Part IV

Planar Pythagorean--hodograph Curves

16
rc-length Parameterization

determin'd by putting it equal to the

Thefluxion of the Length is
the squares of the flurion of the Absciss and
caiaTe-Toot of the s u m of
o f the O r d i n a t e .
Isaac Newton, Flurions (1736)

r(t) (r(t), y(t)) the derivative of arc length

differentiable plane
=
curve
For a

with respect to
the parameter t is given by
s

ds | r ' ( t ) | = Vr2(t) + y2(t).

= (16.1)
dt
it by o(t).
curve, and denote
We call this function the parametric speed of the
c u r v e has the property that o(t) # 0
for all t - in
A regularly-parameterized
general, a point where r(t)) 0 incurs a
=
cusp(tangent reversal) o n the c u r v e .
Ideally, we would like to study curves for which o(t)) = l and hence s = t , i.e.
curves parameterized by a r c length.
We c a n easily achieve this for a straight
But these "simple"
line, and also for a circle (using trigonometric functions).
- w e shall see in §16.1 that no
curve
examples are, unfortunately, exceptional
can be parameterized by rational functions of its arc length.
Failure to achieve the ideal of arc-length or "unit speed" parameterization
cannot make o (t) equal to
should not be for complete despondency. If we
cause
c a n be easily integrated, s o
unity, we can at least try to make it something that
s(t) from (16.1).
We can obtain a closed -form cumulative arc length functio theright in (16.1)
the quantity on
e n r(t) is a degree n polynomial curve, 1). The iudetinite integral
ne square root of a polynomial of degree 2(n
2 (a parabola), aud also the c a s e
Closed-form expression in the case n
=

For higher degrees n,

o f cubics if we appeal to elliptic functions |299|.
however, a closed -form integration is, in general, n o longer possible.
we can circumvent this difficulty by incorporating a special structure into
the wlhich e u s u r e s that the argument of
th Ve hodograph r'(t)= (r'(t), y' ())
u a r e root in (16.1) is the perfect square of a polynoal. Thls property
family of polyuomial
S he Pythaqorean-hodoqraph curves (see $16.3), a
 370 16 Arc length Parameter ization

curvos that among other attractive jproperties-

of their are lengths by simply evaluating a polynouial. 'The r e d
nmit eaact mes
of this book deseribe in detail the lormilations and special pron ainingsureme
clapter
and spatial P'ythagorean hodograph curves, al1gorithms for tho ofpi
and manipulation, and some of their practical applications. lanar

16.1 In Search of an Elusive Ideal

The difforential geometer's

ideal is to parameterize a curve
by its arc
measured from fixed point. With this representation
some
on r(s)ength s.
the intrinsic geometry becomes
exceedingly simple: the
=

tangent (r(«),
are t= dr/ds and K d0/ds, where 6 is the angle that t makesand
=

wit curvatture
direction. However, this "natural parameterization of a
in general.admit closed-form plane
pl. curve doe fixed
expression in terms of elementary functiono
algebraic geometer's notion of an "ideal" parameterization is rather
In algebraic
geometry, curves that can be parameterized in terms of di6e
.e.. rational functions enjoy special distinction. Since "simlple"
can be evaluated any
through a finite sequence of arithmetic operations function that
(see Chap. 3) such is rational
functions' are, in fact, the most general that
evaluate eractly (modulo round off error-see Chap. 12). computers can
For
computer descriptions of geometrical loci, a
perspectives (i.e., curves parameterized by rational reconciliation of these two
would be extremely useful. A trivial functions of the arc length
example is the straight line
(s) to + As
=
y(s) ,
Y0 + uS =

whose direction cosines

satisfy A* +u = 1. Generalizing to curves
polynomials r(s), y(s) of degree n 22 in s, it is obvious that
defined by

a(s) =
r " (s) + y2(s) (16.2)
cannot be
identically equal to unity, since the
a 1on constant
polynomial of degree
argument of the square root
however, that the case 2(n 1) > 2 in s. It is not so odvIO
-

where x(s), y(s) are

allow arc
length rational functions in s does
the proof of this parameterizations either. Despite its innocuous ce,
statement is rather appearai
subtle, involving ideas from ineg ation
theory and complex variables
[187.
Theorem 16.1 It is
a impossible to parameterize anu
straight line, by rational r than
plane curve, o
functions of its arc length.
Proof: Consider a
rational curve

Trigonometric, exponential, e in
prograiming languages logarithmic, and other "ibrary"
functio fu
approrimations of thos nctios
are
actually rational
 16.1 In Search of an Elusi ve Jdeal 371

X(s) Yx)
r(s) W(:) u) (16.3)
W()
ed by polynomialk als W(s), X(s), Y(») with ged(W, X, Y) Constant (since
detin would cancel thcir comuon factors). We aHS1 the curve degrer
otherwise
max(deg(W), deg(A ), deg(Y)) > I s0 ns t excude straight lines
satistics

rentiating
(16.3) we see that, for the speed (16.2) to he identically e1al
nifferen
to 1, the polynomials U= WX' - W'X, V =WY' - W'Y, W mst satisty

U +V2= w. (16.4)
olv U,
U, V,
V, W must
W2 m u s t comprise a Pylhagorcan triple of polynomials. As with
Clearly
les of integers (see Chap. 2), such an equation has only
the Pythagorean triples
tions: we cannot freely choose two of the three polynomials and
pecial" soluti
spe

expect to find a polynomnial solution for the third. Specifically, any polynomial
lution to 16.4) must have the form [292):

U = =(a -6*)c, V = 2 abe, w2 = (a2 +b)c (16.5)

non-zero (real) polynomials a, b, c with ged(a, b) 1. Furthermore. =

for three
if the curve is not a straight line.
a and b cannot both be constants
To define a "unit speed" curve, with arc-length parameterization,
w e must

substitute (16.5) into z = U/W', y = V/W* and integrate this hodograph.

a rational curve, since the integrals of rational
However, this may not yield
functions are not necessarily themselves rational. We will obtain a unit-speed
can find two real n o n - z e r o polynomials a(t) and
b{t)
rational only if we
curve
both constants s u c h that
- relatively prime and not
I(s) =
rs a2 (t) -b°(t) dt =
«(s) -r0
a2(t) +b2(t)
2 a(t)b(t)
1,()
dt = y(s) - y o
(16.6)
Jo a2(t)+b2(t)
are both rational functions of s, where (T0, V0) are integration constants.
that two polynomials a(t), b{t) with
We argue by contradiction. Suppose be found, such that the
gcda,b) constant and max(deg(a), deg(b)) 21a
can
=

rational function fP{t)/q(t).

ntegrals (16.6) a r e both rational. To integrate
e must compute its partial fraction decomposition (see §3.5), of the form

Crs (16.7)

the denominator,
er
Zk are the distinct (real and complex) roots of
wie 1. . .

,
ters the right in
The integrals of the
on

(1 multiplicities m.... . mk.

(16.7) are well known those with s > 1 yield rational ternms ou integration,
 P ' a r a m e t e r i z n t i o n

length
372 16 A
ineur
Iranseelental (i.e., logarithlnie hmic or are
-1
while
enpesios
thos with

|199)
s

Thus if the
inlegral ol (16.7)
I8 to be rational
nngent)
we ust. Inve
lor r 1,...,k.
0
C
i(16.7) are calloet.
of the inverse iiear
terms
thhe residh
Theoefficionts C, 1
p(?)/4(t) at
ils poles 2, (sec h3.5).
funetion
of the rational the ucJinale integral of uln.
The esidues also arise in conputing
"caleulus of resiclies" |235 we ku (t) over
from the that
the entire real line

p )dt - 27i C1
q) Im(z,)>0 (16.8)
-deg{P) 22 (so the integrarnel
when qt) has no real roots and deg(g)
(16.8) is taken over all ti deca
suficiently rapidly
as ||o). The sum im poles
plane.
in the upper half of the complex
of p(t)/q(t) polynomials a(t), b{t) we can
ways choose
Now for two relatively prime
both zero, so that
numbers A and u, not

deg(Aa + pb) < max (deg(a), deg(6)) .

(16.6) are both rational functions, then

If we postulate that the integrals
[Aa(t) + ub(t)]" dt
)- a2(t) + b2 (t)

- ) I,(8) + Mul,() + * +)s

and hence the residues of integrand the
is evidently also a rational function, But this integrand also
must be zero.
(Aa+ub)2/(a2+) at each of its poles since
definite integral to be given by (16.8)
-

satisfies the requirements for its

has no real roots, and by our choice of
gcd(a.b)= 1 by assumption, a +b 2. The vanishing
+ ub)2) 2
A. ju the integrand satisfies deg(a+6) deg((Aa
-

of these residues implies that

rto [Aa(t)+ ub(t) dt =

0,
a ( ) + (0)
- O0

ail
positive value tor
but this is clearly impossible, since the integrand has a

not both and ged(a, b) 1.

a(1), b{t)
=
twhen are zero
Since the supposition that non-zero relatively prime polynomials a(t), O)
exist for which the integrals (16.6) are both rational incurs a contradico
this supposition must be false- the integration of a unit-speed hod08ta
defined by rational functions will always lead us outside the realm of ratio
than
curves. We conclude that it is impossible to create a plane curve, ot
a straight line, parameterized by rational functions of its arc lengtn.
 TG I Searh of an Elnive leal 37%

TOof is speciflically for planar curws, bhit can br extenuderl |1

Theaborv
Consicder a rational space eurve defined by fonur polyenia
s ) , Z() with n= max(deg(W'), deg( N), deg(Y),deg 7))
tant. In order for the sfpreed to be unity.
X. Y, Z) paran
=
E2);

WX'-W'X,
I,
gd( lvnomials WX
nd
WY"- W'Y, iWZ'- w'z, W mst furti
h e l o u

ath quartupe, and hence must be |130 0of the form

a lhthagore

WN'-W'N = u+ --
wY'- WY = 2(uq + vp)
WZ' - W'Z = 2(vg- up).

w2 = u+ + +
We assurme
Pythagorean quartuples of polynomials. see 22.1).
ou
(tor polynomials are relatively prime, since if
these
that

f = gcd(WX'-T W'X, WY' -

W'Y, WZ' -

W'Z. W) # constant

defined
each ot them by fnote that a curve uniquely
is
simply divide under such
by its hodograph, which remains unchanged
dulo translations) since
that (u(t), v{t)) £ (0,0) and (p(t),g(t)) # (0.0)
Ve also assume
division. Wea
and the curve
components y, z vanish identically
hodograph
herwise the
other
to the r-axis.
straight line parallel
to a
degenerates rational functions of the
existence of curves in ° parameterized by
The transtormed into the problem of finding
four polynomials
arc length is then of the integrals
such that all three
ult), vt), p(t),
g(t)

P(t) +v'(t)-P0)_F dt =
r(s) -

ro
Ls) t)+(t)+p*(0) +g*0)
2[u(tg(t) +v(t)p(t)] dt = y(s) - Jo
yS) 2(t)+v2t) +p?(t) +#(t)
2[v(t)g(t) - u(t)pU)dt = :(s) - 0 (16.9)
s) t)+v(0)+P0) +¢0)
exist. with
are rational. Suppose that such polynomials
deg(q))> 1
m = max (deg(u), deg(v), deg(p),
straight line). Now
integrals would be
a
the defined by these
(oTherwise curve
the polynomials
choose three numbers A, p, v such that
We can
always
Ar(t) + vp{t)
-

ug(t)
Au() up(0) + vg{t) bt) =

at) = +

satisfy the conditions

(a(1).b(0)) (0,0).
deg(b) < m and
deg(a) < m,
374 16 Arc length Parameterization

Namely, let un, n , Pn, Gm be the coefficients ofu(t), v{t), p(tN

if (Pmm) = (0, 0) we take A = 0, and (a(t), b(t) # 0(0,0) is easily ), qt). Ther
a suitable choice for jt, v siuee (p(t), q{t))# (0,0). On the other ho
PmPm) # (0,0) we can take
satisfied
Au v = Pn +4 Pmln t ImVm mln-Dr
and the one remaining freedom allows us to ensire that (a(t), b(t))
Now if tlhe three integrals (16.9) are all rational, the indlefinite
b(t)) # (0,0).
gral
a() +(0) dt
I)= u(t) +v2(t) + p*(1)+g(0)*"
=

- *- ) 1 ) Avl,() Aul-(s)
+ -

+ }(a+ +u
must also be rational, and hence the residues at its poles must vanish. Rvs.
residue theorem, this implies that
the

ctoo
a(t) +b() dt=0,
u-(t) + v2(t) +p2(t)+g?(t)
a contradiction, since the integrand is positive for all t. Hence, the supposition
that polynomials u(t), v(t), p(t), q(t) can be found to make all three of the
integrals (16.9) rational is false, and we conclude that no curve in R3 (except
a straight line) can be parameterized by rational functions of its arc length.

16.2 The Rectification of Curves

The rectification (arc-length measurement) of curves has been a problematic

issue from ancient times. The circumference of a circle of unit diameter, for
example, is the transcendental2 number r. The founding of analytic geometry
by Descartes, in his 1637 treatise La Géométrie, did not resolve the problem
Descartes asserted [127] that:
g e o m e t r y should not include lines that are like strings, in that
they are sometimes straight and sometimes curved, since the ratios
betueen straight and curved lines are not known, and I believe cannor
be discovered by human minds, and therefore no conclusion baset upou
such rutios can be accepted as rigorous and exact.
ical"
Descartes made a clear distinction between what he regarded as "geomet
they
curves and "mechanical" curves (we now call the former algebruic, sinee
are definable by finite algebraic equations, and the latter trunscendenta,t h
tey are not). However, his categorical rejection of the possibility of arc e
measurement of curves was soon shown to be erroneous.
by
2
The fact that r cannot be
defined by any finite algebraic equation wasP
Johann Heinrich Lambert (1728 -1777), a colleague of Leonhard Eulet
 16.2 The Rectifiration of Curvea 375

curve
ectifications.
re ct Left: the a r c length of the logarithmie spiral
Early the a r c length
6.1. to the length PQ on the tangent line. Right:
Fi < 0is equal dianeter of the rolling circle.
(16.11) equals four times the
(16.10h ooff tthe cycloid
one
arch

of

transcendental curves (see

curves rectified are actually
a f the earliest
the logaruhmic spral,
defined by the polar equatioon
16.1)-
r(0)=aeko (16.10)

parameterization
with the trigonometric
a ndd the cycloid,
y(0) a(1 cos 0). (16.11)
(0)= a(4 sin6),
=
-

rectified the
Torricelli (1608-1647), a student of Galileo,
Evangelista
Tn 1645 of exhaustion" [56]: he
spiral (16.10) by the Archimedean "method
logarithmic
S 0, the a r c length equals the length of the
tangent
showed that, for - o <
defined by 6 =
0 extended to the point Q =
(0, -a/k)
at the point
P =
(a, 0) result
where it meets
the y-axis namely, a V1+ k-2 (see Fig. 16.1). This
executes a n infinite number
remarkable, since the logarithmic spiral
was quite
before (asymptotically) reaching it!
of gyrations about the origin
is dropped into a hole drilled through
Galileo (207] obserrved that, if a body
execute linear simple harmonic motion
across
the center of the Earth, it will
Torricelli and Newton had
the Earth diameter under the influence of gravity.
due to the earth's rotation
conjectured that, if the initial tangential velocity
into account, the path will be a logarithmic spiral
[21,466-assuming
istaken
the earth's mass exerts gravitational attraction but does not otherwise impede
the motion. This conjecture was argued by Robert
wrong: the correct path, as

Hooke, is an ellipse. The locus (16.10) is also known as the equi-angular spiral,
Since its tangent makes a fixed angle c o t k with the radius vector.
Personne de Roberval
Another curve w a s subsequently rectified by Gilles
602-1675) and Christopher Wren (1632-1723) namely, the cycloicd traced
straight
a fixed point o n a circle of radius a that rolls without slipping
on a

"arch" (0<0< 27) of this

CSee Fig.16.1). They showed that a single cycloi«l the
a 8a. Although it has now lapsed
s lengthground" obscurity, into
Wasproving for new mathematical methods and concepts in the

it caught the attention of all the leading scient ists, and

h century:
rompted international competitions and acrimonious controversies: see g8.3.3
for SCUsSion of its tautochrone and brachistochrone properties
 376 16 Arc length P'araneterization

Fig. 16.2. The cubic (16.12) for various k values, with arc length givo t

It was not long before an algebraic curve succumbed to rectificatin

(16.13).
the scrutiny of Pierre de Ferumat (1601-1665), William Neil (1637-162ndg
Hendrick van Heuraet (1633-1660). This was the cuspidal cubic de6 an

r(t) =
2, y(t) = kts

also known as the "semicubical parabola" (see Fig. 16.2). Its are
(16.12)
measured from t ==0, is an algebraic function of the parameter: length
(9k242+4)3/2 8
s(t) = -
27k2 (16.13)
Ironically, van Heuraet a n associate of Huygens published his results in
an appendix to van Schooten's 1659 Latin version of Descartes, Geometri
Renato Des Cartes. Neil's results also appeared in 1659, in the Tractatus duo.
prior de cycloide, posterior de cissoide published by John Wallis, and Fermat's
work followed in 1660 in De linearum curvarum cum lineis rectis comparatione
dissertatio geometrica- an appendix to a treatise by de Lalouvère (this was
the only publication by Fermat to appear during his lifetime).
Christiaan Huygens (1629-1695), in his Horologium oscillatorium of 1673.
gave a historical account 254) of these rectifications that provoked arguments
over the priority he attributed to van Heuraet and Wren for their discoveries
see Chap. 8 of [242|. This dispute reflects the philosophical importance of the
problem of rectification, which had been considered impossible through long
tradition that originated with Aristotle, was reinforced in the 11th century by'
Ibn Rushd (Averroes), and culminated in Descartes' assertion quoted above.

Huygens' theory of evolutes and involutes, used in the design of his isochronous
pendulum clock (see 8.3.3), offered profound new insights into this age
problem. The cubic (16.12) was recognized as the evolute of a parabold,
the cycloid (8.44) has an identical (displaced) cycloid as its evolute.
All these results preceded the formal development of calculus. Athough
by
the theory of integration resolved the existential issue of curve arc Ieug
defining itfor a (sufficiently smooth) parametric curve («(t),9{t))
(16.14)
st) = V r ( 7 ) +yP(7) dr,
 6.3
act
remained Polynomial P'arametric
w

kWaduction,theeven
eosed tormi reducti
that this
for eurves integral
with
loes not, in
Specd 377
patameter1zation. As
general, adnit
"simple" (polynomial
new field of
stomary in theo diflerential
theoretical argumets to gometry rational) or
t However, assune matured, it
as that. l,ie., the beCame
demonstrated in Gl6.1, s

has only integrand in

parameterization

(other than stra ypothetical existence il. natural

a

raight lines)
this
is
arc
length or

parameteriel by imple inconpatible

Curves

with
funetions.
16.3
Polynomial
Parametric Speed
Athough it
itis
is impossible for
the
polynomial or
rational
lines) t o make
integra
curves
(16.14) identically equal in
(other than straight
can nevertheless be achieved for to unity,
polynomial eurves bysignificant
a
i m p r o v e m e

ment of the square root to be

requiring
the argument

the perfect
nd will then be just square of
the polynomial.
polynomial- rather than the square Sinceofa a

ial- the expression (16.14) root

a po

efinite integral is just

permits a closed-form reduction:
This the
polynomial of degree one
a

higher.
the
is characteristic property of
nce hodograph components r' (t), "' (t)Pythagorean-hodograph (PH) curves.
satisty the Pythagorean equation
(0) + () *(0) =

for some polynomial o(t), and are thus of the

form
(t) =
"(t) v*t), vt)
-

2u(t)u(t), o(t) =u"(t) +u' (t)

where u(t), v{t) are relatively prime polynomials. The arc
obtained by integrating the parametric speed is monotonelength
function
increasing, sinces(t)it
is the integral of a polynomial o(t) = u*(t) +v(t) that is positive for all t. In
motion control applications (e.g-, driving a machine tool, robot end effector.
or measurement probe at a prescribed speed along a curved
path) the digital
controller must repeatedly invert the function s(t), i.e., for any given value
the parameter value t, defined by s(t.)
s.
S, must be accurately and efticiently =

computed. Although s(t) does not in general admit a closed-form inversion.

its monotonicity ensures that accurate and efficient numerical inversion can
be performed using a few Newton-Raphson iterations (see Chap. 29 for a
Comprehensive discussion on the use of PH curves in this context). These
deas are also easily extended from planar curves to space curves.
The polynomial nature of their parametric speed endows PH curves with
many other attractive features. For example, their unit tangents and nornmals
re rational vector functions of the curve parameter. This means, for example,
that the offset curves
ra(t) = r(t) + dn(t)

with unit normaln(t) are rational curves

PH r(t)
tance d from a curve
schenes of
y Can be described exactly within the prevailing representation
 Parameterization
378 16 Are length

neod for heuristiC numerical a .

CAD svstens, without the
Complete details coneerning
this projperty can be
can define
ralionnl adayted frunes ln
found in §17.5.imationst
of spatial PH c u r v e s , o n e
"rotation mnimizing"
frae (see Chap. 30). Planar and snat.nput.
of the energy integral (ia PH
also admit closed form evaluation
a rvey
data they typically yield fairer loci
interpolation of discrete
than "ordinary polynomial curves
with more evethe
s.
curvature distributions Ven
ol PH curvCs, and iIsight into thois
For eficient constructions
and behavior, the adoption of an appropriate computational modo
provide such a model for plan.
properties
The methods of complex analysis PH tical.
curv
while for spatial PH curves we appeal to the algebra of quaternio
nions.

16.4 Algebraically-rectifiable Curves

The arc length s of a parametric curve is said to be an algebraic fm at.

of the parameter t if there exists a bivariate polynomial F(,) such 4146
at
F(s,t) = 0.

This is clearly true of PH curves, since they have the property that s is inucd
a
t. Curves for which s is a more general algebraic function
polynomial in t
were investigated in |385]. Although algebraic functions cannot, in general, ho
specified by simple closed-form expressions, it was shown in |385] that (16.141
is algebraic for a polynomial curve r(t) = (T{t), y{t)) if and only if there exists
a polynomial h(t) such that

(4) +0)] h(t) = h(t). (16.15)

As an immediate consequence, if the arc-length function (16.14) is algebraic,
it must have the simple form

s(t) =
2h(t) + constant.

to
Note that PH curves are subsumed as the special instance corresponding
2

h(t) =
u(r)+(7)dr|
In fact, the cuspidal cubic (16.12) studied by Fermat, Neil, and van Helrat
with
is the sinmplest (non PH) case of such an algebraically-rectifiable curvt

h(t)= (9242 +4)"

2916 k4
385
Moreover, it is the unique non PH cubic that is algebraically rectifiabie
 16.5 Unit
Speed Approximatio8 379
further shown iin 385 that the
It
was
condition (16.15) is satisfiedl
sat by non
lynomials a"(0) + 0) and h(0) if and only if
cost

f other polynon
they are exptxpressible
in terms ofot
f(0) and g(t) in the form

0) SO13/()9(0) + 2/()')1, h0) =f()0) (16.16)

citl may be assumed sequiare Iree. In
where f(0)
orn1 of the
e are
lengtlh lunction for
tenms of f(0) and qt), the rnost
geeral
algebraically roctifiable curves e

s()-29(0)Vr).
details on
on the nature of the
For
further solutions to (16.15), and example of
raically-rectifiable
ulgebrai quartics quinties, see
and
|385

16.5 Unit Speed Approximations

arameterizations of curves by rational functions of the arc length
Since exact paran

impossible, it seems natural to enquire "how close" we can approximate

are
his ideal. Consider, for example, a polynomial curve r(t) of degree n. With
a1,the parameter transformation t e 0,1 7E[0.1] defined by
t= 1-ar
a(1-7) +(1-a)r
oives a rational representation of the same degree, and offers a single degree of
freedom, a, to control the "parameter fHow" over the curve. Using the integral

I =(()-1)? dr (16.17)
as a measure of "closeness" to arc-length parameterization (for which I = 0).
the value of a that minimizes (16.17) can be found [151] as the unique root of
a quadratic equation on (0, 1)-see also [263. Figure 16.3 shows the result of
applying this method to a quadratic Bézier curve (i.e., a parabola segment).
With the original parameterization, the parametric speed varies by a factor of
2 below and above the desired unit speed, o =1. For the optimal rational
parameterization, on the other hand, a remains within 20% of unity over
the entire curve. In general, however, this "optimal parameterization" scheme
ofers rather limited scope for improvement, since fixing the curve degree n
allows only one degree of freedom for the optimization process.
Another approach [154], based on the polynomial are-length function s(t)
0 PH curves, employs the Legendre series to compute a convergent sequence
o(constrained) polynomial approximations t(s),t2(s).... to the inverse of
this function [154|, such that
lim te(s(1)) = 1 for te [0,1],
ko0
380 16 Arc length Parameterization

2.5

2.0
optimurn
tational

1.5

1.0
***'
polynomial
parametenization
....... . . . . .
*****
*
*
*****
*

0.5
polynomial
optimum rational
parametenization 0.0
0.0 0.2 0.4 0.6
0.8

Fig. 16.3. Left: the parameter flow along a quadratic Bezier curve generata
ated by the
original polynomial parameterization and optimal rational parameterizatic
same degree, minimizing (16.17). Right: comparison of parametric speed vo the

given the normalization s E [0,1J. The coefficients of te(s) can be determi.

through closed-form reduction of certain integrals. For sufficiently high k
re parameterized version re(s) = r(ts(s) comes arbitrarily close tothe exact
arc-length parameterization, although it is formally of degree kn.
Methods for approximating arc-length parameterization based on
the s e
of piecewise-polynomial re-parameterizations are described in
[103, 237.34R
17
thagorean-hodograph Curves

translatedto start as
vector of a particle is s0
If the velocity vectors trace out the
center of jorce, then the heads of the
from the considerable antiquity in the history of
a locus of
narticle 's hodograph,
mechanics.
H. Goldstein, Classical Mechanics [213

of a parametric curve r(t) in R"

is just its derivative r'(t).
The hodograph c u r v e r(t) in R"
c u r v e in its o w n right. A polynomial
a s a parametric
regarded c u r v e if the n coordinate components of
its
Pythagorean-hodograph (PH)
is a
elenments of a Pythagorean (n+1)-tuple of
polynomials --i.e.,
hodograph a r e o(t).
coincides with the square of another polynomial
the sum of their squares different approaches
in R' and R" entail quite
Pythagorean-hodograph c u r v e s and quartuples
since Pythagorean polynomial triples
to their characterization,
structures. We a r e concerned here
with just planar
involve disparate algebraic
PH c u r v e s to Part V. A further
PH curves, and defer the treatment of spatial
in Minkowski space R",l with
n
formulations
extension, concerning PH
curve
is addressed in Chap. 24.
space-like and o n e time-like coordinates,
and Bernstein-Bézier representation
The definition, elementary properties,
in terms real variables. In
of planar PH c u r v e s a r e presented below, exclusively
terms of complex variables:
Chap. 19 an alternative formulation is defined in
e n s u r e s its u s e as the basis
the elegance and economy of expression this offers
such as construction of Hermite
torsubsequent planar PH c u r v e algorithms,
of the elastic bending energy integral
interpolants (Chap. 25), computation 27).
(Chap. 26), and solution of the C* spline equations (Chap.

17.1 Planar Pythagorean Hodographs

he key property that distinguishes a planar PH curve r(t) = (r(1), y(t)) fron

is the a pTiori incorporation of a Pythagorean

ordinary" polynomial curve
 Curves
382 17 Pythagorean hodograph

amoly, the components of r(1)

structure in its hodograph
a r e required to satisfy the
conedition

(0)4(0)= a(0)
for some polynonmial a(1). This property 1s chievecl by invokino . (17A
he following
characterization for Pythagorean tripes ol poly1nomials.
conditi0n
Theorem 17.1 7hc Pythagorra1n
a()+b(0)=*(0)
c(1) if and only if they
(17:2
is satisfied by polynomials a(/), b(1),v(1), w(0) in the form be crprCs
can

in terms of other polynonials u(t),

a() = [u"(0)- "(0)] w(t),
b(t) = 2u(t)»{t)w(t),
e(t) = |u(0) + v*(0)] w(1),
(17.3)
uhere u(t) and v(t) are relatively prime.
Proof That the form (17.3) is a sufficient condition for satisfaction of (17
follows immediately from substitution into this equation. To see that it is al
and consider the polynomials
necessary. set w(t) =
gcd(a(t), b(t), c(E))
a(t) =
a(t) b(t) = b(t) (t)=(t)

wt) w(t) w(t)

which are relatively prime and satisty

(t) + P (t)a @(t) =

if a(t). b(t), c(t) are a Pythagorean triple satisfying (17.2). Re-writing this as
Pt) = (0) - ä*(t) = [(t) +ä(t)][E0) - (t)],
since the
we note that (t) +a(t) and t) at) c a n have no c o m m o n roots,
-

Hence.
existence of such roots would imply c o m m o n roots of at), b(t), (t).
a(t), of eren
every root of b(t) must be a root of either (t) + a(t) or c(t) -

multiplicity, and without loss of generality we may write

c(t) +a(t) = 2u(t) and ) -(t) =

2u()
for relatively prinme polynomials u(t) and v(t), such that b2(t) == 4 u ( )

From these three equations, we may deduce that

alt) =
u"(1) - "(0), b(0)= 2u(t)u(t), (t) =
u*() + v°().
and multiplying through by w(t) yields the stated form (17.3).
Pythagorean

This agument from the well known proof for integer

is adapted
triples: see, for exänple, [95. For a more general proof, in the CO
factorization domains of characteristic not equal to 2, see |292|
 17.1 Planar Pythagorran Hodographs 383

with ged(a(t). b{1), r(1))

=

17.1 One can easily

verity that solutions
?emark
arTOspond to taking 7tT) = (Onst ant anl
t a n t c o r r e s p o n d
gei(n(), v(1)} c o n s t a n t
ohutions
so u are called prmitive P'ythagorean triple
Such
(17.3).

in
curve r{1)(r(1), y(1)) is lfind by substitnting
three
rhus.
olanar PH
a planar

nials u(t), v(1), u(1) into the expressions

polynom1als

(174)
) = tu() -v*(1)]w(0). '(1)2n(1)n(!)m(1)
with
otevrating. There
generality in iclentifying r'(t) and /(1)
18 i o loSS of
to
as notei in ,2.2. alternatives
and
b(t), resjpectively. in Theorem 17.1
a(t)and to sati«fy the conveTe

t h e p
als u(t) and v(1) can always be obtained
o l y n o m i a l s

a s s u m e in (17.4) that u(t) and v(t) a r e relatinely prime

ifcation. Also, we non constant c o m m o n fartor o
(u(t). v(t)) constant-Since any
ident =

i.e.. ged(u
be absorbed in w(t). We must also discount certain choices
(t) and v(t) c a n
v(t) that yield "degenerate" PH curves
for w(t). u(t),
lif se(t) = 0 or u(t) =
v{t) =
0, the resulting hodograph r(t) =y/(t) = 0

rather than a continuous locus;

defines a single po1nt
(with constants and at least one of u. v non zero

i f wlt), u(t), v(t) a r e all

w

straight line, trivial PH curve a

we obtain a uniformly-parameterized
not both zero, and w(t) is not a
the constant,

(c)ifu(t) and v(t) a r e constants, is linear but its parametric speed

locusobtained by integrating (17.4) again over intervals delineated
it is multiply-traced
is non-uniform-in general,
roots of w(t);
by odd-multiplicity axis) also
parameterized linear loci (parallel to the
r arise

(d)non-uniformly
if w(t) # 0 and one of u(t) and v(t) is zero.

Henceforth we shall consider only cases where w(t), u(t), v{t) are all non-zero.

and u(t), v(t) are not both constants.

the PH
Remark 17.2 If A =
deg(w(t)) and =max(deg(u(t). deg(r(t)).
A-+24-1.
curveobtained by integrating the hodograph (17.4) of degree
is n =

Lemma 17.1 PH curves of degree n have at most n +3 degrees of frredom,

cs

compared to the 2(n+1) degrees of freedom associated with general polynomial

curves of degree n.

u(). r{t) specitied each

roof: If p max(deg(u(t), v()) the polynomialsassociate
are
=

only \ coetticients
Dat most u +1 coefficients. If A deg(w(t)) assume its leadinng coeticient
we =

wth wt), since without loss of generality we may

unity. Hence, we may freely choose A + 2(/t + 1) coetticiernts when specifving

t ) , w(t) and the integration constants provide two further freedoms

g a total of A + 2/ +4 = n+3 by the preceding rennark.
However, these freedoms are not all available for mauipulating the intrinste
P H curves. Three of them a r e accounted for by choosing
the
origii and
L o n of the coordinate axes, and another two corresponl to treloms t
 384 17 Pythagorcan hodograph Curves

parameterizationsince the substitution t ~ at +b does not alter

or degree of the curve. Discounting these, we see that PH curves of he shape
have n 2 "shape freedons" while general polynomial curves haye
have 2negree3. n
-

17.2 Bézier Control Points of PH Curves

We focus here primarily on hodographs of the form (17.4) with w(t-

ged(u(t), v(t)) = constant-i.e., the primitive Pythagorean hodographs and
hodographs define regular PH curves, satisfying r'(t) # 0 for all t. A Doin
a parametric curve where r'(t) vanishes is a non-regular point- typical on
a
cusp or sudden tangent reversal. The use of a non constant polynomial
in (17.4) incurs cusps (an undesirable feature) on the corresponding PH CUwt)
if w(t) has real roots within the curve parameter domain. PH curves defnod
ine d
by integrating primitive hodographs are of odd degree, n = 24 +1.
The simplest non-trivial PH curves arise from substituting w(t) = 1 and
linear Bernstein-form polynomials

u(t) = uo bo(t) + u1 bi(t), v{t) = vo bg(t) + v1 b}(t)

satisfying uov -u1Uo #0 and (u1 -u0)+(1-vo) # 0, so that u(t), v{t) are
relatively prime and not both constants, into (17.4) to obtain the hodograph

t ) = (u - vo) bG(t) + (ugu - vo1) bi(t) + (u? - vi) bt),

(t) = 2uovo bÙt) + (uov1 t ujvo) b(t) + 2u1v bt).

Integrating this hodograph using (11.7) then yields a PH cubic with Bézier
control points of the form

Pot(u-v,2u02uoto)
P2 P1 +; (uoU1-UoU1,uoV1 + u1Uo)
3

P3P2 + f,2u11), (17.5)

the control point po, defined by the integration constants, being freely chogct
Control polygons of the form (17.5) can also be characterized by intuti
geometrical constraints, that we derive in Chap. 18. Now according to we
discussion of G17.1 the PH cubics possess just one "shape freedom," an we
show in Chap. 18 that this amounts to simply a uniform scaling. Hence the
discount scaling as well as translation, rotation, and re-parameterizat0 as
PH cubics are all segments of a unique curve, known (among other na of
T'schirnhausen's cubic. A complete analysis of this curve, and a discls
its many interesting properties, is deferred to Chap. 18. to
curve design due
Clearly, PH cubiCs are of limited value for free formm mite
their minimal shape flexibilityalthough a method for constructing
 17.2 Bézier Control
Points of PH Curves 385

11sing "double" PH cubics (i.e., pairs of PH cubic segments with

interpolar
res) was described in [l(9). From expressions (17.16) below we note
t h a t t h e c
rVAture of PH cubic segments cannot change sign, i.e., they cannot
u r v a t

dections (since uu'-u'v is simply a constant if u(t), v(!) are linear).

eedoms comparable to those of "ordinary" cubics, with a
exhibi

desire shape
free
we
apability, we must appeal to quintic PH curves.
true inflectional cap.
If

H curves,
define quintic PH we choose quadratic polynomials
)
u(t)= uo bo + u1 bi(t) + uabg(t), »(t) = vob(t) + vy b ) + vab)

integrate, to obtain Bézier

in (17.4)
and
control points of the form

Po + (u6 2u0to),
,

pi

p2 P1 +(u0u1-V0t1,u0U1 +u1to)
p3 P2 + 2 Vi,2u1v1) + Uo2 - vot2,uov2 +u2vo)
1
P4 P3 t u1u2 v1V2, U1 V2 +u2v1) ,

1
PsP4 +(u-v,
5
2u2v2), (17.6)
where po is again arbitrary. In this case, the condition for u(t) and v(t) to be
relatively prime may be phrased as

(u2Uo u g v a ) # 4(ugU1 - u j vo)(u1U2 u 2 1 ) .

(17.7)
The PH quintics are in many respects analogous to the "ordinary" cubics
they may change their sense of curvature, and are capable of interpolating
arbitrary first-order Hermite data, as described in Chap. 25. They are also
the basic components of C2 PH splines (see Chap. 27).
Lemma 17.2 The PH quintic defined by (17.6) has either two real inflections
OT
none, according to whether the quantity
A = (u2vo - uov2) - 4 (ugv1- u1vo) (u1V2 - u2v1)
(17.8)
25
positive or
negative.
Proof: Substituting u(t), v(t) into the numerator k(t) = u(t)v'(t)-u'(t)v(t)
oT expression (17.16) below for the curvature, k(t) is seen to be the quadratic
with Bernstein coefficients

o2(uov1 - u1vo), k1 =-(u2vo- u0v2), k2 =2(u1v2 u2V1).

s diseriminant k-koka is proportional to (17.8), and we observe that A #0

if A 0 there are two real t values
for of the constraint (17.7). Thus, >
w c h K vanishes, while if A < 0 there are none.
386 17 Pythagorean-hodograph Curves
The control point formulae (17.5) and (17.6) characterize PH

quintics in terms of the coefficients of the

two real pol
A significant economy of expression can be realized by interr
olynomials u(t)cuband vt).
and
real and imaginary parts of a single compler polynomial, w(t) =le
preting them
as
the
This approach is developed in Chap. 19, and used u(t)
extenisively thereafte+i
ter.

17.3 Parametric Speed and Arc Length

The parametric speed of a regular PH curve r{t) = (r(t), y(t)) is given
n by

o(t) |r'(t)| =
Va'2(t) +y2() =
u*(t) + v(t).
(17.9)
=

a polynomial in t. If r(t) is of (odd) degree n, ult) and v(t) must be of degre

be written in Bernstein form as
gree
m=
5(n and
-

1) may

u(t) = > ub(t), v(t) = v e b (t).

k=0 k=0

To express (17.9) in the Bernstein form

n-1

a(t) = o b ( t ) , (17.10)
k=0

we invoke the multiplication rule (11.20) for Bernstein-form polynomials. This

gives the coefficients of o(t) in terms of the coefficients of u(t), v(t) as

m m

Ok
min(m,k)
G - k-j+UjUk-j),
(n -1 (4
k =
0,..., n-1.
j=max(0,k-m)
k

For the PH cubics, for example, o(t) is quadratic and has Bernstein coefticiens

Gy u + v, O = UgU1 +UoV1 2 uj +v, (17.11)

while for the PH quintics, a(t) is the quartic with Bernstein coefficients

O1= uoU1 +gU1,

2
2 uj +ví) + (ugu2 + vov2)
(17.12)
3 = u u2 + U1U2, O= u + v.
In order to integrate o(t) and thus obtain the arc length s as a poly

function of the parameter,

 17.3 Parametric Speed and Arc 387
Length

(0) a(T) dr
dhe intcgration rule (1.7) lor the Bernstein basis functions. This gives

) a- (17.13)

where k-1
S0=0 and S
j k=1,...,n.

the total a r c length S is simply

Hence,
O0+o1 + +n-1
S = s(1) = (17.14)

Ta coupute the arc length of any PH curve segment t e [a, b] we need only
b. The
take the difference s(b)-s(a) ofthe polynomial (17.13) evaluated at a,

esnlt is ezact (modulo round-off errors if floating point arithmetic is used),

asdistinct from length computations for "ordinary" polynomial
arc
which require an inherently approximate numerical quadrature.
curves

Similarly, it is much simpler to determine from (17.13) the parameter value

(measured from t =0) has given value i.e..
t, at which the arc length
a s.
-

to solve the equation

s(t.) = S

for t.. Consider, for example, the task of uniform rendering of a parametric
curve. Typically, r(t) is rendered by evaluating at parameter values t o .
..N.
correspondingto a uniform parameter increment At =t-tk-1, k 1,..
However, this yields an uneven spacing (by arc length) of the points r(tk) on
=

the curve, since the parametric speed o (t) is not, in general, constant.
Although the parametric speed of a PH curve is also non-constant, the
simple form (17.13) of s(t) allows us to easily compensate for its variation.
Let to,..., ty be the parameter values of the points uniformly spaced by an
arc-length increment As = S/N, so that

s(tk) = kAs, k = 1,. . . , N - 1 (17.15)

with to = 0 and ty =1. Now since o(t) = ds/dt and o{t) is positive for all t
when GCD(u, v) = 1, s(t) is monotone -increasing with t, and hence for each
equation (17.15) has precisely one simple real root. Clearly, the root ts of
.15) lies between t_i and 1. As an initial approximation to it, we take

As
= tk-1 o(tk-)
 Curves
388 17 Pythagorean hodograjph
As= constant
M = constant

eeoeoeoed
Fig. 17.1. Uniform increments in PH curve parameter (left) and arc
arc llength (right).

and obtain further refinements by applying the Newton-Raphson it.

iteration
_s( r = 1,2,..
a(t-1
As is well known [110], such iterations are quadratically convergent for startin
approximations sufficiently close to tk, and in typical examples the paramete
values tk are obtained to an accuracy of 10** or better in just two or thtee
iterations (see Fig. 17.1) . Since precise uniformity by arc length is probablr
not crucial in many applications, a single iteration often sutfices typicall
this gives uniform spacing to a relative accuracy of about 10-6 or better.
The problem of uniform arc-length rendering of a curve arises naturally in
considering motion at uniform speed along a curve. This is the simplest case
of a broader class of problems addressed by real-time interpolator algorithms
for digital motion controllers. A comprehensive discussion of these problems,
and the advantages of PH curves in solving them, can be found in Chap. 29.

17.4 Differential and Integral Properties

Since the parametric speed of the PH curve r(t) defined by integrating

is the polynomial (17.9) in t, its elementary differential properties-thent
tangent and normal, and the curvature all have a rational dependence ou
the curve parameter. Specifically, they are defined in terms of the polynomas
u(t) and v(t) by

K = 2 UU-u'v
(2 v2, 2uv) n =
(2uv,v u) (17.16)
g2
while
**

For a PH curve, t and n a r e rational functions of degree -

degree n

i s o fdegree n - 3 in the numerator and 2n - 2 in the denominato PH

The fact that the unit normal n(t) is a rational vector functio u r v e s

curve r(t) meaus that its offsets

rational curt
ra(t) at each distane. d a r e f C A D

they admit exact representations in the standard rational Bézier 10 pro*

systems (see §17.5), eliminating the need for data intensive and e
offset curve approximation sehemes [140, 251,252, 280, 362].
 17.5 Rational Ofsets of PH Curves 389

The
ianal curvature uncton i8 another advantage of PH curves over
rational

ordinaryn o
polynomial
l y o m i a l cu
curves, Slnce ntegrals of powers of the curvature: with
a the arc length ds = o dt, defined by
respect to

I " ( ) o() dt for n= 0, 1,2,. (17.17)

luated exactly by a partial fraction decomjposition of the integrand
can be evalua
he 0 gives the total arc length, discussed in G17.3 above.
case n =
§3.5).
(see
=
10/ds, where 0 is the tangent angle with respect to a fixed direction,
Since
t e e r a l I1 defines the net rotation angle (i.e., with anti clockwise rotation
ise rotation) of the tangent along the curve. A modified form
ncelling clockwise

nalyzed in 325.3, with |k| substituted for *, defining the total ahsolute
analyzed
is
ofI
tangent rotatio
rotation - this requires a subdivision of the parameter domain [0. 1]

t values
that entify inflection points of the PH curve.
at the
Finally, the integral h is the bending energy of the curve i.e., the strain
roy stored in a thin, initially straight, elastic beam that is bent into the
shape o fthe
t curve (see $l4.2). Chapter 26 addresses the evaluation of I2 for
DH CUrves in detail. Note that evaluation of the integrals (17.17) by means of
nartial fraction decomposition requires a factorization of the parametric speed
polynomial. Since, by construction, o(t) has no real roots, this factorization
ipvolves only terms that correspond to complex-conjugate root pairs (which
into real quadratic factors).
may be combined

17.5 Rational Offsets of PH Curves

The offsets at each distance d from a PH curve r(t), defined by

ra(t) = r(t) + dn(t), (17.18)

admit exact representation as rational Bézier curves, because the unit normal
n(t) has a rational dependence on the curve parameter t. In the case of cubics
and quintics, the offset curves are of degree five and nine, respectively. Let the
control points of the PH curve r(t) be written in homogeneous coordinates as

P =
(Wk, Xk, Ye) (1,Tk.yk),
= k =
0,.. n.

We define the forward differences of these coordinates by

A P = Pk+1 - Pk = ( 0 , Azk, Ayk), k = 0,. n-1

C T e ATk = Tk+1-Ck, Ayk = Yk+1-Jk, and we set A P ; = (0, J y k rk).

The offset at distance d from the PH curve r(t) is defined by (17.18), where
1Ormal to r(t) is given by (17.16). The offset can be expressed as

(X(1) Y(t) N
ralt)=w' WO)
 Curves
390 17 Pythagorean hodograph

1, whos
polynomials of degree 2n -1, whOse coefficient
where W(0), X(0), Y() are

O= (1V, Xa,Y).
k= 0,.. , 2n 1

rational offset curve.

define the Bézier control points of the
coordinates for the control points of the o . .
Thehomogeneous
the original curve |148.18c1
set may be
terms of those of
concisely expressed in as

min(n-1,k)
Ok (o,Pk-j+dnAP;), k =
0,. . , 2n-1
2n
J=max(0,k-n)
k )
Thus, for PH cubics, the control points of the rational quintic offsets
are

Oo = ooPo + 3d AP,

3aoP1 +
3d (34P +24P)],
O1 =

2a, Po +
O2 =[o2Po + 6a,Pi + 3doP2 + 3d(34P6 +64Pi +AP)].
10

O3 3o2P1+601P2+ OoPs + 3d (AP +64PT +34P)1.

10

O
O 302P2 +201P3 +3d (24P +34P;)],
Os = o2Ps +3d AP
For PH quintics, the rational offsets are of degree 9 with control points

Oo = OoPo + 5d AP% ,

O1 =4a, Po + 5ooP1 +5d (5AP +44P)],

O2 = [3a2Po + 1001P1 +50,P2 +5d (54P +104P +3AP2)].

O3 [20,Po + 15a2P1 +2051P2 +5ooP3

+5d (54P + 204Pt +154P +24P)].
o4Po + 2073P1 + 600,P2 + 4001P3 +500P
126

+5d(5APo +404P +604P+204P+AP4)

Os 105o4Pi
126
+ 4003P2 +60o2P3 +200Pa + ooPs
+5d(AP + 204P + 604P + 40AP+54P)
 17.5 Rational
Ollsets of PH Curves 3991

5oPat 20lo3P"'a +1io,P +20,P

O42
+hd (2AP+15AP; +20AP
+5AP)
50P3 + 1003P t a2P', t 5d (3AP, +10AP + 5AP)I
150P+ doaP's +5l (4AP; + 54AP)I.
9

O=
OP's + 5d AP

17.2 illustrates the control polygons specified by the above fornular,

Figure17.2
ofset
andthe
curves they deline, lor some PH
quinties. Since the offsets are
eurves (each control point having, in general, a different weight W)
ionat

r o l polygons of the offsets are not necessarily intuitive indicators of

of tthe curves they deline.
shape of Figure 17.3 shows that, as d increases, the
oftsetcurve control points move uniformly along straight lines.

Fig, 17.2. Left: Bézier control polygons for degree 9 interior and exterior offsets to a
PH quintic (note that each control point has a different weight). Right: the rational
offsets are exact for every d- even when they develop cusps and self-intersections.

ig.17.3. Control polygons for ofsets at two different distances d from P'tHquint ie:
a
a5 d increases, the offset control points move uniformly along certain straiglt lines.
18
Tschirnhausen's Cubic

I was very much pleased by his manners, and I recognize in that

an outstandng and very promising talent. He showed m e
a017nQ man
i t e a number of [his/ results (inventa), analytic as uwell as geometric
how much
and of reasonably good taste. From this I easily conclude
can be erpected from him

Gottfried Wilhelm Leibniz, letter to Henry Oldenburg

of December 28, 1675 as quoted in 289

Tschirnhaus
18.1 Ehrenfried Walther von

to a n unexpected acquaintance with

The study of planar PH curves brings us
Count Ehrenfried Walther von Tschirnhaust (Fig. 18.1)- a less-prominent
Newton. Born o n April 10, 1651 in
contemporary of Huygens, Leibniz, and
and at
Kieslingswalde (now SBawnikowice in Poland), he was schooled privately
Görlitz Gymnasiumbefore entering the University of Leiden in 1668, where he
studied philosophy, mathematics, and medicine. After completing his studies
in Leiden, he travelled to England carrying a letter of recommendation from
the philosopher Baruch Spinoza to Henry Oldenburg, Secretary of the newly
formed Royal Society of London for the Improvement of Natural Knowledge
(and also editor of its Philosophical Transactions).
From Oldenburg, Tschirnhaus secured further letters of introduction to
close
Huygens and Leibniz in Paris. During 1675-76, Tschirnhaus developed
a

endship with Leibniz and maintained regular correspondence with him after
eaving Paris to visit Italy, where he became engaged in studying the use of
nrrors to achieve very high temperatures by focusing sunlight. Iu this context,

n e occasionally sees [78) his name erroneously rendered as 7'schirnhausen. The

antiquated genitive device in Geruan, facilitating ptonunciat

suffix is ion
en an
as in Tschiruhuusen's cubic.
O1 the possessive form of names ending with an "s
394 18 Tshinhan-on's C'ubir

Fig. 18.1. Left: portrait of Ehrenfried Walther von Tschirnhaus (1651-1708) frr
a small engraving. (Elke Estel, Kupferstich-Kabinett, Staatliche
Kunstsammlunger
Dresden). Right: example of a mirror used by Tschirnhaus to achieve
temperatures
up to 1500 C by focusing sunlight (Jürgen Karpinski, Mathematisch-Physikalisther
Salon. Staatliche Kunstsammlungen Dresden). Reproduced with permission.

he became interested in catacaustics-i.e., the envelopes of parallel light rays

reflected by a mirror. His knowledge of optics played an important role in his
endeavor to develop the manufacture of hard-fired porcelain, in collaboration
with Johann Friedrich Böttger (true porcelain had been produced in China for
several centuries, but the materials and process remained unknown in Europe
at that time). Böttger, an alchemist interested in transmutation of base metals
into gold, was imprisoned in 1700 by Augustus the Strong, Elector of Saxony
who wished to be the exclusive beneficiary of his efforts. Tschirnhaus "saved
Böittger by suggesting they work together on production of porcelain instead.
a collaboration that ledthe now -famous Meissen porcelain.
to
Tschirnhaus died in Dresden on October 11, 1708. Although many-often
tangential- accounts of his abilities, accomplishments, and his appreciation
of contemporary scientific developments are unsympathetic, a less dismissive
contrarian view of his contributions has begun to emerge [2891. By no means
a leading figure in the scientific revolution
engendered by the development o
calculus, Tschirnhaus nevertheless bequeathed some noteworthy results tha
carry his name, In the theory of equations, Tschirnhaus transformations 7
are employed to eliminate certain
lower -order terms from a polynomial pl
degree n. Tschirnhaus proposed this method in his paper "Methodusauferend
onnes
terminos internedios ex data aequatione," Method of eliminating at
or
intermediate terms from a given equation, published in the Acta Erudto
See (146, for an
English translation.
 18.1 Ehrenfried Walther von 'Tsehirnbaus 395

of
1683. Elimination ofthe t -
M a y1 6 8 3 ,
torm from a given degree n polyrnomial
n t ) = ao + + an-1" +a,1" - 0 is cnsily accomplished by the
equa ariables t = T - dn- 1/nan. This was nlrealy kuown to Descartes,
change o f
Tschirnhaus sou ought to remove successive lower order terms.
consider
cubic equatious of the form
He

qt +T, (18.1)
hich the quadratic term has been removed, and he sought to elininate
from wh
term. He achieved this by a transformation t T Cquivalent to
linear
the

t=
2ga 3r +3ar (18.2)
-3a2 - 3T

a is either root of the quadratic equation

where

3 q a - 9ra + q = 0. (18.3)

Substituting (18.2) into (18.1), clearing denominators,

and using (18.3), one

is reduced to the desired form,

finds that the cubic

T =
(272-4q)(2q2-9ra) (18.4)
27q2

quadratic equation, the solutions to the cubic (18.1)

Hence, by first solving a
of
can be obtained by cube
root extractions. Of course, the right-hand side
a is then complex).
(18.4) has a complex value if 27r2-4g3<0(since
t and 7, defines a Möbius
Expression (18.2), extended to complex values of
a general cubic p(t).
transformation (see §4.8) of the complex plane. Given
elimination of the t term is accomplished through a shift along the real axis
that positions the centroid of its three roots at the origin. The Tschirnhaus
roots lie
transform then maps the complex plane in such a way that the
(18.2)
noted in §4.8, MMöbius
symmetrically o n a circle centered at the origin (as
transtormations can map any three points to any other three points).
schirnhaus falls short on the promise, in the title ot his paper, to present

a method that eliminates all intermediate terms from a polynomial cquatiou

PIT)=0 of degree n, thus reducing it to

the form r'" c.
Of course, we now =

OW this is impossible using only arithmetic operations a d root extractions

1824 the brilliant but tragically short lived Norwegian uuathematician

Henrik Abel (1802-1829) published proof that "solution by radicals"

Nels a

studies of the
possible for the quintic equation. Nevertheless, subsequent
Jerrard "reduced" o r
the Bing
C byy Hermite, Klein, and others often use
normal" form,
t+at + b = 0,

traustorinatios.
e d from general quintic by means of Tschirnhaus
a

2003, the Tschirnhaus Gesellschaft was foundedl iu

Dresclen as a
Society
Preserve and promote the achievements of E. W. von Tschirulhaus. The
 396 18 Tschirnhausen's Cubic

society coordinatos the activities of the Mathematisch Physikalischor

the Staatliche Kunstsammlungen Dreselen (Dreslen State Ar
Art Collectior Salon at.
see http://www. tschirnhaus-gesellschaft.de/e_index.htm1.

18.2 Tschirnhaus and Caustic Curves

Tschirnhaus also contributed to the foundations of geometrical optics, thrs

his stndies of caustic curves. Two seminal texts on the geometrical treattme
ment
of light propagation are the Traité de la Lumiére |255| published in 1690
Christiaan Haygens, and the Opticks |346) ofIsaac Newton, published in 704
Huygens employed the wave theory of light, while Newton's exposition
Was
based on light rays (regarded as the paths of discrete "particles" of light)
modern geometrical optics, descriptions of light propagation in terms of rav
and wavefronts are regarded as equally valid, complementary, models.
A fundamental problem of geometrical optics is to describe what happens
when a system of rays or wavefronts, emanating from a finitely or infinitelv
distant point source, is reflected or refracted by a mirror or a lens. Consider
a wavefront W incident on a lens or mirror S (note that W is spherical or
planar according to whether the point source is finitely or infinitely distant).
Prior to reflection/refraction, Huygens' Principle describes the propagation
of W namely, at each subsequent time t, the shape of W is specified by the
offsets (or parallels) to its initial shape. The incident rays are the normals to
this family of parallel wavefronts, and the wavefronts corresponding to fixed
time increments mark off equal distances along the rays.
Suppose W' is the new shape of the wavefront, after refection/refraction
by S. The propagation of W" is again governed by Huygens' Principle, and the
reflected/refracted rays are the normals to the new system of wavefronts, all
parallel to W. These light rays appear to "concentrate" along a certain curve,
that Tschirnhaus called the caustic," from the Greek for burning. The caustic
C is the cnvelope of the family of reflected/refracted rays- i.e., the rays
constitute the family of tangents to the caustic curve. Furthermore, since the
rays are orthogonal to the reflected/refracted wavefront W', the caustic is aso
the evolute - i.e., the locus of centers of curvature of W'. The family of

reflected/refracted wavefronts, parallel to W', are the invoutes of the caustie

C-they all have C as their common locus of centers of curvature. See 8.3
for a review of the geometry of evolutes, involutes, parallel curves, and fanilies
of tangent and normal lines: a more detailed review of their significance in the
context of
geometricaloptics may be found in [159.
Tschirnhaus first discusses caustics in a 1682 Acta Eruditorum paper, the
text of his specch upon admission to the Royal Academy of Sciences in Paris.
Specifically, he considers the caustic for reflection of parallel incident rays

Distinct terms, catucaustic and diacaustic, were originally employed to disting

uish

between caustics by reflection and use.

refraction, but are no longer in cono

 18.2 'T'rhirnhas and C'a1st ir (urves 397

reflection of parallel r a y s by
a

Caustic c u r v e s studied by Tschirnhaus:

Bi, 18.2. reflections a r e considered here
and by a parabola (right). Only primary
circle (left)
before
mirror (Fig. 18.2). This had already been discussed
curve
a Spherical de la Lumiére whose
in 1678, when Huygens read his Traité
the Academy Tschirnhaus w a s
w a s delayed to
1 6 9 0 - b u t the possibility that
publication that he w a s in
has been discounted in view of the fact
guilty of plagiarism direct experience of
read before the Academy, and his
Italy when Huygens
with burning mirrors [289, 390].
caustics through experiments elaborated upon
Eruditorum paper, Tschirnhaus further
In a 1690 Acta 1682
questions concerning his
caustics by reflection, apparently prompted by and hyperbola.
derived caustics for reflection by a circle, parabola,
paper. He the reflected rays." a n
defined by "the crossing of
referring to them a s c u r v e s of
of a curve, regarded a s the envelope
expression that presages the dual form
rather than a locus of points. The envelope
concept
a family of tangent lines Leibniz, but
a 1694 Acta
Eruditorum paper by
was m o r e fully developed in
introduced [390] by Gaspard Mouge
in his
the term was itself apparently first
Applhcation de l'analyse à la géométrie (1795).
18.2. In
of the envelopes apparent in Fig.
Figure 18.3 shows "completions" the locus
caustic is a n cpicycloidi.e..
the case of reflection by a circle, the
rolls without
circumference of a circle that
raced by a chosen point o n the
When the retlecting mirror is of
unit
slipping on the exterior of a fixed circle. and respetively, t he
circles have radii
radius, and the fixed and rolling
the parametric equations
epicycloid is described for 0 E [0, 2r| by
3 cos 6 +cOs 30
3 sin6+ sin 36
y(0) =
r(0)
the conupletet eaustic
parabola y = 2p.r,.
is

the
he of reflection by
case

the cubic curve defined by

16 72 108y+81r =0.
 398 18 Tschirnhausen's Cubic

Fig. 18.3. Completions of the envelopes in Fig. 18.2 defining caustics for refloc
by a circle and a parabola tion
left: an epicycloid, and riglht: Tschirnhausen'scuhi

which has the parameterization

r()= ) = 3(-1)
4 (18.5)
This curve can be found (in slightly different variants) listed in most catalos
of special plane curves [300, 308] as Tschirnhausen's cubici -although it is
also known as l'Hôpital's cubic and the trisectrir of Catalan. The Marquis de
THôpital (1661-1704) discussed it in his book Analyse des infinime
pour l'intelligence des lignes courbes of 1696, commonly regarded as the first
text on the calculus (see Chap. 8), and the Belgian mathematician Eugène
Catalan (1814-1894) showed that it can be used to trisect angles.
To understand the latter use, we need to introduce pedal curves. Given a
smooth curve C and a fixed point o, the pedal curve C of C with respect to o
is the locus of points where perpendiculars drawn from o to the tangent lines
of C meet those tangents. Conversely, given C' and p, one may ask- which
curve C has C' as its pedal curve with respect to o? We call C the negative
pedal of C" with respect to o, and we can construct C as follows. Through
each point p e C', we draw a line orthogonal to the line from o to p: these
lines are the tangents to C, and we can recover C as their envelope.
Tschirnhausen's cubic can also be interpreted as the negative pedal of a
parabola with respect to its focusi.e., if f is the focus and we construet lines
through each point p of the parabola that are orthogonal to the lines between
f and p, the envelope of this family of lines is
precisely the Tschirnhaus cuDIc
Figure 18.4 illustrates this for the parabola y2 = r, with focus f = (i.0)
The trisectrix property of Tschirnhausen's cubie arises from its negative
pedal relation to the parabola (with respect to the focus), as illustrateu
Fig. 18.5. To verify this property, it is convenient to use the polar form

Gomes Teixeira's Traité des Courbes Spéciuales ehes

Remarquables Planes at Gui
cites the Strasbourg 1900 thesis of M. Archibald as establishing this ame 214
 18.2 Tschirnhaus nnd Caust.ice Curves 399

Tschirnhausen's cubic interpreted the negative pedal of the parabola

as
18.4.
Fig focus to points o n
w i t h respect to its focus. Left: radial lines drawn from the
of lines through each point o n the parabola, orthogonal
the parabola. Right:- T sfamily
a

c h i r n h a u s e n ' s cubic is the envelope of this farnily of

lines.
lines
tothe radial

r(0) = secs(}0)

relative to the focus (~,0) as pole and the negative

ofthe Tschirnhaus cubic,
c a n be recovered
the direction 6 =
0. The parameterization (18.5)
r-axis as

from this polar form by using the trigonometric identities

a6 = 3 sin z0 - 4 sin /0, cos 6 = 4 cos*0 - 3 cos 6

tan(0) -V3t.
r(0)sin 0 and setting =

in r =-r(0) cos 6, y =

Tschirnhaus cubic a s a trisectrix. Let O

18.5. Catalan's interpretation of the generic point
Also, let P be a on
e origin and F be the focus of the parabola. tangent to the Tschiruhaus
to FP being
parabola, the line through P orthogonal
cubic the point Q. Then 2PFQ
at ZOFQ defines the angle trisection property.
=
400 18 Tschirnhausen's Cubic

18.3 Unique Pythagorean-hodograph Cubic

Straight lines are, trivially, PH curves. The
simplest non-trivial PH
the cubics, with Bézier control c.
points given in terms of the four nar are
uo,U1 and vo, v1 by expressions (17.5).
The Bézier representa
tation of PH eters
is
especially attractive because it admits intuitive constraints on the. cubi
polygon geometry that serve to identify PH cubics within the set of all
see Fig. 18.6). This structure reflects the fact that, as alluded
PH cubics
to in 81lcs
are actually segments of a unique curve, Tschirnhausen's ch.
P2
L
P1
Le

Lo
P3

Po

Fig. 18.6. The geometrical parameters defining a cubic Bézier control

by Theorem 18.1, this cubic is a PH curve if and only if Li polygon
vVL2Lo and =
62 =
01.

Theorem 18.1 For a plane cubic r(t) vith control points

Po, P1, P2, P3 let
Lo |Apol lPi Pol. L1 |Api| lP2 Pil, L2 |Ap2| P3 P2| be
= = -

= = -

= = -

the lengths of the control polygon

legs, and 61, G2 be the control polygon angles
at the interior vertices Pi, P2 as in Fig. 18.6. Then the conditions

Li = VLa Lo and 62 b 1 (18.6)

are sufficient and necessary for the hodograph r'(t) to be Pythagorean.
Proof: Consider a planar PH cubic with control
points of the form (17.5).
The lengths of the control-polygon legs are
given by

Lo =
1ApPol =(u +),
3

L =

|4pl=V(u+)(u? +u?),
La |Apal
= =

(ui +vi),
and they clearly satisfy the condition Li LzLo. Again using expresSIo =

(17.5), the sines and cosines of the angles 61, 02 are defined by
 18.3 Unique Pytlhagorean hodograph Cubic 101

(Api X
Apo) z
sin= LLo Vu ) ( + v)
(Apz x Ap1)
sin 2
Lal V( )u+v)

c o s 61 = Api ApPo
LLo V(u +(uf ") +

Ap2 APi
cos 62 L2Li

since their
sines and cOSInes are identical, we clearly have 02 =
01
d

Conversely, suppose r(t) is a plane cubic whose control polygon satisfies

that
in
Lo and G2 61 (=6, say). We can adopt a coordinate system
=

= La have the form

which the control-polygon legs

Lo(1,0), AP1 =
VL2Lo(-cos6, sin 0), Ap2 =
L2(cos 20,-sin 20)
Apo

and the quartic polynomial

2
4

=0 k=0

then has the Bernstein coefficients

Co=9L, c=-9LoV L2 cos ,

C2 3L2Lo(2 + cos 20),
C3 = -9L2vLo cos , C4=9L5,
and coincides precisely with the square of the quadratic polynomial

P2

P1

po
Fig.
1 8 , 7 . The conditions (18.6) for a cubic Bézier curve to possess a P'ythagorean
Odograph are equivalent tosimilarity ofthe two triangles ApopiPa and Apip:Pa
402 18 Tschirnhausen's Cubic

a(1)= 3|LobG(t)- VL2Locos0 bf(0) + L2bz)].

Hence, the hodograph of the cubic r(t) is Pythagorean.

One can casily verify that the Bézier control polygon constraints (10
are equivalent [355] to the requirement that the triangles with the co
poimts Po, Pi, P2 and p1, P2, P3 as vertices are similar see
18.7. Fig.
an alternative characterization of the PH cubics, see [340. For
Lemma 18.1 PH cubics defined by the control points (17.5) erhibit a crUnod
(self-intersection) corresponding to the distinct parameter values de

+v6) (ugu +vov1) EV3(ugUi u1 vo)

-

(u6
-

t =
(uo U1) + (vo =- U 1 )

Proof A self-intersection of the curve r(t)= (r{t),y{t)) corresponds to a

pair of parameter values t, t + T such that r { t + T) = r{t) and y(t +T) = y{

with T # 0. In other words, we require the two equations

P(t, T) =F =}"(t) r2 + " ( )7 + «') = 0,

gt, T) = U F T v ( t ) = }/"(t) T2 + t ) r + s'() = 0,
T

to have a common root 7 for some value of t (the division by 7 eliminates the
solution r = 0). Now the condition for the two polynomials p(t, T) and q(t, 7)
to have a common root r is that their resultant with respect to T,

r(t) = Resultant, (p(t, T), q(1, T) ,

must vanish. The resultant can be expressed as the Sylvester determinant

a"t)"(t) rt) 0
0 "()r"(t) r(t)|
r(t)=
") u'(t) (t) 0
0
) "0) V'|
Expanding the determinant reveals that, by a cancellation of leading terin
o), where
r(t) is quadratic in t. We can write it as r(t) = k (czt2 + 2c1t +

k =
(u0V1-u1vo)
c2= [(u1 - uo) + (v1 - vu)*1,
c1= |(u - uo) + (v1 - vo)] [ugu1 + voU - u - vg)

co ugu1 t vov1-u6-61- 3(uov1 u1vo). -

The discriminant of this quadratic is

 18.3
Unique Pythagorean
hodograph Cubie 403
A Ci CoC297t(ui- 4y) 27 +( -

to (ut u -

bysupposition uo®i
liis is positive and r(t)-l1 o #0 and (u1- u
a d

+
$I7.2) has two
distinct, real roots.(v1-vo #0 (sece
Definition 18.1 The standard form of
ade
(real
a crunode (real self
intersection) plane cubic r(0)= (r(t). u{1)
a

parameter
corresponds to choice of coordinates witn
erization in which its conmponents and
a

are of the
form
r(t)= p(t -1). u(0) =
q(t-o)- 1).
(18.7)
The property r(tl) yEl) 0 of the standard form
=
=

de is situated at the
crunode origin, and corresponds to the two implies that the
+1. This fixes the position and parameter values
parameterization
fy the orientation by requiring r(t) to be of the curve, and we then
quadratic rather than cubic: this
means that the curve tangent is asymptotically
plane cubic in standard form has vertical as |t| *. A crunodal
three remaining
freedoms- the r andy
cCales p and q, and the parameter value a of the z-axis
intercept.
Theorem 18.2 In standard form, the
Pythagorean-hodograph cubics
instances of the Tschirnhaus cubic defined are all
by
z(t) =
p(t -1), u(t) =(-1). (18.8)
Proof From the hodograph '(t)
standard-form crunodal cubic (18.7),
=
2pt, y' t) =
q(3t2 -

2at - 1) of the
we obtain
(t) +/(t) =
#[9- 12at+(4 +4a2-6)12+ 4at+11
where f =
p/4. If this is to be the square of a
by comparing like terms we must have
quadratic o(t) q(at2 +bt +c). =

a=9, ab =-6a, 2ac+b 4f +4a-6, be =20, = 1.

From the first three conditions, we deduce that a =
+3, b =
T2a, and c =

j-1). Substituting for b, c into the fourth then gives -a(f-1) = a.

and since
f#0 by assumption, this can be satisfied only if a 0. Finally. =

Substituting for c in the last condition yields f -2)=0, which implies

that f 3, since f 40. Thus q =ty3p (the sigu choice being immaterial
=

SInce it
changes only the sense in which the parameter t inereases along the
Vand for a PH cubic expressions (18.7) reduce to (18.8). a
Figure 18.8 shows Tschirnhausen's cubic in standard form, together with
e
Bézier control polvgons for several finite segments, which all
satisty the
c a l constraints (18.6). As a consequence of these conditions we note
, Unlike "ordinary" cubics, the PH cubies cannot intlect. By the variation
dim
shing property of the Bézier form (see §13.2), a convex control polygon
 404 18 Tsehirnhausen's Cubic

Fig. 18.8. Left: Tschirnhausen's cubic (18.8), expressed in standard form. Rick.
The Bézier control polygons for various finite segments of Tschirnhausen's cubie
the control polygon for any segment satisfies the two geometrical constraints (18.6

implies a convex curve- i.e., the curvature cannot change sign. Evidently, the
freedoms available for designing PH cubic segments amount to just sealing
rotating, and translating different portions of the curve defined by (18.8) for
-<i<+oo. The elegant characterization (18.6) of PH cubics is mitigated
by their inadequate shape freecdoms for most practical design applications.
Pottmann [369) has noted that many of the properties of Tschirnhausen's
cubic (including its interpretation as the caustic for reflection of parallel rays
by a parabola) werediscussed by W. Wunderlich [474] in a study of curves of
constant slope in R$. If we "lift" Tschirnhausen's cubic to R3 by taking its
(polynomial) arc length function s(t) as the z component, the tangent to the
curve r(t) = (r(0), »(4), s(1)) makes the fixed angle T with the (r,y) plane.

18.4 You Mean we Pay you to do That!?

While philosophers may argue over the ezistence of scientific facts or theories
outside the minds of their discoverers and contemplators, it is
incontrovertlD
that such facts and theories are ereated and interpreted by humans operating
under specific social, cultural, and economic circumstances see, for example.
the discussion of the "meaning" of Plimpton 322 in §2.1.
Nevertheless, the accepted norm in scientific writing is to describe proo
theories, discoveries, and inventions in the form of isolated and "imperso
narratives that make little or no reference to the notivations, and

influence S1s
circumstances that may have directly or indirectly contributed to the ge
of an idea or its subsequent interpretation by others. However, one ot the nost

profitable and enjoyable ways of approaching an unfamiliar topic is throu

 18.4 You Mean Thal!? 05
we Pay you to do

. historical development," since the neophyte often faces concepti1al

his
its
study
of md motivational doubts that closely mirror the initial stumbling
difficulties
steps, obstacles ove
acles overcome, and social/cultural impetus through which the ileza,
under consideration first came about.
invention
1e
or
proor,
lotwithstanding their high-minded tone, the intent of these philosophical
heory,
withstanding th to excuse a
is actually brief indulge in a favorite anecdote
alocutions
c i r c u

rning t h e ident
n

lentification of Tschiruhausen's cubic as the unique planar PH

idea of the human dimension of scientific research.
cOnceot conveys some
cubic,
reader prefers who still the "isolated and
impersonal" style of
presentation is at liberty to proceed directly to the next chapter.
O f c o u r s e ,

The author's iinvolvement

with PH curves originated during employment
for the
Research
in t h e Research
Division of a large multi-national company -which,
the
anonymity,
hall refer to as the HAL Corporation. At that time,
w e sha
of world-renowned organization,
Division was a prestigious and
AT. Research advances in science and technology. But
it
many fundamental
with to
of much of its research activity
redited

direct relevance
difficult to discern the in the Manufacturing
commercial interests. The author resided
the Company's as to whether
it
which labored under persistent doubts
Research Department,
with the study of
with "manufacturing" research than
was more
preoccupied solace in the fact
But Manufacturing Research took
manufacturing processes. those of its sister departments:
seemed no less relevant than
that its activities Semiconductor
Sciences, Computer Science,
Mathematical Sciences, Physical enjoyed virtually free
etc. The members of all these departments
Technology, research interests.
of their individual
pursuit
rein in the w a s beset
c o m e to a n end.
After s o m e years, the Company
All good things c a m e under great
and the Research Division
with dire financial circumstances, to
existence. But it is not s o easy
to its relevance and justify its
prove
As part of the endeavor to revamp
pressure
alter a deeply-ingrained culture overnight.
Director w a s appointed. The
new

Research, a n e w Department
Manufacturing had little
environment and, lacking a PhD,
Director hailed from a "shop floor"
And it w a s my fate to
experience with (or tolerance for)
high-falutin' research.
Director.
seminars before the n e w
present one of the first weekly departmental curves
Pythagorean-hodograph
subject of this seminar was, of course,
The including the proof that
Tschirnhausen's
t
contained some theoretical results, discussion
connections, and some
s o m e historical
cuDic the unique PH cubic,
is
Although I w a s
confident of the seminar's
Oapplications to CNC machining.
received at a number
of conferences) I w a s approached
uccess (it had been well
he *really admired iny courage."
who quipped that
Wards by a colleague, surprise that I had not notice
enquired what this meant, he expressed
enI at m e througiout
the seminar.
the in which the Director had glared out of his seat,
at any moment, jump
he were about to lose control
Oh
dexclaim-You mean we pay you to do tinat:
Ancient
Mathematical Thought Jrom
Morris Kline's
mathematical sciences, for this purpose.
excellent r e s o u r c e
tTT
imes 281 is a n
T
 406 18 Tschirnhausen's Culbic

Although this was a rather troubling developnent, one learns

things in stride. Just a few weeks later, however, we were greeted one take such
ne morning
by stunning tragic news: the new Director ed to cardiac
had succumbec
his oflice late the previous night - his soul had concluded its earts Test i
It was rumored that my seminar might have been the proximate caJOurn
hly
untimely demise, but I vigorously repudiated these libelous allegati ohis
case, a new Director of Manufacturing Research was appointed withiny
months. and the department resumed its usual course of business.
Then one day there came an announcement ot a most important "all-h.

meeting," at which the Vice President for Research would propound his
for the complete overhaul and restructuring of the Research Division, tovision
al
its activities with the HAL Corporation's business interests, and thus hel
o
extricate the Company from its pecuniary woes. On the day of the moment.
presentation, a great throng of the entire HAL Research
entous
Staff assembled wit
much anticipation and apprehension in the main lecture hall, to discover their
fate. True to the promise for sweeping reforms, the Vice President announced
that the current "traditional" departments-based on core disciplines such as
Mathematical Sciences, Physical Sciences, Computer Science, Semiconductor
Technology, etc. did not adequately serve the Company's business interests
He proposed, instead, a simplified, streamlined, and modernized departmental
structure, motivated by four "core business principles"- namely,
Department of Efficacy
Department of Responsivity
Department of Timeliness
Department of...
(I forget the fourth, but it was very much in the same vein as the others). By
way of elaboration, the Vice President explained that Efficacy was the science
of minimizing wasted energy or materials, optimizing all operations, reducing
manpower requirements, etc.; that Responsivity was the practice of sincerely
listening to one's customers, and acting on their feedback in the inmprovement
of existing products or the introduction of new product lines; that Timeliness
was the ability to reduce
product design and development cycles, in order
beat the Company's competitors to the marketplace, . . . etc., etc.
Needless to say, this New Vision of Rescarch spawned much
bewilderment
consternation. and incredulity among the assembled Research Staff, burdened
as they were by "traditional" training as mathematicians, computer scientists.
physicists, semiconductor technologists, etc. Nevertheless, it was greeted w
dutiful silence and concealed skepticism- in the prevailing
atmosphere o ind
HAL Corporation, there was an inverse relation between
speaking one s nvery
and one's future career prospects. For my own
part, I recall experiencing a
New Vision of Research nlike
unusual seusation as the
had experienced before. It was only later, on
unfolded, anything
1ow
introspection, that I realized n im
narrowly I had managed to
suppresS a powerful urge to jump up
to the Vice President of Research: "You mean we
and
pay you to do that
19
Complex Representation

The shortest path between two truths in the real domain passes through
the compler domnain.
Jacques Hadamard (1865-1963)

The recognition that geometrical interpretation, a s

complex nunmbers admit a

was a critical step in securing their widespread

noints in the Euclidean plane,
the
acceptance (see 4.1). The "geometric algebra" of points and divide by
in R2 defined

arithmetic rules for complex numbers allows us to multiply points,

the
as well as adding orsubtracting them in the usual vector sense. However,
systematic use of complex numbers as a means of exploring analytic geometry
deserves The Advanced Geometry
in R2 has received less attention than it
C. Zwikker [480] s e e m s to be the
of Plane Curves and Their Applications by
treatise that this approach.
consistently employs
only
The complex representation of R is especially valuable in the analysis of
characterization of the
planar PH curves, since it offers a simple and elegant
in terms of the complex
Pythagorean hodograph property. Any task performed
real variables.
representation could, in principle, be accomplished using only
and also
However, on account of the useful geometrical insights provides-
it
the systems of non-
Significant savings in the cost of formulating and solving
linear equations associated with interpolation problems (Chaps. 25 and 27) -
of planar PH c u r v e s
shall rely extensively on the complex representation
we
henceforth. An analogous representation for spatial PH curves (in term1s of
Tne quaternions rather than complex numbers) is developed in Chap. 22.
n this chapter, we use the complex formulation to establish a one to one

between regular PH curves and "ordinary'" polynomial c u r v e s

Correspondence
, to characterize the control polygons of planar PH curves, and to verify

rotation ivariance and study their intrinsic shape properties.

CA

Dover reprint. Another useful

grettably, this book has not survived even as a

urce, though not directly concerned with the complex representation of plane
arves, is Schwerdtfeger's Geometry of Conpler Nummbers |397|
408 19 Complex Representation

19.1 Complex Curves and Hodographs

Associating the point (z, y) of

the plane with the co.
Euclidean
+iy, we regard these two quantities as interchangeable and s
character z to denote both. Expressions such as 1tZ2, Z1Z2, ngle bold
Z1/z2 follow
usual rules of complex arithmetic (see j4.2), while zj Z2 and
and z,
Z1 X
XZ2 the
scalar and vector products. Complex variables may also appear as
as
denote he
arguments
of transcendental functions sin(z), cos(z). exp(z), log(z). etc.
A plane parametric curve can be specihed by a complex-valued f

r(t) defined on a given (possibly infinite) domain t

E la, b] of a real tion
For example, r(t) =
a1t +ag and r(t)= a2ts + ajt +ag define a
Variable t.
while circle with nter straight line
and a parabola, respectively, a c
and radius
written as r(t) = c+reit for 0 <t<27. Perhaps less obvious is the facre
circles may also be described by the fractional linear (Möbius) form that
at+b
r(t) ct+d
and that, for real a, the expressions

r(t) = cos(t + ia) and r(t) = cos(a + it)

define an ellipse and a hyperbola, respectively. See [480] for a wealth of furthor
ther
details on complex representations of plane curves and their properties.
For our purposes, the most interest ing aspect of such representations is the
ability to create new curves from any given curve by "distorting" the complex
plane in which it resides using a conformal map defined by an analytic function
of z (see §4.5 for the definition and basic properties of such maps). Our concern
here is more subtle than the direct deformation of planar curves by means of
conformal maps. Instead, we introduce a mapping of the hodograph plane-
the plane in which the derivative r'(t) of a
parametric curve r(t) resides. By
this scheme, we establish a one-to-one correspondence between the sets of
regular PH curves and of regular "ordinary" polynomial curves, which offers
a framework for
comparing and contrasting their properties.
Consider a polynomial curve in the complex plane specified in Bézier form
n

r()
P)(1-1k*, te[o,1].
k=0
(19.1)

where the complex values pk Ik

+i yk, k =0,... . n define the control poln
=

The usual control polygon

features-convex-hull confinement, the variati0
diminishing property, and the subdivision and
carry over directly to the complex
degree-elevation algort
representation. The hodograph w{t) r'[t)
of the curve (19.1) can be expressed as a complex Bézier curve of degree in-1.

Note that r(t) is uniquely determined by r'(t), modulo a translation

to the integration constant. corre
 19.2 One to 409
one
Corresponlence

w(t) =
w:
k=0
(1 1 1 , te
[0,1]. (19.2)

w i t h
trol points given by
c o n t r

Wk=1Apk =

7 (Pk+1 -

Pk), k=0,.. . , n
The forward
differences Apk = Pk41 - Pk define the lirecteel "legs" of the
or clarity, we regard curves and their hodogr
control polygon. For graphs as residing
O Separate complex planes, z = t + i y and w = u + i , respectively.

19.2 One-to one Correspondence

Ising C as a representation for R-, let II be the set of all regular polynomial
eurves, and
ll be the set of all regular PH curves. By a regular curve, we mean
parameterization of such a curve (see Remark 19.2). To a s s e s s how
a Specific
PH curves are, it is useful to characterize the relationship between
flexible"
Il and I are both infinite s e t s - they include c u r v e s
these two sets. Although is
II c I, since any regular PH curve
of arbitrary degree-it is clear that but there are regular polynomial c u r v e s
certainly a regular polynomial curve,
the parabola r(t) t+it). We
whose hodographs a r e not Pythagorean (e.g.,
=

P that transforms any differentiable

introduce a simple three-stage procedure
plane curve r(t) into a n e w curve f(t).
procedure P: r(t) > f(t)|
1. differentiate the given curve r(t)
to obtain its hodograph w (t) = r'(t);

2. apply theconformal map w - to w

the hodograph plane, giving w(t) = w (t);
3. integrate the transformed hodograph w(t)
to obtain the new curve f(t) = w(t) dt.

fixed by taking r(0)- r(0) = 0.

translational freedoms are
n this procedure,
co17espondence.
Proposition 19.1 P defines a bijective map, or one-to -0ne
c u r v e s and 1Cgular PH curves.
Detween the sets II and II of regular polynomial

regular polynouial curve with hodograplh

Proof: Let r(t) =
r(t) +iy(t) be a

and y. Since w(t) loes not pass through

wt)u(t)+iv(t), where u =*' v =

constant. Step 2 gives

gcd(u, v)
=

O r g i n , we must have

v"(0) + i2u()e(0)
w(1) =
w (1) = u"() -
is of Pythagorean form with u ancl
the transfored hodograph, which
modulo a translation, a unique
Clatively prime. Integrating w(t) then gives,
410 19 Complex Representation

regular PH curve f(t) corresponling to r(). Conversely, let s ( t .

e a regula
PH curve, Its hodograph must be of the form w(!) u2(t)-
where gcd(u, v) = constant. The inversC lo stejP 2 transforms w .
=

2(t)-u()+i2 u(tyu(t),
into
wt) = /w{1) = + | u(1) + i v(t) ],

a hodograph that does not traverse the origi1. Integrating w(t) gives
(19.3)
translation, a unique regular poly11o1mial curve r(t) corresponding d u l ,
the sense of the paranetrie flow along r(1) is arbilrary, corresponding to
taking the comnplex square root.
Sign ambiguity on

Note that, in general, the iuverse map W*vw n (19.3) does rnot vicl
a
polynonmial hodograph when applied to a general polynomial hodourarn
fact, it will give a polynomial hodograph only when applied to a Putha
hodograph. Thus, regarding P and its inverse P a s mappings between sek
of polynomial curves, we have P(I1) = lI and P-i(17) = II.
We call a regular polynomial curve r(t) and a regular PH curve f(t) a pair
of corresponding curves if they are mapped into each other under the action
of P and P-. Such pairs can be expressed explicitly as

r(t) = u(7) dr + i v(7) dr,

JO

f(t) = uP(r) - f(T)dr + i2u(r)u(r) dr, (19.4)

Jo
where u(t) and v(t) are relatively prime polynomials, and we assune r(0) =
r(0) = 0 to fix translational freedoms.
Remark 19.1 The set II of regular PH curves has the same "cardinality" or
"power" as the set II of regular polynomial curves.
The cardinality or power of an infinite set is a measure of its "size" |230
specifically, two infinite sets have the same cardinality if we can establish a
one-to-one correspondence between their members. Thus, in accordance with

Proposition 19.1, may say that there are "just as

we many" regular PH curves
as there are regular polynomial curves in general.

Remark 19.2 The correspondence between the curves (19.4) refers not oy
to their geometrical loci, but also to the variation of the parameter t along
those loci. Any polynomial curve r(t) may be re-parameterized by a hne
transformation t - a + bt (b 4 0) without altering its geometrical locs,1s
regularity, or its degree. However, if r(t) and f(t) are corresponding curves
re parameterizing the former does not simply yiekd a new parameterzat
of the latter as the corresponding PH curve. Fixing translational freecio
not
by always placing the point of zero parameter value at the origin, it
difficult to verify that the sequence of PH curves corresponding to all
f a
re- parameterizatios of a given curve r(t) are actually uniform scaltngs
unique curve r{1) by the factor b (see Fig. 19.1).
 T9.2 (One to one Correspidenee

1 9 . 1 . Different paraneterizations ofa parabola (leftL) yield different scalings of

a

hiruhaus cubic (right) under the one to one correspondence of Proposition 19.1.

Now all linear re-parameterizations of a given PH curve are clearly also

PH curves. Thus, if I7, is a subset of the space i of regular PH curves whose
aembers are simply the various (fixed-degree) parameterizations of a unique
locus,
it follows from Remark 19.2 that the pre image
II, P-|(Il.) of .
=

1under the map P consists of polynomial curves with mutually distinct loci

Example 19.1 The PH curve corresponding to the parabola r (t) =t+itd

isthe Tschirnhaus cubic f i ( t = t - + i2t2. Consider now the sequence
of parabolae re(t) = kt +i for 0 < k < o , any two members k1 k2 of
which are clearly of different shape. The PH curve corresponding to member
k is easily verified to be f(t) = k*t - Gk543 + i 2k442. By setting t = T/k*,
however, we observe that f(t) = fi(T) for all k. Thus, P maps the family of
of
geometrically distinct parabolae r:(t) into the different
a geometrically unique cubic, fi(t)- see Fig. 19.2.
parameterizations

Fig. 19.2. Parabolas of different shape (left) yield ditlerent paraeterizations of a

Tschirnhaus cubic (right) under the one t o one correspondence of P'roposition 19.1

Lemma 19.1 7The deqrces n and n of corresponding cures r(t) tni r{t) ar
related by n = 2n - 1.

roof: Let u(t) and v(t) be two relatively prime polyuonmials, and lot n =
nax(deg(u), deg(v). Integrating the hodograph w(t) = u(t) + i v(t) then gives
 412 19 Complex Representation

+1. Further, one c a n

polynomial r(t) of degree n m =

(t) == u easily
a curve . .

that max(deg(u?-),deg(2ur)) = 2m, so integrating w(t) t

i2u(t)v(t) gives a PH curve f(1) of degree n = 2 n + l = 2n - 1 .
u"(t) 1(1
-

Clearly, there are no rgular PH curves of even degree. Straight line.

correspond to (different) straight lines in II, but P naps es in Ii

curves of degree > 2 to regular PH curves of higher odd degree (sece T a l . a

regular polyn
.1

Table 19.1. Corresponding curves of low degree.

polynomial curve r(t) PH curve f(t)

n = 1 straight lines straight lines
n=2 parabolae Tschirnhaus cubics
n =3 regular cubics regular PH quintics

The properties of the map w - W = w are described in most textbors

on complex analysis see, for example, |42, pp. 79-84). As is well known [196
it takes the Gaussian integers
or "grid points" a +iß (where a, 3 e } o
the
complex plane into points a2B +i2a/3 whose real and imaginary parts
are members of integer Pythagorean triples. The map is conformal everywhere
except w = 0; this exceptional point need not concern us if we consider onlv
regular curves. Note also that it gives a double covering of the complex plane.
each point w having two pre-images, -w and +w. Writing w =u+ivand
w =ú+i6, the pre-images of the grid lines û = constant and = constant
of the w plane are, respectively, the families of rectangular hyperbolae

and 2uv= ú
in the w plane, with asymptotes u = tv and u = 0, v = 0. On the other hand.
the lines
grid u constant and v = constant of the w plane are
=

mapped into the families of confocal parabolae

respectively
2+4u2ú =4u and -4u =
=444
in the w plane, whose vertices lie along the
positive and negative û axis.
the limits u - 0 and v » 0, we obtain the half lines ú <0 and ú> 0, doubiy
traced, as images of the imaginary and real axes (see Fig. 19.3).
In teris of the velocity vector w(t) = r'(t) = a(t) eivt) at each point of a
given curve r{t), describing motion with speed o(t) at an angle {t) rela tive
to the real axis, the map w w transforms this velocity (see Fig. 19.9) *
squaring the speed and doubling the inclination angle,
ot)= a"(0) and
p(t)=2(t).
 19.2 One to-one Correspondence A13

W W

9.3. The
Fig. 19.3. T1 map w w= w of the hodograph plane suares the lengt.hs of
locity vectors and doubles their inclination angles. Left: grid lines in thee w plane
abped to families of confocal parabolac in the w plane. Right: the pre images
a i d lines in the w plane are families of rectangular hyperholae in the w plane

PropOsition 19.2 The control points of a rgular PH eurve of degree 2n- 1

e qiven in terms of n compler values wo,.,W- 1 by the recursive fomula

min(k,n-1)
Pk T w,Wk- (19.5)
Pk+1
2n-1 j=max(0,k-n+1) (2n-2
k

such that
for k 0, 1, . , 2n-2, where po is arbitrary and wo, ., Wn-1
= ..

the hodograph w(t) defined by (19.2) does not traverse the origin.
.. are

Proof: The square of the complex hodograph (19.2) may be written as

2n-2

w(0) w ( a -1)3n-2-kg* (19.6)

k=0

where, making use of the product rule (11.20) for Bernstein-form polynomials
the coefficients wk are given in terms of Wo,..., Wn-1 by

min(k,n-1)
(n- 1)
W 2n
WWk-j k =
0,... .2n -

2.

j=max(0,k-n+1)
k

To integrate the hodograph (19.6), we employ property (1 I.7) of the Bernstein

basis functions. This yields the quoted form (19.5) for the cont rol points of a
PH curve of
degree 2n 1, with an arbitrary coustiaut ol integration Po.
The requirement that the hodograph (19.2) defined by w o . W-1 u s t
not pass through thhe origin for the control points (19.5) to letine a rrgular
i curve does not, in general, have an intuitive geoetrical interpretatiou in
Crms of the locations of Wo,.. . , w - 1 . This requirement uounts to the grd
t h e real and imaginary parts of (19.2), u(t) and v{t), being a constant
O t h e i r resultant with respect to t being non ero. Criteria specitic to the
cubics and quintics are given in equations (19.10) aned (19.13) below.
414 19 Complex Representation

of Hodographs
19.3 Rotation Invariace
offers a e rotatio
simple proof for the rotaton variance
The complex representation
form
of the sufficient-and-necessary
'(t) = u(1) - v(t), /(t) = 2u(t)v(t), a(1) = u"(t) + u(
t)
for primitive planar Pythagorean hodographs r ( t ) = (r"(1), y ' ( t ) s t : . e

a"(t) + y(0) = a"(0)

isfying
where gcd(u, v) = constant gcd(r',y) = constant. On makine a res
tation
[T'(t)| cos -sin 01'(t)1
sin6 cos '(1)J
through angle 6, we seek to express the rotated hodograph r(t) = ( () #
in terms of two new polynomials u(t), v{t) as

(0) = #0) - *(). ) = 2k(0)o(t).

(19.7)
One can easily see that the transformed hodograph,
( t ) = cos 0 [ u(t) - ( 1 ) ] - sin0 2u(t) u(t),

it) = sin6 [u"(t) -v()] + cos 0 2u(t)u(t),

is obtained by substituting into (19.7) the polynomials

ült) = cos 0 u(t) - sin 0 v(t). (t) = sin 0 u(t) + cos 0 v(t). (19.8)

Using the complex representation r'(t) = w*(t) where w(t) = u(t) +iv(l), the
rotation yields f'(t) = exp(i0) w*(t) = Y(t) = w2(t), the real and imnaginary
parts of w(t) = exp(i50) w{t) = k(t) + iõ(t) being defined by (19.8).

19.4 Pythagorean-hodograph Cubics Revisited

In the first non-trivial instance, that of the PH cubics, expression (19.5) yelas
control points of the form

1
P1PotWo P2 P+WoW1, P3 =
P2 + Wi
(19.9)
3

Here, the complex values Wo and Wi must satisfy

(19.10)
Im(Wow) #0
to obtain a regular curve. If wo uo +ivo and wi
a m o u n t s

=
=
U1+iv1, this a rii
ugU1 # u1 Voi.e., the line from wo
to w should not pass through t
 19.5 Characterization of the PH 115
Quintics
complex forn offers a succinct expression of t.he suflicient and necessary
C o n d i
ns (18.6) characteriziig PH culbics by the control polygon geometry.
Writing aations (19.9) in ters of the control polygon legs as
2
3Apo Wo 3Api WoW 1
3Ap2w
2ay eliminate wo and wi to obtain thc single conmplex constraint

(Ap) App4po (19.11)

ho8e satisfaction distinguishes the PH cubies fromn "ordinary" cubics. To see
that the complex constraint (19.11) is equivalent to the two real constraints
/18.6), we write the directed control polygon legs in polar form as
Apo = lLo e ' o ,
Api = Liek, Ap2= Lze kz |

Substituting into (19.11), we obtain Li = LaLo and 261 = Eo + E2. In terms of

the control polygon interior angles, 61 =T+E1 -Eo and 02 = T +£2 -6i, the
latter condition implies that G2 = 61. Clearly, this derivation is much simpler
than that given in §18.3 using only real variables.
We may regard general cubics as possessing eight degrees of freedom (the
coordinates of their control points), although not all of them are intrinsic shape
freedoms. Each cubic can be regarded as a point in a Euclidean space R" with
coordinates (zo, yo, T1, Y1, T2, 2, T3, V3), and the real and imaginary parts of
(19.11) each define a degree 2 hypersurface (a hyperquadric) in this space. PH
cubics lie on the intersection of the hyperquadrics, a locus of dimension 6.

19.5 Characterization of the PH Quintics

The PH quintics are of great practical interest, since they are the simplest PH
curves that may inflect, and match first-order Hermite data (see Chap. 25).
For regular curves, there is a unique correspondence between PH quintics and
ordinary cubics, so we may expect similar "shape flexibility" anmong these two
Sets of curves. The control points (19.5) for PH quintics are of the tor

Pi Po+ Wo

P2= Pit WoWI

5
1 2w + woW2
P3 P2 t 3

P P3 t WI W2,

P5 P t W2 (19.12)
he possilbility of such a derivation was first noted by Professor Wendelin Degen.
416 19 Complex Representation

illustrates the economy of expreseie sion .

Comparing (17.6) with (19.12)
by the complex representation.
"The c o i s i t r a i n t that ensures regul
be written in the form
regularityallordedthe
PH c u r v e defined by (19.12) may

Im(Ww,)| - 4 Im(Wwi) Im(W W2) # 0,

of
(19.13)
or explicitly in t e r i s of the real
and imaginary par ts Wo, W1, W,

7 4 ( u 0 - u1 Vo) (u1 V2- 121).

(U2 u2o) (19.14)
Although (19.13) does not have an intuitive geometrical meaning, we
an immediate consequence that Wo and w2 must be non zero, i.e., D
is zero at t 0 1).
(otherwise the parametric speed
=
or
and pi ps
form to derive a characterization for PH us
We now use the complex ntics
in terms of the control polygon geometry. we begin by re writing equatior
(19.12) in terms of t h e directed control polygon legS A p k = Pk+1- p: as

w WoW
Apo Api 5
5
2w+wWoW2
Ap2 15
WIW2 W2
Ap3 5
Api = (19.15)
Apo # 0 and Ap4 # 0.
It is understood that, for a regular curve,

be defined
Proposition 19.3 Let the conlrol polygon legs of regular quintic
a

Ap. Then the c u r v e has a Pythugoran hodograph

by compler values Apo,...,
ifand only if these values salisfy

Apo(Aps) = Api(Ap1), (19.16)

consistent wilh the follouwing system of constraints:

and are

0,
3Apo4pi 4p2- (4po)-Ap; -2(4p1)"
=

0,
Ap 4psApz (AP)*Api- 2 (Ap:)"
=
3 -

0,
3 ApoAp,Ap2 Api4po4pi - 2(4pi)*Ap:
=
-

0,
3 ApAp, Ap2 Apo4p,4ps-2 (Ap:)"Api
=
-

0,
9 Apo(AP2)" -6 (Ap:)"Ap2-24poAp Aps- (4p»)"4pa
=

9 Ap(Ap)-6(Ap"Ap 2Ap1ApiAp (pi)"Apo= 0. rder.

are in
Before proceeding with the proof, some o explanatory remarks one
Ordinarily, when Apj and Aps are both non zero, cquation (19.16) ant ons
of the first four of cqualios (19.17) define sufficient and necessary conu
r, (19.16) and the
for a quintic to be a PH curve. If Ap=Aps 0 , however, uplilied
first fouir equations in (19.17) become identities, and we nst take (a sip
form of) cither of the last two of equations (19.17) a h a cOnu
 19.5 Characterization of the PH Qui1t.ics 417

Proposition 19.3: The necessity of nll of the conditions in (19.1)

P r o o fo f P
follows from eliminating Wo, wi, W2 from equat.ions (19. 15). For
and (19. 17)
ample, . we assune Apo,. ..Aps are of the form (19.15), we see from the
first and
last of these cquations that w2 w Apy: Api, while from the
secod a n
n d f o u r t h w e have wo W2Ap Aps. F'rom thes: expressions it
condition (19.16) inust. be satislie«. Likewis, thee neressity
v i d e n t tlhat
nditions (19.17) is Verifiod by substituting directly from (19. 15)
the six conditio.

Concering
icieney of the stated conditions, wr nee to show that
the suflic

.16) and (19.17) are satisficd, then Ap0,

i fequations (19,1 , Ap must he of
the form (19.;15). Consider first equation (19.16). Without losSs of generality,
we
w write Apo a and Apj c where, as noted ahove, a anel c a r e
may write
= =

x numbers for a regular curve. Equation (19.16) then bcones

on 2erw c o m p l e x

a (Apa)* = c* (Ap1)",

and AP3 a r e to satisty the above, they must be of the form

nd if Ap1
Api = ab and Ap3 = bc

for some complex number b (not necessarily non-zero).

a s defining
With these results, we may regard the six equations (19.17)
o n substituting into (19.17) we obtain
the value of Ap2. Specifically,
a b |3 Ap2 - ac - 2b] = 0,

be 3 Ap2 -

ac -

2b] = 0.
a b c | 3 Ap2 - ac - 2b = 0,

abc 3 Ap2 a c - -

2b|= 0.

a 9(4pa)2 -

6b?Ap2 2 ab'c - -

a?c2] =
0.
2 [9 (Ap2)2 - 6b°Ap2 - 2 ab?c - atc°] = 0

and we must distinguish two cases, according to the value of b.

Case (i) b #0: The first four equations above are obviously equivalent and
yield the same solution
Ap2 = 2D+ac
3
for Ap2. The last two equations, on the other hand, give

2b a c
Ap2 Or (19.18)
3 3
but the second solution must be discardel, since if b # 0 it is inconsistent
with the first four equations. Thus, equations (19.16) and (19.17) inply that
4po , Api can be expressecd as

Apo = a, Api = ab, 2b +ac Aps = be

Ap2 3
 418 19 Conplex Rejpresentation

with (Wo, W1, W2) =

V5 (a, b.c
which is clearly of the form (19.15)
of equations (19.17) degeneratete to
to identities
Case (ii) b =
0: The first four
the last two hot1
concerning Apz, whereas
both
and
to 9
yield
(Ap2)2
no

-
information

ac2 =
0, giving solutions simplify
ac
Ap2=

to expressions (19.18)
with b = 0. Hence we have
that correspond

Apo =a, Ap =0, Ap2=tq Aps = 0, Ap4 =

c2

which is also of the form (19.15), with (Wo, W1, wa) = V5 (ta, 0, tc).

were obtained by a Gröbner basis reductin

Equations (19.16) and (19.17) ion
on the system (19.15), employing a lexical ordering with wo, W1, W2 precedinc
and (19.17) comprise all elements of th
Apo. , Ap4. Expressions (19.16)
basis that do not depend on Wo, W1, W2. We distinguish (19.16) from the other
conditions (19.17) since: (i) it has a simple geometrical interpretationse
See
Remark 19.3; and (i) among (19.16) and (19.17), it is the only condition that
is individually invariant under reverse-ordering of the control-polygon legs.
the latter point is that the substitution
The significance of
Apk -Ap4-k k = 0,... ,4 (19.19)
amounts to the re-parameterization t 1 -t, under which the Pythagorean
nature of the hodograph must be preserved. While (19.16) is clearly unaltered
the substitution (19.19), the individual members of the system (19.17)
are
by the equations are grouped into
not (the entire system is invariant, however
pairs whose members clearly map into each other under (19.19)).
Additional necessary control-polygon constraints can be derived, that are

symmetric with respect to the labelling of its legs. The quartic condition

9 (Ape)"ApoApa = [2 Ap1Aps + ApoApa], (19.20)

for example, follows straightforwardly from expressions (19.15). In conjuncton

P
with this does not amount to a sufficient condition
(19.16), however, for a
tneu
curve, since if we choose Apo a2, Api ab, Aps bc, Apa =
= = =
c
equation (19.20) gives two possibilities for the middle leg

Ap2 = +t 2b2+ac
3
of which only the positive sign choice yields a Pythagorean hodograph.
Regarding quintics as residing in a 12-dimensional space defined by their
19.14
control-point coordinates, the real and imaginary parts of equation
this
and of any one of the equations (19.17) define four cubic hypersurfaces p H
space, on whose intersectiona locus of apparent dimension eight-
ape freedoms).

quintics lie (1ot all these dimensions correspond to initinsic shape freeda