.1.1.
Suppose V is a vector space over R. We define the dual space to V to be the set of all
linear functions from V to R. In particular, we denote:
V ∗ = {f : V → R | f (x + y) = f (x) + f (y) and f (cx) = cf (x) ∀x, y ∈ V and c ∈ R}
If α ∈ V ∗ then we say α is a dual vector.
187
188 CHAPTER 9. MULTILINEAR ALGEBRA
I offer several abstract examples to begin, however the majority of this section concerns Rn .
R1
Example 9.1.2. Suppose F denotes the set of continuous functions on R. Define α(f ) = 0 f (t) dt.
The mapping α : F → R is linear by properties of definite integrals therefore we identify the definite
integral defines a dual-vector to the vector space of continuous functions.
Example 9.1.3. Suppose V = F(W, R) denotes a set of functions from a vector space W to R.
Note that V is a vector space with respect to point-wise defined addition and scalar multiplication
of functions. Let wo ∈ W and define α(f ) = f (wo ). The mapping α : V → R is linear since
α(cf + g) = (cf + g)(wo ) = cf (wo ) + g(wo ) = cα(f ) + α(g) for all f, g ∈ V and c ∈ R. We find
that the evaluation map defines a dual-vector α ∈ V ∗ .
Example 9.1.4. The determinant is a mapping from R n×n to R but it does not define a dual-vector
to the vector space of square matrices since det(A + B) 6= det(A) + det(B).
Example 9.1.5. Suppose α(x) = x · v for a particular vector v ∈ Rn . We argue α ∈ V ∗ where we
recall V = Rn is a vector space. Additivity follows from a property of the dot-product on Rn ,
α(x + y) = (x + y) · v = x · v + y · v = α(x) + α(y)
for all x, y ∈ Rn . Likewise, homogeneity follows from another property of the dot-product: observe
α(cx) = (cx) · v = c(x · v) = cα(x)
for all x ∈ Rn and c ∈ R.

Example 9.1.6. Let α(x, y) = 2x + 5y define a function α : R2 → R. Note that
α(x, y) = (x, y) · (2, 5)
hence by the preceding example we find α ∈ (R2 )∗ .

The preceding example is no accident. It turns out there is a one-one correspondance between row
vectors and dual vectors on Rn . Let v ∈ Rn then we define αv (x) = x · v. We proved in Example
9.1.5 that αv ∈ (Rn )∗ . Suppose α ∈ (Rn )∗ we see to find v ∈ Rn such that α = αv . Recall that a
linear function is uniquely defined by its values on a basis; the values of α on the standard
P basis
will show us how to choose v. This is a standard technique. Consider: v ∈ Rn with1 v = nj=1 v j ej
Xn n
X n
X
α(x) = α( xj e j ) = α(xj ej ) = xj α(ej ) = x · v
j=1 j=1 j=1
| {z }| {z }
additivity homogeneity
where we define v = (α(e1 ), α(e2 ), . . . , α(en )) ∈ Rn . The vector which corresponds naturally2 to α
is simply the vector of of the values of α on the standard basis.
1
the super-index is not a power in this context, it is just a notation to emphasize v j is the component of a vector.
2
some authors will say Rn×1 is dual to R1×n since αv (x) = v T x and v T is a row vector, I will avoid that langauge
in these notes.
9.2. MULTILINEARITY AND THE TENSOR PRODUCT 189
The dual space to Rn is a vector space and the correspondance v → αv gives an isomorphism of Rn
and (Rn )∗ . The image of a basis under an isomorphism is once more a basis. Define Φ : Rn → (R)∗
by Φ(v) = αv to give the correspondance an explicit label. The image of the standard basis under
Φ is called the standard dual basis for (Rn )∗ . Consider Φ(ej ), let x ∈ Rn and calculate
Φ(ej )(x) = αej (x) = x · ej
In particular, notice that when x = ei then Φ(ej )(ei ) = ei · ej = δij . Dual vectors are linear
transformations therefore we can define the dual basis by its values on the standard basis.
Definition 9.1.7.
The standard dual basis of (Rn )∗ is denoted {e1 , e2 , . . . , en } where we define ej : Rn → R

to be the linear transformation such that ej (ei ) = δij for all i, j ∈ Nn . Generally, given a
vector space V with basis β = {f1 , f2 , . . . , fm } we say the basis β ∗ = {f 1 , f 2 , . . . , f n } is
dual to β iff f j (fi ) = δij for all i, j ∈ Nn .
The term basis indicates that {e1 , e2 , . . . , en } is linearly independent3 and span{e1 , e2 , . . . , en } =
(Rn )∗ . The following calculation is often useful: if x ∈ Rn with x = nj=1 xj ej then

P
n
X n
X n
X
i i j j i
e (x) = e x ej = x e (ej ) = xj δij = xi ⇒ ei (x) = xi .
j=1 j=1 j=1
The calculation above is a prototype

Pn for many that follow in this chapter. Next, suppose α ∈ (Rn )∗
n j
and suppose x ∈ R with x = j=1 x ej . Calculate,
n
X n
X n
X
i i
α(x) = α x ei = α(ei )e (x) ⇒ α= α(ei )ei
i=1 i=1 i=1
this shows every dual vector is in the span of the dual basis {ej }nj=1 .
9.2 multilinearity and the tensor product

A multilinear mapping is a function of a Cartesian product of vector spaces which is linear with
respect to each ”slot”. The goal of this section is to explain what that means. It turns out the set
of all multilinear mappings on a particular set of vector spaces forms a vector space and we’ll show
how the tensor product can be used to construct an explicit basis by tensoring a bases which are
dual to the bases in the domain. We also examine the concepts of symmetric and antisymmetric
multilinear mappings, these form interesting subspaces of the set of all multilinear mappings. Our
approach in this section is to treat the case of bilinearity in depth then transition to the case of
multilinearity. Naturally this whole discussion demands a familarity with the preceding section.
3
direct proof of LI is left to the reader
9.2.1 bilinear maps
Definition 9.2.1.
Suppose V1 , V2 are vector spaces then b : V1 × V2 → R is a binear mapping on V1 × V2 iff

for all x, y ∈ V1 , z, w ∈ V2 and c ∈ R:
(1.) b(cx + y, z) = cb(x, z) + b(y, z) (linearity in the first slot)

(2.) b(x, cz + w) = cb(x, z) + b(x, w) (linearity in the second slot).
bilinear maps on V × V
When V1 = V2 = V we simply say that b : V × V → R is a bilinear mapping on V . The set of

all bilinear maps of V is denoted T02 V . You can show that T02 V forms a vector space under
the usual point-wise defined operations of function addition and scalar multiplication4 . Hopefully
you are familar with the example below.
Example 9.2.2. Define b : Rn × Rn → R by b(x, y) = x · y for all x, y ∈ Rn . Linearity in each slot

follows easily from properties of dot-products:
b(cx + y, z) = (cx + y) · z = cx · z + y · z = cb(x, z) + b(y, z)
b(x, cy + z) = x · (cy + z) = cx · y + x · z = cb(x, y) + b(x, z).
We can use matrix multiplication to generate a large class of examples with ease.
Example 9.2.3. Suppose A ∈ R n×n and define b : Rn × Rn → R by b(x, y) = xT Ay for all

x, y ∈ Rn . Observe that, by properties of matrix multiplication,
b(cx + y, z) = (cx + y)T Az = (cxT + y T )Az = cxT Az + y T Az = cb(x, z) + b(y, z)
b(x, cy + z) = xT A(cy + z) = cxT Ay + xT Az = cb(x, y) + b(x, z)
for all x, y, z ∈ Rn and c ∈ R. It follows that b is bilinear on Rn .
Suppose b : V × V → R is bilinear and suppose β = {e1 , e2 , . . . , en } is a basis for V whereas
4
sounds like homework
β ∗ = {e1 , e2 , . . . , en } is a basis of V ∗ with ej (ei ) = δij

n
X n
X
i j
b(x, y) = b x ei , y ej (9.1)
i=1 j=1
n
X
= b(xi ei , y j ej )
i,j=1
X n
= xi y j b(ei , ej )
i,j=1
X n
= b(ei , ej )ei (x)j (y)
i,j=1
Therefore, if we define bij = b(ei , ej ) then we may compute b(x, y) = ni,j=1 bij xi y j . The calculation
P
above also indicates that b is a linear combination of certain basic bilinear mappings. In particular,
b can be written a linear combination of a tensor product of dual vectors on V .
Definition 9.2.4.
Suppose V is a vector space with dual space V ∗ . If α, β ∈ V ∗ then we define α⊗β : V ×V →
R by (α ⊗ β)(x, y) = α(x)β(y) for all x, y ∈ V .
Given the notation5 preceding this definition, we note (ei ⊗ ej )(x, y) = ei (x)ej (y) hence for all
x, y ∈ V we find:
n
X n
X
b(x, y) = b(ei , ej )(ei ⊗ ej )(x, y) therefore, b = b(ei , ej )ei ⊗ ej
i,j=1 i,j=1
We find6 that T02 V = span{ei ⊗ej }ni,j=1 . Moreover, it can be argued7 that {ei ⊗ej }ni,j=1 is a linearly
independent set, therefore {ei ⊗ ej }ni,j=1 forms a basis for T02 V . We can count there are n2 vectors
in {ei ⊗ ej }ni,j=1 hence dim( T02 V ) = n2 .
If V = Rn and if {ei }ni=1 denotes the standard dual basis, then there is a standard notation for
the set of coefficients found in the summation for b. In particular, we denote B = [b] where
Bij = b(ei , ej ) hence, following Equation 9.1,
n
X n X
X n
i j
b(x, y) = x y b(ei , ej ) = xi Bij y j = xT By
i,j=1 i=1 j=1
5
perhaps you would rather write (ei ⊗ ej )(x, y) as ei ⊗ ej (x, y), that is also fine.
6
with the help of your homework where you will show {ei ⊗ ej }n 2
i,j=1 ⊆ T0 V
7
yes, again, in your homework
Definition 9.2.5.
Suppose b : V × V → R is a bilinear mapping then we say:
1. b is symmetric iff b(x, y) = b(y, x) for all x, y ∈ V
2. b is antisymmetric iff b(x, y) = −b(y, x) for all x, y ∈ V
Any bilinear mapping on V can be written as the sum of a symmetric and antisymmetric bilinear
mapping, this claim follows easily from the calculation below:

1 1
b(x, y) = b(x, y) + b(y, x) + b(x, y) − b(y, x) .
2 2
| {z } | {z }
symmetric antisymmetric
We say Sij is symmetric in i, j iff Sij = Sji for all i, j. Likewise, we say Aij is antisymmetric in
i, j iff Aij = −Aji for all i, j. If S is a symmetric bilinear mapping and A is an antisymmetric bilinear
mapping then the components of S are symmetric and the components of A are antisymmetric.
Why? Simply note:
S(ei , ej ) = S(ej , ei ) ⇒ Sij = Sji
and
A(ei , ej ) = −A(ej , ei ) ⇒ Aij = −Aji .
You can prove that the sum or scalar multiple of an (anti)symmetric bilinear mapping is once more
(anti)symmetric therefore the set of antisymmetric bilinear maps Λ2 (V ) and the set of symmetric
bilinear maps ST20 V are subspaces of T20 V . The notation Λ2 (V ) is part of a larger discussion on
the wedge product, we will return to it in a later section.
Finally, if we consider the special case of V = Rn once more we find that a bilinear mapping
b : Rn ×Rn → R has a symmetric matrix [b]T = [b] iff b is symmetric whereas it has an antisymmetric
matric [b]T = −[b] iff b is antisymmetric.
bilinear maps on V ∗ × V ∗
Suppose h : V ∗ ×V ∗ → R is bilinear then we say h ∈ T02 V . In addition, suppose β = {e1 , e2 , . . . , en }
is a basis for V whereas β ∗ = {e1 , e2 , . . . , en } is a basis of V ∗ with ej (ei ) = δij . Let α, β ∈ V ∗
n
X n
X
i j
h(α, β) = h αi e , βj e (9.2)
i=1 j=1
n
X
= h(αi ei , βj ej )
i,j=1
X n
= αi βj h(ei , ej )
i,j=1
X n
= h(ei , ej )α(ei )β(ej )
i,j=1
Pn
Therefore, if we define hij = h(ei , ej ) then we find the nice formula h(α, β) = ij
i,j=1 h αi βj . To
further refine the formula above we need a new concept.
The dual of the dual is called the double-dual and it is denoted V ∗∗ . For a finite dimensional vector
space there is a cannonical isomorphism of V and V ∗∗ . In particular, Φ : V → V ∗∗ is defined by
Φ(v)(α) = α(v) for all α ∈ V ∗ . It is customary to replace V with V ∗∗ wherever the context allows.
For example, to define the tensor product of two vectors x, y ∈ V as follows:
Definition 9.2.6.
Suppose V is a vector space with dual space V ∗ . We define the tensor product of vectors
x, y as the mapping x ⊗ y : V ∗ × V ∗ → R by (x ⊗ y)(α, β) = α(x)β(y) for all x, y ∈ V .
We could just as well have defined x ⊗ y = Φ(x) ⊗ Φ(y) where Φ is once more the cannonical
isomorphism of V and V ∗∗ . It’s called cannonical because it has no particular dependendence on
the coordinates used on V . In contrast, the isomorphism of Rn and (Rn )∗ was built around the
dot-product and the standard basis.
All of this said, note that α(ei )β(ej ) = ei ⊗ ej (α, β) thus,
n
X n
X
h(α, β) = h(ei , ej )ei ⊗ ej (α, β) ⇒ h= h(ei , ej )ei ⊗ ej
i,j=1 i,j=1
We argue that {ei ⊗ ej }ni,j=1 is a basis8

Definition 9.2.7.
8 2
T0 V is a vector space and we’ve shown T02 (V ) ⊆ span{ei ⊗ ej }n 2
i,j=1 but we should also show ei ⊗ ej ∈ T0 and
check for LI of {ei ⊗ ej }n
i,j=1 .
Suppose h : V ∗ × V ∗ → R is a bilinear mapping then we say:
1. h is symmetric iff h(α, β) = h(β, α) for all α, β ∈ V ∗
2. h is antisymmetric iff h(α, β) = −h(β, α) for all α, β ∈ V ∗
The discussion of the preceding subsection transfers to this context, we simply have to switch some
vectors to dual vectors and move some indices up or down. I leave this to the reader.
bilinear maps on V × V ∗
Suppose H : V × V ∗ → R is bilinear, we say H ∈ T11 V (or, if the context demands this detail
H ∈ T1 1 V ). We define α ⊗ x ∈ T1 1 (V ) by the natural rule; (α ⊗ x)(y, β) = α(x)β(x) for all
(y, β) ∈ V × V ∗ . We find, by calculations similar to those already given in this section,
n n
j i
Hi j ei ⊗ ej
X X
H(y, β) = Hi y βj and H=
i,j=1 i,j=1
j
where we defined Hi = H(ei , ej ).
bilinear maps on V ∗ × V
Suppose G : V ∗ × V → R is bilinear, we say G ∈ T11 V (or, if the context demands this detail
G ∈ T 1 1 V ). We define x ⊗ α ∈ T 1 1 V by the natural rule; (x ⊗ α)(β, y) = β(x)α(y) for all
(β, y) ∈ V ∗ × V . We find, by calculations similar to those already given in this section,
n
X n
X
G(β, y) = Gi j βi y j and G= Gi j e i ⊗ e j
i,j=1 i,j=1
where we defined Gi j = G(ei , ej ).
9.2.2 trilinear maps

Definition 9.2.8.
Suppose V1 , V2 , V3 are vector spaces then T : V1 × V2 × V3 → R is a trilinear mapping on

V1 × V2 × V3 iff for all u, v ∈ V1 , w, x ∈ V2 . y, z ∈ V3 and c ∈ R:
(1.) T (cu + v, w, y) = cT (u, w, y) + T (v, w, y) (linearity in the first slot)

(2.) T (u, cw + x, y) = cT (u, w, y) + T (u, x, y) (linearity in the second slot).
(3.) T (u, w, cy + z) = cT (u, w, y) + T (u, w, z) (linearity in the third slot).
If T : V × V × V → R is trilinear on V × V × V then we say T is a trilinear mapping on V and

we denote the set of all such mappings T30 V . The tensor product of three dual vectors is defined
much in the same way as it was for two,
(α ⊗ β ⊗ γ)(x, y, z) = α(x)β(y)γ(z)
Let {ei }ni=1 is a basis for V with dual basis {ei }ni=1 for V ∗ . If T is trilinear on V it follows
n
X n
X
i j k
T (x, y, z) = Tijk x y z and T = Tijk ei ⊗ ej ⊗ ek
i,j,k=1 i,j,k=1
where we defined Tijk = T (ei , ej , ek ) for all i, j, k ∈ Nn .
Generally suppose that V1 , V2 , V3 are possibly distinct vector spaces. Moreover, suppose V1 has basis
{ei }ni=1
1
, V2 has basis {fj }nj=1
2
and V3 has basis {gk }nk=1
3
. Denote the dual bases for V1∗ , V2∗ , V3∗ in
i n1 j n1 k n1
the usual fashion: {e }i=1 , {f }j=1 , {g }k=1 . With this notation, we can write a trilinear mapping
on V1 × V2 × V3 as follows: (where we define Tijk = T (ei , fj , gk ))
n1 X
X n2 X
n3 n1 X
X n2 X
n3
i j k
T (x, y, z) = Tijk x y z and T = Tijk ei ⊗ f j ⊗ g k
i=1 j=1 k=1 i=1 j=1 k=1
However, if V1 , V2 , V3 happen to be related by duality then it is customary to use up/down indices.

For example, if T : V × V × V ∗ → R is trilinear then we write9
n
X
T = Tij k ei ⊗ ej ⊗ ek
i,j,k=1
and say T ∈ T2 1 V . On the other hand, if S : V ∗ × V ∗ × V is trilinear then we’d write

n
S ij k ei ⊗ ej ⊗ ek
X
T =
i,j,k=1
and say T ∈ T 2 1 V . I’m not sure that I’ve ever seen this notation elsewhere, but perhaps it could
be useful to denote the set of trinlinear maps T : V × V ∗ × V → R as T1 1 1 V . Hopefully we will
not need such silly notation in what we consider this semester.
There was a natural correspondance between bilinear maps on Rn and square matrices. For a
trilinear map we would need a three-dimensional array of components. In some sense you could
picture T : Rn × Rn × Rn → R as multiplication by a cube of numbers. Don’t think too hard
about these silly comments, we actually already wrote the useful formulae for dealing with trilinear
objects. Let’s stop to look at an example.
9
we identify ek with its double-dual hence this tensor product is already defined, but to be safe let me write it out
in this context ei ⊗ ej ⊗ ek (x, y, α) = ei (x)ej (y)α(ek ).
Example 9.2.9. Define T : R3 × R3 × R3 → R by T (x, y, z) = det(x|y|z). You may not have

learned this in your linear algebra course10 but a nice formula11 for the determinant is given by the
Levi-Civita symbol,
X3
det(A) = ijk Ai1 Aj2 Ak3
i,j,k=1
note that col1 (A) = [Ai1 ], col2 (A) = [Ai2 ] and col3 (A) = [Ai3 ]. It follows that
3
X
T (x, y, z) = ijk xi y j z k
i,j,k=1
Multilinearity follows easily from this formula. For example, linearity in the third slot:
3
X
T (x, y, cz + w) = ijk xi y j (cz + w)k (9.3)
i,j,k=1
3
X
= ijk xi y j (cz k + wk ) (9.4)
i,j,k=1
3
X 3
X
i j k
=c ijk x y z + ijk xi y j wk (9.5)
i,j,k=1 i,j,k=1
= cT (x, y, z) + T (x, y, w). (9.6)
Observe that by properties of determinants, or the Levi-Civita symbol if you prefer, swapping a pair
of inputs generates a minus sign, hence:
T (x, y, z) = −T (y, x, z) = T (y, z, x) = −T (z, y, x) = T (z, x, y) = −T (x, z, y).
If T : V × V × V → R is a trilinear mapping such that
T (x, y, z) = −T (y, x, z) = T (y, z, x) = −T (z, y, x) = T (z, x, y) = −T (x, z, y)
for all x, y, z ∈ V then we say T is antisymmetric. Likewise, if S : V × V × V → R is a trilinear

mapping such that
S(x, y, z) = −S(y, x, z) = S(y, z, x) = −S(z, y, x) = S(z, x, y) = −S(x, z, y).
for all x, y, z ∈ V then we say T is symmetric. Clearly the mapping defined by the determinant
is antisymmetric. In fact, many authors define the determinant of an n × n matrix as the antisym-
metric n-linear mapping which sends the identity matrix to 1. It turns out these criteria unquely
10
maybe you haven’t even taken linear yet!
11
actually, I take this as the definition in linear algebra, it does take considerable effort to recover the expansion
by minors formula which I use for concrete examples
define the determinant. That is the motivation behind my Levi-Civita symbol definition. That
formula is just the nuts and bolts of complete antisymmetry.
You might wonder, can every trilinear mapping can be written as a the sum of a symmetric and
antisymmetric mapping? The answer is no. Consider T : V ×V ×V → R defined by T = e1 ⊗e2 ⊗e3 .
Is it possible to find constants a, b such that:
e1 ⊗ e2 ⊗ e3 = ae[1 ⊗ e2 ⊗ e3] + be(1 ⊗ e2 ⊗ e3)
where [. . . ] denotes complete antisymmetrization of 1, 2, 3 and (. . . ) complete symmetrization:

1 123
e[1 ⊗ e2 ⊗ e3] = e + e231 + e312 − e321 − e213 − e132

6
For the symmetrization we also have to include all possible permutations of (1, 2, 3) but all with +:
1 123
e(1 ⊗ e2 ⊗ e3) = e + e231 + e312 + e321 + e213 + e132

6
As you can see:
a + b 123 b − a 321
ae[1 ⊗ e2 ⊗ e3] + be(1 ⊗ e2 ⊗ e3) = (e + e231 + e312 ) + (e + e213 + e132 )
6 6
There is no way for these to give back only e1 ⊗ e2 ⊗ e3 . I leave it to the reader to fill the gaps
in this argument. Generally, the decomposition of a multilinear mapping into more basic types is
a problem which requires much more thought than we intend here. Representation theory does
address this problem: how can we decompose a tensor product into irreducible pieces. Their idea
of tensor product is not precisely the same as ours, however algebraically the problems are quite
intertwined. I’ll leave it at that unless you’d like to do an independent study on representation
theory. Ideally you’d already have linear algebra and abstract algebra complete before you attempt
that study.
9.2.3 multilinear maps

Definition 9.2.10.
Suppose V1 , V2 , . . . Vk are vector spaces then T : V1 × V2 × · · · × Vk → R is a k-multilinear
mapping on V1 ×V2 ×· · ·×Vk iff for each c ∈ R and x1 , y1 ∈ V1 , x2 , y2 ∈ V2 , . . . , xk , yk ∈ Vk
T (x1 , . . . , cxj + yj , . . . , xk ) = cT (x1 , . . . , xj , . . . , xk ) + T (x1 , . . . , yj , . . . , xk )
for j = 1, 2, . . . , k. In other words, we assume T is linear in each of its k-slots. If T is

multilinear on V r × (V ∗ )s then we say that T ∈ Trs V and we say T is a type (r, s) tensor
on V .
The definition above makes a dual vector a type (1, 0) tensor whereas a double dual of a vector a
type (0, 1) tensor, a bilinear mapping on V is a type (2, 0) tensor and a bilinear mapping on V ∗ is
a type (0, 2) tensor with respect to V .
We are free to define tensor products in this context in the same manner as we have previously.
Suppose α1 ∈ V1∗ , α2 ∈ V2∗ , . . . , αk ∈ Vk∗ and v1 ∈ V1 , v2 ∈ V2 , . . . , vk ∈ Vk then
α1 ⊗ α2 ⊗ · · · ⊗ αk (v1 , v2 , . . . , vk ) = α1 (v1 )α2 (v2 ) · · · αk (vk )
It is easy to show the tensor produce of k-dual vectors as defined above is indeed a k-multilinear
mapping. Moreover, the set of all k-multilinear mappings on V1 × V2 × · · · × Vk clearly forms a
vector space of dimension dim(V1 )dim(V2 ) · · · dim(Vk ) since it naturally takes the tensor product of
the dual bases for V1∗ , V2∗ , . . . , Vk∗ as its basis. In particular, suppose for j = 1, 2, . . . , k that Vj has
nj nj
basis {Eji }i=1 which is dual to {Eji }i=1 the basis for Vj∗ . Then we can derive that a k-multilinear
mapping can be written as
n1 X
X n2 nk
X
T = ··· Ti1 i2 ...ik E1i1 ⊗ E2i2 ⊗ Ekik
i1 =1 i2 =1 ik =1
If T is a type (r, s) tensor on V then there is no need for the ugly double indexing on the basis
since we need only tensor a basis {ei }ni=1 for V and its dual {ei }ni=1 for V ∗ in what follows:
n n
Tij11ij22...i
...js i1
X X
T = r
e ⊗ ei2 ⊗ · · · ⊗ eir ⊗ ej1 ⊗ ej2 ⊗ · · · ⊗ ejs .
i1 ,...,ir =1 j1 ,...,js =1
permutations
Before I define symmetric and antisymmetric for k-linear mappings on V I think it is best to discuss
briefly some ideas from the theory of permutations.
Definition 9.2.11.
A permutation on {1, 2, . . . p} is a bijection on {1, 2, . . . p}. We define the set of permutations
on {1, 2, . . . p} to be Σp . Further, define the sign of a permutation to be sgn(σ) = 1 if σ is
the product of an even number of transpositions whereas sgn(σ) = −1 if σ is the product
of a odd number transpositions.
Let us consider the set of permutations on {1, 2, 3, . . . n}, this is called Sn the symmetric group,
its order is n! if you were wondering. Let me remind12 you how the cycle notation works since it
allows us to explicitly present the number of transpositions contained in a permutation,

1 2 3 4 5 6
σ= ⇐⇒ σ = (12)(356) = (12)(36)(35) (9.7)
2 1 5 4 6 3
recall the cycle notation is to be read right to left. If we think about inputing 5 we can read from
the matrix notation that we ought to find 5 7→ 6. Clearly that is the case for the first version of
12
or perhaps, more likely, introduce you to this notation
σ written in cycle notation; (356) indicates that 5 7→ 6 and nothing else messes with 6 after that.
Then consider feeding 5 into the version of σ written with just two-cycles (a.k.a. transpositions ),
first we note (35) indicates 5 7→ 3, then that 3 hits (36) which means 3 7→ 6, finally the cycle (12)
doesn’t care about 6 so we again have that σ(5) = 6. Finally we note that sgn(σ) = −1 since it is
made of 3 transpositions.
It is always possible to write any permutation as a product of transpositions, such a decomposition

is not unique. However, if the number of transpositions is even then it will remain so no matter
how we rewrite the permutation. Likewise if the permutation is an product of an odd number of
transpositions then any other decomposition into transpositions is also comprised of an odd number
of transpositions. This is why we can define an even permutation is a permutation comprised by
an even number of transpositions and an odd permutation is one comprised of an odd number of
transpositions.
Example 9.2.12. Sample cycle calculations: we rewrite as product of transpositions to deter-

min if the given permutation is even or odd,
σ = (12)(134)(152) = (12)(14)(13)(12)(15) =⇒ sgn(σ) = −1
λ = (1243)(3521) = (13)(14)(12)(31)(32)(35) =⇒ sgn(λ) = 1

γ = (123)(45678) = (13)(12)(48)(47)(46)(45) =⇒ sgn(γ) = 1
We will not actually write down permutations in the calculations the follow this part of the notes.
I merely include this material as to give a logically complete account of antisymmetry. In practice,
if you understood the terms as the apply to the bilinear and trilinear case it will usually suffice for
concrete examples. Now we are ready to define symmetric and antisymmetric.
Definition 9.2.13.
A k-linear mapping L : V × V × · · · × V → R is completely symmetric if
L(x1 , . . . , x, . . . , y, . . . , xk ) = L(x1 , . . . , y, . . . , x, . . . , xk )
for all possible x, y ∈ V . Conversely, if a k-linear mapping on V has
L(x1 , . . . , x, . . . , y, . . . , xp ) = −L(x1 , . . . , y, . . . , x, . . . , xp )
for all possible pairs x, y ∈ V then it is said to be completely antisymmetric or alter-

nating. Equivalently a k-linear mapping L is alternating if for all π ∈ Σk
L(xπ1 , xπ2 , . . . , xπk ) = sgn(π)L(x1 , x2 , . . . , xk )
The set of alternating multilinear mappings on V is denoted ΛV , the set of k-linear alter-
nating maps on V is denoted Λk V . Often an alternating k-linear map is called a k-form.
Moreover, we say the degree of a k-form is k.
Similar terminology applies to the components of tensors. We say Ti1 i2 ...ik is completely symmetric
in i1 , i2 , . . . , ik iff Ti1 i2 ...ik = Tiσ(1) iσ(2) ...iσ(k) for all σ ∈ Σk . On the other hand, Ti1 i2 ...ik is completely
antisymmetric in i1 , i2 , . . . , ik iff Ti1 i2 ...ik = sgn(σ)Tiσ(1) iσ(2) ...iσ(k) for all σ ∈ Σk . It is a simple
exercise to show that a completely (anti)symmetric tensor13 has completely (anti)symmetric com-
ponents.
The tensor product is an interesting construction to discuss at length. To summarize, it is asso-

ciative and distributive across addition. Scalars factor out and it is not generally commutative.
For a given vector space V we can in principle generate by tensor products multilinear mappings
of arbitrarily high order. This tensor algebra is infinite dimensional. In contrast, the space ΛV of
forms on V is a finite-dimensional subspace of the tensor algebra. We discuss this next.
9.3 wedge product

We assume V is a vector space with basis {ei }ni=1 throughout this section. The dual basis is denoted
{ei }ni=1 as is our usual custom. Our goal is to find a basis for the alternating maps on V and explore
the structure implicit within its construction. This will lead us to call ΛV the exterior algebra
of V after the discussion below is complete.
9.3.1 wedge product of dual basis generates basis for ΛV

Suppose b : V × V → R is antisymmetric and b = ni.j=1 bij ei ⊗ ej , it follows that bij = −bji for all
P
i, j ∈ Nn . Notice this implies that bii = 0 for i = 1, 2, . . . , n. For a given pair of indices i, j either
i < j or j < i or i = j hence,
X X X
b= bij ei ⊗ ej + bij ei ⊗ ej + bij ei ⊗ ej
i<j j<i i=j
X X
i j i j
= bij e ⊗ e + bij e ⊗ e
i<j j<i
X X
= bij ei ⊗ ej − bji ei ⊗ ej
i<j j<i
X X
k l
= bkl e ⊗ e − bkl el ⊗ ek
k<l k<l
X
= bkl (ek ⊗ el − el ⊗ ek ). (9.8)
k<l
Therefore, {ek ⊗ el − el ⊗ ek |l, k ∈ Nn and l < k} spans the set of antisymmetric bilinear maps on
V . Moreover, you can show this set is linearly independent hence it is a basis fo Λ2 V . We define
13
in this context a tensor is simply a multilinear mapping, in physics there is more attached to the term
9.3. WEDGE PRODUCT 201
the wedge product of ek ∧ el = ek ⊗ el − el ⊗ ek . With this notation we find that the alternating
bilinear form b can be written as
n
X X 1
b= bkl ek ∧ el = bij ei ∧ ej
2
k<l i,j=1
where the summation on the r.h.s. is over all indices14 . Notice that ei ∧ ej is an antisymmetric
bilinear mapping because ei ∧ ej (x, y) = −ei ∧ ej (y, x), however, there is more structure here than
just that. It is also true that ei ∧ ej = −ej ∧ ei . This is a conceptually different antisymmetry, it
is the antisymmetry of the wedge produce ∧.
Suppose b : V × V × V → R is antisymmetric and b = ni,j,k=1 bijk ei ⊗ ej ⊗ ek , it follows that

P
bijk = bjki = bkij and bijk = −bkji = −bjik = bikj for all i, j, k ∈ Nn . Notice this implies that
biii = 0 for i = 1, 2, . . . , n. A calculation similar to the one just offered for the case of a bilinear
map reveals that we can write b as follows:
X
b= bijk ei ⊗ ej ⊗ ek + ej ⊗ ek ⊗ ei + ek ⊗ ei ⊗ ej
i<j<k

− ek ⊗ ej ⊗ ei − ej ⊗ ei ⊗ ek − ei ⊗ ek ⊗ ej (9.9)
Define ei ∧ ej ∧ ek = ei ⊗ ej ⊗ ek + ej ⊗ ek ⊗ ei + ek ⊗ ei ⊗ ej − ek ⊗ ej ⊗ ei − ej ⊗ ei ⊗ ek − ei ⊗ ek ⊗ ej
thus
n
X X 1
b= bijk ei ∧ ej ∧ ek = bijk ei ∧ ej ∧ ek (9.10)
3!
i<j<k i,j,k=1
and it is clear that {ei ∧ ej ∧ ek | i, j, k ∈ Nn and i < j < k} forms a basis for the set of alternating
trilinear maps on V .
Following the patterns above, we define the wedge product of p dual basis vectors,
X
ei1 ∧ ei2 ∧ · · · ∧ eip = sgn(π)eiπ(1) ⊗ eiπ(2) ⊗ · · · ⊗ eiπ(p) (9.11)
π∈Σp
If x, y ∈ V we would like to show that
ei1 ∧ ei2 ∧ · · · ∧ eip (. . . , x, . . . , y, . . . ) = −ei1 ∧ ei2 ∧ · · · ∧ eip (. . . , y, . . . , x, . . . ) (9.12)
follows from the complete antisymmetrization in the definition of the wedge product. Before we
give the general argument, let’s see how this works in the trilinear case. Consider, ei ∧ ej ∧ ek =
= ei ⊗ ej ⊗ ek + ej ⊗ ek ⊗ ei + ek ⊗ ei ⊗ ej − ek ⊗ ej ⊗ ei − ej ⊗ ei ⊗ ek − ei ⊗ ek ⊗ ej .
14
yes there is something to work out here, probably in your homework
Calculate, noting that ei ⊗ ej ⊗ ek (x, y, z) = ei (x)ej (y)ek (z) = xi y j z k hence

ei ∧ ej ∧ ek (x, y, z) = xi y j z k + xj y k z i + xk y i z j − xk y j z i − xj y i z k − xi y k z j
Thus,
ei ∧ ej ∧ ek (x, z, y) = xi z j y k + xj z k y i + xk z i y j − xk z j y i − xj z i y k − xi z k y j
and you can check that ei ∧ ej ∧ ek (x, y, z) = −ei ∧ ej ∧ ek (x, z, y). Similar tedious calculations prove
antisymmetry of the the interchange of the first and second or the first and third slots. Therefore,
ei ∧ ej ∧ ek is an alternating trilinear map as it is clearly trilinear since it is built from the sum of
tensor products which we know are likewise trilinear.
The multilinear case follows essentially the same argument, note

X i i i i
ei1 ∧ ei2 ∧ · · · ∧ eip (. . . , xj , . . . , xk , . . . ) = sgn(π)x1π(1) · · · xjπ(j) · · · xkπ(k) · · · xpπ(p) (9.13)
π∈Σp
whereas,
X i i i i
ei1 ∧ ei2 ∧ · · · ∧ eip (. . . , xk , . . . , xj , . . . ) = sgn(σ)x1σ(1) · · · xkσ(k) · · · xjσ(j) · · · xpσ(p) . (9.14)
σ∈Σp
Suppose we take each permutation σ and subsitute δ ∈ Σp such that σ(j) = δ(k) and σ(k) = δ(j)
and otherwise δ and σ agree. In cycle notation, δ(jk) = σ. Substitution δ into Equation 9.14:
ei1 ∧ ei2 ∧ · · · ∧ eip (. . . , xk , . . . , xj , . . . )
X i i i i
= sgn(δ(jk))x1δ(1) · · · xkδ(j) · · · xjδ(k) · · · xpδ(p)
δ∈Σp
X i i i i
=− sgn(δ)x1δ(1) · · · xjδ(k) · · · xkδ(j) · · · xpδ(p)
δ∈Σp
i1 i2
= −e ∧ e ∧ · · · ∧ eip (. . . , xj , . . . , xk , . . . ) (9.15)
Here the sgn of a permutation σ is (−1)N where N is the number of cycles in σ. We observed
that δ(jk) has one more cycle than δ hence sgn(δ(jk)) = −sgn(δ). Therefore, we have shown that
ei1 ∧ ei2 ∧ · · · ∧ eip ∈ Λp V .
Recall that ei ∧ ej = −ej ∧ ei in the p = 2 case. There is a generalization of that result to the
p > 2 case. In words, the wedge product is antisymetric with respect the interchange of any two
dual vectors. For p = 3 we have the following identities for the wedge product:
ei ∧ ej ∧ ek = − e|j {z
∧ e}i ∧ek = ej ∧ e|k {z
∧ e}i = − e|k {z
∧ e}j ∧ei = ek ∧ e|i {z
∧ e}j = − e|i {z
∧ ek} ∧ej
swapped swapped swapped swapped swapped
I’ve indicated how these signs are consistent with the p = 2 antisymmetry. Any permutation of
the dual vectors can be thought of as a combination of several transpositions. In any event, it is
sometimes useful to just know that the wedge product of three elements is invariant under cyclic
permutations of the dual vectors,
ei ∧ ej ∧ ek = ej ∧ ek ∧ ei = ek ∧ ei ∧ ej
and changes by a sign for anticyclic permutations of the given object,
ei ∧ ej ∧ ek = −ej ∧ ei ∧ ek = −ek ∧ ej ∧ ei = −ei ∧ ek ∧ ej
Generally we can argue that, for any permutation π ∈ Σp :
ei1 ∧ ei2 ∧ · · · ∧ eip = sgn(π)eiπ(1) ∧ eiπ(2) ∧ · · · ∧ eiπ(p)
This is just a slick formula which says the wedge product generates a minus whenever you flip two
dual vectors which are wedged.
9.3.2 the exterior algebra

The careful reader will realize we have yet to define wedge products of anything except for the dual
basis. But, naturally you must wonder if we can take the wedge product of other dual vectors or
morer generally alternating tensors. The answer is yes. Let us define the general wedge product:
Definition 9.3.1. Suppose α ∈ Λp V and β ∈ Λq V . We define Ip to be the set of all increasing lists
of p-indices, this set can be empty if dim(V ) is not sufficiently large. Moreover, if I = (i1 , i2 , . . . , ip )
then introduce notation eI = ei1 ∧ ei2 ∧ · · · ∧ eip hence:
n
X 1 X 1 X
α= αi1 i2 ...ip ei1 ∧ ei2 ∧ · · · ∧ eip = αI eI = αI eI
p! p!
i1 ,i2 ,...,ip =1 I I∈Ip
and
n
X 1 X 1 X
β= βj1 j2 ...jq ej1 ∧ ej2 ∧ · · · ∧ ejq = βJ eJ = βJ eJ
q! q!
j1 ,j2 ,...,jq =1 J J∈Iq
Naturally, eI ∧ eJ = ei1 ∧ ei2 ∧ · · · ∧ eip ∧ ej1 ∧ ej2 ∧ · · · ∧ ejq and we defined this carefully in the
preceding subsection. Define α ∧ β ∈ Λp+q V as follows:
XX 1
α∧β = αI βJ eI ∧ eJ .
p!q!
I J
Again, but with less slick notation:

n n
X X 1
α∧β = αi i ...i βj j ...j ei1 ∧ ei2 ∧ · · · ∧ eip ∧ ej1 ∧ ej2 ∧ · · · ∧ ejq
p!q! 1 2 p 1 2 q
i1 ,i2 ,...,ip =1 j1 ,j2 ,...,jq =1
All the definition above really says is that we extend the wedge product on the basis to distribute
over the addition of dual vectors. What this means calculationally is that the wedge product obeys
the usual laws of addition and scalar multiplication. The one feature that is perhaps foreign is the
antisymmetry of the wedge product. We must take care to maintain the order of expressions since
the wedge product is not generally commutative.
Proposition 9.3.2.
Let α, β, γ be forms on V and c ∈ R then
(i) (α + β) ∧ γ = α ∧ γ + β ∧ γ distributes across vector addition

(ii) α ∧ (β + γ) = α ∧ β + α ∧ γ distributes across vector addition
(iii) (cα) ∧ β = α ∧ (cβ) = c (α ∧ β) scalars factor out
(iv) α ∧ (β ∧ γ) = (α ∧ β) ∧ γ associativity
I leave the proof of this proposition to the reader.

Proposition 9.3.3. graded commutivity of homogeneous forms.
Let α, β be forms on V of degree p and q respectively then
α ∧ β = −(−1)pq β ∧ α
P 1 I P 1 J
Proof: suppose α = I p! e is a p-form on V and β = J q! e is a q-form on V . Calculate:
XX 1
α∧β = αI βJ eI ∧ eJ by defn. of ∧,
p!q!
I J
XX 1
= βJ αI eI ∧ eJ coefficients are scalars,
p!q!
I J
XX 1
= (−1)pq βJ αI eJ ∧ eI (details on sign given below)
p!q!
I J
= (−1)pq β ∧ α
Let’s expand in detail why eJ ∧ eI = (−1)pq eI ∧ eJ . Suppose I = (i1 , i2 , . . . , ip ) and J =

(j1 , j2 , . . . , jq ), the key is that every swap of dual vectors generates a sign:
eI ∧ eJ = ei1 ∧ ei2 ∧ · · · ∧ eip ∧ ej1 ∧ ej2 ∧ · · · ∧ ejq

= (−1)q ei1 ∧ ei2 ∧ · · · ∧ eip−1 ∧ ej1 ∧ ej2 ∧ · · · ∧ ejq ∧ eip
= (−1)q (−1)q ei1 ∧ ei2 ∧ · · · ∧ eip−2 ∧ ej1 ∧ ej2 ∧ · · · ∧ ejq ∧ eip−1 ∧ eip
.. .. ..
. . .
= (−1)q (−1)q · · · (−1)q ej1 ∧ ej2 ∧ · · · ∧ ejq ∧ ei1 ∧ ej2 ∧ · · · ∧ eip
| {z }
p−f actors
= (−1) e ∧ eI .
pq J

Example 9.3.4. Let α be a 2-form defined by
α = ae1 ∧ e2 + be2 ∧ e3
And let β be a 1-form defined by

β = 3e1
Consider then,
α ∧ β = (ae1 ∧ e2 + be2 ∧ e3 ) ∧ (3e1 )
= (3ae1 ∧ e2 ∧ e1 + 3be2 ∧ e3 ∧ e1 (9.16)
= 3be1 ∧ e2 ∧ e3 .
whereas,
β ∧ α = 3e1 ∧ (ae1 ∧ e2 + be2 ∧ e3 )
= (3ae1 ∧ e1 ∧ e2 + 3be1 ∧ e2 ∧ e3 (9.17)
= 3be1 ∧ e2 ∧ e3 .
so this agrees with the proposition, (−1)pq = (−1)2 = 1 so we should have found that α ∧ β = β ∧ α.
This illustrates that although the wedge product is antisymmetric on the basis, it is not always
antisymmetric, in particular it is commutative for even forms.
The graded commutivity rule α ∧ β = −(−1)pq β ∧ α has some suprising implications. This rule is
ultimately the reason ΛV is finite dimensional. Let’s see how that happens.
Proposition 9.3.5. linear dependent one-forms wedge to zero:
If α, β ∈ V ∗ and α = cβ for some c ∈ R then α ∧ β = 0.
Proof: to begin, note that β ∧ β = −β ∧ β hence 2β ∧ β = 0 and it follows that β ∧ β = 0. Note:
α ∧ β = cβ ∧ β = c(0) = 0
therefore the proposition is true.
Proposition 9.3.6.
Suppose that α1 , α2 , . . . , αp are linearly dependent 1-forms then
α1 ∧ α2 ∧ · · · ∧ αp = 0.
Proof: by assumption of linear dependence there exist constants c1 , c2 , . . . , cp not all zero such
that
c1 α1 + c2 α2 + · · · cp αp = 0.
Suppose that ck is a nonzero constant in the sum above, then we may divide by it and consequently
we can write αk in terms of all the other 1-forms,

−1
αk = c1 α1 + · · · + ck−1 αk−1 + ck+1 αk+1 + · · · + cp αp
ck
Insert this sum into the wedge product in question,

α1 ∧ α2 ∧ . . . ∧ αp = α1 ∧ α2 ∧ · · · ∧ αk ∧ · · · ∧ αp
= (−c1 /ck )α1 ∧ α2 ∧ · · · ∧ α1 ∧ · · · ∧ αp
+(−c2 /ck )α1 ∧ α2 ∧ · · · ∧ α2 ∧ · · · ∧ αp + · · ·
+(−ck−1 /ck )α1 ∧ α2 ∧ · · · ∧ αk−1 ∧ · · · ∧ αp (9.18)
+(−ck+1 /ck )α1 ∧ α2 ∧ · · · ∧ αk+1 ∧ · · · ∧ αp + · · ·
+(−cp /ck )α1 ∧ α2 ∧ · · · ∧ αp ∧ · · · ∧ αp
= 0.
We know all the wedge products are zero in the above because in each there is at least one 1-form
repeated, we simply permute the wedge products till they are adjacent and by the previous propo-
sition the term vanishes. The proposition follows.
Let us pause to reflect on the meaning of the proposition above for a n-dimensional vector space
V . The dual space V ∗ is likewise n-dimensional, this is a general result which applies to all finite-
dimensional vector spaces15 . Thus, any set of more than n dual vectors is necessarily linearly
dependent. Consquently, using the proposition above, we find the wedge product of more than n
one-forms is trivial. Therefore, while it is possible to construct Λk V for k > n we should understand
that this space only contains zero. The highest degree of a nontrivial form over a vector space of
dimension n is an n-form.
Moreover, we can use the proposition to deduce the dimension of a basis for Λp V , it must consist
of the wedge product of distinct linearly independent one-forms. The number of ways to choose p
distinct objects from a list of n distinct objects is precisely ”n choose p”,

n n!
= for 0 ≤ p ≤ n. (9.19)
p (n − p)!p!
Proposition 9.3.7.
If V is an n-dimensional vector space then Λk V is an np -dimensional vector space of p-

forms. Moreover, the direct sum of all forms over V has the structure
ΛV = R ⊕ Λ1 V ⊕ · · · Λn−1 V ⊕ Λn V
and is a vector space of dimension 2n

Proof: define Λ0 V = R then it is clear Λk V forms a vector space for k = 0, 1, . . . , n. Moreover,
Λj V ∩ Λk V = {0} for j 6= k hence the term ”direct sum” is appropriate. It remains to show
dim(ΛV ) = 2n where dim(V ) = n. A natural basis β for ΛV is found from taking the union of the
bases for each subspace of k-forms,
β = {1, ei1 , ei1 ∧ ei2 , . . . , ei1 ∧ ei2 ∧ · · · ∧ ein | 1 ≤ i1 < i2 < · · · < in ≤ n}
15
however, in infinite dimensions, the story is not so simple
But, we can count the number of vectors N in the set above as follows:

n n n
N =1+n+ + ··· + +
2 n−1 n
Recall the binomial theorem states
n
n
X n n−k k
(a + b) = a b = an + nan−1 b + · · · + nabn−1 + bn .
k
k=0
Recognize that N = (1 + 1)n and the proposition follows.
We should note that in the basis above the space of n-forms is one-dimensional because there is
only one way to choose a strictly increasing list of n integers in Nn . In particular, it is useful to note
Λn V = span{e1 ∧ e2 ∧ · · · ∧ en }. The form e1 ∧ e2 ∧ · · · ∧ en is sometimes called the the top-form16 .
Example 9.3.8. exterior algebra of R2 Let us begin with the standard dual basis {e1 , e2 }. By
definition we take the p = 0 case to be the field itself; Λ0 V ≡ R, it has basis 1. Next, Λ1 V =
span(e1 , e2 ) = V ∗ and Λ2 V = span(e1 ∧ e2 ) is all we can do here. This makes ΛV a 22 = 4-
dimensional vector space with basis
{1, e1 , e2 , e1 ∧ e2 }.
Example 9.3.9. exterior algebra of R3 Let us begin with the standard dual basis {e1 , e2 , e3 }.
By definition we take the p = 0 case to be the field itself; Λ0 V ≡ R, it has basis 1. Next, Λ1 V =
span(e1 , e2 , e3 ) = V ∗ . Now for something a little more interesting,
Λ2 V = span(e1 ∧ e2 , e1 ∧ e3 , e2 ∧ e3 )
and finally,
Λ3 V = span(e1 ∧ e2 ∧ e3 ).
This makes ΛV a 23 = 8-dimensional vector space with basis
{1, e1 , e2 , e3 , e1 ∧ e2 , e1 ∧ e3 , e2 ∧ e3 , e1 ∧ e2 ∧ e3 }
it is curious that the number of independent one-forms and 2-forms are equal.
Example 9.3.10. exterior algebra of R4 Let us begin with the standard dual basis {e1 , e2 , e3 , e4 }.
By definition we take the p = 0 case to be the field itself; Λ0 V ≡ R, it has basis 1. Next, Λ1 V =
span(e1 , e2 , e3 , e4 ) = V ∗ . Now for something a little more interesting,
Λ2 V = span(e1 ∧ e2 , e1 ∧ e3 , e1 ∧ e4 , e2 ∧ e3 , e2 ∧ e4 , e3 ∧ e4 )
and three forms,
Λ3 V = span(e1 ∧ e2 ∧ e3 , e1 ∧ e2 ∧ e4 , e1 ∧ e3 ∧ e4 , e2 ∧ e3 ∧ e4 ).
and Λ3 V = span(e1 ∧ e2 ∧ e3 ). Thus ΛV a 24 = 16-dimensional vector space. Note that, in contrast
to R3 , we do not have the same number of independent one-forms and two-forms over R4 .
16
or volume form for reasons we will explain later, other authors begin the discussion of forms from the consideration
of volume, see Chapter 4 in Bernard Schutz’ Geometrical methods of mathematical physics
Let’s explore how this algebra fits with calculations we already know about determinants.
Example 9.3.11. Suppose A = [A1 |A2 ]. I propose the determinant of A is given by the top-form
a b
on R2 via the formula det(A) = (e1 ∧ e2 )(A1 , A2 ). Suppose A = then A1 = (a, c) and
c d
A2 = (b, d). Thus,

a b
det = (e1 ∧ e2 )(A1 , A2 )
c d
= (e1 ⊗ e2 − e2 ⊗ e1 )((a, c), (b, d))
= e1 (a, c)e2 (b, d) − e2 (a, c)e1 (b, d)
= ad − bc.
I hope this is not surprising!
Example 9.3.12. Suppose A = [A1 |A2 |A3 ]. I propose the determinant of A is given by the top-
form on R3 via the formula det(A) = (e1 ∧e2 ∧e3 )(A1 , A2 , A3 ). Let’s see if we can find the expansion
by cofactors. By the definition we have e1 ∧ e2 ∧ e3 =
= e1 ⊗ e2 ⊗ e3 + e2 ⊗ e3 ⊗ e1 + e3 ⊗ e1 ⊗ e2 − e3 ⊗ e2 ⊗ e1 − e2 ⊗ e1 ⊗ e3 − e1 ⊗ e3 ⊗ e2
= e1 ⊗ (e2 ⊗ e3 − e3 ⊗ e2 ) − e2 ⊗ (e1 ⊗ e3 − e3 ⊗ e1 ) + e3 ⊗ (e1 ⊗ e2 − e2 ⊗ e1 )
= e1 ⊗ (e2 ∧ e3 ) − e2 ⊗ (e1 ∧ e3 ) + e3 ⊗ (e1 ∧ e2 ).
I submit to the reader that this

 is precisely
 the cofactor expansion formula with respect to the first
a b c
column of A. Suppose A =  d e f  then A1 = (a, d, g), A2 = (b, e, h) and A3 = (c, f, i).
g h i
Calculate,
det(A) = e1 (A1 )(e2 ∧ e3 )(A2 , A3 ) − e2 (A1 )(e1 ∧ e3 )(A2 , A3 ) + e3 (A1 )(e1 ∧ w2 )(A2 , A3 )
= a(e2 ∧ e3 )(A2 , A3 ) − d(e1 ∧ e3 )(A2 , A3 ) + g(e1 ∧ w2 )(A2 , A3 )
= a(ei − f h) − d(bi − ch) + g(bf − ce)
which is precisely my claim.
9.3.3 connecting vectors and forms in R3

There are a couple ways to connect vectors and forms in R3 . Mainly we need the following maps:
Definition 9.3.13.
Given v =< a, b, c >∈ R3 we can construct a corresponding one-form ωv = ae1 + be2 + ce3
or we can construct a corresponding two-form Φv = ae2 ∧ e3 + be3 ∧ e1 + ce1 ∧ e2
Recall that dim(Λ1 R3 ) = dim(Λ2 R3 ) = 3 hence the space of vectors, one-forms, and also two-
forms are isomorphic as vector spaces. It is not difficult to show that ωv1 +cv2 = ωv1 + cωv2 and
Φv1 +cv2 = Φv1 + cΦv2 for all v1 , v2 ∈ R3 and c ∈ R. Moreover, ωv = 0 iff v = 0 and Φv = 0 iff
v = 0 hence ker(ω) = {0} and ker(Φ) = {0} but this means that ω and Φ are injective and since
the dimensions of the domain and codomain are 3 and these are linear transformations17 it follows
ω and Φ are isomorphisms.
It appears we have two ways to represent vectors with forms in R3 . We’ll see why this is important
as we study integration of forms. It turns out the two-forms go with surfaces whereas the one-
forms attach to curves. This corresponds to the fact in calculus III we have two ways to integrate
a vector-field, we can either calculate flux or work. Partly for this reason the mapping ω is called
the work-form correspondence and Φ is called the flux-form correspondence. Integration
has to wait a bit, for now we focus on algebra.
Example 9.3.14. Suppose v =< 2, 0, 3 > and w =< 0, 1, 2 > then ωv = 2e1 +3e3 and ωw = e2 +2e3 .
Calculate the wedge product,
ωv ∧ ωw = (2e1 + 3e3 ) ∧ (e2 + 2e3 )
= 2e1 ∧ (e2 + 2e3 ) + 3e3 ∧ (e2 + 2e3 )
= 2e1 ∧ e2 + 4e1 ∧ e3 + 3e3 ∧ e2 + 6e3 ∧ e3
= −3e2 ∧ e3 − 4e3 ∧ e1 + 2e1 ∧ e2
= Φ<−3,−4,2>
= Φv×w (9.20)
Coincidence? Nope.
Proposition 9.3.15.
Suppose v, w ∈ R3 then
P3 ωv ∧ ωw = Φv×w where v × w denotes the cross-product which is
defined by v × w = i,j,k=1 ijk vi wj ek .
Proof: Suppose v = 3i=1 vi ei and w = 3j=1 wj ej then ωv = 3i=1 vi ei and ωw = 3j=1 wj ej .

P P P P
Calculate,
X 3 X 3 X3 X
3
ωv ∧ ωw = vi ei ∧ wj e j = v i wj e i ∧ e j
i=1 j=1 i=1 j=1
P3
In invite the reader to show ∧ ei ej
= Φ( k=1 ijk ek ) where I’m using Φv = Φ(v) to make the
argument of the flux-form mapping easier to read, hence,
3 X
X 3 X3 3
X
ωv ∧ ωw = vi wj Φ( ijk ek ) = Φ( vi wj ijk ek ) = Φ(v × w) = Φv×w
i=1 j=1 k=1 i,j,k=1
| {z }
linearity of Φ
17
this is not generally true, note f (x) = x2 has f (x) = 0 iff x = 0 and yet f is not injective. The linearity is key.
Of course, if you don’t like my proof you could just work it out like the example that precedes this
proposition. I gave the proof to show off the mappings a bit more.
Is the wedge product just the cross-product generalized? Well, not really. I think they’re quite
different animals. The wedge product is an associative product which makes sense in any vector
space. The cross-product only matches the wedge product after we interpret it through a pair of
isomorphisms (ω and φ) which are special to R3 . However, there is debate, largely the question
comes down to what you think makes the cross-product the cross-product. If you think it must
pick a unique perpendicular direction to a pair of given directions then that is only going to work
in R3 since even in R4 there is a whole plane of perpendicular vectors to a given pair. On the other
hand, if you think the cross-product in R4 should be pick the unique perpendicular to a given triple
of vectors then you could set something up. You could define v × w × x = ω −1 (ψ(ωv ∧ ωw ∧ ωx ))
where ψ : Λ3 R4 → Λ1 R4 is an isomorphism we’ll describe in a upcoming section. But, you see it’s
no longer a product of two vectors, it’s not a binary operation, it’s a tertiary operation. In any
event, you can read a lot more on this if you wish. We have all the tools we need for this course.
The wedge product provides the natural antisymmetric algebra for n-dimensiona and the work and
flux-form maps naturally connect us to the special world of three-dimensional mathematics.
There is more algebra for forms on R3 however we defer it to a later section where we have a few
more tools. Chief among those is the Hodge dual. But, before we can discuss Hodge duality we
need to generalize our idea of a dot-product just a little.
9.4 bilinear forms and geometry, metric duality

The concept of a metric goes beyond the familar case of the dot-product. If you want a more
strict generalization of the dot-product then you should think about an inner-product. In contrast
to the definition below, the inner-product replaces non-degeneracy with the stricter condition of
positive-definite which would read g(x, x) > 0 for x 6= 0. I included a discussion of inner products
at the end of this section for the interested reader although we are probably not going to need all
of that material.
9.4.1 metric geometry

A geometry is a vector space paired with a metric. For example, if we pair Rn with the dot-
product you get Euclidean space. However, if we pair R4 with the Minkowski metric then we
obtain Minkowski space.
9.4. BILINEAR FORMS AND GEOMETRY, METRIC DUALITY 211
Definition 9.4.1.
If V is a vector space and g : V × V → R is
1. bilinear: g ∈ T20 V ,
2. symmetric: g(x, y) = g(y, x) for all x, y ∈ V ,
3. nondegenerate: g(x, y) = 0 for all x ∈ V implies y = 0.
the we call g a metric on V .

If V = Rn then we can write g(x, y) = xT Gy where [g] = G. Moreover, g(x, y) = g(y, x) implies
GT = G. Nondegenerate means that g(x, y) = 0 for all y ∈ Rn iff x = 0. It follows that Gy = 0
has no non-trivial solutions hence G−1 exists.
Example 9.4.2. Suppose g(x, y) = xT y for all x, y ∈ Rn . This defines a metric for Rn , it is just
the dot-product. Note that g(x, y) = xT y = xT Iy hence we see [g] = I where I denotes the identity
matrix in R n×n .
Example 9.4.3. Suppose v = (v 0 , v 1 , v 2 , v 3 ), w = (w0 , w1 , w2 , w3 ) ∈ R4 then define the Minkowski

product of v and w as follows:
g(v, w) = −v 0 w0 + v 1 w1 + v 2 w2 + v 3 w3
It is useful to write the Minkowski product in terms of a matrix multiplication. Observe that for
x, y ∈ R4 ,
   0
−1 0 0 0 y
0 1 0 0 1
g(x, y) = −x0 y 0 + x1 y 1 + x2 y 2 + x3 y 3 = x0 x1 x2 x3   y  ≡ xt ηy
  
 0 0 1 0 y 2 
0 0 0 1 y3
where we have introduced η the matrix of the Minkowski product. Notice that η T = η and det(η) =
−1 6= 0 hence g(x, y) = xt ηy makes g a symmetric, nondegenerate bilinear form on R4 . The
formula is clearly related to the dot-product. Suppose v̄ = (v 0 , ~v ) and w̄ = (w0 , w)
~ then note
g(v, w) = −v 0 w0 + ~v · w
~
For vectors with zero in the zeroth slot this Minkowski product reduces to the dot-product. However,
for vectors which have nonzero entries in both the zeroth and later slots much differs. Recall that
any vector’s dot-product with itself gives the square of the vectors length. Of course this means that
~x · ~x = 0 iff ~x = 0. Contrast that with the following: if v = (1, 1, 0, 0) then
g(v, v) = −1 + 1 = 0
Yet v 6= 0. Why study such a strange generalization of length? The answer lies in physics. I’ll give
you a brief account by defining a few terms: Let v = (v 0 , v 1 , v 2 , v 3 ) ∈ R4 then we say
1. v is a timelike vector if < v, v > < 0
2. v is a lightlike vector if < v, v > = 0
3. v is a spacelike vector if < v, v > > 0
If we consider the trajectory of a massive particle in R4 that begins at the origin then at any later
time the trajectory will be located at a timelike vector. If we consider a light beam emitted from
the origin then at any future time it will located at the tip of a lightlike vector. Finally, spacelike
vectors point to points in R4 which cannot be reached by the motion of physical particles that pass
throughout the origin. We say that massive particles are confined within their light cones, this
means that they are always located at timelike vectors relative to their current position in space
time. If you’d like to know more I can reccomend a few books.
At this point you might wonder if there are other types of metrics beyond these two examples.
Surprisingly, in a certain sense, no. A rather old theorem of linear algebra due to Sylvester states
that we can change coordinates so that the metric more or less resembles either the dot-product or
something like it with some sign-flips. We’ll return to this in a later section.
9.4.2 metric duality for tensors

Throughout this section we consider a vector space V paired with a metric g : V × V → R.
Moreover, the vector space V has basis {ei }ni=1 which has a g-dual basis {ei }ni=1 . Up to this point
we always have used a g-dual basis where the duality was offered by the dot-product. In the
context of Minkowski geometry that sort of duality is no longer natural. Instead we must follow
the definition below:
Definition 9.4.4.
If V is a vector space with metric g and basis {ei }ni=1 then we say the basis {ei }ni=1 is g-dual
iff
Suppose ei (ej ) = δij and consider g = ni,j=1 gij ei ⊗ ej . Furthermore, suppose g ij are the com-
P
ponents of the inverse matrix to (gij ) this means that nk=1 gik g kj = δij . We use the components
P
of the metric and its inverse to raise and lower indices on tensors. Here are the basic conven-
tions: given an object Aj which has the contravariant index j we can lower it to be covariant by
contracting against the metric components as follows:
X
Ai = gij Aj
j
On the other hand, given an object Bj which has a covariant index j we can raise it to be con-
travariant by contracting against the inverse components of the metric:
X
Bi = g ij Bj
j
I like to think of this as some sort

Pof conservation of indices. Strict adherence to the notation
drives us to write things such as nk=1 gik g kj = δij just to keep up the up/down index pattern.
I should mention that these formulas are much more beautiful in the physics literature, you can
look at my old Math 430 notes from NCSU if you’d like a healthy dose of that notation18 . I use
Einstein’s implicit summation notation throughout those notes and I discuss this index calculation
more in the way a physicist typically approaches it. Here I am trying to be careful enough that
these equations are useful to mathematicians. Let me show you some examples:
Example
P4 9.4.5. Specialize for this example to V = R4 with g(x, y) = xT ηy. Suppose x =
µ µ
µ=0 x eµ the components x are called contravariant components. The metric allows us
to define covariant components by
4
X
xν = ηνµ xµ .
µ=0
For the minkowski metric this just adjoins a minus to the zeroth component: if (xµ ) = (a, b, c, d)
then xµ = (−a, b, c, d).
Example 9.4.6. Suppose we are working on Rn with the Euclidean metric gij = δij and it follows
ij = δ kj = δ j . In this case v i =
P P ij
that
P g ij or to be a purist for a moment k gik g i j g vj =
j δij v j = v i . The covariant and contravariant components are the same. This is why is was ok
to ignore up/down indices when we work with a dot-product exclusively.
What if we raise an index and the lower it backPdown once more? Do we really get back where we
started? Given xµ we lower the index by xν = µ gνµ xµ then we raise it once more by
X X X X X
xα = g αν xν = g αν gνµ xµ = g αν gνµ xµ = δµα xµ
ν ν µ µ,ν µ
and the last summation squishes down to xα once more. It would seem this procedure of raising
and lowering indices is at least consistent.
Example 9.4.7. Suppose we raise the index on the basis {ei } and formally obtain {ej = k g jk ek }
P
on
P the other hand suppose we lower the index on the dual basis {el } to formally obtain {em =
l j j
l gml e }. I’m curious, are these consistent? We should get e (em ) = δm , I’ll be nice an look at
em (ej ) in the following sense:
X X X X X
l jk
gml e g ek = gml g jk el (ek ) = gml g jk δkl = gmk g jk = δm
j
l k l,k l,k k
Interesting, but what does it mean?

18
just a taste: vµ = ηµν v ν or v µ = η µν vν or v µ vµ = η µν vν vµ = ηµν v µ v ν
I used the term formal in the preceding example to mean that the example makes sense in as much
as you accept the equations which are written. If you think harder about it then you’ll find it was
rather meaningless. That said, this index notation is rather forgiving.
Ok, but what are we doing? Recall that I insisted on using lower indices for forms and upper
indices for vectors? The index conventions I’m toying with above are the reason for this strange
notation. When we lower an index we might be changing a vector to a dual vector, or vice-versa
when we raise an index we might be changing a dual vector into a vector. Let me be explicit.
1. given v ∈ V we create αv ∈ V ∗ by the rule αv (x) = g(x, v).
2. given α ∈ V ∗ we create vα ∈ V ∗∗ by the rule vα (β) = g −1 (α, β) where g −1 (α, β) = αi βj g ij .

P
ij
Recall we at times identify V and V ∗∗ . Let’s work out the component structure of αv and see how
it relates to v,
X X X
αv (ei ) = g(v, ei ) = g( v j ej , ei ) = v j g(ej , ei ) = v j gji
j j j
i j
P P
Thus, αv = i vi e where vi = j v gji . When we lower the index we’re actually using an
isomorphism which is provided by the metric to map vectors to forms. The process of raising the
index is just the inverse of this isomorphism.
X X
vα (ei ) = g −1 (α, ei ) = g −1 ( αj ej , ei ) = αj g ji
j j
ie where αi = αj g ji .
P P
thus vα = iα i j
want to change a type (0, 2) tensor to a type (2, 0) tensor. We’re given T : V ∗ × V ∗
Suppose weP
where T = ij T ij ei ⊗ ej . Define T̃ : V × V → R as follows:
T̃ (v, w) = T (αv , αw )
gij ej and
P
What does this look like in components? Note αei (ej ) = g(ei , ej ) = gij hence αei = j
X X X X
k l
T̃ (ei , ej ) = T (αei , αej ) = T gik e , gjl e = gki glj T (ek , el ) = gki glj T kl
k l k,l k,l
Or, as is often customary, we could write Tij = k,l gik gjl T kl . However, this is an abuse of notation
P
since Tij are not technically components for T . If we have a metric we can recover either T from T̃
or vice-versa. Generally, if we are given two tensors, say T1 of rank (r, s) and the T2 of rank (r0 , s0 ),
then these might be equilvalent if r + s = r0 + s0 . It may be that through raising and lowering
indices (a.k.a. appropriately composing with the vector↔dual vector isomorphisms) we can convert
T1 to T2 . If you read Gravitation by Misner, Thorne and Wheeler you’ll find many more thoughts
on this equivalence. Challenge: can you find the explicit formulas like T̃ (v, w) = T (αv , αw ) which
back up the index calculations below?
X X
Tij k = gia gjb T abk or S ij = g ia g jb Sab
a,b a,b
I hope I’ve given you enough to chew on in this section to put these together.
9.4.3 inner products and induced norm

There are generalized dot-products on many abstract vector spaces, we call them inner-products.
Definition 9.4.8.
Suppose V is a vector space. If <, >: V × V → R is a function such that for all x, y, z ∈ V
and c ∈ R:
1. < x, y >=< y, x > (symmetric)
2. < x + y, z >=< x, z > + < y, z > (additive in the first slot)
3. < cx, y >= c < x, y > (homogeneity in the first slot)
4. < x, x > ≥ 0 and < x, x >= 0 iff x = 0
then we say (V, <, >) is an inner-product space with inner product <, >.
Given an inner-product space (V, <, >) we can easily induce a norm for V by the formula ||x|| =
√
< x, x > for all x ∈ V . Properties (1.), (3.) and (4.) in the definition of the norm are fairly obvious
for the induced norm. Let’s think throught the triangle inequality for the induced norm:
||x + y||2 = < x + y, x + y > def. of induced norm

= < x, x + y > + < y, x + y > additive prop. of inner prod.
= < x + y, x > + < x + y, y > symmetric prop. of inner prod.
= < x, x > + < y, x > + < x, y > + < y, y > additive prop. of inner prod.
= ||x||2 + 2 < x, y > +||y||2
At this point we’re stuck. A nontrivial identity19 called the Cauchy-Schwarz identity helps us
proceed; < x, y >≤ ||x||||y||. It follows that ||x + y||2 ≤ ||x||2 + 2||x||||y|| + ||y||2 = (||x|| + ||y||)2 .
However, the induced norm is clearly positive20 so we find ||x + y|| ≤ ||x|| + ||y||.
Most linear algebra texts have a whole chapter on inner-products and their applications, you can
look at my notes for a start if you’re curious. That said, this is a bit of a digression for this course.
19
I prove this for the dot-product in my linear notes, however, the proof is written in such a way it equally well
applies to a general inner-product
20
note: if you have (−5)2 < (−7)2 it does not follow that −5 < −7, in order to take the squareroot of the inequality
we need positive terms squared
9.5 hodge duality

We can prove that np = n−pn
. This follows from explicit computation of the formula for np or

from the symmetry of Pascal’s triangle if you prefer. In any event, this equality suggests there is
some isomorphism between p and (n − p)-forms. When we are given a metric g on a vector space
V (and the notation of the preceding section) it is fairly simple to construct the isomorphism.
Suppose we are given α ∈ Λp V and following our usual notation:
n
X 1
α= αi i ...i ei1 ∧ ei2 ∧ · · · ∧ eip
p! 1 2 p
i1 ,i2 ,...,ip =1
Then, define ∗α the hodge dual to be the (n − p)-form given below:
n
X 1
∗α = αi1 i2 ...ip i1 i2 ...ip j1 j2 ...jn−p ej1 ∧ ej2 ∧ · · · ∧ ejn−p
p!(n − p)!
i1 ,i2 ,...,in =1
I should admit, to prove this is a reasonable definition we’d need to do some work. It’s clearly a
linear transformation, but bijectivity and coordinate invariance of this definition might take a little
work. I intend to omit those details and instead focus on how this works for R3 or R4 . My advisor
taught a course on fiber bundles and there is a much more general and elegant presentation of the
hodge dual over a manifold. Ask if interested, I think I have a pdf.
9.5.1 hodge duality in euclidean space R3

To begin, consider a scalar 1, this is a 0-form so we expect the hodge dual to give a 3-form:
X 1
∗1 = ijk ei ∧ ej ∧ ek = e1 ∧ e2 ∧ e3
0!3!
i,j,k
Interesting, the hodge dual of 1 is the top-form on R3 . Conversely, calculate the dual of the top-
form, note e1 ∧ e2 ∧ e3 = ijk 16 ijk ei ∧ ej ∧ ek reveals the components of the top-form are precisely
P
ijk thus:
3
1 2 3
X 1 1
∗(e ∧ e ∧ e ) = ijk ijk = (1 + 1 + 1 + (−1)2 + (−1)2 + (−1)2 ) = 1.
3!(3 − 3)! 6
i,j,k=1
Next, consider e1 , note that e1 = k

P
k δk1 e hence the components are δk1 . Thus,
X 1 1X 1
∗e1 = ijk δ k1 ei ∧ ej = ij1 ei ∧ ej = (231 e2 ∧ e3 + 321 e3 ∧ e2 ) = e2 ∧ e3
1!2! 2 2
i,j,k i,j
Similar calculations reveal ∗e2 = e3 ∧ e1 and ∗e3 = e1 ∧ e2 . What about the duals of the two-forms?
Begin with α = e1 ∧ e2 note that e1 ∧ e2 = e1 ⊗ e2 − e2 ⊗ e1 thus we can see the components are
9.5. HODGE DUALITY 217
αij = δi1 δj2 − δi2 δj1 . Thus,

X 1
1 2 k 1 X k
X 1
∗(e ∧ e ) = ijk (δi1 δj2 − δi2 δj1 )e = 12k e − 21k e = (e3 − (−e3 )) = e3 .
k
2!1! 2 2
i,j,k k k
Similar calculations show that ∗(e2 ∧ e3 ) = e1 and ∗(e3 ∧ e1 ) = e2 . Put all of this together and we
find that
∗(ae1 + be2 + ce3 ) = ae2 ∧ e3 + be3 ∧ e1 + ce1 ∧ e2
and
∗(ae2 ∧ e3 + be3 ∧ e1 + ce1 ∧ e2 ) = ae1 + be2 + ce3
Which means that ∗ωv = Φv and ∗Φv = ωv . Hodge duality links the two different form-representations
of vectors in a natural manner. Moveover, for R3 we should also note that ∗∗α = α for all α ∈ ΛR3 .
In general, for other metrics, we can have a change of signs which depends on the degree of α.
We can summarize hodge duality for three-dimensional Euclidean space as follows:

∗1= e1 ∧ e2 ∧ e3 ∗ (e1 ∧ e2 ∧ e3 ) = 1
∗ e1 = e2 ∧ e3 ∗ (e2 ∧ e3 ) = e1
∗ e2 = e3 ∧ e1 ∗ (e3 ∧ e1 ) = e2
∗ e3 = e1 ∧ e2 ∗ (e1 ∧ e2 ) = e3
A simple rule to calculate the hodge dual of a basis form is as follows

1. begin with the top-form e1 ∧ e2 ∧ e3
2. permute the forms until the basis form you wish to hodge dual is to the left of the expression,
whatever remains to the right is the hodge dual.
For example, to calculate the dual of e2 ∧ e3 note
e1 ∧ e2 ∧ e3 = e|2 {z
∧ e}3 ∧ |{z}
e1 ⇒ ∗(e2 ∧ e3 ) = e1 .
to be dualed the dual
Consider what happens if we calculate ∗ ∗ α, since the dual is a linear operation it suffices to think
about the basis forms. Let me sketch the process of ∗ ∗ eI where I is a multi-index:
1. begin with e1 ∧ e2 ∧ e3
2. write e1 ∧ e2 ∧ e3 = (−1)N eI ∧ eJ and identify ∗eI = (−1)N eJ .
3. then to calculate the second dual once more begin with e1 ∧ e2 ∧ e3 and note
e1 ∧ e2 ∧ e3 = (−1)N eJ ∧ eI
since the same N transpositions are required to push eI to the left or eJ to the right.
4. It follows that ∗ ∗ eI = eI for any multi-index hence ∗ ∗ α = α for all α ∈ ΛR3 .
I hope that once you get past the index calculation you can see the hodge dual is not a terribly
complicated construction. Some of the index calculation in this section was probably gratutious,
but I would like you to be aware of such techniques. Brute-force calculation has it’s place, but a
well-thought index notation can bring far more insight with much less effort.
9.5.2 hodge duality in minkowski space R4

The logic here follows fairly close to the last section, however the wrinkle is that the metric here
demands more attention. We must take care to raise the indices on the forms when we Hodge dual
them. First let’s list the basis forms, we have to add time to the mix ( again c = 1 so x0 = ct = t if
you worried about it ) Remember that the Greek indices are defined to range over 0, 1, 2, 3. Here
Name Degree Typical Element Basis for Λp R4

function p=0 Pf 1
one-form p=1 α = µ αµ e µ e0 , e1 , e2 , e3
β = µ,ν 12 βµν eµ ∧ eν e2 ∧ e3 , e3 ∧ e1 , e1 ∧ e2
P
two-form p=2
e0 ∧ e1 , e0 ∧ e2 , e0 ∧ e3
1 µ ∧ eν eα e1 ∧ e2 ∧ e3 , e0 ∧ e2 ∧ e3
P
three-form p=3 γ= µ,ν,α 3! γµνα e
e0 ∧ e1 ∧ e3 , e0 ∧ e1 ∧ e2
four-form p=4 ge0 ∧ e1 ∧ e2 ∧ e3 e0 ∧ e1 ∧ e2 ∧ e3
the top form is degree four since in four dimensions we can have at most four dual-basis vectors
without a repeat. Wedge products work the same as they have before, just now we have e0 to play
with. Hodge duality may offer some surprises though.
Definition 9.5.1. The antisymmetric symbol in flat R4 is denoted µναβ and it is defined by the
value
0123 = 1
plus the demand that it be completely antisymmetric.
We must not assume that this symbol is invariant under a cyclic exhange of indices. Consider,
0123 = −1023 flipped (01)

= +1203 flipped (02) (9.21)
= −1230 flipped (03).
In four dimensions we’ll use antisymmetry directly and forego the cyclicity shortcut. Its not a big
deal if you notice it before it confuses you.
Example 9.5.2. Find the Hodge dual of γ = e1 with respect to the Minkowski metric Pηµν , to begin
notice that dx has components γµ = δµ1 as is readily verified by the equation e1 = µ δµ1 eµ . Lets
raise the index using η as we learned previously,

X X
γµ = η µν γν = η µν δν1 = η 1µ = δ 1µ
ν ν
Starting with the definition of Hodge duality we calculate

∗ (e1 ) = 1 1 µ ν α β
P
α,β,µ,ν p! (n−p)! γ µναβ e ∧ e ∧ e
1µ ν ∧ eα ∧ eβ
P
= α,β,µ,ν (1/6)δ µναβ e
ν ∧ eα ∧ eβ
P
= α,β,ν (1/6)1ναβ e
= (1/6)[1023 e0 ∧ e2 ∧ e3 + 1230 e2 ∧ e3 ∧ e0 + 1302 e3 ∧ e0 ∧ e2 (9.22)

+1320 e3 ∧ e2 ∧ e0 + 1203 e2 ∧ e0 ∧ e3 + 1032 e0 ∧ e3 ∧ e2 ]
= (1/6)[−e0 ∧ e2 ∧ e3 − e2 ∧ e3 ∧ e0 − e3 ∧ e0 ∧ e2
+e3 ∧ e2 ∧ e0 + e2 ∧ e0 ∧ e3 + e0 ∧ e3 ∧ e2 ]
= −e2 ∧ e3 ∧ e0 = −e0 ∧ e2 ∧ e3 .
the difference between the three and four dimensional Hodge dual arises from two sources, for one
we are using the Minkowski metric so indices up or down makes a difference, and second the
antisymmetric symbol has more possibilities than before because the Greek indices take four values.
I suspect we can calculate the hodge dual by the following pattern: suppose we wish to find the
dual of α where α is a basis form for ΛR4 with the Minkowski metric
1. begin with the top-form e0 ∧ e1 ∧ e2 ∧ e3
2. permute factors as needed to place α to the left,
3. the form which remains to the right will be the hodge dual of α if no e0 is in α otherwise the
form to the right multiplied by −1 is ∗α.
Note this works for the previous example as follows:
1. begin with e0 ∧ e1 ∧ e2 ∧ e3
2. note e0 ∧ e1 ∧ e2 ∧ e3 = −e1 ∧ e0 ∧ e2 ∧ e3 = e1 ∧ (−e0 ∧ e2 ∧ e3 )
3. identify ∗e1 = −e0 ∧ e2 ∧ e3 (no extra sign since no e0 appears in e1 )
Follow the algorithm for finding the dual of e0 ,
1. begin with e0 ∧ e1 ∧ e2 ∧ e3
2. note e0 ∧ e1 ∧ e2 ∧ e3 = e0 ∧ (e1 ∧ e2 ∧ e3 )
3. identify ∗e0 = −e1 ∧ e2 ∧ e3 ( added sign since e0 appears in form being hodge dualed)
Let’s check from the definition if my algorithm worked out right.
Example 9.5.3. Find the Hodge dual of γ = e0 with respect to the Minkowski metric Pηµν , to begin
notice that e0 has components γµ = δµ0 as is readily verified by the equation e0 = µ δµ0 eµ . Lets
raise the index using η as we learned previously,
X X
γµ = η µν γν = η µν δν0 = η µ0 = −δ 0µ
ν ν
the minus sign is due to the Minkowski metric. Starting with the definition of Hodge duality we
calculate
∗ (e0 ) = 1 1 µ ν α β
P
α,β,µ,ν p! (n−p)! γ µναβ e ∧ e ∧ e
−(1/6)δ 0µ µναβ eν ∧ eα ∧ eβ
P
= α,β,µ,ν
−(1/6)0ναβ eν ∧ eα ∧ eβ
P
= α,β,ν (9.23)
= i,j,k −(1/6)0ijk ei ∧ ej ∧ ek
P
= i,j,k −(1/6)ijk ijk e1 ∧ e2 ∧ e3

P
← sneaky step
= −e1 ∧ e2 ∧ e3 .
Notice I am using the convention that Greek indices sum over 0, 1, 2, 3 whereas Latin indices sum
over 1, 2, 3.
Example 9.5.4. Find the Hodge dual of γ = e0 ∧ e1 with respect to the Minkowski metric ηµν , to
begin notice the following identity, it will help us find the components of γ
X1
e0 ∧ e1 = 2δµ0 δν1 eµ ∧ eν
µ,ν
2
now we antisymmetrize to get the components of the form,

X1
e0 ∧ e1 = 0 1
δ[µ δν] dxµ ∧ dxν
µ,ν
2
0 δ 1 = δ 0 δ 1 − δ 0 δ 1 and the factor of two is used up in the antisymmetrization. Lets raise

where δ[µ ν] µ ν ν µ
the index using η as we learned previously,
X X
γ αβ = η αµ η βν γµν = η αµ η βν δ[µ
0 1
δν] = −η α0 η β1 + η β0 η α1 = −δ [α0 δ β]1
µ,ν µ,ν
the minus sign is due to the Minkowski metric. Starting with the definition of Hodge duality we
calculate
∗ (e0 ∧ e1 ) = 1 1 αβ µ ∧ eν
p! (n−p)! γ αβµν e
= (1/4)(−δ [α0 δ β]1 )αβµν eµ ∧ eν
= −(1/4)(01µν eµ ∧ eν − 10µν eµ ∧ eν )
(9.24)
= −(1/2)01µν eµ ∧ eν
= −(1/2)[0123 e2 ∧ e3 + 0132 e3 ∧ e2 ]
= −e2 ∧ e3
Note, the algorithm works out the same,
∧ e}1 ∧(e2 ∧ e3 ) ⇒ ∗(e0 ∧ e1 ) = −e2 ∧ e3

e0 ∧ e1 ∧ e2 ∧ e3 = e|0 {z
has e0
The other Hodge duals of the basic two-forms calculate by almost the same calculation. Let us make
a table of all the basic Hodge dualities in Minkowski space, I have grouped the terms to emphasize
∗1 = e0 ∧ e1 ∧ e2 ∧ e3 ∗ (e0 ∧ e1 ∧ e2 ∧ e3 ) = −1
∗ (e1 ∧ e2 ∧ e3 ) = −e0 ∗ e0 = −e1 ∧ e2 ∧ e3
∗ (e0 ∧ e2 ∧ e3 ) = −e1 ∗ e1 = −e2 ∧ e3 ∧ e0
∗ (e0 ∧ e3 ∧ e1 ) = −e2 ∗ e2 = −e3 ∧ e1 ∧ e0
∗ (e0 ∧ e1 ∧ e2 ) = −e3 ∗ e3 = −e1 ∧ e2 ∧ e0
∗ (e3 ∧ e0 ) = e1 ∧ e2 ∗ (e1 ∧ e2 ) = −e3 ∧ e0
∗ (e1 ∧ e0 ) = e2 ∧ e3 ∗ (e2 ∧ e3 ) = −e1 ∧ e0
∗ (e2 ∧ e0 ) = e3 ∧ e1 ∗ (e3 ∧ e1 ) = −e2 ∧ e0
the isomorphisms between the one-dimensional Λ0 R4 and Λ4 R4 , the four-dimensional Λ1 R4 and

Λ3 R4 , the six-dimensional Λ2 R4 and itself. Notice that the dimension of ΛR4 is 16 which we have
explained in depth in the previous section. Finally, it is useful to point out the three-dimensional
work and flux form mappings to provide some useful identities in this 1 + 3-dimensional setting.
∗ω~v = −e0 ∧ Φ~v ∗ Φ~v = e0 ∧ ω~v ∗ e0 ∧ Φ~v = ω~v

I leave verification of these formulas to the reader ( use the table). Finally let us analyze the process
of taking two hodge duals in succession. In the context of R3 we found that ∗ ∗ α = α, we seek to
discern if a similar formula is available in the context of R4 with the minkowksi metric. We can
calculate one type of example with the identities above:
∗ω~v = −e0 ∧ Φ~v ⇒ ∗ ∗ ω~v = − ∗ (e0 ∧ Φ~v ) = −ω~v ⇒ ∗ ∗ ω~v = −ω~v

Perhaps this is true in general?
If we accept my algorithm then it’s not too hard to sort through using multi-index notation: since
hodge duality is linear it suffices to consider a basis element eI where I is a multi-index,
1. transpose dual vectors so that e0 ∧ e1 ∧ e2 ∧ e3 = (−1)N eI ∧ eJ
/ I then ∗eI = (−1)N eJ and 0 ∈ J since I ∪ J = {0, 1, 2, 3}. Take a second dual by
2. if 0 ∈
writing e0 ∧ e1 ∧ e2 ∧ e3 = (−1)N eJ ∧ eI but note ∗((−1)N eJ ) = −eI since 0 ∈ J. We find
∗ ∗ eI = −eI for all I not containing the 0-index.
3. if 0 ∈ I then ∗eI = −(−1)N eJ and 0 ∈ / J since I ∪ J = {0, 1, 2, 3}. Take a second dual by
writing e0 ∧ e1 ∧ e2 ∧ e3 = −(−1)N eJ ∧ (−eI ) and hence ∗(−(−1)N eJ ) = −eI since 0 ∈
/ J. We
find ∗ ∗ eI = −eI for all I containing the 0-index.
4. it follows that ∗ ∗ α = −α for all α ∈ ΛR4 with the minkowski metric.

To conclude, I would warn the reader that the results in this section pertain to our choice of notation
for R4 . Some other texts use a metric which is −η relative to our notation. This modifies many
signs in this section. See Misner, Thorne and Wheeler’s Gravitation or Bertlmann’s Anomalies in
Field Theory for future reading on Hodge duality and a more systematic explaination of how and
when these signs arise from the metric.
9.6 coordinate change

Suppose V has two bases β̄ = {f¯1 , f¯2 , . . . , f¯n } and β = {f1 , f2 , . . . , fn }. If v ∈ V then we can write
v in as a linear combination of the β̄ basis or the β basis:
v = x1 f1 + x2 f2 + · · · + xn fn and v = x̄1 f¯1 + x̄2 f¯2 + · · · + x̄n f¯n
given the notation above, we define coordinate maps as follows:
Φβ (v) = (x1 , x2 , . . . , xn ) = x and Φβ̄ (v) = (x̄1 , x̄2 , . . . , x̄n ) = x̄
We sometimes use the notation Φβ (v) = [v]β = x whereas Φβ̄ (v) = [v]β̄ = x̄. A coordinate map
takes an abstract vector v and maps it to a particular representative in Rn . A natural question
to ask is how do different representatives compare? How do x and x̄ compare in our current
notation? Because the coordinate maps are isomorphisms it follows that Φβ ◦ Φ−1
β̄
: Rn → Rn is an
isomorphism and given the domain and codomain we can write its formula via matrix multiplication:
Φβ ◦ Φ−1
β̄
(u) = P u ⇒ Φβ ◦ Φ−1
β̄
(x̄) = P x̄
However, Φ−1
β̄
(x̄) = v hence Φβ (v) = P x̄ and consequently, x = P x̄ . Conversely, to switch to
barred coordinates we multiply the coordinate vectors by P −1 ; x̄ = P −1 x .
9.6. COORDINATE CHANGE 223
Continuing this discussion we turn to the dual space. Suppose β̄ ∗ = {f¯j }nj=1 is dual to β̄ = {f¯j }nj=1
and β ∗ = {f j }nj=1 is dual to β = {fj }nj=1 . By definition we are given that f j (fi ) = δij and
f¯j (f¯i ) = δij for all i, j ∈ Nn . Suppose α ∈ V ∗ is a dual vector with components αj with respect
to the β ∗ basis and Pncomponents with respect to the β̄ ∗ basis. In particular this means we can
ᾱj P
either write α = j=1 αj f or α = nj=1 ᾱj f¯j . Likewise, given a vector v ∈ V we can either write
j
v = ni=1 xi fi or v = ni=1 x̄i f¯i . With these notations in mind calculate:

P P
n
X n
X n
X n X
X n n
X
j i i j i
αi xi

α(v) = αj f x fi = αj x f (fi ) = αj x δij =
j=1 i=1 i,j=1 i=1 j=1 i=1
Pn i.
and by the same calculation in the barred coordinates we find, α(v) = i=1 ᾱi x̄ Therefore,
n
X n
X
i
αi x = ᾱi x̄i .
i=1 i=1
Pn
Recall, x = P x̄. In components, xi = i k
k=1 Pk x̄ . Substituting,
n X
X n n
X
αi Pki x̄k = ᾱi x̄i .
i=1 k=1 i=1
But, this formula holds for all possiblePn vectors vP

and hence all possible coordinate vectors x̄. If we
= δij hence i=1 ᾱi x̄ =P i=1 ᾱi δij = ᾱj . Moreover, i=1 nk=1 αi Pki x̄k =
n
¯ i i
Pn P
consider
Pn Pnv = fj then x̄ P
i n i n i
i=1 k=1 αi Pk δkj = i=1 αi Pj . Thus, ᾱj = i=1 Pj αi . Compare how vectors and dual vectors
transform:
n n
(P −1 )ji xi .
X X
ᾱj = Pji αi verses x̄j =
i=1 i=1
It is customary to use lower-indices onPnthe components of dual-vectors and upper-indices on the

components of vectors: we say x = x i e ∈ Rn has contravariant components whereas
Pn i=1 i
j n ∗
α = j=1 αj e ∈ (R ) has covariant components. These terms arise from the coordinate
change properties we derived in this section. The convenience of the up/down index notation will
be more apparent as we continue our study to more complicated objects. It is interesting to note
the basis elements tranform inversely:
n n
(P −1 )ji f i
X X
f¯j = verses f¯j = Pji fi .
i=1 i=1
The formulas above can be derived by arguments similar to those we already gave in this section,
however I think it may be more instructive to see how these rules work in concert:
n
X n X
X n
x= x̄i f¯i = (P −1 )ij xj f¯i (9.25)
i=1 i=1 j=1
n X
X n n
X
= (P −1 )ij xj Pik fk
i=1 j=1 k=1
Xn X n Xn
= (P −1 )ij Pik xj fk
i=1 j=1 k=1
Xn X n
= δjk xj fk
j=1 k=1
Xn
= xk fk .
k=1
9.6.1 coordinate change for T20 (V )

For an abstract vector space, or for Rn with a nonstandard basis, we have to replace v, w with their
coordinate vectors. If V has basis β̄ = {f¯1 , f¯2 , . . . , fP
¯n } with dual basis β ∗ = {f¯1 , f¯2 , . . . , f¯n } and
v, w have coordinate vectors x̄, ȳ (which means v = ni=1 x̄i f¯i and w = ni=1 ȳ i f¯i ) then,
P
n
X
b(v, w) = x̄i ȳ j B̄ij = x̄T B̄ ȳ
i,j=1
where B̄ij = b(f¯i , f¯j ). If β = {f1 , f2 , . . . , fn } is another basis on V with dual basis β ∗ then we
define Bij = b(fi , fj ) and we have
n
X
b(v, w) = xi y j Bij = xT By.
i,j=1
Recall that f¯i = nk=1 Pik fk . With this in mind calculate:

P
n
X n
X n
X n
X
¯ ¯
B̄ij = b(fi , fj ) = b k
Pi fk , l
Pj fl = Pik Pjl b(fk , fl ) = Pik Pjl Bkl
k=1 l=1 k,l=1 k,l=1
We find the components of a bilinear map transform as follows:
n
X
B̄ij = Pik Pjl Bkl
k,l=1
Chapter 10
calculus with differential forms
A manifold is an abtract space which allows for local calculus. We discuss how coordinate charts
cover the a manifold and how we use them to define smoothness in the abstract. For example,
a function is smooth if all its local coordinate representations are smooth. The local coordinate
representative is a function from Rm to Rn thus we may quantify its smoothness in terms of or-
dinary partial derivatives of the component functions. On the other hand, while the concept of a
coordinate chart is at first glance abstract the usual theorems of advanced calculus all lift naturally
to the abstract manifold. For example, we see how partial derivatives with respect to manifold
coordinates hold to all the usual linearity, product and chain-rules hold in Rn . We prove a number
of these results in considerably more detail than Lecture will bear.
The differential gains a deeper meaning than we found in advanced calculus. In the manifold con-
text, the differential acts on tangent space which is not identified as some subset of the manifold
itself. So, in a sense we lose the direct approximating concept for the differential. One could always
return to the best linear approximation ideal as needed, but the path ahead is quite removed from
pure numerical approximation. First step towards this abstract picture of tangent space is the real-
ization that tangent vectors themselves should be identified as derivations. We show how partial
derivatives give derivations and we sketch a technical result which also provides a converse in the
smooth category1 Once we’ve settled how to study the tangent space to a manifold we find the
natural extension of the differential as the push-forward induced from a smooth map. It turns
out that you have probably already calculated a few push-forwards in multivariate calculus. We
attempt to obtain some intuition for this abstract push-forward. We also pause to note how the
push-forward might allow us to create new vectors fields from old (or not).
The cotangent space is the dual space to the tangent space. The basis which is dual to the partial
derivative basis is naturally identified with the differentials of the coordinate maps themselves. We
then have a basis and dual basis for the tangent and cotangent space at each point on the manifold.
1
it’s actually false for C k manifolds which have an infinite-dimensional space of derivations. The tangent space to
a n-dimensional manifold is an n-dimensional vector space so we need an n-dimensional space of derivations to make
the identification.
225
226 CHAPTER 10. CALCULUS WITH DIFFERENTIAL FORMS
We use these to build vector fields and differential forms just as we used fi and f i in the previous
chapter. Multilinear algebra transfers over point by point on the manifold. However, something
new happens. Differential forms permit a differentiation called the exterior derivative. This
natural operation takes a p-form and generates a new (p + 1)-form. We examine how the exterior
derivative recovers all the interesting vector-calculus derivatives from vector calculus on R3 . Of
course, it goes much deeper as the exterior derivative provides part of the machinery to write co-
homology on spaces of arbitrarily high dimension. Ultimately, this theory of cohomology detects
topological aspects of the space. A basic example of this is the Poincare Lemma.
To understand the Poincare Lemma as well as a number of other interesting calculations we find
it necessary to introduce a dual operation to the push-foward; the pull-back gives us a natural
method to take differential forms in the range of a mapping and transport them back to the domain
of the map. Moreover, this pull-back operation plays nicely with the wedge product and the exterior
derivative. Several pages are devotes towards understanding some intuitive method of calculating
the pull-back. We are indebted to Harold M. Edwards’ Advanced Calculus: A Differential Form
Approach which encouraged us to search for intuition. I also attempted to translate a differential
forms version of the implict mapping theorem from the same text, I’m fairly certain there is some
conceptual error in that section as it stands. It is a work in progress.
The abstract calculus of forms is interesting in it’s own right, but we are happy to find how it
reduces to the familar calculus on R3 . We state2 the Generalized Stokes Theorem and see how the
flux-form and work-form mappings produce the usual theorems of vector calculus as corollaries
to the Generalized Stokes Theorem. The definition of integrals of differential forms is accomplished
by pulling-back the forms to euclidean space where an ordinary integral quantifies the result. In
some sense, this discussion may help answer the question what is a differential form?. We spend
some effort attempting to understand how the form integration interfaces with ordinary surface or
line integration of vector fields.
Finally, the fact that d2 = 0 paired with the nice properties of the pull-back and wedge product
proves to give a technique for study of exact differential equations and partial differential equations.
This final section opens a door to the vast topic of exterior differential systems. In this study,
solutions to PDEs are manifolds and the PDE itself is formulated in terms of wedge products and
differential forms. Here I borrow wisdom from Cartan for Beginners by Ivey and Landsberg as well
as Equivalence, Invariance, and Symmetry by Olver. Please keep in mind, we’re just dipping are
toes in the pond here.
2
the proof is found in Edwards and many other places
10.1. AN INFORMAL INTRODUCTION TO MANIFOLDS 227
10.1 an informal introduction to manifolds

A manifold M is space which locally resembles Rn . This requirement is implemented by the
existence of coordinate charts (χ, U ) which cover M and allow us to do calculus locally. Let me
sketch the idea with a picture:
The charts (χj , Uj ) have to cover the manifold M and their transition functions θij must be
smooth mappings on Rn . Rather going on about the proper definition3 , I’ll show a few examples.
Example 10.1.1. Let M = {(x, y) ∈ R2 | x2 + y 2 = 1}. The usual chart on the unit-circle is the
angle-chart χ = θ. Given (cos t, sin t) = p ∈ M we define θ(p) = t. If x > 0 and x2 + y 2 = 1 then
(x, y) ∈ M and we have θ(x, y) = tan−1 (y/x).
Example 10.1.2. Let M = R2 . The usual chart is simply the cartesian coordinate system χ1 =
(x, y) with U1 = R2 . If (a, b) ∈ R2 then x(a, b) = a and y(a, b) = b. In practice the symbols x, y
are used both as maps and variables so one must pay attention to context. A second coordinate
system on M is given by the polar coordinate chart χ2 = (r, θ) with domain U2 = (0, ∞) × R. I’ll
just
√ take their −1
domain to be the right half-plane for the sake of having a nice formula: (r, θ)(a, b) =
2 2
( a + b , tan (b/a)). You can extend these to most of the plane, but you have to delete the origin
and you must lose a ray since the angle-chart is not injective if we go full-circle. That said, the
coordinate systems (χ1 = (x, y), U1 ) and (χ2 = (r, θ), U2 ) are compatible because they have smooth
transition functions. One can calculate χ1 ◦ χ−1 2
2 is a smooth mapping on R . Explicitly:
χ1 ◦ χ−1 −1
2 (u, v) = χ1 (χ2 (u, v)) = χ1 (u cos v, u sin v) = (u cos(v), u sin(v))
Technically, the use of the term coordinate system in calculus III is less strict than the concept
which appears in manifold theory. Departure from injectivity in a geometrically tractible setting is
manageable, but for the abstract setting, injectivity of the coordinate charts is important to many
arguments.
3
see my 2011 notes, or better yet study Loring Tu’s An Introduction to Manifolds
Example 10.1.3. Define χspherical (x, y, z) = (r, θ, φ) implicitly by the coordinate transformations
x = r cos(θ) sin(φ), y = r sin(θ) sin(φ), z = r cos(φ)
These can be inverted,

p y z
r= x2 + y 2 + z 2 , θ = tan−1 , φ = cos−1 p
x x2 + y 2 + z 2
To show compatibility with the standard Cartesian coordinates we would need to select a subset of R3
for which χspherical is 1-1 and the since χCartesian = Id the transition functions are just χ−1
spherical .
Example 10.1.4. Define χcylindrical (x, y, z) = (s, θ, z) implicitly by the coordinate transformations
x = s cos(θ), y = s sin(θ), z = z
These can be inverted,

−1 y
p
2 2
s = x + y , θ = tan , z=z
x
You can take dom(χcylindrical ) = {(x, y, z) |0 < θ < 2π, } − {(0, 0, 0)}
Example 10.1.5. Let M = V where V is an n-dimensional vector space over R. If β1 is a basis
for V then Φβ1 : V → Rn gives a global coordinate chart. Moreover, if β2 is another basis for V
then Φβ2 : V → Rn also gives a global coordinate chart. The transition function θ12 = Φβ2 ◦ Φ−1β1
is linear hence it is clearly smooth. In short, an n-dimensional vector space is an n-dimensional
manifold. We could put non-linear charts on V if we wished, that freedom is new here, in linear
algebra all the coordinate systems considered are linear.
Example 10.1.6. Let M = {(x, y) ∈ R2 | x2 + y 2 = 1}.
1. Let V+ = {(x, y) ∈ M | y > 0} = dom(χ+ ) and define χ+ (x, y) = x
2. Let V− = {(x, y) ∈ M | y < 0} = dom(χ− ) and define χ− (x, y) = x
3. Let VR = {(x, y) ∈ M | x > 0} = dom(χR ) and define χR (x, y) = y
4. Let VL = {(x, y) ∈ M | x < 0} = dom(χL ) and define χL (x, y) = y

The set of charts A = {(V+ , χ+ ), (V− , χ− ), (VR , χR ), (VL , χL )} forms an atlas on M which gives
the circle a differentiable structure4 . It is not hard to show the transition functions are smooth on
the image of the intersection of their respect domains.√For example, V+ ∩ VR = W+R = {(x, y) ∈
M | x, y > 0}, it’s easy to calculate that χ−1 2
+ (x) = (x, 1 − x ) hence
p p
(χR ◦ χ−1
+ )(x) = χR (x, 1 − x2 ) = 1 − x2
for each x ∈ χR (W+R ). Note x ∈ χR (W+R ) implies 0 < x < 1 hence it is clear the transition
function is smooth.
4
meaning that if we adjoin the infinity of likewise compatible charts that defines a differentiable structure on M
10.1. AN INFORMAL INTRODUCTION TO MANIFOLDS 229
Similar calculations hold for all the other overlapping charts. This manifold is usually denoted
M = S1 .
A cylinder is the Cartesian product of a line and a circle. In other words, we can create a cylinder
by gluing a copy of a circle at each point along a line. If all these copies line up and don’t twist
around then we get a cylinder. The example that follows here illustrates a more general pattern,
we can take a given manifold an paste a copy at each point along another manifold by using a
Cartesian product.
Example 10.1.7. Let P = {(x, y, z) ∈ R3 | x2 + y 2 = 1}.
1. Let V+ = {(x, y, z) ∈ P | y > 0} = dom(χ+ ) and define χ+ (x, y, z) = (x, z)
2. Let V− = {(x, y, z) ∈ P | y < 0} = dom(χ− ) and define χ− (x, y, z) = (x, z)
3. Let VR = {(x, y, z) ∈ P | x > 0} = dom(χR ) and define χR (x, y, z) = (y, z)
4. Let VL = {(x, y, z) ∈ P | x < 0} = dom(χL ) and define χL (x, y, z) = (y, z)
The set of charts A = {(V+ , χ+ ), (V− , χ− ), (VR , χR ), (VL , χL )} forms an atlas on P which gives the
cylinder a differentiable structure. It is not hard to show the transition functions are smooth on the
image of the intersection of their respective domains. For
−1
√ example, V+ ∩ VR = W+R = {(x, y, z) ∈
P | x, y > 0}, it’s easy to calculate that χ+ (x, z) = (x, 1 − x2 , z) hence
p p
(χR ◦ χ−1
+ )(x, z) = χR (x, 1 − x2 , z) = ( 1 − x2 , z)
for each (x, z) ∈ χR (W+R ). Note (x, z) ∈ χR (W+R ) implies 0 < x < 1 hence it is clear the
transition function is smooth. Similar calculations hold for all the other overlapping charts.
Generally, given two manifolds M and N we can construct M×N by taking the Cartesian product
of the charts. Suppose χM : V ⊆ M → U ⊆ Rm and χN : V 0 ⊆ N → U 0 ⊆ Rn then you can define
the product chart χ : V × V 0 → U × U 0 as χ = χM × χN . The Cartesian product M × N together
with all such product charts naturally is given the structure of an (m + n)-dimensional manifold.
For example, in the preceding example we took M = S1 and N = R to consruct P = S1 × R.
Example 10.1.8. The 2-torus, or donut, is constructed as T2 = S1 ×S1 . The n-torus is constructed
by taking the product of n-circles:
Tn = S1 × S1 × · · · × S1
| {z }
n copies
The atlas on this space can be obtained by simply taking the product of the S1 charts n-times.
Recall from our study of linear algebra that vector space structure is greatly elucidated by the
study of linear transformations. In our current context, the analogous objects are smooth maps.
These are natural mappings between manifolds. In particular, suppose M is an m-fold and N is
an n-fold with f : U ⊆ M → N is a function. Then, we say f is smooth iff all local coordinate
representatives of f are smooth mappings from Rm to Rn . See the diagram below:
This definition allows us to discuss smooth curves on manifolds. A curve γ : I ⊆ R → M is smooth

iff χ ◦ γ : I → Rn is a smooth curve in Rn .
Finally, just for your information, if a bijection between manifolds is smooth with smooth inverse
then the manifolds are said to be diffeomorphic. One fascinating result of recent mathematics is
10.2. VECTORS AS DERIVATIONS 231
that R4 permits distinct differentiable structures in the sense that there does not exist a diffeomor-
phism between certain atlases5 . Curiously, up to diffeomorphism, there is just one differentiable
structure on Rn for n 6= 4. Classifying possible differentiable structures for a given point set is an
interesting and ongoing problem.
10.2 vectors as derivations

To begin, let us define the set of locally smooth functions at p ∈ M:
C ∞ (p) = {f : M → R | f is smooth on an open set containing p}
In particular, we suppose f ∈ C ∞ (p) to mean there exists a patch φ : U → V ⊆ M such that f is

smooth on V . Since we use Cartesian coordinates on R by convention it follows that f : V → R
smooth indicates the local coordinate representative f ◦ φ : U → R is smooth (it has continuous
partial derivatives of all orders).
Definition 10.2.1.
Suppose Xp : C ∞ (p) → R is a linear transformation which satisfies the Leibniz rule then
we say Xp is a derivation on C ∞ (p). Moreover, we denote Xp ∈ Dp M iff Xp (f + cg) =
Xp (f ) + cXp (g) and Xp (f g) = f (p)Xp (g) + Xp (f )g(p) for all f, g ∈ C ∞ (p) and c ∈ R.
Example 10.2.2. Let M = R and consider Xto = d/dt|to . Clearly X is a derivation on smooth
functions near to .
∂ ∂

Example 10.2.3. Consider M = R2 . Pick p = (xo , yo ) and define Xp = ∂x p
and Y p = ∂y p
.
2
Once more it is clear that Xp , Yp ∈ D(p)R . These derivations action is accomplished by partial
differentiation followed by evaluation at p.
Example 10.2.4. Suppose M = Rm . Pick p ∈ Rm and define X = ∂x∂ j p . Clearly this is a

derivation for any j ∈ Nm .
Are the other types of derivations? Is the only thing a derivation is is a partial derivative operator?
Before we can explore this question we need to define partial differentiation on a manifold. We
should hope the definition is consistent with the langauge we already used in multivariate calculus
(and the preceding pair of examples) and yet is also general enough to be stated on any abstract
smooth manifold.
5
there is even a whole book devoted to this exotic chapter of mathematics, see the The Wild World of 4-Manifolds
by Alexandru Scorpan. This is on my list of books I ”need” to buy.
Definition 10.2.5.
Let M be a smooth m-dimensional manifold and let φ : U → V be a local parametrization
with p ∈ V . The j-th coordinate function xj : V → R is the j-component function of
φ−1 : V → U . In other words:
φ−1 (p) = x(p) = (x1 (p), x2 (p), . . . , xm (p))
These xj are manifold coordinates. In constrast, we will denote the standard Cartesian
coordinates in U ⊆ Rm via uj so a typical point has the form (u1 , u2 , . . . , um ) and viewed
as functions uj : Rm → R where uj (v) = ej (v) = v j . We define the partial derivative
with respect to xj at p for f ∈ C ∞ (p) as follows:

∂f ∂ ∂ −1

j
(p) = j
(f φ)(u)
◦ = j
f x
◦
.
∂x ∂u u=φ−1 (p) ∂u x(p)
The idea of the definition is simply to take the function f with domain in M then pull it back to
a function f ◦ x−1 : U ⊆ Rm → V → R on Rm . Then we can take partial derivatives of f ◦ x−1
in the same way we did in multivariate calculus. In particular, the partial derivative w.r.t. uj is
calculated by:
∂f d
(p) = f ◦ φ (x(p) + tej )
∂xj dt t=0
which is precisely the directional derivative of f ◦ x−1 in the j-direction at x(p). In fact, Note
f ◦ φ (x(p) + tej ) = f (x−1 (x(p) + tej )).

The curve t → x−1 (x(p) + tej ) is the curve on M through p where all coordinates are fixed except
the j-coordinate. It is a coordinate curve on M.
Notice in the case that M = Rm is given Cartesian coordinate φ = Id then x−1 = Id as well and
the t → x−1 (x(p) + tej ) reduces to t → p + tej which is just the j-th coordinate curve through p on
Rm . It follows that the partial derivative defined for manifolds naturally reduces to the ordinary
partial derivative in the context of M = Rm with Cartesian coordinates. The beautiful thing is
that almost everything we know for ordinary partial derivatives equally well transfers to ∂x∂ j p .
Theorem 10.2.6. Partial differentiation on manifolds
Let M be a smooth m-dimensional manifold with coordinates x1 , x2 , . . . , xm near p. Fur-

thermore, suppose coordinates y 1 , y 2 , . . . , y m are also defined near p. Suppose f, g ∈ C ∞ (p)
and c ∈ R then:
∂f ∂g
1. ∂x∂ j p f + g = ∂x

j p + ∂xj p
∂ ∂f

2. ∂xj p
cf = c ∂xj p
∂ ∂g ∂f

3. ∂xj p
f g = f (p) ∂x j p + ∂xj p
g(p)
∂xi

4. ∂xj p
= δij
Pm ∂xk ∂y i

5. k=1 ∂y j p ∂xk p = δij
∂f Pm ∂xk ∂f

6. ∂y j p
= k=1 ∂y j p ∂xk p
Proof: The proof of (1.) and (2.) follows from the calculation below:

∂(f + cg) ∂ −1

(p) = (f + cg) ◦ x
∂xj ∂uj
x(p)

∂ −1 −1

= f ◦ x + cg ◦ x
∂uj
x(p)

∂ −1
∂ −1

= j
f x
◦ +c j g x ◦
∂u
x(p) ∂u
x(p)
∂f ∂g
= j
(p) + c j (p) (10.1)
∂x ∂x
The key in this argument is that composition (f + cg) ◦ x−1 = f ◦ x−1 + cg ◦ x−1 along side the
linearity of the partial derivative. Item (3.) follows from the identity (f g) ◦ x−1 = (f ◦ x−1 )(g ◦ x−1 )
in tandem with the product rule for a partial derivative on Rm . The reader may be asked to complete
the argument for (3.) in the homework. Continuing to (4.) we calculate from the definition:
∂xi ∂ui

∂ i ◦ −1

= (x x )(u) = = δij .
∂xj p ∂uj x(p) ∂uj x(p)
where the last equality is known from multivariate calculus. In invite the reader to prove it from
the definition if unaware of this fact. Before we prove (5.) it helps to have a picture and a bit
more notation in mind. Near the point p we have two coordinate charts x : V → U ⊆ Rm and
y : V → W ⊆ Rm , we take the chart domain V to be small enough so that both charts are
defined. Denote Cartesian coordinates on U by u1 , u2 , . . . , um and for W we likewise use Cartesian
coordinates w1 , w2 , . . . , wm . Let us denote patches φ, ψ as the inverses of these charts; φ−1 = x
and ψ −1 = y. Transition functions ψ −1 ◦ φ = y ◦ x−1 are mappings from U ⊆ Rm to W ⊆ Rm and
we note
∂y i

∂ i ◦ −1
(y x )(u) =
∂uj ∂xj
Likewise, the inverse transition functions φ−1 ◦ ψ = x ◦ y −1 are mappings from W ⊆ Rm to U ⊆ Rm
∂xi

∂ i ◦ −1
(x y )(w) =
∂wj ∂y j
Recall that if F, G : Rm → Rm and F ◦ G = Id then F 0 G0 = I by the chainrule, hence (F 0 )−1 = G0 .

Apply this general fact to the transition functions, we find their derivative matrices are inverses.
∂y
Item (5.) follows. In matrix notation we item (5.) reads ∂x ∂y ∂x = I. Item (6.) follows from:

∂f ∂ −1

j
= j
(f ◦ y )(w)
∂y p ∂w

y(p)

∂ −1 −1

= j
(f ◦ x ◦ x ◦ y )(w)
∂w y(p)

∂
◦ x−1 (u1 (w), . . . , um (w)) : where uk (w) = (x ◦ y −1 )k (w)

= f
∂wj
y(p)
m
X ∂(x ◦ y ) −1 k
−1

∂(f ◦ x )
= j
: chain rule
∂w
y(p) ∂uk (x ◦ y−1 )(y(p))
k=1
m
∂(xk ◦ y −1 ) ∂(f ◦ x−1 )
X
=
∂wj
y(p) ∂uk x(p)
k=1
m
X ∂xk ∂f
=
∂y j p ∂xk p
k=1
The key step was the multivariate chain rule.
This theorem proves we can lift calculus on Rm to M in a natural manner. Moreover, we should
∂
note that items (1.), (2.) and (3.) together show ∂x i p is a derivation at p. Item (6.) should remind
the reader of the contravariant vector discussion. Removing the f from the equation reveals that
m
∂xk ∂

∂ X
=
∂y j p ∂y j p ∂xk p
k=1
A notation convenient to the current discussion is that a contravariant transformation is (p, vx ) →

∂y
(p, vy ) where vy = P vx and P = (y ◦ x−1 )0 (x(p)) = ∂x x(p)
. Notice this is the inverse of what we see
in (6.). This suggests that the partial derivatives change coordinates like as a basis for the tangent
space. To complete this thought we need a few well-known propositions for derivations.
Proposition 10.2.7. derivations on constant function gives zero.
If f ∈ C ∞ (p) is a constant function and Xp ∈ Dp M then Xp (f ) = 0.
Proof: Suppose f (x) = c for all x ∈ V , define g(x) = 1 for all x ∈ V and note f = f g on V . Since
Xp is a derivation is satisfies the Leibniz rule hence
Xp (f ) = Xp (f g) = f (p)Xp (g) + X(f )g(p) = cXp (g) + Xp (f ) ⇒ cXp (g) = 0.
Moreover, by homogeneity of Xp , note cXp (g) = Xp (cg) = Xp (f ). Thus, Xp (f ) = 0.
Proposition 10.2.8.
If f, g ∈ C ∞ (p) and f (x) = g(x) for all x ∈ V and Xp ∈ Dp M then Xp (f ) = Xp (g).
Proof: Note that f (x) = g(x) implies h(x) = f (x) − g(x) = 0 for all x ∈ V . Thus, the previous
proposition yields Xp (h) = 0. Thus, Xp (f − g) = 0 and by linearity Xp (f ) − Xp (g) = 0. The
proposition follows.
Proposition 10.2.9.
Suppose Xp ∈ Dp M and x is a chart defined near p,

m
X
j ∂
Xp = Xp (x ) j
∂x p
j=1
Proof: this is a less trivial proposition. We need a standard lemma before we begin.
Lemma 10.2.10.
Let p be a point in smooth manifold M and let f : M → R be a smooth function. If
x : V → U is a chart with p ∈ V and x(p) = 0 then there exist smooth functions gj : M → R
∂f
whose values atPpm satisfy gj (p) = ∂xj (p). In addition, for all q near enough to p we have
j
f (q) = f (p) + k=1 x (q)gj (q)
Proof: follows from proving a similar identity on Rm then lifting to the manifold. I leave this as a
nontrivial exercise for the reader. This can be found in many texts, see Burns and Gidea page 29
for one source. It should be noted that the manifold must be smooth for this construction to hold.
It turns out the set of derivations on a C k -manifold forms an infinite-dimensional vector space over
R, see Lawrence Conlon’s Differentiable Manifolds page 49. O
∂f
Consider f ∈ C ∞ (p), and use the lemma, we assume x(p) = 0 and gj (p) = ∂xj
(p):
m
X
j
Xp (f ) = Xp f (p) + x (q)gj (q)
k=1
Xm
Xp xj (q)gj (q)

= Xp (f (p)) +
k=1
m
X
j j
= Xp (x )gj (q) + x (p)Xp (gj (q))
k=1
m
X ∂f
= Xp (xj ) (p).
∂xj
k=1
The calculation above holds for arbitrary f ∈ C ∞ (p) hence the proposition follows.
We’ve answered the question posed earlier in this section. It is true that every derivation of a
manifold is simply a linear combination of partial derivatives. We can say more. The set of deriva-
tions at p naturally forms a vector space under the usual addition and scalar multiplication of
operators: if Xp , Yp ∈ Dp M then we define Xp + Yp by (Xp + Yp )(f ) = Xp (f ) + Yp (f ) and cXp by
(cXp )(f ) = cXp (f ) for all f, g ∈ C ∞ (p) and c ∈ R. It is easy to show Dp M is a vectorspace under
∂f m
these operations. Moreover, the preceding proposition shows that Dp M = span{ ∂x j p }j=1 hence
Dp M is an m-dimensional vector space . 6
Finally, let’s examine coordinate change for derivations. Given two coordinate charts x, y at p ∈ M
we have two ways to write the derivation Xp :
m m
j ∂ k ∂
X X
Xp = Xp (x ) j or Xp = Xp (y ) k
∂x p ∂y p
j=1 k=1
It is simple to connect these formulas. Whereas, for y-coordinates,

m k

j ∂y
X
k
Xp (y ) = Xp (x ) j (10.2)
∂x p
j=1
∂xk ∂
This is the contravariant transformation rule. In contrast, recall ∂y∂ j p = m
P
k=1 ∂y p ∂x p . We
j k
should have anticipated this pattern since from the outset it is clear there is no coordinate depen-
dence in the definition of a derivation.
Definition 10.2.11. tangent space
We denote Tp M = derTp M.
6

∂
technically, we should show the coordinate derivations ∂xj p are linearly independent to make this conclusion. I
don’t suppose we’ve done that directly at this juncture. You might find this as a homework
10.3. DIFFERENTIAL FOR MANIFOLDS, THE PUSH-FORWARD 237
10.2.1 concerning the geometry of derivations

An obvious question we should ask:
How are derivations related to tangent vectors geometrically?
To give a proper answer, we focus our attention to R3 and I follow Barret Oneil’s phenomenal text
on Elementary Differential Geometry, second edition. Consider a function f : R3 → R and calculate
the directional derivative in the v = ha, b, ci-direction at the point p. We make no assumption that
||v|| = 1, this is the same directional derivative for which we discussed the relation with the Frechet
derivative in an earlier chapter. If f is smooth,

∂f

∂f ∂f ∂ ∂ ∂
Df (p)(v) = v (∇f )(p) = a (p) + b (p) +
• (p) = a +b +c f
∂x ∂y ∂z ∂x ∂y ∂z p
Therefore, we find the following interpretation of the derivation Xp = (a∂x + b∂y + c∂z ) |p :
When a derivation Xp = (a∂x + b∂y + c∂z ) |p acts on a smooth function f it describes
the rate at which f changes at p in the ha, b, ci direction. In other words, a derivation
at p generates directional derivatives of functions at p.
Therefore, our work over the last few pages can be interpreted as abstracting the directional deriva-
tive to manifolds. In particular, the partial derivatives with respect to manifold coordinate xj
measure the rate of change of functions along the curve on the manifold which allows xj to vary
while all the other coordinates are held fixed.
10.3 differential for manifolds, the push-forward

In this section we generalize the concept of the differential to the context of manifolds. Recall that
for F : U ⊆ Rm → V ⊆ Rn the differential dp F : Rm → Rn was a linear transformation which best
approximated the change in F near p. Notice that while the domain of F could be a mere subset
of Rm the differential always took all of Rm as its domain. This suggests we should really think of
the differential as a mapping which transports tangent vectors to U to tangent vectors at V . I find
the following picture helpful at times, here I picture the tangent space as if it is attached to the
point p on the manifold. Keep in mind this is an abstract picture:
Often dp f is called the push-forward by f at p because it pushes tangent vectors in the same
direction as the mapping transports points.
Definition 10.3.1. differential for manifolds.
Suppose M and N are smooth manifolds of dimension m and n respective. Furthermore,

suppose f : M → N is a smooth mapping. We define dp f : Tp M → Tf (p) N as follows: for
each Xp ∈ Tp M and g ∈ C ∞ (f (p))
dp f (Xp )(g) = Xp (g ◦ f ).
Notice that g : dom(g) ⊆ N → R and consequently g ◦ f : M → N → R and it follows g ◦ f ∈

C ∞ (p) and it is natural to find g ◦ f in the domain of Xp . In addition, it is not hard to show
dp f (Xp ) ∈ Df (p) N . Observe:
1. dp f (Xp )(g + h) = Xp ((g + h) ◦ f ) = Xp (g ◦ f + h ◦ f ) = Xp (g ◦ f ) + Xp (h ◦ f )
2. dp f (Xp )(cg) = Xp ((cg) ◦ f ) = Xp (cg ◦ f )) = cXp (g ◦ f )) = cdp f (Xp )(g)
The proof of the Leibniz rule is similar. In this section we generalize the concept of the differential
to the context of manifolds. Recall that for F : U ⊆ Rm → V ⊆ Rn the differential dp F : Rm → Rn
was a linear transformation which best approximated the change in F near p. Notice that while the
domain of F could be a mere subset of Rm the differential always took all of Rm as its domain. This
suggests we should really think of the differential as a mapping which transports tangent vectors
to U to tangent vectors at V . Therefore, dp f is called the push-forward.
Definition 10.3.2. differential for manifolds.
Suppose M and N are smooth manifolds of dimension m and n respective. Furthermore,

suppose f : M → N is a smooth mapping. We define dp f : Tp M → Tf (p) N as follows: for
each Xp ∈ Tp M and g ∈ C ∞ (f (p))
dp f (Xp )(g) = Xp (g ◦ f ).
Notice that g : dom(g) ⊆ N → R and consequently g ◦ f : M → N → R and it follows g ◦ f ∈

C ∞ (p) and it is natural to find g ◦ f in the domain of Xp . In addition, it is not hard to show
dp f (Xp ) ∈ Df (p) N . Observe:
1. dp f (Xp )(g + h) = Xp ((g + h) ◦ f ) = Xp (g ◦ f + h ◦ f ) = Xp (g ◦ f ) + Xp (h ◦ f )
2. dp f (Xp )(cg) = Xp ((cg) ◦ f ) = Xp (cg ◦ f )) = cXp (g ◦ f )) = cdp f (Xp )(g)
The proof of the Leibniz rule is similar.

10.3.1 intuition for the push-forward

Perhaps it will help to see how the push-forward appears in calculus III. I’ll omit the point-
dependence to reduce some clutter. Consider f : R2rθ → R2xy defined by f (r, θ) = (r cos θ, r sin θ).
Consider, for g a smooth function on R2xy , using the chain-rule,

∂ ∂ ∂g ∂(x ◦ f ) ∂g ∂(y ◦ f )
df (g) = (g ◦ f ) = +
∂r ∂r ∂x ∂r ∂y ∂r
Note x ◦ f = r cos θ and y ◦ f = r sin θ. Hence:

∂ ∂g ∂g
df (g) = cos θ + sin θ
∂r ∂x ∂y
This holds for all g hence we derive,

∂ x ∂ y ∂
df =p +p
∂r x2 + y 2 ∂x x2 + y 2 ∂y
A similar calculation shows

∂ ∂ ∂
df = −y +x
∂θ ∂x ∂y
We could go through the reverse calculation for f −1 and derive that:

−1 ∂ ∂ sin θ ∂ −1 ∂ ∂ ∂
df = cos θ − and df = sin θ + cos θ
∂x ∂r r ∂θ ∂y ∂r ∂θ
∂2u ∂2u ∂2u 1 ∂u 1 ∂2u

These formulas go to show that ∂x2
+ ∂y 2
= 0 transforms to ∂r2
+ r ∂r + r2 ∂θ2
= 0.
10.3.2 a pragmatic formula for the push-forward

In practice I quote the following result as the definition. However, Definition 10.3.2 is prefered by
many for its coordinate independence. As we’ve seen in the intuition example, coordinates tend to
arise in actual calculations.
Proposition 10.3.3.
If F : M → N is a differentiable mapping of manifolds and (xi ) are coordinates(i = 1, . . . , m)

on M and (y j ) are coordinates (j = 1, . . . , n) on N which contain p ∈ M and F (p) ∈ N
respective then:
m ! m X n j

∂ i ∂(y ◦ F ) ∂
X X
i

(dF )p X = X .
∂xi p ∂xi ∂y j F (p)
i=1 i=1 j=1
Proof: Notice the absence of the f in this formula as compared to Defnition 10.3.2 dF (X)(f ) =
X(f ◦ F ). To show equivalence, we can expand on our definition, we assume here that X =
Pm i ∂

i=1 X ∂xi thus:
p
m
X
i∂
X(f ◦ F ) = X (f ◦ F )
∂xi p
i=1
note f is a function on N hence the chain rule gives:
m n
X X ∂f ∂(y j ◦ F )
X(f ◦ F ) = Xi (F (p))
∂y j ∂xi
i=1 j=1
But, we can write this as:

 
m n j

X X ∂(y ◦ F ) ∂
Xi

X(f ◦ F ) =   (f ).
∂xi ∂y j F (p)
i=1 j=1
Therefore, as f is arbitrary, we have shown the claim of the proposition.
Observe that the difference between Definition 10.3.2 and Proposition 10.3.3 is merely an applica-
tion of the chain-rule.
The push-forward is more than just coordinate change. If we consider a mapping between spaces
of disparate dimension then the push-forward captures something about the mapping in question
and the domain and range spaces. For example, the existence of a nontrivial vector field on the
whole of a manifold implies the existence of a foliation of the manifold.
If the mapping is a diffeomorphism then we expect it will carry the nontrivial vector field to the
range space. However, if the mapping is not injective then there is no assurance a vector field
even maps to a vector field. We could attach two vectors to a point in the range for a two-to-one
map.For example, this mapping wraps around the circle and when it hits the circle the second time
the vector pushed-forward does not match what was pushed forward the first time. It follows that
push-forward of the vector field does not form a vector field in this case:
The pull-back (introducted in Section 10.7) is also an important tool to compare geometry of
different spaces. We’ll see in Section 10.10 how the pull-back even allows us to write a general
formula for calculating the potential energy function of a flux-form of arbitrary degree. This
captures the electric and magnetic potentials of electromagnetism and much more we have yet to
discover experimentally. That said, the pull-back is formulated in terms of the push-forward we
consider here thus the importance of the push-forward is hard to overstate.
Example 10.3.4. Suppose F : R2 → R2×2 is defined by

x cos y − sin y
F (x, y) = e
sin y cos y
Let R2 have the usual (x, y)-coordinate chart and let Z ij defined by
Z ij (A) = Aij
for R2×2 form the global coordinate chart for 2 × 2 matrices. Let us calculate the push-forward of
the coordinate vector ∂x |p :
2 2
∂(Z ij ◦ F ) ∂ ∂(Z ij ◦ F ) ∂

X X
dFp (∂x |p ) = & dFp (∂y |p ) =
∂x ∂Z ij F (p) ∂y ∂Z ij F (p)
i,j=1 i,j=1
Observe that:
Z 11 (F (x, y)) = ex cos y = Z 22 (F (x, y)), Z 21 (F (x, y)) = ex sin y = −Z 12 (F (x, y)).
From which we derive,
dFp (∂x ) = ex cos y(∂11 + ∂22 ) + ex sin y(∂21 − ∂12 )
dFp (∂y ) = −ex sin y(∂11 + ∂22 ) + ex cos y(∂21 − ∂12 )
Here ∂x , ∂y are at p ∈ R2 whereas ∂ij is at F (p) ∈ R2×2 . However, these are constant vector fields
so the point-dependence is not too interesting.
10.4 cotangent space

The tangent space to a smooth manifold M is a vector space of derivations and we denote it by
Tp M. The dual space to this vector space is called the cotangent space and the typical elements
are called covectors.
Definition 10.4.1. cotangent space Tp M∗
Suppose M is a smooth manifold and Tp M is the tangent space at p ∈ M. We define,
Tp M∗ = {αp : Tp M → R | α is linear}.
If x is a local coordinate chart at p and ∂x∂ 1 p , ∂x∂ 2 p , . . . , ∂x∂m p is a basis for Tp M then we denote

the dual basis dp x1 , dp x2 , . . . , dp xm where dp xi ∂k = δik . Moreover, if α is a covector at p then7 :
p
m
X
α= αk dxk
k=1

∂

where αk = α ∂xk and dxk is a short-hand for dp xk . We should understand that covectors
p
are defined at a point even if the point is not explicitly indicated in a particular context. This
does lead to some ambiguity in the same way that the careless identification of the function f and
it’s value f (x) does throughout calculus. That said, an abbreviated notation is often important to
help us see through more difficult patterns without getting distracted by the minutia of the problem.
You might worry the notation used for the differential and our current notation for the dual basis
of covectors is not consistent. After all, we have two rather different meanings for dp xk at this time:
1. xk : V → R is a smooth function hence dp xk : Tp M → Txk (p) R
is defined as a push-forward, dp xk (Xp )(g) = Xp (g ◦ xk )

2. dp xk : Tp M → R where dp xk ∂j p = δjk
It is customary to identify Txk (p) R with R hence there is no trouble. Let us examine how the
dual-basis condition can be derived for the differential, suppose g : R → R hence g ◦ xk : V → R,
∂xk dg

k ∂ ∂ k d
dp x j
(g) = j
(g(x )) = j
= δjk g = δjk g
∂x p ∂x p ∂x p dt xk (p)
dt xk (p)
| {z }
chain rule

d

Where, we’ve made the identification 1 = dt (which is the nut and bolts of Txk (p) R = R ) and
xk (p)
hence have the beautiful identity:

k ∂
dp x = δjk .
∂xj p
7
we explained this for an arbitrary vector space V and its dual V ∗ in a previous chapter, we simply apply those
results once more here in the particular context V = Tp M
10.5. DIFFERENTIAL FORMS 243
In contrast, there is no need to derive this for case (2.) since in that context this serves as the
definition for the object. Personally, I find the multiple interpretations of objects in manifold theory
is one of the most difficult aspects of the theory. On the other hand, the notation is really neat
once you understand how subtly it assumes many theorems. You should understand the notation
we enjoy at this time is the result of generations of mathematical thought. Following a similar
derivation for an arbitrary vector Xp ∈ Tp M and f : M → R we find
dp f (Xp ) = Xp (f )
This notation is completely consistent with the total differential as commonly discussed in multi-
variate calculus. Recall that if f : Rm → R then we defined
∂f ∂f ∂f
df = 1
dx1 + 2 dx2 + · · · + m dxm .
∂x ∂x ∂x
∂f
Notice that the j-th component of df is simply ∂x j . Notice that the identity dp f (Xp ) = Xp (f )
gives us the same component if we simply evaluate the covector dp f on the coordinate basis ∂x∂ j p ,

∂ ∂f
dp f =
∂xj p ∂xj p
10.5 differential forms

In this section we apply the results of Chapter 9 on exterior algebra to the vector space V = Tp M
and its dual space V ∗ = Tp M ∗ . In short, this involves making the identifications:

∂
ei = and ej = dp xj
∂xi p
however, we often omit the p-dependence of dp xj and just write dxj . Observe that:
dxj (∂i |p ) = ∂i (xj ) = δij
Therefore, the coordinate basis {∂1 |p , . . . , ∂m |p } for Tp M is indeed dual to the differentials of the
coordinates {dx1 , . . . , dxm } basis for the cotangent space Tp M ∗ .
In contrast to Chapter 9 we have a point-dependence to our basis. We establish the following

terminology: for a manifold M and p ∈ M ,
(0.) p 7→ R is a 0-form, or function on M
(1.) p 7→ dxj is a 1-form on M
(2.) p 7→ dxi ∧ dxj is a 2-form on M

(3.) p 7→ dxi ∧ dxj ∧ dxk is a 3-form on M
(4.) p 7→ dxi1 ∧ · · · ∧ dxik is a k-form on M
Generally a k-form is formed from taking sums of the basic differential forms given above with
coefficients which are smooth functions. 8
Pm j
(1.) a one-form α = j=1 αj dx has smooth coefficient functions αj .
Pm i
(2.) a two-form β = i,j=1 βij dx ∧ dxj has smooth coefficient functions βij .
Pn
(3.) a k-form γ = i1 ,...,ik γi1 ,...,ik dxi1 ∧ · · · ∧ dxik has smooth coefficient functions γi1 ,...,ik .
The algebra of differential forms follows the same rules as the exterior algebra we previously dis-
cussed. However, instead of having scalars as numbers we now consider scalars as functions. This
comment is made explicit in the theorem to follow:
Theorem 10.5.1.
If α is a p-form, β is a k-form, and γ is a l-form on M then
1. α ∧ (β ∧ γ) = (α ∧ β) ∧ γ
2. α ∧ β = (−1)pk (β ∧ α)
3. α ∧ (aβ + bγ) = a(α ∧ β) + b(α ∧ γ) a, b ∈ R
Notice that in R3 the set of differential forms
B = {1, dx, dy, dz, dy ∧ dz, dz ∧ dx, dx ∧ dy, dx ∧ dy ∧ dz}
is a basis of the space of differential forms in the sense that every form on R3 is a linear combination
of the forms in B with smooth real-valued functions on R3 as coefficients.
Example 10.5.2. Let α = f dx + gdy and let β = 3dx + dz where f, g are functions. Find α ∧ β,
write the answer in terms of the basis defined in the Remark above,
α ∧ β = (f dx + gdy) ∧ (3dx + dz)

= f dx ∧ (3dx + dz) + gdy ∧ (3dx + dz)
(10.3)
= 3f dx ∧ dx + f dx ∧ dz + 3gdy ∧ dx + gdy ∧ dz
= −gdy ∧ dz − f dz ∧ dx − 3gdx ∧ dy
8
Technically, this means the exterior algebra of differential forms is a module over the ring of smooth functions.
However, the exterior algebra at a point is a vector space.
10.5. DIFFERENTIAL FORMS 245
Example 10.5.3. Top form: Let α = dx ∧ dy ∧ dz and let β be any other form with degree p > 0.
We argue that α ∧ β = 0. Notice that if p > 0 then there must be at least one differential inside β
so if that differential is dxk we can rewrite β = dxk ∧ γ for some γ. Then consider,
α ∧ β = dx ∧ dy ∧ dz ∧ dxk ∧ γ (10.4)
now k has to be either 1, 2 or 3 therefore we will have dxk repeated, thus the wedge product will be
zero. (can you prove this?).
The proposition below and its proof are included here to remind the reader on the structure of the
⊗ and ∧ products and components. One distinction, the components are functions now whereas
they were scalars in the previous chapter.
Proposition 10.5.4.
If ω is a p-form in an n-dimensional space is written in the coordinate coframe dx1 , . . . , dxn

at p then the components of ω are given by evaluation on the coordinate frame ∂1 , . . . , ∂n .
at p.
Proof: Suppose ω has component functions ωi1 i2 ...ip with respect to the tensor basis dxi1 ⊗· · ·⊗dxip
for type (0, p) tensors.
Xn
ω= ωi1 i2 ...ip dxi1 ⊗ dxi2 ⊗ · · · ⊗ dxip
i1 i2 ...ip
the functions ωi1 i2 ...ip are called the tensor components of ω. Consider evaluation of ω on a
p-tuple of coordinate vector fields,
n
X
ω(∂j1 , ∂j2 . . . ∂jp ) = ωi1 i2 ...ip dxi1 ⊗ dxi2 ⊗ · · · ⊗ dxip (∂j1 , ∂j2 , . . . , ∂jp )
i1 ,i2 ,...,ip =1
X n
= ωi1 i2 ...ip δi1 j1 δi2 j2 · · · δip jp
i1 ,i2 ,...,ip =1
= ωj1 j2 ...jp
If ω is a p-form that indicated ω is a completely antisymmetric tensor and by a calculation similar to

those near Equation 9.10 we can express ω by a sum over all p-forms (this is not a basis expansion)
n
X 1
ω= ωi i ...i dxi1 ∧ dxi2 ∧ · · · ∧ dxip .
p! 1 2 p
i1 ,i2 ,...,ip =1
But, we just found ω(∂j1 , ∂j2 . . . ∂jp ) = ωi1 i2 ...ip hence:

n
X 1
ω= ω(∂i1 , ∂i2 . . . ∂ip )dxi1 ∧ dxi2 ∧ · · · ∧ dxip .
p!
i1 i2 ...ip
which proves the assertion of the proposition.
Note, if we work with an expansion of linearly independent p-vectors then we can write the con-
clusion of the proposition:
n
X
ω= ω(∂i1 , ∂i2 . . . ∂ip )dxi1 ∧ dxi2 ∧ · · · ∧ dxip .
i1 <i2 <···<ip
10.6 the exterior derivative

The operation ∧ depends only on the values of the forms point by point. We define an operator d
on differential forms which depends not only on the value of the differential form at a point but on
its value in an entire neighborhood of the point. Thus if β ia k-form then to define dβ at a point
p we need to know not only the value of β at p but we also need to know its value at every q in a
neighborhood of p.
You might note the derivative below does not directly involve the construction of differential forms
from tensors. Also, the rule given below is easily taken as a starting point for formal calculations.
In other words, even if you don’t understand the nuts and bolts of manifold theory you can still
calculate with differential forms. In the same sense that highschool students ”do” calculus, you can
”do” differential form calculations. I don’t believe this is a futile exercise so long as you understand
you have more to learn. Which is not to say we don’t know some things!
Definition 10.6.1. the exterior derivative.

1
If β is a k-form and x = (x1 , x2 , · · · , xn ) is a chart and β = I
P
I k! βI dx and we define a
(k + 1)-form dβ to be the form
X 1
dβ = dβI ∧ dxI .
k!
I
Where dβI is defined as it was in calculus III,

n
X ∂βI
dβI = dxj .
∂xj
j=1
Note that dβI is well-defined as

βI = βi1 i2 ···ik
is just a real-valued function on dom(x). The definition in an expanded form is given by
n n n
1 X X X
dp β = ··· (dp βi1 i2 ···ik ) ∧ dp xi1 ∧ · · · ∧ dp xik
k!
i1 =1 i2 =1 ik =1
10.6. THE EXTERIOR DERIVATIVE 247
where
n n n
1 X X X
βq = ··· βi1 i2 ···ik (q)dq xi1 ∧ · · · ∧ dp xik .
k!
i1 =1 i2 =1 ik =1
Consequently we see that for each k the operator d maps ∧k (M ) into ∧k+1 (M ). Also:
Theorem 10.6.2. properties of the exterior derivative.
If α ∈ ∧k (M ), β ∈ ∧l (M ) and a, b ∈ R then
1. d(aα + bβ) = a(dα) + b(dβ)
2. d(α ∧ β) = (dα ∧ β) + (−1)k (α ∧ dβ)
3. d(dα) = 0
Remark 10.6.3.
Warning: I use Einstein’s repeated index notation in the proof that follows. In fact, it’s a
bit worse, I use I to denote a multi-index. This means a repeated I indicates an implicit
sum over all increasing strings of indices of a particular length. This is just a brief notation
to sum over the basis of coordinate p-forms. Indeed, from this point on in the notes there
is occasional use of Einstein’s convention.
Proof: The proof of (1) is obvious. To prove (2), let x = (x1 , · · · , xn ) be a chart on M then
suppose α = αI dxI and β = βJ dxJ
d(α ∧ β) = d(αI βJ ) ∧ dxI ∧ dxJ = (αI dβJ + βJ dαI ) ∧ dxI ∧ dxJ

= αI (dβJ ∧ dxI ∧ dxJ )
+βJ (dαI ∧ dxI ∧ dxJ )
= αI (dxI ∧ (−1)k (dβJ ∧ dxJ ))
+βJ ((dαI ∧ dxI ) ∧ dxJ )
= (α ∧ (−1)k dβ) + βJ (dα ∧ dxJ )
= dα ∧ β + (−1)k (α ∧ dβ) .
To prove (3.) we could resort to a beautiful tensor calculation (see Equation 10.11) or:
dα = dαI ∧ dxI
hence
d(dα) = d(dαI ∧ dxI ) = d(dαI ) ∧ dxI + αI ∧ d(dxI ).
Notice d(dxI ) = d(dxi1 ∧ · · · ∧ dxik ) = d(1) ∧ dxi1 ∧ · · · ∧ dxik = 0. Therefore, we have reduced the
problem to showing d(dαI ) = 0 for a function αI . I leave that problem to the reader. .
10.6.1 exterior derivatives on R3

We begin by noting that vector fields may correspond either to a one-form or to a two-form.
Definition 10.6.4. dictionary of vectors verses forms on R3 .
~ = (A1 , A2 , A3 ) denote a vector field in R3 . Define then,
Let A
ωA = δij Ai dxj = Ai dxi
~ Also define
which we will call the work-form of A.
1 1
ΦA = δik Ak ijk (dxi ∧ dxj ) = Ai ijk (dxi ∧ dxj )
2 2
~
which we will call the flux-form of A.
If you accept the primacy of differential forms, then you can see that vector calculus confuses two
separate objects. Apparently there are two types of vector fields. In fact, if you have studied coor-
dinate change for vector fields deeply then you will encounter the qualifiers axial or polar vector
fields. Those fields which are axial correspond directly to two-forms whereas those correspondant
to one-forms are called polar. Note, the magnetic field is axial whereas the electric field is polar.
Example 10.6.5. Gradient: Consider three-dimensional Euclidean space. Let f : R3 → R then
∂f i
df = dx = ω∇f
∂xi
which gives the one-form corresponding to ∇f .
Example 10.6.6. Curl: Consider three-dimensional Euclidean space. Let F~ be a vector field and
let ωF = Fi dxi be the corresponding one-form then
dωF = dFi ∧ dxi
= ∂j Fi dxj ∧ dxi
= ∂x Fy dx ∧ dy + ∂y Fx dy ∧ dx + ∂z Fx dz ∧ dx + ∂x Fz dx ∧ dz + ∂y Fz dy ∧ dz + ∂z Fy dz ∧ dy
= (∂x Fy − ∂y Fx )dx ∧ dy + (∂z Fx − ∂x Fz )dz ∧ dx + (∂y Fz − ∂z Fy )dy ∧ dz
= Φ∇×F~ .
Thus we recover the curl.
~ be a vector
Example 10.6.7. Divergence: Consider three-dimensional Euclidean space. Let G
1 j k
field and let ΦG = 2 ijk Gi dx ∧ dx be the corresponding two-form then
dΦG = d( 21 ijk Gi ) ∧ dxj ∧ dxk
= 12 ijk (∂m Gi )dxm ∧ dxj ∧ dxk
= 12 ijk (∂m Gi )mjk dx ∧ dy ∧ dz
= 12 2δim (∂m Gi )dx ∧ dy ∧ dz
= ∂i Gi dx ∧ dy ∧ dz
~
= (∇ · G)dx ∧ dy ∧ dz
and in this way we recover the divergence.
10.6. THE EXTERIOR DERIVATIVE 249
10.6.2 coordinate independence of exterior derivative

The Einstein summation convention is used in this section and throughout the remainder
of this chapter, please feel free to email me if it confuses you somewhere. When an index is
repeated in a single summand it is implicitly assumed there is a sum over all values of that
index
It must be shown that this definition is independent of the chart used to define dβ. Suppose for
example, that
βq = β J (q)(dq xj1 ∧ · · · ∧ dq xjk )
for all q in the domain of a chart (x1 , x2 , · · · xn ) where
dom(x) ∩ dom(x), 6= ∅ .
We assume, of course that the coefficients {β J (q)} are skew-symmetric in J for all q. We will
have defined dβ in this chart by
dβ = dβ J ∧ dxJ .
We need to show that dp β J ∧ dp xJ = dp βI ∧ dp xI for all p ∈ dom(x) ∩ dom(x) if this definition is
to be meaningful. Since β is given to be a well-defined form we know
βI (p)dp xI = βp = β J (p)dp xJ .
Using the identities
∂xj i
dxj = dx
∂xi
we have
∂xj1 ∂xjk ∂xjk
βI dxI = β J · · · dxI
∂xi1 ∂xik ∂xik
so that j1
∂x ∂xj2 ∂xjk

βI = β J ··· i .
∂xi1 ∂xi2 ∂x k
Consequently,

∂βJ ∂ i ∂xik
dβJ ∧ dxJ = ∂xλ
(dxλ ∧ dxJ ) = [β ∂x 1 · · · ∂x
∂xλ I ∂xj1
λ
jk ](dx ∧ dx )
J

∗ ∂β I ∂xi1 ∂xik
= ∂x λ ∂xj
1
· · · ∂x jk (dxλ ∧ dxJ )

P ∂xi1 2 ir ∂xik
+β I r ∂xj1 · · · ∂x∂λ x∂xjr · · · ∂xj (dxλ ∧ dxJ )
i ki
∂β I λ ∧ ∂x 1 dxj1 ∧ · · · ∧ ∂x k dxjk
= ∂x λ dx ∂xj1 ∂xjk
∂β I ∂xp
= ∂xp ∂xλ dx ∧ dx 1 ∧ · · · ∧ dxik
λ i

= dβ I ∧ dxI
P
where in (*) the sum is zero since:
r
∂ 2 x ir λ J ∂ 2 x ir
(dx ∧ dx ) = ± [(dxλ ∧ dxjr ) ∧ dxj1 ∧ · · · ∧ dx
d jr ∧ · · · dxjk ] = 0.
∂xλ ∂xjr ∂xλ ∂xjr
It follows that dβ is independent of the coordinates used to define it.
10.7 the pull-back
Another important operation one can perform on differential forms is the “pull-back” of a form
under a map9 . The definition is constructed in large part by a sneaky application of the push-
forward (aka differential) discussed in the preceding chapter. If you are impatient for intuition,
skip ahead to the end of this section and later return to the careful calculations at the outset.
Definition 10.7.1. pull-back of a differential form.
If f : M → N is a smooth map and ω ∈ ∧k (N ) then f ∗ ω is the form on M defined by
(f ∗ ω)p (X1 , · · · , Xk ) = ωf (p) (dp f (X1 ), dp f (X2 ), · · · , dp f (Xk )) .
for each p ∈ M and X1 , X2 , . . . , Xk ∈ Tp M. Moreover, in the case k = 0 we have a smooth

function ω : N → R and the pull-back is accomplished by composition (f ∗ ω)(p) = (ω ◦ f )(p)
for all p ∈ M.
This operation is linear on forms and commutes with the wedge product and exterior derivative:
Theorem 10.7.2. properties of the pull-back.
If f : M → N is a C 1 -map and ω ∈ ∧k (N ), τ ∈ ∧l (N ) then
1. f ∗ (aω + bτ ) = a(f ∗ ω) + b(f ∗ τ ) a, b ∈ R
2. f ∗ (ω ∧ τ ) = f ∗ ω ∧ (f ∗ τ )
3. f ∗ (dω) = d(f ∗ ω)
Proof: The proof of (1) is clear. We now prove (2).
f ∗ (ω ∧ τ )]p (X1 , · · · , Xk+l ) = (ω ∧ τ )f (p) (dp f (X1 ), · · · , dp f (Xk+l ))

P
= σ (sgnσ)(ω ⊗ τ )f (p) (dp f (Xσ1 ), · · · , dp f (Xσ(k+l) ))
P
= σ sgn(σ)ω(dp f (Xσ(1) ), · · · dp f (Xσ(k) ))τ (df (Xσ(k+1) · · · df Xσ(k+l) )
= σ sgn(σ)(f ∗ ω)p (Xσ(1) , · · · , Xσ(k) )(f ∗ τp )(Xσ(k+1) , · · · , Xσ(k+l) )
P
= [(f ∗ ω) ∧ (f ∗ τ )]p (X1 , X2 , · · · , X(k+l) )
9
thanks to my advisor R.O. Fulp for the arguments that follow
10.7. THE PULL-BACK 251
Finally we prove (3).
f ∗ (dω)]p (X1 , X2 · · · , Xk+1 ) = (dω)f (p) (df (X1 ), · · · df (Xk+1 ))

= (dωI ∧ dxI )f (p) (df (X1 ), · · · , df (Xk+1 ))
!
∂ωI

= ∂xλ (dxλ ∧ dxI )f (p) (df (X1 ), · · · , df (Xk+1 ))
f (p) !

∂ωI

= ∂xλ [dp (xλ ◦ f ) ∧ dp (xI ◦ f )](X1 , · · · , Xk+1 )
f (p)
= [d(ωI ◦ f ) ∧ d(xI ◦ f )](X1 , · · · , Xk+1 )
= d[(ωI ◦ f )p dp (xI ◦ f )](X1 , · · · , Xk+1 )
= d(f ∗ ω)p (X1 , · · · , Xk+1 ) .
The theorem follows. .
We saw that one important application of the push-forward was to change coordinates for a given
vector. Similar comments apply here. If we wish to change coordinates on a given differential form
then we can use the pull-back. However, given the direction of the operation we need to use the
inverse coordinate transformation to pull forms forward. Let me mirror the example from the last
chapter for forms on R2 . We wish to convert from r, θ to x, y notation.
Example 10.7.3. Suppose F : R2r,θ → R2x,y is the polar coordinate transformation. In particular,
F (r, θ) = (r cos θ, r sin θ)
The inverse transformation, at least for appropriate angles, is given by

p
−1 2 2 −1
F (x, y) = x + y , tan (y/x) .
Let calculate the pull-back of dr under F −1 : let p = F −1 (q)
F −1∗ (dr)q = dp r(∂x |p)dp x + dp r(∂y |p)dp y
Again, drop the annoying point-dependence to see this clearly:

∂r ∂r
F −1∗ (dr) = dr(∂x )dx + dr(∂y )dy = dx + dy
∂x ∂y
Likewise,
∂θ ∂θ
F −1∗ (dθ) = dθ(∂x )dx + dθ(∂y )dy = dx + dy
∂x ∂y
p
Note that r = x2 + y 2 and θ = tan−1 (y/x) have the following partial derivatives:
∂r x x ∂r y y
=p = and =p =
∂x 2
x +y 2 r ∂y 2
x +y 2 r
∂θ −y −y ∂θ x x
= 2 2
= 2 and = 2 2
= 2
∂x x +y r ∂y x +y r
Of course the expressions using r are pretty, but to make the point, we have changed into x, y-
notation via the pull-back of the inverse transformation as advertised. We find:
xdx + ydy −ydx + xdy

dr = p and dθ = .
x2 + y 2 x2 + y 2
Once again we have found results with the pull-back that we might previously have chalked up to
substitution in multivariate calculus. That’s often the idea behind an application of the pull-back.
It’s just a formal langauge to be precise about a substitution. It takes us past simple symbol
pushing and gives us a rigorous notation for substutions. It’s a bit more than that though, the
substitution we discuss here takes us from one space to another in general.
10.7.1 intuitive computation of pull-backs

Consider the definition below: how can we understand this computationally?
(f ∗ ω)p (X1 , · · · , Xk ) = ωf (p) (dp f (X1 ), dp f (X2 ), · · · , dp f (Xk )) .
In particular, let us consider f (u, v) = (x, y, z). Furthermore, let us consider a one-form to begin
ω = adx + bdy + cdz the pull-back will be formed by a suitable linear combination of du and dv.
We calculate,
∂x ∂ ∂y ∂ ∂z ∂ ∂x ∂ ∂y ∂ ∂z ∂
df (∂u ) = + + & df (∂v ) = + +
∂u ∂x ∂u ∂y ∂u ∂z ∂v ∂x ∂v ∂y ∂v ∂z
Therefore, as ω = adx + bdy + cdz,
∂x ∂y ∂z ∂x ∂y ∂z
ω(df (∂u )) = a +b +c & ω(df (∂v )) = a +b +c
∂u ∂u ∂u ∂v ∂v ∂v
From which we deduce: by Proposition 10.5.4,

∗ ∂x ∂y ∂z ∂x ∂y ∂z
f ω= a +b +c du + a +b +c dv
∂u ∂u ∂u ∂v ∂v ∂v

∂x ∂x ∂y ∂y ∂z ∂z
=a du + dv +b du + dv +c du + dv . (10.5)
∂u ∂v ∂u ∂v ∂u ∂v
| {z } | {z } | {z }
dx for x=x(u,v) dy for y=y(u,v) dz for z=z(u,v)
This shows the pull-back of ω is accomplished by taking ω = adx + bdy + cdz and substituting the
total differentials of x = x(u, v), y = y(u, v) and z = z(u, v) into dx, dy and dz respectively.
Continuing in our somewhat special context, consider Ω = ady ∧ dz + bdz ∧ dx + cdx ∧ dy and use
Theorem 10.7.2 to simplify our life. We already worked out the formula for the one-form case so
we can use it to find the intuitive formula for the two-form:
f ∗ Ω = f ∗ [ady ∧ dz + bdz ∧ dx + cdx ∧ dy]
= f ∗ (ady) ∧ f ∗ dz + f ∗ (bdz) ∧ f ∗ dx + f ∗ (cdx) ∧ f ∗ dy (by Theorem 10.7.2)

∂y ∂y ∂z ∂z ∂z ∂z ∂x ∂x
=a du + dv ∧ du + dv + b du + dv ∧ du + dv
∂u ∂v ∂u ∂v ∂u ∂v ∂u ∂v

∂x ∂x ∂y ∂y
+c du + dv ∧ du + dv (by Equation 10.5)
∂u ∂v ∂u ∂v
The formula above simplifies considerable as certain terms vanish due to du∧du = 0 and dv∧dv = 0
and dv ∧ du = −du ∧ dv. Furthermore, the following standard notation10
∂(y, z) ∂y ∂z ∂y ∂z ∂(x, y) ∂x ∂y ∂x ∂y ∂(z, x) ∂z ∂x ∂z ∂x
= − & = − , & = − .
∂(u, v) ∂u ∂v ∂v ∂u ∂(u, v) ∂u ∂v ∂v ∂u ∂(u, v) ∂u ∂v ∂v ∂u

∂(xi ,xj ) ∂u xi ∂v xi
Generally, ∂(u,v) = det . Ugly equations aside, this allows us to express the pull-
∂u xj ∂v xj
back of the two-form in a way which is easy to remember and is readily generalized.

∗ ∂(y, z) ∂(z, x) ∂(x, y)
f Ω= a +b +c du ∧ dv
∂(u, v) ∂(u, v) ∂(u, v)
I should mention, throughout this calculation I have suppressed the point-dependence. Technically,
a, b, c in the expression above should be understood as a ◦ f, b ◦ f, c ◦ f which are functions of u, v.
The pull-back once complete trades a form in the range coordinates (x, y, z) for a new form in the
domain coordinates (u, v).
The discussion thus far is somewhat limiting since the domain only supports a nontrivial two-form.
To continue, let’s consider pull-backs for the mapping G : R3uvw → R4txyz . In this case, we can
consider the pull-back of
γ = adt ∧ dx ∧ dz + bdx ∧ dy ∧ dz + cdy ∧ dz ∧ dt + mdz ∧ dt ∧ dy
and we will obtain:

∗ ∂(t, x, y) ∂(x, y, z) ∂(y, z, t) ∂(z, t, x)
G γ= a +b +c +m du ∧ dv ∧ dw
∂(u, v, w) ∂(u, v, w) ∂(u, v, w) ∂(u, v, w)
Where we define the coefficients by the natural generalization of the second-order case. Multiplying
out the wedge product of the pull-back of the three one-forms will produce the signs of the following
determinants  
∂u xi ∂v xi ∂w xi
∂(xi , xj , xk )
= det  ∂u xj ∂v xj ∂w xj 
∂(u, v, w)
∂u xk ∂v xk ∂w xk
10
H.M. Edwards Advanced Calculus a Differential Forms approach spends dozens of pages explaining this through
intuitive geometric arguments which we do not pursue here for brevity
In words, the pull-back is formed with coefficients taken from determinants of submatrices of the
Jacobian matrix of G. That probably doesn’t help you much. Perhaps the following two-form
pull-back with respect to G will inspire: an arbitrary two-form on R4txyz has the form:
Ξ = A01 dt ∧ dx + A02 dt ∧ dy + A03 dt ∧ dz + A23 dy ∧ dz + A31 dz ∧ dx + A12 dx ∧ dy
when we pull this back under G to u, v, w-space we will get a form in dv ∧ dw, dw ∧ du and du ∧ dv.
The coefficients are again obtained by appropriate determinants of submatrices of the Jacobian: in
particular: G∗ Ξ =

∂(t, x) ∂(t, y) ∂(t, z) ∂(y, z) ∂(z, x) ∂(x, y)
A01 + A02 + A03 + A23 + A31 + A12 dv ∧ dw
∂(v, w) ∂(v, w) ∂(v, w) ∂(v, w) ∂(v, w) ∂(v, w)

∂(t, x) ∂(t, y) ∂(t, z) ∂(y, z) ∂(z, x) ∂(x, y)
+ A01 + A02 + A03 + A23 + A31 + A12 dw ∧ du
∂(w, u) ∂(w, u) ∂(w, u) ∂(w, u) ∂(w, u) ∂(w, u)

∂(t, x) ∂(t, y) ∂(t, z) ∂(y, z) ∂(z, x) ∂(x, y)
+ A01 + A02 + A03 + A23 + A31 + A12 du ∧ dv.
∂(u, v) ∂(u, v) ∂(u, v) ∂(u, v) ∂(u, v) ∂(u, v)
The arbitrary case is perhaps a bit tiresome. Let’s consider a particular example
Γ = (t2 + x2 )dy ∧ dz
and the mapping G(u, v, w) = (u − 1, v 2 + 1, w3 , uvw) = (t, x, y, z). Observe:
dt = du, dx = 2vdv, dy = 3w2 dw, dz = vwdu + uwdv + uvdw
the pull-back of Γ under G is thus:
G∗ Γ = (u − 1)2 + (v 2 + 1)2 3w2 dw ∧ [vwdu + uwdv + uvdw]

= (u − 1)2 + (v 2 + 1)2 3w2 [vw dw ∧ du + uw dw ∧ dv]

2 2
| {z uw})dv ∧ dw + A23 (u, v)(3w
= A23 (u, v)(−3w | {zvw})dw ∧ du (10.6)
∂(y,z) ∂(y,z)
∂(v,w) ∂(w,u)
In the above I let A23 (u, v) = (u − 1)2 + (v 2 + 1)2 . Let’s check the coefficients of Equation 10.6 in
view of the general claim of G∗ Ξ. Are the coefficients in fact the Jacobians indicated? We calculate:
0 3w2

∂(y, z) yv yw
= det = det = −3w2 uw
∂(v, w) z v zw uw uv
and
3w2 0

∂(y, z) yw yu
= det = det = 3w2 vw.
∂(w, u) z w zu uv vw
Well, that’s a relief. We can either approach the calculation of a pull-back in terms of Jacobian
coefficients or we can just plug in the pull-backs of each coordinate function and multiply it out.
I suppose both techinques have their place. Moreover, when faced with many abstract questions
I much prefer Definition 10.7.1. The thought of sorting through the proof of Theorem 10.7.2 in
Jacobian notation seems hopeless.
10.7.2 implicit function theorem in view of forms

warning: I am not certain my interpretation of the theorem that follows is sound at
this time, however, I may leave this here as a reminder for me to think more on this
family of conceptual problems
Previously we saw this theorem without the benefit of the calculus of differential forms, I hope this
brings new light to the topic. This is the implicit function theorem as presented in H.M. Edwards’
Advanced Calculus: a Differential Forms Approach.
Proposition 10.7.4.
Let us consider for i = 1, . . . m,
yi = fi (x1 , . . . , xn )
where fi are continuously differentiable near a point (x̄1 , . . . , x̄n ). Furthermore, denote
ȳi = fi (x̄1 , . . . , x̄n ) for i = 1, . . . m. Suppose the following conditions are satisfied:
∂(y1 ,...,yr )
1. ∂(x1 ,...,xr ) 6= 0 at (x̄1 , . . . , x̄n )
2. the pull-back of every k-form for k > r is identically zero near (x̄1 , . . . , x̄n )
Then near (x̄1 , . . . , x̄n , ȳ1 , . . . , ȳm ) ∈ Rn+m there exist differentiable functions gi , hi which
solve yi = fi (x1 , . . . , xn ) provided we suppose:
xi = gi (y1 , . . . , yr , xr+1 , . . . , xn ) (for i = 1, . . . , r)
yi = hi (y1 , . . . , yr ) (for i = r + 1, . . . , m).
Condition (1.) implies the rank is at least r then condition (2.) assures us the rank is at most r
hence the rank is r. If the graph y = f (x) is viewed as G(x, y) = y − f (x) then it is seen as a
mapping from Rn × Rm → Rm . We saw before that if G(x̄, ȳ) = 0 and rank(G0 (x̄, ȳ)) = m then the
solution set of G(x, y) = 0 near (x̄, ȳ) forms an (n)-dimensional manifold. This generalizes that in
a sense because it allows for redundant conditions as indicated by r < m. On the other hand, this
is less general than the previous implicit function theorem as it assumes a linear-dependence on
the m-variables y1 , . . . , ym for the m equations defining the level set in Rm × Rn . The formulation
of this Theorem in Edward’s text falls inline with the general pattern of that text to emphasize
equations over mappings. It is both the strength and weakness of the text in my opinion. The next
two examples attempt to work within the confines of the notation put forth in the theorem above,
then we transition to examples where we informally apply the theorem.
Example 10.7.5. Just to remind us of the counting: if y1 = f (x1 , x2 ) = x1 + x2 − 1 then
G(x1 , x2 , y1 ) = y1 − (x1 + x2 − 1) = −x1 − x2 + y1 + 1 = 0
is a plane in (x1 , x2 , y1 )-space. Consider, (x̄1 , x̄2 ) = (2, 3) then ȳ1 = 2 + 3 − 1 = 4. A point on this
plane is (2, 3, 4) we can use (x1 , x2 ) as coordinates hence I expect r = 1. Observe, as G : R3 → R
we consider pull-back of the one-form dt in the range,
G∗ dt = −dx1 − dx2 + dy1
apparently, in our context here:
∂(y1 , . . . , yr ) ∂t
= = −1 6= 0
∂(x1 , . . . , xr ) ∂x1
and clearly there is no 2-form which pulls back under G since the only two-form in t-space is trivial.
We can explicitly write x1 = g1 (y1 , x2 ) = −x2 + y1 + 1 and there is no hi since r + 1 = 2 > m.
Example 10.7.6. Let us attempt another example to unravel the meaning of the Implicit Function
Theorem. Suppose y1 = x21 − x22 and y2 = x21 − x22 clearly these are redundant. The theorem should
deal with this. Let G(x1 , x2 , y2 , y2 ) = (x21 −x22 , x21 −x22 ) so we might hope the solution set is 4−2 = 2-
dimensional however, the rank of G0 is usually 1 so the solution set is likely 3-dimensional. No
need for guessing, let’s work it out.
dy1 = 2x1 dx1 − 2x2 dx2 = dy2
Hence, if we attempt r = 2 then consider:

∂(y1 , y2 ) 2x1 −2x2
= det = 0.
∂(x1 , x2 ) 2x1 −2x2
∂y1
So, we are forced to look at r = 1 for which ∂x 1
= 2x1 6= 0 for x1 6= 0. Consider then a point
such that x1 6= 0 and see if we can derive the functions such that x1 = g1 (y1 , x2 ) and y2 = h2 (y1 ).
Assume x1 , x2 > 0 for convenience (we could replicate this calculation in other quadrants with
appropriate signs adjoined),
q
x1 = y1 + x22 = g1 (y1 , x2 ), & y2 = y1 = h1 (y1 ).
I’m mostly interested in this theorem to gain more geometric insight as to what the pull-back
means. So, a better way to look at the last example is just to emphasize
dy1 = 2x1 dx1 − 2x2 dx2 = dy2
shows the pull-back of the basic one-forms can only give a one-form and where the coefficients are
zero we cannot force the corresponding coordinate to be dependent. For example, 2x1 dx1 is trivial
when x1 = 0 hence x1 cannot be solved for as a function of the remaining variables. This is the
computational essence of the theorem. Ideally, I want to get us to the point of calculating without
reliance on the Jacobians. Towards that end, let’s consider an example for which m = r hence the
h function is not needed and we can focus on the pull-back geometry.
Example 10.7.7. For which variables can we solve the following system (possibly subject some
condition). Let F (x, y, z) = (s, t) defined as follows:
s=x+z−1 (10.7)
t=y+z−2
then clearly F ∗ ds = dx + dz and F ∗ dt = dy + dz.
F ∗ (ds ∧ dt) = (dx + dz) ∧ (dy + dz) = dx ∧ dy + dx ∧ dz + dz ∧ dy
| {z } | {z } | {z }
(1.) (2.) (3.)
Therefore, by (1.), we can solve for x, y as functions of s, t, z. Or, by (2.), we could solve for x, z
as functions of s, t, y. Or, by (3.), we could solve for z, y as functions of s, t, x. The fact that the
rank is maximal is implicit within the fact that ds ∧ dt is the top-form in the range. In contrast,
F ∗ ds = dx + dz
does not mean that I can solve Equation 10.7 by solving for x as a function of s, t, y, z. Of course,
x = s − z + 1 solves the first equation, but the second equation is not contrained by the solution for x
what so over. Conversely, we can solve for z = t − y + 2 but we cannot also solve for z = s − x + 1.
The coefficients of 1 in F ∗ ds = dx + dz are not applicable to the Implicit Function Theorem because
this form is not the highest degree which pulls-back nontrivially. That role falls to ds ∧ dt here.
Example 10.7.8. For which variables can we solve the following system (possibly subject some
condition). Let F (u, v) = (x, y, z) defined as follows:
x = u2 + v 2 (10.8)
y = u2 − v 2
z = uv
then
dx = 2udu + 2vdv
dy = 2udu − 2vdv
dz = vdu + udv
We can calculate,
dy ∧ dz = (2udu − 2vdv) ∧ (vdu + udv) = 2(u2 + v 2 )du ∧ dv
dz ∧ dx = (vdu + udv) ∧ (2udu + 2vdv) = 2(v 2 − u2 )du ∧ dv
dx ∧ dy = (2udu + 2vdv) ∧ (2udu − 2vdv) = −8uv du ∧ dv
What does this tell us geometrically? Well, notice that the top-form dx ∧ dy ∧ dz pulls-back to zero
since the two-form du ∧ dv is the top-form in the domain. Therefore, r = 2, that is F has rank 2.
Honestly, at this point I doubt about 1/3 of my conclusions in this section. I’m
pretty sure I’m missing something big here. I may leave this in the 2013 notes for the
amusement of the reader, but beware my doubt for this last subsection. I’m much
more certain the push-forward calculations are fine.
10.8 integration of forms

The general strategy is generally as follows:
(i) there is a natural way to calculate the integral of a k-form on a subset of Rk
(ii) given a k-form on a manifold we can locally pull it back to a subset of Rk provided the
manifold is an oriented11 k-dimensional and thus by the previous idea we have an integral.
(iii) globally we should expect that we can add the results from various local charts and arrive at
a total value for the manifold, assuming of course the integral in each chart is finite.
We will only investigate items (i.) and (ii.) in these notes. There are many other excellent texts
which take great effort to carefully expand on point (iii.) and I do not wish to replicate that effort
here. You can read Edwards and see about pavings, or read Munkres’ where he has at least 100
pages devoted to the careful study of multivariate integration. I do not get into those topics in my
notes because we simply do not have sufficient analytical power to do them justice. I would encour-
age the student interested in deeper ideas of integration to find time to talk to Dr. Skoumbourdis,
he has thought a long time about these matters and he really understands integration in a way we
dare not cover in the calculus sequence. You really should have that conversation after you’ve taken
real analysis and have gained a better sense of what analysis’ purpose is in mathematics. That
said, what we do cover in this section and the next is fascinating whether or not we understand all
the analytical underpinnings of the subject!
10.8.1 integration of k-form on Rk

Note that on U ⊆ Rk a k-form α is the top form thus there exists some smooth function f on U
such that αx = f (x)dx1 ∧ dx2 ∧ · · · ∧ dxk for all x ∈ U . If D is a subset of U then we define the
integral of α over D via the corresponding intgral of k-variables in Rk . In particular,
Z Z
α= f (x)dk x
D D
where on the r.h.s. the symbol dk x is meant to denote the usual integral of k-variables on Rk . It
is sometimes convenient to write such an integral as:
Z Z
k
f (x)d x = f (x)dx1 dx2 · · · dxk
D D
but, to be more careful, the integration of f over D is a quantity which is independent of the
particular order in which the variables on Rk are assigned. On the other hand, the order of the
variables in the formula for α certainly can introuduce signs. Note
αx = −f (x)dx2 ∧ dx1 ∧ · · · ∧ dxk .

11
we will discuss this as the section progresses
10.8. INTEGRATION OF FORMS 259
How can we reasonably maintain the integral proposed above? Well, the answer is to make a con-
vention that we write the form to match the standard orientation of Rk . The standard orientation
of Rk isR given by
R V olk = dx1 ∧ dx2 ∧ · · · ∧ dxk . If the given form is written αx = f (x)V olk then we
k
define D α = D f (x)d x. Since it is always possible to write a k-form as a function multiplying
V olk on Rk this definition suffices to cover all possible k-forms. For example, if αx = f (x)dx on
some subset D = [a, b] of R,
Z Z Z b
α= f (x)dx = f (x)dx.
D D a
Or, if α(x,y) = f (x, y)dx ∧ dy then for D a aubset of 2
R ,
Z Z Z
α= f (x, y)dxdy = f dA.
D D D
If α(x,y,z) = f (x, y, z)dx ∧ dy ∧ dz then for D a aubset of R3 ,

Z Z Z
α= f (x, y, z)dxdydz = f dV.
D D D
In practice we tend to break

R the integrals
R above down into an interated integral thanks to Fubini’s
theorems. The integrals D f dA and D f dV are not in and of themselves dependent on orientation.
However the set D may be oriented the value of those integrals are the same for a fixed function
f . The orientation dependence of the form integral is completely wrapped up in our rule that the
form must be written as a multiple of the volume form on the given space.
10.8.2 orientations and submanifolds

Given a k-manifold M we say it is an oriented manifold iff all coordinates on M are consistently
oriented. If we make a choice and say φ0 : U0 → V0 is right-handed then any overlapping patch
φ1 : U1 → V1 is said to be right-handed iff det(dθ01 ) > 0. Otherwise, if det(dθ01 ) < 0 then the
patch φ1 : U1 → V1 is said to be left-handed. If the manifold is orientable then as we continue to
travel across the manifold we can choose coordinates such that on each overlap the transition func-
tions satisfy det(dθij ) > 0. In this way we find an atlas for an orientable M which is right-handed.
We can also say M is oriented is there exists a nonzero volume-form on M. If M is k-dimensional

then a volume form V ol is simply a nonzero k-form. At each point p ∈ M we can judge if a given
coordinate system is left or right handed. We have to make a convention to be precise and I do so
at this point. We assume V ol is positive and we say a coordinate system with chart (x1 , x2 , . . . , xk )
is positively oriented iff V ol(∂1 |p , ∂2 |p , . . . , ∂k |p ) > 0. If a coordinate system is not positively
oriented then it is said to be negatively oriented and we will find V ol(∂1 |p , ∂2 |p , . . . , ∂k |p ) < 0 in
that case. It is important that we suppose V olp 6= 0 at each p ∈ M since that is what allows us to
demarcate coordinate systems as positive or negatively oriented.
Naturally, you are probably wondering: is a positively oriented coordinate system is the same idea
as a right-handed coordinate system as defined above? To answer that we should analyze how the
V ol changes coordinates on an overlap. Suppose we are given a positive volume form V ol and
a point p ∈ M where two coordinate systems x and y are both defined. There must exist some
function f such that
V olx = f (x)dx1 ∧ dx2 ∧ · · · ∧ dxk
j
To change coordinates recall dxj = kj=1 ∂x dy j and subsitute,
P
∂y j
k
X ∂x1 ∂x2 ∂xk j1
V ol = (f ◦ x ◦ y −1 )(y)
j j
· · · dy ∧ dy j2 ∧ · · · ∧ dy jk
∂y 1 ∂y 2 ∂y jk
j1 ,...,jk =1

−1 ∂x
= (f ◦ x ◦ y )(y)det dy 1 ∧ dy 2 ∧ · · · ∧ dy k (10.9)
∂y
If you calculate the value of V ol on ∂xI |p = ∂x1 |p , ∂x2 |p , . . . , ∂xk |p you’ll find V ol(∂xI |p ) = f (x(p).
Whereas, if you evaluate V ol on ∂yI |p = ∂y1 |p , ∂y2 |p , . . . , ∂yk |p then the value is V ol(∂yI |p ) =
f (x(p))det ∂x
∂x
∂y (p) . But, we should recognize that det ∂y = det(dθij ) hence two coordinate sys-
tems which are positively oriented must also ∂xbe consistently
∂x Why? Assume V ol(∂xI |p ) =
oriented.
f (x(p)) > 0 then V ol(∂yI |p ) = f (x(p))det ∂y (p) > 0 iff det ∂y (p) > 0 hence y is positively ori-
ented if we are given that x is positively oriented and det ∂x

∂y > 0.
Let M be an oriented k-manifold with orientation given by the volume form V ol and an associated
atlas of positively oriented charts. Furthermore, let α be a p-form defined on V ⊆ M. Suppose
there exists a local parametrization φ : U ⊆ Rk → V ⊆ M and D ⊂ V then there is a smooth
function h such that αq = h(q)dx1 ∧ dx2 ∧ · · · ∧ dxk for each q ∈ V . We define the integral of α
over D as follows: Z Z
α= h(φ(x))dk x ← [?x ]
D φ−1 (D)
Is this definition dependent on the coordinate system φ : U ⊆ Rk → V ⊆ M? If we instead used

coordinate system ψ : Ū ⊆ Rk → V̄ ⊆ M where coordinates y 1 , y 2 , . . . , y k on V̄ then the given
form α has a different coefficient of dy 1 ∧ dy 2 ∧ · · · ∧ dy k

1 2 k −1 ∂x
α = h(x)dx ∧ dx ∧ · · · ∧ dx = (h x y )(y)det
◦ ◦ dy 1 ∧ dy 2 ∧ · · · ∧ dy k
∂y
Thus, as we change over to y coordinates the function picks up a factor which is precisely the
determinant of the derivative of the transition functions.
Z Z
−1 ∂x
α= (h ◦ x ◦ y )(y)det dy 1 ∧ dy 2 ∧ · · · ∧ dy k
D ∂y
ZD
−1 ∂x k
= (h x y )(y)det
◦ ◦ d y ← [?y ]
−1
ψ (D) ∂y
R
We need ?x = ?y in order for the integral D α to be well-defined. Fortunately, the needed equality
is almost provided by the change of variables theorem for multivariate integrals on Rk . Recall,
Z Z
∂x k
k
f (x)d x = ˜
f (y)det d y
R R̄ ∂y
where f˜ is more pedantically written as f˜ = f ◦ y −1 , notation aside its just the function f written
in terms of the new y-coordinates. Likewise, R̄ limits y-coordinates so that the corresponding
x-coordinates are found in R. Applying this theorem to our pull-back expression,
Z Z
k −1
∂x k
h(φ(x)) d x = (h x y )(y)det
◦ ◦ d y.
φ−1 (D) ψ −1 (D) ∂y
Equality of ?x and ?y follows from the fact that M is oriented and has transition functions12 θij
which satisfy det(dθij ) > 0. We see that this integral to be well-defined only for oriented manifolds.
To integrate over manifolds without an orientation additional ideas are needed, but it is possible.
Perhaps the most interesting case to consider is that of an embedded k-manifold in Rn . In this
context we must deal with both the coordinates of the ambient Rn and the local parametrizations
of the k-manifold. In multivariate calculus we often consider vector fields which are defined on an
open subset of R3 and then we calculate the flux over a surfaces or the work along a curve. What
we have defined thus-far is in essence like definition how to integrate a vector field on a surface
or a vector field along a curve, no mention of the vector field off the domain of integration was
made. We supposed the forms were already defined on the oriented manifold, but, what if we are
instead given a formula for a differential form on Rn ? How can we restrict that differential form
to a surface or line or more generally a parametrized k-dimensional submanifold of Rn ? That is
the problem we concern ourselvew with for the remainder of this section.
Let’s begin with a simple object. Consider a one-form α = ni=1 αi dxi where the function p → αi (p)
P
is smooth on some subset of Rn . Suppose C is a curve parametrized by X : D ⊆ R → C ⊆ Rn then
∂

the natural chart on C is provided by the parameter t in particular we have Tp C = span{ ∂t to
}
∗ ∂
where X(to ) = p and Tp C = span{dto t} hence a vector field along C has the form f (t) ∂t and a
differential form has the form g(t)dt. How can we use the one-form α on Rn to naturally obtain a
one-form defined along C? I propose:
n
X ∂X i
α C (t) =
αi (X(t)) dt
∂t
i=1

It can be shown that αC is a one-form on C. If we change coordinates on the curve by reparametriz-
ing t → s it then the component relative to s vs. the component relative to t are related:
n n n
∂X i X dt ∂X i ∂X i

X dt X
αi (X(t(s))) = αi (X(t)) = αi (X(t))
ds ds ∂t ds ∂t
i=1 i=1 i=1
This is precisely the transformation rule we want for the components of a one-form.
12
once more recall the notation ∂x
∂y
is just the matrix of the linear transformation dθij and the determinant of a
linear transformation is the determinant of the matrix of the transformation
Example 10.8.1. Suppose α = dx + 3x2 dy + ydz and C is the curve X : R → C ⊆ R3 defined by

X(t) 2 2
= (1, t, t ) we have x = 1, y2= t and z = t hence dx = 0, dy = dt and dz = 2tdt on C hence
α C = 0 + 3dt + t(2tdt) = (3 + 2t )dt.

Pn 1 i j
Next, consider a two-form β = i,j=1 2 βij dx ∧ dx . Once more we consider a parametrized
submanifold of Rn . In particular use the notation X : D ⊆ R2 → S ⊆ Rn where u, v serve as
coordinates on the surface S. We can write an arbitrary two-form on S in the form h(u, v)du ∧ dv
where h : S → R is a smooth function on S. How should we construct h(u, v) given β? Again, I
think the following formula is quite natural, honestly, what else would you do13 ?
n
X ∂X i ∂X j
β S (u, v) =
βij (X(u, v)) du ∧ dv
∂u ∂v
i,j=1
The coefficient function of du ∧ dv is smooth because we assume βij is smooth on Rn and the local
i ∂X i
parametrization is also assumed smooth so the functions ∂X ∂u and ∂v are smooth. Moreover, the
component function has the desired coordinate change property with respect to a reparametrization
of S. Suppose we reparametrize by s, t, then suppressing the point-dependence of βij ,
n n n
X ∂Y i ∂Y j du dv X ∂X i ∂X j X ∂X i ∂X j
βS=
βij ds ∧ dt = βij ds ∧ dt = βij du ∧ dv.
∂s ∂t ds dt ∂u ∂v ∂u ∂v
i,j=1 i,j=1 i,j=1
Therefore, the restriction of β to S is coordinate independent and we have thus constructed a

two-form on a surface from the two-form in the ambient space.
Example 10.8.2. Consider β = y 2 dt ∧ dx + zdx ∧ dy + (x + y + z + t)dt ∧ dz. Suppose S ⊆ R4 is

parametrized by
X(u, v) = (1, u2 v 2 , 3u, v)
In other words, we are given that
t = 1, x = u2 v 2 , y = 3u, z = v

Hence, dt = 0, dx = 2uv 2 du + 2u2 vdv, dy = 3du and dz = dv. Computing β S is just a matter of
substuting in all the formulas above, fortunately dt = 0 so only the zdx ∧ dy term is nontrivial:
β S = v(2uv 2 du + 2u2 vdv) ∧ (3du) = 6u2 v 2 dv ∧ du = −6u2 v 2 du ∧ dv.

It is fairly clear that we can restrict any p-form on Rn to a p-dimensional parametrized submanifold
by the procedure we explained above for p = 1, 2. That is the underlying idea in the definitions
which follow. Beyond that, once we have restricted the p-form β on Rn to β|M then we pull-back
the restricted form to an open subset of Rp and reduce the problem to an ordinary multivariate
integral.
13 1
include the 2
you say?, we’ll see why not soon enough
Remark 10.8.3. .
Just a warning, Einstein summation convention is used in what follows, by my count there
are over a dozen places where we implicitly indicate a sum.
Our goal in the remainder of the section is to make contact with the14 integrals we study in calculus.
Note that an embedded manifold with a single patch is almost trivially oriented since there is no
overlap to consider. In particular, if φ : U ⊆ Rk → M ⊆ Rn is a local parametrization with
φ−1 = (u1 , u2 , . . . , uk ) then du1 ∧ du2 ∧ · · · ∧ duk is a volume form for M. This is the natural
generalization of the normal-vector field construction for surfaces in R3 .
Definition 10.8.4. integral of one-form along oriented curve:
Let α = αi dxi be a one form and let C be an oriented curve with parametrization X(t) :
[a, b] → C then we define the integral of the one-form α along the curve C as follows,
b
dX i
Z Z
α≡ αi (X(t)) (t)dt
C a dt
where X(t) = (X 1 (t), X 2 (t), . . . , X n (t)) so we mean X i to be the ith component of X(t).
Moreover, the indices are understood to range over the dimension of the ambient space, if
we consider forms in R2 then i = 1, 2 if in R3 then i = 1, 2, 3 if in Minkowski R4 then i
should be replaced with µ = 0, 1, 2, 3 and so on.
Example 10.8.5. One form integrals vs. line integrals of vector fields: We begin with a
vector field F~ and construct the corresponding one-form ωF~ = Fi dxi . Next let C be an oriented
curve with parametrization X : [a, b] ⊂ R → C ⊂ R, observe
b
dX i
Z Z Z
ωF~ = Fi (X(t)) (t)dt = F~ · d~l
C a dt C
You may note that the definition of a line integral of a vector field is not special to three dimensions,
we can clearly construct the line integral in n-dimensions, likewise the correspondance ω can be
written between one-forms and vector fields in any dimension, provided we have a metric to lower
the index of the vector field components. The same cannot be said of the flux-form correspondance,
it is special to three dimensions for reasons we have explored previously.
14
hopefully known to you already from multivariate calculus
Definition 10.8.6. integral of two-form over an oriented surface:
Let β = 21 βij dxi ∧ dxj be a two-form and let S be an oriented piecewise smooth surface with
parametrization X(u, v) : D2 ⊂ R2 → S ⊂ Rn then we define the integral of the two-form
β over the surface S as follows,
∂X i ∂X j
Z Z
β≡ βij (X(u, v)) (u, v) (u, v)dudv
S D2 ∂u ∂v
where X(u, v) = (X 1 (u, v), X 2 (u, v), . . . , X n (u, v)) so we mean X i to be the ith component
of X(u, v). Moreover, the indices are understood to range over the dimension of the ambient
space, if we consider forms in R2 then i, j = 1, 2 if in R3 then i, j = 1, 2, 3 if in Minkowski
R4 then i, j should be replaced with µ, ν = 0, 1, 2, 3 and so on.
Example 10.8.7. Two-form integrals vs. surface integrals of vector fields in R3 : We

begin with a vector field F~ and construct the corresponding two-form ΦF~ = 21 ijk Fk dxi ∧ dxj which
is to say ΦF~ = F1 dy ∧ dz + F2 dz ∧ dx + F3 dx ∧ dy. Next let S be an oriented piecewise smooth
surface with parametrization X : D ⊂ R2 → S ⊂ R3 , then
Z Z
ΦF~ = F~ · dA
~
S S
Proof: Recall that the normal to the surface S has the form,
∂X ∂X ∂X i ∂X j
N (u, v) = × = ijk ek
∂u ∂v ∂u ∂v
at the point X(u, v). This gives us a vector which points along the outward normal to the surface
and it is nonvanishing throughout the whole surface by our assumption that S is oriented. Moreover
the vector surface integral of F~ over S was defined by the formula,
Z Z Z
~ ~
F · dA ≡ F~ (X(u, v)) · N
~ (u, v) dudv.
S D
now that the reader is reminded what’s what, lets prove the proposition, dropping the (u,v) depence
to reduce clutter we find,
Z Z Z
~ ~
F · dA = F~ · N
~ dudv
S Z Z D
= Fk Nk dudv
D
∂X i ∂X j
Z Z
= Fk ijk dudv
D ∂u ∂v
∂X i ∂X j
Z Z
= (ΦF~ )ij dudv
D ∂u ∂v
Z
= ΦF~
S
notice that we have again used our convention that (ΦF~ )ij refers to the tensor components of
the 2-form ΦF~ meaning we have ΦF~ = (ΦF~ )ij dxi ⊗ dxj whereas with the wedge product ΦF~ =
1 i j
~ )ij dx ∧ dx , I mention this in case you are concerned there is a half in ΦF
2 (ΦF ~ yet we never found
a half in the integral. Well, we don’t expect to because we defined the integral of the form with
respect to the tensor components of the form, again they don’t contain the half.
Example 10.8.8. Consider the vector field F~ = (0, 0, 3) then the corresponding two-form is simply
ΦF = 3dx ∧ dy. Lets calculate the surface integral and two-form integrals over the square D =
[0, 1]×[0, 1] in the xy-plane, in this case the parameters can be taken to be x and y so X(x, y) = (x, y)
and,
∂X ∂X
N (x, y) = × = (1, 0, 0) × (0, 1, 0) = (0, 0, 1)
∂x ∂y
which is nice. Now calculate,
Z Z Z
F~ · dA
~ = F~ · N
~ dxdy
S Z Z D
= (0, 0, 3) · (0, 0, 1)dxdy

D
Z 1Z 1
= 3dxdy
0 0
= 3.
Consider that ΦF = 3dx ∧ dy = 3dx ⊗ dy − 3dy ⊗ dx therefore we may read directly that (ΦF )12 =
−(ΦF )21 = 3 and all other components are zero,
∂X i ∂X j
Z Z Z
ΦF = (ΦF )ij dxdy
S D ∂x ∂y
∂X 1 ∂X 2 ∂X 2 ∂X 1
Z Z
= 3 −3 dxdy
D ∂x ∂y ∂x ∂y
Z 1Z 1
∂x ∂y ∂y ∂x
= 3 −3 dxdy
0 0 ∂x ∂y ∂x ∂y
= 3.
Definition 10.8.9. integral of a three-form over an oriented volume:
Let γ = 16 βijk dxi ∧ dxj ∧ dxk be a three-form and let V be an oriented piecewise smooth
volume with parametrization X(u, v, w) : D3 ⊂ R3 → V ⊂ Rn then we define the integral
of the three-form γ in the volume V as follows,
∂X i ∂X j ∂X k
Z Z
γ≡ γijk (X(u, v, w)) dudvdw
V D3 ∂u ∂v ∂w
where X(u, v, w) = (X 1 (u, v, w), X 2 (u, v, w), . . . , X n (u, v, w)) so we mean X i to be the ith
component of X(u, v, w). Moreover, the indices are understood to range over the dimension
of the ambient space, if we consider forms in R3 then i, j, k = 1, 2, 3 if in Minkowski R4 then
i, j, k should be replaced with µ, ν, σ = 0, 1, 2, 3 and so on.
Finally we define the integral of a p-form over an p-dimensional subspace of R, we assume that
p ≤ n so that it is possible to embed such a subspace in R,
Definition 10.8.10. integral of a p-form over an oriented volume:
Let γ = p!1 βi1 ...ip dxi1 ∧ · · · dxip be a p-form and let S be an oriented piecewise smooth
subspace with parametrization X(u1 , . . . , up ) : Dp ⊂ Rp → S ⊂ Rn (for n ≥ p) then we
define the integral of the p-form γ in the subspace S as follows,
∂X i1 ∂X ip
Z Z
γ≡ βi1 ...ip (X(u1 , . . . , up )) ··· du1 · · · dup
S Dp ∂u1 ∂up
where X(u1 , . . . , up ) = (X 1 (u1 , . . . , up ), X 2 (u1 , . . . , up ), . . . , X n (u1 , . . . , up )) so we mean X i

to be the ith component of X(u1 , . . . , up ). Moreover, the indices are understood to range
over the dimension of the ambient space.
Integrals of forms play an important role in modern physics. I hope you can begin to appreciate
that forms recover all the formulas we learned in multivariate calculus and give us a way to extend
calculation into higher dimensions with ease. I include a toy example at the conclusion of this
chapter just to show you how electromagnetism is easily translated into higher dimensions.
10.9. GENERALIZED STOKES THEOREM 267
10.9 Generalized Stokes Theorem

The generalized Stokes theorem contains within it most of the main theorems of integral calculus,
namely the fundamental theorem of calculus, the fundamental theorem of line integrals (a.k.a
the FTC in three dimensions), Greene’s Theorem in the plane, Gauss’ Theorem and also Stokes
Theorem, not to mention a myriad of higher dimensional not so commonly named theorems. The
breadth of its application is hard to overstate, yet the statement of the theorem is simple,
Theorem 10.9.1. Generalized Stokes Theorem:
Let S be an oriented, piecewise smooth (p+1)-dimensional subspace of Rn where n ≥ p + 1

and let ∂S be it boundary which is consistently oriented then for a p-form α which behaves
reasonably on S we have that
Z Z
dα = α
S ∂S
The proof of this theorem (and a more careful statement of it) can be found in a number of places,
Susan Colley’s Vector Calculus or Steven H. Weintraub’s Differential Forms: A Complement to
Vector Calculus or Spivak’s Calculus on Manifolds just to name a few. I believe the argument in
Edward’s text is quite complete. In any event, you should already be familar with the idea from
the usual Stokes Theorem where we must insist the boundary curve to the surface is related to
the surface’s normal field according to the right-hand-rule. Explaining how to orient the boundary
∂M given an oriented M is the problem of generalizing the right-hand-rule to many dimensions. I
leave it to your homework for the time being.
Lets work out how this theorem reproduces the main integral theorems of calculus.
Example 10.9.2. Fundamental Theorem of Calculus in R: Let f : R → R be a zero-form

then consider the interval [a, b] in R. If we let S = [a, b] then ∂S = {a, b}. Further observe that
df = f 0 (x)dx. Notice by the definition of one-form integration
Z Z b
df = f 0 (x)dx
S a
However on the other hand we find ( the integral over a zero-form is taken to be the evaluation
map, perhaps we should have defined this earlier, oops., but its only going to come up here so I’m
leaving it.) Z
f = f (b) − f (a)
∂S
Hence in view of the definition above we find that

Z b Z Z
0
f (x)dx = f (b) − f (a) ⇐⇒ df = f
a S ∂S
Example 10.9.3. Fundamental Theorem of Calculus in R3 : Let f : R3 → R be a zero-

form then consider a curve C from p ∈ R3 to q ∈ R3 parametrized by φ : [a, b] → R3 . Note that
∂C = {φ(a) = p, φ(b) = q}. Next note that
∂f i
df = dx
∂xi
Then consider that the exterior derivative of a function corresponds to the gradient of the function
thus we are not to surprised to find that
Z b
∂f dxi
Z Z
df = i
dt = (∇f ) · d~l
C a ∂x dt C
On the other hand, we use the definition of the integral over a a two point set again to find
Z
f = f (q) − f (p)
∂C
Hence if the Generalized Stokes Theorem is true then so is the FTC in three dimensions,
Z Z Z
~
(∇f ) · dl = f (q) − f (p) ⇐⇒ df = f
C C ∂C
another popular title for this theorem is the ”fundamental theorem for line integrals”. As a final
thought here we notice that this calculation easily generalizes to 2,4,5,6,... dimensions.
Example 10.9.4. Greene’s Theorem: Let us recall the statement of Greene’s Theorem as I
have not replicated it yet in the notes, let D be a region in the xy-plane and let ∂D be its consistently
oriented boundary then if F~ = (M (x, y), N (x, y), 0) is well behaved on D
Z Z Z
∂N ∂M
M dx + N dy = − dxdy
∂D D ∂x ∂y
We begin by finding the one-form corresponding to F~ namely ωF = M dx + N dy consider then that

∂M ∂N
dωF = d(M dx + N dy) = dM ∧ dx + dN ∧ dy = dy ∧ dx + dx ∧ dy
∂y ∂x
which simplifies to,
∂N ∂M
dωF = − dx ∧ dy = Φ( ∂N − ∂M )k̂
∂x ∂y ∂x ∂y
Thus, using our discussion in the last section we recall

Z Z Z
ωF = ~ ~
F · dl = M dx + N dy
∂D ∂D ∂D
where we have reminded the reader that the notation in the rightmost expression is just another
way of denoting the line integral in question. Next observe,
Z Z
∂N ∂M ~
dωF = ( − )k̂ · dA
D D ∂x ∂y
10.9. GENERALIZED STOKES THEOREM 269
~ = k̂dxdy we have
And clearly, since dA
Z Z
∂N ∂M ~ ∂N ∂M
( − )k̂ · dA = ( − )dxdy
D ∂x ∂y D ∂x ∂y
Therefore,
Z Z Z Z Z
∂N ∂M
M dx + N dy = − dxdy ⇐⇒ dωF = ωF
∂D D ∂x ∂y D ∂D
Example 10.9.5. Gauss Theorem: Let us recall Gauss Theorem to begin, for suitably defined
F~ and V , Z Z
F~ · dA
~= ∇ · F~ dτ
∂V V
First we recall our earlier result that
d(ΦF ) = (∇ · F~ )dx ∧ dy ∧ dz
Now note that we may integrate the three form over a volume,
Z Z
d(ΦF ) = (∇ · F~ )dxdydz
V V
whereas, Z Z
ΦF = F~ · dA
~
∂V ∂V
so there it is, Z Z Z Z
(∇ · F~ )dτ = F~ · dA
~ ⇐⇒ d(ΦF ) = ΦF
V ∂V V ∂V
I have left a little detail out here, I may assign it for homework.
Example 10.9.6. Stokes Theorem: Let us recall Stokes Theorem to begin, for suitably defined
F~ and S, Z Z
~ ~
(∇ × F ) · dA = F~ · d~l
S ∂S
Next recall we have shown in the last chapter that,
d(ωF ) = Φ∇×F~
Hence, Z Z
d(ωF ) = (∇ × F~ ) · dA
~
S S
whereas, Z Z
ωF = F~ · d~l
∂S ∂S
which tells us that,
Z Z Z Z
(∇ × F~ ) · dA
~= F~ · d~l ⇐⇒ d(ωF ) = ωF
S ∂S S ∂S
The Generalized Stokes Theorem is perhaps the most persausive argument for mathematicians to
be aware of differential forms, it is clear they allow for more deep and sweeping statements of
the calculus. The generality of differential forms is what drives modern physicists to work with
them, string theorists for example examine higher dimensional theories so they are forced to use a
language more general than that of the conventional vector calculus.
10.10 Poincare lemma

This section is in large part inspired by M. Gockeler and T. Schucker’s Differential geometry, gauge
theories, and gravity page 20-22. The converse calculation is modelled on the argument found in
H. Flanders Differential Forms with Applications to the Physical Sciences. The original work was
done around the dawn of the twentieth century and can be found in many texts besides the two I
mentioned here.
10.10.1 exact forms are closed

Proposition 10.10.1.
The exterior derivative of the exterior derivative is zero. d2 = 0
Proof: Let α be an arbitrary p-form then

1
dα = (∂m αi1 i2 ...ip )dxm ∧ dxi1 ∧ dxi2 ∧ · · · ∧ dxip (10.10)
p!
then differentiate again,

1 m i i i
d(dα) = d p! (∂m αi1 i2 ...ip )dx ∧ dx ∧ dx ∧ · · · ∧ dx
1 2 p
(10.11)
= p!1 (∂k ∂m αi1 i2 ...ip )dxk ∧ dxm ∧ dxi1 ∧ dxi2 ∧ · · · ∧ dxip
=0
since the partial derivatives commute whereas the wedge product anticommutes so we note that
the pair of indices (k,m) is symmetric for the derivatives but antisymmetric for the wedge, as we
know the sum of symmetric against antisymmetric vanishes ( see equation ?? part iv if you forgot.)
Definition 10.10.2.
A differential form α is closed iff dα = 0. A differential form β is exact iff there exists γ
such that β = dγ.
All exact forms are closed. However, there exist closed forms which are not exact.
10.10. POINCARE LEMMA 271
Proof: Exact implies closed is easy, let β be exact such thatβ = dγ then
dβ = d(dγ) = 0
using the theorem d2 = 0. To prove that there exists a closed form which is not exact it suffices
to give an example. A popular example ( due to its physical significance to magnetic monopoles,
Dirac Strings and the like..) is the following differential form in R2
1
φ= (xdy − ydx) (10.12)
x2 + y 2
You may verify that dφ = 0 in homework. Observe that if φ were exact then there would exist f
such that φ = df meaning that
∂f y ∂f x
=− 2 , = 2
∂x x + y2 ∂y x + y2
which are solved by f = arctan(y/x) + c where c is arbitrary. Observe that f is ill-defined along
the y-axis x = 0 ( this is the Dirac String if we put things in context ), however the natural domain
of φ is R n×n − {(0, 0)}.
10.10.2 potentials for closed forms

Poincare’ suggested the following partial converse, he said closed implies exact provided we place
a topological restriction on the domain of the form. In particular, if the domain of a closed form
is smoothly deformable to a point then each closed form is exact. We’ll work out a proof of that
result for a subset of Rn . Be patient, we have to build some toys before we play.
Suppose U ⊆ Rn and I = [0, 1] we denote a typical point in I × U as (t, x) where t ∈ I and x ∈ Rn .

Define maps,
J1 : U → I × U, J0 : U → I × U
by J1 (x) = (1, x) and J0 (x) = (0, x) for each x ∈ U . Flanders encourages us to view I × U as a
cylinder and where the map J1 maps U to the top and J0 maps U to the base. We can pull-back
forms on the cylinder to the U on the top (t = 1) or to the base (t = 0). For instance, if we consider
ω = (x + t)dx + dt for the case n = 1 then
J0∗ ω = xdx J1∗ ω = (x + 1)dx.
Define a smooth mapping K of (p + 1) forms on I × U to p-forms on U as follows:

Z 1
I J
(1.) K(a(t, x)dx ) = 0, (2.) K(a(t, x)dt ∧ dx ) = a(t, x)dt dxJ
0
for multi-indices I of length (p + 1) and J of length p. The cases (1.) and (2.) simply divide the
possible monomial15 inputs from Λp+1 (I × U ) into forms which have dt and those which don’t.
Then K is defined for a general (p + 1)-form on I × U by linearly extending the formulas above to
multinomials of the basic monomials.
It turns out that the following identity holds for K:

Lemma 10.10.4. the K-lemma.
If ω is a differential form on I × U then
K(dω) + d(K(ω)) = J1∗ ω − J0∗ ω.
Proof: Since the equation is given for linear operations it suffices to check the formula for mono-
mials since we can extend the result linearly once those are affirmed. As in the definition of K
there are two basic categories of forms on I × U :
Case 1: If ω = a(t, x)dxI then clearly K(ω) = 0 hence d(K(ω)) = 0. Observe,

X ∂a ∂a
dω = da ∧ dxI = j
dxj ∧ dxI + dt ∧ dxI
∂x ∂t
j
Hence K(dω) is calculated as follows:

X
∂a j I ∂a I
K(dω) = K dx ∧ dx + K dt ∧ dx
∂xj ∂t
j
Z 1
∂a
= dt dxI
0 ∂t
= a(x, 1) − a(x, 0) dxI

= J1∗ ω − J0∗ ω (10.13)
where we used the FTC in the next to last step. The pull-backs in this case just amount to evalu-
ation at t = 0 or t = 1 as there is no dt-type term to squash in ω. The identity follows.
15
dx ∧ dy is a monomial whereas dx + dy is a binomial in this context
Case 2: Suppose ω = a(t, x)dt ∧ dxJ . Calculate,

X ∂a ∂a
dω = dxk ∧ dt ∧ dxJ + dt ∧ dt ∧dxJ
∂xk ∂t | {z }
j zero !
Thus, using dxk ∧ dt = −dt ∧ dxk , we calculate:

X
∂a k I
K(dω) = K − dt ∧ dx ∧ dx
∂xk
k
XZ 1 ∂a
=− k
dt dxk ∧ dxI
0 ∂x
k
at which point we cannot procede further since a is an arbitrary function which can include a
nontrivial time-dependence. We turn to the calculation of d(K(ω)). Recall we defined
Z 1
K(ω) = a(t, x)dt dxJ .
0
We calculate the exterior derivative of K(ω):

Z 1
d(K(ω)) = d a(t, x)dt ∧ dxJ
0
Z 1 X ∂ Z 1
∂
= a(τ, x)dτ dt + a(t, x)dt dx ∧ dxJ
k
∂t 0 ∂xk 0
| {z } k
constant in t
XZ 1
∂a
= dt dxk ∧ dxJ . (10.14)
0 ∂xk
k
Therefore, K(dω)+d(K(ω)) = 0 and clearly J0∗ ω = J1∗ ω = 0 in this case since the pull-backs squash
the dt to zero. The lemma follows. .
Definition 10.10.5.
A subset U ⊆ Rn is deformable to a point P if there exists a smooth mapping G : I ×U → U
such that G(1, x) = x and G(0, x) = P for all x ∈ U .
The map G deforms U smoothly into the point P . Recall that J1 (x) = (1, x) and J0 (x) = (0, x)
hence the conditions on the deformation can be expressed as:
G(J1 (x)) = x G(J0 (x)) = P
Denoting Id for the identity on U and P as the constant mapping with value P on U we have
G ◦ J1 = Id G ◦ J0 = P
Therefore, if γ is a (p + 1)-form on U we calculate,
(G ◦ J1 )∗ γ = Id∗ γ ⇒ J1∗ [G∗ γ] = γ
whereas,
(G ◦ J0 )∗ γ = P ∗ γ = 0 ⇒ J0∗ [G∗ γ] = 0
Apply the K-lemma to the form ω = G∗ γ on I × U and we find:
K(d(G∗ γ)) + d(K(G∗ γ)) = γ.
However, recall that we proved that pull-backs and exterior derivatives commute thus
d(G∗ γ) = G∗ (dγ)
and we find an extremely interesting identity,
K(G∗ (dγ)) + d(K(G∗ γ)) = γ.
If U ⊆ R is deformable to a point P then a p-form γ on U is closed iff φ is exact.
Proof: Suppose γ is exact then γ = dβ for some (p − 1)-form β on U hence dγ = d(dβ) = 0 by

Proposition 10.10.1 hence γ is closed. Conversely, suppose γ is closed. Apply the boxed consequence
of the K-lemma, note that K(G∗ (0)) = 0 since we assume dγ = 0. We find,
d(K(G∗ γ)) = γ
identify that G∗ γ is a p-form on I × U whereas K(G∗ γ) is a (p − 1)-form on U by the very con-

struction of K. It follows that γ is exact since we have shown how it is obtained as the exterior
derivative of another differential form of one degree less.
Where was deformability to a point P used in the proof above? The key is the existence of the
mapping G. In other words, if you have a space which is not deformable to a point then no
deformation map G is available and the construction via K breaks down. Basically, if the space
has a hole which you get stuck on as you deform loops to a point then it is not deformable to a
point. Often we call such spaces simply connected. Careful definition of these terms is too difficult
for calculus, deformation of loops and higher dimensional objects is properly covered in algebraic
topology. In any event, the connection of the deformation and exactness of closed forms allows
topologists to use differential forms detect holes in spaces. In particular:
Definition 10.10.7. de Rham cohomology:
We define several real vector spaces of differential forms over some subset U of R,
Z p (U ) ≡ {φ ∈ Λp U | φ closed}
the space of closed p-forms on U . Then,
B p (U ) ≡ {φ ∈ Λp U | φ exact}
the space of exact p-forms where by convention B 0 (U ) = {0} The de Rham cohomology
groups are defined by the quotient of closed/exact,
H p (U ) ≡ Z p (U )/B p (U ).
the dim(H p U ) = pth Betti number of U.

We observe that simply connected regions have all the Betti numbers zero since Z p (U ) = B p (U )
implies that H p (U ) = {0}. Of course there is much more to say about de Rahm Cohomology, I just
wanted to give you a taste and alert you to the fact that differential forms can be used to reveal as-
pects of topology. Not all algebraic topology uses differential forms though, there are several other
calculational schemes based on triangulation of the space, or studying singular simplexes. One
important event in 20-th century mathematics was the discovery that all these schemes described
the same homology groups. The Steenrod reduced the problem to a few central axioms and it was
shown that all the calculational schemes adhere to that same set of axioms.
One interesting aspect of the proof we (copied from Flanders 16 ) is that it is not a mere exis-
tence proof. It actually lays out how to calculate the form which provides exactness. Let’s call
β the potential form of γ if γ = dβ. Notice this is totally reasonable langauge since in the case
of classical mechanics we consider conservative forces F~ which as derivable from a scalar potential
V by F~ = −∇V . Translated into differential forms we have ωF~ = −dV . Let’s explore how the
K-mapping and proof indicate the potential of a vector field ought be calculated.
Suppose U is deformable to a point and F is a smooth conservative vector field on U . Perhaps you
recall that for conservative F are irrotational hence ∇ × F = 0. Recall that dωF = Φ∇×F = Φ0 = 0
thus the one-form corresponding to a conservative vector field is a closed form. Apply the identity:
let G : I × U → U ⊆ R3 be the deformation of U to a point P ,
d(K(G∗ ωF )) = ωF
Hence, including the minus to make energy conservation natural,
V = −K(G∗ ωF )
16
I don’t know the complete history of this calculation at the present. It would be nice to find it since I doubt
Flanders is the originator.
For convenience, lets suppose the space considered is the unit-ball B and lets use a deformation to
the origin. Explicitly, G(t, r) = tr for all r ∈ R3 such that ||r|| ≤ 1. Note that clearly G(0, r) = 0
whereas G(1, r) = r and G has a nice formula so it’s smooth17 . We wish to calculate the pull-back
of ωF = P dx + Qdy + Rdz under G, from the definition of pull-back we have
(G∗ ωF )(X) = ωF (dG(X))
for each smooth vector field X on I × B. Differential forms on I × B are written as linear combi-
nations of dt, dx, dy, dz with smooth functions as coefficients. We can calculate the coefficents by
evalutaion on the corresponding vector fields ∂t , ∂x , ∂y , ∂z . Observe, since G(t, x, y, z) = (tx, ty, tz)
we have
∂G1 ∂ ∂G2 ∂ ∂G3 ∂ ∂ ∂ ∂
dG(∂t ) = + + =x +y +z
∂t ∂x ∂t ∂y ∂t ∂z ∂x ∂y ∂z
wheras,
∂G1 ∂ ∂G2 ∂ ∂G3 ∂ ∂
dG(∂x ) = + + =t
∂x ∂x ∂x ∂y ∂x ∂z ∂x
and similarly,
∂ ∂
dG(∂y ) = t dG(∂x ) = t
∂y ∂z
Furthermore,
ωF (dG(∂t )) = ωF (x∂x + y∂y + z∂z ) = xP + yQ + zR
ωF (dG(∂x )) = ωF (t∂x ) = tP, ωF (dG(∂y )) = ωF (t∂y ) = tQ, ωF (dG(∂z )) = ωF (t∂z ) = tR
Therefore,
G∗ ωF = (xP + yQ + zR)dt + tP dx + tQdy + tRdz = (xP + yQ + zR)dt + tωF
Now we can calculate K(G∗ ωF ) and hence V 18

∗

K(G ωF )(t, x, y, z) = K xP (tx, ty, tz) + yQ(tx, ty, tz) + zR(tx, ty, tz) dt
Therefore,
Z 1
∗

V (x, y, z) = −K(G ωF ) = − xP (tx, ty, tz) + yQ(tx, ty, tz) + zR(tx, ty, tz) dt
0
Notice this is precisely the line-integral of F =< P, Q, R > along the line C with direction < x, y, z >
from the origin to (x, y, z). In particular, if ~r(t) =< tx, ty, tz > then d~ r
dt =< x, y, z > hence we
identify Z 1 Z
d~r
V (x, y, z) = − ~
F ~r(t) · dt = − F~ · d~r
0 dt C
17
there is of course a deeper meaning to the word, but, for brevity I gloss over this.
18
Note that only the coefficient of dt gives a nontrivial contribution so in retrospect we did a bit more calculation
than necessary. That said, I’ll just keep it as a celebration of extreme youth for calculation. Also, I’ve been a bit
careless in omiting the point up to this point, let’s include the point dependence since it will be critical to properly
understand the formula.
Perhaps you recall this is precisely how we calculate the potential function for a conservative vector
field provided we take the origin as the zero for the potential.
Actually, this calculation is quite interesting. Suppose we used a different deformation

G̃ : I × U → U . For fixed point Q we travel to from the origin to the point by the path t 7→ G̃(t, Q).
Of course this path need not be a line. The space considered might look like a snake where a line
cannot reach from the base point P to the point Q. But, the same potential is derived. Why?
Path independence of the vector field is one answer. The criteria ∇ × F = 0 suffices for a sim-
ply connected region. However, we see something deeper. The criteria of a closed form paired
with a simply connected (deformable) domain suffices to construct a potential for the given form.
This result reproduces the familar case of conservative vector fields derived from scalar potentials
and much more. In Flanders he calculates the potential for a closed two-form. This ought to be
the mathematics underlying the construction of the so-called vector potential of magnetism. In
junior-level electromagnetism19 the magnetic field B satisfies ∇ · B = 0 and thus the two-form
ΦB has exterior derivative dΦB = ∇ · Bdx ∧ dy ∧ dz = 0. The magnetic field corresponds to a
closed form. Poincare’s lemma shows that there exists a one-form ωA such that dωA = ΦB . But
this means Φ∇×A = ΦB hence in the langauge of vector fields we expect the vector potential A
generated the magnetic field B throught the curl B = ∇ × A. Indeed, this is precisely what a
typical junior level physics student learns in the magnetostatic case. Appreciate that it goes deeper
still, the Poincare lemma holds for p-forms which correspond to objects which don’t match up with
the quaint vector fields of 19-th century physics. We can be confident to find potential for 3-form
fluxes in a 10-dimensional space, or wherever our imagination takes us. I explain at the end of this
chapter how to translate electromagnetics into the langauge of differential forms, it may well be
that in the future we think about forms the way we currently think about vectors. This is one of
the reasons I like Flanders text, he really sticks with the langauge of differential forms throughout.
In contrast to these notes, he just does what is most interesting. I think undergraduates need to
see more detail and not just the most clever calculations, but, I can hardly blame Flanders! He
makes no claim to be an undergraduate work.
Finally, I should at least mention that though we can derive a potential β for a given closed form
α on a simply connected domain it need not be unique. In fact, it will not be unique unless we add
further criteria for the potential. This ambuity is called gauge freedom in physics. Mathematically
it’s really simple give form language. If α = dβ where β is a (p − 1)-form then we can take any
smooth (p − 2) form and calculate that
d(α + dλ) = dβ + d2 λ = dβ = α
Therefore, if β is a potential-form for α then β + dλ is also a potential-form for α.
19
just discussing magnetostatic case here to keep it simple
This section will likely get fixed thus shifting all page numbers past here. For this
reason, I would not recommend printing past here until I finish the edit of this section
and a few others in this chapter and add another 10-20 pages to the next chapter.
The chapter on electromagnetism and also the chapter on variational calculus are not
likely to change much, it’s just this part and the next chapter which are the new part.
Also, the final chapter (which we don’t cover) needs some pictures. Hopefully I wrap
up these additions in the first few weeks of class, long before we get here...
10.11 introduction to geometric differential equations

Differential forms, pull-backs and submanifolds provide a language in which the general theory of
partial differential equations is naturally expressed. That said, let us begin with an application to
the usual differential equations course.
10.11.1 exact differential equations

Throughout what follows assume that M, N, I, F are continuously differentiable functions of x, y,
perhaps just defined for some open subset of R2 . Recall (or learn) that a differential equation
M dx + N dy = 0 is said to be exact iff there is a function F such that dF = M dx + N dy. This
is a very nice kind of differential equation since the solution is simply F (x, y) = c. A convenient
test for exactness was provided from the fact that partial derivatives commute. This exchange of
partial derivatives implies that ∂y M = ∂x N . Observe, we can recover this condition via exterior
differentiation:
If a differential equation M dx + N dy = 0 is exact then d(M dx + N dy) = 0 (this zero is

derived from M, N alone, independent of the given differential equation).
Proof: if M dx + N dy = 0 is exact then by definition dF = M dx + N dy hence d(M dx + N dy) =
d(dF ) = 0.
Pfaff did pioneering work in the theory of differential forms. One Theorem due to Pfaff states that
any first order differential equation can be made exact by the method of integrating factors. In
particular, if M dx + N dy = 0 is not exact then there exists I such that IM dx + IN dy = 0 is
an exact differential equation. The catch, it is as hard or harder to find I as it is to solve the
given differential equation. That said, the integrating factor method is an example of this method.
Although, we don’t usually think of linear ordinary differential equations as an exact equation, it
can be viewed as such20 .
A differential equation in x1 , x2 , . . . , xn of the form:
M1 dx1 + M2 dx2 + · · · + Mn dxn = 0

20
see my differential equations notes, it’s in there
10.11. INTRODUCTION TO GEOMETRIC DIFFERENTIAL EQUATIONS 279
can be written as dF = 0 locally iff
d(M1 dx1 + M2 dx2 + · · · + Mn dxn ) = 0.
The fact that exact implies closed is just d2 = 0. The converse direction, assuming closed near a
point, only gives the existence of a potential form F close to the point. Globally, there could be a
topological obstruction as we saw in the Poincare Lemma section.
Example 10.11.2. Problem: Suppose xdy + ydx − xdz = 0. Is there a point(s) in R3 near which
there exists F such that dF = xdy + ydx − xdz?
Solution: If there was the the differential form of the DEqn would vanish identically. However:
d(xdy + ydx − xdz) = dx ∧ dy + dy ∧ dx − dx ∧ dz = dz ∧ dx. Therefore, no.
We can try to find an integrating factor. Let’s give it a shot, this problem is simple enough it may
be possible to work it out. We want I such that I(xdy + ydx − xdz) is a closed one-form. Use the
Leibniz rule and our previous calculation:
d[I(xdy + ydx − xdz)] = dI ∧ (xdy + ydx − xdz) + Idz ∧ dx

= (Ix dx + Iy dy + Iz dz) ∧ (xdy + ydx − xdz) + Idz ∧ dx
= (xIx − yIy )dx ∧ dy + (xIx + yIz + I)dz ∧ dx + (−xIy − xIz )dy ∧ dz
Therefore, our integrating factor must satisfy the following partial differential equations,
(I.) xIx − yIy = 0, (II.) xIx + yIz + I = 0, (III.) − xIy − xIz = 0.
I’ll leave this to the reader. I’m not sure if it has a solution. It seems possible that the differential
consquences of this system are nonsensical. I just wanted to show how differential forms allow us
to extend to higher dimensional problems. Notice, we could just as well have not solved a problem
with 4 or 5 variables.
10.11.2 differential equations via forms

I follow Example 1.2.3. of Cartan for Beginners: Differential Geometry via Moving Frames and
Exterior Differential Systems by Thomas A. Ivey and J.M Landsberg. I suspect understanding this
text makes you quite a bit more than a beginner. Consider,
ux = A(x, y, u) (10.15)
uy = B(x, y, u) (10.16)
Chapter 11
Geometry by frames and forms
In this chapter we restrain ourselves to study three dimensional space and surfaces embedded within
that context. That said, once this chapter is mastered the next natural step to consider is the study
of geometric surfaces in which the ambient space is not used to frame the geometry of the surface.
The idea of this chapter is to take a guided tour through the second edition of Barret Oneil’s Ele-
mentary Differential Geometry. Obviously we do not cover the entire text. Instead, I have carefully
chosen a thread which allows us to see the central argument of the text for embedded surfaces in
three dimensional space.
To begin we review once more the concept of a vector field and we adopt the notation used by
Oneil1 . Frames in R3 are studied and the cartesian frame U1 , U2 , U3 is employed to define the
covariant derivative. This will be almost familar to anyone who has studied vector calculus in
non-cartesian coordinates. Next we develop the connection formulas for R3 which are based on
matrices of differential forms. Once the structure equations are settled we turn to the theory of
surfaces. We quickly introduce the shape operator as it is derived from the normal vector field of
a regular surface. Gauss and mean curvatures are defined. Principle curvature and umbilic points
are described. Isometry of surfaces is introduced and we see how isometric surfaces can be sur-
prisingly different. Finally, we turn to the problem of formulating the theory of surfaces in terms
of the connection form formalism of Cartan. After some fairly simple calculation we arrive at the
result that the shape operator and much of the ambient theory can be replaced with a few simple
formulas. From that point the proof of Gauss’ celebrated theorem on intrinsic curvature is simple.
It is certainly the case that some of the definitions in this chapter have been previously given
in greater generality. However, I intend this chapter to be self-contained as much as is possible.
Furthermore, whenever a definition seems unjust, you can read more background in Oneil when
time permits.
1
in this way I hope reading his text is a natural extension of this study for those interested
281
282 CHAPTER 11. GEOMETRY BY FRAMES AND FORMS
11.1 basic terminology and vector fields

We begin by describing the directional deirvative in R3 as a derivation: if v is a vector at p then

f (p + tv) − f (p)
vp [f ] = lim (11.1)
h→0 h
and for functions f smooth near p we have

3
X ∂f ∂ ∂ ∂
vp [f ] = vi (p) = v1 + v2 + v3 [f ]
∂xi ∂x ∂y ∂z p
i=1
Clearly vp acts on a function to reveal the rate of change of f at p as we allow p to move in the
hv1 , v2 , v3 i-direction. A vector 3 3
∂
field on R∂ is rule which attaches a vector to each point in R 3. For
∂
example, p 7→ ∂x p , p 7→ ∂y p and p 7→ ∂z p . Each of these define a global vector field on R and

together the triple of vector fields is known as the cartesian frame. A frame is set of three vector
fields which provides an orthonormal basis2 for Tp R3 at each p ∈ R3 . Given the importance of
the cartesian frame, we introduce a simple notation for future use:
Definition 11.1.1. notation for the cartesian frame
∂ ∂ ∂
U1 = , U2 = , U3 = .
∂x ∂y ∂z
The cartesian frame provides a convenient background for what follows.
Example 11.1.2. Let v = ha, b, ci then V = aU1 + bU2 + cU3 is a constant vector field on R3
where V (p) = v for all p ∈ R3 .
p
Example 11.1.3. If f (x, y, z) = x2 + y 2 + z 2 then ∇f = hx/f, y/f, z/f i hence ||(∇f )(p)|| = 1
for all p ∈ R3 . We can write this global gradient field as ∇f = √ 2 1 2 2 [xU1 + yU2 + zU3 ].
x +y +z
This vector field is part of the spherical frame field which we discuss in Example 11.1.10.
Vectors as derivations measure rates of change of functions as we move a point. The next logical
study is that of vector fields. How can we capture the change of a vector field W at a point p as we
move in the v-direction? This cannot be a number since a vector field has three components this
will again be a vector3
Definition 11.1.4. covariant differentiation of vector field

Let W be a vector field and vp a vector at p then the covariant derivative of W in the
direction of v at p is defined as follows: (∇v W )(p) = W 0 (p + tv)(0).
2
there is a generalization of this to arbitrary frames in the study of the frame-bundle, we leave that for another
course, all our frames are nice and orthonormal.
3
just like the change in a path in R3 is described by a velocity vector not just a single slope
11.1. BASIC TERMINOLOGY AND VECTOR FIELDS 283
Let the derivative in the definition above be understood as differentiation of each component. You
could see it as the directional derivative of each component in the v-direction. If W = w1 U1 +
w2 U2 + w3 U3 then (by definition)
W 0 (p + tv)(0) = w10 (p + tv)(0)U1 + w20 (p + tv)(0)U2 + w30 (p + tv)(0)U3
however, as we discussed in Equation 11.1 these directional derivatives are best understood in terms
of derivations. We tend to use the following as a method to calculate covariant derivatives:
Proposition 11.1.5.
If W is a vector field defined near p and W = 3i=1P

P
wi Ui the the covariant derivative of W
in the direction of v at p is given by: (∇v W )(p) = 3i=1 v[wi ]Ui .
Given two vector fields W, V we can study the change in W as it moves in the direction of V at
each point. In particular, at p ∈ R3 study how the vector field W evolves in the V (p) direction at
p. This construction suggests the covariant derivative converts a pair of vector fields into an new
vector field defined by the following:
(∇V W )(p) = (∇V (p) W )(p)
Let us pause to describe some useful properties of the covariant derivative. The proofs of these
follow from properties of partial differentiation and Proposition 11.1.5
Proposition 11.1.6. Properties of the covariant derivative
Suppose V, W, Y and Z are vector fields on R3 , f, g are functions and a, b ∈ R,
(a.) ∇f V +gW Y = f ∇V Y + g∇W Y
(b.) ∇V (aY + bZ) = a∇V Y + b∇V Z
(c.) ∇V (f Y ) = V [f ]Y + f ∇V Y
(d.) V [Y • Z] = (∇V Y ) • Z + Y • (∇V Z).
Proof:
P We examine part P(d.). Let Y, Z be vector fields with component functions Yi , Zi . We have
Y = 3i=1 Yi Ui and Z = 3i=1 Zi Ui . We calculate
3
!  3 
3 X3 3
X X X X
Y •Z = Yi Ui •  Zj Uj  = Yi Zj Ui • Uj = Yi Zi .
| {z }
i=1 j=1 i=1 j=1 i=1
δij
Suppose V be a vector field. By Proposition 11.1.5 we calculate:

" 3 # 3 3
X X X
V [Y • Z] = V Yi Z i = V [Yi Zi ] = V [Yi ] Zi + Yi V [Zi ] .
i=1 i=1 i=1
Then, note (∇V Y )i = V [Yi ] and (∇V Z)i = V [Zi ] hence

3
X 3
X
V [Y • Z] = (∇V Y )i Zi + Yi (∇V Z)i = (∇V Y ) • Z + Y • (∇V Z).
i=1 i=1
If you understand this then you can write proofs for the other parts without much trouble .
Definition 11.1.7.
If a triple of vector fields E1 , E2 , E3 on R3 satisfy Ei • Ej = δij for all i, j then they define
a frame field on R3 .
If E1 , E2 , E3 is a frame field on R3 then at each point p ∈ R3 they provide an orthonormal basis
{E1 (p), E2 (p), E3 (p)} for Tp R3 . Therefore, a given frame field allows us to have cartesian-like
coordinates based at each point in R3 . In particular, we can calculate dot-products and vector
lengths with respect to frame field coordinates just as we do with cartesian coordinates. Moreover,
we can select frame field coordinate functions for vector fields via dot-products just as we could
select cartesian coordinates of vectors by dot-products. The choice to work with orthonormal frames
pays off big here. We’ll use this proposition multiple times in our future work.
Proposition 11.1.8.
Suppose E1 , E2 , E3 is a frame. If V is a vector field on R3 then the component
P3 functions of
V w.r.t. the E1 , E2 , E3 frame are given by V • Ej for 1 ≤ j ≤ 3; that is V = j=1 (V • Ej )Ej .
Proof: left to the reader. Or, see my linear algebra notes .
The method I used to find these frames was to calculate the coordinates derivatives by the push-
forward formulas4 we discussed in Subsection 10.3.1.
Example 11.1.9. For cylindrical coordinates x = r cos θ, x = r sin θ and z = z. The cylindrical
frame field is defined as follows:
E1 = cos θU1 + sin θU2
E2 = − sin θU1 + cos θU2
E3 = U3
Example 11.1.10. If spherical coordinates ρ, φ, θ are given by x = ρ cos θ sin φ , y = ρ sin θ sin φ
and z = ρ cos φ then the spherical frame field is given by:

F1 = sin φ cos θU1 + cos θU2 + cos φU3
F2 = − sin θU1 + cos θU2

F3 = cos φ cos θU1 + cos θU2 − sin φU3
4
although, to be honest, I used a better method to calculate the needed partial derivatives. The best way is to
compute the total differentials of each coordinate function and read off the derivatives there. Again, probably a
homework problem
11.2. CONNECTION FORMS ON R3 285
Observe that E1 and E2 of Example 11.1.9 appear here and F2 = E2 as can be expected since
geometrically they play identical roles.
11.2 connection forms on R3

Let E1 , E2 , E3 be a frame field and consider a point p ∈ R3 and a vector v at p. Let us calculate
the covariant derivatives of the frame field at p in the v-direction. Furthermore, we shall express
the covariant derivatives in terms of the E-frame in view of the technique of Proposition 11.1.8
∇v E1 = [(∇v E1 ) • E1 ] E1 + [(∇v E1 ) • E2 ] E2 + [(∇v E1 ) • E3 ] E3

∇v E2 = [(∇v E2 ) • E1 ] E1 + [(∇v E2 ) • E2 ] E2 + [(∇v E2 ) • E3 ] E3 (11.2)
∇v E3 = [(∇v E3 ) • E1 ] E1 + [(∇v E3 ) • E2 ] E2 + [(∇v E3 ) • E3 ] E3
Think about this: for each i, j the coefficient (∇v Ei ) • Ej represents a linear function in v ∈ Tp R3
which assigns a particular real number. In other words, v 7→ (∇v Ei ) • Ej is a one-form at p. If we
allow p to vary then we obtain a one form on R3 for each i, j.
Definition 11.2.1. matrix of connection forms of R3

Let 1 ≤ i, j ≤ 3 and define a one-form ωij by ωij (p) (v) = [(∇v Ei ) • Ej ] (p) for each vector
v at p ∈ R3 . Futhermore, the matrix of one-forms ωij will be denoted by ω.
This matrix of one-forms will be at the center stage of our thinking for most of the remainder of
this chapter. Perhaps the most important property is given next:
Proposition 11.2.2.
Let ωij be the connection forms w.r.t. frame E1 , E2 , E3 then ωij = −ωji for all i, j.
Proof: follows from orthonormality of the frame paired with properties of covariant differentiation.
In particular, fix i, j and let v ∈ Tp R3 . Calculate
v[Ei • Ej ] = [∇v Ei ] • Ej + Ei • [∇v Ej ] .
However, observe that Ei • Ej = δij is a constant function thus v[δij ] = 0. Hence [∇v Ei ] • Ej =
−Ei • [∇v Ej ]. Therefore, ωij (v) = −ωji (v) for arbitrary v and p. The proposition follows. .
The proposition above can be formulated5 as ω T = −ω. In other words, the connection matrix is
antisymmetric. This immediately implies that ω11 = ω22 = ω33 = 0 for any connection form ω.
The frame E1 , E2 , E3 is related to the cartesian frame U1 , U2 , U3 by the attitude matrix. It

turns out this matrix provides a convenient computational tool in what follows.
Definition 11.2.3. attitude matrix of E-frame
5
here Oneil uses the notation t ω for ω T
Let E1 , E2 , E3 be a frame field and let aij be the functions for which
3
X
Ei = aij Uj .
j=1
We say A = (aij ) is called the attitude matrix of the E1 , E2 , E3 frame.

The definition above makes the i-th row of the attitude matrix the components of Ei .
Example 11.2.4. (continuing Ex. 11.1.9) We may calculate the attitude matrix for the cylindrical
frame field by inspection: (zeros and one added for clarity of A-construction)
 
E1 = cos θU1 + sin θU2 + 0U3 cos θ sin θ 0
E2 = − sin θU1 + cos θU2 + 0U3 ⇒ A =  − sin θ cos θ 0 
E3 = 0U1 + 0U2 + 1U3 0 0 1
I invite the reader to verify AAT = I.
Example 11.2.5. (continuing Ex. 11.1.10) The spherical frame F1 , F2 , F3 has attitude matrix:
 
F1 = sin φ cos θU1 + sin θU2 + cos φU3 sin φ cos θ sin φ sin θ cos φ
F2 = − sin θU1 + cos θU2 ⇒ A =  − sin θ cos θ 0 
F3 = cos φ cos θU1 + sin θU2 − sin φU3 cos φ cos θ cos φ sin θ − sin φ
Once more, I invite the reader to verify that AAT = I.
The attitude matrix is a matrix of functions. Notice we can construct a matrix of forms from A by
taking the differential of each component. The matrix dA is then a matrix of one-forms, much like
the connection matrix ω. That said, we should pause to define the operations we need to connect
dA and ω explicitly.
When working with a matrix of one-forms the common notation in such a context is that matrix
multiplication proceeds normally except that the component-wise multiplication is understood to
be wedge products. We also calculate exterior derivatives of matrix-valued forms. The rule is
simply to calculate the exterior derivative of each entry. In summary, if A is an l × m matrix of
p-forms and B is an m×n matrix of q-forms then AB is defined to be a l ×n matrix of (p+q)-forms:
m
X
(AB)ij = Aik ∧ Bkj
k=1
moreover, dA is a l × m matrix of p-forms where we define (dA)ij = dAij . Now we have all the
terminology needed to explicitly connect the exterior derivative of the attitude matrix and the
connection form:
Proposition 11.2.6.
Let A = (aij ) be the attitude matrix for the frame field E1 , E2 , E3 then the connection
matrix ω is given by: ω = dAAT .
Proof: observe ω = dAAT indicates that ω(V ) = dA(V )AT for all vector fields V . In particular,
for each i, j we should show ωij (V ) = (dA(V )AT )ij . Let us begin:
ωij (V ) = (∇V Ei ) • Ej
3
!
X
= ∇V aik Uk • Ej
k=1
" 3 # " 3 #
X X
= V [aik ]Uk • ajl Ul
k=1 l=1
3
XX 3
= V [aik ]ajl Uk • Ul
k=1 l=1
X3
= daik [V ]ajk
k=1
X3
= (dA[V ])ik (AT )kj
k=1
= dA[V ]AT

ij
.
Therefore, as V and i, j were arbitrary the proposition follows .
Example 11.2.7. (continuing Ex. 11.2.4 and Ex. 11.1.9) we calculate the connection form for
the cylindrical frame via Proposition 11.2.6.
    
− sin θdθ cos θdθ 0 cos θ − sin θ 0 0 dθ 0
ω = dAAT =  − cos θdθ − sin θdθ 0   sin θ cos θ 0  =  −dθ 0 0 
0 0 0 0 0 1 0 0 0
Example 11.2.8. (continuing Ex. 11.2.5 and Ex. 11.1.10) we  calculate the connection form for
sin φ cos θ sin φ sin θ cos φ
the spherical frame via Proposition 11.2.6. To begin, recall A =  − sin θ cos θ 0 
cos φ cos θ cos φ sin θ − sin φ
and take the exterior derivative: (the lines are just for convenience)
 
cos φ cos θdφ − sin φ sin θdθ cos φ sin θdφ + sin φ cos θdθ − sin φdφ
dA =  − cos θdθ − sin θdθ 0 
− sin φ cos θdφ − cos φ sin θdθ − sin φ sin θdφ + cos φ cos θdθ − cos φdφ
 
sin φ cos θ − sin θ cos φ cos θ
The transpose is simply AT =  sin φ sin θ cos θ cos φ sin θ . Now, multiply. I’ll just consider
cos φ 0 − sin φ
the interesting terms ω12 , ω13 , ω23 from ω = dAAT . I leave the details to the reader6
 
0 cos φdθ dφ
ω =  − cos φdθ 0 sin φdφ  .
−dφ − sin φdφ 0
We now return to the task of covariant differentiation. This result is important because it relates
the change in the frame field E1 , E2 , E3 in terms of the frame field E1 , E2 , E3 . In some sense, this
is like the Frenet Serret equations from calculus III.
Proposition 11.2.9. Connection equations of frame field E1 , E2 , E3 on R3
Let ωij be the connection forms with respect to the frame field E1 , E2 , E3 on R3 for any
vector field V ,
X3
∇ V Ei = ωij (V )Ej
j=1
P3
Proof: Suppose Ei = j=1 aij Uj where aij is the attitude matrix of an orthonormal frame
E1 , E2 , E3 . Note this implies A = (aij ) is orthonormal T
P3 (A A = I) and so the inverse relation
is given by the transpose and we may express: Ui = j=1 aji Ej .
 
X3
∇ V Ei = ∇ V  aij Uj 
j=1
3
X
= ∇V [aij Uj ]
j=1
3
X
= [V [aij ]Uj + aij ∇V Uj ]
j=1
3
X 3
X
= V [aij ] akj Ek
j=1 k=1
3 X
X 3 3
X 3
X
= daij [V ]akj Ek = (dA[V ]AT )ik Ek = ωik [V ]Ek .
k=1 j=1 k=1 k=1
It is worth expanding these for future reference:
∇V E1 = ω12 [V ]E2 + ω13 [V ]E3

∇V E2 = −ω12 [V ]E1 + ω23 [V ]E3
∇V E3 = −ω13 [V ]E1 − ω23 [V ]E2
6
do not attempt this unless you plan to work carefully, it’s not hard, but it does require attention to detail
Compare these against the Frenet Serret Equations: (assume curve is arclength parametrized)
dT
= κN
dt
dN
= −κT + τ B
dt
dB
= −τ N.
dt
11.2.1 Frenet Serret equations

Let’s pause to see how the Frenet Serret equations are derived from the connection equations for a
frame field. Note that the Frenet Frame on a curve can be extended beyond the curve relative to
some open tubular neighborhood for a regular curve. This is geometrically evident, but I’d rather
not prove it for our purposes here. The derivative with respect to arclength is naturally identified
with ∇T . Let the frame adapted to the curve be given by the Frenet frame in the P usual order:
E1 = T , E2 = N and E3 = B In particular, again using t for arclength, let V = 3i=1 vi Ui be a
vector field along the curve,
dV
∇T V = T (v1 )U1 + T (v2 )U2 + T (v3 )U3 =
dt
ddx ∂ dy ∂ dz ∂ ∂
Here we understand, on the curve, the dt =
dt ∂x + dt ∂y + dt ∂z and U1 , U2 , U3 we recall ∂x = U1 ,
∂ ∂ P 3 dEi •
∂y = U2 and ∂z = U3 . Also, observe: ∇T Ei = j=1 ωij [T ]Ej hence ωij [T ] = dt Ej . But, the
defintion of curvature κ and can be formulated as T 0 = κN hence dT dT •
dt N = κ and dt B = 0.
•
Moreover, torsion τ is defined by dB

dt N = −τ . Therefore, ω12 [T ] = κ, ω13 [T ] = 0 and ω23 [T ] = τ
•
and we see how the connection equations imply the Frenet Serret equations along a curve.
dT
∇T E1 = ω12 [T ]E2 + ω13 [T ]E3 ⇒ = κN
dt
dN
∇T E2 = −ω12 [T ]E1 + ω23 [T ]E3 ⇒ = −κT + τ B
dt
dB
∇T E3 = −ω13 [T ]E1 − ω23 [T ]E2 ⇒ = −τ N.
dt
11.3 structure equations for frame field on R3

∂ ∂ ∂
The cartesian frame U1 , U2 , U3 was defined by ∂x = U1 , ∂y = U2 and ∂z = U3 hence the natural
dual to the cartesian frame is dx, dy, dz. Remember we proved dxi (∂j ) = δij in much greater
generality in a previous chapter.
Definition 11.3.1. coframe dual to a given frame
Let E1 , E2 , E3 be a frame field then coframe field dual to E1 , E2 , E3 is θ1 , θ2 , θ3 which
is a triple of one-forms such that
θi (V ) = V • Ei
for all vector fields V . Naturally, θi (Ej ) = δij .
The term one-form in in some sense unfortunate since what might be a better term is one-form
field. In particular, notice the coframe θ1 , θ2 , θ3 provides a basis {θ1 (p), θ2 (p), θ3 (p)} for (Tp R3 )∗
at each p ∈ R3 . We often suppress the point dependence in what follows.
P3
Suppose γ is a one-form on R3 then γ = j=1 γj θj for some functions γj . Furthermore,
3
X 3
X
γ(Ei ) = γj θj (Ei ) = γj δij = γi .
j=1 j=1
This discussion proves the following analog to Proposition 11.1.8.

Proposition 11.3.2.
Suppose E1 , E2 , E3 is a frame with coframe θ1 , θ2 , θ3 on R3 . If γ is a one-form on R3 then

3
X
γ= γ(Ei )θi .
i=1
Naturally we are curious how an arbitrary coframe relates to the cartesian coframe dx, dy, dz. It
turns out that the frame and coframe share the same attitude.
Proposition 11.3.3.
Suppose E1 , EP
2 , E3 is a frame with coframe θ1 , θ2 , θ3 on R3 . If A is the attitude matrix,
3 P3
meaning Ei = j=1 aij Uj , then θi = j=1 aij dxj .
Proof: Here we use the natural convention x1 = x, x2 = y, x3 = z in the proposition above. The
proof of this proposition filters through the duality condition θi (Ej ) = δij since dxi (UP
j ) = δij and
δij is common ground. In particular, by orthonormality we have inverse relations Uj = 3k=1 akj Ek
3 3 3
!
X X X
θi (Uj ) = θi akj Ek = akj θi (Ek ) = akj δik = aij .
k=1 k=1 k=1
11.3. STRUCTURE EQUATIONS FOR FRAME FIELD ON R3 291
P3
Thus, by Proposition 11.3.3, θi = j=1 aij dxj .
A brief history: classification of curves by curvature and torsion was elegantly treated by Frenet
and Serret independently around 1860. The work on frames attached to curves then prompted
Darboux to try to generalize the method of frames to surfaces. Then the general method of frames
was promoted and championed by E. Cartan around 1900. Research continues to this day.
Theorem 11.3.4. Cartan’s Structural Equations
Suppose E1 , E2 , E3 is a frame with coframe θ1 , θ2 , θ3 on R3 . If ω is the connection form of
the given frame then the following structure equations hold for all i, j
3
X 3
X
(1.) dθi = ωij ∧ θi (2.) dωij = ωik ∧ ωkj .
j=1 k=1
Proof: It is convenient to reformulate these equations in a matrix/column notation. In particular,

   
θ1 dx
θ =  θ2  dξ =  dy 
θ3 dz
P3
The equation θi = j=1 aij dxj gives θ = Adξ. Observe that d2 = 0 is true for matrices of forms
since the exterior differentiation is done component-wise. Thus, d(dξ) = 0. Hence,
dθ = d(Adξ) = dA ∧ dξ + Ad(dξ) = dA dξ = dAAT Adξ.
where we use AT A = I for the last step. Next, apply Proposition 11.2.6 and θ = Adξ to see:
dθ = dAAT Adξ = ωθ.
The equation above is just (1.) in matrix notation. Next consider the exterior derivatve of7
ω = dAAT , again use d2 = 0 and the Leibniz rule,
dω = d(dAAT ) = d2 (A)AT − dAd(AT ) = − dAA T
| {z } AdA
T
| {z }
ω −ω
To see why AdAT = −ω, use the socks-shoes property of the matrix transpose to see that AdAT =
(dAAT )T = ω T . But, by antisymmetry of the connection form says ω T = −ω. We’ve shown
dω = −ω(−ω) = ωω
this is precisely the second struture equation written in matrix notation .
This is one of my top ten favorite favorite theorems. However, to appreciate why I would enjoy
such a strange set of formulas we probably need to spend an hour or two discussing the structure
of a regular surface and a bit on diffeomorphisms. Once those issues are addressed we’ll return and
develop an intrinsic calculus via these equations.
7
note that d(AT ) = (dA)T hence dAT is non an ambiguous notation.
11.4 surfaces in R3
A surface M in R3 is a subset which locally looks like the plane. Moreover, we suppose the surface
is oriented. The orientation is given by a unit-normal vector field which is defined to point in the
upward direction. Let us review a few technical details to make this paragraph a bit more precise.
A surface M ⊂ R3 is called regular if it can be covered by compatible regular patches. A patch

X : D ⊆ R2 → M is regular iff ∂u X × ∂v X 6= 0 for all (u, v) ∈ D. If the vector field ∂u X × ∂v X 6= 0
1
then we may normalize it to form a unit-normal field U = ||∂u X×∂ v X||
∂u X × ∂v X. Furthermore,
if the surface is oriented then the normal vector field on each patch may be smoothly joined to
adjacent patches as to construct a global unit normal field. Throughout this section we use the
notation that U is the unit-normal to the surface considered. It is geometrically clear that U (p) is
orthogonal to the tangent space at p ∈ M . It is analytically clear since
Tp M = span{∂u |p , ∂v |p }
and U (p) is formed from the cross product of these tangent vectors. Of course, we can expand the
vectors in the ambient frame U1 , U2 , U3 if necessary,
∂ ∂x ∂y ∂z ∂ ∂x ∂y ∂z
= U1 + U2 + U3 , = U1 + U2 + U3
∂u ∂u ∂u ∂u ∂v ∂v ∂v ∂v
The shape of the surface is revealed from the covariant derivatives of the normal vector field. It
can be shown that if v is tangent to M then ∇v U is also tangent to M . It follows the definition
below is reasonable:
Definition 11.4.1. shape operator
Let U be the unit-normal vector field to a sufrace M ⊂ R3 . We define a mapping Sp :

Tp M → Tp M by
Sp (v) = −∇v U
for all v ∈ Tp M . This is the shape operator at p.
Naturally, the mapping p 7→ Sp is also called the shape operator. Geometrically, the value of the
shape operator reflects how the normal vector field changes as p moves in the direction of v near p.
Example 11.4.2. Plane: The shape operator on a plane is zero since U is constant.
Example 11.4.3. Sphere: of radius R centered at the origin has U = R1 (xU1 + yU2 + zU3 )
therefore ∇v U = R1 (v[x]U1 + v[y]U2 + v[z]U3 ) = v/R. This equation expresses the geometrically
obvious fact that the normal vector field on a sphere bends in the direction of v as we move from p
in the v-direction. Thus Sp (v) = −v/R.
The sphere example is sufficiently simple in cartesian coordinates that we did not need to do much
calculation. I didn’t even provide a patch for the sphere, we reasoned geometrically. To understand
what follows, please review the cylindrical and spherical frames. We introduced them in Examples
11.1.9 and 11.1.10.
11.4. SURFACES IN R3 293
Example 11.4.4. Cylinder: of radius R with equation x2 + y 2 = R2 has natural parametrization

by θ, z for which x = R cos θ, y = R sin θ and z = z. The cylindrical frame field then provides
vector fields tangent to the cylinder. In particular,
E2 = − sin θU1 + cos θU2 , E3 = U3
are clearly tangent to the cylinder. On the other hand, it’s geometrically clear that the normal
vector field U = E1 = cos θU1 + sin θU2 . Calculate the covariant derivatives in view of the fact
that U = R1 (xU1 + yU2 ),
1 1
∇v U = [v[x]U1 + v[y]U2 ] = (vx U1 + vy U2 ) (11.3)
R R
where I denote v = vx ∂x + vy ∂y + vz ∂z .
I now work on converting the formula into the cylindrical frame. We can invert the equations for
E1 = cos θU1 + sin θU2 and E2 = − sin θU1 + cos θU2 for U1 and U2 . We obtain8 :
U1 = cos θE1 − sin θE2 , U2 = sin θE1 + cos θE2 . (11.4)
Consider v = E2 = − sin θU1 + cos θU2 has vx = − sin θ and vy = cos θ thus:
1
∇E 2 U = (− sin θU1 + cos θU2 ) (by Eqn. 11.3)
R
1
= − sin θ(cos θE1 − sin θE2 ) + cos θ(sin θE1 + cos θE2 ) (by Eqn. 11.4)
R
1
= E1 .
R
1
In summary, if v = aE2 + bE3 where E2 = R ∂θ and E3 = ∂z then we find:
a
Sp (v) = −∇v U = − E1 .
R

−1/R 0
The matrix of this shape operator has the form: [Sp ] = . The shape operator reflects
0 0
our intuition that the cylinder is curved in the θ-direction whereas in the z-direction it is flat.
I will use some calculus which is developed in my calculus III lecture notes in what follows. In
those notes I show how ρb, θb and φb form the spherical frame. The difference between that context
and our current one is that the vector fields were viewed as passive objects. In calculus III we
did not think of vector fields as derivations. That said, all the algebra/geometric formulas derived
for the passive frame hold for the derivation-based frame we consider here. In particular, when I
b p is the normal vector field at each p 6= 0
consider the cone below, it is geometrically obvious that φ|
8
to see U1 , multiply E1 by cos θ and E2 by − sin θ the terms attached to U2 cancel. To see U2 , multiply E1 by
sin θ and E2 by cos θ the terms attached to U1 cancel
b p , ρb|p form an
for the cone M defined by φ = φo . Orthonormality immediately informs be that φ|
orthonormal basis for Tp M at each p 6= 0 on the cone M .
Recall for what follows that: (here 7→ denotes the transition from calculus III notation to our
current derivation-based formalism for vectors)

ρb 7→ F1 = sin φ cos θU1 + sin θU2 + cos φU3
θb 7→ F2 = − sin θU1 + cos θU2

φb 7→ F3 = cos φ cos θU1 + sin θU2 − sin φU3
Example 11.4.5. Cone at angle φo . We consider points other than the origin, the cone is not
smooth at that singular point. To be more accurate, the results we derive in this example apply to
some open subset of the cone. In cylindrical coordinates r = ρ sin(φo ) thus the cartesian equation
of this cone is easily derived from r2 = ρ2 sin2 (φo ) gives x2 + y 2 = sin2 (φo )(x2 + y 2 + z 2 ) hence,
for φo 6= π/2, we find x2 + y 2 = tan2 (φo )z 2 . In cylindrical coordinates this cone has equation
r = tan(φo )z. From spherical coordinates we find a natural parametrization,
x = ρ cos(θ) sin(φo ),y = ρ sin(θ) sin(φo ), z = ρ cos(φo )

Furthermore, this gives U = F3 = cos φ cos θU1 + sin θU2 − sin φU3 . Consider that:
∇v F3 = v[cos φ cos θ]U1 + v[cos φ sin θ]U2 + v[sin φ]U3
On the cone φ = φo we have sin φo is constant thus the U3 -term vanishes. In what follows we
consider tangent vectors on the cone, hence φ = φo throughout. If v = ∂ρ :
∇∂ρ U = ∂ρ [cos φo cos θ]U1 + ∂ρ [cos φo sin θ]U2 = 0
Note F1 = ∂ρ hence ∇F1 U = 0. Continuing, let v = ∂θ and consider:
∇∂θ U = ∂θ [cos φo cos θ]U1 + ∂θ [cos φo sin θ]U2 = cos φo (− sin θU1 + cos θU2 ) = cos φo F2
∂ 1 ∂
However, as ∂θ = −ρ sin φ sin θU1 + ρ sin φ cos θU2 we see F2 = ρ sin φo ∂θ . Thus, by part (a.) of
Proposition 11.1.6,
cos φo
∇F2 U = F2
ρ sin φo
In summary, if v = aF1 + bF2 then
b cos φo
Sp (v) = F2
ρ sin φo
" #
0 0
The matrix of this shape operator has the form: [Sp ] = . Observe that if we hold
0 − ρcos φo
sin φo
θ fixed on the cone we trace out a line as ρ varies. The line is not curved and thus the normal
11.4. SURFACES IN R3 295
is transported parallel to such lines as is reflected in the triviality of the covariant derivative in
the F1 -direction on the cone. On the other hand, if we fix ρ and vary θ on the cone then we
trace out a circle of radius ρ sin φo . The component of the normal which is radial to this circle is
cos φo U . Notice how that component (cos φo U ) of the normal bends in the direction of F2 at a rate
proportional to the reciprocal of the radius (ρ sin φo ) of the circle. This is the same as we saw in
the example of the cylinder. In contrast to the cylinder, we have a family of surfaces to consider
here, as φ varies we have distinct cones. Consider that as φ → π/2 we find Sp = 0. This verifies,
modulo the deleted point at the origin, that the plane does have a trivial shape operator.
Definition 11.4.6.
If S is the shape operator on a regular surface M then the Gaussian curvature K and the
mean curvature H are defined by K = det(S) and H = trace(S).
The eigenvalues of a 2 × 2 matrix correspond uniquely to the trace and determinant of the matrix.
They also have geometric meaning here. Principle curvatures are defined in the theorem:
Theorem 11.4.7.
The shape operator is symmetric thus there exist real eigenvalues and an orthonormal
eigenbasis for Tp M with respect to S at each p ∈ M . The orthonormal eigenvectors de-
fine the principle directions and the corresponding eigenvalues are called the principle
curvatures. If the principle curvatures are k1 , k2 then K = k1 k2 and H = k1 + k2 .
Recall, eigenvalues of a diagonal matrix are simply the diagonal entries.
Example 11.4.8. Plane: the principle curvatures are k1 = k2 = 0. It follows that the mean
curvature H = 0 and the Gaussian curvature K = 0 on the entire plane.
Example 11.4.9. Sphere: the principle curvatures are k1 = k2 = −1/r. It follows that the mean
curvature H = −2/r and the Gaussian curvature K = 1/r2 on the entire sphere.
Example 11.4.10. Cylinder: the principle curvatures are k1 = −1/r and k2 = 0. It follows that
the mean curvature H = −1/r and the Gaussian curvature K = 0 on the entire clylinder.
Example 11.4.11. Cone: (without the point) the principle curvatures are k1 = 0 and k2 =
− ρcos φo cos φo
sin φo . It follows that the mean curvature H = − ρ sin φo and the Gaussian curvature K = 0 on
the entire cone. Here we have an example of a surface which has non-constant mean curvature
whilst9 having constant Gaussian curvature.
Surfaces with constant Gaussian curvature play special roles in many applications. All the examples
above have constant gaussian curvature. As a point of language, we usually intend the unqualified
term ”curvature” to mean Gaussian curvature. It has a deeper meaning than the mean curvature.
For example, any surface with constant Gaussian curvature K = 0 is called flat because it is essen-
tially a plane from an intrinsic point of view. In contrast, the principle and mean curvatures are
9
yes there is doubtless a better way to write this sentence
not intrinsic; the principle and mean curvatures have to do with how the surface is embedded in
some ambient space. Here I’ve touched on an idea you may not have encountered before10 . What
do we mean by intrinsic? One of my goals in the remainder of this chapter is to unwrap that idea.
We’ll not do it here just yet.
Another important concept to the study of surface geometry is the umbilic point. A point is said
to be an umbilic point if the principle curvatures are equal at the point in question. If every
point on the surface is umbilic then the surface is called, shockingly, an umbilic surface. Planes
and spheres are umbilic surfaces. Are there others? We’ll discuss this in the final section of this
chapter where I present a number of theorems without proof11 .
11.5 isometry
10
if you’ve seen Flatland then you may have at least a fictional concept of it
11
most of the proofs are found in Oneil
Chapter 12
Electromagnetism in differential form
Warning: I will use Einstein’s implicit summation convention throughout this section.
I have made a point of abstaining from Einstein’s convention in these notes up to this point.
However, I just can’t bear the summations in this section. They’re just too ugly.
12.1 differential forms in Minkowski space

The logic here follows fairly close to the last section, however the wrinkle is that the metric here
demands more attention. We must take care to raise the indices on the forms when we Hodge dual
them. First we list the basis differential forms, we have to add time to the mix ( again c = 1 so
x0 = ct = t if you worried about it )
Name Degree Typical Element ”Basis” for Λp (R4 )

function p=0 f 1
one-form p=1 α = αµ dxµ dt, dx, dy, dz
two-form p=2 β = 12 βµν dxµ ∧ dxν dy ∧ dz, dz ∧ dx, dx ∧ dy
dt ∧ dx, dt ∧ dy, dt ∧ dz
1 µ
three-form p=3 γ= 3! γµνα dx ∧ dxν ∧ dxα dx ∧ dy ∧ dz, dt ∧ dy ∧ dz
dt ∧ dx ∧ dz, dt ∧ dx ∧ dy
four-form p=4 gdt ∧ dx ∧ dy ∧ dz dt ∧ dx ∧ dy ∧ dz
Greek indices are defined to range over 0, 1, 2, 3. Here the top form is degree four since in four
dimensions we can have four differentials without a repeat. Wedge products work the same as they
have before, just now we have dt to play with. Hodge duality may offer some surprises though.
Definition 12.1.1. The antisymmetric symbol in flat R4 is denoted µναβ and it is defined by the
value
0123 = 1
plus the demand that it be completely antisymmetric.
297
298 CHAPTER 12. ELECTROMAGNETISM IN DIFFERENTIAL FORM
We must not assume that this symbol is invariant under a cyclic exhange of indices. Consider,
0123 = −1023 flipped (01)
= +1203 flipped (02) (12.1)
= −1230 flipped (03).
Example 12.1.2. We now compute the Hodge dual of γ = dx with respect to the Minkowski metric
ηµν . First notice that dx has components γµ = δµ1 as is readily verified by the equation dx = δµ1 dxµ .
We raise the index using η, as follows
γ µ = η µν γν = η µν δν1 = η 1µ = δ 1µ .
Beginning with the definition of the Hodge dual we calculate

∗ (dx) 1
= (4−1)! δ 1µ µναβ dxν ∧ dxα ∧ dxβ
= (1/6)1ναβ dxν ∧ dxα ∧ dxβ
= (1/6)[1023 dt ∧ dy ∧ dz + 1230 dy ∧ dz ∧ dt + 1302 dz ∧ dt ∧ dy

+1320 dz ∧ dy ∧ dt + 1203 dy ∧ dt ∧ dz + 1032 dt ∧ dz ∧ dy]
(12.2)
= (1/6)[−dt ∧ dy ∧ dz − dy ∧ dz ∧ dt − dz ∧ dt ∧ dy
+dz ∧ dy ∧ dt + dy ∧ dt ∧ dz + dt ∧ dz ∧ dy]
= −dy ∧ dz ∧ dt.
The difference between the three and four dimensional Hodge dual arises from two sources, for
one we are using the Minkowski metric so indices up or down makes a difference, and second the
antisymmetric symbol has more possibilities than before because the Greek indices take four values.
Example 12.1.3. We find the Hodge dual of γ = dt with respect to the Minkowski metric ηµν .
Notice that dt has components γµ = δµ0 as is easily seen using the equation dt = δµ0 dxµ . Raising the
index using η as usual, we have
γ µ = η µν γν = η µν δν0 = −η 0µ = −δ 0µ
where the minus sign is due to the Minkowski metric. Starting with the definition of Hodge duality
we calculate
∗ (dt) = −(1/6)δ 0µ ν α β
µναβ dx ∧ dx ∧ dx
= −(1/6)0ναβ dxν ∧ dxα ∧ dxβ
(12.3)
= −(1/6)0ijk dxi ∧ dxj ∧ dxk
= −(1/6)ijk dxi ∧ dxj ∧ dxk
= −dx ∧ dy ∧ dz.
for the case here we are able to use some of our old three dimensional ideas. The Hodge dual of dt
cannot have a dt in it which means our answer will only have dx, dy, dz in it and that is why we
were able to shortcut some of the work, (compared to the previous example).
12.1. DIFFERENTIAL FORMS IN MINKOWSKI SPACE 299
Example 12.1.4. Finally, we find the Hodge dual of γ = dt∧dx with respect to the Minkowski met-
ric ηµν . Recall that ∗ (dt ∧ dx) = (4−2)!
1
01µν γ 01 (dxµ ∧ dxν ) and that γ 01 = η 0λ η 1ρ γλρ = (−1)(1)γ01 =
−1. Thus
∗ (dt ∧ dx) = −(1/2) µ ν
01µν dx ∧ dx
= −(1/2)[0123 dy ∧ dz + 0132 dz ∧ dy]
(12.4)
= −dy ∧ dz.
Notice also that since dt ∧ dx = −dx ∧ dt we find ∗(dx ∧ dt) = dy ∧ dz
The other Hodge duals of the basic two-forms follow from similar calculations. Here is a table of
all the basic Hodge dualities in Minkowski space, In the table the terms are grouped as they are to
∗1 = dt ∧ dx ∧ dy ∧ dz ∗ (dt ∧ dx ∧ dy ∧ dz) = −1
∗ (dx ∧ dy ∧ dz) = −dt ∗ dt = −dx ∧ dy ∧ dz
∗ (dt ∧ dy ∧ dz) = −dx ∗ dx = −dy ∧ dz ∧ dt
∗ (dt ∧ dz ∧ dx) = −dy ∗ dy = −dz ∧ dx ∧ dt
∗ (dt ∧ dx ∧ dy) = −dz ∗ dz = −dx ∧ dy ∧ dt
∗ (dz ∧ dt) = dx ∧ dy ∗ (dx ∧ dy) = −dz ∧ dt
∗ (dx ∧ dt) = dy ∧ dz ∗ (dy ∧ dz) = −dx ∧ dt
∗ (dy ∧ dt) = dz ∧ dx ∗ (dz ∧ dx) = −dy ∧ dt
emphasize the isomorphisms between the one-dimensional Λ0 (M ) and Λ4 (M ), the four-dimensional

Λ1 (M ) and Λ3 (M ), the six-dimensional Λ2 (M ) and itself. Notice that the dimension of Λ(M ) is
16 which just happens to be 24 .
Now that we’ve established how the Hodge dual works on the differentials we can easily take the
Hodge dual of arbitrary differential forms on Minkowski space. We begin with the example of the
4-current J
Example 12.1.5. Four Current: often in relativistic physics we would even just call the four
current simply the current, however it actually includes the charge density ρ and current density
~ Consequently, we define,
J.
~
(J µ ) ≡ (ρ, J),
moreover if we lower the index we obtain,
~
(Jµ ) = (−ρ, J)
which are the components of the current one-form,
J = Jµ dxµ = −ρdt + Jx dx + Jy dy + Jz dz
This equation could be taken as the definition of the current as it is equivalent to the vector defini-
tion. Now we can rewrite the last equation using the vectors 7→ forms mapping as,
J = −ρdt + ωJ~.
Consider the Hodge dual of J ,

∗J = ∗ (−ρdt + Jx dx + Jy dy + Jz dz)
= −ρ∗ dt + Jx ∗ dx + Jy ∗ dy + Jz ∗ dz
(12.5)
= ρdx ∧ dy ∧ dz − Jx dy ∧ dz ∧ dt − Jy dz ∧ dx ∧ dt − Jz dx ∧ dy ∧ dt
= ρdx ∧ dy ∧ dz − ΦJ~ ∧ dt.
we will find it useful to appeal to this calculation in a later section.
Example 12.1.6. Four Potential: often in relativistic physics we would call the four potential
~
simply the potential, however it actually includes the scalar potential V and the vector potential A
(discussed at the end of chapter 3). To be precise we define,
~
(Aµ ) ≡ (V, A)
we can lower the index to obtain,

~
(Aµ ) = (−V, A)
which are the components of the current one-form,
A = Aµ dxµ = −V dt + Ax dx + Ay dy + Az dz
Sometimes this equation is taken as the definition of the four potential. We can rewrite the four
potential vector field using the vectors 7→ forms mapping as,
A = −V dt + ωA~ .
The Hodge dual of A is

∗
A = V dx ∧ dy ∧ dz − ΦA~ ∧ dt. (12.6)
Several steps were omitted because they are identical to the calculation of the dual of the 4-current
above.
Definition 12.1.7. Faraday tensor.

~ = (E1 , E2 , E3 ) and a magnetic field B
Given an electric field E ~ = (B1 , B2 , B3 ) we define a
2-form F by
F = ωE ∧ dt + ΦB .
This 2-form is often called the electromagnetic field tensor or the Faraday tensor. If
we write it in tensor components as F = 12 Fµν dxµ ∧ dxν and then consider its matrix (Fµν )
of components then it is easy to see that
 
0 −E1 −E2 −E3
E1 0 B3 −B2 
(Fµν ) = 
E2 −B3
 (12.7)
0 B1 
E3 B2 −B1 0
12.1. DIFFERENTIAL FORMS IN MINKOWSKI SPACE 301
Convention: Notice that when we write the matrix version of the tensor components we take the
first index to be the row index and the second index to be the column index, that means F01 = −E1
whereas F10 = E1 .
Example 12.1.8. In this example we demonstrate various conventions which show how one can
transform the field tensor to other type tensors. Define a type (1, 1) tensor by raising the first index
by the inverse metric η αµ as follows,
F α ν = η αµ Fµν
The zeroth row,
(F 0 ν ) = (η 0µ Fµν ) = (0, E1 , E2 , E3 )
Then row one is unchanged since η 1µ = δ 1µ ,
(F 1 ν ) = (η 1µ Fµν ) = (E1 , 0, B3 , −B2 )
and likewise for rows two and three. In summary the (1,1) tensor F 0 = Fνα ( ∂x∂α ⊗ dxν ) has the
components below  
0 E1 E2 E3
E1 0 B3 −B2 
(F α ν ) = 
E2 −B3
. (12.8)
0 B1 
E3 B2 −B1 0
At this point we raise the other index to create a (2, 0) tensor,
F αβ = η αµ η βν Fµν (12.9)
and we see that it takes one copy of the inverse metric to raise each index and F αβ = η βν F α ν so
we can pick up where we left off in the (1, 1) case. We could proceed case by case like we did with
the (1, 1) case but it is better to use matrix multiplication. Notice that η βν F α ν = F α ν η νβ is just
the (α, β) component of the following matrix product,
    
0 E1 E2 E3 −1 0 0 0 0 E1 E2 E3
E1 0 B3 −B2    0 1 0 0 = −E1 0 B3 −B2 
(F αβ ) = 
  
 . (12.10)
E2 −B3 0 B1   0 0 1 0   −E2 −B3 0 B1 
E3 B2 −B1 0 0 0 0 1 −E3 B2 −B1 0
So we find a (2, 0) tensor F 00 = F αβ ( ∂x∂α ⊗ ∂x∂ β ). Other books might even use the same symbol F for
F 0 and F 00 , it is in fact typically clear from the context which version of F one is thinking about.
Pragmatically physicists just write the components so its not even an issue for them.
Example 12.1.9. Field tensor’s dual: We now calculate the Hodge dual of the field tensor,
∗F = ∗ (ωE ∧ dt + ΦB )
= Ex ∗ (dx ∧ dt) + Ey ∗ (dy ∧ dt) + Ez ∗ (dz ∧ dt)
+Bx ∗ (dy ∧ dz) + By ∗ (dz ∧ dx) + Bz ∗ (dx ∧ dy)
= Ex dy ∧ dz + Ey dz ∧ dx + Ez dx ∧ dy
−Bx dx ∧ dt − By dy ∧ dt − Bz dz ∧ dt
= ΦE − ωB ∧ dt.
we can also write the components of ∗ F in matrix form:

 
0 B1 B2 B3
−B1 0 E3 −E2 
(∗ Fµν ) = 
−B2 −E3
 (12.11)
0 E1 
−B3 E2 −E1 0.
~ →
Notice that the net-effect of Hodge duality on the field tensor was to make the exchanges E ~
7 −B
and B~ 7→ E.
~
12.2 exterior derivatives of charge forms, field tensors, and their

duals
In the last chapter we found that the single operation of the exterior differentiation reproduces the
gradiant, curl and divergence of vector calculus provided we make the appropriate identifications
under the ”work” and ”flux” form mappings. We now move on to some four dimensional examples.
Example 12.2.1. Charge conservation: Consider the 4-current we introduced in example 12.1.5.
Take the exterior derivative of the dual of the current to get,
d(∗ J ) = d(ρdx ∧ dy ∧ dz − ΦJ~ ∧ dt)

= (∂t ρ)dt ∧ dx ∧ dy ∧ dz − d[(Jx dy ∧ dz + Jy dz ∧ dx + Jz dx ∧ dy) ∧ dt]
= dρ ∧ dx ∧ dy ∧ dz
−∂x Jx dx ∧ dy ∧ dz ∧ dt − ∂y Jy dy ∧ dz ∧ dx ∧ dt − ∂z Jz dz ∧ dx ∧ dy ∧ dt
~ ∧ dx ∧ dy ∧ dz.
= (∂t ρ + ∇ · J)dt
We work through the same calculation using index techniques,
d(∗ J ) = d(ρdx ∧ dy ∧ dz − ΦJ~ ∧ dt)

= d(ρ) ∧ dx ∧ dy ∧ dz − d[ 12 ijk Ji dxj ∧ dxk ∧ dt)
= (∂t ρ)dt ∧ dx ∧ dy ∧ dz − 12 ijk ∂µ Ji dxµ ∧ dxj ∧ dxk ∧ dt
= (∂t ρ)dt ∧ dx ∧ dy ∧ dz − 12 ijk ∂m Ji dxm ∧ dxj ∧ dxk ∧ dt
= (∂t ρ)dt ∧ dx ∧ dy ∧ dz − 12 ijk mjk ∂m Ji dx ∧ dy ∧ dz ∧ dt
= (∂t ρ)dt ∧ dx ∧ dy ∧ dz − 12 2δim ∂m Ji dx ∧ dy ∧ dz ∧ dt
~ ∧ dx ∧ dy ∧ dz.
= (∂t ρ + ∇ · J)dt
Observe that we can now phrase charge conservation by the following equation
d(∗ J ) = 0 ⇐⇒ ∂t ρ + ∇ · J~ = 0.
In the classical scheme of things this was a derived consequence of the equations of electromagnetism,
however it is possible to build the theory regarding this equation as fundamental. Rindler describes
that formal approach in a late chapter of ”Introduction to Special Relativity”.
12.2. EXTERIOR DERIVATIVES OF CHARGE FORMS, FIELD TENSORS, AND THEIR DUALS 303
Proposition 12.2.2.
~ is the vector potential (which gives the magnetic field) and A = −V dt +

If (Aµ ) = (−V, A)
ωA~ , then dA = ωE~ + ΦB~ = F where F is the electromagnetic field tensor. Moreover,
Fµν = ∂µ Aν − ∂ν Aµ .
~ = −∇V − ∂t A and B
Proof: The proof uses the definitions E ~ = ∇×A
~ and some vector identities:
dA = d(−V dt + ωA~ )
= −dV ∧ dt + d(ωA~ )
= −dV ∧ dt + (∂t Ai )dt ∧ dxi + (∂j Ai )dxj ∧ dxi
= ω(−∇V ) ∧ dt − ω∂t A~ ∧ dt + Φ∇×A~
= (ω(−∇V ) − ω∂t A~ ) ∧ dt + Φ∇×A~
~ ∧ dt + Φ∇×A
= ω(−∇V −∂t A) ~
= ωE~ ∧ dt + ΦB~
1
= F = Fµν dxµ ∧ dxν .
2
Moreover we also have:
dA = d(Aν ) ∧ dxν
= ∂µ Aν dxµ ∧ dxν
= 21 (∂µ Aν − ∂ν Aµ )dxµ ∧ dxν + 21 (∂µ Aν + ∂ν Aµ )dxµ ∧ dxν
= 21 (∂µ Aν − ∂ν Aµ )dxµ ∧ dxν .
Comparing the two identities we see that Fµν = ∂µ Aν − ∂ν Aµ and the proposition follows.
Example 12.2.3. Exterior derivative of the field tensor: We have just seen that the field
tensor is the exterior derivative of the potential one-form. We now compute the exterior derivative
of the field tensor expecting to find Maxwell’s equations since the derivative of the fields are governed
by Maxwell’s equations,
dF = d(Ei dxi ∧ dt) + d(ΦB~ )

~ (12.12)
= ∂m Ei (dxm ∧ dxi ∧ dt) + (∇ · B)dx ∧ dy ∧ dz + 21 ijk (∂t Bi )(dt ∧ dxj ∧ dxk ).
W pause here to explain our logic. In the above we dropped the ∂t Ei dt ∧ · · · term because it was
wedged with another dt in the term so it vanished. Also we broke up the exterior derivative on the
~ into the space and then time derivative terms and used our work in example 10.6.7.
flux form of B
Continuing the calculation,

dF ~
= [∂j Ek + 21 ijk (∂t Bi )]dxj ∧ dxk ∧ dt + (∇ · B)dx ∧ dy ∧ dz
= [∂x Ey − ∂y Ex + i12 (∂t Bi )]dx ∧ dy ∧ dt
+[∂z Ex − ∂x Ez + i31 (∂t Bi )]dz ∧ dx ∧ dt
+[∂y Ez − ∂z Ey + i23 (∂t Bi )]dy ∧ dz ∧ dt (12.13)
~
+(∇ · B)dx ∧ dy ∧ dz
~
= (∇ × E + ∂t B) ~ i Φe ∧ dt + (∇ · B)dx
~ ∧ dy ∧ dz
i
= Φ∇×E+∂
~
~
~ ∧ dt + (∇ · B)dx ∧ dy ∧ dz
tB
where we used the fact that Φ is an isomorphism of vector spaces (at a point) and Φe1 = dy ∧ dz,
Φe2 = dz ∧ dx, and Φe3 = dx ∧ dy. Behold, we can state two of Maxwell’s equations as
dF = 0 ⇐⇒ ∇×E ~ = 0,
~ + ∂t B ~ =0
∇·B (12.14)
Example 12.2.4. We now compute the exterior derivative of the dual to the field tensor:
d∗ F = d(−Bi dxi ∧ dt) + d(ΦE~ )

~ (12.15)
= Φ−∇×B+∂
~ ~ ∧ dt + (∇ · E)dx ∧ dy ∧ dz
tE
~ 7→ −B
This follows directly from the last example by replacing E ~ and B
~ 7→ E.
~ We obtain the two
∗
inhomogeneous Maxwell’s equations by setting d F equal to the Hodge dual of the 4-current,
d∗ F = µo ∗ J ⇐⇒ ~ + ∂t E
−∇ × B ~ = −µo J,
~ ~ =ρ
∇·E (12.16)
Here we have used example 12.1.5 to find the RHS of the Maxwell equations.
We now know how to write Maxwell’s equations via differential forms. The stage is set to prove that
Maxwell’s equations are Lorentz covariant, that is they have the same form in all inertial frames.
12.3 coderivatives and comparing to Griffith’s relativitic E & M

Optional section, for those who wish to compare our tensorial E & M with that of
Griffith’s, you may skip ahead to the next section if not interested
I should mention that this is not the only way to phrase Maxwell’s equations in terms of
differential forms. If you try to see how what we have done here compares with the equations
presented in Griffith’s text it is not immediately obvious. He works with F µν and Gµν and J µ none
of which are the components of differential forms. Nevertheless he recovers Maxwell’s equations
as ∂µ F µν = J ν and ∂µ Gµν = 0. If we compare the components of ∗ F with equation 12.119 ( the
matrix form of Gµν ) in Griffith’s text,
 
0 B1 B2 B3
−B1 0 −E3 E2 
(Gµν (c = 1)) =   = −(∗ F µν ). (12.17)
−B2 −E3 0 −E1 
−B3 E2 −E1 0
12.4. MAXWELL’S EQUATIONS ARE RELATIVISTICALLY COVARIANT 305
we find that we obtain the negative of Griffith’s ”dual tensor” ( recall that raising the indices has
the net-effect of multiplying the zeroth row and column by −1). The equation ∂µ F µν = J ν does not
follow directly from an exterior derivative, rather it is the component form of a ”coderivative”. The
coderivative is defined δ = ∗ d∗ , it takes a p-form to an (n−p)-form then d makes it a (n−p+1)-form
then finally the second Hodge dual takes it to an (n − (n − p + 1))-form. That is δ takes a p-form
to a p − 1-form. We stated Maxwell’s equations as
dF = 0 d∗ F = ∗ J
Now we can take the Hodge dual of the inhomogeneous equation to obtain,
∗ ∗
d F = δF = ∗∗ J = ±J
where I leave the sign for you to figure out. Then the other equation
∂µ Gµν = 0
can be understood as the component form of δ ∗ F = 0 but this is really dF = 0 in disguise,
0 = δ ∗ F = ∗ d∗∗ F = ±∗ dF ⇐⇒ dF = 0
so even though it looks like Griffith’s is using the dual field tensor for the homogeneous Maxwell’s
equations and the field tensor for the inhomogeneous Maxwell’s equations it is in fact not the case.
The key point is that there are coderivatives implicit within Griffith’s equations, so you have to
read between the lines a little to see how it matched up with what we’ve done here. I have not en-
tirely proved it here, to be complete we should look at the component form of δF = J and explicitly
show that this gives us ∂µ F µν = J ν , I don’t think it is terribly difficult but I’ll leave it to the reader.
Comparing with Griffith’s is fairly straightforward because he uses the same metric as we have.
Other texts use the mostly negative metric, its just a convention. If you try to compare to such
a book you’ll find that our equations are almost the same up to a sign. One good careful book
is Reinhold A. Bertlmann’s Anomalies in Quantum Field Theory you will find much of what we
have done here done there with respect to the other metric. Another good book which shares our
conventions is Sean M. Carroll’s An Introduction to General Relativity: Spacetime and Geometry,
that text has a no-nonsense introduction to tensors forms and much more over a curved space (
in contrast to our approach which has been over a vector space which is flat ). By now there are
probably thousands of texts on tensors; these are a few we have found useful here.
12.4 Maxwell’s equations are relativistically covariant

Let us begin with the definition of the field tensor once more. We define the components of the
field tensor in terms of the 4-potentials as we take the view-point those are the basic objects (not
the fields). If
Fµν ≡ ∂µ Aν − ∂ν Aµ ,
then the field tensor F = Fµν dxµ ⊗ dxν is a tensor, or is it ? We should check that the components
transform as they ought according to the discussion in section ??. Let x̄µ = Λµν xν then we observe,
α
(1.) Āµ = (Λ−1 )µ Aα
β ∂
−1 β ∂
(12.18)
(2.) ∂∂x̄ν = ∂x
∂ x̄ν ∂xβ = (Λ )ν ∂xβ
where (2.) is simply the chain rule of multivariate calculus and (1.) is not at all obvious. We will
assume that (1.) holds, that is we assume that the 4-potential transforms in the appropriate way
for a one-form. In principle one could prove that from more base assumptions. After all electro-
magnetism is the study of the interaction of charged objects, we should hope that the potentials
are derivable from the source charge distribution. Indeed, there exist formulas to calculate the
potentials for moving distributions of charge. We could take those as definitions for the potentials,
then it would be possible to actually calculate if (1.) is true. We’d just change coordinates via a
Lorentz transformation and verify (1.). For the sake of brevity we will just assume that (1.) holds.
We should mention that alternatively one can show the electric and magnetic fields transform as to
make Fµν a tensor. Those derivations assume that charge is an invariant quantity and just apply
Lorentz transformations to special physical situations to deduce the field transformation rules. See
Griffith’s chapter on special relativity or look in Resnick for example.
Let us find how the field tensor transforms assuming that (1.) and (2.) hold, again we consider
x̄µ = Λµν xν ,
F̄µν = ∂¯µ Āν − ∂¯ν Āµ
α β β α
= (Λ−1 )µ ∂α ((Λ−1 )ν Aβ ) − (Λ−1 )ν ∂β ((Λ−1 )µ Aα )
α β (12.19)
= (Λ−1 )µ (Λ−1 )ν (∂α Aβ − ∂β Aα )
α β
= (Λ−1 )µ (Λ−1 )ν Fαβ .
therefore the field tensor really is a tensor over Minkowski space.
Proposition 12.4.1.
The dual to the field tensor is a tensor over Minkowski space. For a given Lorentz trans-
formation x̄µ = Λµν xν it follows that
∗ α β
F̄µν = (Λ−1 )µ (Λ−1 )ν ∗ Fαβ
Proof: homework (just kidding in 2010), it follows quickly from the definition and the fact we
already know that the field tensor is a tensor.
Proposition 12.4.2.
The four-current is a four-vector. That is under the Lorentz transformation x̄µ = Λµν xν we
can show,
α
J¯µ = (Λ−1 )µ Jα
12.5. ELECTROSTATICS IN FIVE DIMENSIONS 307
Proof: follows from arguments involving the invariance of charge, time dilation and length con-
traction. See Griffith’s for details, sorry we have no time.
Corollary 12.4.3.
The dual to the four current transforms as a 3-form. That is under the Lorentz transfor-
mation x̄µ = Λµν xν we can show,
α β γ
∗¯
J µνσ = (Λ−1 )µ (Λ−1 )ν (Λ−1 )σ Jαβγ
Up to now the content of this section is simply an admission that we have been a little careless in
defining things upto this point. The main point is that if we say that something is a tensor then we
need to make sure that is in fact the case. With the knowledge that our tensors are indeed tensors
the proof of the covariance of Maxwell’s equations is trivial.
dF = 0 d∗ F = ∗ J
are coordinate invariant expressions which we have already proved give Maxwell’s equations in one
frame of reference, thus they must give Maxwell’s equations in all frames of reference.
The essential point is simply that
1 1
F = Fµν dxµ ∧ dxν = F̄µν dx̄µ ∧ dx̄ν
2 2
Again, we have no hope for the equation above to be true unless we know that
α β
F̄µν = (Λ−1 )µ (Λ−1 )ν Fαβ . That transformation follows from the fact that the four-potential is a
four-vector. It should be mentioned that others prefer to ”prove” the field tensor is a tensor by
studying how the electric and magnetic fields transform under a Lorentz transformation. We in
contrast have derived the field transforms based ultimately on the seemingly innocuous assumption
α
that the four-potential transforms according to Āµ = (Λ−1 )µ Aα . OK enough about that.
So the fact that Maxwell’s equations have the same form in all relativistically inertial frames
of reference simply stems from the fact that we found Maxwell’s equation were given by an arbitrary
frame, and the field tensor looks the same in the new barred frame so we can again go through all
the same arguments with barred coordinates. Thus we find that Maxwell’s equations are the same
in all relativistic frames of reference, that is if they hold in one inertial frame then they will hold
in any other frame which is related by a Lorentz transformation.
12.5 Electrostatics in Five dimensions

We will endeavor to determine the electric field of a point charge in 5 dimensions where we are
thinking of adding an extra spatial dimension. Lets call the fourth spatial dimension the w-direction
so that a typical point in space time will be (t, x, y, z, w). First we note that the electromagnetic
field tensor can still be derived from a one-form potential,
A = −ρdt + A1 dx + A2 dy + A3 dz + A4 dw
we will find it convenient to make our convention for this section that µ, ν, ... = 0, 1, 2, 3, 4 whereas
m, n, ... = 1, 2, 3, 4 so we can rewrite the potential one-form as,
A = −ρdt + Am dxm
This is derived from the vector potential Aµ = (ρ, Am ) under the assumption we use the natural
generalization of the Minkowski metric, namely the 5 by 5 matrix,
 
−1 0 0 0 0
0 1 0 0 0
µν
 
0 0
(ηµν ) =  1 0 0  = (η ) (12.20)
0 0 0 1 0
0 0 0 0 1
we could study the linear isometries of this metric, they would form the group O(1, 4). Now we
form the field tensor by taking the exterior derivative of the one-form potential,
1
F = dA = (∂µ ∂ν − ∂ν ∂µ )dxµ ∧ dxν
2
now we would like to find the electric and magnetic ”fields” in 4 dimensions. Perhaps we should
say 4+1 dimensions, just understand that I take there to be 4 spatial directions throughout this
discussion if in doubt. Note that we are faced with a dilemma of interpretation. There are 10
independent components of a 5 by 5 antisymmetric tensor, naively we wold expect that the electric
and magnetic fields each would have 4 components, but that is not possible, we’d be missing
two components. The solution is this, the time components of the field tensor are understood to
correspond to the electric part of the fields whereas the remaining 6 components are said to be
magnetic. This aligns with what we found in 3 dimensions, its just in 3 dimensions we had the
fortunate quirk that the number of linearly independent one and two forms were equal at any point.
This definition means that the magnetic field will in general not be a vector field but rather a ”flux”
encoded by a 2-form.  
0 −Ex −Ey −Ez −Ew
 Ex
 0 Bz −By H1  
(Fµν ) =  Ey −Bz
 0 Bx H2  (12.21)
 Ez By −Bx 0 H3 
Ew −H1 −H2 −H3 0
Now we can write this compactly via the following equation,
F = E ∧ dt + B
I admit there are subtle points about how exactly we should interpret the magnetic field, however
I’m going to leave that to your imagination and instead focus on the electric sector. What is the
generalized Maxwell’s equation that E must satisfy?
d∗ F = µo ∗ J =⇒ d∗ (E ∧ dt + B) = µo ∗ J
where J = −ρdt + Jm dxm so the 5 dimensional Hodge dual will give us a 5 − 1 = 4 form, in
particular we will be interested in just the term stemming from the dual of dt,
∗
(−ρdt) = ρdx ∧ dy ∧ dz ∧ dw
the corresponding term in d∗ F is d∗ (E ∧ dt) thus, using µo = 1

o ,
1
d∗ (E ∧ dt) = ρdx ∧ dy ∧ dz ∧ dw (12.22)
o
is the 4-dimensional Gauss’s equation. Now consider the case we have an isolated point charge
which has somehow always existed at the origin. Moreover consider a 3-sphere that surrounds the
charge. We wish to determine the generalized Coulomb field due to the point charge. First we note
that the solid 3-sphere is a 4-dimensional object, it the set of all (x, y, z, w) ∈ R4 such that
x2 + y 2 + z 2 + w2 ≤ r2
We may parametrize a three-sphere of radius r via generalized spherical coordinates,
x = rsin(θ)cos(φ)sin(ψ)
y = rsin(θ)sin(φ)sin(ψ)
(12.23)
z = rcos(θ)sin(ψ)
w = rcos(ψ)
Now it can be shown that the volume and surface area of the radius r three-sphere are as follows,
π2 4
vol(S 3 ) = r area(S 3 ) = 2π 2 r3
2
We may write the charge density of a smeared out point charge q as,
(
2q/π 2 a4 , 0 ≤ r ≤ a
ρ= . (12.24)
0, r>a
Notice that if we integrate ρ over any four-dimensional region which contains the solid three sphere
of radius a will give the enclosed charge to be q. Then integrate over the Gaussian 3-sphere S 3
with radius r call it M ,
Z Z
∗ 1
d (E ∧ dt) = ρdx ∧ dy ∧ dz ∧ dw
M o M
now use the Generalized Stokes Theorem to deduce,
Z
∗ q
(E ∧ dt) =
∂M o
but by the ”spherical” symmetry of the problem we find that E must be independent of the direction
it points, this means that it can only have a radial component. Thus we may calculate the integral
with respect to generalized spherical coordinates and we will find that it is the product of Er ≡ E
and the surface volume of the four dimensional solid three sphere. That is,
Z
∗ q
(E ∧ dt) = 2π 2 r3 E =
∂M o
Thus,
q
E=
2π 2 o r3
the Coulomb field is weaker if it were to propogate in 4 spatial dimensions. Qualitatively what has
happened is that the have taken the same net flux and spread it out over an additional dimension,
this means it thins out quicker. A very similar idea is used in some brane world scenarios. String
theorists posit that the gravitational field spreads out in more than four dimensions while in con-
trast the standard model fields of electromagnetism, and the strong and weak forces are confined
to a four-dimensional brane. That sort of model attempts an explaination as to why gravity is so
weak in comparison to the other forces. Also it gives large scale corrections to gravity that some
hope will match observations which at present don’t seem to fit the standard gravitational models.
This example is but a taste of the theoretical discussion that differential forms allow. As a
final comment I remind the reader that we have done things for flat space for the most part in
this course, when considering a curved space there are a few extra considerations that must enter.
Coordinate vector fields ei must be thought of as derivations ∂/∂xµ for one. Also the metric is not
a constant tensor like δij or ηµν rather is depends on position, this means Hodge duality aquires
a coordinate dependence as well. Doubtless I have forgotten something else in this brief warning.
One more advanced treatment of many of our discussions is Dr. Fulp’s Fiber Bundles 2001 notes
which I have posted on my webpage. He uses the other metric but it is rather elegantly argued, all
his arguments are coordinate independent. He also deals with the issue of the magnetic induction
and the dielectric, issues which we have entirely ignored since we always have worked in free space.
References and Acknowledgements:
I have drawn from many sources to assemble the content of the last couple chapters, the refer-
ences are listed approximately in the order of their use to the course, additionally we are indebted
to Dr. Fulp for his course notes from many courses (ma 430, ma 518, ma 555, ma 756, ...). Also
Manuela Kulaxizi helped me towards the correct (I hope) interpretation of 5-dimensional E&M in
the last example.
Vector Calculus, Susan Jane Colley
Introduction to Special Relativity, Robert Resnick

Differential Forms and Connections, R.W.R. Darling
Differential geometry, gauge theories, and gravity, M. Göckerler & T. Schücker
Anomalies in Quantum Field Theory, Reinhold A. Bertlmann
”The Differential Geometry and Physical Basis for the Applications of Feynman Diagrams”, S.L.
Marateck, Notices of the AMS, Vol. 53, Number 7, pp. 744-752
Abstract Linear Algebra, Morton L. Curtis
Gravitation, Misner, Thorne and Wheeler
Introduction to Special Relativity, Wolfgang Rindler
Differential Forms A Complement to Vector Calculus, Steven H. Weintraub
Differential Forms with Applications to the Physical Sciences, Harley Flanders
Introduction to Electrodynamics, (3rd ed.) David J. Griffiths
The Geometry of Physics: An Introduction, Theodore Frankel
An Introduction to General Relativity: Spacetime and Geometry, Sean M. Carroll
Gauge Theory and Variational Principles, David Bleeker
Group Theory in Physics, Wu-Ki Tung

Chapter 13
introduction to variational calculus
13.1 history
The problem of variational calculus is almost as old as modern calculus. Variational calculus seeks
to answer questions such as:
Remark 13.1.1.
1. what is the shortest path between two points on a surface ?
2. what is the path of least time for a mass sliding without friction down some path
between two given points ?
3. what is the path which minimizes the energy for some physical system ?
4. given two points on the x-axis and a particular area what curve has the longest
perimeter and bounds that area between those points and the x-axis?
You’ll notice these all involve a variable which is not a real variable or even a vector-valued-variable.
Instead, the answers to the questions posed above will be paths or curves depending on how you
wish to frame the problem. In variational calculus the variable is a function and we wish to find
extreme values for a functional. In short, a functional is an abstract function of functions. A
functional takes as an input a function and gives as an output a number. The space from which
these functions are taken varies from problem to problem. Often we put additional contraints
or conditions on the space of admissable solutions. To read about the full generality of the
problem you should look in a text such as Hans Sagan’s. Our treatment is introductory in this chap-
ter, my aim is to show you why it is plausible and then to show you how we use variational calculus.
We will see that the problem of finding an extreme value for a functional is equivalent to solving
the Euler-Lagrange equations or Euler equations for the functional. Euler predates Lagrange in his
313
314 CHAPTER 13. INTRODUCTION TO VARIATIONAL CALCULUS
discovery of the equations bearing their names. Eulers’s initial attack of the problem was to chop
the hypothetical solution curve up into a polygonal path. The unknowns in that approach were
the coordinates of the vertices in the polygonal path. Then through some ingenious calculations
he arrived at the Euler-Lagrange equations. Apparently there were logical flaws in Euler’s origi-
nal treatment. Lagrange later derived the same equations using the viewpoint that the variable
was a function and the variation was one of shifting by an arbitrary function. The treatment of
variational calculus in Edwards is neither Euler nor Lagrange’s approach, it is a refined version
which takes in the contributions of generations of mathematicians working on the subject and then
merges it with careful functional analysis. I’m no expert of the full history, I just give you a rough
sketch of what I’ve gathered from reading a few variational calculus texts.
Physics played a large role in the development of variational calculus. Lagrange was a physicist
as well as a mathematician. At the present time, every physicist takes course(s) in Lagrangian
Mechanics. Moreover, the use of variational calculus is fundamental since Hamilton’s principle says
that all physics can be derived from the principle of least action. In short this means that nature is
lazy. The solutions realized in the physical world are those which minimize the action. The action
Z
S[y] = L(y, y 0 , t) dt
is constructed from the Lagrangian L = T − U where T is the kinetic energy and U is the potential
energy. In the case of classical mechanics the Euler Lagrange equations are precisely Newton’s
equations. The Hamiltonian H = T + U is similar to the Lagrangian except that the funda-
mental variables are taken to be momentum and position in contrast to velocity and position in
Lagrangian mechanics. Hamiltonians and Lagrangians are used to set-up new physical theories.
Euler-Lagrange equations are said to give the so-called classical limit of modern field theories. The
concept of a force is not so useful to quantum theories, instead the concept of energy plays the
central role. Moreover, the problem of quantizing and then renormalizing field theory brings in
very sophisiticated mathematics. In fact, the math of modern physics is not understood. In this
chapter I’ll just show you a few famous classical mechanics problems which are beatifully solved by
Lagrange’s approach. We’ll also see how expressing the Lagrangian in non-Cartesian coordinates
can give us an easy way to derive forces that arise from geometric contraints. Hopefully we can
derive the coriolis force in this manner. I also plan to include a problem or two about Maxwell’s
equations from the variational viewpoint. There must be at least a dozen different ways to phrase
Maxwell’s equations, one reason I revisit them is to give you a concrete example as to the fact that
physics has many formulations.
I am following the typical physics approach to variational calculus. Edwards’ last chapter is more
natural mathematically but I think the math is a bit much for your first exposure to the subject.
The treatment given here is close to that of Arfken and Weber’s Mathematical Physics text, how-
ever I suspect you can find these calculations in dozens of classical mechanics texts. More or less
our approach is that of Lagrange.
13.2. THE VARIATIONAL PROBLEM 315
13.2 the variational problem

Our goal in what follows here is to maximize or minimize a particular function of functions. Suppose
Fo is a set of functions with some particular property. For now, we may could assume that all the
functions in Fo have graphs that include (x1 , y1 ) and (x2 , y2 ). Consider a functional J : Fo → Fo
which is defined by an integral of some function f which we call the Lagrangian,
Z x2
J[y] = f (y, y 0 , x) dx.
x1
We suppose that f is given but y is a variable. Consider that if we are given a function y ∗ ∈ Fo
and another function η such that η(x1 ) = η(x2 ) = 0 then we can reach a whole family of functions
indexed by a real variable α as follows (relabel y ∗ (x) by y(x, 0) so it matches the rest of the family
of functions):
y(x, α) = y(x, 0) + αη(x)
Note that x 7→ y(x, α) gives a function in Fo . We define the variation of y to be
δy = αη(x)
This means y(x, α) = y(x, 0) + δy. We may write J as a function of α given the variation we just
described: Z x2
J(α) = f (y(x, α), y(x, α)0 , x) dx.
x1
It is intuitively obvious that if the function y ∗ (x) = y(x, 0) is an extremum of the functional then
we ought to expect

∂J(α)
=0
∂α α=0
Notice that we can calculate the derivative above using multivariate calculus. Remember that
∂y 0
y(x, α) = y(x, 0) + αη(x) hence y(x, α)0 = y(x, 0)0 + αη(x)0 thus ∂α = η and ∂y 0 dη
∂α = η = dx .
Consider that:
Z x2
∂J(α) ∂ 0
= f (y(x, α), y(x, α) , x) dx
∂α ∂α x1
∂f ∂y 0 ∂f ∂x
Z x2
∂f ∂y
= + + dx
x1 ∂y ∂α ∂y 0 ∂α ∂x ∂α
Z x2
∂f ∂f dη
= η+ 0 dx (13.1)
x1 ∂y ∂y dx
Observe that
d ∂f d ∂f ∂f dη
0
η = 0
η+ 0
dx ∂y dx ∂y ∂y dx
Hence continuing Equation 13.1 in view of the product rule above,

Z x2
∂J(α) ∂f d ∂f d ∂f
= η+ η − η dx
∂α x1 ∂y dx ∂y 0 dx ∂y 0
∂f x2
Z x2
∂f d ∂f
= 0 η + η− η dx (13.2)
∂y x1 x1 ∂y dx ∂y 0
Z x2
∂f d ∂f
= − η dx
x1 ∂y dx ∂y 0
∂f x2 ∂f ∂f

Note we used the conditions η(x1 ) = η(x2 ) to see that ∂y 0 η x = ∂y 0 η(x2 ) − ∂y 0 η(x1 ) = 0. Our goal
1
is to find the extreme values for the functional J. Let me take a few sentences to again restate
our set-up. Generally, we take a function y then J maps to a new function J[y]. The family of
functions indexed by α gives a whole ensemble of functions in Fo which are near y ∗ according to
the formula,
y(x, α) = y ∗ (x) + αη(x)
Let’s call this set of functions Wη . If we took another function like η, say ζ such that ζ(x1 ) =
ζ(x2 ) = 0 then we could look at another family of functions:
y(x, α) = y ∗ (x) + αζ(x)
and we could denote the set of all such functions generated from ζ to be Wζ . The total variation
of y based at y ∗ should include all possible families of functions in Fo . You could think of Wη and
Wζ be two different subspaces in Fo . If η 6= ζ then these subspaces of Fo are likely disjoint except
for the proposed extremal solution y ∗ . It is perhaps a bit unsettling to realize there are infinitely
many such subspaces because there are infinitely many choices
for
the function η or ζ. In any event,
∂J(α)
each possible variation of y ∗ must satisfy the condition ∂α = 0 since we assume that y ∗
α=0
is an extreme value of the functional J. It follows that the Equation 13.2 holds for
R xall possible η.
Therefore, we ought to expect that any extreme value of the functional J[y] = x12 f (y, y 0 , x) dx
must solve the Euler Lagrange Equations:
Z x2
∂f d ∂f
− = 0 Euler-Lagrange Equations for J[y] = f (y, y 0 , x) dx
∂y dx ∂y 0 x1
13.3 variational derivative

The role that η played in the discussion in the preceding section is somewhat similar to the role
that the ”h” plays in the definition f 0 (a) = limh→0 f (a+h)−f
h
(a)
. You might hope we could replace
arguments in η with a more direct approach. Physicists have a heuristic way of making such
arguments in terms of the variation δ. They would cast the arguments in the last page by just
13.4. EULER-LAGRANGE EXAMPLES 317
”taking the variation of J”. Let me give you their formal argument,
Z x2
0
δJ = δ f (y, y , x) dx
x1
Z x2
= δf (y, y 0 , x) dx
x
Z x2 1
∂f ∂f dy ∂f
= δy + 0 δ dx + δx dx
x1 ∂y ∂y ∂x
Z x2
∂f ∂f d
= δy + 0 δy dx (13.3)
x1 ∂y ∂y dx
∂f x2
Z x2
∂f d ∂f
= 0 δy + − δy dx
∂y x1 x1 ∂y dx ∂y 0
Therefore, since δy = 0 at the endpoints of integration, the Euler-Lagrange equations follow from
δJ = 0. Now, if you’re like me, the argument above is less than satisfying since we never actually
defined what it means to ”take δ” of something. Also, why could I commute the variational δ and
d
dx )? That said, the formal method is not without use since it allows the focus to be on the Euler
Lagrange equations rather than the technical details of the variation.
Remark 13.3.1.
The more adept reader at this point should realize the hypocrisy of me calling the above
calculation formal since even my presentation here was formal. I also used an analogy, I
assumed that the theory of extreme values for multivariate calculus extends to function
space. But, Fo is not Rn , it’s much bigger. Edwards builds the correct formalism for a
rigourous calculation of the variational derivative. To be careful we’d need to develop the
norm on function space and prove a number of results about infinite dimensional linear
algebra. Take a look at the last chapter in Edwards’ text if you’re interested. I don’t
believe I’ll have time to go over that material this semester.
13.4 Euler-Lagrange examples

I present a few standard examples in this section. We make use of the calculation in the last
section. Also, we will use a result from your homework which states an equivalent form of the
Euler-Lagrange equation is
∂f d 0 ∂f
− f −y = 0.
∂x dx ∂y 0
This form of the Euler Lagrange equation yields better differential equations for certain examples.
13.4.1 shortest distance between two points in plane

If s denotes the arclength in 2 2 2
q the xy-plane then the pythagorean theorem gives ds = dx + dy
dy 2
infinitesimally. Thus, ds = 1 + dx dx and we may add up all the little distances ds to find the
total length between two given points (x1 , y1 ) and (x2 , y2 ):
Z x2 p
J[y] = 1 + (y 0 )2 dx
x1
Identify that we have f (y, y 0 , x) =

p
1 + (y 0 )2 . Calculate then,
∂f ∂f y0
=0 and = .
∂y 0
p
∂y 1 + (y 0 )2
Euler Lagrange equations yield,
y0 y0

d ∂f ∂f d
0
= ⇒ p =0 ⇒ p =k
dx ∂y ∂y dx 1 + (y 0 )2 1 + (y 0 )2
where k ∈ R is constant with respect to x. Moreover, square both sides to reveal

r
(y 0 )2 2 0 2 k2 dy k2
0 2
=k ⇒ (y ) = 2
⇒ =± =m
1 + (y ) 1−k dx 1 − k2
where I have defined m is defined in the obvious way. We find solutions y = mx + b. Finally, we
can find m, b to fit the given pair of points (x1 , y1 ) and (x2 , y2 ) as follows:
y2 − y1
y1 = mx1 + b and y2 = mx2 + b ⇒ y = y1 + (x − x1 )
x2 − x1
provided x1 6= x2 . If x1 6= x2 and y1 6= y2 then we could perform the same calculation as above
with the roles of x and y interchanged,
Z y2 p
J[x] = 1 + (x0 )2 dy
y1
where x0 = dx/dy and the Euler Lagrange equations would yield the solution
x2 − x1
x = x1 + (y − y1 ).
y2 − y1
Finally, if both coordinates are equal then (x1 , y1 ) = (x2 , y2 ) and the shortest path between these
points is the trivial path, the armchair solution. Silly comments aside, we have shown that a
straight line provides the curve with the shortest arclength between any two points in the plane.
13.4.2 surface of revolution with minimal area

Suppose we wish to revolve some curve which connects (x1 , y1 ) and (x2 , y2 ) around the x-axis. A
surface constructed in this manner is called a surface of revolution. In calculus we learn how
to calculate the surface area of such a shape. One can imagine deconstructing the surface into a
sequence of ribbons. Each ribbon at position x will have a ”radius” of y and a width of dx however,
because the shape is tilted the area of the ribbon works out to dA = 2πyds where ds is the arclength.
p
If we choose x as the parameter this yields dA = 2πy 1 + (y 0 )2 dx. To find the surface of minimal
surface area we ought to consider the functional:
Z x2 p
A[y] = 2πy 1 + (y 0 )2 dx
x1
Identify that f (y, y 0 , x) = 2πy 1 + (y 0 )2 hence fy = 2π 1 + (y 0 )2 and fy0 = 2πyy 0 / 1 + (y 0 )2 .

p p p
The usual Euler-Lagrange equations are not easy to solve for this problem, it’s easier to work with
the equations you derived in homework,

∂f d 0 ∂f
− f −y = 0.
∂x dx ∂y 0
Hence,
2πy(y 0 )2

d p
0 2
2πy 1 + (y ) − p =0
dx 1 + (y 0 )2
Dividing by 2π and making a common denominator,

d y y
p =0 ⇒ p =k
dx 1 + (y 0 )2 1 + (y 0 )2
where k is a constant with respect to x. Squaring the equation above yields
y2 dy 2
dy 2
= k2 ⇒ y 2 − k 2 = k 2 ( dx )
1 + ( dx )
Solve for dx, integrate, assuming the given points are in the first quadrant,
Z Z
kdy
x = dx = p = k cosh−1 ( ky ) + c
2
y −k 2
Hence,

x−c
y = k cosh
k
generates the surface of revolution of least area between two points. These shapes are called
Catenoids they can be observed in the formation of soap bubble between rings. There is a vast
literature on this subject and there are many cases to consider, I simply exhibit a simple solution.
For a given pair of points it is not immediately obvious if there exists a solution to the Euler-
Lagrange equations which fits the data. (see page 622 of Arfken).
13.4.3 Braichistochrone
Suppose a particle slides freely along some curve from (x1 , y1 ) to (x2 , y2 ) = (0, 0) under the influence
of gravity where we take y to be the vertical direction. What is the curve of quickest descent?
Notice that if x1 = 0 then the answer is easy to see, however, if x1 6= 0 then the question is not
trivial. To solve this problem we must first offer a functional which accounts for the time of descent.
R (x1 ,y1 ) ds
Note that the speed v = ds/dt so we’d clearly like to minimize J = (0,0) v . Since the object is
assumed to fall freely we may assume that energy is conserved in the motion hence
1 p
mv 2 = mg(y − y1 ) ⇒ v= 2g(y1 − y)
2
p
As we’ve discussed in previous examples, ds = 1 + (y 0 )2 dt so we find
Z x1 s
1 + (y 0 )2
J[y] = dx
0 2g(y1 − y)
| {z }
f (y,y 0 ,x)

∂f d 0 ∂f
Notice that the modified Euler-Lagrange equations ∂x − dx f − y ∂y0 = 0 are convenient since
fx = 0. We calculate that
∂f 1 2y 0 y0
= =
∂y 0
q p
1+(y 0 )2 2g(y1 − y) 2g(y1 − y)(1 + (y 0 )2 )
2 2g(y 1 −y)
√
Hence there should exist some constant 1/(k 2g) such that
s
1 + (y 0 )2 (y 0 )2 1
−p = √
2g(y1 − y) 2g(y1 − y)(1 + (y 0 )2 ) k 2g
It follows that,
2
1 1 dy
p = ⇒ y1 − y 1+ = k2
0 2
(y1 − y)(1 + (y ) ) k dx
We need to solve for dy/dx,

2 2
y + k 2 − y1

dy dy
(y1 − y) = k 2 − y1 + y ⇒ =
dx dx y1 − y
Or, relabeling constants a = y1 and b = k 2 − y1 and we must solve

s Z r
dy b+y a−y
=± ⇒ x=± dy
dx a−y b+y
The integral is not trivial. It turns out that the solution is a cycloid (Arfken p. 624):

a+b a+b
x= θ + sin(θ) − d y= 1 − cos(θ) − b
2 2
This is the curve that is traced out by a point on a wheel as it travels. If you take this solution
and calculate J[ycycloid ] you can show the time of descent is simply
r
π y1
T =
2 2g
if the mass begins to descend from (x2 , y2 ). But, this point has no connection with (x1 , y1 ) except
that they both reside on the same cycloid. It follows that the period of a pendulum that follows
a cycloidal path is indpendent of the starting point on the path. This is not true for a circular
pendulum in general, we need the small angle approximation to derive simple harmonic motion. It
turns out that it is possible to make a pendulum follow a cycloidal path if you let the string be
guided by a frame which is also cycloidal. The neat thing is that even as it loses energy it still
follows a cycloidal path and hence has the same period. The ”Brachistochrone” problem was posed
by Johann Bernoulli in 1696 and it actually predates the variational calculus of Lagrange by some
50 or so years. This problem and ones like it are what eventually prompted Lagrange and Euler to
systematically develop the subject. Apparently Galileo also studied this problem however lacked
the mathematics to crack it.
13.5 Euler-Lagrange equations for several dependent variables

We still consider problems with just one independent parameter underlying everything. For prob-
lems of classical mechanics this is almost always time t. In anticipation of that application we
choose to use the usual physics notation in the section. We suppose that our functional depends on
functions y1 , y2 , . . . , yn of time t along with their time derivatives ẏ1 , ẏ2 , . . . , ẏn . We again suppose
the functional of interest is an integral of a Lagrangian function f from time t1 to time t2 ,
Z t2
J[(yi )] = f (yi , ẏi , t) dt
t1
here we use (yi ) as shorthand for (y1 , y2 , . . . , yn ) and (ẏi ) as shorthand for (ẏ1 , ẏ2 , . . . , ẏn ). We
suppose that n-conditions are given for each of the endpoints in this problem; yi (t1 ) = yi1 and
yi (t2 ) = yi2 . Moreover, we define Fo to be the set of paths from R to Rn subject to the conditions
just stated. We now set out to find necessary conditions on a proposed solution to the extreme
value problem for the functional J above. As before let’s assume that an extremal solution y∗ ∈ Fo
exists. Moreover, imagine varying the solution by some variational function η = (ηi ) which has
η(t1 ) = (0, 0, . . . , 0) and η(t2 ) = (0, 0, . . . , 0). Consequently the family of paths defined below are
all in Fo ,
y(t, α) = y ∗ (t) + αη(t)
Thus y(t, 0) = y ∗ . In terms of component functions we have that
yi (t, α) = yi∗ (t) + αηi (t).
∗ ∗
that δyi = yi (t, α) − yi (t) = αηi (t). Since y is an extreme solution we should
You can identify
∂J
expect that ∂α = 0. Differentiate the functional with respect to α and make use of the
α=0
chain rule for f which is a function of some 2n + 1 variables,
Z t2
∂J(α) ∂
= f (yi (t, α), y˙i (t, α), t) dt
∂α ∂α t1
Z t2 X n
∂f ∂yj ∂f ∂ y˙j
= + dt
t1 j=1 ∂yj ∂α ∂ y˙j ∂α
Z t2 X n
∂f ∂f dηj
= ηj + dt (13.4)
t1 j=1 ∂yj ∂ y˙j dt
n n
∂f t2
Z t2 X
X ∂f d ∂f
= η + − ηj dt
∂ y˙j t1 t1 ∂yj dt ∂ y˙j
j=1 j=1
Since η(t1 ) = η(t2 ) = 0 the first term vanishes. Moreover, since we may repeat this calculation for
all possible variations about the optimal solution y ∗ it follows that we obtain a set of Euler-Lagrange
equations for each component function of the solution:
Z t2
∂f d ∂f
− = 0 j = 1, 2, . . . n Euler-Lagrange Eqns. for J[(yi )] = f (yi , y˙i , t) dt
∂yj dt ∂ ẏj t1
13.5. EULER-LAGRANGE EQUATIONS FOR SEVERAL DEPENDENT VARIABLES 323
Often we simply use y1 = x, y2 = y and y3 = z which denote the position of particle or perhaps
just the component functions of a path which gives the geodesic on some surface. In either case
we should have 3 sets of Euler-Lagrange equations, one for each coordinate. We will also use non-
Cartesian coordinates to describe certain Lagrangians. We develop many useful results for set-up
of Lagrangians in non-Cartesian coordinates in the next section.
13.5.1 free particle Lagrangian

For a particle of mass m the kinetic energy K is given in terms of the time derivatives of the
coordinate functions x, y, z as follows:
m
ẋ2 + ẏ 2 + ż 2

K= 2
Construct a functional by integrating the kinetic energy over time t,

Z t2
m
ẋ2 + ẏ 2 + ż 2 dt

S= 2
t1
The Euler-Lagrange equations for this functional are

∂K d ∂K ∂K d ∂K ∂K d ∂K
= = =
∂x dt ∂ ẋ ∂y dt ∂ ẏ ∂z dt ∂ ż
∂K ∂K ∂K
Since ∂ ẋ = mẋ, ∂ ẏ = mẏ and ∂ ż = mż it follows that
0 = mẍ 0 = mÿ 0 = mz̈.
You should recognize these as Newton’s equation for a particle with no force applied. The solution
is (x(t), y(t), z(t)) = (xo + tvx , yo + tvy , zo + tvz ) which is uniform rectilinear motion at constant
velocity (vx , vy , vz ). The solution to Newton’s equation minimizes the integral of the Kinetic energy.
Generally the quantity S is called the action and Hamilton’s Principle states that the laws of physics
all arise from minimizing the action of the physical phenomena. We’ll return to this discussion in
a later section.
13.5.2 geodesics in R3
A geodesic is the path of minimal length between a pair of points on some manifold. Note we
already proved that geodesics in the plane are just lines. In general, for R3 , the square of the
infinitesimal arclength element is ds2 = dx2 + dy 2 + dz 2 . The arclength integral from p = 0 to
q = (qx , qy , qz ) in R3 is most naturally given from the parametric viewpoint:
Z 1p
S= ẋ2 + ẏ 2 + ż 2 dt
0
We assume (x(0), y(0), z(0)) = (0, 0, 0) and (x(1), y(1), z(1)) = q and it should be clear that the
integral above calculates the arclength. The Euler-Lagrange equations for x, y, z are

d ẋ d ẏ d ż
p = 0, p = 0, p = 0.
dt ẋ2 + ẏ 2 + ż 2 dt ẋ2 + ẏ 2 + ż 2 dt ẋ2 + ẏ 2 + ż 2
It follows that there exist constants, say a, b and c, such that

ẋ ẏ ż
a= p , b= p , c= p .
ẋ2 + ẏ 2 + ż 2 ẋ2 + ẏ 2 + ż 2 ẋ2 + ẏ 2 + ż 2
These equations are said to be coupled since each involves derivatives of the others. We usually
need a way to uncouple the equations if we are to be successful in solving the system. We can
calculate, and equate each with the constant 1:
ẋ ẏ ż
1= p = p = p .
a ẋ2 + ẏ 2 + ż 2 b ẋ2 + ẏ 2 + ż 2 c ẋ2 + ẏ 2 + ż 2
But, multiplying by the denominator reveals an interesting identity

p ẋ ẏ ż
ẋ2 + ẏ 2 + ż 2 = = =
a b c
The solution has the form, x(t) = tqx , y(t) = tqy and z(t) = tqz . Therefore,
(x(t), y(t), z(t)) = t(qx , qy , qz ) = tq.
for 0 ≤ t ≤ 1. These are the parametric equations for the line segment from the origin to q.
13.6 the Euclidean metric

The square root in the functional of the last subsection certainly complicated the calculation. It
is intuitively clear that if we add up squared line elements ds2 to give a minimum then that ought
to correspond to the minimum for the sum of the positive square roots ds of those elements. Let’s
check if my conjecture works for R3 :
Z 1
S= ( ẋ2 + ẏ 2 + ż 2 ) dt
0 | {z }
f (x,y,z,ẋ,ẏ,ż)
This gives us the Euler Lagrange equations below:
ẍ = 0, ÿ = 0, z̈ = 0
The solution of these equations is clearly a line. In this formalism the equations were uncoupled
from the outset.
13.6. THE EUCLIDEAN METRIC 325
Definition 13.6.1.
The Euclidean metric is ds2 = dx2 + dy 2 + dz 2 . Generally, for orthogonal curvelinear
1 1 1
coordinates u, v, w we calculate ds2 = ||∇u|| 2 2 2
2 du + ||∇v||2 dv + ||∇w||2 dw . We use this as a
guide for constructing functionals which calculate arclength or speed

The beauty of the metric is that it allows us to calculate in other coordinates, consider
x = r cos(θ) y = r sin(θ)
For which we have implicit inverse coordinate transformations r2 = x2 + y 2 and θ = tan−1 (y/x).
From these inverse formulas we calculate:
∇r = < x/r, y/r > ∇θ = < −y/r2 , x/r2 >
Thus, ||∇r|| = 1 whereas ||∇θ|| = 1/r. We find that the metric in polar coordinates takes the form:
ds2 = dr2 + r2 dθ2
Physicists and engineers tend to like to think of these as arising from calculating the length of
infinitesimal displacements in the r or θ directions. Generically, for u, v, w coordinates
1 1 1
dlu = du dlv = dv dlw = dw
||∇u|| ||∇v|| ||∇w||
and ds2 = dl2u + dl2v + dl2w . So in that notation we just found dlr = dr and dlθ = rdθ. Notice then
that cylindircal coordinates have the metric,
ds2 = dr2 + r2 dθ2 + dz 2 .
For spherical coordinates x = r cos(φ) sin(θ), y = r sin(φ) sin(θ) and z = r cos(θ) (here 0 ≤ φ ≤ 2π
and 0 ≤ θ ≤ π, physics notation). Calculation of the metric follows from the line elements,
dlr = dr dlφ = r sin(θ)dφ dlθ = rdθ
Thus,
ds2 = dr2 + r2 sin2 (θ)dφ2 + r2 dθ2 .
We now have all the tools we need for examples in spherical or cylindrical coordinates. What about
other cases? In general, given some p-manifold in Rn how does one find the metric on that manifold?
If we are to follow the approach of this section we’ll need to find coordinates on Rn such that the
manifold S is described by setting all but p of the coordinates to a constant. For example, in R4
we have generalized cylindircal coordinates (r, φ, z, t) defined implicitly by the equations below
x = r cos(φ), y = r sin(φ), z = z, t=t
On the hyper-cylinder r = R we have the metric ds2 = R2 dθ2 + dz 2 + dw2 . There are mathemati-
cians/physicists whose careers are founded upon the discovery of a metric for some manifold. This
is generally a difficult task.
13.7 geodesics
A geodesic is a path of smallest distance on some manifold. In general relativity, it turns out that
the solutions to Eistein’s field equations are geodesics in 4-dimensional curved spacetime. Particles
that fall freely are following geodesics, for example projectiles or planets in the absense of other
frictional/non-gravitational forces. We don’t follow a geodesic in our daily life because the earth
pushes back up with a normal force. Also, do be honest, the idea of length in general relativity is a
bit more abstract that the geometric length studied in this section. The metric of general relativity
is non-Euclidean. General relativity is based on semi-Riemannian geometry whereas this section
is all Riemannian geometry. The metric in Riemannian geometry is positive definite. The metric
in semi-Riemannian geometry can be written as a quadratic form with both positive and negative
eigenvalues. In any event, if you want to know more I know some books you might like.
13.7.1 geodesic on cylinder

The equation of a cylinder of radius R is most easily framed in cylindrical coordinates (r, θ, z); the
equation is merely r = R hence the metric reads
ds2 = R2 dθ2 + dz 2
Therefore, we ought to minimize the following functional in order to locate the parametric equations
2 2
of a geodesic on the cylinder: note ds2 = R2 dθ
dt2
+ dz
dt2
dt2 thus:
Z
S = (R2 θ̇2 + ż 2 ) dt
Euler-Lagrange equations for the dependent variables θ and z are simply:

θ̈ = 0 z̈ = 0.
We can integrate twice to find solutions
θ(t) = θo + At z(t) = zo + Bt
Therefore, the geodesic on a cylinder is simply the line connecting two points in the plane which is
curved to assemble the cylinder. Simple cases that are easy to understand:
1. Geodesic from (R cos(θo ), R sin(θo ), z1 ) to (R cos(θo ), R sin(θo ), z2 ) is parametrized by θ(t) =
θo and z(t) = z1 + t(z2 − z1 ) for 0 ≤ t ≤ 1. Technically, there is some ambiguity here since I
never declared over what range the t is to range. Could pick other intervals, we could use z
at the parameter is we wished then θ(z) = θo and z = z for z1 ≤ z ≤ z2
2. Geodesic from (R cos(θ1 ), R sin(θ1 ), zo ) to (R cos(θ2 ), R sin(θ2 ), zo ) is parametrized by θ(t) =
θ1 + t(θ2 − θ1 ) and z(t) = zo for 0 ≤ t ≤ 1.
3. Geodesic from (R cos(θ1 ), R sin(θ1 ), z1 ) to (R cos(θ2 ), R sin(θ2 ), z2 ) is parametrized by
θ(t) = θ1 + t(θ2 − θ1 ) z(t) = z1 + t(z2 − z1 )
z2 −z1
You can eliminate t and find the equation z = θ2 −θ1 (θ − θ1 ) which again just goes to show
you this is a line in the curved coordinates.
13.8. LAGRANGIAN MECHANICS 327
13.7.2 geodesic on sphere

The equation of a sphere of radius R is most easily framed in spherical coordinates (r, φ, θ); the
equation is merely r = R hence the metric reads
ds2 = R2 sin2 (θ)dφ2 + R2 dθ2 .
Therefore, we ought to minimize the following functional in order to locate the parametric equations
2 2
of a geodesic on the sphere: note ds2 = R2 sin2 (θ) dφ
dt2
+ R2 dθ
dt2
dt2 thus:
Z
S= ( R2 sin2 (θ)φ̇2 + R2 θ̇2 ) dt
| {z }
f (θ,φ,θ̇,φ̇)
d
Euler-Lagrange equations for the dependent variables φ and θ are simply: fθ = dt (fθ̇ ) and fφ =
d
dt (fφ̇ ) which yield:

2 2 d 2 d 2 2
2R sin(θ) cos(θ)φ̇ = dt (2R θ̇) 0= 2R sin (θ)φ̇ .
dt
We find a constant of motion L = 2R2 sin2 (θ)φ̇ inserting this in the equation for the azmuthial
angle θ yields:
2 2 d 2 d 2 2
2R sin(θ) cos(θ)φ̇ = dt (2R θ̇) 0= 2R sin (θ)φ̇ .
dt
If you can solve these and demonstrate through some reasonable argument that the solutions are
great circles then I will give you points. I have some solutions but nothing looks too pretty.
13.8 Lagrangian mechanics

13.8.1 basic equations of classical mechanics summarized
Classical mechanics is the study of massive particles at relatively low velocities. Let me refresh
your memory about the basics equations of Newtonian mechanics. Our goal in this section will be
to rephrase Newtonian mechanics in the variational langauge and then to solve problems with the
Euler-Lagrange equations. Newton’s equations tell us how a particle of mass m evolves through
time according to the net-force impressed on m. In particular,
d2~r
m = F~
dt2
If m is not constant then you may recall that it is better to use momentum P~ = m~v = m d~
r
dt to
set-up Newton’s 2nd Law:
dP~
= F~
dt
In terms of components we have a system of differential equations with indpendent variable time
t. If we use position as the dependent variable then Newton’s 2nd Law gives three second order
ODEs,
mẍ = Fx mÿ = Fy mz̈ = Fz
where ~r = (x, y, z) and the dots denote time-derivatives. Moreover, F~ =< Fx , Fy , Fz > is the sum
of the forces that act on m. In contrast, if you work with momentum then you would want to solve
six first order ODEs,
P˙x = Fx P˙y = Fy P˙z = Fz
and Px = mẋ, Py = mẏ and Pz = mż. These equations are easiest to solve when the force is
not a function of velocity or time. In particular, if the force F~ is conservative then there exists a
potential energy function U : R3 → R such that F~ = −∇U . We can prove that in the case the
force is conservative the total energy is conserved.
13.8.2 kinetic and potential energy, formulating the Lagrangian

Recall the kinetic energy is T = 12 m||~v ||2 , in Cartesian coordinates this gives us the formula:
1
T = m(ẋ2 + ẏ 2 + ż 2 ).
2
If F~ is a conservative force then it is independent of path so we may construct the potential energy
function as follows: Z ~
r
U (~r) = − F~ · d~r
O
Here O is the origin for the potential and we can prove that the potential energy constructed in
this manner has F~ = −∇U . We can prove that the total (mechanical) energy E = T + U for
a conservative system is a constant; dE/dt = 0. Hopefully these comments are at least vaguely
familiar from some physics course in your distant memory. If not relax, calculationally this chapter
is self-contained, read onward.
We already calculated that if we use T as the Lagrangian then the Euler-Lagrange equations
produce Newton’s equations in the case that the force is zero (see 13.5.1). Suppose that we define
the Lagrangian to be L = T −U for a system governed by a conservative force with potential energy
function U . We seek to prove the Euler-Lagrange equations are precisely Newton’s equations for
this conservative system1 Generically we have a Lagrangian of the form
1
L(x, y, z, ẋ, ẏ, ż) = m(ẋ2 + ẏ 2 + ż 2 ) − U (x, y, z).
2
1
don’t mistake this example as an admission that Lagrangian mechanics is limited to conservative systems. Quite
the contrary, Lagrangian mechanics is actually more general than the orginal framework of Newton!
13.8. LAGRANGIAN MECHANICS 329
R
We wish to find extrema for the functional S = L(t) dt. This yields three sets of Euler-Lagrange
equations, one for each dependent variable x, y or z

d ∂L ∂L d ∂L ∂L d ∂L ∂L
= = = .
dt ∂ ẋ ∂x dt ∂ ẏ ∂y dt ∂ ż ∂z
Note that ∂L ∂L ∂L
∂ ẋ = mẋ, ∂ ẏ = mẏ and ∂ ż = mż. Also note that
∂L
∂x = − ∂U
∂x = Fx ,
∂L
∂y = − ∂U
∂y = Fy
and ∂L ∂U
∂z = − ∂z = Fz . It follows that
mẍ = Fx mÿ = Fy mz̈ = Fz .
Of course this is precisely m~a = F~ for a net-force F~ =< Fx , Fy , Fz >. We have shown that
Hamilton’s principle reproduces Newton’s Second Law for conservative forces. Let me take a
moment to state it.
Definition 13.8.1. Hamilton’s Principle:
If a physical system has generalized coordinates qj with velocities q˙j and Lagrangian L =
T − U then the solutions of physics will minimize the action S defined below:
Z t2
S= L(qj , q˙j , t) dt
t1
Mathematically, this means the variation δS = 0 for physical trajectories.

This is a necessary condition for solutions of the equations of physics. Sufficient conditions are
known, you can look in any good variational calculus text. You’ll find analogues to the second
derivative test for variational differentiation. As far as I can tell physicists don’t care about this
logical gap, probably because the solutions to the Euler-Lagrange equations are the ones for which
they are looking.
13.8.3 easy physics examples

Now, you might just see this whole exercise as some needless multiplication of notation and for-
malism. After all, I just told you we just get Newton’s equations back from the Euler-Lagrange
equations. To my taste the impressive thing about Lagrangian mechanics is that you get to start
the problem with energy. Moreover, the Lagrangian formalism handles non-Cartesian coordinates
with ease. If you search your memory from classical mechanics you’ll notice that you either do
constant acceleration, circular motion or motion along a line. What if you had a particle con-
strained to move in some frictionless ellipsoidal bowl. Or what if you had a pendulum hanging off
another pendulum? How would you even write Newtons’ equations for such systems? In contrast,
the problem is at least easy to set-up in the Lagrangian approach. Of course, solutions may be less
easy to obtain.
Example 13.8.2. Projectile motion: take z as the vertical direction and suppose a bullet is fired
with initial velocity vo =< vox , voy , voz >. The potential energy due to gravity is simply U = mgz
and kinetic energy is given by T = 21 m(ẋ2 + ẏ 2 + ż 2 ). Thus,
1
L = m(ẋ2 + ẏ 2 + ż 2 ) − mgz
2
Euler-Lagrange equations are simply:

d d d ∂
mẋ = 0 mẏ = 0 mż = (−mgz) = −mg.
dt dt dt ∂z
Integrating twice and applying initial conditions gives us the (possibly familiar) equations
x(t) = xo + vox t, y(t) = yo + voy t, z(t) = zo + voz t − 21 gt2 .
Example 13.8.3. Simple Pendulum: let θ denote angle measured off the vertical for a simple
pendulum of mass m and length l. Trigonmetry tells us that
x = l sin(θ) y = l cos(θ) ⇒ ẋ = l cos(θ)θ̇ y = −l sin(θ)θ̇
Thus T = 12 m(ẋ2 + ẏ 2 ) = 21 ml2 θ̇2 . Also, the potential energy due to gravity is U = −mgl cos(θ)
which gives us
1
L = ml2 θ̇2 + mgl cos(θ)
2
Then, the Euler-Lagrange equation in θ is simply:

d ∂L ∂L d g
= ⇒ (ml2 θ̇) = −mgl sin(θ) ⇒ θ̈ + sin(θ) = 0.
dt ∂ θ̇ ∂θ dt l
In the small angle approximation,

p sin(θ) = θ then we have the solution θ(t) = θo cos(ωt + φo ) for
angular frequency ω = g/l
Chapter 14
leftover manifold theory
In this short chapter I collect a few discussions which extend constructions which we have only
considered for low-dimensional cases. In addition, I include a few other calculations which I saw fit
to omit from the 2013 notes as they did not fit the narrative this semester. I leave them here for
the interested reader, you don’t need to print these for the 2013 offering of Advanced Calculus.
I intend to give you a fairly accurate account of the modern definition of a manifold1 . In a nutshell,
a manifold is simply a set which allows for calculus locally. Alternatively, many people say that
a manifold is simply a set which is locally ”flat”, or it locally ”looks like Rn ”. This covers most
of the objects you’ve seen in calculus III. However, the technical details most closely resemble the
parametric view-point.
1
the definitions we follow are primarily taken from Burns and Gidea’s Differential Geometry and Topology With
a View to Dynamical Systems, I like their notation, but you should understand this definition is known to many
authors
331
332 CHAPTER 14. LEFTOVER MANIFOLD THEORY
14.1 manifold defined

The definition I offer here is given in terms of patches, you can also state it in terms of charts. We
primarily used charts earlier in these notes.
Definition 14.1.1.
We define a smooth manifold of dimension m as follows: suppose we are given a set M ,
a collection of open subsets Ui of Rm , and a collection of mappings φi : Ui ⊆ Rm → Vi ⊆ M
which satisfies the following three criteria:
1. each map φi : Ui → Vi is injective
2. if Vi ∩ Vj 6= ∅ then there exists a smooth mapping
θij : φ−1 −1
j (Vi ∩ Vj ) → φi (Vi ∩ Vj )
such that φj = φi ◦ θij
3. M = ∪i φi (Ui )
Moreover, we call the mappings φi the local parametrizations or patches of M and the
space Ui is called the parameter space. The range Vi together with the inverse φ−1 i is
called a coordinate chart on M . The component functions of a chart (V, φ−1 ) are usually
denoted φ−1 1 2 m j
i = (x , x , . . . , x ) where x : V → R for each j = 1, 2, . . . , m. .
We could add to this definition that i is taken from an index set I (which could be an infinite
set). The union given in criteria (3.) is called a covering of M . Most often, we deal with finitely
covered manifolds. You may recall that there are infinitely many ways to parametrize the lines
or surfaces we dealt with in calculus III. The story here is no different. It follows that when we
consider classification of manifolds the definition we just offered is a bit lacking. We would also like
to lump in all other possible compatible parametrizations. In short, the definition we gave says a
manifold is a set together with an atlas of compatible charts. If we take that atlas and adjoin
to it all possible compatible charts then we obtain the so-called maximal atlas which defines a
differentiable structure on the set M . Many other authors define a manifold as a set together
with a differentiable structure. That said, our less ambtious definition will do.
We should also note that θij = φ−1

i
◦ φ hence θ
j
−1 −1 ◦
ij = (φi φj )−1 = φ−1
j
◦ φ = θ . The functions
i ji
θij are called the transition functions of M. These explain how we change coordinates locally.
I now offer a few examples so you can appreciate how general this definition is, in contrast to the
level-set definition we explored previously. We will recover those as examples of this more general
definition later in this chapter.
Example 14.1.2. Let M = Rm and suppose φ : Rm → Rm is the identity mapping ( φ(u) = u for
all u ∈ Rm ) defines the collection of paramterizations on M . In this case the collection is just one
14.1. MANIFOLD DEFINED 333
mapping and U = V = Rm , clearly φ is injective and V covers Rm . The remaining overlap criteria
is trivially satisfied since there is only one patch to consider.
Example 14.1.3. Let U = Rm and suppose po ∈ Rp then φ : U → Rp ×Rm defined by φ(u) = po ×u

makes φ(U ) = M an m-dimensional manifold. Again we have no overlap and the covering criteria
is clearly satisfied so that leaves injectivity of φ. Note φ(u) = φ(u0 ) implies po × u = po × u0 hence
u = u0 .
Example 14.1.4. Suppose V is an m-dimensional vector space over R with basis β = {ei }ni=1 .
Define φ : Rm → V as follows, for each u = (u1 , u2 , . . . , um ) ∈ Rm
φ(u) = u1 e1 + u2 e2 + · · · + um em .
Injectivity of the map follows from the linear independence of β. The overlap criteria is trivially
satisfied. Moreover, span(β) = V thus we know that φ(Rm ) = V which means the vector space is
covered. All together we find V is an m-dimensional manifold. Notice that the inverse of φ of the
coordinate mapping Φβ from out earlier work and so we find the coordinate chart is a coordinate
mapping in the context of a vector space. Of course, this is a very special case since most manifolds
are not spanned by a basis.
You might notice that there seems to be little contact with criteria two in the examples above.
These are rather special cases in truth. When we deal with curved manifolds we cannot avoid it
any longer. I should mention we can (and often do) consider other coordinate systems on Rm .
Moreover, in the context of a vector space we also have infinitely many coordinate systems to
use. We will have to analyze compatibility of those new coordinates as we adjoin them. For the
vector space it’s simple to see the transition maps are smooth since they’ll just be invertible linear
mappings. On the other hand, it is more work to show new curvelinear coordinates on Rm are
compatible with Cartesian coordinates. The inverse function theorem would likely be needed.
Example 14.1.5. Let M = {(cos(θ), sin(θ)) | θ ∈ [0, 2π)}. Define φ1 (u) = (cos(u) sin(u)) for
all u ∈ (0, 3π/2) = U1 . Also, define φ2 (v) = (cos(v) sin(v)) for all v ∈ (π, 2π) = U2 . Injectivity
follows from the basic properties of sine and cosine and covering follows from the obvious geometry
of these mappings. However, overlap we should check. Let V1 = φ1 (U1 ) and V2 = φ2 (U2 ). Note
V1 ∩ V2 = {(cos(θ), sin(θ)) | π < θ < 3π/2}. We need to find the formula for
θ12 : φ−1 −1
2 (V1 ∩ V2 ) → φ1 (V1 ∩ V2 )
In this example, this means

θ12 : (π, 3π/2) → (π, 3π/2)
Example 14.1.6. Let’s return to the vector space example. This time we want to allow for all
possible coordinate systems. Once more suppose V is an m-dimensional vector space over R. Note
that for each basis β = {ei }ni=1 . Define φβ : Rm → V as follows, for each u = (u1 , u2 , . . . , um ) ∈ Rm
φβ (u) = u1 e1 + u2 e2 + · · · + um em .
Suppose β, β 0 are bases for V which define local parametrizations φβ , φβ 0 respective. The transition
functions θ : Rm → Rm are given by
θ = φ−1
β
◦φ 0
β
Note θ is the composition of linear mappings and is therefore a linear mapping on Rm . It follows
that θ(x) = P x for some M ∈ GL(m) = {X ∈ R m×m | det(X) 6= 0}. It follows that θ is a smooth
mapping since each component function of θ is simply a linear combination of the variables in Rm .
Let’s take a moment to connect with linear algebra notation. If θ = φ−1 β β
−1
◦ φ 0 then θ ◦ φ 0 = φ
β
−1
β
hence θ ◦ Φβ 0 = Φβ as we used Φβ : V → Rm as the coordinate chart in linear algebra and φ−1 β = Φ β.
Thus, θ ◦ Φβ 0 (v) = Φβ (v) implies P [v]β 0 = [v]β . This matrix P is the coordinate change matrix from
linear algebra.
The contrast of Examples 14.1.3 and 14.1.6 stems in the allowed coordinate systems. In Example
14.1.3 we had just one coordinate system whereas in Example 14.1.6 we allowed inifinitely many. We
could construct other manifolds over the set V . We could take all coordinate systems that are of a
particular type. If V = Rm then it is often interesting to consider only those coordinate systems for
which the Pythagorean theorem holds true, such coordinates have transition functions in the group
of orthogonal transformations. Or, if V = R4 then we might want to consider only inertially related
coordinates. Inertially related coordinates on R4 preserve the interval defined by the Minkowski
product and the transition functions form a group of Lorentz transformations. Orthogonal matrices
and Lorentz matrices are simply the matrices of the aptly named transformations. In my opinion
this is one nice feature of saving the maximal atlas concept for the differentiable structure. Manifolds
as we have defined them give us a natural mathematical context to restrict the choice of coordinates.
From the viewpoint of physics, the maximal atlas contains many coordinate systems which are
unnatural for physics. Of course, it is possible to take a given theory of physics and translate
physically natural equations into less natural equations in non-standard coordinates. For example,
look up how Newton’s simple equation F~ = m~a is translated into rotating coordinate systems.
14.1.1 embedded manifolds

The manifolds in Examples 14.1.2, 14.1.3 and 14.1.5 were all defined as subsets of euclidean space.
Generally, if a manifold is a subset of Euclidean space Rn then we say the manifold is embedded
in Rn . In contrast, Examples 14.1.4 and 14.1.6 are called abstract manifolds since the points in
the manifold were not found in Euclidean space2 . If you are only interested in embedded manifolds3
then the definition is less abstract:
Definition 14.1.7. embedded manifold.
We say M is a smooth embedded manifold of dimension m iff we are given a set

M ⊆ Rn such that at each p ∈ M there is a set an open subsets Ui of Rm and open subsets
Vi of M containing p such that the mapping φi : Ui ⊆ Rm → Vi ⊆ M satisfies the following
criteria:
2. each map φi : Ui → Vi is smooth
3. each map φ−1

i : Vi → Ui is continuous
4. the differential dφx has rank m for each x ∈ Ui .
You may identify that this definition more closely resembles the parametrized objects from your
multivariate calculus course. There are two key differences with this definition:
1. the set Vi is assumed to be ”open in M” where M ⊆ Rn . This means that for each point p ∈ Vi
there exists and open n-ball B ⊂ Rn such that B ∩M contains p. This is called the subspace
topology for M induced from the euclidean topology of Rn . No topological assumptions were
given for Vi in the abstract definition. In practice, for the abstract case, we use the charts
to lift open sets to M, we need not assume any topology on M since the machinery of the
manifold allows us to build our own. However, this can lead to some pathological cases so
those cases are usually ruled out by stating that our manifold is Hausdorff and the covering
has a countable basis of open sets4 . I will leave it at that since this is not a topology course.
2. the condition that the inverse of the local parametrization be continuous and φi be smooth
were not present in the abstract definition. Instead, we assumed smoothness of the transition
functions.
One can prove that the embedded manifold of Defintition 14.1.7 is simply a subcase of the abstract
manifold given by Definition 14.1.1. See Munkres Theorem 24.1 where he shows the transition
2
a vector space could be euclidean space, but it could also be a set of polynomials, operators or a lot of other
rather abstract objects.
3
The defition I gave for embedded manifold here is mostly borrowed from Munkres’ excellent text Analysis on
Manifolds where he primarily analyzes embedded manifolds
4
see Burns and Gidea page 11 in Differential Geometry and Topology With a View to Dynamical Systems
functions of an embedded manifold are smooth. In fact, his theorem is given for the case of a
manifold with boundary which adds a few complications to the discussion. We’ll discuss manifolds
with boundary at the conclusion of this chapter.
Example 14.1.8. A line is a one dimensional manifold with a global coordinate patch:
φ(t) = ~ro + t~v
for all t ∈ R. We can think of this as the mapping which takes the real line and glues it in Rn along
some line which points in the direction ~v and the new origin is at ~ro . In this case φ : R → Rn and
dφt has matrix ~v which has rank one iff ~v 6= 0.
Example 14.1.9. A plane is a two dimensional manifold with a global coordinate patch: suppose
~ B
A, ~ are any two linearly independent vectors in the plane, and ~ro is a particular point in the plane,
~ + vB
φ(u, v) = ~ro + uA ~
for all (u, v) ∈ R2 . This amounts to pasting a copy of the xy-plane in Rn where we moved the
origin to ~ro . If we just wanted a little paralellogram then we could restrict (u, v) ∈ [0, 1] × [0, 1],
then we would envision that the unit-square has been pasted on to a paralellogram. Lengths and
angles need not be maintained in this process of gluing. Note that the rank two condition for dφ says
the derivative φ0 (u, v) = [ ∂φ ∂φ ~ ~
∂u | ∂v ] = [A|B] must have rank two. But, this amounts to insisting the
~ B
vectors A, ~ are linearly independent. In the case of R3 this is conveniently tested by computation
~×B
of A ~ which happens to be the normal to the plane.
Example 14.1.10. A cone is almost a manifold, define
φ(t, z) = (z cos(t), z sin(t), z)
for t ∈ [0, 2π] and z ≥ 0. What two problems does this potential coordinate patch φ : U ⊆ R2 → R3
suffer from? Can you find a modification of U which makes φ(U ) a manifold (it could be a subset
of what we call a cone)
The cone is not a manifold because of its point. Generally a space which is mostly like a manifold
except at a finite, or discrete, number of singular points is called an orbifold. Recently, in the
past decade or two, the orbifold has been used in string theory. The singularities can be used to
fit various charge to fields through a mathematical process called the blow-up.
Example 14.1.11. Let φ(θ, γ) = (cos(θ) cosh(γ), sin(θ) cosh(γ), sinh(γ)) for θ ∈ (0, 2π) and γ ∈ R.
This gives us a patch on the hyperboloid x2 + y 2 − z 2 = 1
Example 14.1.12. Let φ(x, y, z, t) = (x, y, z, R cos(t), R sin(t)) for t ∈ (0, 2π) and (x, y, z) ∈ R3 .
This gives a copy of R3 inside R5 where a circle has been attached at each point of space in the two
transverse directions of R5 . You could imagine that R is nearly zero so we cannot traverse these
extra dimensions.
Example 14.1.13. The following patch describes the Mobius band which is obtained by gluing
a line segment to each point along a circle. However, these segments twist as you go around the
circle and the structure of this manifold is less trivial than those we have thus far considered. The
mobius band is an example of a manifold which is not oriented. This means that there is not a
well-defined normal vectorfield over the surface. The patch is:

1 t 1 t 1 t

φ(t, λ) = 1 + 2 λ cos( 2 ) cos(t), 1 + 2 λ sin( 2 ) sin(t), 2 λ sin( 2 )
for 0 ≤ t ≤ 2π and −1 ≤ λ ≤ 1. To understand this mapping better try studying the map evaluated
at various values of t;
φ(0, λ) = (1 + λ/2, 0, 0), φ(π, λ) = (−1, 0, λ/2), φ(2π, λ) = (1 − λ/2, 0, 0)
Notice the line segment parametrized by φ(0, λ) and φ(2π, λ) is the same set of points, however the
orientation is reversed.
Example 14.1.14. A regular surface is a two-dimensional manifold embedded in R3 . We need
φi : Ui ⊆ R2 → S ⊂ R3 such that, for each i, dφi u,v has rank two for all (u, v) ∈ Ui . Moreover, in
this case we can define a normal vector field N (u, v) = ∂u φ × ∂v φ and if we visualize these vectors
as attached to the surface they will point in or out of the surface and provide the normal to the
tangent plane at the point considered. The surface S is called orientable iff the normal vector field
is non-vanishing on S.
14.1.2 diffeomorphism
At the outset of this study I emphasized that the purpose of a manifold was to give a natural
languague for calculus on curved spaces. This definition begins to expose how this is accomplished.
Definition 14.1.15. smoothness on manifolds.
Suppose M and N are smooth manifolds and f : M → N is a function then we say f is
smooth iff for each p ∈ M there exists local parametrizations φM : UM ⊆ Rm → VM ⊆ M
and φN : UN ⊆ Rn → VN ⊆ N such that p ∈ UM and φ−1 N
◦f ◦φ
M is a smooth mapping
m n
from R to R . If f : M → N is a smooth bijection then we say f is a diffeomorphism.
Moreover, if f is a diffeomorphism then we say M and N are diffeomorphic.
In other words, f is smooth iff its local coordinate representative is smooth. It suffices to check one
representative since any other will be related by transition functions which are smooth: suppose
we have patches φ̄M : ŪM ⊆ Rm → V̄M ⊆ M and φ̄N : ŪN ⊆ Rn → V̄N ⊆ N such that p ∈ ŪM ,
φ−1 ◦f ◦φ
−1
M = φN ◦ φ̄N ◦ φ̄−1 ◦ f ◦ φ̄
M ◦ φ̄−1 ◦φ
M
| N {z } | {z } | N {z } | M {z }
local rep. of f trans. f nct. local rep. of f trans. f nct.
follows from the chain rule for mappings. This formula shows that if f is smooth with respect to a
particular pair of coordinates then its representative will likewise be smooth for any other pair of
compatible patches.
Example 14.1.16. Recall in Example 14.1.3 we studied M = {po } × Rm . Recall we have one
parametrization φ : Rm → M which is defined by φ(u) = po × u. Clearly φ−1 (po , u) = u for all
(po , u) ∈ M. Let Rm have Cartesian coordinates so the identity map is the patch for Rm . Consider
the function f = φ : Rm → M, we have only the local coordinate representative φ−1 ◦ f ◦ Id to
consider. Let x ∈ Rm ,
φ−1 ◦ f ◦ Id = φ−1 ◦ φ ◦ Id = Id.
Hence, φ is a smooth bijection from Rm to M and we find M is diffeomorphic to Rm
The preceding example naturally generalizes to an arbitrary coordinate chart. Suppose M is a

manifold and φ−1 : V → Rm is a coordinate chart around the point p ∈ M. We argue that φ−1 is a
diffeomorphism. Once more take the Cartesian coordinate system for Rm and suppose φ̄ : Ū → V̄
is a local parametrization with p ∈ V̄ . The local coordinate representative of φ−1 is simply the
transition function since:
φ̄ ◦ φ−1 ◦ Id = φ̄ ◦ φ−1 .
We find φ−1 is smooth on V ∩ V̄ . It follows that φ−1 is a diffeomorphism since we know transition
functions are smooth on a manifold. We arrive at the following characterization of a manifold: a
manifold is a space which is locally diffeomorphic to Rm .
However, just because a manifold is locally diffeomorphic to Rm that does not mean it is actually
diffeomorphic to Rn . For example, it is a well-known fact that there does not exist a smooth
bijection between the 2-sphere and R2 . The curvature of a manifold gives an obstruction to making
such a mapping.
14.2 tangent space

Since a manifold is generally an abstract object we would like to give a definition for the tangent
space which is not directly based on the traditional geometric meaning. On the other hand, we
should expect that the definition which is given in the abstract reduces to the usual geometric
meaning for the context of an embedded manifold. It turns out there are three common viewpoints.
1. a tangent vector is an equivalence class of curves.
2. a tangent vector is a contravariant vector.
3. a tangent vector is a derivation.
I will explain each case (except derivations, those are discussed earlier in Section 10.2) and we will
find explicit isomorphisms between each language. We assume that M is an m-dimensional smooth
manifold throughout this section.
14.2. TANGENT SPACE 339
14.2.1 equivalence classes of curves

I essentially used case (1.) as the definition for the tangent space of a level-set. Suppose γ :
I ⊆ R → M is a smooth curve with γ(0) = p. In this context, this means that all the local
coordinate representatives of γ are smooth curves on Rm ; that is, for every parametrization φ :
U ⊆ Rm → V ⊆ M containing p ∈ V the mapping5 φ−1 ◦ γ is a smooth mapping from R to
Rm . Given the coordinate system defined by φ−1 , we define two smooth curves γ1 , γ2 on M with
γ1 (0) = γ2 (0) = p ∈ M to be similar at p iff (φ−1 ◦ γ1 )0 (0) = (φ−1 ◦ γ2 )0 (0). If smooth curves
γ1 , γ2 are similar at p ∈ M then we denote this by writing γ1 ∼p γ2 . We insist the curves be
parametrized such that they reach the point of interest at the parameter t = 0, this is not a severe
restriction since we can always reparametrize a given curve which reaches p at t = to by replacing
the parameter with t − to . Observe that ∼p defines an equivalence relation on the set of smooth
curves through p which reach p at t = 0 in their domain.
(i) reflexive: γ ∼p γ iff γ(0) = p and (φ−1 ◦ γ)0 (0) = (φ−1 ◦ γ)0 (0). If γ is a smooth curve on M
with γ(0) = p then clearly the reflexive property holds true.
(ii) symmetric: Suppose γ1 ∼p γ2 then (φ−1 ◦ γ1 )0 (0) = (φ−1 ◦ γ2 )0 (0) hence (φ−1 ◦ γ2 )0 (0) =
(φ−1 ◦ γ1 )0 (0) and we find γ2 ∼p γ1 thus ∼p is a symmetric relation.
(iii) transitive: if γ1 ∼p γ2 and γ2 ∼p γ3 then (φ−1 ◦ γ1 )0 (0) = (φ−1 ◦ γ2 )0 (0) and (φ−1 ◦ γ2 )0 (0) =
(φ−1 ◦ γ3 )0 (0) thus (φ−1 ◦ γ1 )0 (0) = (φ−1 ◦ γ3 )0 (0) which shows γ1 ∼p γ3 .
The equivalence classes of ∼p partition the set of smooth curves with γ(0) = p. Each equivalence
class γ̃ = {β : I ⊆ R → M | β ∼p γ} corresponds uniquely to a particular vector (φ−1 ◦ γ)0 (0) in
Rm . Conversely, given v = (v 1 , v 2 , . . . , v m ) ∈ Rm we can write the equation for a line in Rm with
direction v and base-point φ−1 (p):
~r(t) = φ−1 (p) + tv.
We compose ~r with φ to obtain a smooth curve through p ∈ M which corresponds to the vector v.
In invite the reader to verify that γ = φ ◦ ~r has
(1.) γ(0) = p (2.) (φ−1 ◦ γ)0 (0) = v.
Notice that the correspondence is made between a vector in Rm and a whole family of curves.
There are naturally many curves that share the same tangent vector to a given point.
Moreover, we show these equivalence classes are not coordinate dependent. Suppose γ ∼p β rel-
ative to the chart φ−1 : V → U , with p ∈ V . In particular, we suppose γ(0) = β(0) = p and
(φ−1 ◦ γ)0 (0) = (φ−1 ◦ β)0 (0). Let φ̄−1 : V̄ → Ū , with p ∈ V̄ , we seek to show γ ∼p β relative to the
chart φ̄−1 . Note that φ̄−1 ◦ γ = φ̄−1 ◦ φ ◦ φ−1 ◦ γ hence, by the chain rule,
(φ̄−1 ◦ γ)0 (0) = (φ̄−1 ◦ φ)0 (φ−1 (p))(φ−1 ◦ γ)0 (0)
5
Note, we may have to restrict the domain of φ−1 ◦ γ such that the image of γ falls inside V , keep in mind this
poses no threat to the construction since we only consider the derivative of the curve at zero in the final construction.
That said, keep in mind as we construct composites in this section we always suppose the domain of a curve includes
some nbhd. of zero. We need this assumption in order that the derivative at zero exist.
Likewise, (φ̄−1 ◦ β)0 (0) = (φ̄−1 ◦ φ)0 (φ−1 (p))(φ−1 ◦ β)0 (0). Recognize that (φ̄−1 ◦ φ)0 (φ−1 (p)) is an in-
vertible matrix since it is the derivative of the invertible transition functions, label (φ̄−1 ◦ φ)0 (φ−1 (p)) =
P to obtain:
(φ̄−1 ◦ γ)0 (0) = P (φ−1 ◦ γ)0 (0) and (φ̄−1 ◦ β)0 (0) = P (φ−1 ◦ β)0 (0)
the equality (φ̄−1 ◦ γ)0 (0) = (φ̄−1 ◦ β)0 (0) follows and this shows that γ ∼p β relative to the φ̄ coordi-
nate chart. We find that the equivalence classes of curves are independent of the coordinate system.
With the analysis above in mind we define addition and scalar multiplication of equivalence classes
of curves as follows: given a coordinate chart φ−1 : V → U with p ∈ V , equivalence classes γ˜1 , γ˜2
at p and c ∈ Rm , if γ˜1 has (φ−1 ◦ γ1 )0 (0) = v1 in Rm and γ˜2 has (φ−1 ◦ γ2 )0 (0) = v2 in Rm then we
define
(i) γ˜1 + γ˜2 = α̃ where (φ−1 ◦ α)0 (0) = v1 + v2
(ii) cγ˜1 = β̃ where (φ−1 ◦ β)0 (0) = cv1 .
We know α and β exist because we can simply push the lines in Rm based at φ−1 (p) with directions
v1 + v2 and cv1 up to M to obtain the desired curve and hence the required equivalence class.
Moreover, we know this construction is coordinate independent since the equivalence classes are
indpendent of coordinates.
Definition 14.2.1.
Suppose M is an m-dimensional smooth manifold. We define the tangent space at p ∈ M

to be the set of ∼p -equivalence classes of curves. In particular, denote:
curveTp M = {γ̃ | γ smooth and γ(0) = p}
Keep in mind this is just one of three equivalent definitions which are commonly implemented.
14.2.2 contravariant vectors

If (φ̄−1 ◦ γ)0 (0) = v̄ and (φ−1 ◦ γ)0 (0) = v then v̄ = P v where P = (φ̄−1 ◦ φ)0 (φ−1 (p)). With this in
mind we could use the pair (p, v) or (p, v̄) to describe a tangent vector at p. The cost of using (p, v)
is it brings in questions of coordinate dependence.
The equivalence class viewpoint is at times quite useful, but the definition of vector offered here is
a bit easier in certain respects. In particular, relative to a particular coordinate chart φ−1 : V → U ,
with p ∈ V , we define (temporary notation)
vectTp M = {(p, v) | v ∈ Rm }
14.2. TANGENT SPACE 341
Vectors are added and scalar multiplied in the obvious way:
(p, v1 ) + (p, v2 ) = (p, v1 + v2 ) and c(p, v1 ) = (p, cv1 )
for all (p, v1 , (p, v2 ) ∈ vectTp M and c ∈ R. Moreover, if we change from the φ−1 chart to the φ̄−1 co-
ordinate chart then the vector changes form as indicated in the previous subsection; (p, v) → (p, v̄)
where v̄ = P v and P = (φ̄−1 ◦ φ)0 (φ−1 (p)). The components of (p, v) are said to transform
contravariantly.
Technically, this is also an equivalence class construction. A more honest notation would be to
replace (p, v) with (p, φ, v) and then we could state that (p, φ, v) ∼ (p, φ̄, v̄) iff v̄ = P v and P =
(φ̄−1 ◦ φ)0 (φ−1 (p)). However, this notation is tiresome so we do not pursue it further. I prefer the
notation of the next viewpoint.
14.2.3 dictionary between formalisms

We have three competing views of how to characterize a tangent vector.
1. curveTp M = {γ̃ | γ smooth and γ(0) = p}
2. vectTp M = {(p, v) | v ∈ Rm }
3. derTp M = Dp M
Perhaps it is not terribly obvious how to get a derivation from an equivalence class of curves.
Suppose γ̃ is a tangent vector to M at p and let f, g ∈ C ∞ (p). Define an operator Vp associated to
γ̃ via Vp (f ) = (f ◦ γ)0 (0). Consider, (f +cg) ◦ γ)(t) = (f +cg)(γ(t)) = f (γ(t))+cg(γ(t)) differentiate
at set t = 0 to verify that Vp (f +cg)(p) = (f +cg) ◦ γ)0 (0) = Vp (f )(p)+cVp (g). Furthermore, observe
that ((f g) ◦ γ)(t) = f (γ(t))g(γ(t)) therefore by the product rule from calculus I,
((f g) ◦ γ)0 (t) = (f ◦ γ)0 (t)g(γ(t)) + f (γ(t))(g ◦ γ)0 (t)
hence, noting γ(0) = p we verify the Leibniz rule for Vp ,
Vp (f g) = ((f g) ◦ γ)0 (0) = (f ◦ γ)0 (0)g(p) + f (p)(g ◦ γ)0 (0) = Vp (f )g(p) + f (p)Vp (g)
In view of these calculations we find that Ξ : curveTp M → derTp M defined by Ξ(γ̃) = Vp is

well-defined. Moreover, we can show Ξ is an isomorphism. To be clear, we define:
Ξ(γ̃)(f ) = Vp (f ) = (f ◦ γ)0 (0).
I’ll begin with injectivity. Suppose Ξ(γ̃) = Ξ(β̃) then for all f ∈ C ∞ (p) we have Ξ(γ̃)(f ) = Ξ(β̃)(f )
hence (f ◦ γ)0 (0) = (f ◦ β)0 (0) for all smooth functions f at p. Take f = x : V → U and it follows
that γ ∼p β hence γ̃ = β̃ and we have shown Ξ is injective. Linearity of Ξ must be judged on
the basis of our definition for the addition of equivalence classes of curves. I leave linearity and
surjectivity to the reader. Once those are established it follows that Ξ is an isomorphism and
curveTp M ≈ derTp M.
The isomorphism between vectTp M and derTM was nearly given in the previous subsection. Es-
sentially we can just paste the components from vectTp M onto the partial derivative basis for
derivations. Define Υ : vectTp M → derTp M for each (p, vx ) ∈ vectTp M, relative to coordinates x
at p ∈ M,
X m X m
k k ∂

Υ p, vx e k = vx k
∂x p
k=1 k=1
Note that if we used a different chart y then (p, vx ) → (p, vy ) and consequently
X m X m m
k ∂ k ∂
X
k

Υ p, vy e k = vy k = v .
∂y p ∂xk p
k=1 k=1 k=1
Thus Υ is single-valued on each equivalence class of vectors. Furthermore, the inverse mapping is
simple to write: for a chart x at p,
m
X
−1
Υ (Xp ) = (p, Xp (xk )ek )
k=1
and the value of the mapping above is related contravariantly if we were to use a different chart y
m
X
Υ−1 (Xp ) = (p, Xp (y k )ek ).
k=1
See Equation 10.2 and the surrounding discussion if you forgot. It is not hard to verify that Υ
is bijective and linear thus Υ is an isomorphism. We have shown vectTp M ≈ derTp M. Let us
summarize:
vectTp M ≈ derTp M ≈ curveTp M
It is my custom to assume Tp M = derTp M for most applications. This was the definition we
adopted earlier in these notes.
14.3 Gauss map to a sphere

Suppose S ⊂ R3 is an embedded two-dimensional manifold. In particular suppose S is a regular
surface which means that for each parametrization φ : U → V the normal vector field N (u, v) =
(φu × φv )(u, v) is a smooth non-vanishing vector field on S. Recall that the unit-sphere S2 = {x ∈
R3 | ||x|| = 1} is also manifold, perhaps you showed this in a homework. In any event, the mapping
U : S → S2 defined by
φu × φv
G(u, v) =
||φu × φv ||
provides a smooth mapping from the surface to the unit sphere. The change in G measures how
the normal deflects as we move about the surface S. One natural scalar we can use to quantify
14.3. GAUSS MAP TO A SPHERE 343
that curving of the normal is called the Gaussian curvature which is defined by K = det(dG).
Likewise, we define H = trace(dG) which is the mean curvature of S. If k1 , k2 are the eigen-
values the operator dp G then it is a well-known result of linear algebra that det(dp G) = k1 k2 and
trace(dp G) = k1 + k2 . The eigenvalues are called the principal curvatures. Moreover, it can be
shown that the matrix of dp G is symmetric and a theorem of linear algebra says that the eigenvalues
are real and we can select an orthogonal basis of eigenvectors for Tp S.
~ B,
Example 14.3.1. Consider the plane S with base point ro and containing the vectors A, ~ write
~ + vB
φ(u, v) = ro + uA ~
to place coordinates u, v on the plane S. Calculate the Gauss map,
φu × φv ~×B
A ~
G(u, v) = =
||φu × φv || ~ × B||
||A ~
This is constant on S hence dp G = 0 for each p ∈ S. The curvatures (mean, Gaussian and
principles) are all zero for this case. Makes sense, a plane isn’t curved!
Let me outline how to calculate the curvature directly when G is not trivial. Calculate,
∂(y j ◦ G)

∂ j ∂ j◦
dp G (y ) = y G =
∂xk ∂xk ∂xk
Thus, using the discussion of the preceding section,
2
∂(y j ◦ G) ∂

∂ X
dp G =
∂xk ∂xk ∂y j
j=1

∂(y j ◦ G)
Therefore, the matrix of dp G is the 2 × 2 matrix ∂xk
with respect to the choice of coordinates
x1 , x2 on S and y 1 , y 2 on the sphere.
√
Example 14.3.2. Suppose φ(u, v) = ( u, v, R2 − u2 − v 2 ) parameterizes part of a sphere SR
of radius R > 0. You can calculate the Gauss map and the result should be geometrically obvious:
1 p
G(u, v) = u, v, R2 − u2 − v 2
R
Then the u and v components of G(u, v) are simply u/R and v/R respective. Calculate,
∂ u ∂ u
1
∂u [ R ] ∂v [R] R 0
[dp G] = =
∂ v ∂ v
∂u [ R ] ∂v [ R ]
0 R1
Thus the Gaussian curvature of the sphere K = 1/R2 . The principle curvatures are k1 = k2 = 1/R
and the mean curvature is simply H = 2/R. Notice that as R → ∞ we find agreement with the
curvature of a plane.
Example 14.3.3. Suppose S is a cylinder which is parametrized by φ(u, v) = (R cos u, R sin u, v).
The Gauss map yields G(u, v) = (cos u, sin u, 0). I leave the explicit details to the reader, but it can
be shown that k1 = 1/R, k2 = 0 and hence K = 0 whereas H = 1/R.
The differential is actually easier to frame in the equivalence class curve formulation of Tp M. In
particular, suppose γ̃ = [γ] as a more convenient notation for what follows. In addition, suppose
F : M → N is a smooth function and [γ] ∈ curveTp M then we define dp F : curveTp M →
curveTF (p) N as follows:
dp F ([γ]) = [F ◦ γ]
There is a chain-rule for differentials. It’s the natural rule you’d expect. If F : M → N and
G : N → P then, denoting q = F (p),
dp (G ◦ F ) = dq G ◦ dp F.
The proof is simple in the curve notation:

dq G ◦ dp F ([γ]) = dq G dp F ([γ]) = dq G([F ◦ γ]) = [G ◦ (F ◦ γ)] = dp (G ◦ F )[γ].
You can see why the curve formulation of tangent vectors is useful. It does simply certain questions.
That said, we will insist Tp M = Dp M in sequel.
14.4 tensors at a point

Given a smooth m-dimensional manifold M and a point p ∈ M we have a tangent space Tp M and
a cotangent space Tp M∗ . The set of tensors at p ∈ M is simply the set of all multilinear mappings
on the tangent and cotangent space at p. We again define the set of all type (r, s) tensors to be
Tsr Mp meaning L ∈ Tsr Mp iff L is a multilinear mapping of the form
L : Tp M × · · · × Tp M × Tp M∗ × · · · × Tp M∗ → R.
| {z } | {z }
r copies s copies
Relative to a particular coordinate chart x at p we can build a basis for Tsr Mp via the tensor
product. In particular, for each L ∈ Tsr Mp there exist constants Lji11ij22...i
...js
r
such that
m
∂ ∂
(Lji11ij22...i
...js
X
Lp = )(p)dp xi1 ir
⊗ · · · ⊗ dp x ⊗ j1 ⊗ · · · ⊗ js .
r
∂x p ∂x p
i1 ,...,ir ,j1 ,...,js =1
The components can be calculated by contraction with the appropriate vectors and covectors:

j1 j2 ...js ∂ ∂ j1 js
(Li1 i2 ...ir )(p) = L , . . . , ir , dp x , . . . , dp x .
∂xi1 p ∂x p
14.5. SMOOTHNESS OF DIFFERENTIAL FORMS 345
We can summarize the equations above with multi-index notation:

I i1 i2 ⊗ ir ∂ ∂ ∂
dp x = dp x ⊗ dp x · · · ⊗ dp x and = ⊗ · · · ⊗ js
∂xJ p ∂xj1 p ∂x p

X ∂
Consequently, Lp = LJI (p)dp xI ⊗ J . We may also construct wedge products and build the
∂x p
I,J
exterior algebra as we did for an arbitrary vector space. Given a metric gp ∈ T20 Mp we can calculate
hodge duals in ΛMp . All these constructions are possible at each point in a smooth manifold6 .
14.5 smoothness of differential forms

In this section we apply the results of the previous section on exterior algebra to the vector space
∂
V = Tp M. Recall that { ∂x i |p } is a basis of Tp M and thus the basis {ei } of V utilized throughout
the previous section on exterior algebra will be taken to be
∂
ei = | , 1≤i≤n
∂xi p
in this section. Also recall that the set of covectors {dxi } is a basis of Tp∗ M which is dual to { ∂x
∂
i |p }
j
and consequently the {e } in the previous section is taken to be
ej = dxj , 1≤j≤n
in the present context. With these choices the machinery of the previous section takes over and
one obtains a vector space ∧k (Tp M ) for each 1 ≤ k and for arbitrary p ∈ M . We write ∧k T M
for the set of ordered pairs (p, α) where p ∈ M and α ∈ ∧k (Tp M ) and we refer to ∧k (T M ) as the
k-th exterior power of the tangent bundle T M . There is a projection π : ∧k (T M ) → M defined by
π(p, α) = p for (p, α) ∈ ∧k (T M ). One refers to (∧k T M, π) as a vector bundle for reasons we do not
pursue at this point. To say that α̂ is a section of this vector bundle means that α̂ : M → ∧k (T M )
is a (smooth) function such that α̂(p) ∈ ∧k (Tp M ) for all p ∈ M . Such functions are also called
differential forms, or in this case, k-forms.
Definition 14.5.1. vector field on open subset of Rn .
To say that X is a vector field on an open subset U of M means that

∂ ∂ ∂
X = X1 1
+ X2 2 + · · · Xn n
∂x ∂x ∂x
where X 1 , X 2 , · · · , X n are smooth functions from U into R.
6
I assume M is Hausdorff and has a countable basis, see Burn’s and Gidea Theorem 3.2.5 on page 116.
Note that in this context we implicitly require that differential forms be smooth. To explain this
we write out the requirements more fully below.
If β is a function with domain M such that for each p ∈ M , β(p) ∈ ∧k (Tp M ) then β is called a
differential k-form on M if for all local vector fields X1 , X2 , · · · , Xk defined on an arbitrary open
subset U of M it follows that the map defined by
p → βp (X1 (p), X2 (p), · · · , Xk (p))
is smooth on U . For example if (x1 , x2 , · · · , xn ) is a chart then its domain V ⊂ M is open in M

and the map
p → dp xi
is a differential 1-form on U . Similarly the map
p → dp xi ∧ dp xj
is a differential 2-form on U . Generally if β is a 1-form and (x1 , x2 , · · · , xn ) is a chart then there

are functions (bi ) defined on the domain of x such that
X
β(q) = bi (q)dq xi
µ
for all q in the domain of x.

Similarly if γ is a 2-form on M and x = (x1 , x2 , · · · , xn ) is any chart on M then there are smooth
functions cij on dom(x) such that
n
1 X
γp = cij (p)(dp xi ∧ dp xj )
2
i,j=1
and such that cij (p) = −cji (p) for all p ∈ dom(x).
Generally if α is a k-form and x is a chart then on dom(x)
X 1
αp = ai i ···i (p)(dp xi1 ∧ · · · ∧ dp xik )
k! 1 2 k
where the {ai1 i2 ···ik } are smooth real-valued functions on U = dom(x) and αiσ1 iσ2 ···iσk = sgn(σ)ai1 i2 ···ik ,
for every permutation σ. (this is just a fancy way of saying if you switch any pair of indices it
generates a minus sign).
14.6 tensor fields

Since the tangent and cotangent space are defined at each point in a smooth manifold we can
construct the tangent bundle and cotangent bundle by simply taking the union of all the tangent
or cotangent spaces:
14.6. TENSOR FIELDS 347
Definition 14.6.1. tangent and cotangent bundles.
Suppose M is a smooth manifold the we define the tangent bundle T M and the cotan-
gent bundle T M∗ as follows:
T M = ∪p∈M Tp M and T M∗ = ∪p∈M Tp M∗ .
The cannonical projections π, π̃ tell us where a particular vector or covector are found on the
manifold:
π(Xp ) = p and π̃(αp ) = p
I usually picture this construction as follows:
XXX– add projection pictures
Notice the fibers of π and π̃ are π −1 (p) = Tp M and π̃ −1 (p) = Tp M∗ . Generally a fiber bundle
(E, M, π) consists of a base manifold M, a bundle space E and a projection
π : E → M. A local section of E is a mapping s : V ⊆ M → E such that π ◦ s is injective.
In other words, the image of a section hits each fiber over its domain just once. A section selects
a particular element of each fiber. Here’s an abstract picture of section, I sometimes think of the
section as its image although technically the section is actually a mapping:
XXX– add section picture–

Given the language above we find a natural langauge to define vector and covector-fields on a
manifold. However, for reasons that become clear later, we call a covector-field a differential one-
form.
Definition 14.6.2. tensor fields.
Let V ⊆ M, we define:
1. X is a vector field on V iff X is a section of T M on V
2. α is a differential one-form on V iff α is a section of T M∗ on V .
3. L is a type (r, s) tensor-field on V iff L is a section of Tsr M on V .
We consider only smooth sections and it turns out this is equivalent7 to the demand that the
component functions of the fields above are smooth on V .
7
all the bundles above are themselves manifolds, for example T M is a 2m-dimensional manifold, and as such the
term smooth has already been defined. I do not intend to delve into that aspect of the theory here. See any text on
manifold theory for details.
14.7 metric tensor

I’ll begin by discussing briefly the informal concept of a metric. The calculations given in the first
part of this section show you how to think for nice examples that are embedded in Rm . In such
cases the metric can be deduced by setting appropriate terms for the metric on Rm to zero. The
metric is then used to set-up arclength integrals over a curved space, see my Chapter on Varitional
Calculus from the previous notes if you want examples.
In the second part of this chapter I give the careful definition which applies to an arbitrary manifold.
I include this whole section mostly for informational purposes. Our main thrust in this course is
with the calculus of differential forms and the metric is actually, ignoring the task of hodge duals,
not on the center stage. That said, any student of differential geometry will be interested in the
metric. The problem of paralell transport8 , and the definition and calculation of geodesics9 are
fascinating problems beyond this course.
14.7.1 classical metric notation in Rm

Definition 14.7.1.
The Euclidean metric is ds2 = dx2 + dy 2 + dz 2 . Generally, for orthogonal curvelinear
1 1 1
coordinates u, v, w we calculate ds2 = ||∇u|| 2 2 2
2 du + ||∇v||2 dv + ||∇w||2 dw .
The beauty of the metric is that it allows us to calculate in other coordinates, consider
x = r cos(θ) y = r sin(θ)
For which we have implicit inverse coordinate transformations r2 = x2 + y 2 and θ = tan−1 (y/x).
From these inverse formulas we calculate:
∇r = < x/r, y/r > ∇θ = < −y/r2 , x/r2 >
Thus, ||∇r|| = 1 whereas ||∇θ|| = 1/r. We find that the metric in polar coordinates takes the form:
ds2 = dr2 + r2 dθ2

Physicists and engineers tend to like to think of these as arising from calculating the length of
infinitesimal displacements in the r or θ directions. Generically, for u, v, w coordinates
1 1 1
dlu = du dlv = dv dlw = dw
||∇u|| ||∇v|| ||∇w||
and ds2 = dl2u + dl2v + dl2w . So in that notation we just found dlr = dr and dlθ = rdθ. Notice then
that cylindircal coordinates have the metric,
ds2 = dr2 + r2 dθ2 + dz 2 .

8
how to move vectors around in a curved manifold
9
curve of shortest distance on a curved space, basically they are the lines on a manifold
14.7. METRIC TENSOR 349
For spherical coordinates x = r cos(φ) sin(θ), y = r sin(φ) sin(θ) and z = r cos(θ) (here 0 ≤ φ ≤ 2π
and 0 ≤ θ ≤ π, physics notation). Calculation of the metric follows from the line elements,
dlr = dr dlφ = r sin(θ)dφ dlθ = rdθ
Thus,
ds2 = dr2 + r2 sin2 (θ)dφ2 + r2 dθ2 .
We now have all the tools we need for examples in spherical or cylindrical coordinates. What about
other cases? In general, given some p-manifold embedded in Rn how does one find the metric on
that manifold? If we are to follow the approach of this section we’ll need to find coordinates on
Rn such that the manifold S is described by setting all but p of the coordinates to a constant.
For example, in R4 we have generalized cylindircal coordinates (r, φ, z, t) defined implicitly by the
equations below
x = r cos(φ), y = r sin(φ), z = z, t=t
On the hyper-cylinder r = R we have the metric ds2 = R2 dθ2 + dz 2 + dw2 . There are mathemati-
cians/physicists whose careers are founded upon the discovery of a metric for some manifold. This
is generally a difficult task.
14.7.2 metric tensor on a smooth manifold

A metric on a smooth manifold M is a type (2, 0) tensor field on M which is at each point p a
metric on Tp M. In particular, g is a metric iff g makes the assignment p → gp for each p ∈ M where
the mapping gp : Tp M × Tp M → R is a metric. Recall the means gp is a symmetric, nondegenerate
bilinear form on Tp M. Relative to a particular coordinate system x at p we write
X
g= gij dxi ⊗ dxj
i,j
In this context gij : V → R are assumed to be smooth functions, the values may vary from point to
point in V . Furthermore, we know that gij = gji for all i, j ∈ Nm and the matrix [gij ] is invertible
by the nondegneracy of g. Recall
Pm we use the notation g ij for components of the inverse matrix, in
particular we suppose that k=1 gik g kj = δij .
Recall that according to Sylvester’s theorem we can choose coordinates at some point p which
will diagonalize the metric and leave diag(gij ) = {−1, −1, . . . , −1, 1, 1, . . . , 1}. In other words, we
can orthogonalize the coordinate basis at a paricular point p. The interesting feature of a curved
manifold M is that as we travel away from the point where we straightened the coordinates it is
generally the case the components of the metric will not stay diagonal and constant over the whole
coordinate chart. If it is possible to choose coordinates centered on V such that the coordinates are
constantly orthogonal with respect the metric over V then the manifold M is said to be flat on V .
Examples of flat manifolds include Rm , cylinders and even cones without their point. A manifold
is said to be curved if it is not flat. The definition I gave just now is not probably one you’ll find
in a mathematics text10 . Instead, the curvature of a manifold is quantified through various tensors
which are derived from the metric and its derivatives. In particular, the Ricci and Riemann tensors
are used to carefully characterize the geometry of a manifold. It is very tempting to say more
about the general theory of curvature, but I will resist. If you would like to do further study I can
recommend a few books. We will consider some geometry of embedded two-dimensional manifolds
in R3 . That particular case was studied in the 19-th century by Gauss and others and some of the
notation below goes back to that time.
Example 14.7.2. Consider a regular surface S which has a global parametrization φ : U ⊆ R2 →

S ⊆ R3 . In the usual notation in R3 ,
φ(u, v) = (x(u, v), y(u, v), z(u, v))
Consider a curve γ : [0, 1] → S we can calculate the arclength of γ via the usal calculation in R3 .
The magnitude of velocity γ 0 (t) is ||γ 0 (t)|| and naturally this gives us ds 0
dt hence ds = ||γ (t)||dt and
the following integral calculates the length of γ,
Z 1
sγ = ||γ 0 (t)||dt
0
Since γ[0, 1] ⊂ S it follows there must exist some two-dimesional curve t → (u(t), v(t)) for which
γ(t) = φ(u(t), v(t)). Observe by the chain rule that

0 ∂x du ∂x dv ∂y du ∂y dv ∂z du ∂z dv
γ (t) = + , + , +
∂u dt ∂v dt ∂u dt ∂v dt ∂u dt ∂v dt
du dv
We can calculate the square of the speed in view of the formula above, let dt = u̇ and dt = v̇,
||γ 0 (t)||2 = x2u u̇2 + 2xu xv u̇v̇ + x2v v̇ 2 ,

yu2 u̇2 + 2yu yv u̇v̇ + yv2 v̇ 2 ,
zu2 u̇2 + 2zu zv u̇v̇ + zv2 v̇ 2

(14.1)
Collecting together terms which share either u̇2 , u̇v̇ or v̇ 2 and noting that x2u + yu2 + zu2 = φu · φu ,
xu xv + yu yv + zu zv = φu · φv and x2v + yv2 + zv2 = φv · φv we obtain:
||γ 0 (t)||2 = φu · φu u̇2 + φu · φv u̇v̇ + φv · φv v̇ 2
Or, in the notation of Gauss, φu · φu = E, φu · φv = F and φv · φv = G hence the arclength on S is

given by Z 1p
sγ = E u̇2 + 2F u̇v̇ + Gv̇ 2 dt
0
10
this was the definitin given in a general relativity course I took with the physicisist Martin Rocek of SUNY Stony
Brook. He then introduced non-coordinate form-fields which kept the metric constant. I may find a way to show you
some of those calculations at the end of this course.
14.7. METRIC TENSOR 351
We discover that on S there is a metric induced from the ambient euclidean metric. In the current
coordinates, using (u, v) = φ−1 ,
g = Edu ⊗ du + 2F du ⊗ dv + Gdv ⊗ dv
p
hence the length of a tangent vector is defined via ||X|| = g(X, X), we calcate the length of a
curve by integrating its speed along its extent and the speed is simply the magnitude of the tangent
vector at each point. The new thing here is that we judge the magnitude on the basis of a metric
which is intrinsic to the surface.
If arclength on S is given by Gauss’ E, F, G then what about surface area?. We know the magnitude
of the cross product of the tangent vectors φu , φv on S will give us the area of a tiny paralellogram
corresponding to a change du in u and dv in v. Thus:
p
dA = ||φu × φv ||2 dudv
√
However, Lagrange’s identity says ||φu × φv ||2 = ||φu ||2 ||φv ||2 − φu · φv hence dA = EF − G2 du dv
and we can calculate surface area (if this integral exists!) via
Z p
Area(S) = EG − F 2 du dv.
U
I make use of the standard notation for double integrals from multivariate calculus and the integra-
tion is to be taken over the domain of the parametrization of S.
Many additional formulas are known for E, F, G and there are entire texts devoted to exploring
the geometric intracies of surfaces in R3 . For example, John Oprea’s Differential Geometry and
its Applications. Theorem 4.1 of that text is the celebrated Theorem Egregium of Gauss which
states the curvature of a surface depends only on the metric of the surface as given by E, F, G. In
particular,
−1 ∂ Ev ∂ Gu
K= √ √ + √ .
2 EG ∂v EG ∂u EG
Where curvature at p is defined by K(p) = det(Sp ) and Sp is the shape operator is defined
by the covariant derivative Sp (v) = −∇v U = −(v(U1 ), v(U2 ), v(U3 )) and U is simply the normal
vector field to S defined by U (u, v) = φu × φv in our current notation.
It turns out there is an easier way to calculate curvature via wedge products. I will hopefully show
how that is done in the next chapter. However, I do not attempt to motivate why the curvature is
called curvature. You really should read something like Oprea if you want those thoughts.
Example 14.7.3. Let M = R4 and choose an atlas of charts which are all intertially related to
the standard Cartesian coordinates on R4 . In other words, we allow coordinates x̄ which can be
obtained
P3 from a Lorentz transformation; x̄ = Λx and Λ ∈ R4×4 such that ΛT ηΛ = η. Define
g = µ,ν=0 ηµν dx ⊗ dx for the standard Cartesian coordinates on R4 . We can show that the
µ ν
metric is invariant as we change coordinates, if you calculate the components of g in some other
coordinate system then you will once more obtain ηµν as the components. This means that if we
can write the equation for the interval between events in one coordinate system then that inter-
val equation must also hold true in any other inertial coordinate system. In particle physics this
is a very useful observation because it means if we want to analyze an relativistic interaction then
we can study the problem in the frame of reference which makes the problem simplest to understand.
In physics a coordinate system if also called a ”frame of reference”, technically there is something
missing from our construction of M from a relativity perspective. As a mathematical model of
spacetime R4 is not quite right. Why? Because Einstein’s first axiom or postulate of special relativity
is that there is no ”preferred frame of reference”. With R4 there certainly is a preferred frame, it’s
impicit within the very definition of the set R4 , we get Cartesian coordinates for free. To eliminate
this convenient set of, according to Einstein, unphysical coordinates you have to consider an affine
space which is diffeomorphic to R4 . If you take modern geometry you’ll learn all about affine space.
I will not pursue it further here, and as a bad habit I tend to say M paired with η is ”minkowski
space”. Technically this is not quite right for the reasons I just explained.
14.8 on boundaries and submanifolds

A manifold with boundary11 is basically just a manifold which has an edge which is also a manifold.
The boundary of a disk is a circle. In fact, in general, a closed ball in (n + 1)-dimensional euclidean
space has a boundary which is the Sn sphere.
The boundary of quadrants I and II of the xy-plane is the x-axis. Or, to generalize this example,
we define the upper-half of Rn as follows:
n
H = {(x1 , x2 , . . . , xn−1 , xn ) ∈ Rn | xn ≥ 0}.
The boundary of H n is the x1 x2 · · · xn−1 -hyperplane which is the solution set of xn = 0 in Rn ; we

can denote the boundary by ∂H n hence, ∂H n = R n−1 × {0}. Furthermore, we define
n
H + = {(x1 , x2 , . . . , xn−1 , xn ) ∈ Rn | xn > 0}.
It follows that H n = H+n ∪ R n−1 × {0}. Note that a subset U of H n is said to be open in H n
iff there exists some open set U 0 ⊆ Rn such that U 0 ∩ H n = U . For example, if we consider R3
then the open sets in the xy-plane are formed from intesecting open sets in R3 with the plane; an
open ball intersects to give an open disk on the plane. Or for R2 an open disks intersected with
the x-axis give open intervals.
Definition 14.8.1.
11
I am glossing over some analytical details here concerning extensions and continuity, smoothness etc... see section
24 of Munkres a bit more detail in the embedded case.
14.8. ON BOUNDARIES AND SUBMANIFOLDS 353
We say M is a smooth m-dimensional manifold with boundary iff there exists a family
{Ui } of open subsets of Rm or H m and local parameterizations φi : Ui → Vi ⊆ M such
that the following criteria hold:
2. if Vi ∩ Vj 6= ∅ then there exists a smooth mapping
θij : φ−1 −1
j (Vi ∩ Vj ) → φi (Vi ∩ Vj )
such that φj = φi ◦ θij
3. M = ∪i φi (Ui )
We again refer to the inverse of a local paramterization as a coordinate chart and often
use the notation φ−1 (p) = (x1 (p), x2 (p), . . . , xm (p)). If there exists U open in Rm such that
φ : U → V is a local parametrization with p ∈ V then p is an interior point. Any point
p ∈ M which is not an interior point is a boundary point. The set of all boundary points
is called boundary of M is denoted ∂M.
A more pragmatic characterization12 of a boundary point is that p ∈ ∂M iff there exists a chart
at p such that xm (p) = 0. A manifold without boundary is simply a manifold in our definition
since the definitions match precisely if there are no half-space-type charts. In the case that ∂M is
nonempty we can show that it forms a manifold without boundary. Moreover, the atlas for ∂M is
naturally induced from that of M by restriction.
Proposition 14.8.2.
Suppose M is a smooth m-dimensional manifold with boundary ∂M 6= ∅ then ∂M is a

smooth manifold of dimension m − 1. In other words, ∂M is an m − 1 dimensional manifold
with boundary and ∂(∂M) = ∅.
Proof: Let p ∈ ∂M and suppose φ : U ⊆ H m → V ⊆ M is a local parametrization containing
p ∈ V . It follows φ−1 = (x1 , x2 , . . . , xm−1 , xm ) : V → U is a chart at p with xm (p) = 0. De-
fine the restriction of φ−1 to xm = 0 by ψ : U 0 → V ∩ (∂M) by ψ(u) = φ(u, 0) where U 0 =
{(u1 , . . . , um−1 ) ∈ R m−1 | (u1 , . . . , um−1 , um ) ∈ U }. It follows that ψ −1 : V ∩ (∂M) → U 0 ⊆ R m−1
is just the first m − 1 coordinates of the chart φ−1 which is to say ψ −1 = (x1 , x2 , . . . , xm−1 ). We
construct charts in this fashion at each point in ∂M. Note that U 0 is open in Rm−1 hence the mani-
fold ∂M only has interior points. There is no parametrization in ∂M which takes a boundary-type
subset half-plane as its domain. It follows that ∂(∂M) = ∅. I leave compatibility and smoothness
of the restricted charts on ∂M to the reader.
Given the terminology in this section we should note that there are shapes of interest which simply
do no fit our terminology. For example, a rectangle R = [a, b] × [c, d] is not a manifold with bound-
12
I leave it to the reader to show this follows from the words in green.
ary since if it were we would have a boundary with sharp edges (which is not a smooth manifold!).
I have not included a full discussion of submanifolds in these notes. However, I would like to
give you some brief comments concerning how they arise from particular functions. In short, a
submanifold is a subset of a manifold which also a manifold in a natural manner. Burns and Gidea
define for a smooth mapping f from a manifold M to another manifold N that
a p ∈ M is a critical point of f if dp f : Tp M → Tf (p) N is not surjective. Moreover, the image
f (p) is called the critical value of f .
b p ∈ M is a regular point of f if p is not critical. Moreover, q ∈ N is called a regular value
of f iff f −1 {q} contains no critical points.
It turns out that:
Theorem 14.8.3.
If f : M → N is a smooth function on smooth manifolds M, N of dimensions m, n
respective and q ∈ N is a regular value of f with nonempty fiber f −1 {q} then the fiber
f −1 {q} is a submanifold of M of dimension (m − n).
Proof: see page 46 of Burns and Gidea. .
The idea of this theorem is a variant of the implicit function theorem. Recall if we are given
G : Rk × Rn → Rn then the local solution y = h(x) of G(x, y) = k exists provided ∂G ∂y is invertible.
But, this local solution suitably restricted is injective and hence the mapping φ(x) = (x, h(x)) is a
local parametrization of a manifold in Rk × Rn . In fact, the graph y = h(x) gives k-dimensional
submanifold of the manifold Rk × Rn . (think of M = Rk × Rn hence m = k + n and m − n = k so
we find agreement with the theorem above at least in the concrete case of level-sets)
Example 14.8.4. Consider f : R2 → R defined by f (x, y) = x2 + y 2 . Calculate df = 2xdx + 2ydy

we find that the only critical value of f is (0, 0) since otherwise either x or y is nonzero and as a
consequence df is surjective. It follows that f −1 {R2 } is a submanifold of R2 for any R > 0. I think
you’ve seen these submanifolds before. What are they?
Example 14.8.5. Consider f : R3 → R defined by f (x, y, z) = z 2 − x2 − y 2 calculate that df =
−2xdx − 2ydy + 2zdz. Note (0, 0, 0) is a critical value of f . Furthermore, note f −1 {0} is the cone
z 2 = x2 + y 2 which is not a submanifold of R3 . It turns out that in general just about anything
can arise as the inverse image of a critical value. It could happen that the inverse image is a
submanifold, it’s just not a given.
Theorem 14.8.6.
If M be a smooth manifold without boundary and f : M → R is a smooth function with a
regular value a ∈ R then f −1 (−∞, a] is a smooth manifold with boundar f −1 {a}.
14.8. ON BOUNDARIES AND SUBMANIFOLDS 355
Example 14.8.7. Suppose f : Rm → R is defined by f (x) = ||x||2 then x = 0 is the only crit-
ical value of f and we find f −1 (−∞, R2 ] is a submanifold with boundary f −1 {r2 }. Note that
f −1 (−∞, 0) = ∅ in this case. However, perhaps you also see B m = f −1 [0, R2 ] is the closed m-ball
and ∂B m = Sm−1 (R) is the (m − 1)-sphere of radius R.
Theorem 14.8.8.
Let M be a smooth manifold with boundary ∂M and N a smooth manifold without bound-
ary. If f : M → N and f |∂M : ∂M → N have regular value q ∈ N then f −1 {q} is a smooth
(m − n)-dimensional manifold with boundary f −1 {q} ∩ ∂M.
This theorem would seem to give us a generalization of the implicit function theorem for some
closed sets. Interesting. Finally, I should mention that it is customary to also allow use the set
L1 = {x ∈ R | x ≤ 0} as the domain of a parametrization in the case of one-dimensional manifolds.

Advanced Calculus 13

Uploaded by

Copyright:

Available Formats

Advanced Calculus 13

Uploaded by

Document Information

Original Description:

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Advanced Calculus 13

Uploaded by

Copyright:

Available Formats

Lecture Notes for Advanced Calculus

introduction and motivations for these notes

X continuity in view of the metric topology in n-space.

X the concept and application of the derivative and differential of a mapping.

X inverse function theorem

X implicit function theorem

X tangent space and normal space via gradients or derivatives of parametrizations

X multivariate Taylor series.

X critical point analysis for multivariate functions

X dual space and the dual basis.

X metric dualities and Hodge duality.

X the work and flux form mappings for R3 .

X basic manifold theory (don0 t let me get too deep, please...)2

X vector fields as derivations.

X differential forms and the exterior derivative

X generalized Stokes’s Theorem.

X push-fowards and pull-backs

X how differential forms and submanifolds naturally geometrize differential equations

allows you to communicate analytical concepts to other like-educated individuals.

James Cook, August 18, 2013.

1 sets, functions and euclidean space 11

3 topology and limits 51

4.6.2 calculus of paths in R3 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98

5 inverse and implicit function theorems 117

6 two views of manifolds in Rn 143

7 critical point analysis for several variables 161

8 convergence and estimation 185

9 multilinear algebra 187

10 calculus with differential forms 225

10.11introduction to geometric differential equations . . . . . . . . . . . . . . . . . . . . . 278

11 Geometry by frames and forms 281

12 Electromagnetism in differential form 297

13 introduction to variational calculus 313

14 leftover manifold theory 331

14.2 tangent space . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 338

sets, functions and euclidean space

1.1 set theory

Definition 1.1.1. real numbers

(A1) addition commutes; a + b = b + a for all a, b ∈ R.

(A2) addition is associative; (a + b) + c = a + (b + c) for all a, b, c ∈ R.

(A3) zero is additive identity; a + 0 = 0 + a = a for all a ∈ R.

(A4) additive inverses; for each a ∈ R there exists −a ∈ R and a + (−a) = 0.

(A5) multiplication commutes; ab = ba for all a, b ∈ R.

(A6) multiplication is associative; (ab)c = a(bc) for all a, b, c ∈ R.

(A7) one is multiplicative identity; a1 = a for all a ∈ R.

(A8) multiplicative inverses for nonzero elements;

(A9) distributive properties; a(b + c) = ab + ac and (a + b)c = ac + bc for all a, b, c ∈ R.

(A10) totally ordered field; for a, b ∈ R:

(i) antisymmetry; if a ≤ b and b ≤ a then a = b.

We often make use of the following standard sets:

natural numbers (positive integers); N = {1, 2, 3, . . . }.

natural numbers up to the number n; Nn = {1, 2, 3, . . . , n − 1, n}.

integers; Z = {. . . , −2, −1, 0, 1, 2, . . . }. Note, Z>0 = N.

non-negative integers; Z≥0 = {0, 1, 2, . . . } = N ∪ {0}.

negative integers; Z<0 = {−1, −2, −3, . . . } = −N.

rational numbers; Q = { pq | p, q ∈ Z, q 6= 0}.

open interval from a to b; (a, b) = {x|a < x < b}.

half-open interval; (a, b] = {x | a < x ≤ b} or [a, b) = {x | a ≤ x < b}.

closed interval; [a, b] = {x | a ≤ x ≤ b}.

ijk = jki = kij = −kji = −jik = −ikj .