(Mamoru Hosaka (Auth.) ) Modeling of Curves and Su

Computer Graphics -
Systems and Applications
Managing Editor: J. L. Encarna~ao

Editors : K.B0 J. D.Foley R.A.Guedj
P. J. W. ten Hagen F. R.A. Hopgood M. Hosaka
M.Lucas A.G.Requicha
M. Hosaka
Modeling of
Curves and Surfaces
in CAD/CAM
With 90 Figures
Springer¥erlag
Berlin Heidelberg New York
London Paris Tokyo
Hong~ong Barcelona
Budapest
Mamoru Hosaka
Tokyo Denki University
Kanda-Nishikicho 2-2
Chiyoda-ku
Tokyo 101, Japan
ISBN-13:978-3-642-76600-8 e-ISBN-13:978-3-642-76598-8
DOl: 10.1007/978-3 -642- 76598-8
Library of Congress Cataloging-in-Publication Data
Hosaka, M. (Mamoru)
Modeling of curves and surfaces in CAD/CAMIM. Hosaka. p. cm. - (Computer graphics
-systems and applications) Includes bibliographical references and index.
ISBN-13:978-3-642-76600-8
1. Computer graphics. I. Title. II. Series: Symbolic computations. Computer graphics -
systems and applications. T385.H6958 1992
670.285'66 - dc20 91-43266 CIP
This work is subject to copyright. All rights are reserved, whether the whole or part of the
material is concerned, specifically the rights of translation, reprinting, reuse of illustrations,
recitation, broadcasting, reproduction on microfilms or in any other way, and storage in data
banks. Duplication of this publication or parts thereof is permitted only under the provisions
of the German Copyright Law of September 9,1965, in its current version, and permission for
use must always be obtained from Springer-Verlag. Violations are liable for prosecution under
the act of the German Copyright Law.
© Springer-Verlag Berlin Heidelberg 1992

Softcover reprint of the hardcover 1st edition 1992
'IYPesetting: Camera ready by author

45/3140-543210-Printed on acid-free paper
Preface
1 Aims and Features of This Book

The contents of t.his book were originally planned t.o be included in a book en-
titled Geometric lIIodeling and CAD/CAM to be written by M. Hosaka and F.
Kimura, but since the draft. of my part of the book was finished much earlier than
Kimura's, we decided to publish this part separately at first. In it, geometrically
oriented basic methods and tools used for analysis and synthesis of curves and
surfaces used in CAD/CAM, various expressions and manipulations of free-form
surface patches and their connection, interference as well as their qualit.y eval-
uation are treated. They are important elements and procedures of geometric
models. And construction and utilization of geometric models which include
free-form surfaces are explained in the application examples, in which the meth-
ods and the techniques described in this book were used. In the succeeding book
which Kimura is to write, advanced topics such as data structures of geometric
models, non-manifold models, geometric inference as well as tolerance problems
and product models, process planning and so on are to be included. Conse-
quently, the title of this book is changed to Modeling of Curves and Surfaces in
CAD/CAM.
Features of this book are the following. Though there are excellent text
books in the same field such as G. Farin's Curves and Surfaces for CAD /CAM[l]
and C.M. Hoffmann's Geometric and Solid Modeling[2]' this book differs from
them in t.he methods of analysis, interpretation and range of the problems to
be treated. Though most of the theories and the methods treated in this book
were developed from academic points, they were applied to real problems which
were performed for some Japanese companies during the period from mid-1960
to 1987. Since before the mid-1970s the literature in this field outside Japan was
hardly available for the author and the necessity of the development was urgent,
t.he theories and the methods were developed independently of those from outside
Japan. Since most of them appeared only in Japanese journals and proceedings
and the application results have been company assets, some of the contents of
this book may be new in the English language. They have been successfully
applied in the companies and the intuit.ive approach of the analysis has enabled
the practical engineers there to understand the essence of the methods and to
become interested in their application. Moreover the contents treat current and
future problems of importance, so the author believes the contents are worth
publishing.
Nowadays, there are many commercial software systems for CAD/CAM.
VI Preface
They are useful for ordinary products, but they do not seem "to be adequate
for application to products with special features attractive to customers, be-
cause there have still been companies which asked the author of this book for
advice as to the application of the system to their special product developments.
The author thinks the contents of the book include some hints on how to respond
to these questions.
2 Structure and Contents of the Book
This book is written for practical engineers who need knowledge about theories
of curves and surfaces and their processing methods for the design and man-
ufacture of engineering objects of high quality. The author has tried to make
the description as easily understandable and applicable as possible. Accordingly,
though the various methods are systematically formalized, they are also graph-
ically interpreted so that one can analyze problems just using pencil and paper
with the help of the drawings. This improves one's understanding of the problem
he is trying to solve.
The contents of the book are divided into the following six parts.
- Chapters 1-4. The basic mathematical tools and linear and quadratic curves
and surfaces.
- Chapters 5-8. The fundamental theory of curves and surfaces extracted from
the theory of differential geometry and its application. Also the curve passing
through a given point sequence.
- Chapters 9-12. Curves and surfaces with control points and their regular
connection.
- Cha.pters 13-15. Non-regular connection of surface patches and triangular
patches.
- Chapter 16 and Appendix. Interference of surfaces. Tracing methods of inter-
section curves by numerical calculation.
- Chapter 17. Geometric models and engineering drawings. Feature input,
construction of geometric models which include free-form surfaces, and their
utilization.
In Chap. 1, the theory of vectors and matrices together with their notations
which are used in this book are concisely summarized with application examples.
In Chap. 2, coordinate transformations and movement of a solid body are treated.
Since it is rather difficult for a person to grasp intuitively the three-dimensional
movement of a body, we formalize its description. And for its intuitive under-
standing, we introduce mirror reflections to represent rotations and translations.
In Chap. 3, objects which are represented by linea.r equations are treated. Since
this is related to the control polygons and nets, we introduce Menelaus' theorem,
which is a classic theorem of geometry. Its variants are used in various places
Preface VII
afterwards. A topology of faces, edges and vertices of a solid body is explained

simply.
In Chap.4, conics and quadrics are treated, because they are represented
by second degree equations and their properties are well investigated. Since the
shape of a free-form surface around, a point on it can be expressed approximately
by a quadric surface and also the shape of intersection curves of surfaces near
their singular point is conic, knowledge of quadrics and conics is important. And
many mechanical parts have shapes consisting of quadrics and planes.
Chapters 5 and 6 are excerpts from the differential geometry to be adapted
to analysis and interpretation of curves and surfaces described in this book.
Geometric quantities such as tangent and normal vectors, curvature and torsion
are introduced together with the formulas of Frenet-Serret. The shape of a
curve near a point on it is described approximately with these values. When an
elevation and a plan of a space curve to be designed are given independently, a
problem of determining an expression of the space curve is treated as a practical
example. In the last part, a method of converting a parametric expression of
a plane curve into an implicit form is described. Conics are represented by a
rational Bezier curve of degree two, which is treated in Chap. 12.
For a surface, its basic vectors, fundamental mi1gnitudes, curvatures and
lines of curvature are explained, which determine the shape around a point on
the surface. Weingarten's formula and Rodrigues' equation, which describe the
behavior of the normal vector, are introduced and used in Chap. 7, where various
curves on a surface are defined to express its characteristics. These curves are
classified according to their features: curves intrinsic to its shape, those depen-
dent on direction of interest, those related to its environment and the direction,
and those defined to search for special points on the surface. Umbilics, lines
of cllIvature, equi-brightness curves, silhouette patterns, highlight curves, ridge
valley curves, and extremum search curves and others are discussed. The funda-
mental magnitudes and curvatures of offset surfaces and ruled surfaces are also
determined , which are to be used afterwards.
In Chap. 8, curves which pass through a given point sequence are treated.
Lagrange's formula which uses a polynomial is explained together with a method
of calculating interpolated or extrapolated points. Its rational equivalent is also
introduced. Though these methods are not directly used in CAD/CAM op-
erations, they are included in an excellent differential equation solver used in
Chap. 16 and explained in the Appendix. Then curve segments expressed by
parametric polynomials are described. The point sequence can be connected
smoothly by the curve segments with the pseudo-elastic energy stored in all the
segments being kept minimum. In these connections of parametric curves, the
importance of scale ratios which relate to relative real magnitudes of the curve
segments is shown by examples of connected curve shapes. So far, fundamental
knowledge on curves and surfaces has been explained, then in Chaps. 9-15, that
which is used in CAD/CAM is treated.
VIII Preface
In Chap. 9, so-called Bezier curves and surface patches are defined from the
special characters of control points . Here the shifting operators for subscripts of
control point.s are introduced for simplicity of the expressions as well as of their
manipula.tion. Geometric properties of a Bezier curve are intuitively obt.ained
from a Bezier polygon which consists of t.he control points for the segment. Divi-
sion and degree elevation of the curve segment. correspond to those of the Bezier
polygon. A Bezier net for a surface patch is obtained by the tensor product of
t.wo Bezier polygons for curve segment.s. The net determines geometric propert.ies
of t.he patch.
In Chap. 10, first, the meaning of degree of freedom for the connect.ion of
Bezier polygons is explained; then a Connection Defining Polygon (CDP) which
det.ermines connection of the adjacent curves in c(n-l) is defined. When conti-
nuity of derivatives at the junction of two curves of degree n is gua.r anteed up
to t.he (n - I)-t.h derivative, the condition of connection is called c(n-l). Since
at each junction o(connecting Bezier polygons in c(n-l), there exists a CDP, we
explain that the adjacent CDP are related by another polygon called the spline
polygon. Inversely, from a given spline polygon we can decluce a sequence of
c(n-l) connected Bezier polygons. We explain practical met.hods of determining
t.hese Bezier polygons from an arbitrary spline polygon whose shape is a rough
picture of the shape of the generated connected curve. The spline polygon is
considered to cont.rol the shape of connected curves of degree n and the c(n-l)
condit.ion means t.hat t.he shape control is local. In a Bezier polygon , the control
is global.
Then we explain a method of division of a spline polygon which is equivalent
to increase of vert.ices of the spline polygon. Next, for a spline polygon of degree
three we show a met.hod to acld more freedom for shape control by changing the
connecting condit.ion from C(2) to C(2), which only guarant.ees continuit.y of the
curvat.ure vect.or at t.he junctions instead of t.he second derivative. In the last
part. of this chapt.er, effects of the scale ratios on the shape of the generated shape
of the curve and t.hat of the radius of curvature profile, whose shape is used as
an aesthet.ic crit.erion by designers , are explained.
In t.he first section of Chap. 11, by the met.hod of tensor product Bezier nets
are determined from a given spline net. In the succeeding sect.ions, proofs of t.he
methods of det.ermining Bezier polygons from a spline polygon are given. Here
Menelaus' theorem is extensively used .
In Chap. 12, rat.ional Bezier and rational spline curves are treated. First
comes t.he relation between linear division of a line segment and rational division
of the same segment with a different weight value attached to each of it.s end
point.s. The geometric properties of a Bezier polygon can easily be transferred to
the rational Bezier polygon and vice versa. A simple graphical mapping method
between th~ respective polygons is given. In the similar mapping from a rational
spline polygon we obtain rational Bezier polygons connected in c(n-l). We give
examples of effects of the weights on the shape of the generat.ed curve and surface.
Preface IX
Since a rational Bezier of degree two is converted to an implicit form which is

the same as a conic, we discuss the effect of the weight on the class of conics. In
the last sections we introduce the use of negative values of a weight and a curve
fitting method by conics. We show that t.he optimum number of segments for
fit.ting a given point sequence is obtained.
In Chaps. 13-15, shapes which cannot be expressed by spline surfaces are
treated. Such shapes are often needed in special regions of surface expressed by
regulaT connection of patches. In Chap. 13 it is shown that regions of convex
corner and convex-concave corners can be rounded by using only non-regular
connection of Bezier patches. In Chap. 14, methods to correct a mismatch be-
tween freedoms of given boundary conditions and that of a patch to be defined
are treated. At first the classic Coons' method is explained, then a patch con-
sisting of a weighted mean of patches, each of which partially satisfies boundary
condit.ions, and a. p'atch with double twist vectors or double inner control points
are introduced.
Next , a method of correction patches which compensate the mismatch of
cross-boundary tangents and higher derivative vectors is described with various
examples. In the last section the method of a rolling ball for smooth connection
of surfa.ces is explained.
In Chap. 15, a three-sided patch in operator form is treated, which has the
similaT characters as those of a usual Bezier patch. But their connection is
much more constrained because of the lesser number of control points. The
correction patch similar to that for the four-sided patch is explained for their
flexible connection. Division and degree elevation of the patch can be performed
in a similaT way.
In Chap. 16, interference of surfaces of vaTious types is treated. Intersection
of free-form surfaces is a difficult and important problem in the real application
of surface design. This chapter was completely revised from the original draft,
because a new method which has proved to give good results was found during
the writing of the draft. Differential equations which describe the intersection
curves of surfaces of various types, including offset surfaces, are given and nu-
merical met.hods of solving these different.ial equations with adapt.ive stepsize aTe
successfully applied for tracing the int.ersection curves.
All the starting points of the curve are determined by other methods de-
scribed in this section, and truncation errors in the integration process are kept
wit.hin allowable tolerances. We are freed from the problem of determining step-
size to obt.ain a next point and we can use the method even in the very near
region of a singular point where the normal vectors of both surfaces are coinci-
dent. Around a singulaT point we can investigate the behavior of the intersection
curve by considering only higher degree terms of the differential equations, which
are usually neglected. Various examples of solved results are given.
In Chap. 17, main features of two examples of the successful integrated
CAD /CAM systems are described· using data from Toyota Motor Corporation
x Preface
and Sony Corporation. In motor ca.r body design and manufa.ct.ure, there are two
groups of persons who are least adaptable to comput.erizat.ion. One is a group of
sensitive st.yle designers, who are art.ists, and the ot.her is the group of expert.s on
die-face design of stamping presses, whose accumulated knowledge on stamping
press processes is valuable for successful die-making. If human interfaces and
int.eraction procedures which these persons would agree to use willingly ca.n be
provided, a successful integrated CAD/CAM system can be const.ructed. Its
other various and highly automated operations can be implemented technically
even if they are complicated. We describe such parts of the successful system.
The other feature is that the mast.er models which have been used in various
st.ages of manufacturing are replaced by the numerical master model from which
various standard and control information is extra.ct.ed for use in manufacturing
the st.amping dies and inspection of t.he manufactured objects and the welding
fixtures and jigs.
The next example is a very small commercial product such as an in-the-ear
headphone and its case. Though the size of the objects is very small, similar
design and evaluation techniques are used, and we explain them briefly. In both
examples the theories and methods which are described in this book were applied.
In t.he Appendix, numerical met.hods of differential equat.ion solving with
adapt.ive stepsize, which are used in this book for tracing or drawing various
curves on a. surface including int.ersection curves, are explained. We have t.ried
to make the descript.ion understandable wit.hout special knowledge of numerical
integration.
3 Short Historical Note

Many of the materials for this book were taken from the author's research work
performed from the mid-1960s to the late 1980s in Japan. A short historical note
on t.his research and development is described to clarify the background.
The aut.hor's practical experience on design task dates back to the period of
1943-1945, when he engaged in the structural design of airplanes. After 1947 he
engaged in research and development work at the Research Instit.ute of Nat.ional
Railways of Japan. Among various works performed there the most famous
one was his proposal for and complet.ion of t.he on-line real t.ime seat reservat.ion
system for t.rains [3], which began its successful field operation from the beginning
of 1960. With this system he learned a great. deal about. computer system design
and the importance of man-machine interaction.
After his transfer to the University of Tokyo in 1959, he had an interest in
graphics and designed graphics hard wares devices which at that time were not
available in Japan. These included graphical input and output devices and a
unique Digital Differential Analyzer [4] . In lat.e 1964, he read papers by Sut.her-
land and Coons [5] in the proceedings of 1963 SJCC, and was greatly shocked
especially by the article by S.Coons in which a revolutionary int.elligent ma-
Preface XI
chine for interactive designing would be completed in five years. Stimulated

by Sutherland's SKETCHPAD, he designed a graphic system whose hardware
was constructed by Hitachi Ltd. in Japan and which surprised people who knew
nothing about the tricks of programming of graphics [6].
With these experiences he knew that the prediction by Coons was only a
dream and Coons' surface expressions were not adequate for high quality surface
description. So, he developed his own theory of free-form surfaces [7], for which
two la.rge motor car makers in Japan approached him, because only his theory
among all others passed their tests, as each of them informed him separately. At
that time he knew nothing about car industry. One company wanted to apply
his theory to drawings in car body design. The other company, Toyota, wanted
to apply it to NC machining of stamping dies for car bodies. For the latter he
proposed the use of a data structure for describing the shape of a part of a car
body, which would. be an internal model of the real object. This was an idea of
geometric modeling.
At that time, concerning information creation and communication and trans-
fer to downstream of the design tasks, he felt the necessity for an information
description which can be treated directly by the computer and replace the con-
ventional engineering drawings. With this idea and his surface theory, Toyota
developed a system called TINCA (Toyota Integrated Numerical Control Ap-
proach), completed around 1972 [8], which produced cutter paths for machining
stamping dies from the internal model.
While the model was used in the company, his idea on modeling led to his
design of an interactive language called GIL (Generalized Interactive Language)
which operated on the laboratory-made minicomputer at the University of Tokyo.
With these facilities, creation, manipulation and display of geometric models of
polyhedra became possible [9]. This work was performed independently without
knowledge of works of LBraid and M.E.Engli on geometric modeling [10]. An
English paper on this work was published [11]. In 1974 the author published a
book on computer graphics and its application based on his research work during
1964-1972 [12].
The work of geometric modeling was further developed by F.Kimura into the
system GEOMAP (GEOmetric Modeling And Processing) at the ETL (Electric
Technical Laboratory) of the Japanese Government, where powerful computers
were available. The GEOMAP system supported a unique two-dimensional in-
put method of data and instruction with pen and paper. Works concerning
GEOMAP were published in English [13][14].
The author of this book learned about the Bezier curve for the first time
while he was in Europe attending IFIP Congress '74. He was interested in it
and soon found a shift operator could be used to simplify its expression and
manipulation [15]. This expression has been used ever since then [15][16].
Around 1976, the Toyota Moto.r Corporation again asked the author to help
with their new project, which was to assist style designers' activities using a
XII Preface
computer to shorten the design time and to make the design shape information
a direct input to the various downstream tasks performed by the computers such
as the structure design system of car bodies and the stamping die design and
manufacturing system (TINCA system). This project seemed very attractive but
it was difficult to attain its objective in a short time.
At that. time Toyota had a system called NTDFB (New Toyota Data File
system for Body engineering) which took in measured data from a clay model
made by designers and constructed awire frame model of a car body with inter-
polat.ion by Coons' pat.ches. But the quality of the surface was not satisfactory
and it was an enormous task to correct and to convert the data for use by TINCA
system for manufacturing stamping dies.
For the new project, at Toyot.a the designers' conventional methods were sys-
temat.ically scrut.inized and analyzed. According to this information new meth-
ods of surface synt.h~sis and evaluat.ion were developed with the collaboration of
the style designers. The new methods were a kind of simulation of designers '
activities during their creative works of constructing a clay model which was the
expression of their intentions. Accordingly, the system had to provide various
methods and tools for construction and evaluation of the model in order for the
designers and modelers to feel they were using their accustomed methods.
Through the experiments, armed with the theoretical works and experience
on a prototype and after more than three years development with much effort
and investment by Toyota, the system was completed and has been used since
1981 and an epoch-making enhancement of productivity of the design as well as
high product quality of the final products have been attained. By means of this
system and downstream systems integrated with it, all the physically constructed
master models were abolished and the number of tryouts of the stamping press
with real models was extremely diminished [18].
During cooperation on the development of the system, the author found
many problems to be solved, but owing to the limited developing time they
were not all solved satisfa.ctorily. These were problems related to the evaluation,
interference and connection of free-form surfa,ces, some of which were analyzed
more elegantly afterwards [17][19][20]; these topics are treated in this book. In
1981 the author left the University of Tokyo according to the age-limit rule there
and tra,nsferred to Tokyo Denki University, where he served as the director of
the university research center. After the term of office in 1988, he had time to
make his efforts t.o write this book in which most of his practically useful work
on free-form curve and surfaces are included in a systematic way, although some
of the complicat.ed analyses were omitted owing to the introductory character of
the book.
After the successful development work of the first Seat Reservation System of
Japanese N a.t.ional Railways, the author had to be engaged in consulting work on
the development of the large on-line stock information and transaction system of
the Tokyo Stock Exchange. They Doth have quite different characters from the
Preface XIII
CAD/CAM system of Toyota, but the common factor was that they involved
expert persons without whom the systems could not be operated successfully.
The author has realized that human considerations and good interfaces with the
machine were among the most important factors for the success of computerized
systems together with their high reliability and high performance.
4 Acknowledgements
The author owes his research work in this field to many people who gave problems
to him and discussed with him and tested research results. Especially, he thanks
very much the people at Toyota Motor Corporation and Prof. emeritus T. Sata
of the University of Tokyo, who first introduced me to this field, and Prof. F.
Kimura of the University of Tokyo, who was a student of my laboratory in the
same university . and afterwards has been my collea.gue in research work in this
field.
For materials in Chap. 17, Mr. M. Ohara and Dr. M. Higashi (now at Toyota
Technological Institute) of Toyota Motor Corporation and Dr. T . Kuragano of
Sony Corporation permitted me free use of various results of the developments
performed there. All the figures except those in Chap. 17, were produced by
Mr. T. Saitoh of Tokyo Denki University and Mr. T. Kushimoto and Mr. H.
Kobayasi of Stanley Electric Corporation by calculating equations in this book.
Their resea.rch works with the author are also included in the book.
By the effort of Mr. J.Andrew Ross of Springer-Verla.g, ma.ny errors, typo-
graphical and grammatical, of the original draft were corrected; I thank him very
much for his contribution.
Tokyo, August 1991 Mamoru Hosaka

Contents
1 Excerpts from Vector and Matrix Theory 1

1.1 Introduction . . . . . . . . . . . . . . . . . 1
1.2 Notations of Vectors and Vector Arithmetic 1
1.3 Product of Vectors . . . . 3
1.3.1 Inner Product of Vectors. 3
1.3.2 Vector Product . . . . 3
1.4 Triple Products . . . . 4
1.4.1 Scalar TrIple Product 4
1.4.2 Vector Triple Product 5
1.4.3 Application Examples 7
1.4.4 Oblique Coordinate System 7
1.5 Differentiation of Vectors . 8
1.6 Matrix Notations and Simple Arithmetic of Matrices 9
1.7 Products of Matrices . . . . . . . . . . . 10
1.7.1 Mult.iplication of a Vector and a Matrix . . . . . . . 10
1.7.2 Product of Matrices . . . . . . . . . . . . . . . . . . 11
1.8 Squa.re Matrix, Inverse Matrix and Other Related Matrices 12
1.9 · Principal Directions and Eigenvalues . . . . . . . . . . 14
2 Coordinate Transformations and Displacements 15

2.1 Introduction . . . . . . . . . . . . . . 15
2.2 Coordinate Transformation Matrix 1 . 15
2.3 Calculation of Transformation Matrix 16
2.4 Coordinate Transformation Matrix 2 . 17
2.5 Movement and Coordinate Transformations 18
2.6 Application Examples . . . . . . . . . . . . 19
2.6.1 Successive Rotations in Space . . . . . . . . 19
2.6. 2 Rotation of a Body Around a Line in Space 21
2.6.3 Calculation of Geometric Constraints. . . . 22
2.7 Expressions of Movement of a Body by Reflection . 24
2.7.1 Translation . . . . . . . . . 25
2.7.2 Rotation Around an Axis . . . . . . . . . . . . . . 25
2.7.3 Movement by Four Mirrors . . . . . . . . . . . . . 27
2.7.4 Determination of Screw Axis, Rotation Angle and Translation
Distance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
2.7.5 Displacement Matrix S a,n d Mirror Matrix M . . . . . . . . . . . 29
XVI Contents
3 Lines, Planes and Polyhedra . . . . . . . . . . . . . 31

3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . 31
3.2 Equations of Straight Line and Intersection of Line Segments 31
3.3 Control Polygons and Menelaus' Theorem . . . .. . . 33
3.4 Equations of Plane and Intersection of Line and Plane 35
3.5 Polyhedron and Its Geometric Properties 1 37
3.6 Polyhedron and Its Geometric Properties 2 .. . 39
3.7 Interference of Polyhedra . . . . . . . . . . . . . 41
3.8 Local Operat.ions for Deformat.ion of Polyhedron 42
4 Conics and Quadrics 45

4.1 Introduction . . . . 45
4.2 Conics . . . . . . . . . . 45
4.2.1 Equat.ion of Conics . . . 45
4.2.2 Transform~t.ion of Equation 45
4.2.3 Classificat.ion of Conics 47
4.2.4 Int.ersection of Conics . . . 48
4.3 Quadrics . . . . . . . . . . 49
4.3.1 Coordinate Transformation 49
4.3.2 Classification . . . . . . . . 51
4.4 Intersection of Two Quadrics 53
5 Theory of Curves . . . . . 55
5.1 Introduction . . . . . . . . . 55
5.2 Tangent and Curvature of Curve 55
5.3 Binormal and Torsion of Curve . 57
5.4 Expressions with Parameter t .. 59
5.5 Curvature of Space Curve and Its Projection 61
5.6 Implicit Expression of a Parametric Curve 63
6 Basic Theory of Surfaces 67

6.1 Introduction . . . . . . . . . . . . . . . . 67
6.2 The Basic Vectors and the Fundamental Magnitudes 68
6.3 Normal Section and Normal Curvature .. . 69
6.4 Principal Curvatures . . . . . . . . . . . . . . . . . . 71
6.5 Principal Directions and Lines of Curvature . . . . . 73
6.6 Derivatives of a Unit Normal and Rodrigues' Formula 74
6.7 Local Shape of Surface. . . . . . . . . . . . . . . . . 76
7 Advanced Applications of Theory of Surfaces 79

7.1 Introduction . . . . . . .. . . . . . 79
7.2 Umbilics . . . . . . . . . . . . . . . . 79
7.3 Characteristic Curves on a Surface 1 83
Contents XVII
7.3.1 General Remarks . 83

7.3.2 Lines of Curvature 84
7.3.3 Extremum Search Curves 85
7.3.4 Cont.our Curves and Their Orthogonal Curves. 88
7.3.5 Equi-gradient Curves. . . . . . . . . . . 88
7.3.6 Silhouette Curve and Silhouette Pattern 91
7.3.7 Highlight Curves . . . . . . . . . . . . . 92
7.4 Characteristic Curves on a Surface 2 . . 95
7.4.1 Gradient. Extremum Curves or Ridge-Valley Curves. 95
7.4.2 Loci of Zero Gaussian Curvature and Loci of Extremum Principal
Curvatures . . . . . . .. . . . . . . . . . . . . . . . . . . . . 96
7.5 Offset Surfaces 97
7.6 Ruled Surfaces . . . . . . . . . . . . . . . . . 102
8 Curves Through Given Points, Interpolation and

Extrapolation .. . . . . . . . . . . . . . . . . . . . . 103
8.1 Int.roduct.ion. . . . . . . . . . . . . . . . . . . . . . . . 103
8.2 Polynomial and Rational Interpolation and Extrapolation 103
8.2.1 Lagrange's Formula . . . . . . . . . . . . . . . . . 103
8.2.2 Numerical Methods of Interpolated Points. . . . . . . . . 104
8.2.3 Rational Function Interpolation and Extrapolation . . . . 107
8.3 Polynomial Interpolation with Constraints of Derivatives. 110
8.4 Elastic Curves with Minimum Energy . . . . . . . . . . . 111
8.5 Interpolation by Parametric Curves. . . . . . . . . . . . . 113
8.6 Appendix. Derivation of Equations by Elastic Beam Analogy 115
9 Bezier Curves and Control Points 117

9.1 Introduct.ion. . . . . . . . . . . . . . 117
9.2 Curve Segment and Its Control Points 117
9.3 Bezier Curve and Its Operator Form . 118
9.4 Different Expressions of B Curve . . . 121
9.5 Derivat.ives at Ends of a Segment and Hodographs 122
9.6 Geomet.ric Properties of B Curve . . . . . . . . . . 126
9.7 Division of a Curve Segment and Its B Polygon . . 128
9.8 Cont.inuit.y Conditions of Connection of B Polygons. 1.31
9.9 Elevation of Degree of a Curve Segment 131
9.10 Expression for a Surface Patch . . . . . . 133
9.11 Geometric Properties of a Patch . . . . . 136
9.12 Division and Degree Elevation of a Patch 137
9.13 A ppendix. The Original' Form of the Bezier Curve 138
XVIII Contents
10 Connection of Bezier Curves and Relation to Spline

P~ygo~ ............. . 141
10.1 Introduction . . . . . . . . . . . 141
10.2 Connection of B Curve Segments 142
10.2.1 Scale Ratios. . . . . . . . . . . . 142
10.2.2 Conditions of C(i) Connection .. 143
10.2.3 c(n-1) Connection and Control Points 144
10.2.4 Connection Defining Polygon . . . . . 146
10.3 Introduction of 5 Polygon . . . . . . . 148
10.3.1 Locating B Points from an 5 Polygon 148
10.3.2 Increase of Vertices of an 5 Polygon . 152
10.4 B points under Geometric Connecting Condition (JC2) 154
10.5 Curvature Profile Problem in Design . . . . . . . . . . 156
10.5.1 Geometric interpretation of Dividing Ratios . . . . . . 156
10.5.2 Control of Curvature Distribution of Connected Curves. 158
11 Connection of Bezier Patches and Geometry of Spline

Polygons and Nets . . . . . . . . . . 161
11.1 Introduction . . . . . . .. . . . . . . . 161
11.2 Spline Nets and Connected Bezier Nets 161
11.2.1 Tensor Product Surfaces . . . . . . 161
11.2.2 Division of an 5 Net . . . . . . . . . . . 164
11.3 . Geometric Structure of 5 Polygons . . . 165
11.4 Menelaus Edges and Their Dividing Points 168
11.4.1 Dividing Points and Sub-edges . . . . . . . 168
11.4.2 Relations Among Dividing Points and Menelaus Edges 168
11.5 Derivation of B Polygons from an 5 Polygon 170
11.5.1 Reduced 5 Polygons . . . . . . . . . . . . . . . . . . . 170
11.5.2 Examples . . . . . . .. . . . . . . . . . . . . . . . . . 172
11.5.3 Locating B Polygons from Reduced-Truncated 5 Polygons . 173
11.5.4 Division of an 5 Polygon and Insertion of a Vertex . . . . . 174
11.6 General Formulas for Locations of B Points . . . . . . . . . 175
11.6.1 Rules of Location Symbols of B Points and Their Properties . 175
11.6.1.1 Level of Menelaus Edges and Dividing Points 175
11.6.1.2 Symbols for Location of Control Points. . 176
11.6.2 General Expressions of B Point Locations 177
11.6.2.1 Application of Location Symbols 177
11.6.2.2 Formulas for Location Symbols . . . . . . 178
11.6.2.3 Level of Edges of a B Polygon . . . . . . 178
11.7 Appendix. Orthodox Approach to a B Spline Curve 179
Contents XIX
12 Rational Bezier and Spline Expressions 181

12.1 Introduction . . . . . . . . . . .. . . . . . . 181
12.2 Rational Bezier Curves 182
12.2.1 Rational Diyision Between Two Points and Its Perspective Map 182
12.2.2 Rational Bezier Curves and Their Canonical Perspectives 184
12.2.3 Effects of Weights . . . . . . . . 187
12.2.4 Division and Degree Eleyation 189
12.2.5 Derivatives at Ends of a Segment 190
12.3 Rational Bezier Patches . . . . . 191
12.4 Rational Splines . . . . . . . . . 19.3
12.4.1 Rational B Polygons from a Rational S Polygon 193
12.4.2 G(2) Connection of Curves from a Rational S Polygon 196
12.5 Rat.ional Spline Nets and Bezier Nets. 197
12.6 Expressi9ns for Conics . . . . . 199
12.6.1 Conversion to an Implicit Form. 199
12.6.2 Classification by Weight . . . . . 201
12.6.3 Sphere and Surface of Revolution 204
12.7 Interpolation and Extrapolation with Conics. 204
12.7.1 Weight of a Control Point and Parameter Values 204
12.7.2 Division of a Rational Polygon . . . . . . .. . 205
12.7.3 Extension of a Curve Segment . . . . . . . . . 206
12.7.4 Dist ance Between a Conic and a Point Near It 209
12.7.5 Curye Fitting by Conics . . . . . . . . . . . . . 210
13 ' N on-regular Connections of Four-Sided Patches and

Roundings of Corners . . . . . . . 213
13.1 Introduction . . . . . .. . . . . . . . . . . . . . . . . . . . 213
13.2 General C(1 ) Connection of B Patches . . . . . . . . . . . . 213
13.3 Example of Closed Surface of Minimum Number of Patches 215
13.4 Three or Five-Sided Patch in Regular Patch Nets 217
13.4.1 Rounding of a Convex Region . . . . . . . . . . . 217
13.4.2 Rounding of a Conyex-Concave Mixed Region 1 . 220
13.5 Rounding of a Convex-Concaye Mixed Region 2 . 224
13.6 Rounding with a Rolling Ball . . . . . . . . . . . 228
13.7 Appendix .. .. . . . . . . . . .. . . . . . .. . 2.30
13.7.1 Connection in a Triangular Region: General Case 230
13.7.2 Connection of a Pentagonal Region: General Case 2.32
14 Connections of Patches by Blending 233

14.1 Introduction . .. . . . . . . . . . 233
14.2 Coons) Patch . . . . . . . . . . . . 233
14.3 Independent Boundary Conditions 236
14.3.1 Blending by Weighted Slm1 . . . . 236
xx Contents
14.3.2 Two-Valued Twist Vectors and Floating Inner Control Points 237
14.4 Correction of Cross-Boundary Tangent Vectors 238
14.4.1 Connection of Four-Sided Patches 238
14.4.2 Evaluation and Comparison of Methods 243
14.4.3 Three-Sided Patches . . 246
14.5 Case of C(2) Connection 250
14.5.1 Four-Sided Patches. 250
14.5.2 Three-Sided Patches . . 252
15 Triangular Surface Patches and Their Connection 253

15.1 Int.roduction . . . . . . . . . . . . . . 253
15.2 Operator Form of a Triangular Patch. 253
15.2.1 Triangular Bezier Patches . . . 253
15.2.2 Rational Triangular Patches . . . . . . 255
15.2.3 Tangent.s on Patch Boundaries . . . . 256
15.3 CO) Connection of Triangular Patches 257
15.4 Arbitrary Connection of Three-Sided Patches 261
15.5 Division of a Triangular Pa.tch 262
15.6 Elevation of Degree .. 265
16 Surface Intersections 267

16.1 Int.roduction . . . . . . 267
16.2 Intersection of a Curved Surface and a Plane 268
16.2.1 General Remarks . . . . . . . . . . . . . . . . 268
16.2.2 A Practical Method of Obtaining a Point on an Intersection
Curve . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 269
16.2.3 Curve Tracing by Differential Equation Solving 270
16.3 Points on Intersection of Two Curved Surfaces 272
16.3.1 General Remarks . . . . . . . . . . . . . . . . . 272
16.3.2 Met.hod wit.h an Auxiliary Plane . . . . . . . . 273
16.3.3 Init.ial Start.ing Points and Critical Contact. Points 275
16.3.3.1 Det.ection of Int.ersection Loops . . . . . . . . . . . 275
16.3 .3.2 Critical Points . . . . . . . . . . . . . . . . . . . . 279
16.4 Int.ersection Curves Described by Differential Equa.tions 282
16.4.1 Both Surfaces with Parametric Expressions . . . . 282
16.4.2 Surfaces with Implicit and Parametric Expressions 284
16.4.3 Both Surfaces with Implicit Expressions . . 285
16.5 Intersection Near a Probable Singular Point 286
16.6 Intersection of Offset Surfaces . . . 290
16.6.1 Intersect.ion with a Plane . . . . . 290
16.6.2 Int.ersection of Two Offset Surfaces 291
16.6.2.1 Two Parametric Surfaces . . . . . 291
16.6.2.2 A Parametric Surface and· a Surface of Implicit Form 293
Contents XXI
16.6.2.3 Two Surfaces with Implicit Expressions 294
17 Applications of the Theories in Industry 297

17.1 Int.roduction .. . . . . . . . . . . . . . . . . 297
17.2 Engineering Drawings and Geometric Models 297
17.3 Examples of Integration . 299
17.3.1 Conventional processes . . 299
17.3.2 New Integrated Processes 301
17.4 Style Design System . . . 303
17.4.1 Importance of Shape Design . 303
17.4.2 Two Aspects of Style Design 303
17.4.3 Computer-Aided Style Design. 304
17.4.3.1 Input of Simplified Drawings . 305
17.4.3.2 Classes a.nd Types of Surfaces in Style Design of Motor Cars . 307
17.4.3.3 Evalua.ti~n of Curve and Surface Quality .. .. . . 308
17.4.4 Die-Face Design System . . . . . . . . . . . . . . . 312
17.5 CAD/CAM of Free-Form Injection-Mold Products 317
Appendix. Numerical Methods of Differential Equation Solving 323

A.l Introduction. . . . . . . . . . 323
A.2 Adaptive Runge-Kutta Method . . 324
A.2 .1 Runge-Kutta Step . . . . . . . . . 324
A.2 .2 Runge-Kutta with Quality Control 324
A.2.3 Runge-Kutta-Fehlberg Method . . 325
A.3 Variable Stepsize Predictor-Corrector Method 326
A.4 Bulirsch-Stoer Method. . 327
A.4.1 Principle of the Method . 327
A.4.2 Outline of the Procedures 328
A.4.3 Integration Procedure . . 328
A.4.4 Polynomial Extrapolation 329
A.4 ..5 Rational Extrapolation . 331
A.5 Examples and Evaluation 332
Bibliography . 337
Subject Index 347

1 Excerpts from Vector and Matrix Theory
1.1 Introduction
In treating geometric objects or quantities, we use vector and matrix notations
and operations throughout this book to simplify their description and also to
make the reader's geometric understanding easier. We show in this chapter only
the essence of vector and matrix operations with their application examples,
which will be referred to from various places in this book.
1.2 Notations of Vectors and Vector Arithmetic

A vector is a quantity which has magnitude and direction. In diagrams, a vector
is expressed by a line segment with an arrow head: the line length represents
its magnitude and the arrow head indicates its direction. Usually its absolute
position need not be considered except in the case of a position vector.
In a three-dimensional space, the degree of freedom of a vector is three: the
direction and the magnitude or the three components in a coordinate system in
the space. Coordinate systems are not necessarily orthogonal, but for simplicity,
at first we use an orthogonal coordinate system and afterwards we treat an
oblique coordinate system. Three unit vectors, each of which corresponds to
one a.xis of the coordinate system, are used as the basis. Let i, j and k be the
unit vectors. Their expressions are given by their components values on the
coordinate axes:
i = (1,0,0), j = (0, 1,0), k = (0,0, I), (1.1)
and if we let components of an arbitrary vector a be ax, ay and az , then a is

expressed as
(1.2)
Multiplication of a vector a by a constant c, addition and subtraction of two
vectors a and b are defined as the same operations between their components:
c = ca = (ca x, cay, ca z ), (1.3)

c a±b=(ax±bx,ay±by,az±bz)' (1.4)
In Fig. 1.1, the above operations are shown graphically. Using equations (1.1),
(1.2), (1.3) and (1.4), a vector a can be written in a form:
(1.5)
2 1 Excerpts from Vector and Matrix Theory
a+ b
;/ /-
;f;?/ b
/. 6- b
b
Figure 1.1 Graphic representation of vectors and their arithmetic
Since the position of a. point P in the space is determined by its distance and
the direction from the origin of the coordinate system, it is a vector. Usually
when a vector represents a position, the tail of the vector p is fixed at the origin;
its component values are not explicitly subtracted by the corresponding value
(0,0,0) of the origin. Therefore, the coordinates values (Px, PY' pz) ofthe position
P are the components of its position vector p:
(1.6)
To express a vector, its components are written horizontally and enclosed by a

pair of parentheses. The other form is to write its components vertically enclosed
by a pair of large parentheses thus:
(~) (1.7)
Since this latter form occupies a large space and is not suitable for usual type-
writ.ing, when this form is needed, the following alternative is frequently used:
(1.8)
The superscript T indicates transposition.

When a reference point of a position vector p is at a position q, a displace-
ment vector d is defined by
(1.9)
Its magnitude is the distance or the length between the points p and q, which
is denoted by
(1.10)
1.3 Product of Vectors 3
The magnitude of a vector is the square root of the sum of the squares of its
components. It is also called the absolute value of a vector. A vector it whose
magnitude is 1 and whose direction is same as that of a , is called a unit vector
in the direction of a . It is derived from a vector a by dividing by its absolute
value lal :
, a
a = jaf' (1.11)
The unit vectors which have the directions of the coordinates axes x, y and z are
i, j and k as used above.
1.3 Product of Vectors

1.3.1 Inner Product of Vectors
Next , an inner pro"duct a· b of two vectors a and b is defined . The inner product
is a scalar value and is equal to the product of the absolute values of two vectors
multiplied by the cosine of the angle e between them:
a .b = lal . Ibl cos e. (1.12)
An inner product a . a, which is written a 2 , is the same as the square of the

absolute value of a. Hence,
lal = va:-a. (1.13)
If non-zero vectors a and b are orthogonal, a . b is equal to zero. Naturally, we
have j2 = y = k 2 = 1 and i . j = j . k = k . i = 0 by definition. Since a and
b can be expressed in the form (1.5), a· b is given by the sum of the products
of corresponding components of the two vectors in the orthogonal coordinate
system:
(1.14)
If the inner product of two non-zero vectors is positive, or zero or negative, then
the angle between them is smaller than, or equal to, or greater than 90 degrees.
The inner product does not depend on the order of multiplication.
1.3.2 Vector Product

A vector product a x b of two vectors a and b is a vector. Let it be c, whose
magnitude is equal to the product of the absolute values of the two vectors a
and b mult.iplied by absolute value of the sine of the angle e between them:
Icl = lal·lbllsinel· (1.15)
When the tails of a and b are coincident, lei gives the area of the parallelogram
made by them. The direction of c is equal to that of the right screw in a rotation
of a towards b by the angle e. Hence, the vector c is orthogonal both to a and
b. When the order of multiplication is reversed, the direction of the resulting

vector is also reversed:
a x b = -b x a. (1.16)
If a vector product of two non-zero vectors is zero, they are parallel. The normal
direction of a plane is determined by the vector product of two non-parallel
vectors in the plane. The front side and the rear side of a plane are defined by
the direction of its normal vector.
When there is a relation
c = a x b, ( 1.17)
the components of the product c can be expressed by those of a and b. Let their
components be
(1.18)
By the two orthogonal conditions a . c=O and b . c=O, we have
(1.19)
From the above two linear equations for the three variables cx , cy and cz , we
obtain the ratios among the three components of c,
( 1.20)
where 'Y is a constant independent of a and b. To determine the value of the

constant 'Y, we choose a = (1,0,0) and b = (0, 1,0), then c is determined by the
definition of the vector product as (0,0,1). Comparing the left-hand side and the
right-hand side of eq. (1.20), we can fix the value of 'Y equal to 1. It follows that
the components of the result vector of ax b are given by eq. (1.20) with 'Y = 1.
They can collectively be written:
J k
c = cxi + cyj + czk = ax ay a z (1.21)
bx by bz
1.4 Triple Products

1.4.1 Scalar Triple Product
The scalar triple product is a kind of inner product of two vectors, one of which
is represented by a vector product of another two vectors. Suppose the tails of
three vectors a, band c coincide. See Fig. 1.2. The value of la x bl is equal to
1.4 Triple Products 5
axb
Figure 1.2 Scalar triple product
that of the area of a parallelogram whose adjoining sides are the vectors a and
b. The unit normal n of the parallelogram is given by
(a x b)
n - -:---...,- ( 1.22)
- la x bl'
Since the inner product (n . c) is the normal component of c, the expression
la x bl(n . c), which represents a product of the bottom area and the height of
the parallelogram, is equal to the volume of the parallel piped whose adjoining
three edges are a, band c. If its value is zero, the three vectors are coplanar. If
it is positive, the vector c is in the front side of a plane determined by a x b, if
negative, it is in the rear side of the plane.
In the evaluation of the volume of the paraUelpiped, the area defined by a
and b are considered to be its bottom face. But as anyone of the areas defined
by two vectors of the three can equally well be taken as a bottom face, its volume
is given also by (b x c) . a or by (c x a) . b as well as by (a x b) . c. They are
all denoted by [a, b, c] and are called the scalar triple product of the vectors and
have the relations:
[a, b, c] = [b, c, a] = [c, a, b] = -[b, a, c]. (1.2.3 )
When any two element vectors of a scalar triple product have the same or opposite
direction or when its three element vectors are coplanar, the value of the scalar
triple product is zero. Its sign is used to determine in which side of a plane a
given point exists.
1.4.2 Vector Triple Product

A vector triple product has forms (a x b) x c or a x (b x c). These forms appear
frequently in manipulation of vectors in geometric modelling. The direction of
a vector (a x b) is normal to the plane determined by a and b. Therefore the
direction of a vector (a x b) x c is normal to (a x b), that is, it is in the plane
defined by a and b. Therefore, it is expressed by a linear combination of a and
b such as .
axb
(axb)xc Figure 1.3 Vector triple product
(axb) xC=Aa+f.lb, (1.24)

where A and f.l are scalar constants to be determined. See Fig. 3. By multiplying
both sides of eq. (1.24) with the vector c, the left-hand side becomes [c, a x b, c],
which is zero, because of the two identical elements.
The right-hand side becomes
A(a· c) + J.l(b· c) = o. (1.25)
The following relations satisfy the above equation:
A = -,),(b· c), f.l = ')'(a· c) ,
where,), is a constant independent of a, band c. To determine ,)" we set a and b
orthogonal and b equal to c, then we evaluate both sides of eq. (1.24) using the
relations .\ = _,),b 2 and f.l = 0 which are obtained from the above setting, and
determine that the constant')' is equal to 1. It follows that eq. (1.24) becomes
(a x b) xc = (a· c)b - (b· c)a. (1.26)
By changing the multiplication order and then exchanging the names of variables,
we get
a x (b x c) = (a· c)b - (a· b)c. (1.27)
Another vector product formula may be obtained. Multiplying both sides of
eq. (1.26) by a. vector d, we have
{(a x b) xc}· d = (a· c)(b· d) - (b· c)(a· d).
Its left-hand side is a scala.r triple product [(a x b), c, d]. By applying eq. (1.23),
it is equal to (a x b) . (c x d). Therefore, the formula
(a x b) . (c x d) = (a· c)(b· d) - (b· c)(a· d) (1.28)
is obtained.
1.4 Triple Products 7
1.4.3 Application Examples
Example 1. Construction of a Coordinate System

The vector triple product is used to construct a coordinate system in a space.
Let n be a normal unit vector to a plane determined by the vectors a and b:
axb
( 1.29)
n = la x bl'
Then writing ( = n, ~ = it, we make iJ orthogonal to ( and ~:
••• (axb)xa
rJ = ( x ~ = la x bl . lal . ( 1.30)
Using eq. (1.26), the numerator of the right-hand side of the above equation is
changed to . .
(a x b) x a = (a· a)b - (b· a)a = a 2b - (b· a) a,
so the expression for iJ is

. b - (b. it)it
rJ = lit x bl .
The orthogonal unit vectors ~, f] and ( thus determined make an orthogonal

coordinate system.
Example 2. Use of Eq. (1.28)
Let there be three vectors a, band c, whose tails coincide, let the plane
defined by a and b be orthogonal to the plane defined by band c, and let the
angles between a and b, between band c and between a and c be e1 , e2 and e3
respectively. Then the following relation holds:
To prove this, set d = b in eq. (1.28); since a x band b x c are orthogonal, the
left side of eq. (1.28) is zero. By rewriting the inner product with absolute values
and the cosines of the angles between the vectors concerned, the above relation
is obtained.
1.4.4 Oblique Coordinate System
A vector is usuaUy resolved into its components in an orthogonal coordinate

system, but sometimes it is required to be resolved in an oblique coordinate
system. Let ell e2 and e3 be the unit vectors of a..'Ces of the coordinate system,
let the a.ngle between el and e2 be w, and that between e3 and the normal of the
plane made by el and e2 be e.
Now we are to express an arbitrary unit vector a by its components in the

oblique coordinat.e system. Since the vector a can be expressed as
(1.31)
we det.ermine A, f1 and v in terms of el, e2, e3, a, e and w.

Multiplying both sides of eq. (1.31) by e2 x e3, we obtain
and simila.rly
a·(e3 xe l) f.1.e2 . (e3 x el) = f1[el, e2, e3],

a· (el x e2) ve3 . (el x ez) = v[el' e2, e3] '
Since the triple product [el, e2, e3] is the volume of a parallelpiped of unit
lengt.h edges , which is equal to sin w cos e, we have
A = [a, e2 , e3]/s~nwcose , }
f1= [a, e3 , el]/ SlllW cos e, (1.32)
v = [a,el , ez]/sinwcose.
e
If w = 90 degrees and = 0, then the unit vectors el, e2 and e 3 are mutually
orthogonal. Accordingly, the above equations are reduced to
1.5 Differentiation of Vectors

When a vector is a function of paTameters, it is often necessary to differentiate
the vector with respect to the parameters. For example, when r(t) represents a
curve, dr(t)/dt is a vector expressing the tangent direction of the curve. This is
obvious from the definition of differentiation.
When differentiating a vector product or an inner product of two vectors,
the process is the same as differentiation of a product of scalar variables. But
the order of differentiation must not be changed for the vector product, and we
have
d(a· b)
(1.33)
dt
d(a x b)
(1.34)
dt
Examples.
Let t , b and 11 be unit vectors orthogonal to each other and functions of a
parameter s , which is an aTC length, and let b be defined by b = t x 11 . Since
1.6 Matrix Notations and Simple Arithmetic of Matrices 9
t 2 = 1, we have t(dt/ds) = 0 which indicates that dt/ds is orthogonal to t.

Introducing a constant. K, we define n such that dt/ds = Kn. By differentiating
the relations b . b = 1 and t . b = 0 wit.h respect to s, we obtain
db dt db
b· ds = 0, ds' b + t· ds = O.
As dt/ds = Kn,
we have t.he relation (dt/ds)· b = 0 and hence we get
t . (db/ds) = O.
This means (db/ds) is orthogonal to t as well as to b, so it
has a. component of n only. Introducing a proportional constant -1', we can
express it as
db
ds = -rn.
Making the vector product of t with both sides of b = t x n, and using eq. (1.26),
we get a relation .n = b x t. Differentiating this by s, we obtain
dn db dt
---xt+bx-
ds - ds ds'
which is converted to
dn
ds = -rn x t + Kb x 11 = rb - d.
This equation is called Serret-Frenet's formula. See Sect. 5.3.
1.6 Matrix Notations and Simple Arithmetic of Matrices

Matrix notations and operations are used in this book for coordinate transforma-
tions, displacements of solid objects, and for expressions denoting objects. The
reason is that they are concise and systematic as well as visually understandable.
So, we explain t.heir notations and some of operations in this section.
Letting m and n be integers, m x n real numbers or vectors or symbols of
the same attribute are arranged in m rows and in n columns enclosed by a pair
of brackets as shown below:
A= (1.35)
The right-hand side arrangement is called a matrix of size m x nand aij are
elements of the matrix. Matrices can have names: the name of the above matrix
is A. For simplicity, an notation A = [ai j ] may be used.
A vector of m components can be thought of a matrix of size 1 x m, or its
transpose is a matrix of size m x L
Let the i-th row of the matrix A be denoted by ai:
ai = (ail ai2 ... ain), (1 ~ i ~ m). ( 1.36)
Then the matrix A can be written:
A= (1.37)
But to save space and to type it easily, the following expression is frequently
used:
( 1.38)
Let each column. of the matrix A be a transposed vector or matrix of m x 1:
(1.39)
Then the matrix A can be denoted by
(l.40)
When the size of matrices A, Band C are the same, the following arithmetic
operations:
C = A ± B, C = cA , (1.41)
are defined in such a way that the elements of C are equal to the arithmetic
operations between the corresponding elements of the matrices A and B,
( 1.42)
where c is a scalar constant.
1. 7 Products of Matrices
1.7.1 Multiplication of a Vector and a Matrix
Multiplying a vector b of m components by a matrix A of m x n, we get a vector
c of n components,
( 1.43)
A component of c corresponds to an inner product of the vector b and the
corresponding column vector of the expression eq. (1.40) of A.
Since they are matrices of size 1 x m and size m x 1, their inner product is
denoted by ba? Using this notation and eq. (1.40), we can express eq. (1.43) in
the following form,
(1.44)
1. 7 Products of Matrices 11
Comparing eq. (1.43) and (1.44), we have

Ci = ba: T, (1 ~ i ~ 11). ( 1.45)
Example. Coordinate transformation
When a position vector p is defined on the coordinate system (1 and it is
required t.o indicate this fact explicitly, p and its components are written with a
superscript indicating the coordinate system (1, such as
pel) = (Px, PY' Pz)(I).
Usually the superscript. is omitted when there is no confusion. If the same point
p is to be described by the coordinate system (2, there arise the necessity of
coordinate transformation from the system (1 to the system (2 . The vector pel)
is transformed t.o p(2) by multiplying the transformation matrix TI2 with p(l),
p(2) = p(I)TI2 . (1.46)
The detailed description on the transformation is given in the next chapter.
1.7.2 Product of Matrices

The product of a matrix B of size k x m and a matrix A of size m x 11 produces
a matrix of size k x 11. For k = 1, the multiplication is explained in the previous
section. This is extended to any integer k. Let the product matrix be C. Then
we use the notation:
C=BA. (1.47)
As in eq. (1.38), B is composed of k row vectors b j of m elements and C of k
row .vectors Cj of n elements, (1 ~ j ~ k):
B = [b l b2 ••• bk y, C = [CI C2 ••• Ck y. (1.48)
Since BA is [b l b2 ••• b k FA, the product C becomes
C = [bIA b 2 A ... bkA y. (1.49 )

Hence, the i-th row of C is
Ci = biA. (1.50)
The j-th component Cij of Ci is equal to an inner product of a vector b i and the
j-th column vector of A.
An element Cij of C, which is the product of two matrices A and B, is
constructed from the inner product of i-th row vector of the matrix Band j-th
column vector of the matrix A.
Example.
Let a matrix T12 be the coordinate transformation matrix from the system
(1 to the system (2, and T23 be the coordinate transformation matrix from the
system (2 to the system (3. The coordinate transformation matrix TI3 from the
system (1 to the system (3 is the product of TI2 and T 23 , such as
(1.51)
1.8 Square Matrix, Inverse Matrix and Other, Related

Matrices
A matrix whose numbers of rows and columns are equal is called a square matrix.
A unit matrix is one whose diagonal elements are all 1 and the other elements
are all zero. The unit matrix I has the following properties,
bI = b, AI = I A = A. (1.52)
The unit matrix I corresponds to 1 in algebraic multiplication.

When product of two square matrices A and B is the unit matrix, A is called
the inverse matrix of B or vice versa. A matrix whose rows and columns are
interchanged from the original matrix A is called the transposed matrix of A, and
is denoted by AT. Considering a vector b as 1 x n matrix, b T is an n x 1 matrix
or a column vector.
The transposed matrix of a product AB is equal to the product of the trans-
posed matrices BT and AT, that is ,
( 1.53)
Multiplying a vector b with a matrix B generates a vector:
c=bB. ( 1.54)
This is the form of the post-multiplication of a matrix.

When the pre-multiplication form is required, transposition of both sides of
eq. (1.54) gives
(1.55)
where BT is denoted by C.
Example 1. Matrix expressions of conics
The general expression for a conic is given by the following implicit equation:
ax 2 + 2hxy + by2 + 2gx + 2fy + d = 0, (1.56)
where a, b, d, 9 and h are constants. This can be expressed in matrix form as
(xyl)· [ ah hb 9f 1.(xyl)T=O. (1.57)

9 f d
Example 2. An expression for a tensor product surface patch

Let each element Pij of a n x n square matrix C represent a characteristic
value or a vector , such as a control point, of a surface patch. By multiplying Pij
by a weighting function fi(U)fJ(v) and taking the sum of all of them, we obtain
a surfa,ce patch expression, where. u and v are the paTameters. To simplify
1.8 Square Matrix, Inverse Matrix and Other Related Matrices 13
the expression and its numerical calculation, instead of the double summation
L L Ji(U)JJ(V)Pij, the equivalent matrix representation is frequently used. Let
b(u) be a vector whose components are Ji(u),(i = 1··· n) and b(v) be a vector
whose components are JJ(v),(j = 1·· ·n), then the expression of the surface
patch is formally written as
s(u, v)= b(v)· C· b(u)T = b(u) . C· b(v)T, (1.58)
where C = [PiJ
This is one of expressions for a tensor product surface patch. The derivation of
eq. (1.58) is explained in Chap. 9.
The inverse matrix of a square matrix is a square matrix which produces
the unit matrix I by its multiplication with the original square matrix. If the
relation
AB = BA = I ( 1.59)
holds, a matrix A is' called the inverse of a matrix B, or matrix B is the inverse
of matrix A. They are denoted by
(1.60)
In this case, the matrices A and B are non-singular. The inverse matrix does
not always exist: if value of determinant of a square matrix is zero, the matrix
is called singular.
'When a vector b is unknown, a vector c and a matrix B are known, and
there exists a relation:
c= bB, (1.61)
the l:lllknown vector b ca.n be obtained by post-multiplying the inverse matrix
B- 1 , if it exists, with both sides of eq. (1.61):
cB- 1 = bBB- 1 = bI = b. (1.62)
This is the solution of eq. (1.61).
The inverse of a product of two matrices is equal to
( 1.63)
When the inverse and the transposed matrices of a given matrix are the same,
then the given matrix is called the orthogonal matrix. A coordinate transforma-
tion matrix of rotation about any axis is the orthogonal matrix. For example, let
T 1 ,2 be the transformation matrix of rotation around the z axis, which is given
by
T12 = [~~:~ ~~~~e ~ 1' (1.64)

o 0 1
then the relation Ti-;/ = TZ1 = T1~ holds. Since a rotation matrix about any
axis is expressed by a product of three orthogonal matrices of rotation around
the respective coordinate axes, its inverse is equal to its transpose because of
equations (1.53) and (1.63).
1.9 Principal Directions and Eigenvalues

If a product of a vector e and a square matrix A gives a vector of the same
direction as that. of e, this is called the principal direction and the vector is
called the eigenvector of the matrix A. It is determined from the equation given
by eA = Ae, that is,
e(A - U) = 0, ( 1.65)
where A is t.he constant to be determined. For the existence of a non-null vector
e , an equation
det(A - U) = 0 ( 1.66)
must hold. This is an algebraic equation for A: the characteristic equation of the
matrix A.
When A is symmetric, the equation (1.65) has real roots which are called
the eigenvalues of {t. For each eigenvalue, the equation (1.32) gives the ratio of
the component.s of e, from which the principal directions, the eigenvectors, are
determined. The eigenvectors are proved to be orthogonal each other.
Example.
a.nd e = (ex, ey). Then e(A - AI) = °is resolved into its components:
(a - ,\)e x + hey = 0, = O.
hex + (b - A}ey ( 1.67)
If these simultaneous equations have solutions other than ex = e y = 0,
det(A-AI)-! (a-A) h !-A2-(a+b)A+ab-h2-O (1.68)

- h (b - A) - -
must hold. The above equation is a quadratic equation for A and its roots Al,),2
are given by
(1.69)
Corresponding to t.hese AI, A2, the directions of the eigenvectors el (exll eyl) and
e2(e x2, ey2) can be determined from eq. (1.67) by
ey; (\ - a) h
ex; = h =(\-b)'
i = 1,2. (1. 70)
From equations (1.69) and (1.70), we obtain

eyl ey2 (>.1 - a)
--= =-1 (1. 71)
exlex2 (A2-b) .
Accordingly, the inner product of el and e2 becomes zero:
(1.72)
Therefore el and e2 are orthogonal.
2 Coordinate Transformations and
Displacements
2.1 Introduction
In CAD and CAM, objects to be designed and manufactured have to be mod-
elled in the computer. In dealing with such models, geometric relations among
the objects have to be described and their calculations systematically performed.
Vector and matrix operations are very useful in these tasks. Among their ap-
plications, coordinate transformation and the movements of solid bodies are the
fundamental tools in treating 3D objects mathematically. We explain them in
this chapter. A position vector p of a point is denoted by p(1) to indicate that it
is seen from the coordinate system (1. The same is applied for its components:
(2.1)
When this point is to be seen from the coordinate system (2, it must be written
in the same form as eq. (2.1):
(2.2)
We 'have to est.ablish the relation between p(l) and p(2). When a body moves to
other location changing its orientat.ion, a point p(l) of the body moves to q(l)
seen from the same coordinate system. We have to also establish their relation.
Recognition and manipulation of the geometrical relations of objects in 3D
space are rather difficult tasks for a person without using real models or other
suitable visualizing methods. So, we formalize the mathematics used in three-
dimensional movements of bodies and coordinate transformations in a systematic
way so that they may easily be applied for computer processing.
In the last section of this chapter, simulation of the displacement of objects
by mirror reflections is treated, because sometimes movements of a body in space
are easily understood by analogy with mirror reflections.
2.2 Coordinate Transformation Matrix 1

Suppose the coordinate system (2 is set by rotating the coordinate system C1
around a line through its origin by a given angle. A case of translation with the
rotation of the system (2 is treated afterwards.
Since there is no translation, the components of p(2) can be expressed by
a linear combination of the components of p(1), that is, the former is obtained
16 2 Coordinate Transformations and Displacements
from the latter by multiplying with a 3 x 3 matrix, which is called the coordinate
transformation matrix T12 . So we have a relation:
p(2) = P(l}T12 . (2.3)
The position p(l) must be derived by formally applying T21 to p(2):
p(l) = p(2)T21 . (2.4)
Replacing p(l} of eq. (2.3) by eq. (2.4), we get
p(2) = p(2)T21 T 12 . (2.5)
From this the following relation must hold,
(2.6)
where I is the unit matrix of size 3 x 3. The transformation matrix T21 is the
inverse of T 12 :
(2.7)
Seeing the same point p from the coordinate system C3, we have the following
relation:
p(3) = p(1)T13 = p(2)T23 . (2.8)
Replacing p(2) in the right-hand side of eq. (2.8) by eq. (2.3), we get
P(l}T13 = P(l}T12T 23 , (2.9)

from which we obtain a law of multiplication of the transformation matrices,
(2.10)
Taking its inverse and applying the relation (2.7), we have
(2.11)
This is the same relation as eq. (2.10). We can state that in the sequence of
coordinate transformations, the intermediate coordinates have no effect. So, its
general form is
(2.12)
2.3 Calculation of Transformation Matrix

The size of a transformation matrix T12 is 3 x 3. Let the matrix T12 be
(2.13)
2.4 Coordinate Transformation Matrix 2 17
To determine its elements tij, we set p(l) = ii1) = (1,0,0)(1). Then using eq. (2.3),
we obtain the first row of eq. (2.13):
ii2) = (tll t12 t 13 )
= ((i 1 · iz) (i1 · j2) (i 1 · k 2) ).

Similarly, for the second and third rows:
ji2) = ((j1' i2) (j1' j2) (j1' k 2 ) ),
ki2) = ((k1 · i 2 ) (k1 · j2) (k1 · k 2 ) ).
Thus the matrix T12 can be written:
T12 -- [.(2) .(2) k(2)]T - [1' J' k ](2)T (2.14)
11 J1 1 - 1 .
This expression means that values of elements of the transformation matrix T12
are obtained from those of a transposed matrix of the basis vectors of the system
(1, which are seen from the system (2. Similarly, T21 is given by
1 .J k ](l)T
T 21 -- [ . 2' (2.15)
This matrix T21 is the inverse of T 12 , and also the transpose of T 12 . The trans-
formation matrix, therefore, is the orthogonal matrix.
2.4 Coordinate Transformation Matrix 2

We treat a case where the origins of the systems (1 and (2 do not coincide. The
ori~n 01 of (1 has non"zero components seen from the system (2:
(2) ( )(2)
0 1 = t 41 t42 t43 1 . (2.16)
In order to express the transformation in the same form as eq. (2.3), equations
(2.1), (2.2) and (2.16) are extended to the following forms,
p(1) (x y z 1 )(1), (2.17)
p(2) = (x y z 1 )(2), (2.18)
(2.19)
then the transformation matrix T12 takes the form:
T _ [
tll
t21
t12
t22
t 13
t23
0
0
j
12 - t31 t32 t33 0 .
(2.20)
t41 t42 t 43 1
The relations (2.5) - (2.12) hold without modification, equations (2.14) and
(2.16) are modified to
T12 = [i J k 0 ]i2)T, (2.21 )
T21 = [ i· j k 0 ]~l)T. (2.22)
y
x
z
Figure 2.1 Displacement of a body
2.5 Movement and Coordinate Transformations
In the previous section a point is fixed, but the coordinate systems (1 and (2 are
different, or the coordinate system (1 moves to a different position by rotation
and translation to become the system (2. In this section, a coordinate system
C1 is fixed, but a body B moves to a~other position by rotation and translation.
We are to formulate its movement.
A coordinate system (2 fixed in a body B becomes the coordinate system
(3 when the body B moves to another location. See Fig.2.1. Any point of the
body seen from the body coordinate systems (2 and (3 does not change though
its position p changes to q if seen from the coordinate system (1. So, we can
express
(2.23)
If we see q from the system (2, we consider q(2) has moved from p(2) and we
describe this fact by using the movement matrix S,
(2.24)
On the other hand, multiplying T32 with eq. (2.23) to get q(2), we have
q(2) = q(3)T32 = p(2)T32 . (2.25)
Comparing the above two equations (2.24) and (2.25), we have
(2.26)
2.6 Application Examples 19
We can state t.hat the value of S, which is the movement matrix of a body
1 it.s init.ial coordinate system C2, is the same as that of the coordinate trans-
)fInation matrix from C3 to C2.
If we use anot.her coordinate system C1 in eq. (2.24), we can change eq. (2.25)
o
(2.27)
rom which we have
(2.28)
~his is t.he formula. of the movement matrix in the system C1. We use S(l) in
,lace of T12 . S . T21 , then the above equation becomes
(2.29)
:ince S introduced in eq. (2.24) is the movement matrix in the system C2, it
hould be writ.ten S(2), and the coordinate transformation of S from the system
:2 to C1 becomes
(2.30)
,yhen there are successive movements of a body, which include rotations, this
an be expressed mathematically by the successive multiplication of relevant S
natrices. As it is rather difficult to grasp this kind of movement visually, the
,ther approach which uses the mirror (reflection) transformation is described in
'ect. 2.7.
L6 Application Examples
~.6.1 Successive Rotations in Space
n rot.at.ion of a body in space, the coordinate system attached to the body also
otates, so human understanding of the motion of the body becomes difficult.
3ut rot.at.ion around a coordinate axis is easier to grasp, and also construction
)f a. mechanism of rot.ation around a. fixed axis is easier, so in most cases general
otat.ion is resolved int.o successive application of single axis rotations.
With the coordinate system attached to a body, at first the body rotates
I.round its z axis by an angle '1jJ in the direction from the x to the y axes, then it
otates around its y a.xis by an angle e in the direction from the z to the x axes,
inally it rotates around its x axis by an angle ifJ from the y to the z a.xes.
Let consider t.he coordinate systems CO, Cl, C2 and C3, the origins of which
:oincide and their definitions are (refer to Fig. 2.2).
:::0: t.he init.ial coordinate system fixed in the space.

::1: the system defined by a rotation of CO around its z axis by an angle '1jJ.
::2: the system defined by a rotation of Cl around its y axis by an angle e.
::3: t.he system defined by a rotat.ion of C2 around its x axis by an angle ifJ.
Z = Zl
./ y
,------
x Figure 2.2 Rotation of coordinates
Then the transformation matrices from one system to the next are con-
structed using eq. (2.14):
T01(1j; : z)
[005'1> - ,in 1 01
sin 1j; cos 1j; 0 , (2.31 )
o 0 1
[ 00'o 0 0 ,in 0 1
T 12 (e : y) 1 0 , (2.32)
- sine 0 cos e
T23(¢>: x) = [1o cos0¢> - sin0 ¢> 1. (2.33)

o sin ¢> cos ¢>
The transformation matrix of the synthesized rotations is given by
(2.34)
For example, for 1j; = 45·, e= 60·, ¢> = 30·,
1/2./2 -.;3/4./2
T03 = [ 1/2./2 3.;3/4./2 5/4./2]
1/4./2 . (2.35)
-.;3/2 1/4 .;3/4
Its transpose is equal to the inverse, because it is the orthogonal matrix. The
squared sum of each row or column of the above matrix is equal to 1, because
each row or column of the above matrices is the unit vector as are shown in
eq. (2.14). When the origins of CO and C1 do not coincide, T03 becomes a 4 x 4
matrix, whose 4th row is the vector of the origin of CO seen from C3 and the 4th
column is the vector (000 1)T.
L
Cl
/ (w
Q(.
e • y
Figure 2.3 Rotation of a body around

x a line in space
2.6.2 Rotation of a Body Around a Line in Space

When a body B rotates by an angle w around a line L which passes a fixed point
Po in the space, we are to determine its final position B by using the coordinate
transformation.
Refer to Fig. 2.3. We use coordinate systems CO,Cl,C2 and C3 explained
below.
co: the original coordinate system in the space.

C1: The origin is at Po and the axes are parallel to those of CO.
C2: the system made by Cl rotating around its z axis by an angle '1j;, which is
the azimuth angle of the line L.
C3: the system made by C2 rotating around its y axis by an angle e, which is
the elevation angle of L.
In the system C3, the body rotates around the x axis by an angle w. The
final position is seen from C1 and then from co. Let p(O) be a point of the body
B. In C1 it is pel) = p(O) - p~O). After the rotation, p(l) moves to q(I), which is
expressed by p(1)S(1). From eq. (2.30), 5(1) is equal to (T13 ·5(3) . T31)' So we
have
q(1) = P(1)(TI3 ·5(3) . T31)' (2.36)
Since TI3 = T 12 T23 , which is given by T('1j; : z) . T(e : y), and 5(3) is equal to
T( -w : x) for it is a rotation around the x axis, and q(1) = q(O) - p~O), the final
position q(O) is given by
q(O) = p~O)+(p(O)_p~O))T('1j;: z).T(e: y).T(-w: x)·T(-e: y)·T(-'1j;: z). (2.37)
Exercise. Successive 90-degree rotations of a cube.

The center of a cube of edge size 2 is placed at the origin of the fixed co-
ordinate system with its edges parallel to the coordinate axes. Let rotations of
Ia
b
/'
C2
Figure 2.4 Interference of two cylinders
± 90 degrees around three axes be written X , XI , Y , yl, Z and ZI . Show the

following relations by experiments or calculations:
(i) X)(I=I, X 4 =y4=Z4=I,
where I is t.he unit matrix and X 2 = XX.
(ii) XZ = YX = ZY, XY = YZ I = ZIX, X = YXZ I = ZYZI,
XY Zy l ZX I = ZXI.
2.6.3 Calculation of Geometric Constraints

In calculation of int.ersection curves of surfaces or motions of mechanisms , selec-
tion of the appropriate coordinate systems is necessary for a simple description
of the objects concerned. With coordinate transformation matrices, relations
among the variables of both the bodies or components are established and the
equations to be solved are determined. An example is given below.
The int.ersection curve( s) of two circular cylinders is (are) to be determined.
Let the radii of two cylinders be PI and P2 , unit vectors of the axes of the cylinders
be a and b, and their distance be d. Coordinate systems C1 and (2 are used for
the two cylinders.
Referring to Fig. 2.4, the center axes are the z axes of both systems, the
common perpendicular of the center axes is the x axis common to (1 and (2 , and
the y axis is chosen to be orthogonal to both the z and x axes. The parameter ei
is an angle between the xz plane and a radial plane of a cylinder i. Each cylinder
is expressed as
1'(i)
1
-- ( p.t cosePt sin
f.
e
z·t 1 )(i) ,
f. (2.38)
where i is 1 or 2. On their int.ersection curve, a relation

l .(I)T - r(Z) (2.39)
1 lZ - 2
must. hold. From this equation, relat.ions between 8i and Zi are obtained.
To solve this equat.ion, first the transformation matrix T 1Z has to be con-
structed. Then from t.he equation r~I)TIZ = r~2), variables ZI and 81 for the
cylinder 1 being eliminated, the relation of Z2 and 8z is obtained. The following
values are the constant.s calculated from the given data.
(a·b)=cos¢,laxbl=sin¢.
The unit vectors (iI, jl' k 1) of the coordinate axes of the system C1 are deter-
mined:
kl = a, (2.40)
11 = -(a x b)/ sin ¢, (2.41 )
jl kl XiI = -a x (a x b)/ sin¢ = (b - cos¢a)/ sin¢. (2.42)
The vector triple product appearing in jl is changed by using the formulas

given in Sect.. 1.4.2. The unit vectors (iz, j2' kz) for (2 a..'(es are
kz = b, (2.43)
iz (a x b)/sin¢, (2.44)
jz k z x i z = b x (a x b)/sin¢ = (a - cos¢b)/sin¢. (2.45)
Since the transformation matrix has the form (see eq. (2.22)):
Tl2 = (il jl kl 01 )(2)T, (2.46)
we obtain
-1 0
Tl = [ 0 0 00]
-cos¢ sin¢ (2.47)
z 0 sin ¢ cos <jJ 0 .
d 0 o 1
Then from the equation rl1)Tl2 = r~Z), we get t.he following three equations:
-PI cos 81 + d pz cos 8z , (2.48)

-PI sin 81 cos <jJ + ZI sin <jJ = pz sin 8z , (2.49)
Plsin81sin¢+ zlcoS¢ = Zz· (2.50)
Squaring the first and the second equations and adding them, we obtain the
following quadratic equation for ZI.
sin z ¢. z~ - 2Pl sin 81 cos ¢sin ¢. ZI + (PI sin 81 cos <jJ)Z + (-PI cos 81 + d? - p~ = 0
(2.51 )
By setting the value of 81 and solving the equation, we obtain the value of Z1.
Taking the ratio of the second to the first equation, we have
-PI sin 81 cos ¢ + Zl sin ¢
tan 82 = . (2.52)
-PI cos 81 +d
We can determine the value of 82 from eq. (2.52), and Z2 from eq. (2.50). The
intersection curve of the two cylinders is thus calculated. According to given
values PI, P2, ¢ and d, the shape of intersection curve has two, one and no
loop(s) and shows special shapes between their transitions, which has singular
points where dzI/d8 l becomes indeterminate.
A more general treatment of the intersection is explained in Chap. 4 and in
Chap. 16.
2.7 Expressions of Movement of a Body by Reflection

When a body moves in space, it is difficult for a person to grasp its positions
and attitudes without visualizing it, for one cannot imagine correctly a motion
of simultaneous translation and rotation. Mathematically, successive displace-
ments of a solid body can be expressed by the successive multiplication of the
corresponding 4 x 4 matrices of displa.cement. And a discrete displacement of a
solid body can be expressed by a translation along an axis and a rotation around
it, that is , by a screw motion. For geometric understanding of its screw motion,
a. method using mirror reflections seems suitable, so it is explained.
Let us consider a mirror M, a body B and its mirror image B' in M. The
mirror M is the perpendicular bisection plane of a line segment connecting the
corresponding points of Band B'. Since the coordinate system CO in the space
also has its image CO', the description of B by the system CO is the same as that
of B' by the system CO'. However, the description B' by the system CO is not
the same, but has the relation
B'=B·M, (2.53)
where Band B' are position vectors of bodies Band B', and M is a. kind of
displacement matrix produced by the mirror M. Hereafter, these definitions of
B, B, B', B' and M, M are used rather loosely so long as confusion does not
occur. Refer to Fig. 2..5. Let p' be the image of p by M, whose surface is defined
by its unit normal vector n and its distance do from the origin of CO, and the
position of p' is given by
p' = p + 2n(do - p. 11). (2.54)
To express the above equation in a matrix form, the fourth component 1 has to
be added to the three components of a position vector, such as in
p = (Px py pz 1), (2.55)

2.7 Expressions of Movement of a Body by Reflection 25
co Z'
x CO'
Figure 2.5 Reflection by a mirror
because p' consist.s of component.s proport.ional to p and a constant component

in eq. (2.54). The mat.rix M has the form
(2.56)
which is derived from eq. (2.54). The value of the determinant of the matrix M
is equal to -1. This indicat.es that a point p' is an imaginary point seen from
the real coordinat.e system CO.
2.7;1 Translation
Consider t.wo parallel mirrors M} and Jvh separated by the distance h. A body B
is moved to B' by M}, and B' t.o B" by M 2 • Refer to Fig. 2.6(a). These processes
are described by
B" = B' . M2 = B . M} . M2. (2.57)
The distance bet.ween B a.n d B" is 2h.
This is equivalent to the translation of 2h in the direction from M} to M2
and does not depend on the absolute position of the mirror pair, that is, the
mirror pair (M}, M 2 ) can be moved anywhere in the perpendicular direction.
If t.he distance bet.ween two mirrors is zero, this is equivalent to no mirrors.
Accordingly, t.ranslations by even numbers of parallel mirrors are equivalent to
those by the two parallel mirrors.
2.7.2 Rotation Around an Axis

Let two mirrors M} and M2 intersect at. a line L and their intersection angle be ().
Then a body B rotates around L by an angle 2(). See Fig.2.6(b). This rotation
does not depend on the absolute position of the mirror pair (M}, M2 ) around the
( a)
(b)
Figure 2.6 Displacement of a body: (a) two parallel mirrors, (b) two intersecting mirrors
axis. The direction of the rotation is determined by the order of applica.tion of

the mirrors.
If the mirrors are orthogonal, the rotation angle is 180 degrees, so the ap-
plication order can be interchanged. MI' Jvh = M2 . MI' The rotation by even
numbers of mirrors intersecting at the line L is equivalent to rotation by the two
mirrors around it. Translation is the case of the intersection line at infinity.
Exercise. Explain t.he behavior of one's own image in an orthogonal mirror
standing at his front. right-hand side while he walks.
2.7 Expressions of Movement of a Body by Reflection 27
2.7.3 Movement by Four Mirrors

There are the following three cases.
1. Four mirrors M 1,M2,M3 and M4 are all parallel or intersect in one line, then
they are equivalent to two mirrors.
2. If the intersection line of M1 and M2 intersects with that of M3 and M 4,
then M2 and M3 can be eliminated by making them coincide with suitable
rotations of the pair (M1' M2) and the pair (M3, M 4) around the respective
axes. Another interpretation is that successive rotations of a body around
lines through a fixed point are equivalent to a rotation around one a..xis
through that fixed point.
3. If the intersection line of M1 and M2 does not intersect with that of M3
and M 4 , the displacement is described finally by two parallel mirrors and
two intersecting mirrors which are orthogonal to the parallel mirrors . The
displacement ·consists of a rotation and a translation, around and along the
axis determined by the two intersecting mirrors.
This displacement is represented by a screw motion along a,n d around the screw
aXlS.
Proof of 3. Refer to Fig. 2. 7( a). Let a line L1 be the intersection line of M1 and
M 2, and a line L2 be that of M3 and M 4, and the common perpendicular of L1
and L2 be L3.
If M1 and M2 are parallel, then there is no intersection line L 1, and rotating
the pair (M3' M 4) around L2 until M3 becomes orthogonal to M2, then inter-
changing the order of application of ll'h and M3, we get two pairs (M1' M 3)
and (M2' M 4), which we consider the initial states of M1, M 2, M3 and M4 by
renaming.
Then by the following procedures the rotation axis and the rotation angle
are determined.
1. Rotate the pair (M1' M2) around L1 until M2 includes L3 which is the
common perpendicular of L1 and L2. Similarly rotate the pair (M3, M 4 )
around L2 until M3 contains L 3. The two planes M2 and M3 intersect along
L 3 • See Fig.2.7(b).
2. Then rotate t.he pair (M2' M 3) around L3 by 90 degrees without moving the
mirrors M1 and M 4. We have two orthogonal pairs (M1, M2) and (M3, M 4),
whose axes are L4 and L 5. See Fig. 2.7(c).
3. Rotate each of the orthogonal pairs around the respective axes L4 a,n d L5
to make M1 and M3 parallel. The intersection of M2 and M4 becomes
orthogonal to M1 and M3.
4. Cha,nge the application order from M2 . M3 to M3 . M2. This does not
affect the result, because they are orthogonal. Then we have a parallel pair
(M1 . M 3) and an intersecting pair (M2 · M 4) orthogonal to the former. The
screw axis and the rotation angle and the translation distance are determined
from these pairs. See Fig.2.7(d).
( a) (b)
(Af3, M4) is
an orthogonal pair
( M 1, M2) is
"',''''~'"'---';-..,.--~
an orthogonar-palr--- ! L3
,
(c) (d)
Figure 2.7 Conversion of two pairs of mirrors to a parallel pair and an in tersectillg pair
orthogonal to the former: (a) initial locations of two pairs, (b) rotation of each pair to include
the common perpendicular L 3 , (c) rotation of (M2' M 3 ) around L 3 , (d) final state of two pairs
2.7 Expressions of Movemen t of a Body by Reflection 29
2.7.4 Determination of Screw Axis, Rotation Angle arid

Translation Distance
Suppose a. body in the initial posit.ion B moves to the final position B", then its
screw axis, rot.ation angle and translation distance can be determined by four
mirrors which give t.he same effect.
First, we give a simple theorem concerning two mutually mirror-imaged bod-
ies B' and B", which have a common point p'.
T1Jeorem. By a. rot.ation around an axis through p' and a reflection by a mirror

through p' and orthogonal to the axis, two bodies B' and B" can be made to
coincide.
This mirror plane M4 is determined by three points in the space: p' and
two mid-points of any two line segments joining corresponding points of the two
bodies B' and B". The axis of rot.ation which is represented by the mirror pair
(M2' M 3) passes t.hrough the point. p' and is orthogonal to the mirror M 4. The
rotat.ion angle by the mirror pair (M2' M 3) is determined by an angle between
two planes which include the axis of the pair and each of the corresponding points
of the bodies.
Using the above theorem, we can determine the screw a..xis and the rotation
angle between the initial position B and the moved position B". The body B is
moved to B' by a mirror M1 which is the perpendicular bisection plane of the
corresponding points of Band B". Since B' and B" are mutually mirror-imaged
bodies with a common point, the displacement of Band B" are represented by
the four mirrors M I , M 2, M3 and M 4. By rotating the pair (M2' M 3) around
their intersection line to make M2 orthogonal to MJ, we have two orthogonal
pairs (MI , M 2) and (M3, M 4), from which we can determine the screw axis, the
rotation angle and the displacement by the method of Sect.. 2.7.3.
2.7.5 Displacement Matrix S and Mirror Matrix M

The displacement. matrix or screw matrix S, which is introduced in Sect. 2.5, can
be expressed by two parallel mirrors and two intersecting mirrors orthogonal to
the parallel mirrors. The product of two displacement matrices 51 and 52 can
be synthesized by their corresponding mirror matrices.
Let 51 = MI . M2 . M3 . M4 where M1 and M2 are parallel and their screw
axis is L I , and let 52 = M 5 · Ah· M7 . Ms where M5 and M6 are parallel and their
screw axis is L 2 , and let the common perpendicular of LI and L2 be L 3 •
Perform the following procedures:
(MI , M 2 ) and (A15, M 6 ) along the respective screw axes to

1. Move t.he pairs
make M2 and M5 include their common perpendicular L 3 •
2. Rotate the pair (M2' M 5 ) around L3 by 90 degrees, then M2 includes L1

and M5 includes L 2 •
3. Rotate (M3, M 4 ) around L1 to eliminate M2 and M3 and rotate (M7' M 8 )
around L2 to eliminate M5 and M 7 • Then two orthogonal pairs (Mb M 4 )
and (M 6 , M 8 ) remain.
Then, as in Sect. 2.7.3, the new screw axis and the four mirrors are determined.
3 Lines, Planes and Polyhedra
3.1 Introduction
In a.dvanced "CAD, its software has to process information on the models of 3D
objects as though a person treats real ones. In construction and processing of
the models , their information must be sufficient and have no contradiction, and
their processing methods must be robust, efficient and flexible, because the re-
sult is directly connected to real processes of physical objects. From this point
of view, a polyhedron is a fundamental object to be modelled in the computer.
Its information structure and processing methods are clear, and numerical ap-
proximation enters least compared with those of other 3D objects. Polygons and
polyhedra or wire-frames are also useful in describing objects with free-form sur-
faces, because free-form curves and surfaces are defined by their control points
(refer to Chap. 9). So we explain in this chapter their basic equations as well as
their geometric and topological properties and their .interference.
3.2 Equations of Straight Line and Intersection of Line

Segments
A line which passes a point P and has a direction of a. unit vector it is expressed
with a parameter t by
r(i) = p+a·i, -00 < i < 00. (3.1)
This is a line of infinite length. An equation of a line segment joining two points
Po and Pl is
r(i) = Po' (1 - i) + Pl' i , O~i~l. (3.2)
Next, we are to determine an intersection point of two line segments Ll and L2
both of which are on a plane. Equations of these line segments joining Po and
Pl' and joining P2 and P3 are given by
rl(i) = Po' (1- i) + Pl 'i, (3.3)

r2(i) = P2 . (1 - i) + P3 . i. (3.4)
Since these two lines are on a plane , three vectors made from the end points
of the line segments must be coplanar, that is
(3.5)
32 3 Lines, Planes and Polyhedra
Po Po
Po
Figure 3.1 Intersection of two line segments
The unit normal vector of the plane is given by
n = (PI - Po) X (P3 - P2)

(3.6)
I(PI - Po) X (P3 - P2) I
For the two line segments Ll and L2 to intersect, the two end points of Ll
must be on both sides of L2 and those of L2 must be on both sides of L1 . These
conditions are formulated in the following way: referring to Fig.3.1, if signs of
two scalar triple products
are different, and also signs of
are different, the line segments Ll and L2 intersect.

The intersection point is determined by the equation rl(i 1 ) = r2(i 2), which
IS
(3.7)
Multiplying (PI - Po) x n with the above equation to eliminate ill we obtain
i _ [ Po - P2' PI - Po, n 1 (3.8)

2 - [ P3 - P2' PI - Po, n l·
Similarly, multiplying (P3 - P2) x n, we obtain
i - [ Po - P2' P3 - P2, n 1
1 - [ P3 - P2' PI - Po, n r (3.9)
With these parameter values, the intersecting point of two line segments Ll and
L2 is given by rl(t 1 ) or r2(h) of eq. (3.3) and eq. (3.4).
3.3 Control Polygons and Menelaus' Theorem 33
Figure 3.2 Generation of a surface

patch of degree one
3.3 Control Polygons and Menelaus' Theorem

In eq. (3.2). if the two end points Po and PI are not static points, but move on
further line segments La and Lb as linear functions of t, then r(t) becomes the
equation of a, curve of degree two of t. If the end points of the line segments La
and Lb move also as linea,r functions of t on yet further line segments, a, curve
equation of degree three can be obtained.
When the points Po and PI move on lines as linear functions of another
parameter s independently of t, r( t) becomes also a function of s, that is, r(t, s)
represents an equation of a surface patch of degree one. See Fig. 3.2. If the
corner points of a surface patch r(t, s) move as functions of t and s on other
patches of degree one, then the surface patch r(t, s) becomes of degree two. In
these kinds of curves or surfaces, their end points or corner points determine
their shapes. These points are closely related to the control points, which have
important functions for analysis and synthesis of curves and surfaces. They are
treated in detail in Chap. 9.
Two useful theorems in the primary geometry for treating the control points
are explained in the following.
(I) Refer to Fig. 3.3.
- Lo and Ll a,re adjacent edges of a polyline made from three points Po, PI and
P2. Let a point qo on La and a point qI on L1 divide each edge in the ratio
t : 1 - t, where 0 ~ t ~ 1, and let the line segment joining qo and q1 be L 2 .
- Let a points q~ on Lo and q~ on L1 also divide the edges Lo and L1 in the
ratio i' : 1 - t', where 0 ~ t' ~ 1, and let the line segment connecting q~ and
q~ be L~.
- Then L2 divides L~ in the ratio t : I-t and L~ divides L2 in the ratio t' : I-t'.
The inverse theorem says that a point q2 which divides L2 in the ratio t' : 1 - i'
is collinear with q~ and q~.
Since these theorems are closely related to Menelaus' theorem which is ex-
plained below, in this book we cail such divisions of edges a Menelaus division of
Po
Figure 3.3 Menelaus division of edges I
edges. For t.he applica.tion of t.he Menelaus division, t.he following procedure (re-
fer t.o Chap. 10) is used in the connection of Bezier polygons or the construction
of a spline polygon:
- Let. t.hree consecutive vertices of a. polygon be Po,' PI and P2, and one edge
(POPl) be divided into n sub-edges in the dividing ra.tio (AI : A2 : ... : An) at
the dividing point.s Po i for i = 1 ... n - 1.
- The ot.her edge (PlP2) is divided int.o n sub-edges in the ratio (A2 : A3 : ... :
A.n+l) at the dividing points Pl,i for i = l..n - 1.
- \Ve make an opposit.e edge of the vertex PI by connecting PO,i' and Pl ,i-l, and
divide this line in the ra.tio (A2 : A3 : ... : An) at the dividing points PO,i,j' for
j = l..n - 2.
- Then PO,j, PO,i,)-l' and Pl,)-l are collinear, or PO,J,i-l and PO,i,)-l are coincident,
where PO,j,i is the i-th dividing point of a line joining PO,j and Pl,j-l divided
in the rat.io (.\2 : A3 : ... : An).
(II) Refer t.o Fig. 3.4. Let POPIP2 be a triangle, whose edges are ao, al and a2,
and point.s qo, ql and q2 divide ao, al and a2 in the ratio (to: I-to), (tl : I-t l )
and (t2 : 1 - t 2) as shown in the figure, where 0 :::; to, t l , t2 :::; 1. If
then line segment.s Poqo, PI ql and P2Q2 are coincident. This is Ceva's theorem.
The inverse of the Ceva's theorem is as follows: let an intersection point of line
°
segment.s Poqo, Plql be 0, then the three points P2' and q2 are collinear. For
the applicat.ion of t.his theorem, see Sect. 13.4.1.
The above theorems can be proved easily by applying Menelaus' Theorem.
Referring to Fig. 3.5, it states thal for a triangle abc and any line intersecting
3.4 Equations of Plane and Intersection of Line and Plane 35
Po P2 Figure 3.4 Ceva's theorem
a e
Figure 3.5 Menelaus' theorem
with three edges be, ea and ab or their extensions at d, e, and I, where the
distance between points band d is denoted by bd or db, the relation
bd ee al = 1
(3.10)
de ea Ib
holds. Inversely, three points d, e and I (or d, e and b) are collinear if the above
relation holds in the triangle abc (or the triangle a Ie).
3.4 Equations of Plane and Intersection of Line and Plane

To distinguish the front side from the rear side of a plane, the direction of the
unit normal vector n of the plane is defined to be from its rear side to its front
side. The direction of a boundary line or curve which encloses a part of a front
do
y
x
Figure 3.6 Symbols for a plane
plane is defined to be positive when the enclosed part of the plane is on the
left-hand side of the boundary.
Refer to Fig. 3.6. Let r be the position vector of a point on a plane, and
let the distance from the origin to the plane be do (do is defined positive, if the
origin looks at the rear side of the plane, so the distance from the plane to the
origin is -do), then the relation
r·n=do (3.11)
is an equation of the plane. Another expression
r( u, v) = Po + uel + ve2 (3.12)
is sometimes used for a plane which passes through Po and contains two different
non-zero basis vectors el and e2, u and v being parameters.
A line through Po with its direction a is given by
r(t) = Po + a .t. (3.13)
By substituting eq. (3.13) into eq. (3.11), we obtain the parameter value of the
intersection point of the line and the plane:
(3.14)
The numerator of the above equation expresses the direct distance
(3.15)
from the point Po to the plane, and the denominator the cosine of the angle
between the line and the normal of the plane. It follows that t gives the distance
along the line from the point Po to the plane.
Figure 3.7 Convex and concave edges
If the distance d is positive, Po looks at the rear side of the plane, if negative,
it looks at its front side and if zero, it is on the plane.
Exercise. Determine the common perpendicula,r of two lines in space. (Hint:
the direction of the common perpendicular is that of the vector product of the
directions of the two lines. Determine a plane containing one of the lines and
pa,rallel to the common perpendicular, The intersection of the plane and the
other line is on the common perpendicular.)
3.5 Polyhedron and Its Geometric Properties 1

A polyhedron is a body enclosed by plane segments called faces. If all the dihedral
angles between a.dja~ent faces of a polyhedron, which are measured between the
front sides, are greater than 180 degrees, it is called a convex polyhedron. Any
points which do not look at the front sides of any faces of a convex polyhedron
are inside the polyhedron. If the plane equations of all the faces of a convex
polyhedron are given, the polyhedron is uniquely defined .
But if a general polyhedron is to be uniquely defined, not only the equations
of its face planes but also the relation of adjacency of its faces have to be given.
From these data, the topological relations between its edges , vertices and its
faces as well as their geometric values are obtained.
Let the equation of a plane 7rl including a face Fl be rl . nl = dl and that of
its adjacent face F2 be r2 . n2 = d2 • The direction of the intersection line of the
two adjacent faces Fl and F2 (refer to Fig. 3.7) is given by
(3.16)
If the dihedral angle is greater than 180 degrees, the edge between them is
named a convex edge, if smaller than 180 degrees, it is a concave edge. Whether
Figure 3.8 In tersection of three

planes
an edge is convex or concave is determined not by the vector a only, but by it

together with the direction of the edge which is a boundary of a face Fl' If the
directions of the intersection line and the edge are the same, the edge is convex,
otherwise concave. For the adjacent face F2 , as the edge direction is reversed, its
boundary line vector it should be written
(3.17)
instead of eq. (3.16).

Next, when the intersection line L12 of the planes 7rl and 7r2 is given, let this
line L12 intersect with the third plane 7r3: r3' ll3 = d3 • The intersecting point is
determined as follows. The value of the parameter t from a point Po on L12 to
the intersection point with the plane 7r3 is given by eq. (3.14). Accordingly, the
intersection point of the three planes 7rl ,7r2 and 7r3 is determined by eq. (3.14)
and eq. (3.16) being substituted into eq. (3.13) as
(d 3 - Po' ll3)(lll X ll2)
r = Po + [lll, ll2, ll3]
. (3.18)
From this point, two intersection lines of the planes (7r1,7r3) and (7r2,7r3) start.
Their directions are given by eq. (3.16) or (3.17) with ll3 substituting III or ll2 .
Example. Refer to Fig. 3.8. Edge tracing and an intersection point of three
planes 7r1o 7r2 and 7r3'
Let a. plane through the origin and parallel to the plane 7rl be 7r~, and a
similar plane parallel to the plane 7r2 be 7ri. Starting from the origin along the
intersection line of 7r~ and 7r~, which is denoted by (7r~, 7r~), one can reach the
intersecting point PI with 7r3, which is denoted by (7r~,7r;,7r3)' PI is given by
eq. (3.18) with Po = 0:
(3.19)
Starting from PI along the intersection line (Jr3, JrD of the planes Jr3 and Jr~,
we get to the intersecting point P2 with Jr2, which is (Jr~,Jr2,Jr3) and given by
From P2 along the intersection line (Jr2, Jr3) of Jr3 and Jr2 , we reach the intersection
point P3 with Jrl ,which is (JrbJr2,Jr3) and given by
(d l - P2 . lll)(ll2 X ll3)
P3 = P2 + . (3.20)
[llb ll2, ll3]
In the above equations, (Pl' ll2)' (Pl' lll) and (P2 . lll) are zero because they
include factors [llbll2,ll2] and [lll,lll,ll3]' So the intersection point P3 ofthe three
planes 11"1 ,11"2 and Jr3 is given by
(3.21)
This is the solution of linear simultaneous equations of three variables, or the

intersection point of three planes.
3.6 Polyhedron and Its Geometric Properties 2

Each face of the polyhedron is separated from its adjacent faces by the edges.
Both ends of an edge are vertices. For a closed polyhedron, if the numbers of its
faces , its vertices and its edges are F, V and E, the following Eulerian condition
holds:
F + V - E= 2. (3.22)
In a general polyhedron, its edges are convex or concave. Refer to Fig. 3.9.
Its vertices, to which the edges gather, are convex, concave or mixed, according
to the characteristics of the gathered edges. Its faces which are bounded by
the edges are also convex, concave or mixed according to the properties of the
bounding edges. A polyhedron of all convex vertices can be converted to a
concave polyhedron by inverting the direction of its face normals. Generally,
convex vertices of a polyhedron do not connect directly to its concave vertices,
but do so through the mixed vertices. The same is true for its faces; there are
mixed domains between convex and concave regions.
A polyhedron with a mixed domain can be divided by a plane, which coincides
with a face, into two parts. If the dividing plane is large enough, the divided
parts may be more than two, but the dividing process is stopped when one loop
of the dividing lines is generated. With this process, concave edges contained
in the dividing plane are eliminated. When the polyhedron becomes hollow, its
openings must be closed by putting in face planes. With this repeated division
of the polyhedron, the original one is resolved into a set of convex polyhedra.
.: Convex vertex
0: Concave vertex
A: Mixed vertex
Figure 3.9 Edges of a polyhedron
n n
Pl :Outside point
P 2 :Inside point Figure 3.10 Points and a convex
n
polyhedron
The geometric characteristics of a convex polyhedron are simple. See

Fig. 3.10, where some examples are shown.
- A point is inside of the polyhedron if it looks at rear sides of all the faces
('~ooks at a rear side" means that the sign of inner product of the normal
vector of a. face and the vector from the point to the face is positive). A point
is outside of the polygon if it looks at both the rear side and the front side
faces, and it is on a face if it looks at all the rear side faces except the one on
which the point exists.
- A line segment joining two points outside a convex polyhedron does not inter-
sect with it, if the two end points look at a common face of the polyhedron or
if any point P on the line segment and one of its end points look at a common
face of the polyhedron, and the same point P and the other end point look at
3.7 Interference of Polyhedra 41
Figure 3.11 Silhouette edge loop
another common face. In this case each of the line segments is outside of the
convex polyhedron.
Looking from a 'point outside of a. polyhedron, one can classify its faces into
two groups; those which the point looks and those which it does not. The
bounda.ries of the two groups necessarily make a loop, otherwise its ends go
to infinit.y. This loop is a silhouette line of the polyhedron looking from the
point. For a convex polyhedron this loop is simple. In a general polyhedron ,
there may be more than one boundary loop which interfere with each other or
with themselves on the projected plane. The whole loops or parts of the loops
become the silhouette lines of the polyhedron. See Fig. 3.11.
3.7 Interference of Polyhedra
An edge of a polyhedron is an intersecting line of neighboring face planes 11"1

and 11"2, and its end point, which is a vertex, is an intersecting point of the edge
with the their common adjacent face plane 11"3. Edge traversing of a polyhedron
is intersection line traversing of face planes of the polyhedron. The process of
traversing intersection lines of two polyhedra is similar to that of tracing on its
own edges, but the intersection lines are those generated by face planes 11"1 and
11"2 which belong to different polyhedra B1 and B 2 •
Suppose that a pair of faces 11"1 and 11"2 of the interfering polyhedra has been
determined and names of adjacent faces of each face are given. The end point of
the intersection line of the pair is an intersecting point with one of the candidate
faces, which a.re the adjacent faces of 11"1 or 11"2' See Fig. 3.12. By applying
eq. (3.14) to each of the candidate faces, we select the nearest true intersecting
point wit.h one of t.he adjacent faces. Then one of the current faces 11"1 or 11"2
is replaced by the new one and the new pair of interfering planes generates an
int.ersection line which joins the previous intersection line segment. In the figure,
pairs of int.ersecting faces are (11"111"2) , (11"211"3) , (11"311"10) , (11"1011"5) , (11"511"8) , ••••
In t.he usual case one of the face planes of the pair is replaced by a new face
Figure 3.12 Intersection of faces of polyhedra
plane t.o make a new pair, but in the special case when the previous intersection
line passes through. a vertex of either polyhedron or when the vertices of the
two polyhedra coincide, the previous face pair ends there and a new face pair
starts, which is selected from the faces around the vertex or vertices concerned.
Continuing the above process of determining the intersection until we reach the
starting int.ersection line, we complete an intersection loop.
Though the procedure stated above can apply to general polyhedra, if we
divide the original polyhedra into sets of convex polyhedra, we can determine
their intersection loops easier than by attacking the problem without dividing .
3.8 Local Operations for Deformation of Polyhedron

The shape of a polyhedron can be modified by so-called local operations such
as sweeping a face, moving an edge and its vertices, generating a new vertex
and new edges or eliminating some of them from the original polyhedron so as
not to violate the Eulerian relation. This process does not require interference
calculations, which usually take a long time. Since the Eulerian relation is derived
only from a topological net offaces, edges and vertices, and does not concern with
realizability of a polyhedron, it is sometimes necessary to check the realizability
or interference of the generated parts of the polyhedron after the modification of
shape by the local operations.
Figure 3.13 shows a simple example of creation of a polyhedron by surface
sweeping, which is one of the local operations.
(i) Sweep a square, which is on the xy plane and of size a, in the z direction
by length a. This makes a cube and increases the number offaces, vertices
and edges by 4, 4 and 8 respectively,
(ii) Make a small square of size b at the corner of the front face A of the cube.
This creates one face, three vertices and four edges.
(iii) Sweep the new small square on the face A in the direction x by length b.
This creates two faces, three vertices and five edges.
3.8 Local Operations for Deformation of Polyhedron 43
/x
(iii),(iv)
(ii)
( vij )
(v),( vi)
Figure 3.13 Example of Euler operations
(iv) Make a b x a rectangle at the corner on the side face B of the body. This
creates one face, two vertices and three edges.
(v) Sweep the new face in the y direction by length (a - b). This creates one
face, two vertices and three edges.
(vi) Make an a x a square on the upper face C . This creates one face, one
vertex and two edges.
(vii) Sweep the a x a square on the face C in the direction z by length (a - b).
This creates two faces, one vertex and three edge.
The final body created is the same as that of one cube. of size a interfering with
another cube of the same size displaced by (a ~ b) in the x, y and z directions.
In each process described above, the increase in the number of faces and vertices
is equal to the increase in the number of edges. So, the processes do not violate
the Eulerian relation of eq. (3.22).
A general form of eq. (3.22) is given by
F +V - E = 2(5 - H),
where Sand H are the numbers of shells (connected surfaces) and holes of the
solid object. An advanced explanation and discussion of the topology and its use
for construction of data structure of geometric models are to be treated in the
book by Kimura.
4 Conics and Quadrics
4.1 Introduction
Conics and quadrics are frequently used in various parts of shapes in engineering
products. Though they are only one degree higher than lines and planes, their
expressive capability is far greater than the latter. Since their geometric and
algebraic properties have been fully investigated theoretically, we can use them
with confidence. Knowledge of them is necessary not only for their appropriate
use, but also for estimating characteristics of the shape around a point on a free-
form surface and understanding various techniques of analyzing surface problems.
Moreover, solid bodies can interfere in practice and this is a rather compli-
cated problem, but using quadrics of special types their interference analysis is
not difficult. So we treat such cases in this chapter and leave general ones to
Chap. 16. Conics are also be expressed as the rational Bezier curves of degree
two, which are tr~ated in Chap. 12..,
4.2 Conics
4.2.1 Equation of Conics
A general implicit expression of a conic is given by
ax 2 + 2hxy + by2 + 2gx + 2fy + d = 0, (4.1)
where a,b,f,g and h are constants, and a ~ 0 can be assumed. Its matrix form is
(x y 1)· [~ } ~ j. (x Y I)T ~ O. (4.2)
Using a symbol C for the above central matrix, which we call the characteristic
matrix of a conic, we write the above equation:
(x y 1) . C· (x Y l)T = O. (4.3)
4.2.2 'I\'ansformation of Equation

We transform this general equation to the special forms in which it is easy to
grasp their characteristics. At first we translate the origin of this coordinate
46 4 Conics and Quadrics
system (1 to (2, whose origin is at (xo, Yo) of (1. Its transforma:tion matrix and
its inverse are given by
o
1 TZ1 = [~ ~ ~].
-Yo Xo Yo 1
Applying t.he following relation
(x y 1)(1) = (x Y 1)(2)T2,1
to eq. (4.2), we obtain
(x y 1)(2). TZ1 . e· T~ . (x Y 1)(2)T = O.
Given to no risk of confusion, we omit. the superscript indicating the coordinate

system (2. The matrix part T21 . e . T~ of eq. (4.2) is changed to
a h g,]
[ h b l' ( 4.4)
g' l' d' '
where
g'=axo+hYo+g, h'=hxo+bYo+f, d'=gxo+fYo+d. (4.5)
From the first. two equations of (4.5), we can determine Xo and Yo so as to make
g' and h' vanish if the value of the determinant taken from the coefficients of
Xo and Yo is not zero. Denoting the left-upper 2 x 2 sub-matrix of e a:s D and
assuming its determinant is not zero:
(4.6)
we obtain
-I-g
Xo - _f hi·
b • det(D), _I
Yo - ha -f I·
-g .det(D), (4.7)
and substitut.ing the above values int.o eq. (4.5), we have
d' = det(e). (4.8)

det(D)
Then, eq. (4.3) becomes
(x y 1)· [~ ~ ~]. (x Y l)T = O. (4.9)

O· 0 d'
4.2 Conics 47
By rotating the coordinate system around its origin, we can transform the matrix
(4.9) into a diagonal form. We explained in Sect. 2.6 the eigenvalue method for
obtaining a diagonal matrix. Using the eigenvalues .AlJ .A2 of the upper-left 2 x 2
submatrix in eq. (4.9), the equation (4.9) is transformed to the following form:
.AI 0 0
(xy1)· [ 0 .A2 0
1.(xy1)T=O, (4.10)
o 0 d'
where
( 4.11)
Corresponding to these .Al,.A2, the directions and e e + 1r/2 of the eigenvectors
are given by
(.AI-a) h
tane= h = (.AI-b) (4.12)
e
Exercise. Show the same results are obtained by rotation of the conic around
its origin.
Hint.. Since the rotation matrix S (refer to Sect. 2.6.1) is given by
5 = T( -e : z) = [ eo~ e
- sm casee ]
e sin
the rotational part of eq. (4.8) is
4.2.3 Classification of Conics

Equation (4.10) is
a'x 2 + b'y2 + d' = 0, (4.13)
where a' = .AI and b' = .A2.
For ab - h2 =1= 0, conics ha.ve a center given by eq. (4.5). We can classify these
comes:
- an imaginary ellipse if a', b' and d' have all the same sign,
- an ellipse if a' > 0, b' > 0 and d' < 0,
- a. hyperbola if a' > 0, b' < 0 and d' < 0,
- two straight lines if d' = 0, which is equivalent to det( C) = 0 from eq. (4.8).
For det(D) = ab - h2 = 0, eq. (4.1) is transformed into
(Vax + .../by j2 + 2gx + 2fy + d = O. (4.14)

Using the coordinate system (u, v) instead of (x, y):
u y'ax + Vby + (g/y'a + f /Vb)/2,

v y'ax - Vby + (g/y'a + f /Vb)/4 - d/(g/y'a + f /Vb),
we transform eq. (4.14) into
u2 + (g /.;a - f /Vb)v = o.
This equat.ion is classified as
- a parabola if f /g #- ~,
- a real line if (g / y'a - f /Vb) = 0,
- t.wo real or imaginary lines if f = 9 = O.
The conics are expressed in parametric forms in various ways. A rational Bezier
expression of degree two covers all kinds of conics. This is treated in Chap. 12.
Exercise. Prove that eq. (4.1) is resolved into a product of two linear functions of
x and y by the condition det( C) = 0 . In this case a conic becomes two straight
lines.
4.2.4 Intersection of Conics

Intersection point.s of two conics are generally given by root.s of a quartic equa-
tio!l which is produced by eliminating one of the variables from the given two
conics. The number of intersecting points is at most four. Since the number
of int.ersection points of a line and a conic is two, if we can reduce the problem
of the intersection of two conics to that of the intersection of two lines and one
conic, we need not solve a quartic equation.
Let charaderistic matrices of two conics be C I and C 2 • For brevity, we call
these conics also C I and C 2 . Now we consider a conic C3 , whose characteristic
matrix is a linear combination of C I and C 2 , for instance C3 = C 2 - p,CI where
p, is a constant.. The conic C 3 intersects with the conics C I or C 2 at the same
points as those of intersection of the conics C I and C 2 , because the intersecting
points satisfy each equation of the conics C I , C 2 and C 3 . The conic C3 is called
a pencil of C I and C 2 •
If we can make an appropriate pencil which becomes straight lines, our ob-
jective is at.tained. The condition that a conic equation is resolved into a product
of two straight. lines is that the value of the determina.nt of its characteristic ma-
trLx is zero. This is proved from equations (4.8) and (4.9). The characteristic
matrix C3 of the pencil is equal to that of C 2 - p,CI , where p, is a constant to be
determined. The condition that the pencils are straight lines is
(4.15)
4.3 Quadrics 49
...... .... ..... . . . . .J
Figure 4.1 Three set of pencils of two ellipses
Since the degree of the above equation for jJ, is three, there is at least one real
root, which corresponds to the two lines. Intersections of these lines and one of
the conics give desired points. The number of intersecting points is from zero
to four. When t.here are three real roots, we have t.hree set of two straight lines
which represent the pencils. Their intersecting points are those of the conics C 1
and C2 . An example is shown in Fig. 4.1.
4.3 Quadrics
4.3.1 Coordinate Transformation
A quadric is expressed in a matrix form as
(x, y, z, 1) . Q . (x, y, Z, l)T = 0, (4.16)
where
a h g l
Q= [ h b f m
1.
(4.17)
g fen
l m n d
Let the left-upper 3 x 3 sub-matrix of the matrix (4.17) be
D=[~ ~f ~l.
. g c
(4.18)
If det(D) is not zero, we can translate the origin ofthe coordinates to (xo, Yo, zo),
which are given by
(4.19)
The matrix Q is changed to
Q' =[
a
h
h g
b f 0
01
(4.20)
g f co'
o 0 0 d'
where
d' = lxo + myo + nzo + d = det(Q}fdet(D). (4.21 )
To transform the matrix (4.18) into a diagonal form, we have to determine eigen-
values of the mat.rix D, whose characteristic equation is (refer to Sect. 2.6):
(4.22)
This is a cubic equat.ion in ).:
).3 _ (a + b + C).2 + (ab + be + ca - f2 -l- h2),\ - det(D) = O. (4.23)
It has three real roots ).1, ).2 and ).3, which are the eigenvalues. The coefficients
of ).2 y and )..0 of eq. (4.23),
(a + b + c), (ab + be + ca - f2 -l- h2), det(D), (4.24)

are used for determining the eigenvalues, which are independent of the coordinate
systems used, and consequently they are invariant and are equal to
Let the principal directions corresponding to ,\; be (cxi,;3i, l'i); from eq. (4.21),
their ratio is given by
.. .. . _I
cx, . ;3, . 1'. - (b -f \) (c -f ).i)
I:. - I hg f
(c _,\;) 1·1. hg (b -f ).i) I. (4.26)
4.3 Quadrics 51
Since the three eigenvectors (principal directions) are mutually 'orthogonal, they
make a coordinat.e system. Then the matrix (4.20) becomes diagonal, elements
of which are A1J A2, A3 and d'. Accordingly, the equation of a quadric is given
by
(4.27)
4.3.2 Classification
The shapes of quadrics are classified as follows:
1. A1JA2,,\3 and d' have the same sign: an imaginary ellipsoid.
2. Al "\2"\3 are of the same sign and d' has the opposite sign: a real ellipsoid.
3. one or two of A'S are negative and d' is not zero: a hyperboloid , which is
classified into two types according to the sign of the A's and d':
(i) ,\} > 0, ~2 < 0, A3 < 0 and d' < 0: eq. (4.27) becomes a hyperboloid of
two sheets.
(ii) A}, A2 > 0, A3 < 0 and d' < O:eq. (4.27) represents a hyperboloid of
one sheet a.n d can be factored into
{{i::x + V- A 3Z H{i::x - V- A 3Z} = {vi-d' + {i:zyHvi-d'- {i:zy}.
(4.28)
Accordingly, there are two sets of generator lines which satisfy the above
equation,
{Ax + vi- A3 Z } ,{vi-d' + Ay} } (4.29)
{Ax-vi- A3 Z } (lhHvi-d'-Ay} '
{Ax + vi- A3 Z } = ,{vi-d' - AY} } (4.30)
{Ax - vi- A3 Z } = (lhHvi-d' + Ay} ,
where, is an arbitrary constant.
4. one or two of the ,\'s are negative and d' is zero: a cone. A cone also has a
generator. Those quadric surfaces which have generators are ruled surfaces
(refer to Sect. 8.9).
In the above cases, since the point (xo, Yo, zo) is determined from eq. (4.19), these
quadrics are called central quadrics.
When det(D) = 0, the center cannot be determined, but eq. (24) becomes
A3 - (a + b + c),\2 + (ca + ab + bc - f2 -l- h 2 )A = O. (4.31)
One of its roots '\3 is zero; let the others be A}, A2. Taking the corresponding
principal directions (O!i' i3i, 'i) as the coordinate axes, we transform eq. (4.17)
into the following form:
1
r~ ~ o I'
m'
(x y Z 1) . o nd' . (x y Z l)T = 0, (4.32)
I' in' n'
where (I', m', n') are given by

CYI
(l, m, n) [ /31 (4.33)
')'1
Equation (4.32) is
(4.34)
Translat.ing t.he coordinate origin of eq. (4.34) to eliminate its t.erms including x
and y, we have one of the following forms:
- AIX2 + A2y2 = -2n' Z : elliptical or hyperbolic paraboloid.
- Al x 2 + A2y2 = d' : elliptical or hyperbolic cylinder.
When t.wo eigenvalues are zero, eq. (4.32) for A2 = 0 can be transformed into
the following forms by a transla.tion parallel to the x axis and a rotation around
the x axis:
- AIX2 = 2m'y : parabolic cylinder.
- AIX2 = d' : two paTallel planes.
They aTe non-central quadrics. A hyperbolic paraboloid has the generators:
{Ax + V- A2Y} = ')'( -2n')z } (4.35)
{Ax - V-A3Z} = (1/1') ,
{Ax + V- A2Y} (4.36)
{Ax - V- A3 Z } : (1/1')( -2n')z }.
All cylinders have the generators which are parallel to the Z axis. General
quadrics can be t.ransformed into the canonical forms of central or non-central
quadrics. Some of them have generat.or lines.
The quadrics mostly used in CAD are t.hose of revolution, which have two
equal eigenvalues. Their equations and names are
sphere x 2 + y2 + Z2 = r2
circular cylinder x 2 + y2 = r2
circular cone x 2 + y2 = C2Z2
circular paraboloid x 2 + y2 = c2z
circular prolate ellipsoid x 2 + y2 = r2 _ C2Z2 c2 <1
circular oblate ellipsoid x 2 + y2 = r2 _ C2Z2 c2 >1
circulaT hyperboloid of one sheet x 2 +y2=r2+c 2z 2
circular hyperboloid of two sheets x 2 +y2=c2z 2 _r2
(4.37)
Since the above bodies of revolution are parameterized, calculation of their in-
terference becomes a little easier.
4.4 Intersection of Two Quadrics 53
4.4 Intersection of Two Quadrics

The intersection of two planes can easily be obtained, because both of their equa-
tions are linear. If the value of one of three variables (x, y, z) of the intersection
line is given, the other two are determined by its linear functions . When one of
the intersecting surfa,ces is a quadric , two variables of the intersection curve are
determined as root.s of a quadratic equation of the other one variable. When both
of the surfaces are quadrics, generally an equation to determine their intersection
curve becomes quaTt.ic. So, it cannot be solved easily. However, as we explained
in the previous section some of the quadrics have generators which are expressed
by two linear equat.ions, that is , a straight line, so if one of the surfaces belongs
to such a class, solution of their int.ersection curve is obtained from the roots of
the quadratic equations.
Even when bot.h of the intersecting quadrics lack generators, we can produce a
quadric with generators, which intersects with the same intersection curve as that
of the original quadrics [4][5]. The method of producing such a quadric is similaT
to that of producing a pencil of conics from two intersecting conics explained in
Sect . 4.2.4. In this case, the pencil must be a non-central quadric, excluding an
elliptical paTaboloid which does not have generators. For this purpose the method
requires numerical manipulations of the equations SUGh as affine transformations
together with solving a characteristic equation of a, matrix of degree four. After
making generators of a desired pencil, intersecting points of a modified quadric
with the generators have to be calculated and then transformed to the original
coordinate system. These operations are apt to introduce numerical errors, so
this method of ca.lcula,ting intersect.ion curves [7] is not always the best one .
Alternatively, the following met.hod is easier in the calculation [8]. If we cut
the intersecting quadrics by a plane pa,rallel to one of the coordinate planes,
we get two intersecting conics, whose intersecting points can be obtained by
the pencil method. The intersecting points of a conic and one or two lines are
point.s on the intersection curve in 3D. By moving the cutting plane in its normal
direction, we can trace the intersection points with the cutting planes.
Let QI and Q2 be two intersecting quadrics. Let the equations of two conics
generated by the cut of QI and Q2 wit.h a plane z = Zo be C 1 and C 2 , which are
obtained from eq. (4.16).
h•
b• i = 1,2, (4.38)
Jf(zo)
where gH zo), !f(zo) and d:(zo) are functions of zoo Let the characteristic equa-
tion of their pencil be D(fJ, zo) = 0, which is a cubic function of fJ and has at
least one real root Po. For the root, the pencil become a straight line, whose
int.ersection with one of the conics gives intersection points of the two conics.
For z = Zo + 6z, we obtain fJ = Po'+ 6p by using the Newton-Raphson met hod
Figure 4.2 Interference of two ellipsoids, Figure 4.3 Interference of three ellip-
intersection curves are shown soids, intersection curves are also shown
/\
{
( a) (b)
Figure 4.4 Fillet generation by rolling ball method: (a) interference of two cylinders, (b)
interference of a cylinder and a cone, ball-contact curves are also shown
starting from Zo and !-lo. For a new !-l, we obtain new intersecting points. Ex-
amples of intersection of two ellipsoids obtained by this method are shown in
Fig. 4.2.
This method is ext.ended easily t.o a. case of int.ersection of multiple quadrics.
Figure 4.3 shows an example of t.he intersection of three ellipsoids. When the
intersecting quadrics are circula.r cones or cylinders, their offset surfaces are also
the same type. Since the intersection curves of their offset surfaces are easily
obtained, they are used as loci of the center of a rolling ball to make fillet surfaces
along the intersecting corners. Figure 4.4 shows these examples.
Another method of intersection tracing of two surfaces represented by two
implicit equations is to use the numerical methods of solving differential equa-
tions. Details of this method are explained in Sect. 16.3. It is efficient and robust
and critical cases can also be t.reated.
5 Theory of Curves
5.1 Introduction
Curves are not only the fundamental geometric elements, but also they give
us various information on shapes. In CAD/CAM, free-form curves have to be
mathematically defined or extracted so that their properties can be controlled
and evaluat.ed by the computer. Accordingly, knowledge of the fundamentals
of the differential geometry of curves is required to apply geometry to practical
problems and it also gives the concepts and methods needed to understand the
theory of surfaces which is explained in the next two chapters [1]. The curves are
usually expressed in parametric forms, and arc length of the curve is used for
the parameter in theoretical treatments because of its simplicity of expression,
but for practical uses the parameter is cha.nged from arc length s to a more
manageable variable parameter t which monotonically increases with arc length.
When treating interference problems of curves or surfaces, the curves expressed in
implicit forms are sometimes convenient in numerical calculations or theoretical
analysis. Therefore, we introduce a method of converting from a parametric form
to an implicit one in the last section.
5.2 Tangent and Curvature of Curve

\Nhen a position vector r is a function of one variable t, r represents a curve in
space:
r(t) = {x(t) , y(t), z(t)}. (5.1)
The independent variable is any parameter which is convenient to calculate, but
when the geomet.ric properties of a curve are to be described concisely, an arc
lengt.h s measured along the curve is used as the parameter. So, initially we use
s and express the curve as r(s), and afterwards convert it to r(t) for practical
uses.
We introduce a unit tangent vector t, a unit normal vector n and a curvature
K, to describe the behavior of a curve r( s) in the vicinity of a point on the curve.
A unit tangent vector t of a curve is a unit vector from r(s) to r(s + ds), which
is given by
t = ~ = lim r( s + ds) - r( s) = r' ~ (5.2)
Idrl d.-+olr(s+ds)-r(s)1 Idrl
Here, a prime mark indicates differentiation with respect to the arc s. When ds
becomes infinitesimal, Idrl is equal to ds ,
Idrl = ds, (5.3)
56 5 Theory of Curves
n t
t + dt
~dt
~
t + dt · Figure 5.1 Tangent vectors and their
difference and normal vector
and we obtain from eq. (5.2)

t = rIo (5.4)
A unit tangent vector of a curve is equal to the first derivative with respect to S.
Let n be the unit vector orthogonal to t in the plane determined by t and
t + dt and let its direction be 90 degrees anti-clockwise to that of t.
The curvature K, at a point on a curve is the ratio of the anti-clockwise
rotation angle of the unit tangent vector t to the infinitesimal increment ds of
arc length. Let t and t + dt be the unit tangent vectors on the curve separated
by ds, and d¢ an angle between these vectors, then the curvature K, is defined by
d¢
K, = ds' (5.5)
As 'shown in Fig. 5.1, lo¢1 ~ (Iotl)/Itl, and It I = 1 by definition, so the relation
lo¢1 ~ loti holds. It follows that we get ot ~ no¢, and from eq. (5.5) we obtain
the relat.ion between the unit tangent and the unit normal vectors:
dt
ds = K,n. (5.6)
The curvat.ure K, is positive if dt and n are in the same direction. When a curve
proceeds anti-clockwise, its curvature is positive. From equations (5.4) and (5.6),
we obtain
r" = K,n. (5.7)
The second derivat.ive of a curve by an arc s is equal to the normal vector
multiplied by the value of its curvature.
The plane determined by t and t + dt is considered to be the plane deter-
mined by three points r( s), r( s + ds) and r( s + 2ds) in their limit and is called
the osculating plane at the point on the curve. The normal vector n is in the
osculating plane and orthogonal to t in the anti-clockwise direction. The vector
n is called the principal normal vector. Let the radius of a circle passing through
these three points determining the osculating plane be p. Since there is a relation
pd¢ = ds, (5.8)
5.3 Binormal and Torsion of Curve 57
n+dn
t + dt
b+db Figure 5.2 Binormal and torsion
we obtain from equations (5.5) and (5.8)
1
p =- . (5.9)
K
The radius p is called the radius of curvature, and the center of the circle is called
the center of curvature, whose location is
rc = r + pn.
5.3 Binormal and Torsion of Curve

There is a unit vector b orthogonal to t and n, called the binormal vector,
which is the unit normal vector of the osculating plane and has the following
relation from equations (5.4), (5.7) and (5.9):
b = t x n = p(r' x r"). (5.10)
By differentiating the relations b . b = 1 and t .b = 0 with respect to s, we obtain

b . b' = 0, t'. b + t . b' = O.
From eq. (5.6) and from eq. (5.10), (t' . b) becomes K[n, t, n], which equals zero,
and from the above relation , we get (t . b/) = O. This means b' is orthogonal to
t as well as to b, so it has a component of n only. We can express it using the
proportional constant r as
db
- = -rn. (5.11)
ds
This is interpreted as follows: when advancing along the curve, the rate of ro-
tation of the osculating plane is r and the rotation is around the tangent in the
right screw direction if r is positive. See Fig. 5.2. The parameter r is called the
torsion.
Different.iating n = b x t, which is obtained froin eq. (5.10), we get
dn db dt
- = - xt+bx-.
ds ds ds
Substituting equat.ions (5.6) and (5.ll) into the above, we obtain
dn
ds = -Tn x t + II:b x 11 = Tb - II:t. (5.12)
Equations (5.6), (5.ll) and (5.12) are called Serret-Frenet's formulas. They are
written collectively:
t' = II:n }
n' = rb - II:t . (5.13)
b' = -rn
Like the expression of curvature in eq. (5.7), we make an expression for torsion.
Differentiat.ing r( s) repeatedly, we obt.ain
r' = t, r" = II:n, rill = lI:'n + lI:(rb - II:t). (5.14)
The following triple product
[r', r", rill] = (r' x r") . r'"
is transformed to
From t.his we have

1
T = (2)[1", r", rill]. (5.15)
II:
Next, we determine an approximate sha.pe of a curve near a point on it. By
expanding res), with reO) = 0 as the origin, we get using equations (5.14),
To express the shape of a curve near the origin with t, nand b as the coordinate
axes, we take t.he first term of each coefficient of t, nand b in eq. (5.16) and
replace t, nand b by x, y and z, then we have
(5.17)
Eliminating s in eq. (5.17), we obtain
1 2 3 9 II: 2 1 3
Y= "2I1:X , Y ="2 r2 Z , Z = 611:rx . (5.18)
5.4 Expressions with Parameter t 59
xy plane
yz plane
zx plane
Figure 5.3 Shape of a curve near its origin. Its projections Oil three coordinate planes are
shown
Referring to Fig. 5.3, from the above equations, we can understand the local
shape of a curve. The projection of the curve on the osculating plane( the xy
plane) is a symmetrical second-degree curve with a tangent to the x axis at
the origin. The projection on the normal plane (the yz plane) is symmetrical
with respect to the z axis and has a cusp at the origin. The projection on the
tangential plane (the zx plane) is anti-symmetrical with respect to the origin
and has a tangent to the x axis. If the curvature does not take an extremum
value there, the shape of the curve in the 11 direction distorts anti-symmetrically,
because of the term ""s3/6 in eq. (5.16).
5.4 Expressions with Parameter t
So far, we have treat.ed r as a function of t.he arc s, but in practice it is difficult to

describe curves as functions of s, because we do not know the function of a curve
before designing a shape which determines the independent va.riable s. Accord-
ingly, instead of s, for convenience, we use a parameter t which monotonically
increases with s.
With the use of parameter t, calculation of position on the curve is not
difficult, but derivat.ion of the geometric quantities of the curve is not as simple
as doing so in terms of s. So, in t.he following, the derivation of r',r" and 1""
as functions of t.he parameter t is described first. For brevity, we denote dr/dt,
d2 r/dt 2 and d3 r/dt 3 by r, f and r(3). Changing the independent va.riable from s
to t , we have
. dr ds ds
r=-·-=t·- (5.19)
ds dt dt'
Hence, a vector r has the same direction as t. By squa.ring both sides of the
above equation, we get

(5.20)
so we have
ds r:--:-r = 1'r1 ,
dt = vr· (5.21)
a.nd the unit tangent vector is

i'
t = li-f' (5.22)
Differentiating eq. (5.19) by t, we obtain
d2 r ds 2 dr ~s ds 2 ~S
i; = ds 2 . ( dt) + ds . dt2 = 11:11 • ( dt) + t· dt2' (5.23)
Multiplying 11 with both sides of the above equation, we have an expression for
the curvature by using eq. (5.21),
(11 • i:)
II: = -'-2-' (5.24)
r
And differentiating eq. (5.21) and using eq. (5.22) we have
(5.25)
From eq. (5.23) and eq. (5.25), the vector 11:11 becomes
(5.26)
Next, the torsion is treated. By differentiating eq. (5.23) by t,
3
. -~s + t . -d s
3
-d r3 = 11' (d)
-
'2)
(11:1' + 11:11 ,(dS)3
• - + t ,(ds)
. -
2
dt dt dt dt dt dt 3 '
then eliminat.ing t' and 11' by using Serret-Frenet's formulas (5.13), and multi-
plying b with its result, we obtain
From this, the torsion is given by
(5.27)
For a curve of degree two, there is ·no torsion because d3 r / dt 3 vanishes.

5.5 Curvature of Space Curve and Its Projection 61
Projected curve
Figure 5.4 Space curve determined

x by its two orthogonal projections
5.5 Curvature of Space Curve and Its Projection
This section is an example of the application of the previous sections. When

measurement data of a space curve in a real three-dimensional model are not
a.vailable, designers have to define a space curve they intended to make with
drawings of its projection curves. See Fig. 5.4. Consequently, sometimes it is
required from these orthogonally projected curves to construct a space curve
expression for use in the computer. The expression must have the same geometric
propert.ies as t.hose t.he designers expressed on the drawings .
. For t.he projection curves, we assume their expressions have already been
det.ermined independently. They are denoted by r xy and r yz . And their respec-
tive curvatures at t.he ends K,xy and K,yz are assumed known. From the tangent
directions and the curvatures of t.he corresponding end points of the projected
curves, we are to obtain the tangent vector t, the normal and binormal vectors
nand b and the curvature K, of the space curve segment.
Since a space curve near the origin is given by eq. (5.16), its x, y, z components
are, higher degrees than S2 being neglected,
x = stx + (1/2)K,s 2nx , y = sty + (1/2)K,s2ny,

z = stz + (1/2)K,s nZ)
2 (5 .28)
where t = (tx, ty, tz), n = (n x , ny, nJ.
The component.s of t are obtained from the tangent directions of the projected
curves, because of the relation (dx : dy : dz) = (tx : ty : t z ) at t.he end point.
The curvatures of the curves on the projected (x, y) and (y, z) planes are
expressed by
(5.29)
Substituting the equations in (5.28) into the above equations , we obtain
(5.30)
By setting
A = (ex + e)3/2K,
y xy, B = (ey + e)3/2K,
z yz> (5.31)
eq. (5.30) becomes
(5.32)
T~. king the ratio of t.he two equations of (5 .32), we obtain
(5.33)
Since t and n are orthogonal, there is a. relation:
(5 .34)
From (5.33) and (5.34) the component.s of n can be written in the forms:
(5.35)
The constant C is determined from the normalizing condition: n; + n; + n; = 1,

as
(5.36)
We can det.ermine n by equations (5.35) , (5.31) and (5.36) . The curvature K, is

given from equations (5.32) and (5.35) by
(5.37)
The direction of t.he tangent vector t of the space curve is easily determined
from the tangent vectors of the projected curves, and the normal vector n is
calculat.ed from equations (5.31) , (5.36) and (5.35) . Accordingly, the binormal
vector b at t.he stating point is obt.ained from n x t . The curvature of the space
curve is given by eq. (5.37) and the shape of the curve near its starting point in
space is given by eq. (5.28.). Since these values are determined at both ends of
the space curve segment, it can be expressed by a Bezier curve of degree five or
higher.
5.6 Implicit Expression of a Parametric Curve

In paTametric expressions of plane curves r1 (t) and 1'2 (t), there is no relation
between parameters t though the same symbol is used. When considering the
intersection of the two curves, the two t's have to be changed to t1 and t2 to
avoid confusion. The intersection points aTe determined from the solution of the
following vector equation
Generally this represents a set of simultaneous non-linear equations, from which

t1 and t2 are obtained as values of the intersection points. Except for straight
line segments, their solution is hard to obtain.
In a case involving plane curves, if either of their expressions can be converted
into an implicit form f(x, y) = 0, then we substitute the expressions for x(t)
and y(t) of the ot,her curve into f(x, y) = 0 to obtain an algebraic equation
f{x(t), y(t)} = 0 in a single variable t. So long as its degree in t is not very high,
we can use well established solving methods for algebraic equations.
Then the problem is implicitization of a paTametric expression. We present
a method of conversion [5][6]. From the x and y components of r(t) we define
two polynomial equations X(t) = 0 and Y(t) = 0:
X(t) = ant n + an_1tn-1 + ... + ao = 0, (5.38)

Y(t) = bnt n + bn_1t n- 1 + ... + bo = O. (5.39)
When 1'( t) is expressed by a polynomial of t, the variable x is contained in ao

and the variable y in bo only. When r( t) is a rational expression which is treated
in Chap. 12, x is contained in all ai and y in all bi, but a coefficient of x in each
ai and that of y in each bi are the same.
If a value of t satisfies both equations (5.38) and (5.39), a pair (x, y) is on
the curve r(t). This value of t is a common root of equations (5.38) and (5.39).
Even if we make an equations of degree n - 1 from them by elimination of tn, the
result. contains the common roots. There are n ways of elimination, as shown
below.
An operation bn x X(t) - an x Y(t) gives
CO,n_1tn-1 + CO,n_2 tn - 2 + ... + co,o = 0, (.5.40)
where the coefficients CO,n-1, CO,n-2,"', co,o of powers of t contain only linear
functions of x, y and canst and are easily determined.
By paTenthesizing the first two terms of eq. (5.38) and eg. (5.39) with common
t n - 1 outside we change them into
X(t) = (ant + an_1)t n- 1 + ... + ao = 0, (5.41)

Y(t) (bnt -f bn_1W- 1 + ... + bo = O. (5.42)
An operation {(bni + bn- l ) x (S.41) - (ani + an-d x (S.42)) eliminates the first
terms and gives
Cl,n-l t -
n l
+ Cl,n-2 tn-2 + ... + Cl,O = 0, (S .43)
where Ci,n-i are constants containing only linear functions of x, y and const.
Generally, by parenthesizing the first i + 1 terms of eq. (S.38) and eq. (S.39), we
get
X(t) ( an t i+ an-l ti-l + ... + an-i )t n - i + ... + ao = 0 , (S.44)
Y(t) (b nt i + bn-l t i - l + ... + bn-i )t n .- i + ... + b0= 0 . (S.4S)
By a similar elimination process of coefficients for t n - i we obtain an equat.ion of
degree n - 1 of the form:
(S.46)
Cont.inuing t.hese parenthesizing and eliminating processes, finally we obtain n
equations of degree n - 1 in t, which have a common root:
CO,n-l CO,n-2 CO,O
ttn-l
n- 2 1
r Cl,n-l Cl,n-2 Cl,O
x r . = o. (S.47)
Cn-;,n-l Cn-l,O to
We int.erpret these n equations as n homogeneous linear equations for n variables
(tn-I, t n - 2 , " ' , t, 1). And if these linear simultaneous equat.ions have solutions
other than zero, t.he determinant of the coefficient matrix must be zero .
. An element of this determinant is a linear function with respect to x and y.
The development of t.his determinant gives an algebraic equation of degree n in
x and y, which is the implicit form of 1'(t).
Example. Implicitization of a rational Bezier expression of degree two, which is
treated in Sect. 12.6.
Let control point.s be Po = (xo, Yo), PI = (x}, YI), PZ = (X2, Y2), and the
parameter t be 0 ~ t ~ 1.
A rational Bezier curve is given by
(1 - t)2pO + 2t(t - l)WPI + 3t2P2
l' ()
t = ~~~~~~~~~~----~
(1-tP+2t(1-t)w+t2 .
(S.48)
For simplicity, the parameter t is transformed to '11 by a relation t = '11/(1 + '(1),

and eq. (S.48) becomes
1'(t) = Po + '~'11WPI
) + '112 P2.
( S.49)
1 + 2'11w + '112
Since l' is (x, y), the above equation is transformed to
X('11) '11 2(x - X2) + 2'11w(x - xd + (x - xo) = 0,

Y(u) '11 2(y - Y2) + 2'11w(y - YI) + (y - Yo) = o.
Applying the elimination met.hod described above, we obtain the following two
equations of u:
(y - Y2)X(U) - (x - X2)Y(U)
= 2w{(x - xr)(y - Y2) - (x - X2)(Y - yd}u +
{(x - xo)(Y - Y2) - (x - X2)(Y - Yo)}
0,
{(y - Y2)U + 2w(y - YI)}X(U) - {(x - X2)U + 2w(x - XI)}Y(U)

{(x - xo)(Y - Y2) - (x - X2)(Y - Yo)}u
+2w{(x - xo)(Y - YI) - (x - XI)(Y - Yo)}
= o.
Eliminating u from the above two equations, we obtain
4w 2{(YI - Y2)X + (X2 - XI)Y + XIY2 - x2yIl

{(Yo - ydx + (Xl - xo)y + XOYI - XIYO}
-{(Yo - Y2)X + (X2 - xo)y + XOY2 - X2YO}2 = 0.
Now, for simplicity, we put Yo = Y2 = 0, YI = h, -X2 = Xo = a, Xl = b, and the

above equation becomes
This is a conic equation, which is t.reated in Chap. 5.

6 Basic Theory of Surfaces
6.1 Introduction
Quadric surfaces, especially bodies of revolution such as spheres, cylinders, cones,

or also ruled surfaces have been widely used in engineering products. This is be-
cause their geometric properties are easily specified by their designers and the
shapes are easily manufactured by machines. On the other hand, free-form sur-
faces have only been used in special cases and their design and manufacture
have required speci-al talent and skill. But in recent years their uses have in-
creased because of strong demands for high performance and aesthetic quality
in engineering products. And computers have been used for their design and
manufacture . Accordingly, various mathematical expressions of free-forms for
engineering use have been developed.
With these advanced requirements, appropriate mathematical expressions
for free-form surfaces and methods of design and evaluation for their shapes
have been studied. Accordingly, knowledge of the fundamental geometric theory
of surfaces is indispensable for engineers in this field. Since the theory of the
differential geometry of surfaces is not new, but has been treated from a mathe-
matical point of view and not from engineering one, practical engineers have not
wanted to approach it. In textbooks on differential geometry, tensor notation
is widely used, because it is concise in analysis and description of geometry in
n-dimensional space. But since uses of the surfaces in CAD is limited to three
dimensions (mathematically two dimensions), adoption of the tensor notation
has little merit for practical engineers unless it is used advantageously in other
domains of their tasks. Moreover, since we emphasize geometrical intuition and
understanding rather than elegance of the mathematical formalism, we use the
conventional notations in this book.
First, we introduce the basic framework for description of the geometric
characteristics of the surfaces, then explain the features which determine the
local shape of surfaces such as the tangential plane and the principal curvatures
and the lines of curvature.
Since the behavior of normal vectors of the surface gives important infor-
mation for analysis and understanding of the shape features of the surface, we
explain the formulas for the differential of a unit normal vector, and the local
shape around a point on the surface. In this chapter we treat the fundamental
parts of the theory of surfaces. Its advance topics, such as various curves which
characterize the surfaces and the offset surfaces, are left to the next chapter.
68 6 Basic Theory of Surfaces
Figure 6.1 Surface patch with u-v

parameter net
6.2 The Basic Vectors and the Fundamental Magnitudes

A surface segment or a surface patch is denoted in a parametric form by
s(u,v) = {x(u,v),y(u,v),z(u,v)} 0 s u,v S 1. (6.1 )
If there is any functional relation between u and v, s( u, v) represents a curve

on the surface. Values of s( u, v) of u = const make a family of curves on the
surface and those of s( u, v) of v = const generate another family of curves. They
constitute a net of curves on the surface. Usually, any point on the surface is
determined by the intersection of two curves, each of which belongs to one of the
fa.inilies. See Fig.6.1. A partial derivative os/ou represents a tangent vector
of the curves v = canst and os/ov is a tangent vector of the curves u = canst.
They are denoted by Su and Sv for brevity and are called the basic vectors of the
surface.
They span the tangent plane of the surface. The vector ds which connects
two points on the surface from s(u, v) to s(u + du, v + dv) is given by
ds = sudu + svdv. (6.2)
The square of the absolute value of ds is
(ds? = ds· ds = S~(dU)2 + 2s u ' svdudv + S~(dV)2. (6.3)
From the basic vectors of the surface, the following quantities are defined.
E = s~, F = Su • sv, G = s~. (6.4)
They are called the fundamental magnitudes of the first order. Then eq. (6.3)
becomes
ds 2 = E(du? + 2Fdudv + G(dvJ2. (6.5)
6.3 Normal Section and Normal Curvature 69
Let an angle between the basic vectors s,. and Sv be w, then their inner
product F and the absolute value H of vector product of the basic vectors can
be expressed by the fundamental magnitudes of the first order:
F Is,.l· Isvl cos W = (V EG) cos w, (6.6)

H Is,. x svl = Is,.I·lsvl sinw
= JEG( 1 - cos2 w) = VEG - F2. (6.7)
Since usually the angle w is not equal to 90 degrees, the parameter net on the sur-
face constitutes an oblique coordinate system with its basic vectors. In analysis
and description in the oblique coordinate systems, the tensor notation is conve-
nient, but in this case, as the oblique coordinate system is only two-dimensional,
its merit is lost to those who are not familiar with the tensor notation and its
operations.
The third coordinate axis of the local coordinate system is taken in the
direction orthogonal to s,. and sv, that is, along the normal vector of the surface.
The unit normal vector of a point on the surface is given by
(6.8)
6.3 Normal Section and Normal Curvature
A pencil of tangent vectors at P on the surface is in the tangent plane there.

One 'of the unit tangent vector t is expressed from eq. (6.2) as
(6.9)
where ds = Ids!, and a plane determined by t and 11 is called a normal plane.

The intersection of the surface by a normal plane is called a normal section. It
is a plane curve. See Fig. 6.2.
The curvature K at P of the normal section is called the normal curvature
at P, which is obtained as follows. Differentiating t with respect to the arc s of
the normal section, we obtain
dt
ds
(6.10)
This is equal to K11 from eq. (5.6) of Chap. 5, when we consider the sign of cur-
vature. Multiplying the normal vector 11 with the above equation (6.10), we can
eliminate the first and the second terms of the right-hand side, because they are
Figure 6.2 Normal section and

tangential plane
in the tangential plane. Introducing the following symbols called the fundamental
magnitudes of the second order,
(6.11)
we obtain the relation:
(6.12)
Using eq. (6.5), eq. (6.12) becomes
L(du? + 2Mdudv + N(dv)2

(6.13)
E(du)2 + 2Fdudv + G{dv)2 .
~- ~~~----------~~~
-
Values of the fundamental magnitudes of the first and the second order
E, G, F, L, M and N are fixed on a point of the surface and do not depend
on the direction of the normal plane through the point P. But the direction
depends on the ratio du : dv. So the normal curvature ~ has different values in
different directions of the normal section.
When a cutting plane passing through a point on the surface is not orthogonal
to the tangential plane at the point, that is, the section is not a normal section,
the curvature ~c of this section at the point is given by
~
~c = cos e'
(6.14)
e
where ~ is the curvature of the normal section and is the angle between the
normal pla.ne and the oblique cutting plane. See Fig. 6.3. This relation is called
Meunier's theorem.
6.4 Principal Curvatures 71
Figure 6.3 Non-normal section and

Meunier's theorem
Its proof is given as follows. Let n l be the unit normal of the cutting section
at the point, t be the unit tangent there and "'c be the curvature of the section.
There is a relation "'cnl = dt/ds, whose right-hand side is given by eq. (6.10).
Multiplying bot.h sides of the equation by n, we obtain ("'en· n/) = (n· dt/ds),
which is "'ccose = "'. So we have (6.14).
Exercise. Prove t.hat. the curvature "'e of an int.ersection curve of two surfaces 81
and 82 is given by
"'e = V("'I + "'~ - 2"'1"'2 cos ¢)(sin- I ¢),

where "'I and "'2 are curvatures of the normal sections of 81 and 82 , which include
t.he t.angent of the intersection curve, and ¢ is the angle between the normals of
81 and 82 at the point.
6.4 Principal Curvatures

Next, we consider t.he extremum values of the normal curvatures. When the
normal plane rotates around the normal, the shape of the normal section changes
with its tangent. direction, and so does the normal curvat.ure. The shape returns
to the initial state after half a rotation of the normal plane. Now, setting
dv
'Y= - (6.15)
du
and rewriting'" as ",b), a funct.ion of 'Y , we get the following equation from
eq. (6.13),
{L - ",b), E} + 2{M - ",b), Ph + {N - "'b)' Gh 2 = O. (6.16)

This equation gives t.he relation between", and 'Y. With the rotation of the normal
plane, "'b) takes t.he extremum values, which are determined by d",b)/d'Y = O.
Differentiating eq. (6.16) with respect to 'Y together with the above extremum
condition, and t.hen writing K, and 'Y as 'K; and 1', we obtain an equation:
(M - 'K;F) + (N - 'K;Gh = 0, (6.17)
and substituting this into eq. (6.15), we get another equation:
(L - 'K;E) + (M - 'K;h = o. (6.18)
Then from the above equations, we obtain the relations:
-(M - 'K;F) -(L - 'K;E)

(6.19)
l' = (N - 'K;G) = (M - 'K;F) ,
(M + 1'N) (L + 1'M)
( 6.20)
(F + 1'G) (E + 1'F)'
From eq. (6.19) we have the equation for 'K;,
(EG - F2)'K;2 - (EN + LG - 2MF)'K; + LM - M2 = O. (6 .21)
The coefficient of 'K;2 of the above equation is positive because of eq. (6.7), and
this equation is the second degree for 'K;. Let its roots be K,l and K,2, which are
called the principal curvatures of surface, then we get the following expressions
from the relation between roots a.nd coefficients of a quadratic equation,
!( ) _ 1 (EN + LG - 2M F)
2 K,l + K,2 - 2 (EG _ F2) , (6.22)
(LN - M2)
Kg = K,1K,2 = (EG _ F2) . ( 6.23)
The expression ](m is called the mean curvature and the expression Kg is the
Gaussian curvature of the surface. The local shape of a surface and the sign of
its Gaussian curvature have the following relation:
Kg > O. The normal curvature has the same sign in all directions, so the tangent
plane touches the surface at one point. The usual convex or concave surfaces
correspond to this.
Kg < O. The normal curvature becomes zero twice during the half rotation of
the normal plane around the normal. The tangent plane intersects with the
surface in t.hese directions of zero curvature.
Since the denominator EG - F2 of Km and Kg is positive from eq. (6.7), the
sign of the Gaussian curvature is determined from the sign of the numerator
LN - M2. So, when L or N or both are zero, or Land N have different
sign, the Gaussian curvature is surely negative. A surface patch s( u, v) of
degree one, the inner ring surface of a torus and a hyperboloid of one sheet
are examples.
6.5 Principal Directions and Lines of Curvature 73
Kg = o. When the Gaussian curvature is zero everywhere, the surface is devel-

opable, because at least one of the principal curvature is zero, hence the line
of curvature on the surface is straight (refer to the next section). Such sur-
faces include general cones, cylinders and the surface swept out by a tangent
to any twisted space curve.
From equations (6.22) and (6.23), we have
Kl = Km + J(I{;" - Kg), ( 6.24)
for the two principal curvatures.
6.5 Principal Directions and Lines of Curvature

Eliminating K, from eq. (6.20), we obtain the equation of:Y (hereafter we omit the
tilde mark for simpiicity):
(MG - N F):y2 + (GL - N E):Y + F L - ME = 0,
or (MG - N F)dv 2 + (GL - N E)dudv + (F L - M E)du2 = O. (6.25)
This is the equation of lines of curvature which is explained in this section. Since
the equation is of degree two, it has two roots '1'1 and "12 and has the relation:
-(GL-NE) (FL - ME)
"11 + "12 = (MG - N F) , "11"12 = (MG - N F)·
(6.26)
At ~. point on the surface, in the directions determined by "11 and "12' the nor-
mal curvatures take extremum values. Since a tangent vector on the surface is
(sudu + svdv) , the inner product of the two tangent vectors corresponding to
"11 and "12 is given by
(6.27)
With equations (6.4) and (6.26), we convert the inside of { } of the above ex-
pression to
{E(MG - N F) - F(GL - N E) + G(F L - M E)}/(MG - N F),
and this is equal to zero.
It follows that the two tangent directions of the normal sections of the princi-
pal curvature are orthogonal. These directions are called the principal directions.
Except for singular points on the surface, any point has principal directions. If
the tangents of curves on a surface coincide with the principal directions there,
the curves are called the lines of curvature. Examples are shown in Fig. 7.1 and
Fig. 7.5.
The lines of curvature make an orthogonal net. The pattern of lines of
curvature of a surface is an inherent attribute independent of the coordinate
systems used. They are discussed in more detail in the next chapter.
6.6 Derivatives of a Unit Normal and Rodrigues'

Formula
For designing and manufacturing surfaces in engineering products or for evaluat-
ing the surfaces from technical and aesthetic viewpoints , analysis of the behavior
of their normal vector becomes important. The surface normal vector n is a
function of the parameters u and v , and since its derivative vectors n" and nv
.are orthogonal to n, that is, in the tangential plane at that point, they can be
expressed by the basic vectors of the surface.
(6.28)
The constants 0',(3, a' and (3' are determined in the following way. Multiplying
s" and Sv to the first equation of (6.28) and considering L = n . s,," = -n" . s"
and M = n· s"v =. -nv . s,,' which are derived from differentiation of the relation
n . s" = 0 wit.h respect to u or v, we obtain the equations:
-L = O'E + (3F, -M = O'F + (3G.

By solving these simultaneous equations for 0' and (3, we obtain:
(MF - LG) (3 = (LF ~ ME) (6.29)
0' = H2 ' H2 '
where H2 = EG - F2 . In the same way, using the relation n . Sv = 0 we obtain
from the second equation of (6.28),
a' = (NF - MG) ,_ (MF - NE)

(3 - H2 . (6.30)
H2
The equations (6.28) with (6.29) and (6.30) are called Weingarten's equations,
which express the derivatives of the normal n in terms of the basic vectors.
We can deduce the following relat.ions using equations (6.28)-(6.30) and the
expressions of mean curvature Km and Gaussian curvature Kg:
n~ 2KmL - KgE, (6.31)

11" . nv 2KmM - KgF, (6.32)
11" X nv = K g H11. (6.33)
Multiplying n by both sides of eq. (6.33), we obtain another expression of the

Gaussian curvature
y [n, n", nv]
19 = H (6 .34)
When the differential dn of the normal vector is given at a point on a surface
s( v, v), we are required to determine the location s( u + du, v + dv) which has the
specified normal vector 11 + dn. Since there is a relation:
(6.35)
6.6 Derivatives of a Unit Normal and Rodrigues' Formula 75
we can determine du and dv by multiplying n x nv or n x nu and'using eq. (6.34):
du = dv = [n, n u, dnl. (6.36)

HKg
Accordingly, if an absolute values of the Gaussian curvature is small, values of

du and dv become large, which may violate the assumption of the Weingarten's
equation.
Next we treat the differential dn of a unit normal vector in the principal
direction in terms of a principal curvature i';,. The relation is given by Rodrigues'
formula:
dn + i';,ds = o. (6.37)
This is proved as follows. Substituting the Weingarten's equations into eq. (6.35),
we obtain
(6.38)
where 'Y = dv/du. On the lines of curvature 'Y becomes ;y, which is given by
eq. (6.19). Using this and equations (6.29) and (6.30), the right-hand side of
eq. (6.38) is transformed to -i';,sudu - i';,svdv, which is equal to -i';,ds. Hence we
have eq. (6.37).
In a general space curve, Frenet-Serret's formula (5.12) gives the relation
dn/ds + Id = Th, (t = ds/ds),
where n is the principal normal of the curve, so dn and ds are not always parallel.
But along a line of curvature Rodrigues's formula (6.37) assures that dn, which
is variation of the unit normal of the surface, is parallel to ds: dn is in the same
or opposite direction to ds according to negative or positive principal curvature.
In the region of negative Gaussian curvature, there is two directions in which dn
IS zero.
If the ta.ngent t of a curve on a surface is parallel to dn, then the curve is

a line of curvature of the surface. This is the inverse of Rodrigues' formula a.nd
proved as follows: for an arbitrary constant A, if dn = Ads, then using eq. (6.38),
we get
For the above equation to hold everywhere, the insides of { } must be zero, that
is, the following relations must hold.
A = ex + ex' 'Y = (3 h + (3'.

Substituting equations (6.29), (6.30) into the above equation, we obtain equation
(6.25), which indicates the lines of curvat ure.
6.7 Local Shape of Surface

We can express an approximate shape of a surface around a point taken as the
origin by expanding s:
(6.39)
Multiplying the unit normal n with this equation and using the fundamental
magnitudes, we obtain an equation of sha.pe near the origin, z being the distance
from the tangential plane at the origin,
z = (1/2)(Ldu 2 + 2Mdudv + Ndv 2 ). (6.40)
We transform this into an orthogonal coordinate system, whose axes coincide

with the principal direction of the surface, since the unit vectors i and j of the
principal directions are given by
(6.41)
where
( 6.42)
and 11 and 12 are determined from eq. (6.25). Let it point (du, dv) on the tan-
gential plane expressed by the basic vectors s" and Sv be a point (x, y) expressed
by i and j. There is a relation:
s"du + svdv = xi + yj.
Multiplying n x s" or n x Sv with both sides of the above equation and considering
eq. (6.41), we obtain
du = x/A + y/B, ( 6.43)
Then eq. (6.40) is transformed to

1
z = 2(Ldu 2 + 2M dudv + N dv 2 )
1 x2
= 2{(L+ 2Ahl+ N 1?)(A2)
xy
+2(L + Mhl + 12) + N11 12)(AB)
y2
+ (L + 2M 12 + N 1~)( B2 n· (6.44)
From equations (6.26) we know the coefficient of xy in eq. (6.44) is zero. So,
eq. (6.44) takes the following form:
( 6.45)
6.7 Local Shape of Surface 77
"g <0
Figure 6.4 Section patterns of a surface by its offset tangential planes
where
, __ (L + 2M'Yi
. },',' + N'Y;) f or z. = 1, 2 . (6.46)
(E + 2F'Yi + G'Y;)
Considering t.hat. eq. (6.46) corresponds t.o eq. (6.13) and t.he "Ii's to t.he principal
direct.ions , we have Kl = 1\;1, K2 = 1\;2 and eq. (6.45) becomes
Z
1 l X 2 + 1\;2Y-") .
= 2(l\; (6.47)
Since t.he curvature of t.he normal section is given by eq. (6.13),
(Ldu 2 + 2M dudv + N dv 2 )
( 6.48)
- (Edu 2 + 2Fdudv + Gdv 2 ) ,
I\; - ~----------------~
and using eq. (6.4.3), t.his is also transformed to
( 6.49)
e
where is an angle between the x (Lxis and the tangent of the normal section at
the origin. Equation (6.49) is called Euler's formula .
Equation (6.47) represents the sha.pe of the surface in the vicinity of the
origin. The intersection of the surface with a plane z = c which is paTallel to the
xy plane has three kinds of form:
- Kg{ = 1\;11\;2} > 0 : t.he intersection is an ellipse for c > 0 or nothing for c < O.
Its long and short radii are V(2c/l\;d and V(2C/1\;2)'
- Kg < 0: The int.ersection curve is a. hyperbola, which intersects with the x
axis or the y axis according to the sign of c. If c equals zero, the intersection
is two straight lines through the origin, which become its asymptotes.
- Kg = 0: the intersection is two straight lines equidistant from one of the axes.
See Fig. 6.4; this diagram is called Dupin 's indicatrix. Equation (6.47) is used
also for approximate calculation of the intersection curve of surfaces.
7 Advanced Applications of Theory of
Surfaces
7.1 Introduction
Special topics on the geometric properties of surfaces are treated in this chapter.
They are developed from the fundamental theory of surfaces which has been
explained in the previous chapter. Knowledge of these topics is useful for treating
free-form surfaces in advanced problems. First we discuss the umbilics and lines
of curvature. On a free-form surface, there are points and regions which have
the special chara.ciers inherent to its shape, and whose locations do not depend
on the coordinate system adopted. The lines of curvature make orthogonal nets
on a surfa.ce, and the pattern they form exhibits inherent features of the surface.
The umbilics are singular points, or curves or regions seeing from the lines of
curvature. On the free-form surface the umbilics appear more frequently than
our expectation. There are other curves on the surface which depend not only
on its inherent features, but also on its orientation with respect to its observers
or its environments or its surfa.ce physical properties. These curves are useful for
describing or evaluating the objects from engineering and aesthetic criteria.
After treating these problems, we explain the application of the surface the-
ory to two special surfaces: an offset surface and a ruled surface. Determining
the offset surfaces of an original surface is important in engineering, but the solu-
tion in closed analytical form is very difficult to obtain. Since in determining the
intersection curves of a surface with an offset surface, we need the fundamental
magnitudes of the first and the second order of the latter and not necessarily its
whole shape, we can deduce them from the properties of the original surface. In
the last section, an equation of a ruled surface is introduced and its geometric
properties are treated.
7.2 Umbilics
An umbilic is a point at which the principal curvatures are equal. It follows
that all the normal curvatures there are equal: it represents an infinitesimally
small region of a spherical or planar surface. The condition of equal principal
curvatures is given from the condition of equal roots of eq. (6.21) of the previous
chapter, which determines the principal curvatures of a surface. From the zero
discriminant value of the quadratic equation, we have
(EN + GL - 2F M)2 - 4(LN - M 2 )(EG - F2) = O. (7.1)

80 7 Advanced Applications of Theory of Surfaces
This is transformed to
(EN - GL? -4(EM - FL)(FN - GM) = 0,
which is the same to the condition of equal roots of the equation of the lines
of curvature which is given by eq. (6.25) . This equation gives differential form
dv/du = -(GL - EN)/2(GM - FN), but the double roots mean the same di-
rections of tangents of the orthogonal curves. This is a contradiction. So the
roots must be indefinite. A relation
E:F:G=L:M:N (7.2)
is sufficient for the above condition.
A line element on a surface is given from eq. (6.5) by
ds 2 = Edu 2 + 2Fdudv + Gdv 2. (7.3)
If we set du : dv in the principal directions, using eq. (6.19), ds 2 is transformed
to
(EG - F2)(M - "'iF)("'l + "'2 -
2"'i), where i =1 or 2. (7.4)
At the umbilic this becomes zero. So we cannot trace the line of curvature there
by the differential equation solvers.
Examples of the umbilics which constitutes a curve are shown. If in a small
region of a surface, its shape is smooth and mirror-symmetric (reflection), the
intersection curve of the surface and the mirror plane is the line of curvature,
because the differential of the normal dn along the intersection is in the plane of
symmetry and the inverse of Rodrigues' formula assures us that the intersection
is the line of curvature. If another line of curvature of the surface intersects with
this intersection curve obliquely, the intersection point is an umbilic, because
there are more than two lines of curvature through the point. In this case all the
points of the symmetrical intersection curve are umbilics.
On a surface of revolution, when the curvature at a point on its meridian
curve is equal to that of the parallel of latitude through that point, all the points
on the parallel are umbilics. Surfaces on which all the points are umbilics are
a sphere or a. plane. On these surfaces any orthogonal net can give the lines of
curvature.
On free-form surfaces the umbilics appears frequently. For example, in the
patch shown in Fig. 7.1 there seem to be four umbilics, but we cannot recognize
them from its contour curves shown in Fig. 7.2.
If we take the origin at an umbilic of a surface and the z axis in the direction
of the normal, the shape of the surface around the umbilic is given approximately
by
z n's= S(du,dv) = (1/2)"'ds 2 +
+ (1/6)(Suuudu3 + 3Suuvdu 2 dv + 3Suvv dudv 2 + SvvvdV3), (7.5)
where", is a constant normal curvature.
7.2 Umbilics 81
Figure 7.1 Lines of curvature and

locations of umbilics of the patch
Contour curve
Figure 7.2 Contour curves and their

orthogonal curves
Examples of shapes given by eq. (7.5) are shown in Fig. 7.3 and Fig. 7.4 to-
gether with their patterns of the lines of curvature. In Fig. 7.3 (a) and (b), the
umbilics are in region of positive Gaussian curvature and the pattern (a) appears
in a region of elliptical shape. The pattern (b) appears also in a region of nearly
elliptical shape with disturbing third-degree terms. When we observe shapes of
the surfaces by their contour curve patterns, we cannot imagine that the patterns
of their lines of curvature are so different. Clearly we can see an invisible singu-
lar curve on which tracing the lines of curvature by a differential equation solver
fails. This is an umbilic of the curve type. In Fig. 7.4 an umbilic is an isolated
point of zero curvature surrounded by a region of negative Gaussian curvature.
A pattern like (c) appears also in 'a region of positive Gaussian curvature.
Contour curves Lines of curvature

(a)

(b)
Figure 7.3 (a), (b) Contour curves of two patches and their lines of curvature in a region of
positive Gaussian curvature
Figure 7.4 Contour curves of a patch and its lines of curvature in region of negative Gaussian
curvature
7.3 Characteristic Curves on a Surface 1
7.3.1 General Remarks
T4ere are curves on a surface which characterize its features. Some of them
are useful for understanding the shape of the surface or evaluating its surface
characteristics. We introduce their equations and examples in the following sub-
sections [18]. Generally, a curve on a surface s( u, v) is determined by a relation
between the parameters u and v. The relation is mostly given by a differential
equation of a form:
f(u, v)du + g(u, v)du = o. (7.6)
Otherwise, a relation F( u, v) = 0 may be given. If solving F( u, v) for successive

values of given u or v is not easy, it is converted into the form of eq. (7.6) from
its differential. To trace the curve, after solving the above equation numerically
with the initial condition, we calculate s( u, v) step by step for the obtained set
of u and v.
If the shape of a curve is simple to trace, we can integrate eq. (7.6) with u or
v as the independent variable. Otherwise, the above equation is transformed into
the following simultaneous differential equations with the independent variable
of arc length s.
du dv
ds = 1jJ(u, v)g(u, v), ds = -1jJ(u, v)f(u, v), (7.7)
where
1
1j;(u,v) =± .
J(Eg2 - 2Ffg + Gj2)
The last equation is obtained from eq. (7.3). Using appropriate numerical meth-
ods of solving the differential equations with given initial conditions, we can trace
the curve. As for differential equation solvers, refer to the Appendix.
The characterist.ic curves of the surfaces are divided into three categories:
curves inherent to the surface geomet.ry, curves which are dependent on their
positions and orientations in space and certain curves, which also depend features
of t.heir environment such as the incident light and the reflection properties of
their faces. Curves of the former category are the lines of curvatures, curves
of zero Gaussian or zero mean curvature and curves of the extremum principal
curvature. Those of the second category are the contour curves, the contour
orthogonal curves, .t he equi-gradient curves, the extremum gradient curves, the
extremum search curves and so on. Some of these curves represent features of
the surface.
For example, the extremum gradient curves are loci of the maximum or
minimum gradient point on each of the contour curves. These curves are the
analog of ridge or valley lines in a t.opographical map. The third category includes
silhouette patt.erns, equi-brightness curves, equi-highlight curves and reflection
patt.erns on surfaces , which relate to the incident light, the view points, the
reflection characteristics of the surface, and the environmental as well as the
geometric conditions of the surfaces.
The points on the surface which are determined by the conditions f (u, v) = 0,
°
g( u; v) = of eq. (7.6) are singular points of the family of the curves described
by t.his different.ial equation. These points have certain special properties.
7.3.2 Lines of Curvature

The equation of lines of curvature is given by eq. (6.25):
(GM - FN)dv 2 + (GL - EN)dudv + (FL - EM)du 2 = 0. (7.8)
Since this equation gives an orthogonal net which is inherent to a surface, features
of the surfa.ce seems to be extracted from its pattern. But we have not yet
succeeded in their categorization. Equation (7.8) is resolved into a product of
two equations of type (7.6) , each of which corresponds to one family of lines of
curvature.
Since each point of a surface determines its principal directions, when trac-
ing a line of curvat.ure by numerical integration we must be careful not to be
deviated from the curve by numerical errors. If we introduce an arc length s as
the independent variable for the integration of the equation, the amount of cal-
culation increases but control of the steps of the integration procedure becomes
easier than the integration witholit it. In small regions around the umbilics,
Figure 7.5 Lines of curvature of

patches connected in C(J)
the calculation becomes unstable. Beyond that point the sign of ds has to be
reversed.
It should be noticed that the umbilics make the lines of curvature form special
patterns. Figure 7.1 is an example of the lines of curvature of a surface whose
contour curves are shown in Fig. 7.2. This is a Bezier patch of degree 3 x 3.
We can notice a few special features indicated by arrows in Fig.7.1. These are
local patterns of the lines of curvature showing the existence of umbilics, but in
Fig. 7.2 we cannot recognize any special features at the same locations. Figure
7.5 shows t.he pattern of the lines of curvature of four patches connected in e(l).
The pat.t.ern shows clearly the connecting boundary curves of the patches and
also t.he differences of their surface characters. In Fig. 7.6(a) and (b), examples
of variations of patterns of the contour curves and the lines of curvature for the
same displacement 6z = 6% of the control point P22 of the Bezier patch used in
Fig. 7.1, which is of degree 3 x 3. Varia.tion of the curve patterns is more sensitive
in t.he line of curvature t.han in the cont.our curve.
Uses of t.he lines of curvat.ure are currently rather limited, because of their
cumbersome calculation and lack of experience for interpreting their patterns.
Since the lines of curva.ture of an offset surface are easily determined from those
of t.he original surface, they can be used in the analysis of offset surfaces. This
is treated in Sect. 7.5.
7.3.3 Extremum Search Curves
We frequent.ly need t.o find locations of the maximum or the minimum height or
the saddle points or paths from the lowest to the highest points of a surface to
use these data for other purposes. Hence, we need to obtain this information as
simply as possible.
Let n f be t.he unit vector directing upwards and let a component of s( u, v)
Contour cur ves
Figure 7.6 Variation of pattern of contour curves and lines of curvature with a displacement
=
/:::"Z 6% of a control point P2,2
in that direction be ¢( u, v).
¢(u,v) = nj ·s(u,v). (7.9)
The extremum values of s( u, v) are obtained by solutions of the simultaneous

equations:
¢,,(u,v) =0, ¢v(u,v)=O, (7.10)
Sin'ce usually it is difficult to solve these equations directly, we trace the curves
represented by them individually by transforming them into the differential equa-
tions with appropriate initial conditions:
(7.11)
These curves are the loci of the extremum points on the curves of constant
paTameter u or v of s(u, v). The intersecting points of these curves indicate
the locations of the extremum points on the surface. Their initial conditions
are given on the boundary curves or on constant parameter curves inside the
boundaries.
These curves not only give the extremum points at their intersection, but
also roughly depict the shape of the surface together with its contour curves.
On these curves one can determine the starting points of the contour curves
which do not reach the boundary of the surface. Figure 7.7 shows examples of
the extremum search curves of a complicated surface, and Fig. 7.8 shows their
overlay with the contour curves of the surface. Eight cross points of the curves
correspond to the lowest and the highest points and the saddle points of the
surface shape. When drawing the contour curves, these extremum search curves
were very helpful, otherwise some 'extremum points would be missed.
Fig u r e 7.7 Extremum search cu rves
Contour cur ve
Figure 7.8 Contour curves of a patch

with the extremum search curves
To trace the extremum curves given by equations (7.11), we transform the

equation int.o the form (7.7) int.roducing the arc length s as the independent
variable . The initial conditions of the above equations are given usually on the
boundaries of t.he patch using equations (7.10).
Other than equations (7.10), if we trace the curves given by the equations
¢uu(u, v) = 0, ¢vv(u, v) = 0 (7.12)
from t.he boundaries, we can find the intersections with internal loops made by
(7.11), because curves specified by equations (7.12) are the loci of the inflection
points of the constant parameter curves. Locus curves of the derivatives of

higher degree show simple behavior because of the lower degree of the powers of
their polynomials and because they intersect with the loops of the locus of the
derivative of one degree less.
The extremum search curves given by equations (7.10) divides the surface into
regions with a monot.onic gradient and each point determined by the intersection
of the two curves given by (7.12) is the maximum gradient point in each one of
the regions.
The idea of extremum search curves can be applied for similar problems.
7.3.4 Contour Curves and Their Orthogonal Curves

The contour curves of a surface are represented by the intersection of the surface
wit.h equi-dist.ance parallel planes. They have widely been used to represent the
shapes of free-form-surfaces in two dimensions. Let nf be the unit normal vector
of the cut.ting planes. The equation of the contour curves is given by
z = nf . s( u, v) == ¢( u, v) = const == h (7.13)
Differentiating eq. (7.13), we have
(7.14)
This is the differential equation of the contour curves of the patch s( u, v). Its
extremum search curves are given in the previous subsection.
, We introduce a group of curves orthogonal to the contour curves for detailed
descript.ion of the surface shape. We call this group here the contour orthogonal
curves, whose equation is given by
(7.15)
Since the inner product of the tangent vectors derived from eq. (7.14) and
eq. (7.15) becomes zero, the curves given by the above equation are orthogonal
to the contour curves. In Fig. 7.2, both the contour curves and their orthogonal
curves are shown.
7.3.5 Equi-gradient Curves

Let a be an angle between the unit normal n of the patch s and the vertical
direction nf' An equation of points of constant a is given by
cos a = nf . n = const. (7.16)
Sets of these points make the equi-gra.dient curves. If nf represents a direction

of parallel rays and the diffused reflection of the surface is assumed, these curves
are the same as equi-brightness curves. When a is 90 degrees, the curve includes
Figure 7.9 Equi-gradient curves of the Figure 7.10 Curves orthogonal to the
patch shown in Fig.7.6 equi-gradient curves
shade boundary curves of the surface or represents its silhouette curves when
the viewing point is at infinity. Differentiating eq. (7.16) and using Weingarten's
equation, we obtain the differential equation of the curves:
(a¢u + (3¢1))du + (a'¢u + (3'¢v)dv = 0, (7.17)
where ¢ = (nJ . s).

The equation of curves orthogonal to the equi-gradient curves is given by
{(au + (3'¢v)E - (a¢u + (3¢v)F}du+

{(a' ¢u + (3'¢v)F - (a¢u + (3¢v)G}dv = O. (7.18)
As eq. (7.1.5) is deduced from eq. (7.13), so eq. (7.18) is obtained by ¢u and ¢v
in eq. (7.15) being replaced with (a¢u + (3¢1)) and (a'¢u + (3'¢v) of eq. (7.17).
Accordingly, the inner product of tangent vectors derived from equations (7.17)
and (7.18) is zero. Examples of the equi-gradient curves and their orthogonal
curves are shown in Fig. 7.9 and Fig. 7.10.
An equi-gradient curve crosses an umbilic in a horizontal direction, because
the shape of t.he surface becomes spherical at an umbilic. Accordingly, the equi-
gradient curves take an extremum height at. an umbilic.
The pattern of t.he equi-gradient curves can be used to evaluate surface
smoothness including connecting conditions.
Figure 7.11(a) shows the equi-gradient curves of e(l) connected patches and
(b) shows those of e(2) connected patches. In the former pattern, we can clearly
recognize cusps at the patch boundaries. "
( a)
(b)
Figure 7.11 Equi-gradient curves for composite patches: (a) C(l) connection, (b) C(2)
connection
37°
z Direction of
A.
x
¥ projection:n,
Figure 7.12 Example of a silhouette pattern
7.3.6 Silhouette Curve and Silhouette Pattern

In eq. (7.16) for a = 90 degrees and 11, being the vector of the line of sight from
infinity, eq. (7.17) gives silhouette curves. For a designed surface, the silhouette
curve pattern, which is generated by repeated small angle rotations of 11" gives
information on the smoothness of the surface and for the evaluation of its aes-
thetic qua1ity. Figure 7.12 is an example of the silhouette pattern. In patches
with C(1) connection, the silhouette curves show tangent discontinuity on the
patch boundaries, though their contour curves are smooth. Figure 7.13(a.) shows
an example and Fig. 7.13(b) shows the contour curves of the same patches. The
pa.ttern of the lines of curvature of the same surface is shown in Fig. 7.5. Compar-
ing these two curves the silhouette pattern seems more sensitive to discontinuity
of the curvature on the patch boundaries than that of the lines of curvature.
When a. viewing point is not at infinit.y, but at Pel the condition of the
silhouette curve is given by
(7.19)
instead of eq. (7.16). Differentiating eq. (7.19), we have
(7.20)
The solut.ion of eq. (7.20) gives true silhouette curves. Figure 7.14(a) is a parallel
projection of a patch from a direction other than the viewing direction to show
the loci of t.he silhouette curves together with highlight equi-brightness curves
described in t.he next section. Figure 7.14(b) is the perspective projection from
the viewpoint Pe. The silhouette curves bound the highlighted surface as well as
the shaded part of the surface beside the visible boundary.
(a) (b)
Figure 7.13 (a) Silhouette pattern and (b) contour curves, for C(I) connected patches
In most of the shaded pictures produced by computers, the quality of their

silhouette curves is poor, for the computers do not draw the silhouette curves
explicitly and t.he boundaries of the object are shown only in the scan lines.
7.3.7 Highlight Curves

The highlight equi-brightness curve is defined as the locus of points on a surface
where the bright.ness of the surface to a human observer's eye is constant. The
brightness is caused by uniform ambient light, diffusion of the incident light and
also by rays specularly reflected to an observer's eye.
See Fig. 7.15. Let the angle between a normal vector of the surface and the
e
incident ray be and the angle between the line of sight and the specularly
reflect.ed ray be cp, and a, a' and b be the unit vectors of the incident ray, the
reflected ray and the line of sight. There are relations:
- a· n = cos e, a'· b = cos cp = canst. (7.21)
We are to express a' by a and n. The normal components of a and a' have the
same magnitude and opposite directions: they are (a· n)n and -(a· n)n. The
tangential components of a and a' are the same: a - (a· n)n. So we have
a' = a - 2(a· n)n.
One of the equations (7.21) becomes
- 2(a . n)(b . n) + (a· b) = cos cp. (7.22)

(a)
Silhouette curves
(b)
Figure 7.14 True silhouette curves and equi-highlight curves: (a) parallel projection to show
curves clearly, (b) projection from the view point
n
b
Figure 7.15 Model of specular

reflection
(a) (b)
Figure 7.16 Highlight equi-brightness curves and their orthogonal curves: (a) case of only
diffusion reflection, (b) case of diffusion and specular reflection included
The perceived brightness Ep of the surface consists of the uniform brightness

Eo by the ambient light, the brightness Ed by diffuse reflection and the brightness
Es by specular reflection of the incident light I:
(7.23)
where Ed, Es(e) and n are factors and an exponent value dependent on the
conditions and the material of the surface. If we assume they are constant, we
can make a. differential equation of the equi-brightness curves from eq. (7.23)
using the relations (7.21) and (7.22) ,
- p.d(a· n) + nvd{ -2(a' n)(b· 11) + (a· b)y-l = 0, (7.24)
where p. and IJ are rela.tive constants to determine the ratio of the contributions
of diffusion and specular reflections.
Figures 7.16(a) and (b) are examples of the equi-brightness curves together
with their orthogonal curves for the same composite patch shown in Fig. 7.11(b) .
Figure 7.16(a) shows a case of only diffusion reflection, but Fig. 7.16(b) shows a
case including both diffusion and specular reflection. Highlighted region appears
clea.rly.
n Contour curve
Figure 7.17 Vertical component of a

Line of curvature unit normal and extremum slope
7.4 Characteristic Curves on a Surface 2

7.4.1 Gradient Extremum Curves or Ridge-Valley Curves
There are points of extremum height on an equi-gradient,urve. The loci of these
points on all equi-gradient curves are called here the gradient extremum curves,
because at an intersecting point of this curve with a contour curve the downward
gradient along the contour takes an extremum value. Their differential equation
is obtained from eq. (7.17) and the condition of extremum height of the points,
dnf . s(u, v) = 0, which becomes
(7.25)
By eliminating du and dv from equations (7.17) and (7.25), and using equations
(6.28) and (6.29), we obtain
(GM - F N)¢~ - (GL - EN)¢u¢v + (LF - EM)¢; = O. (7.26)
The above equation can also be derived from the equation of the lines of
curvature and the condition (7.25) of the extremum height of the points on it.
And the condition (7.25) is the same as eq. (7.15).
It follows that at these extremum points the equi-gradient curve, the line of
curva.ture and the contour curve have a common tangent. Accordingly, varia-
tion dn of the unit. normal vector n along the contour curve becomes zero from
Rodorigues' formula. So, its component in the nf direction is zero at this point.
d(nf . n) = nf . dn = O. (7.27)
Since (nf . n) is cosO' (refer to Fig.7.17) and this is a measure of the slope
along the cont.our curve, at this point the slope takes an extremum value. This
is another interpretation of the gradient extremum curve. An example of this
curve is shown in Fig. 7.18 together with other characteristic curves.
The curvature k of a contour curve also takes an extremum value at this point
because the tangent direction of the contour curve coincides with the principal
Equi-gradient extremum curve
Contour curve c;!> - Region of nega.tive

Ga.ussian curvature
lit - Umbilic
Locus of zero
mean curvature
Figure 7.18 Equi-gradient extremum curves and other characteristic curves
direction of the surface. The shape of the surface around this point is represented
by a quadric surface from eq. (6.42) given by
z = (1/2)(K:IX2 + K:2y2)
and the contour curve is a section curve given by a plane which includes the
principal axis, for instance x axis, of this quadric surface. The section curve is
symmetrical about the yz plane, that is, its curvature takes an extremum value.
The model surface in Fig. 7.18 is the same as that used in Figs. 7.5, 7.6, 7.9,
a.nd 7.13. The locus passes always through a point of maximum or minimum
slope on each level, which coincides with a. point of extremum curvature of a
contour curve. This is an image of a ridge or a valley path in a topographical
map. So we call the loci the ridge-valley curves. Their equation is given by
eq. (7.26).
Since it is rather difficult to obtain all the curves from step-by-step solution
of eq. (7.26), we convert it into the form of differential equations, which we solve
with the help of a differential equation solver with adaptive stepsize control.
Figure 7.19 is a more complicated case, which is the same surface as in Fig. 7.4.
The curves correctly represent the ridge or valley paths of the terrain. The
even number of curves start or end at points of the extremum heights, and the
curves pass the umbilics which are the locations of the extremum heights on
equi-gradient curves.
7.4.2 Loci of Zero Gaussian Curvature and Loci of Extremum

Principal Curvatures
A locus of zero Gaussian curvature divides a surface into positive and nega-
tive Gaussian curvature regions. In Fig. 7.18, the regions of negative Gaussian
Contour curve
Figure 7.19 Ridge-valley curves and

con tour curves
curvature are shown in light grey. At any points in the negative Gaussian cur-
vat.ure regions t.here are two directions of zero curvature, which are given by
e = ±tan- 1 V(-,.;,I/";'2). At a point where mean curvature is zero, the angle e
becomes 4.5 and 135 degrees. There is a net of curves whose tangent directions
coincide wit.h zero curvature directions. Currently, this net does not seem useful.
Along a line of curvature, the value of the principal curvature changes and
there exist.s a. point. of extremum value of the curvature. The loci of these ex-
tremum curvature points for all lines of curvature can be considered. They
are called here t.he principal curvature extremum curves. Their equations are
obt.ained from t.he equation of the lines of curvature (7.8) and the curvature
ext.remum condition d,.;, = ,.;,,,d'U + ";'vdv = 0, that is,
(7.28)
Since direct. calculation or use of the differential equation form of eq. (7.28) is
very complicated, we determine the extremum curvature points numerically while
t.racing each line of curvature, then we make their interpolation curves. An
example is shown in Fig. 7.20.
7.5 Offset Surfaces

There is a set. of points, ea{:h of which has a constant distance d from a corre-
sponding point of a surface s( 'U, v) in the direction of its normal vector at that
point. They const.itut.es a surface s+('U, v). This surface is called the offset sur-
face of the original one. A point of s+ ('U, v) is located on the normal of the
corresponding point. of s( 'U, v), so we have
s + ( 'U, v) = s( 'U, v) + nd. (7.29)
Loci of extIemum
principal CUIvature
Line of CUIvatuIe
Figure 1.20 Loci of extremum principal curvatures
Figure 1.21 Normal vectors and lines

of curvature between offset surfaces
The normal planes of both the original and the offset planes coincide because
they include their common normals. Now let the parameters u and v be taken
along the lines of curvature of s( u , v). See Fig.7.21. Differentiating eq. (7.29)
with respect to u and using Rodrigues' formula., we obtain
(7.30)
and similarly,
(7.31)
where ;;;1 and ;;;2 are the principal curvatures in the principal directions of s( u, v),
which are the same as those of constant. parameter curves in this case. Notice
that. sign of curvature of a surface obeys the definition given in eq. (5.6). For
example, when the normal directs upwards and t.he normal section is convex
upwards, the curvature is negative.
Any point on the lines of curvature of s( '11, v) has a corresponding point on

the offset surface s( '11, v)+. The parameters '11 and v can be considered those of
the offset surface s('11,v)+. Using equations (7.30) and (7.31), an inner product
of its basic vectors is
Since Su . St, = 0, the above value is zero. It follows that the '11, v parameter net
of t.he offset. surface is orthogonal and t.hey constitute also the lines of curvature
net. Since unit normal vectors n+ and n on the corresponding points are equal,
their increment.s along a line of curvat.ure are the same, that is, dn+ = dn and
also t.he differentials of distances along the respective line of curvature have the
relation ds = ds+ /(1 - "'id), which is derived from equations (7.30) and (7.31).
Rodrigues' formula applied to s( '11, v) is
dn + "'ids = o.
Substitut.ing dn and ds into the above equation, we obtain
+
dn + (1 - "'i"'i d) ds + = O. (7.32)
This is Rodrigues' formula for s( '11, v)+. Accordingly, the principal curvatures of
s+('11,v) are
",.+ =
"'i
,for i = 1,2. (7.33)
• (l-"'i d )
So its Gaussian and mean curvatures are given by
J{+ =
9
(7.34)
(7.35)
In the above description of the surface, the lines of curvature are taken as its
paramet.ric curves for easy underst.anding. Since values of the curvatures are
independent of parameterization, so are the expressions of "'t, J{: and J{;:;. "'i,
Next, in the usual paramet.ric expression of a surface, we are to derive the
fundamental magnitudes of the first and the second order of its offset surface.
Since the basic vectors of the offset surface are given by
s;!' = Su + llud, s; = + llvd

Sv
s;!'u = Suu. + lluu d, s:v = " Suv + lluv d,
we can derive their fundamental magnit.udes from their definitibns,
s+ X s+/ls+ v = 11
x s+1
11+ u v u
E+ E - 2Ld + 1l~d2
F+ F - 2M d + (11" • llv)cP
C+ = C - 2N d + ll~cP
(7.36)
H+ Is~ x s~1 = H(l - 2K",d + KgcP)
L+ = L -ll~d
M+ M - (11 .. 'llv)d
N+ N - ll~d
For derivat.ion of t.he above equations the following relations are used. By differ-
ent.iating the relations 11 • 11 = 1, 11 . Su = 0, we obtain
11 • lluu = -ll~, 11 • lluv = -llu' llv, 11 . llvv = -ll~

llu • S1J = -11' Suu = - L, llv' Sv = -11' Svv = - N,
and ll~, 11~ and 11u • llv in eq. (7.36) are given by equations (6.31) and (6.32),
which are:
(M2 E + L 2C - 2LM F)I H2 = -KgE + 2KmL }
(APC + N2 E - 2M N F)I H2 = -KgC + 2KmN .
= (LJlIC + M N E - M2 F - LN F)I H2 = -KgF + 2KmM
(7.37)
Since we can det.ermine the fundamental magnitudes of the offset surface, we
can calculate the int.ersection curves a.nd various characteristic curves as in the
original surfa.ce.
So long as s+ (u, v) does not intersect with itself, a sphere of radius d, the
center of which is on the offset surface, touches the original surfa.ce at one point,
but if there are regions bounded by the self-intersection curve of the offset surface,
the sphere whose cent.er is in these regions not only touches the original surface,
but also intersects with it at other places.
Since the fundamental magnitudes of an offset surface can be determined
from the data of the original surface, the interference calculations for the offset
surface are performed wit.hout having the explicit equation of the offset sur-
face(refer to Sect.. 16.5). Figure 7.22 shows a surface and its offset one. Problems
of obt.aining t.he int.ersection curves of offset surfaces are t.reat.ed in Sect. 13.6 and
Sect.. 16.5.
Set.s of t.he normal lines emanat.ed from points on the net of the lines of
curvature of a surface s( u, v) constitute the mutually orthogonal surfaces Sl and
S2 in three-dimensional space. The surfaces s+( u, v) and s( u, v) are orthogonal
to t.hem. Since t.he offset dist.ance d can take any values, s( u, v)+ and s( u, v)
can be considered to belong to a set of the surfaces So. Then the space can be
divided by t.he three mutua.lly orthogonal sets of the surfaces So, Sl and S2 .
Offset surface
Original surface
(a)
Offset surface
(b)
Figure 7.22 Original and offset surfaces: (a) 3 X 3 patch, (b) C(l) connected patches
Since the parametric expressions of surfaces do not describe explicitly the

relation of their mutual locations, it is difficult to search for the global rela.tions
among them. The orthogonal set of surfaces explained above, which is generated
from one of the parametric surfaces, serves to give the distance relation to the
other surfaces. This can be used to obtain all the intersection curves between
two parametric surfaces.
Exercise. Show that the expressions of the Gaussian and the mean curvatures
of an offset surface derived from its fundamental magnitudes are equal to those
given by equations (7.34) and (7.35).
7.6 Ruled Surfaces
A ruled surface is generated by a sweeping line which is guided by a space curve.

Let p(u) be t.he given space curve and a(u) a unit vector of the sweeping line
and let them be functions of a parameter u. An equation of the ruled surface
s( u, v) is defined by
s(u,v) = p(u) + va(u), (7.38)
where u and v are the parameters of the surface. From this we obtain
s" = pI + val, Sv = a,
su" = pll + vall, s"v = aI, Svv = 0,
where a prime mark indicates differentiation by u.

Its fundamental magnitudes of the first order are
E p,2 + 2vp'a' + v2a'2,

F p'a, G = a2 .
Its unit normal vector is by definition
n = (s" x sv)IH
(p' x a + va' x a)1 H. (7.39)
The fundamental magnitudes M and N of the second order of the surface are
[p', a, a']
M = 11 • s"v = =-----=-
H
N =0. (7.40)
Its Gaussian curvature is given by
(7.41 )
which is negative or zero. If a is const.ant or parallel to pI, which is the tangent

direction of t.he curve, then M becomes zero, hence the Gaussian curvature is
zero. The surface is developable.
A helical surface is developable, because it is produced by a sweeping tangent
to a helical curve, but a hyperboloid of one sheet a,nd a hyperbolic paraboloid
are not developable, because their generators are not parallel to tangents of their
guiding curves.
8 Curves Through Given Points,
Interpolation and Extrapolation
8.1 Introduction
Methods of constructing a curve which passes through all the points of a given
sequence are described. In CAD, input data for a curve are values of a point
sequence supplied by automatic measuring instruments or values from digitizers
used by a person or supplied from other processes. Since those data contain
errors of various sor~s, sometimes smoothing procedures such as methods of least
squaTes or appropriate filtering procedures are required .
For storage and use of data by computers, input data finally have to be
converted to mathematical expressions which preserve features and accuracy of
the original data.. There are several curve expressions, but except for parametric
polynomial or rat.ional expressions, they are not suitable for use in CAD for rea-
sons of difficulty of application and non-favorable properties of the curves. Non-
paTa.metric expressions are not widely used, but since polynomial and rational
interpolations and extrapolations are used in the latter sections for numerical
methods of solving differential equations, these two representations with their
calculation methods aTe included in this chapter [16].
In CAD, a curve segment is usually represented by a parametric polyno-
mial expression for ease of manipulation. We introduce Hermite expressions for
polynomials and then simulation methods of an elastic beam for smooth inter-
polation. The simulat.ed method has to do with spline curves, but control point
methods, which aTe a very useful technique to represent and control the shapes
of curve segments, aTe mentioned. The detail is treated in Chaps . 9 - 12. Curve
fitting by the rational expression of degree two or the conics in implicit forms is
treated in Chap. 12.
8.2 Polynomial and Rational Interpolation and

Extrapolation
8.2.1 Lagrange's Formula
Given n + 1 sets of coordinates values (Xi, Yi), where the values of Xi are m
ascending or descending order, a polynomial of degree n,
n
p(x) = Lajx j , (8.1 )
j=O
104 8 Curves Through Given Points, Interpolation and Extrapolation
which passes through points
Yi = p(Xi)' i = O... n, (S.2)
is to be determined. The coefficients aj of eq.(S.l) can be obtained by solving

the simultaneous equations (S.2) in n + 1 unknown aj. But this method is not
useful for applications, for it requires solving the simultaneous equations each
time the data are changed, and the coefficients a) do not give direct information
on the shape of the curve.
Now, we introduce functions gj(x) , {j = 0 .. . n} of x, which have favorable
properties for their determination. We express a curve by a linear combination
of gi(X):
n
p(x) = L: Cjg)(x), (S.3)
)=0
whose coefficients Cj are determined to satisfy the conditions (S.2). Lagrange's

interpolation formula has the form of eq.(S.3), in which Cj = Yj and
(x - xo)'" (x - Xj_l)(X - Xj+l)'" (x - xn)
gj ()
x = (S.4)
(Xj - xo) "'(Xj - Xj_l)(Xj - Xj+l) "'(Xj - xn)
Since the value of gj( x) becomes 1 at x = x j and is zero at the other Xi of the
given points, changes of the values and the number of the given points do not
require solving of the simultaneous equations. However, since the degree of the
polynomial is proportional to the number of points to pass, it easily becomes
a higher degree polynomial. Then an unwanted oscillation in the interpolated
curve frequently appears.
Example. See Fig. S.l. Values of Y = Vx are given at x = (1.0), (1/2), (1/4),
(1/16), {1/16)2, (1/16)4. Interpolation curves by the polynomials are shown in
solid lines, whereas those in dotted lines are curves by the rational interpolations
which are explained in the next section. Figure S.l(a) shows a curve derived
from the first four points, (b) a curve from the first five points, and (c) a curve
from all six points. As the order of the curve increases, there appear increasing
oscillations in the polynomial interpolations.
8.2.2 Numerical Methods of Interpolated Points

When additional input data (Xi, Yi) are introduced one by one to increase the
accuracy of an interpolated or extrapolated value Y at x, use of eq.(S.4) is not
adequate for the calculation of the updated value, but the following method
which utilizes the previous value of y is suitable.
Let Pi(i+l) be the value at x of the polynomial of degree one, which passes
through two points (Xi, Yi) and (Xi+l, Yi+l)' Likewise let Pi(i+l) ... (i+n) be the
value at x of the polynomial of degree n which passes through n + 1 points
(Xi, Yi), (Xi+l> Yi+l),"', (Xi+n' Yi+n)' Analogously, since Pi is considered the value
a,t x of the polynomial of degree ' zero, we define Pi = Yi. .
y y
/
Rational
interpolation
Polynomial
interpolation
x x
(a) 4 points (b) 5 points
(c) 6 points
Figure 8.1 Polynomial and rational interpolation and extrapolation for y = sqrt(x)
A value of the polynomial of degree m at given x can be deduced from values
of the polynomials of degree m - 1 using the following recurrence formula:
p.. . . _ (x - Xi+m)Pi ... (i+m-l) + (Xi - x)P(i+l) ... (i+m)

(8.5)
.(.+1) .. .(.+",) - ( )
Xi - Xi+m
It is easily proved from the definition of P; ... (i+m) that the right-hand side of
eq.(8.5) takes values Yi, Yi+l,"', Yi+m, Yi+m+l at x = Xi, Xi+b"', Xi+m+l and
is a polynomial of degree m. These values are produced systematically in the
table below: for the case i = 1 and m = 4, From the first column of the above
table the second column is produced using eq.(8.5). If the elements of the first
column are only PI and P 2 , the result value is P 12 . If P 3 is added, then P 23 is
obtained and from P 12 and P 23 the final interpolated value P l23 is determined .
When this value is not so much different from g2 or P23 , we may not take in
Table 8.1 Value of each Pi(i+l) .. . (i+m)
PI
g2
P2 P l23
P 23 P l234
P3 P 234 Pl234S
P 34 P 2345
P4 P 34S
P 4S
Ps
new input data P 4 and P s . But if we have to use all the data from PI to P s for
interpolation or extrapolation, we have to calculate P 1234S '
Our objective is to evaluate Y at x from a given sequence of points (Xl, yd,
••• , (Xn, Yn). vVe assume that from YI , •.• , Yn-l, a value Y = P I .. n - 1 has
already been determined and now a new point Yn at Xn is added. First we fill
the entries of Ta.ble 8.1 to produce P2 .. n , then we obtain P12..n from P12 .. (n-l)
and P2 .. n . To make the above process systematical, at first we define differences
between Pi(i+l) ... (i+m) and the lower adjacent entries:
Ci ,m P; ... (i+m) - Pi ...(i+m-l), )

Di,m Pi ... (i+m) - P(i+l) ... (i+m), (8.6)
Ci,a P; = Yi, Di,a = P; = Yi
There aTe relations between these differences:
Ci,m+l Xi,m(Ci+l,m - Di ,m)' (8.7)

Di,m+l (1 - Xi,m)(Ci+l,m - Di,m)' (8.8)
Xi - X
where Xi,m
Xi - Xi+m+l
We const.ruct two triangular Tables 8.2 and 8.3, whose entries contain Ci,m
or Di,m' The first index i indicates the i-th row of the difference tables and the
second index m shows the m-t.h column of the tables. Entries in the tables are
filled step by st.ep as each new value Yi = Ci,a = Di,a is added in the both entries
of the first columns (i , 0). At each level m, the Ci,ffi and Di,m are the correction
terms for the interpolation value one order higher. The final value PI ,2'" n is
equal to the sum of any Yi plus Ci,m or Di,m in the ascending order of m from
m = 2 to m = n according to eq.(8.6). Application of the tables appears in the
A ppendix of this book.
Table 8.2 Value of each Ci,j
Cl,Q Cl,1 C 1,2 C 1,3 C l ,4

C 2 ,Q C 2,1 C2 ,2 C2 ,3
C3 ,o C 3,1 C3 ,2
C 4 ,o C 4 ,1
Cs,o
Table 8.3 Value of each Di,j
Dl,o Dl,l D1,z D 1,3 D 1 ,4

D2 ,o DZ,l D2 ,2 D2 ,3
D3 ,o D3 ,1 D 3 ,2
D4 ,o D4 ,1
Ds,o
8.2.3 Rational Function Interpolation and Extrapolation
Approximations by polynomial functions are easily calculated and their deriva-

tives and integrals are easily obtained, but they oscillate around the desired
cur\Tes so t.hat t.hey frequently exceed given error bounds. Approximat.ions by
rational functions usually show smaller error bounds than by polynomials. In
Fig. 8.2, polynomial and rational curves which pass t.hrough t.he same six point.s
are shown. The rational curve seems good, but the polynomial one shows large
oscillation. However, the rational int.erpolat.ions do not. always give good results;
y Rational
in terpolation
Polynomial
in terpolation
x Figure 8.2 Polynomial and rational

interpolation for six points
Y Rational
interpolation
.. / / / \ Polynomial
interpolation
x Figure 8.3 Polynomial and rational

interpolation for six points - unfavor-
able case
for instance, Fig. 8.3 shows very bad interpolation curves for both polynomial and
rational cases.
In numerical methods of solving ordinary differential equations, the Bulirsch-
Stoer method [16][17] applies rational instead of polynomial extrapolation. It is
an excellent method of solving ordinary differential equations and we use it in
Chap. 16; its detail is explained in Appendix of this book. Since its procedure to
obtain an interpolated or extrapolated value is similar to that of the polynomial
procedure described above, we treat the case of a rational function here .
. Let R i(i+1),,(i+n) be a rational function which passes through n + 1 points:
A recurrence formula analogous to the polynomial case eq.(8.5) is
R i(i+l)",( i+n) = R(i+1)" ,( i+n) +

R(i+1).,,(i+n) - R i",(i+n-1)
, (8.9)
x-x
----'-'-{1 _ ,.+1:,
R( ' ) (',+n )-R, .,,' (,+n-1
' ) } _ 1
x - Xi+n R(,+1)",(,+n)-R(,+1)",(.+n-1)
Ri = Yi, R i(i-1) == O.
For x = Xi, the first term is R(i+1) ,,(i+n) and the second term gives
-{R(i+1) ,,(i+n) - R i ,(i+n-l)}'
The result. is Ri ,,(i+n-l), which is Yi.

For x = Xi+1,"" Xi+n, the first term gives Yi+1,"" Yi+n and the second term
gives zero. Accordingly, eq.(8.9) is the recurrence formula we want.
For n = 1, we obt.ain a rat.ional curve which passes two given points. From
eq.(8.9) we have
R i+1 - Ri
= Ri + 1 + X -
---{I _
x,' R
i+1 -
R
i} _ 1
x - Xi+n R i+1
YiYi+1(Xi+1 - Xi)
=
The denominat.or is a polynomial of degree one, whereas its numerator is con-

stant., that is, a polynomial of degree zero. Since by definition R i(i+1) passes two
points (XiYi) and (Xi+1' Yi+1), we have the extrapolated value at x = 0 :
(S.10)
For n = 2, a. rational curve which passes given three points is obtained from
eq.(S.9) by
R i(i+1)(i+2) R(i+1)(i+2) +
(x - Xi+2)(R(i+1)(i+2) - R i(i+1))(R(i+1)(i+2) - Yi+1)
We wa.nt t.o obtain R 12 ... n by constructing two triangular tables similar to Tables
S.2 a.n d S.3 of the polynomial case. For this purpose, we define the difference
functions bet.ween the adjacent rat.ional functions similar to eq.(S.6):
Ci,m = Ri ... (i+m) - R i ... (i+m-1) ,

Di,m = Ri ... (i+m) - R(i+1) .. (i+m), (S.l1)
Ci,o = R i, Di ,o = Ri.
Inst.ead of equat.ions (S.7) and (S.S), we have the recurrence formulas for C i,m
and Di,m for the rational case:
X i ,m D i,m(Ci+1,m - Di,m)
C i ,m+l (S.12)
(Xi,mDi,m - C i +1,m)
C i,m(Ci+1,m - Di,m)
D i,m+l (S.13)
(Xi,mDi,m - C i+1,m) ,
(x - x;)
where Xi ,m = (x - Xi+m+1)"
Then we can const.ruct a table similar to Tables 8.2 and S.3 step by step each
time a new pair (Xi, Yi) is added. If x is given, we can determine Y = R 12 ...n by
a similar method as for Y = P12 ... n."
When x is out.side of x n , the corresponding value of y is an extrapolated

value which is given by
R 12 ... n = Rn + Dn- 1,1 + Dn- 2 ,2 + ... + D1,n-1' (8.14)
Example. For t.he examples given in Fig.8.1, we obtain extrapolated values of

y = ..;x at x = 0, when the correct y values are given at seven points of Xi:
x = 1.0, 1/2, 1/4, 1/16, (1/16)2, (1/16)\ (1/16)8.
Relations between the extrapola.ted values at x = 0 and the number of the given
points taken from t.he head of the above sequence are shown for the polynomial
and the rational ext.rapolations.
number of points 4 5 6 7
polynomial 0.1417 0.04699 0.00366 0.000015 (8.15)
ra.tional 0.1249 0.04096 0.00357 0.000015
Though the curves gi ven by the polynomials oscillate extremely around the true
curve as the order increases, the values of y given by the two extrapolations at
x = 0 do not differ considerably for this sequence oevalues of x.
8.3 Polynomial Interpolation with Constraints of

Derivatives
The Hermit.e int.erpolat.ion method specifies points to pass as well as derivatives

t.here. In t.he simplest case, a curve segment is to be determined by its two
end points. A curve which passes through many points is composed from curve
segments , each of which passes through the aclja.cent points.
Let r(t) , {O ::; t ::; I}, be a curve segment with a parameter t between two
points r(O) = 1'0 and r(l) = r1, where derivatives up to the m-th with respect to
t are specified, t.hen the degree of the curve is at least 2m + 1. We are to express
r(t) in terms of t.hese 2m + 2 derivat.ives with respect to the parameter at the
end points, multiplied by suitable weight.ing functions. We use notations r~i) and
rii) for i-th deriva.tives of 1'(t) at t = 0 and t = 1 and the notations gi(t; 2m + 1)
and hi (t; 2m + 1) for the weighting functions of r~i) and 1'ii).
The curve segments of degree 1, 3 and 5 are expressed by
m=O r(t) = r~O)go(t; 1) + 1'10)ho(t; 1), (8 .16)

m= 1 r(t) = r~O)go(t; 3) + riO)ho(t; 3) + r~l )g1 (t ; 3) + ril)hl(t; 3), (8.17)
m=2 r(t) = r~O)go(t; 5) + 1'iO)ho(t; 5)
+ r~1) gl (t; 5) + rF) hl (t; 5) + 1'~2) g2(t; 5) + ri2)h2(t; 5), (8.18)
8.4 Elastic Curves with Minimum Energy ill
where the weighting functions gi(t; 2m + 1) and hi(t; 2m + 1) are given by

m=O go(t; 1) = 1 - t, ho(t; 1) = t,
m= 1 go(t; 3) = (1- t)3 + 3t(1 - t)2 = 1- 3t2 + 2t 3,
ho(t; 3) = t 3 + 3t 2(1- t) = 3t 2 - 2t 3,
gl(t; 3) = t(l- t)2, hl(t; 3) = -t 2(1- t),
m=2 go(t; 5) = (1- t)3(1 + 3t + 6t 2), ho(t; 5) = t 3(10 - 15t + 6t 2),
gl(t; 5) = t(l- t)3(1 + 3t), h1(t; 5) = -(1- tW(4 - 3t),
g2(t; 5) = (1/2)t 2(1- t)3, h2(t; 5) = (1/2)(1- t)2t 3.
(8.19)
The general form of r(t) of degree 2m+1 is [2]:
m
r(t) = L r~i)gi(t; 2m + 1) + r~i)hi(t; 2m + 1), (8.20)
j=O
where
gi( t; 2m + 1) = ~ f
Z. j=O
Cr+1- i • t i+j (1 - t )2m+l-i- J, (8.21)
h(f'
t ,
')m + 1) =
IiJ
.!.',L...J
~ C 2m +1- i . t 2m +1- i-j(1 _ t)i+J
J ' (8.22)
Z. J=O
and cj represent the binomial coefficients.

These functions have the following relations,
go(t; 2m + 1) + ho(t; 2m + 1) = (t + 1 - t?m+1 = 1,
hi(t;2m+ 1) = (-1)igi(1-t;2m+ 1).
The shape of a curve segment is expressed by the weighted sum of the derivatives
of r(t) from O-th t.o m-th degree at both ends of t.he curve. Though values of the
derivatives with respect to a parameter are difficult to give explicitly, the above
expressions are used t.o connect with the adjacent curve segments.
8.4 Elastic Curves with Minimum Energy

The shape of an elastic wire, which is deformed by external constraints or forces,
is determined t.o make it.s stored elastic energy minimum. So, its shape is con-
sidered natural and smooth. We want. to use this elastic wire for interpolation
of a given point sequence.
Let n+ 1 point.s P;(Xi' Yi) with Xi < Xi+l, {i = 0 ... n}, be given and let Yi,i+1
be the shape of interpolated curve between Xi and Xi+}, then we are to determine
all Yi .i+1 connected wit.h continuous curvatures. Refer to Fig. 8.4. Using notations
di.i+1 == Xi+1 - Xi and t == (x - x;)/(di.i+1 ) , the slope in the i-th section is written
as
( dY ) 1 (d Yi )
dx 'i = di •i +1 dt '
,
Pi Yi Yi,i+1 Pi+1 y'
.+1
1
~di,i+1~
. .
Xi Xi+1 x
Figure 8.4 Elastic beam model for smoothing
Using t.he weight.ing functions given by eq.(8.19), we express the shape Yi,i+1 of
each curve segment in the following form:
Yi ,i+1 = YigO(t) + Yi+1 h O(t) + y;di,i+1g1(t) + y;+A.i+1 h1(t) , (8.23)

for i = 0 ... n - 1.
There are n + 1 unknown Y! which are derivatives with respect to x at the

junctions Xi. They are determined to minimize the §itored elastic energy. This
condit.ion leads to t.he following equat.ions (refer to Sect. 8.6):
y:'---dXi) (8.24)
Y~,l(XO)
which stipulate that a,t each junction, the second derivatives of the curve seg-
ments are continuous and at the end points they are zero. Substituting eq.(8.23)
into equations (8.24), we obtain the following three-term simultaneous equations:
, ( 1) '( 1 1), ( 1 )
Yi-1 ~ +2Yi ~+~ +Yi+1 ~
1- 1,1. 1-1 ,1 1,1.+1 1,,1+1
= 3 (Yi - Yi-1)
2
+ 3 (Yi+1
c-EYi)
' for
i = 1 . . . n - 1, (8.25)
di- 1•i i,i+1
and the conditions at both ends are
2yb do.1 + y~dO. 1 = 3(Y1 - Yo), } (8.26)

Y~_ldn-1.n + 2y:,dn- 1,n = 3(Yn - Yn-1)'
Solving the equations, we determine values of Y! and the shape of each segment
from eq.(8.23).
If at bot.h ends of the curve its slopes yb and y~ are fixed, the number of
unknown Y! is n - 1, which is equal to that of equations (8.25). So, we can solve
them and calculate the shape from eq.(8.23).
8.5 Interpolation by Parametric Curves 113
8,5 Interpolation by Parametric Curves

In the previous section, a curve is expressed in the form y( x). Though this
form is easy to treat, a limitation of the expression arises when y of the curve
takes multi values. Accordingly, for shape design, curve segments with parametric
expressions are used. For a curve segment 1'i(t), {O ::; t ::; I}, with its end points
Pi and Pi+1 and t.he tangent vectors i'i(O) and ri(l) there, its expression is given
from eq.(8.17) for m = 1 by
(8.27)
with the conditions
(8.28)
which specify the .continuity of the tangent at the junctions, where ki+J is a
constant which we take as the ratio of t.he chord lengths of two curves. Since
in t.his section, the independent variable t of the curve expressions is normalized
between 0 and 1, the ratios k i have to be taken into consideration.
For each interval of the point sequence Po, PI' P2"", P n , a curve segment
of the above form is assigned and connected to its adjacent curve segments with
tangential cont.inuity. To determine the value of i'i(O) at the connecting points
to make the whole shape as smooth as possible, pseudo-elastic energy stored in
the curve is made a. minimum. Let a chord vector of a segment be
(8.29)
and let it.s absolute value lail be ai and assume that shape of the curve does
not deviat.e much from its chord line, then ds ~ la;ldt and the curvature IS
proportional t.o i:;ja} , and the stored pseudo-elastic energy is given by
E = L(1jan f i:i(t)2dt. (8.30)
In the above equation j:i is the function of i'i(O) and i'i+ l(l) by equations (8.27)
and (8.28). From the minimum condition 8Ej8i'i(0) = 0, we obtain similar
results as in the previous section; continuity of curvature at the intermediate
connecting points of curve segments and zero curvatures at the end points,
ri -:: 1 (1)
= ~?i\(0). , i = 1 . . . n - 1, }
1'0(0) = 1'n-l(l) = 0, (8 .31)
where ki = ai-da,.
From these equat.ions, which correspond to equations (8.23) and (8.24), we get
the simultaneous equations for the tangent vectors at the connecting points,
i'i-l + 2i'i(ki + kn + i'i+lk;ki +1 = 3(Pi - Pi-I) + 3(Pi+l - p;)k;,(8.32)

for i = 1 ... n - 1.
and at the both ends
2ro + r1k1
r n-I + 2i'n (8 .33)
where ri = i'i(O),
°
When a.t the end points the directions of the tangents are specified instead of
zero curvatures, the magnitudes of the tangents are unknown. So, aE/a\i'o\ =
is needed instead of aE faro = 0, that is,
(8.34)
Similarly for the other end,
" . rn-l
rn + 2'2
rn = 3rn'
' (Pn - Pn-l ) . (8.35)
These boundary conditions are used instead of eq.(8.33). Since we can use the
Bezier expression (refer to Chap. 9) in place of eq.{8.23), a curve segment ri(t)
of degree t.hree is represented by its control polygon with three edges e3i , e3i+I
and e3i+2 (in Chap. 9, a sYl11bol a is used instead of the symbol e for an edge of
a cont.rol polygon). The tangent vectors at the ends of the segment are given by
its edges vectors:
It follows that equations (8.32) and (8.33) are converted to the relations
among the polygon edges as below,
r
e3i-3 + 2e3i{ki + kf) + e3i+3 k ki+1 = (Pi - Pi-I) + (Pi+l - p;)k; , }
i = 1 ... n - 1, (8.36)
2eo + e3 k I = (PI - Po), e3n-3 + 2e3n-l = (Pn - Pn-l)'
Equations (8.32) and (8.33) or (8.34) and (8.35) are three-term simultaneous
equat.ions whose solution is easily obtained. When a point sequence is given,
the chord vectors of adjacent point. pairs and the ratios of their absolute values
k i = ai-I! ai are all the data for the coefficients of the equations for determining
fi's. Each curve segment is calculated from eq.{8.27). Their connection continu-
ity is C(2).
When the dist.ances between adjacent points of the sequence are almost the
same, then const.ants k; may well be taken equal to 1, but when their variation
is great, k; should be taken as defined. In Fig. 8.5 two examples are shown with
the same given point.s: (a) constants k i are taken as ratios of the chord lengths,
(b) all ki are equal to 1. The resulting shapes are quite different.
8.6 Appendix. Derivation of Equations by Elastic Beam Analogy 115
(a) k; = ratio of adjacent chord lengths (b) k; = 1
Figure 8.5 Curve shapes and scale ratios: (a) k; = ratio of adjacent chord lengths, (b) k; = 1
8.6 Appendix. Derivation of Equations by Elastic Beam

Analogy
Equations (S.24) and (S.25) are derived in the following way. For deflect.ion of
a wire or slender beam, a bending moment applied to its axis contributes most
to the stored elast.ic energy, and any other elastic energy such as that caused by
uniform compression need not be considered for the present purpose. The stored
elastic energy of a. wire by bending is expressed by(I/2) J G,,?ds, where G is the
rigidity of bending and K, is the curvature of the wire a.xis, and the integral is
taken along its axis.
. Minimizing this elastic energy under given constraints is difficult, because of
the nonlinearit.y of the curvature expression and the dependence of the bending
moment on the deformed shape of the wire. This is the kind of problem of
Elastica [19], and even if it could be solved any way it may not be practical.
Since its energy is used for a shape criterion only, its expression need not be
very accurate so long as its characteristics are preserved. Accordingly, instead
of a. flexible wire, an elastic beam of small deflection is assumed. In this case,
its curvature is approximated by J2y/dx 2 with the initial beam axis in the x
direction and its deflection in the y direction. This means the reaction forces
from the supports are directed vertically, so the bending moments at a section
of the beam between the supports is a linear function of the distance from a
support point. Since t.he bending moment divided by the bending rigidity of the
beam is equal to the curvature at that point, its deflection y is expressed by a
polynomial of third degree in x, because of the double integration of the linear
function of x representing the bending moment of a section of the beam.
This polynomial is expressed by the vertical positions (Yi, Yi+l) and the slopes
Y:and Y:+lat the end points Xi and Xi+l, multiplied by their corresponding
weighting function 90, ho, 91 and hI given in Sect. S.4. So we have eq.(S.23). We
construct a continuous elastic beam which consists of n curve segments connected
ll6 8 Curves Through Given Points, Interpolation and Extrapolation
at n - 1 support.s. The total elastic energy st.ored in the deformed beams is
where Y::i+l is the second derivat.ive of the deflection between the i-th and
(i + 1)-t.h supports, which represents the curvature there. The total energy is a
funct.ion of t.he slopes yi at the supports. Differentiating it with respect to yi and
setting the result to zero, we have
~G
~
J It
Y.,.+l
/) /I
Yi,i+l d
>l I
UYi
X
0, i = 0 ... n.
By evaluating t.he above equat.ions, we obtain equations (8.25) and (8.26).

9 Bezier Curves and Control Points
9.1 Introduction
There are various expressions for describing a shape, but the number of practical
ones is limited. A polynomial expression in parametric form with control points
is most useful in most applications, for it is not only mathematically simple,
but also it has favorable properties for engineering design; its expression does
not depend on coordinate systems, geometrical properties are easily grasped and
manipulated by control points, unwanted waviness usually does not appear in
construction and modification of shapes, and long curves and large areas can be
designed by connecting short or small segments.
Accordingly, we treat single segments of Bezier curves and surfaces which
have the above stated characteristics. Their expressions are a little different from
those in other textbooks, but because of our adoption of shifting operators for
manipulation of control points, our expressions for Bezier curves are simpler and
easier to process [5][8]. Then in the next chapter, B spline curves and surfaces
which have similar favorable properties in design are introduced as the special
connection of segments of Bezier curves and surfaces. We think this approach
is the easiest for design engineers to understand and use. The extension of the
approach to special applications is explained in later chapters.
9.2 Curve Segment and Its Control Points

We expla.in the relat.ion bet.ween the shape of a curve segment and its control
point.s together with their weighting functions. Let a curve segment be denoted
by r(t) whose components are x(t), y(t), z(t), where the parameter t takes values
o :::; t :::; 1. A curve segment is expressed by the summation of its control
points Po, PI' P2' P3'" Pn multiplied by the corresponding weighting functions
fo(t), fI(t), h(t),"', fn(t) :
r(t) = Pofo(t) + pdI(t) + pzh(t) + ... + Pnfn(t). (9.1)
A polyline made of the control points Po, PI, P2, P3 ... ,Pn is called a control
polygon. For the expression (9.1) to be useful, it needs to have the following
properties:
- The expression is independent of coordinate systems used.
- The shape of the control polygon approximates to that of r( t).
- The geometric characteristics of the curve are obtained from the control poly-
gon only. .
118 9 Bezier Curves and Control Points
To meet these requirements, the following relation must hold for the weighting
functions independently of t:
10(t) + 11(t) + 12(t) + ... + 1n(t) = 1. (9.2)
When all the control points converge to a same point P, the expression (9.1)
becomes
r(t) = p{10(t) + 11(t) + h(t) + ... + 1n(tn· (9.3)
Since both sides of the above expression must be independent of t and equal
to P, eq. (9.2) is derived. This is also the condition for independence of coordi-
nate transformation. A transformation applied to the left-hand side of eq. (9.1)
must be the same as the curve with the transformed control points and the same
weighting functions. Since a transformation consists of a rotation and a transla-
tion, the form of eq. (9.1) is invariant for a rotational transformation and together
with t.he condition. eq. (9.2) it is invariant for translation of the coordinates.
Properties of the weighting functions 1i(t) are described next. When all the
control point.s except an arbitrary Pj converge to one point P, the expression
(9.1) becomes
r(t) = p{10(t)+ 11(t)+ h(t)+···+ 1n(tn+(Pj-p)fJ(t) = P+(Pj-p)fJ(t), (9.4)
because of eq. (9.2). If r(t) is to be within the line segment joining P and Pj' a
weighting function 1j(t) must have a value between 0 and 1: hence for all j
o :::; fJ(t) :::; 1. (9.5)
The equality holds only for t = 0 or t = 1, where only one weighting function
is 1 and the others must be aU zero. Without these conditions, the curve is not
within the line segment (p, Pj)' And all the weighting functions must have a
single pea.k in the range, otherwise the relation between the control points and
the curve shape becomes complicated and the shape cannot be inferred from the
control polygon . Then the roles of the control points cannot be satisfied.
With these properties of fJ(t), the curve r(t) exists within the convex hull
made by all the control points (Po PI P2 P3 . "Pn)' See Fig. 9.1. If we take the
x-axis to coincide with one of the edges of the convex hull, the signs of y values
of all the control points are the same or zero, consequently the value of the curve
generated by the cont.rol points multiplied by their weighting functions has the
same sign. This means that the curve is on one side of the edge of the convex
hull. For all edges a similar property holds, hence the curve cannot go out of the
convex hull. It follows that if the control polygon is planar, the curve is planar,
and if the edges of the control polygon are collinear, so is the curve.
9.3 Bezier Curve and Its Operator Form

We explain a. curve segment with its control points expressed in a compact oper-
ator form and the merits of using the operator. To indicate explicitly the degree
9.3 Bezier Curve and Its Operator Form 119
PI PI P2
.. .,
,r;a- - -- - ----- -- ----~
.,.
,,
.,,
,
,: ...
/// \,,\,
/,/ \\
Po Po
( a) (b)
'"
:' ...
/ \\
'~
Po Po
(c) (d)
Figure 9.1 Control polygons and their generating curves
n of the polynomial of t and the first control point Pi of r(t), we use a notation
ri(t; n) (9.6)
for representing the curve segment. The control points of this curve begin from
Pi and end at Pi+n'
A curve for n = 0 is considered a single point, so the following expressions
can be used:
ro(t; 0) = Po , ri(t; 0) = Pi . (9.7)
For n = 1, since the number of control points of a curve segment is two, a
curve ro(t; 1) is considered a line segment connecting them, and the degree of its
two weight.ing functions is 1. Owing t.o the conditions imposed on them, their
simplest forms are:
fo(t) = (1 ~ t), (9.8)
Then, we have the expression
(9.9)
This is rewritten using eq. (9.7) as
ro(t; 1) = ro(t; 0)(1 - t) + 1'1(t; O)t. (9.10)
A point given by the left-hand side of the above expression is the point dividing
the line segment joining 1'o(t; 0) and 1'1(t; 0) in the ratio t : 1- t.
For n = 2, increasing all the degree pa.rameters of the curve expressions in
eq. (9.10) by 1, we get
ro(t; 2) = ro(t; 1)(1- t) + 1'1(t; l)t. (9.11)
The expression r1(t; 1) is the line segment connecting P1 and P2, which is for-
mally derived from 1'o(t; 1) by increasing its subscript by one or from 1'1(t; 0) by
increasing its degree parameter by one. For fixed t, eq. (9.11) is considered to
give a point dividing the line segment joining ro(t; 1) and 1'1(t; 1) in the ratio
t : 1 - t. When t changes, the locus of 1'o(t; 2) represents a curve segment of
degree two, whose control points are Po , P1 and P2 . Generalizing eq. (9.11) for
a curve segment of degree n, we obtain
ro(t; n) = ro(t; n - 1)(1 - t) + r1(t; n - l)t. (9.12)
The left-ha,n d side of the above equation is the locus of a point dividing the line
segment connecting 1'o(t; n - 1) and 1'1(t; n -1) in the ratio t : 1- t. The curves
1'o(t; n - 1) and 1'1(t; n - 1) are determined respectively from the control points
(PO'P1,·"Pn-1) and (P1,P2,···Pn)·
Now, to simplify the manipulation of mathematical expressions of curve seg-
ments generated from the control points, we introduce the shift operator E which
operat.es on ident.ifiers of t.he sequent.ially ordered objects of a group such as the
cont.rol point.s of a curve. Its basic property is to increase or decrease the number
which is used as a subscript. of an object in the group in the following way:
E- 1 Pi+1 = Pi' (9.13)
From these, we obtain the rules of E for integers m and n,
(9.14)
where I is the identit.y opera.tor which does not change its operand, just like 1
in arithmet.ic multiplication. The rules (9.14) indicate that the shift operator E
obeys the law of multiplication in algebra, hence we can trea.t E as an algebraic
const.ant.
9.4 Different Expressions of B Curve 121
Since ri(t; n) is expressed by summation of terms each of which consists of

one of the consecut.ive control points Pi' Pi+l " ", Pi+n multiplied by its corre-
sponding weight.ing function, the shift operator E can operate on ri(tj n) as well.
So we can express eq. (9.12) as
ro(tj n) = ro(tj n - 1)(1 - t) + Ero(tj n - l)t

= (1 - t + Et)ro(t; n - 1). (9.15)
By applying the above equation to its right-hand side factor ro(t; n - 1) repeat-
edly, we finally obtain the following form for a curve segment of degree n:
ro(t;n) (1- t + Ettro(t; 0) (9.16)

or ro(t; n) (1 - t + Ettpo . (9.17)
To express the ab~ve equation by the last control point P n , the right-hand side
of eq. (9 .17) is multiplied by E-nE n = I and then it is converted to the desired
form.
ro(t; n) = (1 - t + EttE-nEn po
= {(1-t)E- 1 +t}npn- (9.18)
In the right-hand side , P n appears explicitly instead of Po, so we denote this

expression as r,,(t; n) instead of ro(t; n),
(9.19)
Equations (9 .17) and (9.19) are the operator expressions of a curve segment of
degree n with n + 1 control points. This curve segment is the same as what
is called a Bezier curve, whose form is given by eq. (9.20) in the next section .
This operator form of a Bezier curve is convenient not only for its conciseness,
but also for its mathematical manipulation. We call the control points Pi Bezier
points or B points for brevity. Instead of B point sequence, we use the term B
polygon with the same meaning, but when its edges are emphasized, the term B
polygon is more appropriate than B point sequence.
9.4 Different Expressions of B Curve

Expanding eq. (9.17) and noticing that EipO = Pi' we get
n
ro(t; n) = L Ci(l - t)n-iCPi (9.20)
i=O
where Ci are the binomial coefficients , which are equal to n!/{i!(n - i)!}. The
weighting function of eq. (9.1) in t.he B curve is
(9.21)
!i(t) !o(t)
fI(t) Nt)
h(t)
o 1
Figure 9.2 Shapes of weighting functions j;(t) for n =3
The shapes of functions !i(t) are shown in Fig. 9.2.

The expression (9.20) is a standard form of a Bezier curve or a form using
Bernstein polynomials [3]. The original expression by Bezier was a rather com-
plicated one; see Sect. 9.13. In this book we use eq. (9.17) for the Bezier curve,
because it is easy for algebraic manipulation. For numerical calculation, the
polynomial form of t is sometimes convenient, which is obtained by expanding
eq. (9.17) in t,
n
{I + (E - l)t}npo = L: Ciqiti, (9.22)
i=O
t
where qi (E - l)ipo = L:( -l)iC;pj'

)=0
9.5 Derivatives at Ends of a Segment and Hodographs

The tangent direction of a B curve is obtained by differentiation of eq. (9.17)
with respect to t :
dro(t;n)
i'o(t; n) = n(l - t + Et)n-I(E - l)po
dt
n(l - t + Ett-I(PI - Po)
n{rl(t; n - 1) - ro(t ; n - I)}. (9.23)
This shows that the tangent direction at a point on ro(t; n) is that of a vector
from ro(t; n - 1) to rl(t; n - 1), while according to eq. (9.12) the point ro(t; n)
divides this vector in the ratio (t : 1 - t).
9.5 Derivatives at Ends of a Segment and Hodographs 123
Po P3
(a) control polygon (b) its hodograph
Figure 9.3 Construction of a hodograph
Next we explain the hodograph of a curve, which provides detailed information

on its derivative vectors. We use notations ao, aI, a2,··· an-l for edges of a control
polygon of degree n. Since (E -1)po is equal to ao, we can rewrite eq. (9.23) with
edges a; of the polygon:
ro(t ; n) = n(1- t + Et)n-Iao . (9.24)
If we consider the edge vectors ai as position vectors, a B polygon can be con-

structed from a; , such as shown in Fig. 9.3. This B polygon produces a curve
proportional to i'o(t; n) which is called a hodograph of the original curve ro(t; n).
The hodograph is useful for examining the behavior of its original curve. For
example in Fig. 9.4, whether each of three curves (a.), (b) and (c) of degree three
generated from almost similar B polygons has a loop or not can be determined
by using their hodographs (d), (e) and (f) . The hodograph (d) of the curve (a)
passes to the right-hand side of the origin, that of (b) passes to the left-hand
side, whereas that of (c) passes through the origin. The direction of the tangent
vector of the curve (a) changes by far less than 180 degrees clockwise, that of (b)
by far more than 180 degrees counter-clockwise, and the curve (c) has a singular
point where its tangent vector vanishes .
The second derivative of ro(t; n) is given by
Just as the hodograph of ro(t; n) is constructed from edges of the B polygon

of ro(t; n), the hodograph of ro(t; n) is made from edges of the B polygon of
i'o(t; n). It shows graphically the . behavior of i;o(t ; n) . At inflection points of
Q,~.----- - --- ----- ---------- - -,{;J
\ .-,/'
(a) curve with no loop (d) hodograph of (a)
Gt:~_--------------------- ---- ------------- ____ _______ ::,i:J

-'
(b) curve with a loop (e) hodograph of(b)
~ -------- ---- --- ------ ---- ---------.~~,::j;J
""'" ., -
(c) curve with a cusp (f) hodograph of (c)
Figure 9.4 Examples of three hodographs

9.5 Derivatives at Ends of a Segment and Hodographs 125
\\,~1
t = 0.25 \, t = 0.5
a}/
Po \/ P2
(a) r : original curve and its control polygon
:' ':
\ ;
a1 .... \
\:
"
\~
(b) i' : hodograph of r (c) r : hodograph of i:
FigUl'e 9.5 Behavior of i' and r
ro(t ; 11,) two vectors i'o(t; 11,) and ro(t; 11,) become parallel, because its curvature
vector is proportional to the normal component of ro(t; 11,) (refer to eq. (5.24))
and it becomes zero. Figure 9.5 shows B polygons and vectors of 1'o(t; 4), i'o(t; 4)
and j:o(t; 4) from t=O to t=l. There are two values of t where i'o(t; 4) and j:o(t; 4)
become parallel. These values correspond to inflection points of 1'o(t ;4).
The higher-order derivatives are obtained similarly:
(9 .25)
They also have the form of the B curve expression.

The derivatives at the end points of the curve are given by setting t = 0 or
t = 1 in the above equation:
(9.26)
.(j)( . ) _ n! ( E-1)J (9.27)

10 1, n - (n _ j)! 1 - Pn'
The j-th derivative at an end point is determined by the j + 1 consecutive control
points from the end point. Conversely, the j-th control points Pj and Pn-j from
the end points Po and Pn are determined by the j + 1 derivatives from zero-th
to j-th at the respective ends:
J
Pj = (1 + E - l)1po = I:(q)(E - l)kpo
k=O
1 J . (k)
, I: (CD(n - k)!ro (0; n), (9.28)
n. k=O
J
p"-J {I - (1- E- 1)}1Pn = I:(q)(-l)k(l- E-1?Pn
k=O
1 J . (k)
= ,I:(q)( -l)k(n - k)!ro (1; n), (9.29)
n. k=O
for j = O,···n.
The relation between the control points of connecting curves at their junction
is determined by the continuity conditions of derivatives there. Equations (9.26)
and (9.27) are also used for this purpose.
9.6 Geometric Properties of B Curve

The geometric properties of a B curve can be expressed by the edges of its control
polygon. At an end point of a curve segment, we can express important values
of geometric features of the curve in terms of the first three edges of the polygon.
From eq. (.5.22) in Chap. 5, its unit tangent vector t is expressed by
(9.30)
Since the osculat.ing plane is made by t and t + dt, it coincides with the plane
made by the first and the second edges of the polygon. The unit normal vector
11 is in the osculating plane and 90 degrees anti-clockwise from the first edge
vector. This is derived from eq. (5.26) and the definition of the unit normal.
From equation (5.24), the curvature at the end point is given by
n - 1 (al . 11)
K, =- n
--- ~.
- (9.31)
The sign of K, is determined by the sign of (aJ . 11). The unit binormal vector b is
det.ermined by t x 11 and the torsion at the end point is given from eq. (5.27) by
(9.32)
9.6 Geometric Properties of B Curve 127
al
ao
Po '-<::,:-----------i-------!:-~--------------:--}::-
"" i Pz -,
P'~" i
~o'
"'.0
Figure 9.6 Locating the center of curvature
At the starting point of a curve segment, its tangent direction, curvature and
torsion are represented by the first three edges of its Bezier polygon with forms
very similar to their original definitions: t has the direction of ~, the change
of the tangent direction de of the curve is considered to be proportional to the
normal component of al divided by the length of ao, and the arc length ds is
represented by t.he length of ~, that is, de/ds which defines the curvature is
given by eq. (9.31). The change of torsion angle db is represented in the B
polygon as the binormal component of a2 divided by the normal component of
aI, accordingly, db/ds which defines the torsion is given by eq. (9.32).
When the B polygon of a curve is given, its tangent, curvature and torsion at
its end points are obtained graphically. For instance, a procedure for determining
the center of curvature is as follows. Letting p be the radius of curvature and p'
be equal to (n - l)p/n , we get from eq. (9.31) the following relation:
p' : laol = laol : (al . n),
from which p' is determined graphically. Referring to Fig. 9.6,
1. Draw a perpendicular at PI to the edge ao, to which draw a perpendicular
from P2 to p~. Draw a line Ll joining its foot P~ and Po.
2. Draw a perpendicular from PI to the line Ll and extend it to intersect with
the normal line from Po. Let the intersection point be 0'.
3. Extend the vector Poo' by n/(n - 1) times from Po to determine o. Then 0
is the center of the curvature at the point Po and the length of Poo is the
radius of curvature.
A similar procedure can be applied to locate the center of torsion graphically by
using the following relation:
(n - 2)Pr/n ~ I~I = (al . n) : (a2 . b),
where PT is the radius of torsion. The geometric properties stated above are
t.hose at t.he end points of a curve segment. Those at the other point on the
curve can be expressed in the same form, because any point on the curve can be
converted to an end point of a curve segment made by division of the original
segment, and t.he B polygon of the divided segment can be obtained from that
of the original B polygon.
9.7 Division of a Curve Segment and Its B Polygon

Equat.ion (9.17) is the expression for the B curve segment for t = 0 to 1. Divid-
ing it. int.o two segments, we can give them similar expressions, that is, we can
construct two B polygons for the divided segments.
Suppose we divide the curve ro(t; n) at t = t* int.o t.wo divided segments
1'g(t; n) wit.h a B ·polygon Ba = {Po,"" p~} and 1'g(t; n} wit.h a B poly-
gon Bb = {pg, ... ,p~}. To deduce 1'o(t; n} in the standard operator form from
eq. (9.17), we change the parameter t to 7 (this is not the torsion of the curve)
so that
7 = tlt*, (9.33)
and we get an intermediate expression:
(9 .34)
Here, we introduce a shift operator Ea for the control points pa i of the divided
first segment. The relat.ion between E and Ea is defined by
(1 - t* + Et*)'p a =
- E'a paa = pa.' (9.35)
Using this and denot.ing the left-hand side of eq. (9.34) by 1'0(7; n), we have an
expreSSlOn
(9.36)
Since 7 takes a value between 0 and 1, a parameter t is used in place of 7 and
the above eq. (9.36) becomes
(9.37)
This is the equat.ion of t.he first part of the divided curves.

Next, we treat the second pa.rt. We use eq. (9 .19) in which the la.st control
poin t appears.
(9.38)
Changing the parameter t to 7, which is given by
7 = (t - t*)/(1 - t*), (9.39)

9.7 Division of a Curve Segment and Its B Polygon 129
Po P3
Figure 9.7 Division of a B polygon
we obtain from eq: (9.38) an intermediate expression:

i'nCr; n) = [{1- (1 - t*)r - t*}E-l + (1- t*)r + t*]nPn
= [{(I - t*)E- l + i*}(1 - r) + rrp n.
Here, we define a new shift operator Eb for the control points pr. The relation
between P~'-i and Pn-i is determined by
{(I - t *)E-l + t*}i p" -= E-b i p"b = P,,-i
b
' (9.40)
Subst.ituting t.his in the right-hand side ofi- n ( r; n) and changing r to t , we obtain
(9.41 )
This is t.he equation of t.he second pa.rt of the divided curves.
A Bezier curve ro(t; n) is divided at t = t* into two B curves r(j(t; n) and
rg(t; n) = i,~(t; n), the locations of whose B points are given respectively by
pi (1- t* + Et*)ipo, (9.42)
P~-i = {(I - t*)E-l + tTPn, for i = 0,··· n. (9.43)
From eq. (9.42) we can see that the location of i-th B point pi of Ba is at a
point. ro(t*; i), whose B polygon vertices are (Po, PI' . ", pJ. In the same way,
the locat.ion of (n- i)-th B point pb n-i of Bb is a point rn(t*; i), whose B polygon
vertices are (Pn-i, Pn-i+l,"', Pn)·
We call this process of producing B polygons Ba and Bb from the polygon
B o the division of a B polygon in the dividing ratio {to : (1 - t·)}.
We explain a graphical method of determining the location of the new B
point.s pai and pbn_i' Refer to Fig. 9.7. .
1. Divide each edge ai of Bo at. p)1 ) in the ratio (t* : 1 - to), which we call
the dividing point of level one. A number in the superscript indicates the
level. The original vertices are denoted by p;O). A case for n = 3 is shown
. h (1) (1) 0)) .
III parent. eses as Po , PI , P2
Po
Figure 9.8 Reconstruction of an original B polygon from its divided one
2. Connect pIll and P\~l by a line segment, which we caU the dividing line of
level one, and divide it at p;2) in the ratio (t* : 1- to). Then p\2) is called a
dividing point. of level two. ( For n = 3, p~2) is determined on the dividing
line p~l)p~l), and p~2) is on p~l)p~l). )
3. Connect p\2) and P;~l by the dividing line of level two and divide it at p;3)
in the ra.tio(t* : 1 - to). (For n = 3, p~3) is the only dividing point of level
three, and this is final.)
4. Repeat this procedure until division of the dividing line of level (n - 1) at
the dividing point p~n) of level n occurs. Then we determine locations of the
control points of Ba = (Po, .. , , p~) and Bb = (pg, ... p~) as follows.
(0) (1) (2) (n-l) (n)
Po Po Po Po Po ,
(n) (n-l) (n-2) (1) (9.44)
Po PI P2 Pn-l p~O).
The location of the dividing point of level n is a point on the original curve:
p~ = pg = p~n) = ro(t*; n) = (1 - t* + Et*tpo.

When one of the divided B polygons Ba or Bb and a dividing ratio (t* : 1 - to)
a.re given, we can construct the other B polygon Bb or Ba and also the original
polygon Bo uniquely. The graphical procedure is just the inverse of the division
procedure of the polygon Bo. For inst.ance, referring to Fig. 9.8,
1. By ext.ending each edge of Ba by (l-t*)jt*times, we determine the locations

piO), pill and pi2).
2. By ext.ending edges p~O)p~l) and p~1)p~2) by (1 - t*)jt* times, we det.ermine
. d
t he 1ocatlOns P2 an P2 ,
(0) (1)
3. Finally, (1 - to) It* times extension of a vector p~O)p~l) fixes the location of
(0)
P3 .
9.8 Continuity Conditions of Connection of B Polygons 131
Thus, all the locations of the polygons Bb and Bo are determined from Ba.
The dividing lines and points of a B polygon are closely connected to those
of a B spline polygon, which is explained in the next chapter.
9.8 Continuity Conditions of Connection of B Polygons

At a dividing point ro(t*; n) of a Bezier curve segment of degree n the condition
of cont.inuit.y is naturally c(n); continuous up to n-th derivatives, for i = 0, ... n,
(djdtrro(t* - 0; n) = (djdtrro(t* + 0; n). (9.45)
If these conditions are to be expressed by the divided segments r~(I; n) and

rg( 0; n), t.he differentiat.ion with respect to t has to be converted to that by t.heir
own normalized parameters T which are defined by equations (9.33) and (9.39) .
Accordingly, when comparing their differentiated values at t.heir junction we have
to consider the scale ratio of the parameters, which is t* : (I-t*). Then eq. (9.45)
becomes
. d .b
d)i a
( dt 1'0
1
t=1 = k'( dt )'1'0 It=o, i = 0, ... n, (9.46)
t*
where k
1 - t*
We call k the scale rat.io of the adjacent. parameters.
If eq. (9.46) holds for i = 0 to i = j, the j + 1 control points from p~ to
P~-f of 1'~(t; n) and t.he j + 1 control points from pg(= p~) to pj of 1'g(t; n) are
constrained. The other control points of 1'~(t; n) and 1'g(t; n) are free. These
constrained cont.rol points can be derived from the division of a B polygon of
degree j in the ratio t* : (1 - to). For instance, in Fig. 9.7, the control points
(pj, pj, P3 = pg, p~, p~) are determined by the division of a B polygon
(p~l) pill p~l)) in t.he ratio t* : (1 - to). This assures the connection of 1'0(1; n)
and rg(l; n) to be C(2).
Though a curve segment 1'o(t; n) is divided into many segments, each of which
has n + 1 control point.s, they are equivalent to the original curve segment so
long as the continuity constraints at their connecting points are not reduced. If
they are reduced by one degree at ea{;h junction, then the amount of freedom
increases, that is, control points which can be moved freely are generated. This
kind of problem is treated in Chap. 10.
9.9 Elevation of Degree of a Curve Segment

Increasing the number of control points of a curve segment without changing its
shape is attained by dividing the curve segment as well as by increasing degree
of the curve. If one wants to get more freedom for control of a shape consisting
Figure 9.9 Degree elevation of B

polygon
of connecting segments, the method by division is accompanied by decrease of

degree of the conthuity condition, but the method of degree elevation increases
the degree of the segment, that is, increases the number of control points.
Referring to Fig. 9.9, let (Po PI P2 P3) be a B polygon of degree three, divide
each of it.s edges int.o four equal sub-edges and let the third dividing point of the
first edge be ql' the second dividing point of the second edge be q2 and the first
dividing point of the third edge be q3' then the polygon (Po ql q2 ~ P3) is of
degree four, but generates a curve of the same shape as that derived from the
original B polygon.
We prove the above rule in general form. Multiplying the right-hand side
of eq. (9.17) by {(I - t) + t}, we make its degree increase apparently by one
and denot.e the left-hand side by fo(t; n + 1), which has new control points
(qo' ql' q2' ... qn' qn+l) with their new shift operator E. Then this new expression
and its expansion are given by
n+1
fo(t; n + 1) = (1- t + Et)n+Iqo = "L)C;+1)(I- t)n+l-iti qi . (9.47)
i=O
To obtain t.he relation bet.ween point.s Pi and qi, we expand {(1-t) +t} x ro(t; n)
in terms of {(I - t)n+l-it i } for i = 0··· (n + 1). Then it becomes
{(I - t) + t} x ro(t; n)
n n
= ~]C;)(I- tt-i+ICPi + 2:JC;)(I- t)n-iti+I Pi
i=o i=O
n+1
= ~]C;+1)(l - t)n+l-iti{(C;)/(C;+1)PiO~+I,i
i=O
(9.4S)
where
0' . = {O for i = j (9.49)
t,). 1 for i -I j
9.10 Expression for a Surface Patch 133
fl'
p
"
Po P3 Figure 9.10 B polygon by repeated
degree elevations and a curve segment
Comparing equations (9.47) and (9.48), we get the locations of the new control
points in terms of the old control points:
qi =
{ n n++1 -1 i} Pi 8'n+l,i + { i} 8'
n + 1 Pi-l O,i ' (9.50)
The relation is that both of the first control points Po and qo are the same
point., ql is located at the dividing point of the first edge ao of the original
polygon in the rat.io n : 1, qi is located at the dividing point of the i-th edge ai
in t.he ratio (n + 1 - i) : i and the last point q,,+l coincides with the last point
p" of t.he original polygon.
By repeat.ing this degree increasing process indefinitely, the corresponding
B polygon gradually approaches the shape of the curve segment it defines. See
Fig. 9.10.
9.10 Expression for a Surface Patch

We introduce an expression for a Bezier surface patch in an operator form by
extending that of a B curve. Instead of a B polygon, a control point net is used
to define a four-sided surface called a. Bezier patch, B patch for brevity. A B
patch, whose parameters are u and v and degrees are m and n, is denoted by
soo(u, v; m, n) and its control point net is given by an m x n matrix {Pi;l:
POO POI PO,n-l PO,n

PIO PH Pl,n-l Pl,n
(9.51)
Pm-l,O Pm-I,! Pm-l,n-l Pm-l,n
Pm,O Pm,l Pm,,,-l Pm,,,
To deduce a.n operator expression for soo(u,v;m,n), we use a tensor product

method. Considering each column of (9.51) as an independent B polygon of
degree m, we have (n + 1) B curves
(9.52)
For fixed u, we make a row of (n + 1) consecutive points from eq. (9.52):
(9.53)
Considering this 1 x (n + 1) array as a B polygon of degree n, we construct a B

curve of degree n with a parameter v,
(1 - v + Fvtro(u; m), (9.54)
where F is the shift operator used for the elements of the row (9.53), which
are expressed by .eq. (9.52), and operates on the second subscript of the control
points Pij' The role of F is just the same as that of E for the first subscript.
Since eq. (9.54) is a function of parameters u and v, whose degrees are m and
n, and represents a patch, we denote it by SOO( u, v; m, n). Substituting eq. (9.52)
into eq. (9.54) we obtain
SOO( u, v; m, n) = (1 - v + Fvt(1- u + Eu)mpoo

= (1 - u + Eu)m(l - v + Fvtpoo, 0 ~ u, v ~ 1. (9.55)
This is an operator expression for a Bezier surface patch constructed from the
(m + 1) x (n + 1) control points net {PiJ, for i = 0··· n, j = 0··· m. An
example of Soo (u, v; 3, 3) is shown in Fig. 9.1l.
Since the equation of a B patch has a similar structure to that of a. B curve,
it has similar characteristics:
SOO( u, v; 0, 0) Poo, (9.56)

soo(u,v; 1, 1) = (1- u + Eu)(l - v + Fv)poo (9.57)
= (1- u)(l - v)poo + (1- U)VPOI + u(1- V)PI0 + UVPll,
soo(u,v;2,2) = (l-u+Eu?(1-v+Fv?poo (9.58)
(1 - u + Eu)(l - v + Fv)soo(u, v; 1, 1)
= (1 - u)(l - v)soo(u, v; 1, 1) + (1 - U)VSOl(U, v; 1, 1)
+ u(l- V)SlO(U, v; 1, 1) + UVSll(U, v; 1, 1).
Equation (9.57) is a surface patch of degree 1 x 1, whose control points are at the
four corners of a general twisted quadrilateral. A surface patch of degree 2 x 2
is equivalent to a special surface patch of degree 1 x 1, whose four control points
are on the four patches Soo, SOl, S10 and Sl1 of degree 1 x 1 respectively. Similarly
a Bezier surface patch of degree m x n is made from a patch of degree 1 x 1,
whose four control points are replaced by expressions for four patches Soo, SOl, SlO
9.10 Expression for a Surface Patch 135
P30
p,o,2:c~~~ - '... . .. l?::. .",

... P31
/ .",/ PH "" ';. :'

""" "',:,:-:;:/P2;"'""" , "r P33
,/ P12\, ""
, , ,,/ \" ',,~
.. ---_ __ ,/ , P23
PO~ "''',- . , : .. ,,: /
POl " ..:: \ ...
'~/
P02"" , ./ P13
\ . /
P03
(a) control net
(b) B patch
Figure 9.11 B patch and its controinet
and S11 of degree (m - 1) x (n - 1), The general form of eq, (9.57) is
SOO( 7.1, v ; m, n) = (1 - 7.1)( 1 - v)soo( 7.1, v; m - 1, n - 1)

+(1 - U)VSOl(U, v; m - 1, n - 1)
+ 7.1.(1 - V)SlO(U, v; m - 1, n - 1)
+ UVS11 (7.1, v; m -1, n - 1), (9.59)
The expression for a B patch in operator form is concise and useful for
theoretical manipulation, but if we want to write all the control points explicitly,
for instance in numerical calculation, then the following matrix form is used
(refer to Sect. 1.8, Example 2):
800(U , v; m , n) = F(u; m){PiJ}F(v; n)T, (9.60)
where F(u;m) {fo(u;m) fl(u;m)···fm(u;m)}
F(v; n) {fo( v;n) fl(v;n)"'fn( v; n)}
fi(t; m) = (9.61)
9.11 Geometric Properties of a Patch

One of t.he four boundaries of 800(U , v; m, n) is obtained by setting one of the
parameter values u = 0 or 1, or v = 0 or 1, for example,
800(0, v; m, n) = (1 - v + Fv)npoo (9.62)
This is t.he boundary curve of u = O. Let the edges of the B polygon of this
curve be
(boo bOl ... bo,n-l) ,
and similarly, let. the edges of the B polygon of the boundary curve v = 0 be
(aoo alO ... an-l,o) .
The i-th derivative of the surface patch in the direction u along the boundary
curve u = 0 is given by
(:)i 800 ( u, v; m, n) 1,,=0= {(m7: i)! HI - v + Fv)n(E - l)ipoo' (9.63)
For i = 1, t.his is the tangent vector across the boundary u = 0, which we call
the cross-bounda.ry ta.ng ent vector T".
t
T" = m( 1 - v + Fv aoo . (9.64)
Similarly, we get. t.he cross-boundary tangent vector at the boundary v = 0,
Tv = n(1 - u + Eu)mb oo . (9.65)
The normal of t.he surface patch at the corner point PO~ is
aoo x boo
n - .,..-----,- (9.66)
- laoo x bool '
The cross-boundary t.angent vectors T" and T v are not constant along the bound-
aries and their rates of variat.ion along the boundaries at t.he corner point Poo
are given by
a~" Iv=o = mn(aOI - aoo) = mn{(pll - POl) - (PIO - Poo)} , (9.67)
aJuv 1,,=0 mn(blO - boo) = mn{(Pll - PIO) - (POI - Poo)}. (9.68)

9.12 Division and Degree Elevation of a Patch 137
They are equal to 02S00(0, OJ m, n)/ouov and 02soo(0, OJ m, n)/ovou and are
called twist vectors at the corner, and take the same value when the corner point
is not singular.
As stated in the surface theory, the fundamental magnitudes of the first and
the second order at Poo are given by
E T~ = m2~0, F = Ttl' T v = mnaooboo ,
. 02soo(0, OJ m, n) _ ( _ )d
L n ou 2 - m m 1 02, (9 .70)
. 02soo(0, OJ m, n) _ d
M n OUOV - mn '11, (9.71)
02soo (0,Ojm,n) ( )d
N n· ov 2 = n n - 1 20, (9.72)
where d02 , dl l and d20 are distances from P02' P11 and P20 to the tangential plane
at Poo' The Gaussia'n curvature, t.he mean curvature and the principal directions
can be calculated from the locat.ions of the control points. As in the case of
B curves, any point on a B surface patch can be made a corner point of a.
divided pat.chj the control points around it. are determined and the fundamental
magnitudes at that point can be obt.ained accordingly.
As in the case of B curve, either of derivatives at t.he boundary or the cont.rol
point.s of the B patch can be expressed by the others. For instance, Pn is
expressed by the derivatives at the corner,
Pn EFpoo = {((E - 1) + l)((F - 1) + l)}poo

{(l/mn)stLV + (l/m)su + (l/n)sv + s} lu=o,v=o . (9 .73)
9.12 Division and Degree Elevation of a Patch

Since each control polygon corresponding to a column of the cont.rol point matrix
{PiJ} can be divided at u = u* and becomes two control polygons of the same
degree connected in c(m) without changing the shape of the curves generated, the
matrix {Pij} is changed into two matrices {pij} and {prJ of the same magnitude ,
which generate t.wo patches s~o and s~o according to the division of soo( u, Vj m, n)
at u = u*. Similarly two matrices {piJ} and {prj} can be divided at v = v* into
four matrices {pin, {pi!}, {ptr} and {prj} · Their corresponding patches divide
soo(u, Vj m , n) at u = u* and v = v* .
Each control point. in the above matrices is derived from the original control
points , such as
pi/ (1- u* + Eu*r(1- v* + Fv*)1poo, (9.74)

be
P(m-ilJ {(I - u*)E- 1 + u*)}i(l - v* + Fv*)1Pmo, (9.75)
ad
Pi(n-Jl (1- u* + Eu*n(1- v*)F- 1 + v*}lPon ' (9 .76)
bd
P(rn,-i)(n-J) = {(I - u*)E- 1 + u*r{(l - V*)F-l + v*}J Pmn . (9.77)
They have also tensor product forms.

A surface patch of (m + 1) x (n + 1) degree, which is the same shape as the
patch of degree m x n, is obtained by the same degree increasing process as that
adopted in a curve. Its new control point.s qij are given by the tensor product
form of eq. (9.50),
qi) = {1/(m + 1)}{I/(n + 1)}{(m + 1 - i)b;"+l,i+l + ib~,iE-l}

x{(n + 1- j)8~+1,)+1 + jbb,jF-1}Pij' (9.78)
where Pij is t.he cont.rol point of the original pa.tch, and bi,j = 0 for i = j,
ot.herwise 8L
= 1, which is the same as eq. (9.49). The control points at patch
corners before and after t.he degree elevation are the same, but their indices are
different:
qoo = Poo, . qm+l,O = Pm,O' qO,n+l = PO,n' q",+l,n+l = Pm,n' (9.79)
9.13 Appendix. The Original Form of the Bezier Curve

The original form of t.he expression for t.he Bezier curve [3]' which Bezier gave
using polygon edges (aI , a2, ... an), was
(9.80)
This is a very difficult. expression, from which R. Forrester deduced the Bernstein
polynomial form (9.20) in 1970 [3]' The form (9.17) was introduced by M. Hosaka
in 1976 [5] .
Equation (9.80) can be obtained by conversion from eq. (9.17) as follows: by
putting
al = (E - l)po, (1 - E)t == x, (9.81)
we get from eq. (9.17)
rO(t; n) - Po = {(1- t + Ett - l}po = -t[{(I- xt - 1}/x]al'
Inside [ 1in t.he above equat.ion is a polynomial of degree (n - 1) in x, hence from

eq. (9.81) it is a polynomial of degree (n - 1) in E. So its Taylor expansion in E
gives
Since from eq. (9.81) x becomes equal to t for E = 0, we have
(9 .82)
9.13 Appendix. The Original Form of the Bezier Curve 139
And since (dx/dE)i = (-t)i, eq. (9.82) becomes
n-l d . (1 - W- 1 Eial
ro(t;n)-po=-tL:(-d)'{ }(-t)'-.-, . (9.83)
i=O t t 2.
Expressing Eial as ai+1 and changing the summation range from (0 to n-1)
to (1 to n), we obtain the equation (9.80) which Bezier gave without explanation.
10 Connection of Bezier Curves and Relation
to Spline Polygons
10.1 Introduction
In design of free-form shapes, use of low-degree curves is desirable rather than

adopting higher-degree ones or other complex expressions, so long as the objective
of the design is attained by connecting low-degree curves. Curves of low degree
are easy in analysis, synthesis and calculation of their geometry and also in
detailed control of ~esigning shapes, though there are some problems with their
connections.
Bezier curve segment.s have favorable features for CAD application as we have
explained in the previous chapter. But when a curve of complex shape is required
to be represented by a single Bezier curve segment, it becomes necessarily of high
degree and moreover local shape control by its polygon is difficult, because of the
global nature of t.he influence of each of its control points. So, if we can connect
B polygons with appropriate constraints which meet a slightly lower continuity
condition at each junction than that expected from a single segment, we will
be able to limit the influence of one segment over others. And if we can find
a g~ometric relat.ion between these constraints, we will be able to control local
shape without violat.ing the continuity conditions and can synthesize a favorable
shape.
The vector representation of a basis spline curve, which is usually called B
spline (refer to Sect. 11.7), has the characterist.ics st.at.ed above. In ordinary text-
books, t.he basis spline functions are introduced mathematically at first [22][33].
Hence, some amount of mathematical preparation and background are needed.
Moreover numerical calculation of the basis spline curves and surfaces, due to
t.heir recurrence formula, is far more complicated when calculation of interfer-
ence or evaluation of shapes is required, compared to that of Bezier curves and
surfaces which have simple analyt.ical expressions as well as a simple graphical
interpretation.
Since a basis spline curve consists of piecewise polynomial curves, each curve
segment can be expressed as a Bezier curve which is defined by a cont.rol polygon.
So in this book, we take a different approach for synthesizing a curve or a surface,
in which they are composed of Bezier curves or surface patches of degree n
connected in c(n-l) [12]. For automat.ic adjustment of control points of adjacent
Bezier curves to satisfy the connecting condition, we introduce constraints by a
polygon with a scale ratio list., which determines connections a.nd scale rat.ios of
ta.ngent vectors at junctions of co~necting Bezier curves. We call this polygon
142 10 Connection of Bezier Curves and Relation to Spline Polygons
an S polygon (S stands for spline), from which B polygons (B stands for Bezier)
connected in c(n-l) are deduced.
We give formulas for obtaining B points from the S polygon and also the
method of inserting vertices of the S polygon without changing the shape of its
generating curve. The reason for adopting B polygons is that they are geomet-
rically clear and simple and their expressions are easy to manipulate compared
with the recursive formulas used in B splines. Especially in advanced design
activities, calculation problems such as interference and quality evaluation of
surfaces require analytical expressions of shapes, so simple closed forms of Bezier
type are suitable.
The description of spline curves and surfaces is divided into Chaps. 10 and 11.
In this chapter, we first treat a constraint on the connection of two Bezier curves
and then introduce the S polygon from which a row of B polygons connected
in c(n-l) can be generated. Omitting the theoretical aspects of the S polygon,
which are treated In the next chapter, we explain practical methods of locating
B polygons from a given S polygon and of inserting new vertices into it.
Then we loosen the connecting condition, from C(2), which requires conti-
nuity up to the second derivative vectors at the junction, to G(2) which satisfies
continuity only of curv,a ture vectors there. This allows a little freedom to move
locations of the B points without changing the S polygon of degree three. The
spline curve satisfying this condition is usua11y called the Beta spline, but the
method of deriving it adopted in this cha,pter is far simpler and the result is easy
to use.
In the last part of this chapter, we treat a problem of effects of scale ratios
on-generated shapes and that of control of curvature distribution along a curve.
The reader with good knowledge of Bezier curves and surfa,ces, who just
wants to make shapes by connecting them, need not know the theoretical aspects
treated in the next chapter, but may refer to Sects. 10.3 and 10.4 for curve
synthesis and Sect. 11.1 for surface synthesis.
10.2 Connection of B Curve Segments
10.2.1 Scale Ratios
For various curve segments of parametric expressions, there may be great dif-
ferences in increments of position vectors for the same incremental value of the
respective parameters, because usually the parameters are normalized between
o and 1 irrespective of the real magnitude of the objects they represent. This
does not cause a problem so long as curve segments are independent. But when
connecting them, we need to take account of these differences of scale for each
segment, because their derivatives with respect to the parameter cannot be com-
pared with each other without correction for their scale differences. Accordingly,
we introduce a list of ratios of the scale factors for connecting segments. For
convenience, we consider that each constituent segment has its 'own scale factor
A.
Let there be a row of Bezier curve segments of degree n, each connected in
C(i) .
. . . , L1(t;n) , ro(t;n) , r1(t;n), ....

Instead of a curve segment rj(t; n), we treat its B polygon B;n):
(Pjn, Pjn+l, ... , Pjn+n), (10.1 )

or (ajn' ajn+1,···, ajn+n-l) (10.2)
where Pjn and a}n indicate the first control point and the first control edge of
the control polygon BJn ). Its superscript indicates the degree of the curve, and
if this is clear it may be omitted. Two curve segments rj_1(t; n) and rj(t; n) are
connected continubusly up to the i-th derivative at their ends.
Though we use the same symbol t for the parameter of each curve segment,
they are substantially independent each other. However, when two curves are
connected smoothly, it is natural to consider that the element ds of arc length
of both segments t.ake the same value at the junction. Therefore, we postulate
(refer to Sect. 5.2) a condition, setting j = 0 for simplicity:
ds = IL1(1; n)l(dt)-l = Iro(O; n)l(dt)o, ( 10.3)
where (dt)_l and (dt)o indicate that they belong to L1(t; n) and ro(t; n) at the
junction respectively.
Introducing constants A-I and AO which are proportional to the magnitudes
of tangent vectors i·_ 1(1; n) and ro(O; n) at the junction, we define the ratio
ko = A-dAo and obtain the following relation from eq. (10.3):
(10.4)
This ratio is considered the scale factor ratio of the parameters at the junc-
tion. For connecting curve segments ( ... , r -1, ro, r1, r2, ... ), a list of scale factors
(... , A-I, Ao , AI, A2,···) must be provided to ensure a natural shape. When curve
segments are of almost equal magnitude, all factors Ai may well take the value
1.
10.2.2 Conditions of C(i) Connection

If we interpret the connection of continuity C(i) geometrically, the following con-
ditions must hold at the junction of two curve segments r_1(t; n) and ro(t; n):
(10.5)
Substituting eq. (10.4) into eq. (10.5), we rewrite eq. (10.5) as

d d
(d/L1(t;n) It=l= kMd/ro(t;n) It=o. (10.6)
Let the two connected curve segments Ll(t; n) and ro(t; n) be represented by
Bezier curves in the operator forms(refer to equations (9.20) and (9.17)):
L l(t; n) = {(1- t)E- 1 + t}npo, (10.7)

ro(t;n) = (l-t+EttPo· (10.8)
Substituting these into eq. (10.6), we have, for j = 0, .. . i,
(10.9)
This is the condit.ion of C(i).

For example,
j=O Po = Po,
j =1 Po - P-l = kO(Pl - Po) or a_I = koao,
j=2 (Po - P-l) - (P-l - P-2) = k~{P2 - PI) - (PI - Po},
or a_I - a_2 = k~(al - ao) (10.10)
and so on.
These relations are the geometric constraints, as shown in Fig. 10.1 for C(1),
C(2), C(3) and C(4). The important fact is that in the C(i) connection, on both
sides of the junction the i consecutive control point.s of one segment are com-
pletely determined by the i consecutive control points of the other segment. This
sequence of 2i + 1 control points defines a B polygon of degree i. Application of
this fact is explained in the following subsection.
10.2.3 c(n-l) Connection and Control Points

Let t.here be c(n-l) connected curve segments of degree n. We examine the
influence of displacement of one control point of one of its constituent curve
segments on the shape of the whole curve.
For C(i) connection, there are i + 1 equations (10.9) for the i + 1 consecutive
control points of both sides of one junction. Accordingly, for c(n) connection
of these segments, all the n + 1 control points of one segment are completely
determined by the n + 1 control points of the adjacent segment. Consequently,
displacement of one control point of a segment influences all the control points
of the other segments connected one after another to keep the c(n) condition at
all the junctions. So, there is no freedom of local deformation of shape of the
curve.
But for c(n-l) connection of n+1 consecutive segments, the number of equa-
tions (10.9) at all the n junctions is"equal to n 2 while the number of control points
Po
.0
.~~-.:.-I_-:&\',,-0-~PJ. P-I.... ··/ Po ·'' ·''Pl

C(I) P-I , C(2) PI
.... P-2
P2\\.,
\:,-3 P~,p
;'P- 4
(a) (b)
0-- ---- ... .. _- .... :::0.-_ . • .. ---_ .. _.. _____ ~., 1."'•.. --- ... -- -- -- -- -:.G.-: -- ..... - --~
1'-1, / Po " """ PI
P""-,,,I_- .....
; _(:-) _~PI '. :,: ,\
P-2 P2 /'
1'-3 P3 '
, 1'-3
0 1'_4 1'-4
(c) (d)
Figure 10.1 Constraints of control points on both sides of a junction of two B polygons
of all the n + 1 segments is equal to (n + 1)2, so their difference of 2n + 1 control

points are free. Therefore, then + 1 control points of the first segment and the
n last control points of the (n + 1)-th segment can be considered independent.
Accordingly, displacement of the control points of the first segment does not
influence the (n+2)-th segment or further ones. For instance, let Bo B} B2 B 3 ,'"
be Bezier control polygons of curve segments of degree three and let them connect
in C(2). See Fig. 10.2. Then any displacement of control points of Bo does not
influence the location of the last three control points of B 2, while those of Bo
and B} have to be adjusted to keep the C(2) condition at the junctions, and B3
does not change at all. We are to find a method of automatic adjustment of
the control points of Bo , B} and B2 to keep them in C(2) connection. Thus the
c(n-l) connection enables local control of shape of connected curve segments of
degree n.
146 10 Connection of Bezier Curves and Relation t o Spline Polygons
• e e e e e o •
Bo Bl B2
( a)
Y',
: "
!BO "
Bl
B2
(b)
R
:
/\
,
\
\
I.,
Bo
B2
(c)
Figure 10.2 Propagation of displacement over other segments connected in C(2)
10.2.4 Connection Defining Polygon

We introduce a special B polygon which stipulates the condition of connection
of two B polygons. We call it the Conn ection Defining Polygon (CDP), whose
properties are explained in this section.
Referring to Fig.10.1(b), for connection of Bi!{ = (P-3, P-2, P-l, Po) and
B~3) = (PO , P1,P2,P3) in C(2), their five control points from P-2 to P2 have the
two relations of eq. (10.9) for j = 1 and j = 2, so two of the five control points
are determined from the other three. Or there exist s a B polygon iJ~;,o which
has three control points:
(P-l , Po, Pl)(2), (10.11)
division of which in the ratio P'-1 : >'0) produces two B polygons
(10.12)
Their control points are from P_2 -to P2'

We call this B polygon B~; 0 the connection defining polygon or the CDP of
(3) (3) , - (2). . ' (2)
B_1 and Bo , because the polygon B_ I0 determmes the connectlOn of B_1 and
B'(2).
om.
C(2)
. ,(2) . (3)
The control pomts of B_1 are the same as those of the control pomts of B_1
,(2) (3)
except. for t.he left.most P-3' and those of Bo are the same as those of Bo except
for the rightmost. P3' So we use notations • B~i and B O
(3) in place of B~; and
B~2) when emphasizing their missing leftmost or rightmost control point. We
call these B polygons the left-end or right-end incomplete B polygons.
Refer to Sects. 9.7 and 9.S. Since these control points are determined by the
division of B~; 0 in the ratio (A-I: AO)' their locations are given, for 0 5 j 5 2,
by
(2)
P-2+j = + EtFp~i ,
(1 - t (10.13)
p(2)
2-J
{(1- t)E- l + t}jpi2 ), (10.14)
A-I ko
where t + (10.15)
A-I + AO = 1 k o'
According to the dividing procedure for a B polygon explained in Sect. 9.7, it is

understood that each edge of the CDP is considered to correspond to (A-I + AO),
each edge of it.s divided polygon • B~{ to >--1 and that of BO(3) to AO'
The above stated relations in C(2) connection can be extended to those in
c(n-l) connection. For the connection of two B polygons B~"l and B6n) in c(n-l)
wifh the scale ratio (A-I: AO), there is a CDP B~I-;)' each edge of which cor-
responds to (A-l + AO). By division of the CDP i'n the ratio (A-I: Ao), the
incomplete B polygons· B<::l and Bo(n) are obtained, where· B<::l indicates the
cont.rol point.s of B~~) with its first one missing and Bo(n) the control points of
B~n) with its last one missing.
If there exists a series of CDPs of degree (n - 1):
.·.B_I0, BOI, B12, B 23, ··· , (10.16)
by dividing each CDP of (10.16) in the ratio
we obtain a series of incomplete B polygons:
By removing duplicate B points from the above series, we obtain a row of B

polygons of degree n connected in -c(n-l):
(10.17)
whose scale factors are

(10.18)
If we can make a polygon which controls the disposition of a series of CDPs
and directly gives a series of B polygons (10.17), we are freed from problems of
individually connect.ing B polygons in c(n-l), yet may utilize favorable charac-
terist.ics of Bezier expressions. We introduce such a polygon in the next section,
and call it the 5 polygon for brevity.
10.3 Introduction of S Polygon

10.3.1 Locating B Points from an 5 Polygon
Since an 5 polygon stipulates the disposition of CDPs, we can provide methods
of directly det.ermining its constituent B polygons without considering each CDP
at a junction. We explain such practical methods in this section.
Since two adjacent CDPs B_ 1 •o and BO.l are resolved into (* B_IB~) and
(* BoBn, from their common part we can extract a (bot.h-ends) truncated B
polygon * B~. Similarly we have adjacent truncated B polygons * B:' l and * B~.
From the condition (10.10) for j = 1, the location of their common control point
Po is the dividing point between P-l and Po in the ratio (,\-1 : AD) and that of P n
is the dividing point between Pn-l and Pn+l in the ratio (AD: AI)' Accordingly, if
we can determine all the both-end-truncated B polygons, their missing common
B points can be supplied easily.
Now we int.roduce a. polygon 5 which constrains the row of CDPs given by
eq. '(10.16) with t.he list of scale factors. Let the vertices of an 5 polygon be
denoted by
(10.19)
and it.s edges by
(10 .20)
Each of it.s edges b j is divided into n sub-edges whose dividing ratio is equal to
the ratio of scale factors of n consecutive B polygons (10.17). The dividing ratio
list for an edge b J is given by a list. of n elements A denoted by
(10.21)
Accordingly, each sub-edge has its own scale factor. The value j of the index of
its last element of Ljnl is the same as t.hat. of the edge b j to be divided . This
rule is det.ermined logically; the reason is explained in Sect. 11.3. When terms
a. 'scale list' or a 'A' list are used, they simply refer to a list such as eq.(10.18). A
set of consecutive n elements of a scale list belonging to an 5 polygon of degree
n becomes a dividing ratio list of an edge of the polygon.
Next, a unit 5 polygon of degree n, which is denoted by S1 n ), is defined as
the n consecutive edges starting at the vertex qi of an 5 polygon. Since all the
10.3 Introduction of S Polygon 149
dividing rat.io lists of Sr n) are (L;n), ... , L;~~_l)' the scale factor Ai is the only
common element. in them. Accordingly, there is only one sub-edge, which has
the same common scale factor Ai, in each edge of a unit S polygon Sr n ). We call
the common scale factor the main scale factor of the unit S polygon.
Leaving t.he t.heoretical aspect of the S polygon to Sects. 11.3- 11.5, we give
its features useful for derivation of B polygons.
1. A unit S polygon sf generates three successive B polygons denoted by
n)
(* B i - 1 Bi B;+l)(n): the central one, whose scale is \, is complete and both

side ones are incomplete B polygons of degree n with scale factors Ai-1 and
\+1 respectively.
2. An incomplete S polygon s;(n) which misses the right-end edge of Sr n ) gen-
erates a CDP 13;-1,i' and its adjacent incomplete S polygon * sIn) which
misses the left-end edge of S}n) generates Bi ,i+1'
3. From a both-end-truncated S polygon * s;(n), which misses both end edges of
sf n ), an S polygon of degree (n - 2) with its main scale Ai which is the same
as that. of S1 n ), can be deduced. The method is explained in Chap. 11. It

is called t.he reduced-truncated S polygon of sfnl,
for which we use a symbol
S~n-2). The dividing ratio list of its first edge is denoted by
L;n-2) = (,\(i-n+3),"', Ai), (10.22)
instead of L;n-2) for clarity. The reduced-truncated S polygon finally pro-

duces a truncated B polygon * B;(n).
4. Location of a common control point Pn(i+1) of two adjacent B polygons
Br n ) and B~:~l) whose scales are Ai and \+1 is determined at the (\ : Ai+1)
dividing point of the distance between the end control points P{n(i+1)-1} and
P{n(i+1l+1} of the adjacent truncated B polygons * B;(n) and * B(i+1)(n).
Applying the above properties, especially (3), we can determine locations of

B polygons connect.ed in c(n-1) from a given S polygon. We give examples. See
Fig. 10.3.
n = 1 : By the definition the following relations are determined.
S6 1 ) = (QOql) = b o, L~1) = (AO)
So*(1) -- 11_
'W'
*5,0 (1 ) -- Q1, *S*(l)
0
-- null .
Each CDP coincides with each vertex of the S polygon, because the in-
complete S polygons coincide with the vertices. Hence, each edge of S is
a B polygon of degree one, which connects with adjacent B polygons at
the vertices.
(1 ) (-1) (1)
= (POP1) = b o = (QOQ1), = null, = null.
v
Bo So • B;
Since the dividing ratio list is L~1) = (AD) and the number of its element
is one, there are no dividing points on the edges.
PI
5(1).
o .
B (l).
o .
Po
.~---------- i>
ql
q••o'-----_ _ _ _ _ _ _ _ _ _~
•
PI
q,
~
S~')
S(o) .
, , 'b
•
B P2
o ) :.
(a) n =1 (b) n = 2
q2
q1.2 q2.2
q3 ql • 5;;(4) q3
S~l) : '5(2) .
o . 6---'
B~3) : •• • B;;(4) :
(c) n = 3 (d) n = 4
v (3)
So : Br--~Dr--_BO--D
* B *(5) .
o .
(e)n = 5
Figure 10.3 Locations of B points for n =1, ... 5

10.3 Introduction of S Polygon 151
n=2 The following relations are easily obtained.

S~2) (qo ql q2) = (b o b 1 ), L~2)
S;(2) (qo q1), * S~2) = (q1 q2), *5*(2) =
o
v(O) v(O)
50 Lo
* B*(2) * B* (2)
o -1
* B;(2) = P3 = Q2,
Since the reduced-truncated 5 polygon is a vertex of an S polygon, the

truncated B polygon becomes the same point. The common B control
point Po between B1 and Bo is located at the dividing point between qo
and q1 in the ratio (A_I: Ao). Similarly the common B control point P2
between Bo and B1 is located at the dividing point between q1 and ~
in the ratio (-\0 : AI)' We denote their locations symbolically as qO.1 and
q1.1' .
n=3 By the definition we have
3) 56
(~ q1 q2 q3) = (b o b 1 b 2), L~3)
v (1)
(q1 q2) = b 1, Lo
Its reduced-truncated 5 polygon S~l) is necessarily the Ao sub-edge of b 1
which is * B;(3) = (PI P2)' Accordingly, locations of PI and P2 are the
Li
dividing points of the edge b 1 with the ratio 3 ) = (A-I Ao AI)' Similarly,
locations of P-2 and P-1 and those of P4 and Ps are determined at the
dividing points of b o and b 2 by the ratios L~3) and L~3) respectively. The
location of Po is the (A-I: Ao) dividing point between locations P-1 and
PI ' and the locat.ion ofp3 is the (Ao : AI) dividing point ofP2 and P4' We
denote the locations of the control points (Po PI P2 P3) of B~3) as
(10.23)
These symbolic notation for locations of the dividing points are explained
in Sect. 11.4, but they are not needed in this chapter.
n =4 :
(Ao Al A2 A3)
(,\1 Ao).
v(2)
Its reduced-truncated 5 polygon 50 is constructed as follows . Refer to
Fig.10.3(d) . Make two new edges: one by connecting the second dividing
point q1.2 of b 1 to the first dividing point q2.1 of b 2, and the other from
the third dividing point q1.3 of b 1 to the second dividing point q2.2 of
b 2· Let their intersection be q. Then the polygon (q1.2 q q2.2) is the
v (2)
reduced-truncated 5 polygon 50 ,from which the truncated B polygon
• BO(4) = (PI P2 P3) is determined by the procedure for n = 2. In the lower
part of the figure, relation between S~2) and' BO(4) is shown schematically.
AO
Ao
Figure 10.4 Relation between loca-

tions of old vertices (qj' q2) and those
of new vertices (qa, qj (2)
n=5
Make a reduced-truncated 5 polygon S of degree three, whose vertices

are produced from the dividing points (indicated by small blank squares
in Fig.10.3(e)) of the edges of the truncated 5 polygon of degree five.
• h LV(3)
From 5v (3)
0 WIt 0 an d usmg
. t h e proced ure for n = 3,we d etermllle
.
four B points indicated by small black circles, which are the locations of
* B O(5) = (PI P2 P3 P4)' In the lower part of the figure, relation between
v(3)
So and * Bo (5) .
IS shown.
The above st.ated met.hods can be ext.ended to degree n: from an 5 polygon
and it.s edge dividing ratio lists we can obtain its reduced-t.runcated polygon S
of degree (n - 2). Repeating the degree reduction process, we can determine
locations of t.he B polygons of degree n from the lower-degree 5 polygons. B
polygons thus obt.ained from an 5 polygon of degree n are connected in e(n-l).
10.3.2 Increase of Vertices of an 5 Polygon

56
From an n ) polygon with a scale list ("', "-1, AO AI,"') we obt.ain a series of
B polygons connected in e(n-l):
( . .. , B(n)
-1'
B(n ) B(n ) ... )
0 , 1, ,
whose scale ratios are ( ... : "-1 : Ao : Al : ... ). If we divide B~n) in the ratio
(t : 1- t) int.o two B polygons Ba~n) and Bb~n), we can consider that their scales
are '\ot and Ao(1-t). They connect with their adjacent B polygons B~n2 and Bin)
in e (n-l), because t.he original Bbn ) connects with the same adjacent B polygons
in ern-I).
5b
Now, from the original unit. 5 polygon n ) we produce two new unit 5
polygons 5a~n) and 5b~n) by the following procedure: see Fig. 10.4, which is a
case of n = 3.
5b
In each edge of n ) there is a sub-edges of the dividing ratio AO which is the
main scale of the polygon. We divided each sub-edge into two parts in the ratio
10.4 B points under Geometric Connecting Condition G(2) 153
--- --'--
-- .... ~-------
r: ::~
,./ \
:/ \
~
"-
/ \
: 1- t
~ -- - - -- -- -------- - - - - ---- - - --o Gr- ---0 13--------- ---- ----------------0 <r- ---<>
(a) increase of vertex of S polygon (b) displacement of a vertex and deformation

of curve shape
Figure 10.5 Illserti~n of a vertex
(t : 1 - t), then we have n dividing points (<to .. 'qn-l) ' We replace the
original n - 1 vertices (ql ... qn-l) by these new n vertices and consider there
generate two unit S polygons:
We can prove these two unit S polygons have the main scales >"ot and >"0(1 - t)
respectively and generate the B polygons B,,~n) and Bb~nl, which are the same
B polygons generated from the division of B6n ) in the ratio (t : 1 - t). (This is
treated in Sec. 11.5.4).
These B polygons B,,~n) and Bb~n) connect in c(n) if no displacement of
the vertices occurs. Even when vertices of the new S polygon are moved, the
condition of c(n-l) connection of the generated curve segments is not violated.
An example is shown in Fig. 10.5. We call this process the division of an S
polygon or the insertion of a. vertex.
10.4 B points under Geometric Connecting

Condition G(2)
So far we have treated c (n-l) connection of B polygons of degree n, considering

scale ratios of the parameters at their junctions. Now we are to loosen the
condition of C(2). Instead of continuity of the second derivative vectors at the
junctions, we use the continuity condition of curvature there, which is called the
geometric continuity condition C(2). We restrict the degree of curve segments to
three. This makes the constraints stipulated by the Connection Defining Polygon
/i : .
q2
"
p,
/~2 P3 \\a4
n.a\~
____ ' P5
P I ----
''0 q3
Figure 10.6 Control vectors at ajunc-

tion in G(2) connection
of degree two hold' only in the normal direction at the junction. It follows that
the dividing ratio list of each edge of an S polygon can be adjusted to modify
the shape of connected curve segments without breaking the condition of G(2).
Refer to Fig. 10.6. Let the connecting B polygons of degree three be expressed
by their edges
and t.heir CDP be

- (2)
B O•1 = (PI q2 P5)'
In C(2) connection, division of the CDP in the ratio (Ao : AI) produces two
incomplete B polygons * Bo and B; and the following relations hold:
(10.25)
In G(2) connection, instead of the continuity condition of the second derivatives
at t.he junction, which is given by
( 10.26)
only the continuity of t.he curvatures there is allowed. Referring to the expression
(9 .31) for curvature at the left end of Bo, which is proportional to (n . al)/ii,
we have the equation of equality of the curvatures at the junction Bo and B 1 :
(n·al) (n·~)
(10.27)
This is convert.ed to
(10.28)
by the condition of continuity of the tangent vectors a2 = k1a3. Equation (10.28)
is the normal component of eq. (10.26). Using eq. (10.25), from eq. (10.28), we
have
(10.29)
10.4 B points under Geometric Connecting Condition G(2) 155
Accordingly, in G(2) connection, the tangential component of eq: (10.26) need not
be satisfied, but only its normal component eq. (10.29) is the condition. Here we
introduce an arbitrary positive constant 0", which we call the stretch factor, in
the above equation thus:
(10.30)
Since the left-hand side of eq. (10.29) represents the normal component of q2P2
and the right-hand side that of Q2P4' they are naturaUy equal. Equation (10.30)
now shows that they can be modified to
Q2P2 al (O"),d ),0)
~P4 ~(0"),0/),1) '
This means that locations of the control points on the edges b 1 and b 2 can be
moved along the r'espective edges. If 0" = 1, t.he connection is C(2), but if 0" is an
arbitrary positive constant, C(2) reduces to G(2).
To indicate that 0" is used at the junction of two B polygons Bo and Bl
connecting in G(2J, we attach a subscript to the strtch factor 0" such as 0"0. For
connection of Bi and B i+1 we use O"i . Then the dividing ratio list of an edge ai
is modified from L;3 ) for C(2) to L,;3) for G(2):
L,;3) = (O"i-l),i-2, ),i-1> ),iO"i). (10.31)

When applying G(2 ) connection, notice that L;2) = (),i-I,),;) does not change.
Wf; can consider that a stretch factor O"i belongs to a vertex Qi+l of an S polygon
of degree three . We can determine the locations of B polygons connecting in
G(2) using L,;3) and L';~1 for the cont.rol points on b i and on b i +! instead of L;3)
(3)
and L i+1 •
For example, consider the effects of changing values of 0" on the shape of
connected curve segments. In Fig.10.7(a) where all 0" = 1, the variation of
radius of curvature (relative values) is shown along the curve. Its profile shows
moderate shape change. In (b) where all 0" = 0.3, the curve shape approaches
that of the S polygon. The radius of curvature becomes small at the junctions.
In (c) where aU 0" = 1.3, the curve segments approach a flat shape at their
junctions. The effect of 0" is to stretch the shape of connecting curves around
their junction it belongs to.
In t.he literature [11][22], paramet.ers ca.lled t.ension (,82) and bias (,81) are in-
cluded in coefficients of the polynomial expressing a curve segment which con-
nects in G (2). The curves including these parameters are called Beta splines. In
deriving the expressions for Beta splines, a.utomatic algebraic symbol manipula-
tion by a computer was used. The bias corresponds to ), in this section, but the
tension is a complicated function of geometric entities and is taken as a constant
at aU junctions. When it varies at each junction, the expression of the curve
segment becomes a complex rational representation of high degree. Compared
(a)allu=l (b) all u = 0.3
( c) all u = 1.3
Figure 10.7 Effects of values of u on shapes of connected curves. Patterns of radius of

curvature are also shown
with the Beta spline, our expression of curves with G(2) connection is simple,
that. is, t.hey are connected Bezier polygons of degree t.hree, whose control point
locations are easily determined from the S polygon with the modified dividing
ratio lists of degree t.hree. Effects of values of the stretch factor (J on the location
of t.he control points are easily determined even without a computer.
10.5 Curvature Profile Problem in Design

10.5.1 Geometric Interpretation of Dividing Ratios
We explain in this section the meaning of the dividing ratio list and a practical
method to determine the values in the list. We defined ratio of scale factors), in
Sect. 10.2 as
( 10.32)
This is t.he ratio of scales of the parameters of connecting curves at their junction
or the ratio of absolute values of the tangent vectors there. If the value of
t.his scale ratio is not equal to 1, the magnitudes of the tangent vectors of the
10.5 Curvature Profile Problem in Design 157
connecting curves at the junction are not equal, also their derivatives of higher
degree with respect to the parameter are not the same, though the geometrical
quant.ities such as unit t.a,ngent directions, curvatures and t.orsions are continuous
there.
When we want to construct shapes by connecting curve segments, it is diffi-
cult to determine their ratios exactly in advance . Fortunately their exact ratio
values are not needed. We explain two methods of estimating ko at junctions.
1. When a curve segment r(t; n) or its B polygon is divided at t = t* into two

curve segment.s L1 and ro or B_1 and Bo, eq. (10.32) holds exactly. The
value of ko is equal to A-I! Ao . Accordingly, if the shape of two connecting
curve segments can roughly be simulated by one curve segment, and its
division at t = t* gives also an approximation to the desired connecting
curve shape; then t* /(1 - to) is a good estimate for the desired scale ratio.
2. The other method is to take ko equal to the ratio of their chord lengths. In
this case the list of A is proportional to the list of chord lengths of connecting
curve segments. This is effective when the shapes of connected segments
(more precisely their B polygons) are expected to be almost similar.
When we design a curve shape using an 5 polygon, we determine the Ai to be

all equal or proportional to expected chord lengths of the segments. The former
is simple, but when there are considerable differences in the chord lengths of
th~ generat ed curve segments, the distribution of radius of curvature is affected .
Figure 10.8 shows them for the same 5 polygon: (a) profile of radius of curvature
when the list of scale factors A is (1,1,1,1,1,1,1), (b) that for A list (1,1,1,0.3,1,1,1),
(c) that for .\ list (1,1 ,1,2,1,1,1). The region of small radius of curvature at the
central part of the curve seems proportional to the value of the central element
of the A list. As we explain in the next section, this profile is important for
evaluation of the shape quality of curves.
The shape of a curve generated from an 5 polygon is predictable, because
the polygon represents its approximate shape, but the reverse does not always
give predictable results. The former process is to get points Pi from points
qi, the latter is its reverse. In the latter case, simultaneous equations for the
vertices of an 5 polygon have to be solved. Curves generated from the 5 polygon
show sometimes quite different shapes from those expected due to inappropriate
selection of the dividing ratios. This is the same problem with the scale ratios
discussed in Sect. 8.5. The equations to be solved and examples are given there.
We can have more freedom of shape design by applying the C(2) connection
treated in Sect. 10.6 or the rational expressions explained in Chap. 12, but we do
not yet have good research results on controlling the curvature profile along the
curve.
(a). = (1,1,1,1,1,1,1) (b), =(1,1,1,0.3,1,1,1)
(c). = (1,1,1,2,1,1,1)
Figure 10.8 Effects of values of scale ratios on shapes of connected curves. Patterns of radius
of curvature are also shown
10.5.2 Control of Curvature Distribution of Connected Curves

When connecting two curve segments in C(2) or in C(2), usually the shape of the
distribution of curvature along the curve has not been treated. But when quality
of curve shape has to be considered, the shape of the curvature profile becomes
important. It is very sensitive to locations of control points. For example,
Fig. 10.9(a) and (b) show the radius of curvature distribution for lists of ,\ ratios
(1,1,1,1,1,1) and (1,1,1,0.5,0.5,1) for the same S polygon. The profiles are
not smooth in either case, though they are continuous. As shown in Fig. 10.9(c),
a slight displacement of one vertex changes the profile rema.rkably.
When we connect curve segments to synthesize a larger segment, frequently
its profile of curvature distribution takes shapes as shown in Fig. 10.8 or Fig. 10.9.
This type of connection is not always accepted by caT style designers though the
curvature is continuous. So we have to devise a new met.hod to synthesize a
curve shape with a good curvature profile by connecting two segments. At first
we ma.ke a segment of degree five to adjust its control points to have a. good
curvature profile. Then we divide the segment at t* to approximate each part by
a segment of degree three keeping the scale ratio of the approximated segments
to be nearly equal to t* /(1 - t*). When this process is performed well, we have
10.5 Curvature Profile Problem in Design 159
(a) A =(1,1,1,1,1,1) (b) A = (1,1,1,0.5,0.5,1)

Displaced vertex
(c) A =(1 , 1,1,0.5,0.5,1) and displacement of a vertex

Figure 10.9 Profiles of radius curvature. Effects of variation of A, and of displacement of
one vertex
(a)
(b)
(c)
Figure 10.10 Selection of an optimum scale ratio at a junction. Profiles of radius of curva-
=
ture(logarithmic values) are shown: (a) scale ratio, 1: 1, (b) scale ratio chord ratio, (c) scale
ratio obtained from division of higher degree curve
a synthesized shape with a good profile of curvature.

Examples are shown in Fig.lD.lD . In (a), the scale ratio of two segments is
taken as 1, in (b) it is equal to the ratio of the chord lengths of the segments.
The scale rat.io in (c) is determined from the division of an approximated curve
represented by a Bezier curve segment of degree five. As for their shapes, cases
of (a) and (b) are not considered good, but (c) is evaluated as good by car body
stylists. Evaluation of surface quality is more difficult; its methods are described
in Chap. 7 and in Chap. 17.
11 Connection of Bezier Patches and '
Geometry of Spline Polygons and Nets
11.1 Introduction
In Sect. 1L 2 of this chapter, we treat surfaces which consist of Bezier patches con-
necting in c(n-l) in all directions. Techniques for obtaining B polygons from an S
polygon are extended to those for generating nets of Bezier control points from a
given spline net. For simplicity of mathematical manipulation and programming,
we adopt a tensor product techn,ique for producing Bezier nets. Connection in
C(2) and insertion" of vertices of an S net are also included in the first section.
In the next three sections, we explain the theoretical aspects of the S polygon,
which are the bases of the practical methods in the previous chapter.
Properties of S polygons are discussed using Menelaus' theorem in the pri-
maTY geometry. Then we deduce two general algorithms to determine B polygons
from an S polygon and a method of inserting a vert~x in an S polygon. We in-
troduce a general formula for locating B points from a given S polygon of degree
n. When n is equal to or lower than 3, methods of locating B points from an
S polygon are well known, but in higher degree such methods are rarely found.
So we explain our method, which can be applied for arbitrary n. In this way
we. treat the theory of connection a.nd control in curves and surfaces using our
knowledge of Bezier curves and patches as much as possible. Finally in Sect. 11.7,
we explain an orthodox approach to the B spline curve, using the B spline basis
functions.
11.2 Spline Nets and Connected Bezier Nets

11.2.1 Tensor Product Surfaces
Just as a row of B polygons of degree n connected in c(n-l) can be generated
from an S polygon with its dividing ratio lists L;n) of degree n, an array of c(n-l)
connected B nets of degree n x n can be generated from an S net with its dividing
ratio lists in u and v directions Lu;n) and Lv;n) of degree n. See Fig.Il.l.
Let mesh points or vertices of the S net be qiJ. Their first subscript i belongs
to the u direction and the second subscript j to the v direction. In figures, the
u direction is taken vertically and the v direction horizontally. This coincides
with the order of subscripts of elements in a matrix. When we treat a matrix of
vertices of a unit S net or the corresponding matrix of B points of a complete
B net derived from the unit S net, we call them an S matrix or a B matrix for
simplicity.
162 11 Connection of Bezier Patches and Geometry of Spline Polygons and Nets
Figure 11.1 A unit 5 net and its

derived B net
In an S polygon, its basic element is an edge with its dividing ratio list, while
in an S net, its basic element is a ruled surface of degree 1 x 1 with its u and
v dividing ratio lists . The ruled surface is defined by four vertices of a mesh.
Dividing points of edges in an S polygon correspond to dividing lines in the v
and u directions, which coincide with generating lines of the ruled surface in the
v and u directions. The dotted lines in Fig.1l.1 are the dividing lines. Dividing
ratio lists Lu;n) of the dividing lines in v direction are common to all the ruled
surfaces in the v direction (row), and Lv;n) of the dividing lines in the u direction
are common to all the ruled surfaces in the u direction (column). Accordingly,
the dividing lines in the v direction divide the dividing lines in the u direction
in the same dividing ratios as given by Lu;n) and vice versa.
. Now, we explain a tensor product method for determining locations of ele-
ments of a B matrix from a given unit S matrix.
1. Convert each column of a unit S matrix with the common dividing ratio list
Lu;n) into the corresponding column of the B polygons, then the S matrix
is changed to an intermediate matrix. If we make a surface from the control
points defined by this matrix, it consists of Bezier patches of degree n x 1
connected in the u direction in c(n-l) and in the v direction in C(O). The
u = canst lines of this composite patches make polygons in the row direction
and can be considered S polygons with the common dividing ratio list Lv;n).
2. Convert each row of the intermediate matrix with the common Lv;n) into
the corresponding complete B polygons, then we have a. matrix of vertices
of a complete B net, from which we can produce a B patch. This B patch
connects with its adjacent B patches of the same kind in c(n-l).
Example. Let a unit S matrix of soo( u, v; 3, 3) be
r
~:: ~:~ ~:~ ~:].
q20 q21 ~2 q23
(11.1)
Q30 ' Q31 ~2 ~3

11.2 Spline Nets and Connected Bezier Nets 163
Converting each column of the above array with the dividing ratio lists
i = 0,1,2, (11.2)
into a B polygon of degree three, we obtain the intermediate matrix. Let it be
qbo qbl qb2 qb3]

[ q~o q~1 q~2 q~3 (11.3)
q~o q~1 q~2 q~3 '
cho chI q~2 q~3
where the elements of each column of the above matrix are obtained according
to the location symbols used in the item (3) of Sect. 10.3. as follows,
,
qOj = (~).2.1'
, (q1.j) .b
qIj
,
q2j = (q1.j ).2,
, j = 0, . . . 3.
q3j (q1.j).2.1 ,
Then, considering each row of the above intermediate matrix as a unit S

polygon with the dividing ratio lists
j = 0,1,2,
a.nd converting it into a B polygon of degree three , we obtain a B point matrix
where locations of the B polygon for each row of the above matrix are given by
PiO = (q:.Ohl,
Pil (q:,lh,
Pi2 = (q:.Ib,
Pi3 = (q:.Ihb i = 0, . . . 3.
The scale of the B net thus determined is ).; x J..l)" Figure 11.1 shows a unit S
net of degree three and its derived B nets.
For a connection of 0(2) the dividing ratio lists L'U;3) and L'V;3) should be
used instead of LU;3) and LV;3) (refer to eq. (10.31)) where L'U;3) and L'vJ 3) are
the dividing ratio lists in the u and v directions for 0(2) connection.
For a higher-degree S net, the conversion process is similar to the above.
The order of the conversion procedures, for columns or rows, does not affect the
resulting surface.
( a) Spline net
(b) Derived B nets (c) B patches
Figure 11.2 Shape of an S net and its derived B nets and six generated patches connected
in C(2)
In Fig. 11.2, an S net, its generating array of B nets and the shape of its
surface are shown in t.he same scale. For the S net, the shape of the generated
surface is globally similar, but its detail is quite different, whereas for the array
of B nets, the shapes of both the genera.ted surface and the B nets are visually
almost the same, and t.heir true shapes and geometric properties can easily be
calcula.ted. In the S net, shape modifica.tion by displacement.s of their vertices is
possible, but in the B nets it is not allowed. Accordingly, in the design of surface
shape, representations and displays of both the S net and the B net should be
used int.erchangeably.
11.2.2 Division of an S Net

To divide an S net or to increase its vertices, we use a method similar to one
used in derived B net.s from a S net..
For each column of a.n S net, we ma.ke new vertices in all sub-edges corre-
sponding t.o a cert.ain scale factor, for instance AO, by dividing them in the ratio
(u : 1 - u). We replace the columns of the old vertices by the new ones. This
procedure corresponds to the division of an S polygon. The number of rows
increases by one. For each row of this new net, we make new vertices in all
sub-edges corresponding to a certain scale factor , for insta.nce !-la, by dividing
t.hem in t.he ra.tio (v : 1 - v) and -replace the old vertices by the new ones.
11.3 Geometric Structure of S Polygons 165
c:d
a:b a:b
a: b = SIlO: (1 - 5)11'0
c : d = tAo: (1 ~ t)Ao
(a) Locations of new vertices (b) New S net with increased vertices
Figure 11.3 Division of S net
This is the increase of vertices of an S net. A B net corresponding to

AO x fJo is divided into four B nets corresponding to AOU x fJoV, Ao(1- u) x fJoV,
AOU x fJo(l - v) and Ao(l - u) x fJo(1- v) connected in c(n-1).
An example of insertion of control points in an S net is shown in Fig. 11.3.
The·initial 4 x 4 vertices of an S net of degree 3 x 3 are increased to 5 x 5 vertices.
In Fig. l1.3(a), the original unit S net is shown together with the dividing lines
in the u and v directions and locations of the new vertices. Figure l1.3(b) shows
the resulting S net with increased vertices.
11.3 Geometric Structure of S Polygons
In the upper part of Fig. 11.4, an S polygon with its vertices qo ql q2 q3 is

schematically shown in a straight line with the vertices indicated by small blank
circles. In the following explanation, any number n is used, but if n is taken equal
to three, the schematic diagrams in Fig. 11.4 become helpful for understanding.
In Fig. 11.5 the dividing ratios and sub-edges are shown for n=4.
First, we assume that each of n edge vectors of a B polygon Bo of degree
n be constructed by a linear combination of consecutive n edge vectors of an S
polygon and the similar B polygon Bl be determined from the one-edge right-
shifted n consecutive edges of the same S polygon, and that Bo and Bl connect
in C( 11-1), which is guaranteed by the Connection Defining Polygons (CD Ps)
explained in Sect. 10.2.4. Since we assume the degree of B polygons Bo a.nd Bl
is n, and that of their constraining ·CDP BO.l is n - 1, the degree of freedom of
qo ho ql hI q2 h2 q3
5(3)
o .. 6 ~ ~. ~
a
5- 1,0 a
50,1
a 0
B_ 1 ,0
Iil
BO,1
*B-l Bo
o-----@----liJ
*Bo Bi
• ®
® ®
Bo
Figure 11.4 Schematic diagram for relations between S polygon and its various derived
polygons: a unit S ' polygon, truncated S polygons, CDPs, truncated B polygons and a B
polygon
the edge vectors of the connected Bo and Bl is 2n - (n - 1) = n + 1.

Accordingly, we take consecutive n + 1 edges of the 5 polygon, from which
c(n-l) connected B polygons Bo and B 1 , can be constructed. We consider that
in these consecut.ive n + 1 edges of the S polygon, the first n edges produce Bo
wit.h scale Ao and t.he last n edges produce Bl with scale AI. There are n - 1
overlapped edges. Then it is appropriate to assume that a CDP Bo,1 with scale
('\0 + AI) is generated from their common overlapped n - 1 edges. Since the S
polygon ext.ends in both directions indefinitely, any consecut.ive n - 1 edges of
the S polygon can generate one CDP.
Let edges of the S polygon be . . . , ho, hI! h2' ... , and a B polygon Bo be
generated from consecutive n edges (h o ··· h n - 1 ) which begin at a vertex qo.
Then, the S polygon must have the following properties:
1. A list of scale factors (Ai-n+l> Ai-n+2, ... , \) of n elements belongs to each

edge hi of the S polygon. The index of the last element of the list is the
same as that of the edge to which it belongs. We call this list the dividing
ratio list of the edge hi and denote it by
(11.4)
To assign an edge hi the data of its dividing ratio list., we divide the edge
into n sub-edges in the ratio proportional to each element of L~n) given by
(11.4) . For simplicity, we call this process "divide hi by LIn)>> or "dividing
ratio of hi is LIn)>>. A "sub-edge of A/ is a sub-edge which is a Aj part of
an edge divided by LIn).
The eq.(l1.4) is derived as follows . Referring to Fig. 11.5, since a polygon
Bo whose scale is Ao is constructed from n consecutive edges (ho ··• h n - 1 )
11.3 Geometric Structure of S Polygons 167
Figure 11.5 Dividing ratios and sub-edges
of the S polygon, we consider that the polygon Bo is constructed from all

the Ao sub-edges, each of which belongs to each of the above consecutive n
edges of the S polygon, but not from the other edges.
Its a.dja.cent polygon B-1 is produced by A-I sub-edges from each of n edges
(b_ 1 ••• b n- 2 ), and similar relations continue until B-(n-1) is constructed by
A_(n-I) sub-edges from each of n edges (b-(n-l)'" b o). Therefore, the edge
b o influences the n B polygons (B-(n-1)'" B o), that is, the edge b o has the
dividing ratio list L~n), which contains only n elements of A from A-n+l to
Ao.
We call a set of n consecutive edges (bi ... b i +n - l ) beginning at a vertex q ;
a. unit S polygon sin) and its scale Ai the main scale of sin), which is the
common element of the Ljn) of all its n edges for j = i,'" (i + n - 1). The
index of the main scale is the same as that of the unit S poly~on.
2. The common part of adjacent unit S polygons s;~i and S; n) has n - 1
consecutive edges and common scales A;-1 and A;. So the common part
can be considered a unit S polygon of degree n - 1 with the main scale
(,\;-1 + ,Xi). We use a notation 5i - 1,; for it and call it the reduced S polygon
of second kind. It produces a CDP 13;-1,;. The reduced S polygon of the
first kind appears in the farther section.
3. Referring to Fig. 11.4 again, a unit S polygon S;n) generates three B poly-
gons * B;-l, B;, BiHof degree n, the central one complete and the other two
incomplete. They connect in c(n-l) and their scales are Ai-I, Ai and '\;+1.
This is proved as follows. A unit S polygon sin) is looked upon as union of
two common parts: the common edges of {Si~i and sin)} , which make 5i - 1 ,; ,
and the common edges of {Sin) and S;~i}, which form 5;,i+l' The reduced S
polygon 5;-1,i generates a CDP'13;_l,;, whose division in the ratio (\-1 : A;)
gives incomplete B polygons {* B i - 1 , Bn and similarly from Si,i+1, a e DP

Bi ,i+l is obtained, whose division by ratio (>'0 : >'d produces {* B i , Bi+l}'
Their union is * B i - 1 , B i , Bi+l' So the statement is correct.
4. The common edges (b 1 ••• b n - 2 ) of three unit S polygons s~t S~n) and
sin) constitute also a unit S polygon of degree (n - 2) with the main scale
(>'-1 + >'0 + >'1)' This is also the common part of 5_ 1 ,0 and So,}' which
can generate two adjacent CDPs B_ 1 ,o and Bo,l' Their common part is the
both-end truncat.ed incomplete B polygon * B~. So we use a nota.tion * S~
for this trunca.ted S polygon, whose main scale is (,\-1 + >'0 + >'d and the
degree is (n - 2).
If we can deduce from this *S~ a unit S polygon of degree (n - 2) with the
main scale >'0, we can easily determine the locations of the B points from
an S polygon of degree n, because t.he degree of the given S polygon can be
reduced rapidly with the same main scale, and a unit. S polygon of degree
one or zero has the same vertices or vertex as a B polygon of the same
degree. Examples of the method are explained in Sect. 11.4.3.
11.4 Menelaus Edges and Their Dividing Points

To determine B polygons connected in ern-I) from an S polygon, we apply
Menelaus' theorem (refer to Sect. 3.3) to the geometric structure of the S polygon,
which we explain in this section.
11.4.1 Dividing Points and Sub-edges

The dividing ratios lists of all edges of the unit S polygon S~n) are
{L6n ), ••• , L~~I}' whose common element is >'0. As an edge of the polygon ad-
vances to t.he right., the relative posit.ion of a sub-edge of >'0 in each edge shift.s by
one position leftwards from the right-end position of the first edge to the left-end
position of the last edge. See Fig. 11.5. Except for these consecutive n edges
there are no sub-edges of >'0 in the S polygon.
We divide an edge b o by the ratio L~n) and the next edge b I by Lin) each
into n sub-edges and let the dividing points of b o be (qO.l QO.2··· Qo.n-l) and
those of b 1 be (Ql1 Q12'" Q1.n-l)·
11.4.2 Relations Among Dividing Points and Menelaus Edges
We define Menelaus edges which are used in the process of locating B points
from an S polygon and explain their properties. Owing to t.he division of edges
into sub-edges by t.he scale lists and the shifting of the dividing ratios in the
consecut.ive edges, we can apply Menela.u s' theorem (Sect. 3.3) to each pair of
adjacent edges.
11.4 Menelaus Edges and Their Dividing Points 169
_:0.2
(
qO.1
qO.2.2 = QO.3.I
,1,-4 QO.2.3 = QO.4.I
qo "
QO.3.3 = Q0.4.2
Figure 11.6 Dividing points and Menelaus edge
Refer to Fig. 11.6. We select one dividing point qO.i on the edge b o = qOql
and the other point ql.i-I on the adjacent edge b i = qIq2' The subscripts of
qi.i-I are determined by increasing the first index and decreasing the second
index of qO.i each by one. We call this indexing operation skew shifting. We
divide a line segment joining ~.i and its skew shifted q1.i-I' in the ratio L~n-I)
into n-l sub-edges at the dividing points (~.i.1 qO.i.2··· qO.i.n-2) ' In each symbol
of the dividing points of this array, the first two index numbers indicate the index
of the starting point of this line segment which starts at qO.i and ends at q1.i-1
and 'the third one indicates the sequence number of the dividing points.
The following relations hold for the dividing points.
(i) Three points qO.] , qO.i.j-1 and q1.j-1 are collinear. When two of the three
point are determined first, a line joining these two points or its extension
intersects a line, on which the third point is to be placed. This intersecting
point gives t.he location of the third point.
(ii) Two points, whose location symbols are qO.i.j and ~.i+k.j-k are the same
point if j =I- i - I and k < n - i.
These two relations are easily proved by application of Menelaus' theorem
and it.s inverse (refer to Sect. 3.3).
A line segment , which joins t.he point.s qo.j and q1.j-1 each on the adjacent
edges and have n - 2 dividing points by the ratio L~n-I), appears frequently. For
convenience, we call this type ofline segment the Menelaus edge, which subtends
an angle at the vertex qi'
Anyone of the Menelaus edges which subtends an angle made by the two
edges b i and b i + 1 can replace the parts of these two edges cut off by the Menelaus
edge on their vertex side, because it preserves geometric properties of the cut-off
parts by t.he propert.ies stated above in (i) and (ii).
170 11 Connection of B6zier Patches and Geometry of Spline Polygons and Nets
q~
qO.4 q1.3 - q2.1
= ql.O q1.2 . ..... ::,,-<: ...... Q2.2
= ql """" Q3
"
qO.3 ---
5(4) o>--- - - - e - - -----B9>------""e>--------<E)

o
Q - -- - - - -- --- - -Er - - - - - - - - - - - - - - - -Er - - - - - - - - - - - - - - - -El
. .
------------ ---- -- ----- --------- . -------------.
Figm'e 11.7 Relation between a unit S polygon S6 4 ) and its reduced S polygons
Sa(>'O) , SbeAO)
11.5 Derivation of B Polygons from an S Polygon

We explain procedures or formulas which can determine the locations of B points
from a given 5 polygon with its scale list. First we introduce auxiliary polygons,
which we call reduced 5 polygons, derived from the given 5 polygon. With the
help of these polygons, we can locate the B points as well as insert a new vertex
in the given 5 polygon.
11.5.1 Reduced 5 Polygons

A unit 5 polygon of degree n is resolved into unit 5 polygons of degree n - 1,
which can generate all t.he B point.s belonging to the original 5 polygon. These
resolved 5 polygons are called the reduced 5 polygons of a unit 5 polygon 5~n),
for convenience. There are t.wo kinds of the reduced 5 polygon. One is used for
determining the locations of B polygons of degree n from the unit 5 polygons
of degree n - 1. The other is directly related to a CDP as stated in item 3 of
Sect. 11.3 and also used for insert.ion of new vertices to increase freedom of the
shape design.
5b
1. Refer to Fig. 11.7. On each edge of n ) there exists a sub-edge corresponding
to the main scale >'0 . On the first edge ho, two dividing points qO.n-l and
qO.n {= ql} are t.he end points of this sub-edges. We construct two polygons
5 a and 5 b of 11 vert.ices by repeated skew shifting from ~.n-l and qO.n' which
are the start.ing points.
5a : qO.n-l q1.n-2
(11.5)
5b : qO.n . ql.n-l
11.5 Deri vation of B Polygons from an S Polygon 171
where q71-1.0 = qn-l and qO.n = ql'

The degree of 5 a and 5 b is n-1 and their corresponding edges intersect owing
to their construction. The intersection of these two polygons determines a
both-end-truncated B polygon. Proof of this and use of this polygon are
described in Sect. 11.5.3.
2. We divide each of n - 1 edges of 5 a into n - 1 sub-edges by the dividing
ratios L~,,-l), ... , L~n_~l). For 5 b we apply the same procedure. Then the
two polygons are the unit 5 polygons of degree n - 1 with the main scale
AO, so notations 5 a(AO) and 5 b (AO) are used for them.
3. Each edge of 5 a {-\0) and 5 b(>.0) becomes a Menelaus edge which subtends
a vertex of 56 n). Hence, edges eit.her of 5 a (AO) or of 5 b (AO) can be used as
substitut.es of parts of the edges of 56n ) which are cut off by these Menelaus
edges.
Since 5 a (-\o) and 5 b ( Ao) hold aU the geometric data of the sub-edges corre-
sponding to AO: t.hey can generate B6 n ) which is originally to be generated
from 56") . Since the last edge of 56n ) does not belong to edges of 5 a (>.0) and
the first edge of 5a n ) is also not included in 5 b(>.0), we can conclude:
5 a (AO) can generate an incomplete B polygon B~(n) and 5 b(>.0) can generate
*B(n)
a .
It follows t.hat if we can determine locations of a complete B polygon of
degree n - 1 from a unit 5 polygons of degree n - 1, we can determine
a complete B polygon of degree n from the two incomplete B polygons of
degree n. Those incomplete B polygons are derived from the two reduced
.5 polygons of degree n - 1.
Since for degree 1, a unit 5 polygon and the complete B polygon are the
same, in the case of n = 1 the above statement holds, so it holds by induction
for general n.
4. In 56n ), there are still two unit 5 polygons of degree n - 1 other than those
st.ated above. One is a polygon wit.h vertices whose first index values are
one less than those given by 5 b in (11.5). The locations of its vertices are:
(11.6)
Since the main scale of this unit 5 polygon (11.6) of degree n - 1 is A-I, we
can use the notation 5 b (A_l) for this polygon, which can generate B: 1(n).
5. Similarly, a unit 5 polygon of degree n - 1 with vertices whose first index
values are one more than those given by 5 a in (11.5) is:
(11. 7)
It has the main scale Al with the right-end point coinciding with that of
56 n ). This polygon is denoted by 5 a (Al) and generates * Bl(n).
Thus 5 b(>.-I), 5 a(AO)' 5 b(AO), 5 a(Al) are the reduced 5 polygons of the first
kind, and from them four incomplete B polygons * B-ll B~, * Bo, B; can be
produced . The union of B~, * Bo is the complete B polygon Bbn ).
6. Besides the reduced S polygons stat.ed above, from item (3) of Sect. 11.3.1
there are anot.her two reduced S polygons of the second kind: 5_ 1 ,0 and
S6
50 ,1 of degree n -1, which are deduced from n ). They generate two CDPs
B_ I0 and Bo 1·
7. Division of B'-I,O gives ·B_ l and B o, which are also obtained from Sb(Ld
and Sa(>'O). Similarly, division of BO,1 gives· Bo and Bi which are also gen-
erated from Sb().O) and Sa().I)' Accordingly, 5_ 1 ,0 and 50 ,1 are the reduced
S S6
polygons of n ). They are of the second kind. The B polygons B-1, Bo
and BJ are connected in c(n-l). From (6) and (7), we can prove that a
procedure of division of an S polygon, or insertion of a vertex in a.n S poly-
gon, does not change the shape of the original curve. This is explained in
Sect. 11.5.4.
Thus, all the condit.ions for S polygons given in Sect. 11.3.1 are satisfied without
cont.radiction. .
11.5.2 Examples
We apply the rule in item (3) of Sect. 11.5.1 for locating B polygons from a unit
S polygon.
B6
For n=l, it is evident that 1) = (Po PI) = (qo ql) is the same as Sal) = (qo ql)'
For n=2, a unit S polygon of degree two is S6 2 ) = (qo ql q2) whose dividing ratios
of two edges are (>'-1 : ).0) and ().o : ).1)' Its reduced S polygons are given by
(11.8)
The former gives the location of BO(2) and the la.tter determines that of • B~2),
. because they are actually of degree one and the result of the case of n= 1 IS
directly applied. Accordingly we have
(11.9)
Similarly we obt.ain from the reduced S polygons:
(11.10)
(11.11)
By incrementing t.he first index of the dividing points of B~2) and Sa2), we have
Bi
the right-hand adjacent polygon 2 ):
(11.12)
This step is more t.horoughly treated in Sect. 11.6.

11.5 Deri vation of B Polygons from an S Polygon 173
11.5.3 Locating B Polygons from Reduced-Truncated S Polygons

If a both-end-incomplete B polygon • B; is simply obtained from a truncated
polygon· S;, it is very useful for determining B points of higher degree from the
corresponding S polygon, because polygons • B; are sufficient for obtaining all
the locations of B points from the given S polygon. Though· S; is of degree
n - 2, it.s main scale is not Ai. Accordingly, it cannot easily be used for repeated
reduction of S polygons of higher degree to those of lower degree of the same
main scale.
The objective of this section is to deduce an S polygon of lower degree with
the same main scale from a given S polygon of higher degree. Wit.hout. losing
v (n-2) v (n-2)
generality, we set i = 0 and use a not.ation So and Lo for such an S
polygon and its dividing ratio of the first edge. We call this polygon S~n-2) a
reduced-truncated S polygon. This polygon was already used in Sect. 10.4.1.
Since SaP,O) an'd Sb(AO) generate B~ and· Bo, if we can produce an equivalent
S polygon with the common part of Sa (>.0) , Sb(Ao), we can obtain the common
part. of B~ and· Bo, which is • B~. Since the requirements of such an S polygon
have been given , at first we explain a method of locating vertices of this reduced-
truncated S polygon.
Refer t.o Fig. 11.7. There are n - 1 intersections between Sa(AO) and Sb(AO),
because each pair of their corresponding edges intersect. Let the intersecting
points be 1'0, r},··· r n -3, l'n-2. They make a polygon S of degree n - 2. Its first
vertex 1'0 is at ql.(n-2) and the last one l' n-2 is at ~n-2).2 by definition.
Ea.ch of t.he int.ermediate vertices is det.ermined at an intersecting point of
the ·t.wo Menelaus edges, which are edges of Sa(AO) and Sb(>.O) and subtend a
S6
common angle at a vertex of n ). Their locations are determined by using
v(n-2)
Menelaus' theorem. Thus, all the locations of the vertices of So are given by
1'0 = ql.(n-2)
ri = Q i.(n-i-l).(n-i-l) = Qi.( n-i ).(n-i-2), i=I···n - 3 }. (11.13)
r n -2 = Q(n-2).2
(In Fig. 11.7, since n=4, the degree of t.he reduced-truncated S polygon is two,
so there exist 1'0 = Q1.2 ' 1'1 = Q1.2 .2 and 1'2 = Q2.2 only.)
We prove that the polygon thus determined has the desired dividing ratio
lists and t.he main scale.
v(n- 2)
1. The first. edge of So is the same as the second edge of Sa(>.O) except for
the rightmost sub-edge which corresponds to AI. Since the dividing ratio
list of the second edge of Sa(AO) is L~n-l) = (A_ n+3,"', AO, Ad, the dividing
ratio list. of the first edge of S~n-2) becomes L~n-2) = (A_ n+3,"' , AO), which
v (n - 2)
we can denote as Lo
v(n-2)
2. The last. edge of So is the (n-2)-th edge of Sb(AO) except for the left.most
sub-edge which corresponds to A-I' Since the (n - 2)-th edge of S(AO) has
Figure 11.8 Resolution of 8- 1 ,0 into Sb(Ld and SaPO)
the dividing ratio list L~n_-.}) = (>'-1, >'0,"', >'n-3), the dividing ratio list of
v(n-2) v(n-2)
the last edge of So becomes L n- 2 = (>'0,"', >'n-3)'
3. The intermediate edges of S are Menelaus edges of level one with respect
to the vertices of S,,(>'o) and also to those of Sb(>'O). For either case the
v (n-2) v (n-2) v (n-2)
dividing ratio lists of So are L1 ... L n_3 . From the properties stated
above, the polygon S~n-2) is a unit S polygon of degree n - 2 with the main
scale >'0, and its degree is n - 2. Hence, S~n-2) is the S polygon which we
desired, that is, the common part of S,,(,\o) and Sb(>'O)'
Application examples have already appeared in the previous chapter in Sect. 3.4.
11.5.4 Division of an S Polygon and Insertion of a Vertex

Refer to Fig. 11.8. Two unit polygons Sb(>'-1) and S,,(>'o) are the divided S
polygons of 5- 1,0 in the ratio (>'-1 : >'0). As stated in Sect. 11.5.1(6) and (7),
Sb{-\-1) and S"Po) produce two B polygons • B-1 and B~, while 5- 10 produces
B polygon B_ 1,o, division of which in the ratio (>'-1 : >'0) produces the same B
polygons as those produced from Sb(>'-d and S,,(>'o).
But t.he number of vertices of 5- 1 ,0 is n and that of Sb(>'-d and S,,(>,o),
which have n - 1 common edges, is n + 1 except for the duplicate ones. The
division of S polygon 5- 1,0 into Sb(>'1) and S,,(>'o) is equivalent to increase or
insert.ion of a vert.ex of the S polygon wit.hout change of shape of the generated
curve segment.s.
In this case we considered division of a unit S polygon of degree n - 1 with
main scale ('\-1 + >'0) into two unit S polygons of the same degree with the main
scales >'-1 and '\0 respectively. If we are to divide a unit S polygon of degree n
11.6 General Formulas for Locations of B Points 175
Figure 11.9 Rules on expressions of locations of B points
with the main scale Ai into (t : 1 - t), we make n dividing points, each of which
divides each sub-edge of Ai in the ratio (t : 1 - t). There are n such dividing
points which replace the n - 1 intermediate vertices of the old unit S polygon.
This is the increase of the vertices of an S polygon. An example has already
appeared in Sects. 10.5 and 11.2.2.
11.6 General Formulas for Locations of B Points

We give formulas of locations of B polygons connected in c(n-l) from a given S
polygon with its list of dividing ra.tios. In lower-degree cases, for example n = 2
and n = 3, methods of determining locations of B points are well known. And
the method using reduced-truncated S polygons, which is given in Sect. 11.5.3
is effective for higher-degree polygons. But in this section, we present a general
formula to give locations of B points from an S polygon of arbitrary degree n.
11:6.1 Rules of Location Symbols of B Points and

Their Properties
The location of each dividing point on a Menelaus edge is determined by (i) the
sequence number from the left starting point together with its dividing ratio list
which determines division of the edge concerned, and (ii) the location of the left
starting point of the edge, which is also a dividing point of a Menelaus edge of
one level lower.
Referring to Fig. 11.9, for instance, q13.2 is the second dividing point of an
edge starting from q1.3 and ending at Q2.2' The left end point Q1.3 is the third
dividing point of an edge joining Ql and Q2' The location of the right end of a
Menelaus edge is not explicitly shown, but it is the point whose index is skew
shifted from that. of the left end point of the same edge.
11.6.1.1 Level of Menelaus Edges and Dividing Points
As for the level of an edge and a dividing point, the lowest level is defined as level
zero. Edges and vertices of the level zero are those of the original S polygon. A
dividing point of level one is a point on an edge of level zero, and a Menelaus
edge of level one starts from a dividing point of level one. Dividing points of level
two are on a Menelaus edge of level one. The dividing ratio list is inherited as
follows. For instance, if the level zero edge joining q1 and q2 has dividing ratio
list Li4 ) = P'-2, '\-11 AO, A1), the dividing ratio list of the level one edge joining
q1.3 and q2.2 is Li3 ) = (A-11 Ao, A1)' That is, the first element of Li4 ) is removed
and the rest is the same. The dividing ratio list of level two edge joining q1.3.2
and q2 .2.1 has the dividing ratio list Li2 ) = (.\0, A1)'
11.6.1.2 Symbols for Location of Control Points
Now we generalize the above notation rule for dividing points. Let i be an index e
variable indicating a sequence number of a dividing point of level i. The location
of a dividing point of level i is given in the form:
(11.14)
The characteristics of this expression are :
- The location of this dividing point is on a Menelaus edge of level i-I whose left
end is qe .e
o 1.8 2 .••• 8i- 1. Its dividing ratio list is given by L8~n-i+1), because
the lowest level is zero by definition and its superscript of the dividing ratio
list eq.(llA) is (n). Accordingly for the level i - \ the superscript becomes
(n-(i-l)).
- The subscript of the dividing ratio list is coincident with values of the first
index 80 of the location symbol, because the last element of the dividing ratio
Qf a Menelaus edge is inherited from the last element of the dividing ratio of
the Menelaus edge of one level lower. Accordingly it is the same as that of the
lowest. level: it. is t.he scale factor of the rightmost sub-edge of an edge of the
original polygon.
- A vertex of t.he polygon is represented by q8 0 and a dividing point on its edge
is q8 0 .81 . The value of 80 can be any number because it represents vertices
of the originalS polygon, but for t.he ot.her 8k , k > 0, its maximum value is
n- k.
- The operation of skew shifting increases value of 80 by one and decreases each
values of 8k by one for k > O. If the values of indices of two dividing points
are the same t.o a certain level and the values of the following t.wo indices are
(i,j) and t.he ot.hers are (i + k,j - k) and if j =I i - I and k < n - i, the two
dividing point.s are the same point. Refer to Sect. 11.4.2. For example, three
dividing point.s wit.h location symbols
(i) qO.3.3.2 (ii) q04.2.2 (iii) qO.4.3.1
represent. t.he same location, because in the expressions (i) and (ii), qo is com-
mon and i = 3, j = 3, k = 1. In (ii) and (iii), qO.4 is common and i = 2, j = 2,
k = 1. They satisfy the condit.ions given above.
11.6 General Formulas for Lo cations of B Points 177
Table 11.1 Values of indices for B point locations
In I Po P5
1 0 1
2 0.1 1 1.1
3 0.2.1 1.1 1.2 1.2.1
4 0.3.2.1 1.2.1 1.2.2 1.3.2 1.3.2.1
5 0.4.3.2.1 1.3.2.1 1.3.2.2 1.3.3.2 1.4.3.2 1.4.3.2.1
11.6.2 General Expressions of B Point Locations
11.6.2.1 Application of Location Symbols
From locat.ion symbols of vertices of a B polygon of degree n, which are derived

from a unit. S polygon of the same degree , we can deduce vertex locations of a B
polygon of degree n + 1 derived from a unit S polygon of degree n + 1. For n = 1
and n = 2, we have shown the deduction process in Sect. 11.4.2. The method is
similar for general n. In Table 11.1 the locations of B point.s for n = 1,' . ·5 are
shown . We give rules for deducing • B~i+l) and B(j(i+1 ) from B~i) :
- To convert location symbols for B points of B~i) , which is assumed known, into
t)lOse of Bo(i+I), which is one degree higher and misses the rightmost point,
replace the first index position of the location symbol in B~i) by two index
numbers: if the first one is 0, it is replaced by O.i and if the first one is 1, it is
replaced by 1.(i - 1). Do not change the subscripts of the control points Pj of
B polygons.
For exampIe, Iet us assume for z· = 3 we know Po(3) = qO.2.I , and PI(3) = q1.1'
Using the above rule we obtain p~4) = ~.3.2.I and pi4) = P1.2.I'
- To convert. location symbols for B~i ) into those of • B~i+1), replace the first
index position of the symbol by t.wo index numbers: if the first one is 0, it is
replaced by 1 and if the first one is 1, it is repla.ced by l.i. Increase values of
the subscript j of the control points Pj by one .
For exampI e, f rom Po = qO.3.2.1, we get PI = q1.3.2.I and from PI = QU.I'
(4) (5) (4)
we obtain p~5) = Q1.4 .2.1, which is equivalent to Q1.3.3.I and Q1.3.2.2' The last
one is written in the Table 11.1.
The number of B points in B~n+l) is n + 2, but the total number of B

point.s obtained by the above method is 2n + 2, so there are n duplicates. For
convenience, the first n - 1 points a.re taken from those of B·~n+l ) and the last
two points are from those of· B~n+.l), which are shown in Table 11.1.
11.6.2.2 Formulas for Location Symbols
The general formulas for location symbols of B points of B~n) are given in the
form (11.14):
• the first B point Po:

{ eeok == 0n - k : for k = 1 ... n - 1
• the second B point PI:

{ eoe
= 1
k = n - 1 - k : for k = 1 ... n - 2
• the intermediate B points Pj (j = 2 ... n - 2) :
eo = 1
{ e =n -
k 1 - k : for k = 1...n - j - 1
(11.15)
ek = n - k : for k = n - j ... n - 2
• the penultimate point Pn-I :

{ eoe
= 1
k = n - k : for k = 1 ... n - 2
• the last point P n :
{ eoe =
k
1
= n - k : for k = 1 ... n - 1
To determine the real location of a B point from its symbolic notation, we
just translat e it according to the definition.
For example, we take PI and P2 of BaS). According to the above formulas
(11.15) for n = 5, j = 1 and j = 2 we have PI = q1.3.2.1 and P2 = q1.3.2.2' They
are the first and the second dividing points of the level two Menelaus edge divided
in t.he ratio Li3 ), because n = 5, i = 3 and eo = 1 and Le~n-i+l) according to the
first. item of Sect. 11.6.1.2. The end point.s of this Menelaus edge are ql.32 and
its skew shifted point q2.2.1'
The point q1.3.2 is the second dividing point of the Menelaus edge of level
one, which joins q1.3 andq2.2, divided in Li4 ). The point q1.3 is the third dividing
point of the edge of level zero between ql and q2 divided in LiS). The point q2.2.1
is the first dividing point of the edge of level two between q2.2 and ~.1' divided
in L~4). The points Q2.2 and Q3.1 are the second and the first dividing points of
the edges (Q2Q3) and (Q3Q4) of level zero divided by L~S) and L~S) respectively.
11.6.2.3 Level of Edges of a B Polygon
The relation bet.ween t.he levels of the Menelaus edges and edges of a B polygon
IS:
11. 7 Appendix. Orthodox Approach to a B Spline Curve 179
- Bot.h end edges of t.he B~n) polygon deduced from an 5 polygon of degree n
belong t.o t.he sub-edges of the Menela.us edges of level (n - 2).
- The other int.ermediate (n - 2) edges belong to the Menelaus edges of level
(n - 3).
- Each of t.he edges of a B;n) polygon, whose scale factor is Ai, is the sub-edge
of Ai of the respective Menelaus edge of level (n - 2) or the level (n - 3).
B6
For example, the first and the last edges of 5 ) are on level 3 and the other
three are on level 2, and their sub-edge belong to t.he Ao part of the respective
Menelaus edges.
Exercise. When t.he scale factors Ai are all equal, prove that the locations of B
point.s are as shown below.
For all n,
P (n)
n
_
-
(1/'))(
~ P n +1
+ P n -1 )(n) . (11.16)
For 11 = 2,
P(2) -q
1 - l'
For 11 = 3,
For n = 4,
For n = 5,
11. 7 Appendix. Orthodox Approach to a B Spline Curve
In order for basis functions or weight.ing functions for control points not to have
a global influence on the shape they express, they must have the property that
each one has value zero outside a restricted range called a support. Each of the
basis functions of t.his property together wit.h derivative continuity conditions can
be represent.ed not by a single polynomial, but by the connection of polynomials
defined bet.ween successive points called knots. The B spline curve is a linear
combination of t.he B spline basis functions, each of which is multiplied by a

control point vector qi' such as
n
r(x) = LqiNi(n)(X). (11.17)
i=O
The control point.s qi make a B spline control polygon. The normalized B spline
basis functions N;nl(x) of degree n are piecewise polynomials of degree n which
are defined over t.he knots ... < Xo < Xl < X2 < ... and have the following
properties:
n
LN(n)i(X) = 1,
i=O
N(n)i(X) > 0,
~T(n),: (x)
"' = 0 1'f X < Xi an d X > Xi+n+l,
N(n)i(X) IS (n - 1) t.imes continuously differentiable.
The recurrence formula by deBoor is
N (j-l)( ) N U-l)( )
Ni
(Jl () (
X = X - Xi
)'
(
X
) + ( Xi+J+l -
) . +1 X
X ( ). (11.18)
Xi+j - Xi Xj+i+1 - Xi+l
where
N,(O)(x) = {01 ifotherwise.
< <
Xi X Xi+l
The vertices qi and the knot.s ... Xo, Xl , X2,'" having been supplied, an ort.hodox
met.hod of calculation of a curve expression (11.17) at given X is to obtain values
of t.he spline basis functions by using the recurrence formula (11.18) given above.
Compared with the method of det.ermining Bezier points from the vertices of
the B spline control polygon with the given scale ratio list, the orthodox method
of B spline curves is not easy for understanding the geometric properties of the
curve or for it.s manipulat.ion. Moreover, when we need to calculate various
characteristic curves on a spline surfaces or to determine interference curves
between spline surfaces, their surface expressions derived from the tensor product
of eq.( 11.17) are not adequate because calculation of a large number of surface
point.s is required. For further det.ail on splines consult the references [22][33].
12 Rational Bezier and Spline Expressions
12.1 Introduction
Bezier expressions for curves and surfaces are easy to understand and have fa-
vorable features for use in CAD and CAM. But they cannot exactly express
conics and quadrics such as circles and spheres. By extending the expressions
from polynomials to rational forms, we can provide them not only with the abil-
ity to express conics and quadrics, but also with possibly increased freedom for
design [28][29][32].
A Bezier curve is constructed by repeated linear interpolation between two
points, that is, division of a line joining two points in the ratio t : (l-t). Whereas
its counterpart, a rational Bezier, uses repeated rational interpolation which
divides a line joining two end points in the ratio WIt: wo(1 - t), where Wo and
WI are weights of the respective end points, and the weight of the intermediate
dividing point is wo(1 - t) + WIt. The end points and their weights may be
functions of t. For the same end points and the same value of the parameter t,
o :::; t :::; 1, the linear and the rational interpolations represent different points.
Similarly a ruled surface is expressed by a tensor product of two line segments in
linear or rational interpolation forms. For the same values of parameters u and
v, ~heir generators are different. But the rational and the linear interpolations
have close relations. For instance, in a perspective projection (see Fig. 12.1), the
control polygon (Po PI P2 P3) of a rational Bezier and the corresponding control
polygon (Po PI P2 pJ of a non-rational Bezier give the same projection seen
from a view point o. A ratio of distances from 0 to the corresponding control
points, 0Pi : 0Pi, is the weight Wi belonging to Pi.
Analogous to this correspondence between the control points, a point on a
rational Bezier curve can be mapped to a point on the usual Bezier curve or on
the corresponding control polygon. Many simple and elegant properties which
characterize Bezier and spline polygons are inherited by their rational equivalents
unless the projection violates the original relations. An S polygon of degree n is
mapped to a rational S polygon of the same degree. An S polygon of degree three
with C(2) connection is also mapped to a rational S polygon of degree three with
C(2) connection. The corresponding locations of B points are easily determined.
This approach makes their mathematical analysis and geometric understanding
much easier than other methods which use 4D space (wx, wy, wz, w) or only
rational forms in 3D.
Currently, the merits in use of the rational curves and surfaces are not al-
ways recognized except for representation of conics. This is because appropriate
selection of weight values used in the rational forms is not always clear and their '
182 12 Rational Bezier and Spline Expressions
Pi
Figure 12.1 Perspe,c tive relation between control polygons of a rational and a non-rational
Bezier
shape calculation is more complicated than non-rational ones [32]. But owing to
their generalized features, their use has begun to be considered recently. Accord-
ingly, we explain in this chapter characteristics and uses of the rational Bezier
and spline curves and surfaces, applying a new geometric interpretation, though
much space in the last section is allocated to rational Bezier curves of degree
two, which are practically useful. An optimum curve fitting method with conics
is explained, which uses rational expressions of degree two with positive as well
as negative weights.
12.2 Rational Bezier Curves

12.2.1 Rational Division Between Two Points and
Its Perspective Map
In a rational polygon each point has its weight together with its coordinate values.
When a point. on an edge of the polygon is determined by division of its two end
points, let the linear dividing ratio between the two end points bea : b, then the
rational dividing ratio is defined as wla : wob and the weight of the corresponding
rational dividing point has a value determined by the linear dividing ratio of the
weights Wo and Wl of the end points of the edge.
We explain a graphical method of locating a dividing point between two
points Po and Pl in the ratio Wlt : wo(l - t), which we call a rational division
with weights Wo and Wl' Referring to Fig, 12.2(a), we define points with their
symbols in the figure.
1. Let 0 be an arbitrary point which we call a virtual origin. The following

procedure takes a point Po on a line joining 0 and Po or its extension to
~Wlt_ r___ wo(l-t)=-=YI
~.~~:!
. . . .\/~!/
w6· . .
\&
()
(a)
(b) (c)
Figure 12.2 (a) Rational division of a line segment, (b) figure for the proof (when both
weights are positive), (c) figure for the proof (when one weight is negative)
make a ratio {oPo : oPo = 1 : wo}, similarly takes a point PIon OPI so
that {OPI : OPI = 1 : WI}.
2. Let P be a dividing point on POPI in the ratio (t : 1 - t), which we call a

linear division, and let P be an intersecting point of the line POPI and a
line joining 0 and P or its extension.
3. Then we can prove that the point P divides the line POPI in the ratio
{WIt: wo(1- tn,
which is a rational division, and the point P has a weight
(1 - t)wo + tWI, that is, a ratio [op : op = 1 : {(I - t)wo + tWI} holds.
Line segments, for instance p~p~, each of whose end points are on oPo and OPI
or their extensions, are perspectives of the original line POPI with respect to the
point O. Among the perspectives the line POPI is special, because locations of
its end points are directly associated with the end points of the original segment
with weights Wo and WI' Accordingly, let us call this line POPI the canoriical
perspective of the original line POPI with respect to the point O. Any points P
between Po and PI correspond to their canonical perspective point p, which is a
dividing point between Po and PI in the ratio t : 1 - t. We can prove that the
corresponding point P is the dividing point of the original line POPI in the ratio
tWl : (1 - t)wo. The point P has the weight tWl + (1 - t)wo just as both end
points Po and PI have the weights Wo and WI' The canonical perspective has an
important role in the construction of rational Bezier or spline curves.
We give a proof for the above stated relations between the original line and
its canonical perspective line. Referring to Fig. 12.2(b), draw a line from Po
parallel to POPI and let its intersection with op and OPI be Pa and Pb, which
divide op and OPI in the same ratio Wo : 1 - Wo0 Then there is a relation of
length ratios:
(12.1 )
Let op divide POPb in the ratio x : y and POPI divide oPa in the ratio u : V.
Applying Menelaus' theorem to a triangle 6POPbPl and a line op, we have
(l~o~al)(I~~ol)(I~lApl) = (~)(wo)(l-t) = 1.
IPaPbl IOPII IpPol y WI t
From this we obtain
(12.2)
Therefore, the point P divides POPI in the ratio tWl : (1 - t)wo.
Then applying the theorem to a triangle 6PaOPb and a line POPI' we have
( I~~I )(I~a~ol)(l~b~ll) = (~)(_x_)(wo - WI) = 1. ( 12.3)

IPPal IpOPbl IpI 0 1 v x +Y WI
From these geometrical relations (12.2) and (12.3) another relation
lopl: IOpl = (u + V)jwo : u = 1 : {(1- t)WO + twd, (12.4)

is obtained. This indicates that a point P has a weight {(l-t )WO+tWI}. At t = 0,
the weight of the point Po is Wo and at t = 1, the weight of the point PI is WI.
The above stated relat.ions can be proved even if one or both of the weights has
or have negative value(s). See Fig. 12.2(c). When one of the weights is negative,
the line segment start.ing from Po and reaching at PI or vice versa should be
considered to pass an infinite point. The application is found in Sect. 12.7.3.
12.2.2 Rational Bezier Curves and Their Canonical Perspectives

Since the expression for a dividing point between two end points represents an
equation of a line connecting them, a line POPI is considered to be a usual Bezier
curve of degree one given by
ro(tj 1) = (1- t·)po + tpi = (1 - t + Et)po' ( 12.5)

We use the shift operator E for concise description.

Similarly, from the definition of the rational dividing point, a line POPI is
represented by a.n equation
(l-t)wOPO+tWIPl
ro (t; 1) = () , ( 12.6)
1 - t Wo + tWl
which we call a rational Bezier curve of degree one. We use the notation ro(t; 1)
for it. Since there are relations Po = WoPo and PI = WIPI together with equations
(12.5) and (12.6), we have
ro(t; 1) = {(I - t)wo + tWl}rO(t; 1).
We use the notation wo(t; 1) for the weight of a point ro(t; 1):
Wo(t; 1) = (1- t)wo + tWl = (1- t + Et)wo. (12.7)
Substituting the above equation into eq. (12.6), we have
. (. ) _ (1- t + Et)(woPo)
(12.8)
Iot,l- Wo (
t; 1
).
Equation (12.8) is a rational Bezier of degree one constructed from rational

Beziers of degree zero, Po = ro(t; 0) and PI = rl (t; 0), together with their weights.
In .Fig. 12.3, we show a construction process of a rational Bezier of degree three
from four control points (Po, PI' Pz, P3) with their weights (Wo WI W2 W3)' Two
rational Beziers of degree two are produced during the process.
1. From the control polygon with its weights and a virtual origin 0, make its
canonical perspective (Po PI P2 P3)' See Fig. 12.3(a).
2. Determine a point on a Bezier curve P from the canonical control poly-
gon as shown in Fig. 12.3(b). Three points Pa, Pb and Pc are linear in-
terpolated points on the edges (POPl)' (PIP2) and (PZP3)' These points
are ro(t; 1), rl(t; 1) and r2(t; 1). Two dividing points Pd and Pe' which are
ro(t; 2) and rl(t; 2), are linea,r interpolated points on (PaPb) and (PbPJ.
Then a point P, which is ro(t; 3), is a linear interpolated point of (PdPe)'
3. Referring to Fig. 12.3( c), take the corresponding points Pa' Pb, Pc, Pd, Pe and
P on the edges of the rational control polygon and its derived edges. Then
P is a point on a rational Bezier curve ro(t; 3), and Pd, P e are ro(t; 2) and
rl(t; 2),and Pa , Pb' Pc are ro(t; 1), rl(t; 1), rz(t; 2).
Since the weights of the points Pa,Pb and Pc are wo(t; 1),Wl(t; 1) and W2(t; 1),
the weights of the points Pd and Pe are wo(t;2) and wl(t;2), whichare the lin-
ear interpolations of wo(t; 1) and Wl(t; 1), and Wl(t; 1) and wz(t; 1) respectively.
Rational
/ Bezie, polygon
Canonical
/ perspective
Po
P3
P3 ...... /
\'. ... . ./ / 0) /
1"'- . . . ... . ' .,/ 1
/
(a)
Po Po
P3 P3
(b) (c)
Figure 12.3 (a) Rational polygon and its canonical perspective, (b) determining a point
from Bezier polygon of degree three, (c)corresponding point of rational polygon
Similarly that of pis wo(t; 3), which is the linear interpolation of wo(t; 2) and
w}(t; 2) . So we can use the expressions:
wo(t; 2) = (1 - t)wo(t; 1) + tWl(t; 1) = (1- t + Et)wo(t; 1),

wo(t; 3) = (1- t)wo(t; 2) + tWl(t; 2) = (1 - t + Et)wo(t; 2).
Hence, we have
(1- t)wo(t ; 1)ro(t; 1) + tWl(t; 1)rl(t; 1)

ro(t; 2) =
(1 - t)wo(t; 1) + tWl(t; 1)
(1- t + Et)(wo(t; 1)ro(t; 1))
=
wo( t; 2)
(1- t + Et)2(woPo)
(12.9)
(1- t"+ Et)2wo
(1- t + Et)(wo(t; 2)ro(t; 2))

ro(t; 3)
wo(t; 3)
(1- t + Et)3(WOPO)
(12.10)
(1 - t + Et)3 wo
With a similar procedure, we can obtain a rational Bezier curve of degree n

from two rat.ional Bezier curves of degree n - 1:
(1 - t + Et)(wo(t; n - l)ro(t; n - 1))

ro(t; n) =
(1 - t + Et)wo(t; n - 1)
(1 - t + Et)n(wopo)
(12.11)
wo(t; n)
where
wo(t ; n) = (1 - t + Et)nwo. ( 12.12)
Since the graphical construction of a rational Bezier curve corresponds to that of
a linear Bezier curve, we can easily understand the effects of weights of vertices
of t.he rat.ional polygon using its canonical perspective.
12.2.3 Effects of Weights

A rat.ional Bezier curve is a linear combination of the weighting functions mul-
tiplied by t.heir respective control points. The weighting function of the j-th
control point P J is the coefficient of Pj in eq. (12.11) which is given by
and the summation of all the weight.ing functions for j = 0,··· n is 1. The
denominator wo(t; n) is considered a scalar Bezier curve, whose control points
are determined by values (wo , WI , ... , wn) given at t = 0 , (l/n), ... ,
« n - 1) / n) , 1. See Fig. 12.4. If the weights are equal, the denominator becomes
a. const.ant.. Then eq. (12.11) becomes a usual Bezier curve.
Since all the weights are assumed positive, each of the weighting functions for
the cont.rol points has positive values less than one, but it is not always of single
peak type, though the summation of the functions is equal to one. Anyway a
rational Bezier curve is in a convex hull of its control points by the same reason
as that for non-rational one. If a certain Wj is much larger than the others, then
the denominator becomes large at t = (j/n) and the curve r(t;n) approaches
P J whose weight is w)' This is more precisely explained in the following. If
we increase Wj to Wj + 6wJ and let ro(t; n) become r~(t; n), then eq. (12.11) is
transformed to
I ( . ) _ ro(t; 11) + ,),6wjpj
ro t, n -:- (12.13)
1 + ')'UWj
A ,
W
WI
./.,.,.........,."",."".,
~ ' .....
Wo ........ ,~.. _~~,,-,,---""
! W2
o I
:3
2
:3 1 Figure 12.4 Behavior of a weight
function of degree three
PI P3
~
\ 1
Po T~(t;:)\~
\ , -yAWl
TO(t; ~;\)
Figure 12.5 Variation of shape of a rational Bezier curve with weight
where 'Y =Cj U(l - t)n- j /wo(t; n). The right-hand side of this equation can be
interpreted as a. rational interpolation between the original ro(t; n) and a control
point. Pj' that is, division in the ratio ("(6wj : 1). See Fig. 12.5.
Accordingly, all the points of the original curve move towards Pj' Whereas
if a control point. Pj moves by 6pj' all the points move in the same direction as
that of 6pj.
So far, we have assumed that the weights are greater than zero, but here we
treat a case of a weight of zero. If Wi = 0, then the value of the denominator
becomes small at t = (i/n), but always positive. Though the control point Pi
has no influence on the shape of the curve, the summation of the weighting
functions of the other control points is equal to 1. So the generated curve t(t, w)
is cont.ained in a convex hull made by the control points without the vertex Pi'
Now we consider a case in which a control point Pi is moved to infinity in a
direct.ion 6Pi as t.he weight Wi approaches zero, while a product Wi Pi converges
to a. constant relative vector Lpi' Then a relative vector Lr(t) which is equal
to Lpi multiplied by its weight.ing function affects the shape of the curve r in
t.he following way,
r(t, w) = f(t, w) + Lr(t), (12.14)
where Lr(t) = Lpi(Ci)(1 - t)itn-i/wo(t;n). An example of the use of zero
weight is given in Sect. 12.6.2.
For curves of degree two, the weights classify t he curves into an ellipse, a
parabola. and a hyperbola. This is discussed in more detail in Sect. 12.6.
12.2.4 Division and Degree Elevation

Division and degree elevation of a rat.ional Bezier correspond to those of its linear
one, which are easily performed. When fo(t; n) is divided at t = r int.o two Bezier
curves f~(t; n) and 'f~(t; n) , the locations of their control points are given by, from
equat.ions (9 .24) and (9.25),
pi = ro(r;i) = (1- r+ Er)ipo,

P~-i = rn( r ; i) = {(I - r)E- 1 + r fPn, (12.15)
0, . .. n.
A rational Bezier curve ro(t; n) is divided at t = r into two rational Beziers

r~(t; n) and rg(t; n) , the locations of whose control points are easily determined
from eq. (12.15) or by the graphical method.
pi ( .') _ (1 - r + Er)i wopo (12.16)

r o r , 2 - ( 1- r + E)"
r 'wo
r (7' i) = {(I - r)E- + r }'wnPn
1
P~-i n, {(1-r)E- 1 +r}'wn
Their weights are
wf(r;i) = (l-r+Er)iwo,
w~_i(r; i) = {(I - r)E- 1 + rrw n.
For n=3, the locations of the control points of divided rational Bezier polygons
are shown in Fig. 12.3. They are (Po, P a , Pd, p) and (p, P e, Pc, pJ, co~responding
to (Po, Pa, Pd, p) and (p, Pe, Pc, P3)'
For elevation of degree of a rational Bezier, since new control points P; of
the corresponding linear Bezier are given from eq.(9.50) by
• I
Pi = n +1 1 { 2Pi-l
.•
+ (n + 1 - ')'}
2 Pi , i = 0, ... n + 1,
.........;:... =
··~·;--P3 P4
=
P3 P4 Rational
Canonical Bezier polygon
perspective
I"igure 12.6 Degree elevation of a rational Bezier polygon
their rational equivalents P: become

iWi-1Pi_1 + (n + 1 - i)WiPi
I
Pi (12.17)
iWi-1 + (n + 1 - i)Wi
iWi_1 + (n + 1 - i)Wi
,
Wi
n+1
= 0, .. . n + 1.
Alternatively, they are easily determined by the graphical method. See Fig. 12.6.
12.2.5 Derivatives at Ends of a Segment

The i-th derivat.ives at an end point of a linear Bezier curve are expressed by
i + 1 control points from the end. In a rational Bezier, the same statement holds,
but its derivatives cannot be expressed by simple projection of those of the linear
Bezier curve.
To avoid differentiation in a rational form, we change eq. (12.11) into the
form:
(1- t + Ettworo(t ; n) = (1- t + EttwoPo. (12.18)
Differentiating eq. (12.18) i times by t and setting t = 0, we have
,(il ~ i n! n! i
WOIO + j=l n J.
j .i-j _
~CJ( _ .),(E -1) Wolo - ( _ .),(E -1) woPo ·
n z.
From the above relation, ro , r~1), r~2) , ... can be obt.ained successively.
ro = Po,
12.3 Rational Bezier Patches 191
'Wor~1) + n( WI - wo)Po = n( WIPI - WoPo),

wor~2) + 2n2(Wl - wo)WdWO(Pl - Po) + n(n - 1)(w2 - 2Wl + wo)po
= n(n - 1)(w2P2 - 2WIPI + WoPo)'
Then the derivatives at one end t = 0 are:
ro = Po,
(1) WI
ro n-(Pl - Po) (12.19)
Wo
(2) W2
ro = n(n - 1)-(p2 - PI),
Wo
{2nwr - (n - l)WOWI - 2wowd( )
+n 2 PI - Po . (12.20)
Wo
With a similar manipulation, the derivatives at t = 1 are given by
ro p",
r(l)
o n - - (p" - P,,-l )
W,,-l (12.21)
Wn
r~2) = -n( n - 1) W n -2 (Pn-2 - Pn-l),

Wn
{2nW~_1 - (n - l)w n - 2wn - 2Wn -lW n } ( ) ( )

+n 2 Pn - Pn-l . 12.22
Wn
The tangent direction at an end of a rational Bezier expression is that of

the first control edge of its polygon. The direction of the second derivative is
determined by the difference of the second and the first edges, each multiplied by
a. different factor. Hence, its direction is not the same as that of the non-rational
Bezier expression. Its magnitude of curvature at the end point is equal to that
of a non-rational Bezier multiplied by a factor (wowdwi).
12.3 Rational Bezier Patches

Since a linear Bezier patch of degree one is given by
Soo(u,v; 1, 1) = (1- u+ Eu)(l- v + Fv)poo, (12.23)
its rational equivalent is
( 'll)_(l-u+Eu)(l-v+Fv)wooPoo
800 u, v,, - (1 - u + Eu )(1 - v + Fv )
Woo '
(12.24)
where {Pij} for i, j = 0,1 is its control point matrix and {Wij} is its corresponding
weight matrix. See Fig. 12.7. Equation (12.23) is a canonical perspective of
eq. (12.24). Locations of the control points Pij of eq. (12.23) are determined on
the line joining a virtual origin 0 'and Pij so as to make OPij : OPij = 1 : Wij'
192 12 Rational Bezi er and Spline Expressions
Rational Bezier
of degree one
p,/1
/ Pu
I P ll .
; .I
Canonical
perspective
Figure 12.7 Rational Bezier patch of degree one
Equation (12.24) is derived also in the following way. A rational line segment
800(0, v) joining Poo and POI is given by
(1 - v + Fv)wooPoo
800 ( 0, V ) = (1-v+Fv)woo , (12 .25)
and anot.her rational line segment 800(1, v) joining PlO and Pll is
(l-v+Fv)WlOPlO
800 ( 1, v ) = . (12.26)
(1 - v + FV)WlO
For v = canst a rational line segment. joining 800(0 , v) and 800( 1, v), whose weights
are (1- v + Fv)woo and (1- v + FV)WIO, is given, referring to eq. (12.8) formaIIyby
(l-u+Eu)(wb(v)p~(v))
800 (u, v) = -'---;----'-:::--:'-'--:~:-:--'-'-- (12.27)
(1- u + Eu)w!J(v)
where
(1 - v + FV)(WiOPiO)
(1 - v + FV)WiO
= (l-v+Fv)wio(i=O , l).
Substit.uting the above equat.ions into (12.27), we obtain eq. (12.24) .

For n = 2, we make a rational Bezier patch of degree one whose four vertices
are given by Soo(u,v;l, 1),81O(U,v;1, 1) , 801(U,V; 1, 1) and 811(U , v; 1, 1), that is,
12.4 Rational Splines 193
each point is on one of four adjacent rational Bezier patches of degree one, with
weights WiJ( u, v; 1,1) , i, j = 0, 1. This makes a rational Bezier patch of degree
two:
( ) (1 - u + Eu)(l- v + Fv)(woo(u, Vj 1, l)soo(u, Vj 1, 1)) (12.28)

soo u,v;2 , 2 = ( 1 - u + Eu )( 1 - v + Fv ) Woo ( u, Vj 1, 1)
where woo(u, v; 1, 1) = (1- u + Eu)(l- v + Fv)woo. The same expression can be

obtained from the tensor product of two rational Bezier curve of degree two.
We make a general form of a rational Bezier patch of degree n when its
control point matrix {Pij} and weight matrix {Wij} are given. Since a canonical
perspective of a Bezier surface patch is
the corresponding .rational Bezier surface patch is given by
(u, v; n, n ) = (1- v + Fv)n(1 - u + Eu)n(woopoo)

( ( ) . (12.29)
1 - v + Fv)n 1 - u + Eu nwoo
Soo
Figure 12.8 shows examples of the shape of rational Bezier patches of degree
three. The contour curves are shown to indicate variation of shapes according to
the variation of values of weights at the different vertices. Effects of the weights
on its shape are similar to those in the rational Bezier curve: each point on the
patch is moved in the direction of a control point with increased weight.
1Z.4 Rational Splines
12.4.1 Rational B Polygons from a Rational S Polygon

From a spline polygon of degree n with its scale list, we can produce a series
of Bezier polygons of degree n connecting in c(n-l). To distinguish a rational
S polygon from a usual spline polygon, vertices of the latter are denoted by q
instead of q. As explained in Chap. 11, locations of vertices of these B polygons
are det.ermined from those of the S polygon(··· Ct-l' Cto, Cti> <b," .). Nowattach-
ing a weight k i (to distinguish the weights of vertices of a rational S polygon
from the weights Wi of vertices of Bezier polygons, different symbols k i are used)
to each vertex qi of t.he rational S polygon we can deduce a series of rational
Bezier polygons of degree n connecting in c(n-l).
Though we can use a canonical perspect.ive of a rational spline polygon as
we did in a rational Bezier, we utilize directly the result of the rational division
for locating the rational B polygons from the rational S polygon.
As explained in Sect. 12.3 and here, locations of Bezier control points can be
det.ermined by repeated linear division of Menelaus edges of the given S polygon.
Similarly, locations of rational Bezier control points are determined by repeated
rat.ional division of rational Men~la.us edges of the rational S polygon. The
,····--·---·----·,------·--·-·.-,-------··- -·.·' 1
t.__._ . __. t. ... _._._ .. _._ .. ~._. ___ .. _._. ___ l
(a) control net and surface (b) all weights are equal
i..··········----:----·--------i--··········---j
,
!DJ'~
1ifS}-.; l
!~!
: , , ;
:_____________.;___________ .. ___.. ______ ... _____ .i ~_ ... __ ... _.. __..:.. ___ .. ____ .__ .. _; __ ... _... _______i
(c )increase weight of Pn (d) increase weights of Pn and P22
Figure 12.8 Effects of weights on shape of a rational patch of degree three. Contour curves
and the control poin t( s) of increased weigh t( s) indicated by dots are shown
rat.ional Menelaus edge is similar t.o the Menelaus edge and its end points are
rational dividing point.s of the rat.ional Menelaus edge of one level lower. The
rat.ional dividing point. has a weight. which is determined by linear division of the
weights of t.he end points of the edge. The location of a ra.tional dividing point is
at the dividing point in the ratio (w' a : wb) of the edge, when its corresponding
linear dividing point divides the edge in the ratio (a : b) and has the weight 1.
For inst.ance, t.he rational dividing points qO.1 and QO.2 of an edge of a rational
S polygon of degree three are located at the dividing point.s of ratios (k 1 A_2 :
kO(>'-l + '\0)) and (k1(A_2 + A-d : ko'\o) of the edge (QOQ1)' Their weights are
the dividing value in t.he ratios ('\-2 : ('\-1 + AO)) and (('\-2 + '\-1) : '\0) of ko
and kb which are t.he weights of the vertices Qo and Q1'
Accordingly, formulas of weights of the rational B points have the same form
as those of the B points derived from the linear S polygon, but the symbols of
vert.ices are replaced by its weights of the vertices. Formulas for the rational B
point.s are similar to those of the B points from the linear S polygon, in which
the linear dividing ratios are replaced by the rational dividing ratios. We give
an example for n = 2.
Referring to the formula given in Chap. 11 and Fig. 12.9, vertices of a Bezier
12.4 Rational Splines 195
....
Rational
spline polygon
..... \ i ! ..."" Canonical
....:~/ .. perspective Figure 12.9 Connected rational
Bezier polygons and curves of degree
two from a rational spline polygon
polygon of degree t.wo are determined from the unit 5 polygon of degree two:
.Aoqo + .A-I ql
Po
.Ao + .A-I
PI ql' (12.30)
.AI ql + .AOq2
P2 =
.AI + .Ao
Replacing Pi by ki ' we obtain
.Aoko + .A_lkl
Wo
.A-I + .Ao
WI = kI, (12.31)
.Alkl + .AOk2
W2 =
.Ao + .AI
We obt.ain t.he locations of the rat.ional Bezier control points from eq. (12.30)
by replacing the linear dividing ratios with the rational ones .
.Aokoqo + .A-l k lql

Po
.Aoko + .A_lkl
PI = ql' ( 12.32)
.Alkl ql + .AOk2q 2
P2 =
.Alkl + .AOk2
If a weight kl of the vertex ql increases, Bezier points PO,P2 approach ql
and their weights Wo, W2 increase. This makes the curve ro( t; 2) shrink and its
adjacent L l and rl grow. An example is shown in Fig. 12.10.
Since the relat.ion between a unit 5 polygon of degree n a.nd its derived B
polygon is formulated in Chap. 11, it is easy to write down its rational equivalents
such as eq. (12.31) and eq. (12.32) for any degree n . Applicat.ion of the graphical
method for det.ermining points on a· rational spline curve is also easy.
Figure 12.10 Shape variation of a

spline curve with an increasing weight
kl
Rational Bezier curves derived from a rational S polygon of degree n connect

in e(n-I). This is proved as follows. In the canonical perspective, ern-I) con-
nection of Bezier polygons of degree n is guaranteed by the S polygon of degree
n. There is a con.nection defining polygon (CDP, refer to Sect. 10.2.4) of degree
n - 1 at each junction of the Bezier polygons, which can be constructed from
the S polygon. Division of the eDP determines the Bezier points of the ern-I)
connecting B polygons . These B points together with the S polygon are mapped
to the corresponding rational B points and the rational S polygon. Accordingly,
the rational CDP guarantees e(n-l) connection of rational B polygons derived
from the rational S poly.gon. Of course, the condition of e(n-l) connection is
also proved by manipulation of the expressions of rational Beziers deduced from
the rational S polygon.
12.4.2 G(2) Connection of Curves from a Rational S Polygon
If two rational Bezier polygons of degree three connect in 0<2), the condition of
curvat.ure continuity at the junction of Bo and Bl gives the relation from the
derivatives given in Sect. 12.2.5:
This equat.ion is similar to eq. (10.27) of Sect. 10.4. Therefore, we obtain an

equation similar to eq. (10.30) of Sect. 10.4. From this relation we can deduce an
arbit.rary const.ant (J by which locations of the control points can be adjusted.
If we change t.he linear dividing ratios from (A-2 : A-I: '\0) and (A-I: Ao :
Ad to ((JoL 2 : A-I: (JIAO) and ((JIA-l : Ao : (J2Al), where (Ji are the stretch
fact.ors, t.hen by t.he rational division of the edges of level zero using the above
revised rat.ios we determine the rational dividing points of level one and their
weights. And we determine t.he rational B points of level two considering that
their corresponding linear dividing ratios are (A-I: Ao) and (Ao : AI)'
Effect.s on derived shapes of displacing control points by changing values of
(Ji are similar t.o t.he non-ration a.! case. See Fig. 12.11.
12.5 Rational Spline Nets and Bezier Nets 197
(a)O"=l
(b) 0" = 0.3
(c) 0" = 3.0
Figure 12.11 G(2)connection of rational Bezier curves, showing variation of radius of curva-
ture profiles
12.5 Rational Spline Nets and Bezier Nets
From a rational spline net we can determine its canonical perspective from which
we obtain Bezier nets. Accordingly, we can easily determine their rational equiv-
alent.s. Or we can use the tensor product method for the derivation of a rational
spline surface. Anyway we can determine rational Bezier nets connecting in
c(n-l ) with t.heir adjacent rational Bezier nets.
An example is shown in Fig. 12.12. Given a vertex matrix {qi) and a weight
matrix {kiJ of a. rat.ional spline net of degree 2 x 2, we deduce those of a rational
Bezier net of degree 2 x 2 by a tensor product scheme. From each column of
{qij} and {kiJ, for j = 0,1,2, we obtain intermediate Bezier points P:j and their
k:
weight.s j .
I AokojqOj + A-l kljqlj

POj
Aokoj + A-I klJ
= ql;'
I
Plj
Bezier patch Q20
Figure 12.12 A rational Bezier net of degree 2 x 2 from a rational spline net of the same
degree
).l kljQlj + ).okzJ~j

).1 k1j + )'okZj
).okoJ + ).-1 k1j
k'OJ --
).-1 + ).0
Wlj,
).1 klJ + ).Ok2j
).0 + ).1
. Bezier points {PiJ of the patch and their weights {Wi;} are determined
from each row of the array of intermediate points and their weights obtained
above, which we consider as rational spline polygons with the common scale list
(/-l-I, /-lo , ttd for i = 0, 1,2:
/-lokIopio + /-l-l k IIPil

PiO =
/-lokio + /-l-l k Il
Pil P:l
/-l1 kil Pil + ttokizpiz
PiZ =
/-llkil + /-lokiz
/-lokIo + p-l k il
WiO =
/-lo + P-l
Wil k: 1,
Pl k il + ttoki2
WiZ =
PI + Po
Using the above data, the rational Bezier patch of degree 2 x 2 is given by
( . 2 ) _ (1- u + Eu)Z(1 - v + Fv)ZwooPoo

Soo u, v, ,2 - ( 1 - u + Eu ) Z( 1 - v + Fv ) zWoo (12.33)
12.6 Expressions for Conics 199
This patch connects with the adjacent patches S2i,2j( U, v), for i, j = -1, 0, 1,
in C(1) . A t.echnique similar to the above is applicable for the derivation of a
rat.ional Bezier pat.ch of higher degree from a rational spline net of the same
degree. Figure 12.13 shows a rational spline net and its generating rational
Bezier patches connected in C(l). Projections of boundary curves together with
paramet.ric curves of t.he Bezier pat.ches are also shown. Though they are very
dist.ort.ed by values of the weights, the contour curves of the surfaces are not so
dist.ort.ed and the shape variation is smooth. This shows that we cannot recognize
t.he shape of surface only from its parametric curves.
Figure 12.14 shows a composit.e surface consist.ing of a surface of revolution
smoothly connecting with a surface of elliptical cross section. This is an example
of ability of the rational spline surfaces.
12.6 Expressions for Conics

12.6.1 Conversion to an Implicit Form
Now we examine the case of a rational Bezier expression of degree two, which is
given by
( . ')) _ (1 - t)2woPo + 2t(1 - t)WIPl + t2w2P2

r t, - - (1 _ t)2wo + 2t(1 - t)Wl + t 2W2 .
To observe effects of weights, transforming the above equation by
(12.34)
and then replacing r by t in the above equation, we have
r(t w) = (1 - t)2pO + 2t(1 - t)WPl + ep2 (12.35)

, (1 - tF + 2t(1 - t)w + t 2
We change the not.ation from r(t; 2) to r(t, w), beca.use the single weight w is the
import.a.nt. parameter and the degree two is already assumed.
The three weight.s are reduced to one. Now eliminating a. pa.ramet.er t , we
make an implicit. expression of eq. (12.35) in terms of x and y. The general
met.hod is explained in Sect. 5.6. For convenience, we int.roduce a variable U
which replaces t:
t
U=--. (12.36)
1-t
Then eq. (12.35) is changed to
(12.37)
(a) spline net (b) derived Bezier patch
(c) k22 = 1.0
(d) k22 = 0.3 (e)k22 = 3.0
Figure 12.13 Effects of a weight k22 on the shape of a patch. Dotted lines are parameter
lines, solid lines are boundary curves of a patch. Shapes of the patches may be grasped from
their contour curves, but not from their parametric curves
Let coordinates of t.he points in eq. (12.37) be
0=(0,0), Po = (-c,O), Pl=(b,h), P2=(C,0), r=(x,y). (12.38)
Resolving eq. (12.37) into its components and rearranging them, we obtain
u 2 (c-x)+2wu(b-x)-(c+x) 0, (12.39)
u 2 (-y)+2wu(h-y)-y = O.
Considering u2 and u as unknowns, we get the solution of eq. (12.40):
u2 = '2w{(c+x)h-y(b+c)}/D, (12.40)
(a) Oblique view (b) Top view
Figure 12.14 Smooth connection of a surface of revolution with another surface
u 2ey/D, (12.41 )
where D 2w{(e - x)h + y(b - en.
Eliminating u from the above equation, finally we obtain
(12.42)
This represents a. general conic equation which is descdbed in Chap. 4.
12.6.2 Classification by Weight

Since the discriminant of this quadratic equation is given by e2 h2 (I/w 2 - 1), the
conic· it expresses is an ellipse for w < 1, a parabola for w = 1 and a hyperbola
for w > 1. By differentiating eq. (12.42) with respect to x and setting it to zero,
we obtain a relation hx = by. This means the extremum value of y is on the
median OPl and its extension. See Fig. 12.15. Its value Yo and the corresponding
value to are obtained by using equations (12.42) and (12.41).
wh 1
Yo = 1 + w' to ="2. (12.43)
In eq. (12.42), w is not necessarily limited to a positive value. Equation (12.35)

represent.s a curve segment for w > o. If we want to extend the segment, we
can do it by changing w to -w. The extended curve segment is also a part of
the conic expressed by eq. (12.42). We can express this part together with the
original segment. in the parametric form of eq. (12.35) with new control points
Po, Pl and P2· This technique is treated later in Sect. 4.4.2.
We examine the effects of the weight w on the shape of curve of eq. (12.42).
Refer to Fig. 12.16 .
• w > 1 or w < -1 : the coefficient of y2 is negative, so the curve is a hyperbola,
a.nd the y coordinate values Yo of its intersecting points with the median are,
:,,'
:
,/
/~
// "'/ _i___
, ,,/
------ - ---- - ---- - - - ---------~------ - - - - - ----- - - - --- - ---- -

Po = (- e,O) ([,0) P2=(e,0)
Figure 12.15 A rational Bezier control polygon of degree two and a conic segment
/'
Iwl > 1 j
(hyperbola) j PI
"\r / '\""""
------------ --- --------,------ --------- ---------
: l'
Iwl < 1
~ Iwl =
j (ellipse)
1 (parabola)
Figure 12.16 Conics and their weight
if 00 > w > 1, between hand 00, and between hand h/2,

if -1 > W > -00, between h/2 and h, and between 00 and h.
• w = ±1 : the curve is a parabola and crosses the median at Yo = h/2.
• -1 < w < 1 : the curve is an ellipse, its intersecting points with the median
are Yo = wh/(1 + w).
• When an absolute value of w is 00 , the curve expressed by eq. (12.42) becomes
{h(x - e) - y(b - e)}{h(x + c) + y(b + en = 0,
which is two straight lines through (Po , PI) and (PI ' P2) '
The conditions that the ellipse given by eq. (12.42) is a, circle are that the
h
~/,;I
d,//_ _~ w=~
/
-c
w =-~
Figure 12.17 Representation of

circular arcs
coefficients of x 2 and y2 are equal and that of xy is zero. Hence we obtain

h2 = c2 (1lw 2 - 1) and b = o. Let d be distance between Po and PI' then we
have
c c
w ± =±- (12.44)
../h 2 + c2 d
ch -ch
Yo = c+d or d - c· (12.45)
If 11/ = cld, eq. (12.3.5) expresses the upper part. of a circle above the chord POP2'
and if w = -cld, it. represents the lower part of the circle. See Fig. 12.17.
The case of 111 = 0 is treated in another way from the definit.ion of t.he curve.
Eq.(12.42) wit.h b = 0 is
( 12.46)
Keeping the product of wand h equal to a constant c, if we increase h to infinity,

then the above equation becomes: x 2 + y2 = c2 , which is the equation of a circle
of radius c. Thus, we can also make a rational Bezier expression (12.37) express
a circle of radius c with its center at the origin by taking PI to infinity while WPI
converges to a vector ±cj and setting Po and P2 to -ci and ci:
(1 - tFpo + 2t(1 - t)WPI + t 2P2

r =
(1- t)2 + 2t(1- t)w + t2
c{(2t - l)i ± 2t(1 - t)j}
( 12.47)
(1 - t)2 + t 2
Exercise 1. Prove that eq. (12.47) represent an exact circle of radius c.

Exercise 2. Prove that in eq. (12.47), if j is replaced by a vector parallel to OPI '
the resulting equation represents an ellipse with Popz and OPl as its conjugate
axes.
Exercise 3. Prove that a conic is uniquely determined if two points with directions
of tangents there and any other point on the curve are given.
12.6.3 Sphere and Surface of Revolution

A surface of revolution can be expressed by control points of its meridian curve
moving along circular orbits of latitudes. As a simple example, we give an equa-
tion of a sphere [17].
In the xy plane a half-circle of radius r with its center at the origin and its
diameter on t.he y axis is expressed fromeq. (12.47) by
(1 - u)Zpo + 2u(1 - U)Pl + uZPz

r- (12.48)
(1 - u)2 + U Z
~--~~~--~--~~--~~
. - '
where Po = (0, -r, 0), Pl = (r, 0, 0) and Pz = (0, r, 0).

If the cont.rol point Pl of the above equation rotates around the y axis with
a paramet.er v, t.hen r is a function of u and v and represent.s a sphere with it.s
center at the origin. The locus of Pl is given by
(1 - V)ZPl.O ± 2v( 1 - V)Pl.l + VZPl.Z (12.49)

(1- v)2 + v2
. .
where PlO = (r, 0, 0), Pll. = (0,0, r), PlZ = (-r, 0, 0).
Subst.it.ut.ing t.he above expression into Pl of eq. (12.48) , we obtain a spherical
surface:
(1 - U)2po + u 2 pz
s( u, v) -'--:--~:7--~ +
(1 - u)2 + uZ
+ 2u(1 - u)(1 - V)ZPl.O ± 2v(l- V)Pl.l + VZPl.Z
( 12.50)
{(I - u)2 + u2 }{(1 - v)2 + v2 }
To check it.s spherit.y, we form the squared sum of the x, y and z components of
eq. (12.50) and obt.ain rZ which is equal to the squared value of the radius of the
sphere.
If three point.s Po , Pl and Pz of eq. (12.48) rotate around a certain axis, a
torus surface is constructed. If the control points of a plane curve are rotated
around an arbitrary axis, then a surface of revolution is easily produced.
12.7 Interpolation and Extrapolation with Conics

12.7.1 Weight of a Control Point and Parameter Values
To make t.he geometric structure of r(t, w) clear we change the right-hand side
of eq. (12.3.5) into the following form. Refer to Fig. 12.18. .
12.7 Interpolation and Extrapolation with Conics 205
PI
Pc = r(t, w) ,/
,X---
..-./
, /'
, /'
..-
/
..-
Po
Figure 12.18 Struc~ure of a conic segment
(1 - t? + t2 (1 - t)2pO + t 2P2
1'(t, w) = (1 _ t)2 + 2wt(1 - t) + t2 (1 - t)2 + t2 +
2t(1- t)WPI
(12.51)
(1- t)2 + 2wt(1- t) + t 2'
Let Pc be a point on 1'(t, w) and q a dividing point of t.he chord POP2 in the
ratio t2 : (1 - t? The second factor of the first term of the above equation is
equal t.o q . Accordingly, the right-hand side of eq. (12.51) gives a dividing point
Pc petween PI and q in the ratio (1- t)2 + t 2 : 2t(1- t)w.
We can determine a value of a parameter t of a point Pc on a curve and
t.he weight of the curve from length ratios Poq : qp2 (= m : n) and PIPc : P8
(= m' : n') :
..;;n;
= ( 12.52)
..;;n; + ...fni'
w = (-) ~-+ -m,} '
n {n'
2m rg (12 ,53)
If q is the mid-point of POP2' a vector Pcq is a camber vector of the curve with
the chord POP2' This gives the ext.remum y value.
12.7.2 Division of a Rational Polygon

We divide the rational Bezier curve of eq. (12 .35) at t = 7 into two parts 1' a and
1'b wit.h t.heir control polygons (Po PI P2) and (pg pt p~) . Their locations and
weight.s given from eq. (12.16) are
p~ = 1'o(7jO) = Po,
(l-r+Er)wopo
r ( T" 1) - -'-:-_ _----::::-'-:-_.0-
p~ ( 12.54)
o , - (l-r+Er)wo '
( ) (1 - r + Er)2 wopo
p~ = ror;2 = (l-r+Er)2wo'
whose weight.s are
w0a 1,
w1a (1 - r) + wr, (12.55)
w2a (1 - r)2 + 2wr(1- r) + r2 ,
and
{(I - r)E- 1 + rFw2P2

pg r2(r; 2) =
{(1- r)E- 1 + rpw2
,
{(I - r)E- 1 + r}w2P2
p~ = r2(r;l) = , (12 .56)
{(1- r)E 1 + r}w2
p~ = r2(r; 0) = P2'
whose weight.s are
w0b = + 2wr(1- r) + r2 ,
(1 - r)2
w 1b = (1- r)w + r, (12.57)
w2b 1.
In t.his section t.he weight of the first and third control points of the rational
Bezier polygon of degree two must have the value 1. Accordingly, we have to
make a correction for the expression of the weights given by equations (12.55)
and (12.57). By the relation (12.34)' the weights (12.55) and (12.57) are changed
to
wr + (1 - r)
wg = 1, w~ = -,=.============2 w~ = 1, (12.58)
)(1- r)2 + 2wr(l- r) + r '
r+w(l-r)
wg = 1, w~ = -,=.============2 W 2b -- 1• (12.59)
)(1- rp + 2wr(l- r) + r '
12.7.3 Extension of a Curve Segment
When we try t.o fit. a. given sequence of points by connected curve segments of
conics, we make a segment as large as possible within a given tolerance to reduce
the number of conics, whose information has to be stored [34][35].
Accordingly, we consider extension of a curve segment which has been deter-
mined. Refer t.o Fig. 12.19. The segment r(t, w) starts from Po and ends at P2 in
Figure 12.19 Extension of conic segment
the convex hull (PO-Pl P2)' If we change the sign ofthe weight., the curve r(t, -w)
start.s from Po and ends at P2, but. it is outside of the convex hull. Hence, if we
want. t.o ext.end the curve segment r(t, w) beyond P2 in the same conic, we use
r(t, -w) from t = 1 to t' which corresponds to a segment from P2 to p~. We
want to combine the original curve r(t, w) from t = 0 to t = 1 and the extended
segment r(t, -w) from t = 1 to t = t' into one segment re(t, we) with its control
points (Po p~ p;) and the weight we of the second control point p~. Suppose
that value of t' has been determined, then a procedure to determine re(t, we) or
p~ and we is given in the following:
1. The three control points (Po p~ p~) are obtained by the division of r(t, -w)
. at t = t'. They a.re given by eq. (12.54) with T = t' and Wl = -w.
2. The weight we is equal to -wi, which is given by (12.58), with T = t' and
the sign of w reversed.
-wt' + (1 - t')
we = -
/(1- t')2 - 2wt'(1 - t') + t'2 . (12.60)
The reason is as follows. At t = t' we divide r(t, -w) into two parts, ra(t, WO)
starting at Po and ending at p~, and rb(t, w b), which starts at p~ and ends at P2
and is the extended part adjacent to r(t, w). Therefore, a curve segment whose
weight is equal to -Ix weight of ra(t,w a), that is, ra(t, _w a), starts at Po and
passes through P2 and ends at p~. This is the curve segmentre(t, we) which we
want.
To underst.and this dividing processes we can use the canonical perspectives
of these rat.ional Beziers. Refer to Fig. 12.20( a). We set the virtual origin 0 at the
mid-point between Po and P2' We construct a (l/w)-times magnified canonical
Bezier polygon (Po Pl P2) of the rational Bezier polygon (Po Pl P2) with respect
to the virt.ual origin O. Then we have
oPo = (l/w)opo, OP2 = (l/w)op2'
OPl = (l/wjwoPl = 0Pl'
(a) Division of positive weight curve
PI
(b) Division of negative weight curve
Figure 12.20 Division of a rational polygon with a positive or a negative weight
The (l/w)-times magnification used above is just for convenience.

By the rational mapping from a ca.nonical Bezier curve ro(t; 2) produced from
the ahove polygon to a rational one r(t; 2), we obtain the rational Bezier curve
starting from Po and ending at P2 in the convex hull (Po PI P2). This is r(t, w).
Division of t.he B polygon (Po PI P2) at t produces two polygons (Po Pa pJ and
(Pc Pb P2)' Their inverse mapping produces the divided ra.tional B polygons
(Po Pa pJ and (Pc Pb P2) of the ra.tional B polygon (Po PI P2)'
A similar dividing procedure is applied for the rational B polygon with a

negative weight. Now t.o make r(t, -w), we construct the (l/w)-times magnified
canonical Bezier polygon with the weight of its second cont.rol point changed
from w to -w. Refer to Fig. 12.20(b). Then we have
OPo (l/w)opo , OP2 = (1/w)op2'

op~ = (l/w)( -W)OPI = -OPI'
The points Pa and Pb are the dividing points of PoP~ and P~P2 in the ratio
t : 1 - t . And Pc is the dividing point of PaPb in the same ratio.
The intersecting point of oPa and the extension of PIPO is Pa'
The int.ersecting point of OPb and the extension of PIP2 is Pb'
The intersecting point of PaPb and the extension of oPc is Pc'
The polygons POPaP c and P cPbP2 are the divided polygon of POPIP2 with
the weight -w in' the ratio t : 1 - t. They are the divided rational B polygons
a
( Po PIa P 2a) an d (b
Po PIb P2b) .
12.7.4 Distance Between a Conic and a Point Near It

When a. given point whose coordinate values are (~, 1]) is near a conic curve
f(x, y) = 0, we can evaluate its nearest distance from the curve.
Let us define differences ox, oy and a criterion F to be minimized:
Ox x- ~, oy = y - 17, (12.61 )
F ox 2 + oy2 + 2)...f(x, y), (12.62)
°
where .\ is an auxiliary variable. We assume Ox and oy are small values .
From IJF/IJ(ox) = 0, IJF/IJ(oy) = and IJF/IJ)... = 0, we obtain
Ox +)...fx 0, (12.63)
oy + )...fy 0, (12.64)
f(x, y) = f(~ + ox, 1] + oy)
f(~, 1]) + fA~, 1])ox + fy(~, 1])oy = 0. (12.65)
Eliminating)..., we have
fyox fxoy = 0, (12 .66)

fxox + fyoy = - f, (12 .67)
from which we get
-ffx
ox = f2x + P' (12 .68)
y
. "".
qi = r (O,w ) r(t,w)
qJ --7 ,.
=r(l,w) '.
= 1'(1, -w)t~
'1k:.l ~/ 1'(t,-w)
;.. r
. ·qk = 1'(t', -w)
(a) Extension of curve segment
(b) New segment obtained by curve division
Figm'e 12.21 Extension of a conic segment for minimum number of fitting segments
and from eq. (12.61), the nearest point x, y on the curve is determined. The
distance d is given by
(12.69)
This is used as a measure to select points within a given tolerance.
12.7.5 Curve Fitting by Conics
When a series of points qi of a shape outline is given and we want to make a curve
expression to approximate the shape for storage and processing, then to reduce
the storage space, the curve is approximated by the least number of connected
conics keeping the shape of the curve within a given tolerance. We use the range
extension technique of a conic given in Sect. 12.7.3 and the tolerance evaluat.ion
technique in the previous section.
The out.line of the curve fitting procedure is (refer to Fig. 12.21):.
1. From the given five consecutive points, determine the tangent direction at
the mid-point and set the first control point at the mid-point. Alternatively,
the point.s may be given by the expression for the previous segment.
2. Approximate determination of location of the third control point and the
t.angent. direction there.
3. Determine the second control point from the intersection of the tangents at
the first. and the third control points. Then estimate the weight of the curve
segment. r using eq. (12.53).
4. Ext.ension of the coverage of the conic:
(a) Det.ermine the last point in the given sequence which is within the
t.olerance, using the eq. (12.69).
(b) Revise t.he old control points and the weight for the new rational Bezier
r(t, w) by the method explained in Sect. 12.7.3.
5. Determine th€ tangent direction from dr/dt at the end point of the ext.ended
segment. This is also the direction of the tangent of the starting point of
t.he next segment. Repeat from t.he beginning.
When a point. sequence
is given, a t.angent. direction ti at qi is approximated by linear combination of

two vectors (qi+ 1 - qi) and (qi - qi-l):
where (1- Gi) : Ci is the ratio of contribution of the adjacent chord vectors to the
tangent vector at. their junction qi' The constant Ci is assumed to be given by
The second control point Pl is determined by the intersection of to and

t z . The weight. of the conic is determined from eq. (12.53). We take in the
new point.s t.o ext.end the cornc and calculate the dist.ances d between t.he conic
f(x, y) = 0 and t.he given point.susing eq. (12.69) until the distance exceeds the
tolerance. Then the new cont.rol points and the weight are calculated for the new
rational Bezier r(t, w} which covers qo,"', qj using the method in Sect. 12.7.3.
To increase t.he accuracy of the fitting, we recalculat.e weight values Wi for each
point.s using eq. (12.52) and eq. (12.53) and then we take their weighted mean as
the new weight. The weighted mean is calculated by
Leiwi
w =
Lei
where ei = 8f(Xi, Yi)/8wl w =w;'
Figure 12.22 An example of a vector

font (boundaries of curve segments are
shown with dots)
The value e i i's the sensitivity of pseudo-distance change from the conic
f(x, y) = 0, which is given from eq. (12.42) with the weight Wi . This value be-
comes small near the end control points where error of W becomes large.
Figure 12.22 shows an application of this met.hod to the const.ruction of a
vector font. The initial number of cont.our points produced from the data by t.he
scanner is 69.5 and the number of segments determined by the method is 1.5.
13 Non-regular Connections of Four-Sided
Patches and Roundings of Corners
13.1 Introduction
Usually in design of free-form surfaces, four-sided patches are connected to syn-
thesize required shapes. For this purpose, S nets can be used advantageously
because of their simplicity and the ease of shape control with the automatic
adjustment of defined continuity along the boundaries of the patches.
However, shapes with special features cannot always adequately be expressed
with S nets. For i~stance, box-like shapes, or fillets and roundings along the in-
tersecting boundaries are difficult. To cope with them, we introduce another
technique of smooth connection and synthesis of non-four-sided pa.tches. They
are produced by non-regular connection of B patches, blending of patches, bound-
ary correction pa.tches and triangular patches. These techniques are treated in
this chapter and the following two chapters.
In the next three sections, we describe methods of the non-regular connec-
tion of BEber pa.tches to cover non-four-sided regions. Though only continuity of
the tangential planes on the boundaries between adjacent patches is guaranteed,
the non-regular connection of patches is useful for expressing rounding corners
of va,rious types which appear in products or for aesthetic representation in de-
sign. To easily determine the locations of control points of patches, we introduce
polyhedra as the basic forms, from which the desired shape of patches can be
constructed.
First, we explain notations used in this chapter as well as a basic technique
and its simple application to make a closed surface with the minimum number of
patches of degree two. We keep the degree of patches as low as possible to make
calculation easy. After treating three types of non-regular regions, we explain
the rolling ball method to blend a connecting region with an envelop surface of
a rolling ball.
13.2 General C U) Connection of B Patches
Let two Bezier patches 8( U, v) and 8'( U, v) connect along a common boundary
which is expressed by 8(0, v) or 8'(0, v) and the parameter U of the both patches
increase in the opposite directions along their v = canst curves. The connection
of two B control nets of degree two is schematically shown in Fig. 13.1. We use
the same symbols for the same geometric quantities of the patches, discriminating
one from the other by attaching a ·prime mark.
214 13 Non-regular Connections of Four-Sided Patches and Roundings of Corners
PI2 P'12
a2
S S'
el
PII 81 a'i p'lI

POI
eo
ao Poo a' 0
Figure 13.1 Notation of control
PIO
uLu P'IO points and edges for connection of B
patches
The common B control points of the boundary of both patches a,re POO, POI,
P02 or p~o, P~l' P~2 and their corresponding points coincide. The expressions of
both patches are given by
s(u, v) (1- u + Eu)2(1- v + FV)2pOO' (13.1)
s'(u,v) = (l-u+Eu)2(I-v+Fv?p~o, O~u , v~1. (13.2)
On each boundary of a patch, we define the two tangent vectors: the cross-
boundary tangent and the along-boundary tangent. The cross-boundary tangents
are given by s,,(O,v) and s~(O,v) on the common boundary:
s,,(O,v) = 2(I-v+Fv)2ao , (13.3)
s~(O,v) = 2(I-v+Fv?a~, (13.4)
where aj = PI) - POj and a'j = P'lj - p'Oj for j = 0,1,2 , which are the control
vectors of the cross-boundary tangents. The along-boundary tangents on the
common boundary u = 0 are given by
sv(O, v)= s'v(O, v) = 2(1- v + Fv)eo, ( 13.5)
where ej = PO(j+l) - POj, j = 0,1. (13,6)
The conditions of smooth connection of the tangential planes of both patches at
any points on their boundary curve are:
(i) the cross-boundary tangent vectors of both patches are collinear, or
(ii) the two cross-boundary tangent vectors and the along-boundary tangent
vector are coplanar.
For these conditions, the following equations (13.7) and (13.8) must hold for any
values of v,
.\s,,(O, v) + .\'s',,(O, v) = {av + (1 - v),B}sv(O, v), (13.7)
.\ + .\' = 1, (13.8)
13.3 Example of Closed Surface of Minimum Number of Patches 215
Figure 13.2 Geometric relation of

vectors: ).a + ).'a'
where a, j3, A and N are constants to be determined. From condition (i), a

and j3 must be 0, and from (ii) a scalar triple product [8",8'", 8 v ] is required to
be O. The latter condition is easily proved by substituting the left-hand side
of eq. (13.7) into f'lv in the triple product, in which two elements become the
same one. Substituting equations (13.3) and (13.4) for 8,,(0, v) and 8',,(0, v) into
eq. (13.7), and equating the coefficients of the terms of (1- v)2, v(l - v) and v 2
to zero, we obtain three equations:
Aao + X a' 0 aeo, (13.9)

Aal + A'a'l = (ael + j3eo)/2, (13.10)
Aa2 + Xa'2 j3el' (13.11)
Referring to Fig. 13.2, the tails of ai and a: are made coincident for i = 0, 1,2.
Because of the relation (13.8) between A and N, the left-hand sides of equations
(13'.9)-(13.11) are vectors, whose heads are at the dividing point of the line join-
ing the heads of ai and a:, i = 0, 1,2, in the ratio X : A. Giving appropriate values
for the constants a, j3 and,\, we can determine the control vectors eo, el, ai and
a: and the tangent plane along the boundary of the patches.
13.3 Example of Closed Surface of Minimum Number of

Patches
As a simple example of the patch connection which cannot be expressed by an S
net, we show six identical four-sided patches making a closed surface with C(1)
connection, This is not to make a sphere, which is made simply by two rational
pat.ches of degree two as shown in Sect. 12.6.3. Setting a = -1/2, j3 = 1/2 and
A = N = 1/2 in equations (13.9)-(13.11), we have
ao + a~ = -eo (13.12)
al + a~ = (-el + eo)/2 (13.13)
a2 + a~ el (13.14)
and we make the absolute values 6f vectors eo, ell ao and a2 all equal.
P 12 -=-----:-=1/# P'12
Figure 13.3 Locations of control

P10 F - - - - - - - " ' e P'10 points
Pll
Figure 13.4 A regular octahedron

and locations of control points
Referring to Fig. 13.3, since vectors ao, a~ and eo must be coplanar and their
absolute values are equal, they make an angle of 120 degrees with each other.
The control points PlO' p~o and POI make an equilateral triangle 1::::.0 with Poo at
its centroid. Three vectors a2, a~ and -el also are coplanar, hence P12' P~2 and
POI make an equilateral triangle 1::::.2 with P02 at its centroid. Since six patches of
the same shape enclose a space, there are eight corner points where three patches
meet. The eight corner points of the patches become the centroids of the eight
triangles I::::. i . Suppose a given regular octahedron with eight equilateral triangu-
lar faces. See Fig. 13.4. We set each of the corner points Poo, P02' P20' P22, etc.,
at each centroid of the faces of the octahedron, and we put the vertices of I::::. i at
the three mid-points of the edges of the face.
On each face of the octahedron, a small triangle like 1::::.0 exists and its vertices
are common with those of the similar small triangles on the adjacent faces. There
are eight such triangles which control the connection of B patches. If we can
determine six points Pll to satisfy eq. (13.13), we can obtain all the B points of
the connected six patches in C(1). · Let the length of an edge of the octahedron
(a) (b)
Figure 13.5 Shape of connected patches: (a) parameter lines, (b) contour curves
be 2c. Refer to Fig. 13.4. Extending outwards a line joining the center 0 of the
octahedron and one of its vertices by length c/(3V2), we set a point Pll at its
end. Since al = Pll - POl and a~ = P~l - POl' the left-hand side of eq. (13.13)
is a vector from the center 0 to POl with the absolute value c/3. The right-hand
side of eq. (13.13) becomes also the same vector, because the absolute values of
el and eo are (c/..;3).
Thus all the B points of the patches of the second degree are determined. Fig-
ure 13 .5(a) shows the constituent patches with constant parameter curves. The
contour curves of the synthesized surface in Fig. 13.5(b) show smooth connection
of the patches though their parametric curves have corners on the boundaries.
13.4 Three or Five-Sided Patch in Regular Patch Nets

Next we will explain methods to determine B points when there are elements of
three or five-sided meshes in a regular four-sided S net. We treat cases appearing
frequently in the rounding of corner paTts.
13.4.1 Rounding of a Convex Region

We explain a rounding procedure of a region which is roughly approximated by
the shape around a convex vertex with three convex edges. Refer to Fig. 13.6.
By cutting each of the convex edges by a plane parallel to its edge, and also
cutting off the central pyramid generated by the previous cutting procedures,
we can make a. primary body as shown in Fig. 13.6. In the center, there is a
triangular region shown in grey, which is to be treated specially for the rounding
of the corner. The surface outside of this region is easily rounded by using the
S net. The rounded surface of the central triangular region is synthesized by
the three connected B patches Bl, B 2 and B 3 . We assume all B patches are of
second degree and the dividing ratios of each S net in the u and v directions
are both 1 : 1 for simplicity. Figure 13.7 is its schematic diagram indicating the
Dividing
line
S net --~ . --.--

B2 I
I .-
----r---
i
Figure 13.6 A triangular region at a convex corner
PIO P oo P '1O
Figure 13.7 Locations of B points of three patches
locations of central control points Pl1 and the constituent B nets Bll B2 and B3.
We are to determine the locations of the control points of each B patch from the
positions of the vertices of the cut. block.
The B control points of B 1 , B2 and B3 are determined by the following pro-
cedures:
1. The cont.rol points Pl1 and P~l of Bl and B2 are two vertices of the central
triangle made by the cutting process, and the third vertex corresponds to
Pll of B3 ·
2. Refer to Fig. 13.7. Let a mid-point of the edge of the triangle be P~l. A
lower corner control point of Poo, which is the same as P~o, is fixed on the
symmet.ricalline as shown in the figure. A dividing point between P~l and
Figure 13.8 Contour curves of con-

nected patches
POO in the ratto (1 : 4) becomes POI. Similar points to this POI exist between
B2 and B3 and between B3 and B l . They are the control points P~2 and
P12·
3. The centroid of these three points POI' P~2 and P12 becomes P02' which is
the meeting point of the three boundaries of B patches B l , B2 and B 3.
4. The control points PlO' P20 and p~o, p~o of Bl and B2 are determined from
the surrounding S net. Those of B3 are similarly determined.
A proof of the C(l) connection of these patches is: on the t.hree boundaries of
B l , B2 and B3, equations (13.9)-{13.11) with values of the constants
A = A' = 1/2, a = 0, f3 = 1/2,
become
aO + a~ = 0, (13.15)
{1/2)al + (1/2)a~ = (1/4)eo, (13.16)
a2 + a~ = el. (13.17)
For inst.ance, we consider the connection of patches Bl and B 2. Since P02 is the
cent.roid of three points POI' P12 and P~2 of the patches Bl and B 2, eq. (13.17)
holds. The left-hand side of eq. (13.16) is equal to POlPOl and the right-hand
side is a quarter of POOPOl' hence they are equal by the definition of the location
of POI. Equation (13.15) holds naturally. An example is shown in Fig. 13.8.
Smoothness check by the contour curves gives good result.
The above is the case when the dividing line ratios A : A' of the surrounding
S net are taken as 1 : 1, but even when the dividing ratios of the net are (AI : AD,
(A3 : A~) , (A5 : A~) as indicated in Fig. 13.9, we can determine B points of three
B patches connect.ed in C(1), if the following condition holds among the dividing
ratios:
(13.18)
Dividing
line
S net
Figure 13.9 A triangular region at a convex corner for non-equal), (cf. Fig. 13.6)
. . .•. A: 8 = A'5 A'I: A5

• ... 83
,(1
eo
ao
a'o
Figure 13.10 Location of B point of three patches for non-equal ). (d. Fig. 13.7)
Figures 13.9 and 13.10 correspond to Figures 13.6 and 13.7. The proof is given
in Sect. 13.7.1.
13.4.2 Rounding of a Convex- Concave Mixed Region 1

We explain a method of rounding around a conca.ve vertex with mixed concave
and convex edges. See Fig. 13.11. For simplicity, we assume the two edges are
convex a.nd the one concave, and the shape is symmetrical with respect to the
bisecting plane of the concave angle. We want to cover this region by five B
patches of degree two connected in C(1). This process is rather complicated. If
we are allowed to use patches of rational type, we can divided the region into
two instead of five(refer to Chap. 14).
(a)
B5
D3 D4 FZ
B3
AO
BO BZ
(b)
Figure 13.11 A concave corner and five patches: (a) basic block, (b) modified block
Before making five B patches, we modify the basic block shown in Fig. 13.11(a)
to adapt to suitable B polygons as follows.
1. Make two new faces by cut ting off the convex edges to decrease the convexity.
The new faces are parallel to their original convex edges. The new planes
intersect ea.ch other making a new concave edge XY which connects with
the original concave edge Y Z. Fill the concave space along the connected
concave edges with a triangular prismatic column and a truncated pyra.mid
to decrease their concavity. See Fig. 13.11(b).
2. Four new faces are produced with the above procedures: in the central part,
a trapezoidal face (EoE}, D4 D3) whose height is h; in the left and right-
hand sides two bands of faces (AoD3, BoEo) and (A 2 D4 , B 2 E 1) parallel to
the respective convex edges, and let their widths be h'. We call them the
convex edge-bands; in the lower vertical part there is a face (EoA}, E1B 1)
parallel to the original concave edge, which is the concave edge-band. When
the slopes of the convex edge-bands and that of the trapezoidal face are
equal, their widths are equal,' h = h'. These edge-bands connect with the
H,
B4
D',
c,
B1 B2
F,
Figure 13.12 Locations of all control
points
trapezoid along EoEl, EoD3 and E I D4. Let the length of the upper edge
D3D4 of the trapezoid be m and that of the lower edge be n, and let the
mid-point.s of EoEI' EoD3 and EID4 be Do, DI and D2 , and the mid-point
of DID2 be O. The ratio m : n can be a.rbitrary, but if we take 1 : 3, the
values of various geometric relations become simple.
3. We draw mid-lines FoDo, CODI and C2D2 in the edge-faces and extend them
into the trapezoidal face . These lines int.ersect at the center point O. We
draw also lines D3FI and D4F2' parallel to the convex edges, that is, C2D2
and CODI' Connect OD3, and OD4.
4. There a.re five regions separated by the five boundaries: (ODoFo), (ODICO),
(OD 3FI ), (OD 4F2) and (OD 2 C2). We name these regions B I , B 2, B 3, B4 and
Bs. From these five regions we are to make B nets which define smoothly
. connected B surface patches.
Figure 13.12 shows the completed disposition of their control points. The equa-
tions (13.9)-(13.11) are applied for each boundary of a pair of the adjacent B
patches. Since the two bottom vect.ors ao and a~ are collinea.r (refer to Fig. 1),
a =0 in eq. (13.9). The equations which stipulate the connections for each pair
of Bi a.re:
).ao + )./a~ = 0 (13.19)
).al + )./a~ = (3eo/2 (13.20)
).a2 + )./a~ = (3el' (13.21)
Though we can take arbitrary values for h, hI and m and n, hereafter we assume
h = hI and 3m = n, for simplicity of explanation. Then the slopes of the convex
edge bands and the central trapezoidal pa.rt a.re equal and ratio of the upper
edge and the lower edge of the trapezoid is 1 : 3. This makes determining the
locations of B points simple. General cases are described in Sect. 13.7.2. We
treat three pairs of connection: (BJ and B 2), (B2 and B 3) and (B3 and Bs). The
other connections (B2 and B 4) and (B4 and Bs) are the same as (B2 and B 3)
and (B3 and Bs). We explain the method of locating the control points of each
pair.
.; \ ...... . ----
..........
...................................:>
.... ~ ............. .
PlI
eo
i
PIO' aD a' i P'IO

POD
Figure 13.13 Connection of BJ and
B2
Figure 13.14 Connection of BJ and

B3
- We treat first the pair (BI and Bz). Refer to Figures 13.13 and 13.12. The
boundary is (ODoFo). The points Eo and EI become Pu and P~I' The point
Do is POI and Fo is Poo which is determined from outside conditions. Then
D I , Dz become the control points P12' P~2' mid-point of which is 0 or P02' For
this case, ). = ).' = 0.5, j3 = 0 in equations (13.19)-(13.21).
- We treat the pair (BI and B3)' Refer to Figures 13.14 and 13.12. The prime
mar k of a vector indicates the vector belong to B I . Their common boundary
is (ODICO)' The control vectors are el = DIO, eo = CODI' a2 = OD3 ,
a~ = 0 Do. From the configuration of the central trapezoid, the constants )., ).'
and j3 in eq. (13.21) are determined:
). = ).' = 1/2, j3 = -1/4. (13.22)
Accordingly, from eq. (13.20) whose left-hand side is equal to DIDL the length
of eo is determined as 8DID~. Since P~I is at Eo, the location of Pu is its
symmetrical point Go with respect to COD I . Thus all the B points are fixed.
- We consider the pair B3 and Bs. Refer to Figures 13.15 and 13.12. The prime
mark belongs to the patch B 3 • The boundary is (OD 2 F I ). The control point
P~I of B3 is the same location Go as that of Pu of B3 of the pair (Bll B3) '
Figure 13.15 Connection of B3 and

Bs
Figure 13.16 Contour curves of the

synthesized region
The control edges are eo = D3Fl' el = D30 and ~ = OD}, a2 = OD4 . We

nave P02 = 0 and POl = D3 , P~l = Go and Pll = Ho, where Ho is located on
the line whose distance from a2 is the same as that of P~l and e2 and in the
symmetrical bisecting plane of the initial concave angle. From these relations
and (13.21) we can determine the following values:
), = ),' = 1/2, f3 = -1/2 (13.23)
In this way all the B points of the five B nets are determined as shown in
Fig. 13.12 and the C(1) connections of the five patches Bl . B5 are attained.
Figure 13.16 shows the contour curves of the synthesized surface.
13.5 Rounding of a Convex-Concave Mixed Region 2

Next. we treat a case of rounding around a vertex at which six edges meet. Its
block model is shown in Fig. 13.17. Six orthogonal faces intersect to make three
convex edges and three conca.ve edges interlaced. We define its three convex edges
as the coordinate axes with its origin o. Extensions of the convex edges beyond
the origin coincide with the concave edges. We treat the case of orthogonal edges,
but the met.hod can be generalized for non-orthogonal cases.
13.5 Rounding of a Convex-Concave Mixed Region 2 225
Figure 13.17 Block model of a mixed

corner
A mixed region of this type cannot be synthesized by patches of degree

two only, because of the insufficient number of control point.s of the const.ituert
patches, but for pat.ches of degree three there is a surplus of cont.rol points.
Therefore, double control points in a patch are to be considered.
A basic block modification is performed as follows.
1. At first. we apply the procedure of concave space filling and convex edge cut-
ting t.o t.he original shape. Referring to Fig. 13.17, we fill the three concave
edge spaces with the corresponding prismatic columns shown in grey. The
cross section of the column is a rect.angle isosceles triangle. Two side faces
of the column are coplanar with the coordinates planes, and the three top
edges of each of the columns make an equilateral triangle with vertices on
each of t.he coordinat.e axes.
2: We cut t.hese fillers by a plane parallel to the equilateral triangle stated
above. The cut.t.ing plane reaches the side edges of three prismatic columns
as shown in Fig. 13.18. Let the intersecting points be (q6 and ql), (q2 and
q3), (q4 and qs)·
3. Then we cut the convex edge of t.he original body by a plane paraUel to
the vert.ical edge and passing through the line q6~' and remove the cut
part of original convex edge. We caU the produced face (q6~q43~1) the
convex edge-fa.ce and also we call a prismatic filler face (qSq4q32q4l) the
concave edge-face. Repeating t.he procedure for the other two convex edges,
we obt.ain t.he shape shown in Fig. 13.18. In the central region, there is a
hexagon with vertices q6, ql, q2, Q3, Q4 and Q5·
Let t.he cent.er of the hexagon be Qa and the mid-points of it.s edges be (Qm6, Qml,
Qm2' Qm3 ' Qm4 and Qm5); and the center lines in the edge-faces, each passing
through Qmi and t.he center Qa, be L i . We are now to make six B nets connected
to cover the cent.ral region, which generate B patches connected smoothly. Let
the corresponding B nets be B l , B 2 , B 3 , B 4 , B5 and B 6 • We define their outer
boundaries by the intersecting lines with the six cutting planes, each of which is
parallel to one of the coordinate planes. We name the main node points on the
Figure 13.18 Modified block and

names of specific points
PI3 P'!3
eZ
az a 'z
PIZ P'IZ
POZ
el
al a'l
PII P'II
POI
eo
Figure 13.19 Relations between

PIO P'IO control points of connecting patches of
degree th ree
outer boundaries. They are
as shown in Fig. 13.18. In this case, it is impossible to make a smooth region

by connecting B patches of second degree, because no possible disposition of B
points satisfies equations (13.9)-(13.11) for a shape of this type. Accordingly, we
assume B patches of degree three.
We take a pair of adjacent patches to discuss their connection. Refer to
Fig. 13.19. The patch expressions are given by
s(u,v) = (l-u+Eu?(l-v+Fv?poo (13.24)

s'(u, v) (1 - u + Eu) 3(1 - v + FV)3p~O. ( 13.25)
13.5 Rounding of a Convex-Concave Mixed Region 2 227
On their boundary u = 0, the tangent vectors are
s .. (O, v) = 3(1 - v + Fv?ao (13.26)

s~(O, v) = 3(1 - v + Fv?a~, (13.27)
sv(O,v) = 3(1-v+Fv)2eo . (13.28)
For the e(l) connection, the following relation must hold, just as in the case of
the second degree patches,
'\su(O, v) + ,\/S~(O, v) = {a(l - v) +,Bv }sv(O, v). (13.29)
The coefficients of (1 - v)3, v( 1- v)2, v2(1- v) and v3 should be zero, accordingly

we have
'\ao + ,\/a~ = aeo (13.30)

'\al + ,\/a~ (1/3)(2ael + ,Beo) (13.31)
'\a2 + ,\/a~ (1/3)(2,Bel + ae2) (13.32)
'\a3 + ,\/a~ ,Be2' (13.33)
We set ,\ = ,\' = 1/2, a = 0, since Bo and Bl are symmetrical with respect to
their boundary, and ao and a~ are collinear. The absolute values of a3, a~ and
e2 must be equal, and the angles between a3 and e2 and between e2 and a~ are
120 degrees. Accordingly, ,B = -1/2 satisfies eq. (13.33). Then eq. (13.31) and
eq. (1-3.32) become
(1/2)(al + a~) = -(1/6)eo (13.34)

(1/2)(a2 + a~) = -(1/3)el' (13.35)
Referring to Fig. 13.20, at first we determine the control points of Bl net as

follows:
PO~ = qm13' POl = qml' P02 = qml' P03 = qo

PlO = q13' Pll = (1/6)(q13 + 5ql)' P12 = ql' P13 = qm6'
P20 = (1/2)(q13 + q12)*' P21 = (1/2)(q12 + ql)*' (13.36)
P22 = (1/6)(qll + 5ql), P23 = qm6'
P30 = q~2' P31 = (1/2)(qn + qd*, P32 = qll, P33 = qm63
Control points with a superscript * mark can be taken arbitrarily. Their indicated
values in eq. (13.36) are examples. The control points of B2 net are
P~o = qm13' P~l = qml' P~2 = qml' P~3 = qo

P~o = Q21' P~l = (1/6)(Q21 + 5Q2), P~2 = Q2' P~3 = Qm2
P~o = (1/2)(~3 + Q22)*' P~l = (1/2)(Q22 + Q2)* ' ( 13.37)
P~2 = (1/6)(Q23 + 5Q2), P~3 = Qm2,
P~o = Q;2' P~l = (1/2)(Q23 + Q22)*, P~2 = Q23, P~3 = Qm23,
b
08
..··P.;:::··"""
/ 0 'Bt BJ \\\Q
POD
P30 · P30
Figure 13.20 B patches in hexagonal region and their control points
The symbol * indicates arbitrary locations. From the above equations we get the
relations
(13.38)
Substituting the ahove equations in the left-hand side of eq. (13.35) , we get zero
vector. The right-hand side is el = P02 - POI = qml - qml = O. Therefore,
eq. (13 .35) holds. Also we have
al = Pll - POI = (1/6)(qI3 + 5ql) - q",l' (13.39)

a~ = P~l - P~l = (l/6)(q21 + 5q2) - qml' (13.40)
Substituting the above equations in the left-hand side of eq. (13.34)' we get
The right-hand side is -1/3(qml - qmlJ, so eq. (13.34) holds. Accordingly, with
the control points(13.36) and (13.37) the patch BI and the patch B2 connect in
C(l). With a similar disposition of the control points, any two adjacent B patches
of this arrangement connect in C(1) . The control points P30' P20 ' P31' P32 and
P30' P~o, P~l' P~2 of the patches are not concerned with this connection, hence
they can be taken to adjust connections with the outside regions. Figure 13.21
shows an example of contour curves of this rounding. The curves cut by the
parallel planes show smooth connection of the patches.
13.6 Rounding with a Rolling Ball

In practice, edge corners and vertices of connected patches are rounded by en-
velopes of a moving sphere which contacts with the bounding surfaces to elim-
inate their intersecting edges and vertices. We can determine the locus of the
Rounding with a Rolling Ball 229
Figure 13.21 Contour curves in the

mixed region
center of the ball and also the loci of contact points on the surfaces by solving
the differential equation for the intersection of the offset surfaces of the original
ones. The solving process is explained in Sect . 16.6.2.
The rounding procedure by a rolling ball around a vertex made by three
surfaces follows the rules:
1. Around a convex vertex, usually three intersection curves made by each two
Of three offset surfaces meet at one point, from where the center of the ball
starts along each one of the intersection curves. If the loci of the contact
points of the ball and the original surfaces are continuous, an envelop of the
moving ball becomes the smooth blending surface of the intersecting original
surfaces. Figure 13.22 shows the locus of the center of the ball and loci of
the contact points.
2. Around a convex-concave vertex, which is generated by one convex (concave)
edge and two concave (convex) edges , the ball moves first along the single
concave (convex) edge and then the ball moves along the consecutive convex
(concave) edges touching only two surfaces, a pa.rt of which is the envelop
surface already produced.
A point on the locus of the center o( s) of the ball and the two contacting
point.s Po(s) and P2(S) with the surfaces to be rounded are functions of the arc
length s of the locus of the center. Let the intersecting point of the tangents at
Po(s) and P2(S) in the plane OPOP2 be Pl' Then vertices of the triangle POPIP2
are considered three control points of a rational Bezier polygon of degree two,
whose weights are 1, (1/2), 1. Since these three control points are functions of
s, an envelop surface is determined by the moving ball.
230 13 Non-regular Connections of Four-Sided Patches and Houndings of Corners
Inner offset surface
Contact curve
Loci of th e center
of a rolling ball
(a)
Patch of 3 x 3 degree
Envelope surface
Loci of t he cen ter

of a rolling ball
(b)
Figure 13.22 (a) Intersection curves of three inner offset surface, (b) Locus of the center of
a rolling ball and its envelope surface
13.7 Appendix
13.7.1 Connection in a Triangular Region: General Case
Referring to Figures 13.9 and 13.10, the B control points of Bll B2 and B3 are
determined as follows:
13.7 Appendix 231
1. The cont.rol points Pl1 and P~l of B1 and B2 are two vertices of the central
triangle made by the cutting process, and the third vertex corresponds to
the control point Pl1 of B 3 . These points are the same as those of the case
of the equal,\; 's .
2. As in Fig. 13.7 t.he control point. p* 01 is the dividing point of Pll and P~l '
but its dividing ratios is (>.~ : ).d instead of (1 : 1), and POI is the dividing
point of P"Ol and PO~ in the ratio (2)'~ : ).1).5) instead of (1: 4).
A control point Poo, which is the same point as p~o, is the corner point of B1
and B2 as shown in the figure. The control points P12 and P~2 are similar
to POI' They are determined in a similar procedure, but the corresponding
dividing ratios are ().~ : ).5), (2)'~: ).5).3) for P12, and (>.~ : ).3), (2)'~ : ).3).1)
for P~2' On each side of the cent.ral triangle (POlP~2P12), there is a dividing
point of t.he side in the ratio, each given by ().~ : ).1) or (>.~ : ).3) or (>.~ : ).5)
as shown in the figure, then P02 is the coincident point of three lines from
each vertex to the dividing point on its opposite side, because of condition
(13 .18). The coincidence is guaranteed by Ceva's theorem, refer to Sect. 3.4.
3. The cont.rol points PIO' P20 and p~o, p~o of BI a.nd B2 are determined from
the surrounding 5 net. Those of B3 are similarly determined.
Three Pll points of each of B patches are the three vertices of the triangle
in t.he 5 nets. The control points PIO, P20 and P21 of the central three B patches
are determined as in usual 5 net. The connection of these patches with the
surrounding B patches with e(!) is guaranteed by the 5 net properties.
The positions of B points thus determined satisfy the conditions of e(1)
connection of the three central B patches. We give its proof on the bounda.ry of
the patches B1 and B 2 .
In the triangle made by three points POl' the three vectors from the center
P02 to its vertices can be looked upon as the control vectors a2, el and ~.
The following relation holds between them by definition ( by applying Menelaus'
theorem):
I I ).1).5
).l a 2 + ).l a 2 = 2>.' el·
5
As for ao and al , the following relations exist by the definition,
).1 ao+ ).~ a~ 0, (13.41 )

).1'\5 /
).la1 + ).Ial =
I I
~eo 2. (13.42)
2/\5
These relations are the same as equations (13.11), (13.9), (13.10), which stipulate
the e(l) connection at the boundary edge of the two patches Bl and B 2 , if we
set
a = 0, (J = ).1).5.
2)'~
The e(1) connections between B2 and B3 and between B3 and Bl are similarly
proved.
13.7.2 Connection of a Pentagonal Region: General Case

When the ratio of lengths of the upper edge and lower edge of the trapezoid
is not 1 : 3 and the height h of the trapezoid and the width h' of the convex
edge-band are not equal, the values given by eq. (13.22) become
m
). = ).' = 0.5, f3 = ---,
m+n
then the following expressions must be used for the values:
2D~Dl
f3
and values given -by eq. (13.23) are changed to
m+n 4m
). , ).' = ,
5m+n 5m+n
m+n
f3 5m+n
Let the distance between P~l of B3 and eo, which is the boundary vector
between B3 and Bs , be d. Then the distance between Pll of Bs and eo is
d· ().II).) = 4mdlm + n. The location of Pll is determined at a point which has
this distance from eo and also has similar distance from the boundary vector eo
between Bs and B 4 •
14 Connections of Patches by Blending
14.1 Introduction
Generally, when we have to define a surface patch, its degree of freedom is not
always equal to that of given external conditions. If the latter is greater than
the former, there must be some relations among the given conditions, or some
methods must be provided to increase the degree of freedom of the patch to
satisfy the given conditions.
For a polynomiaJ surface s( u, v) of degree 3 x 3 of tensor product type, there
are sixteen constant vectors to be determined corresponding to the terms uiv j
{i,j = 0,1,2, 3}, or the sixteen control points, or the conditions of one position
vector, two tangent vectors and one twist vectors at each of the four corner points
of a patch. If two curves and two cross-boundary tangent vectors expressed in
degree three are given on the boundaries at u = 0 and u = 1, all the constants
of a patch are determined, because the sum of degrees of freedom of the given
boundary conditions is equal to sixteen. Hence we cannot give the conditions of
the cross-boundary tangent vectors along the boundaries v = 0 and v = 1. They
are automatically fixed . Accordingly, so long as a patch is expressed by a 3 x 3
polynomial, we cannot specify all the boundary conditions even up to the first
derivative vectors.
To construct a patch with such boundary conditions, we have to devise some
methods to increase the degree of freedom. We treat four methods. First, Coons'
approach is explained concisely, for it is a classic one , though not practical. It
uses a Boolean sum of three patches to meet given boundary conditions. Another
method is to blend two patches, each of which satisfies boundary conditions on
one pair of boundaries only, with weighting functions multiplied to eliminate
unwanted effects. Then methods of using patches with variable corner twist
vectors or variable inner Bezier points are introduced. Finally, a method of using
correcting patches , which compensate the derivative vectors along the boundary
curves of a patch to satisfy given conditions, is explained.
14.2 Coons' Patch
Coons' patch expression consists of three blending patches. It is not used today,
though its name is famous [1]. So we explain it concisely. Refer to Fig. 14.l.
On the boundaries 1 and 3 which correspond to the parameter values u = 0
and u = 1, the boundary curves rl.(v) and r3(v) , the cross-boundary tangents
tl (v) and t 3(v) are given and also on the boundaries 2 and 4 corresponding to
234 14 Connections of Patches by Blending
Figure 14.1 Boundary curves and cross-boundary tangents of Coons' patch
v = 0 and v = 1, the boundary curves 1'2(U), 1'4(U) and the cross-boundary

tangents t 2 ( u) and t 4 ( u) are specified. By adding two patches which partially
satisfy the given conditions, and subtracting a patch, which compensates the
surplus boundary wnditions, Coons constructed a patch expression satisfying all
the boundary conditions.
For a fixed v on the boundaries 1 and 3, we consider points 1'1 (v) and 1'3( v),
and cross-boundary tangents t1(V) and t3(V), which give the boundary conditions
of a curve segment r( 11.) in the u direction. This is because we can express r( u)
by a weighted sum of the positions and the tangent vectors at its end points.
Their weighting functions were already given by eq. (8.19), which are expressed
in 1 x 4 matrix form by
H(t) = {fo(t) f1(t) g(t) h(t)}, for t = u, v. (14.1 )
For the variable v moving from 0 to 1, the curve segment f( u) becomes an

expression of a patch sn( u, v):
( 14.2)
Similarly, from the data on the boundaries 2 and 4, we have a surface patch
Sb(U , v):
(14.3)
On the boundaries 2 and 4, the position and cross-bounda.ry tangent vectors of
sa are automatica.lly fixed for j = 0 and 1 as
sa(u,j)
( 14.4)
s~(u,j)
On the boundaries 1 and 3, those of Sb are fixed for i = 0, 1 as
Sb( i, v)
(14.5)
s~(i, v)
14.2 Coons' Patch 235
Using equations (14.4) as the boundary conditions, we can define a surface

patch sab(u., v) which is given by
(14.6)
Substituting equations (14.4) into the above equation and letting matrix Mab be
1'3(0)
== 1'3(1)
lv[ab (14.7)
1'3v(0)
1'3v(1)
we ha.ve
( 14.8)
Similarly from equations (14.5), we can define a surface patch which is given
by
SbaCu.,v) = H(u.){Sb(O,v) sb(l,v) s~(O,v) s~(l,v)Y

( 14.9)
= {Sb(O, v) sb(l, v) s~(O, v) s~(l, v)}H(u.)T .
Letting Mba represent
(14.10)
we have
(14.11)
The two patches sab and sba are expressed by the properties at their four corners
only, which are the elements of the matrices Mab and Mba . Now examine their
properties. The elements of the upper-left 2 x 2 sub-matrix of Mab or Mba
represent the four corners of the patch, those of the lower-left and upper-right
2 x 2 sub-matrices represent the tangent vectors in the u. and the v directions.
The corresponding elements of the sub-matrices are equal . The elements of the
lower-right 2 x 2 sub-matrix represent the lateral change of the tangent vectors
at the corners which are the so-called twist vectors.
Since in a regular surface patch t.he order of differentiation does not influ-
ence on t.he result, the corresponding element.s of this sub-matrix are also equal.
Accordingly, Al ab = Mba and consequently sab = sba.
So, the boundary conditions of the patch sab or sba are considered equal to
the sum of those of the non-defined boundaries of sa and Sb. Accordingly, in the
following expression of s( u., v), the surplus boundary condit.ions are canceled:
(14.12)
This satisfies the given boundary conditions on all its four boundaries. This is
called Coons' patch expression by Boolean sum. Usually it is difficult to give two
pairs of aIbitraIY cross-boundary tangent vectors in the u a.n d the v directions
along the respective boundaries.
When t.he bounda.ry curves and the cross-boundary tangent vectors aIe ex-
pressed by the zero-th and the first derivatives at their end points together with
the Hermite weighting functions of degree three, the three expressions sa( u, v),
Sb( u, v) and sab( u, v) become the same one. So we have
(14.13)
This patch of 3 x 3 degree is expressed by its corner vectors:
s(u,v) = H(v)MH(u)T, (14.14)
where
s(O,O) s(l,O) su(O,0) su(l,O)
s(O,l) s(l,l) su(O,l) su(1,l)
1
M= [
sv(O,O) sv(l,O) suv(O,O) suv(1,O) .
sv(O,l) sv(l,l) suv(O, 1) suv(1, 1)
Usually, it is still difficult to give appropriate values of the vectors Suv at the
corners of a patch.
14.3 Independent Boundary Conditions
14.3.1 Blending by Weighted Sum

There is another method of eliminating disturbing boundary values of the blend-
ing patches by multiplying with appropriate weighting functions [4]. Let each of
two patches sa(u, v) and Sb(u, v) satisfy the given boundary conditions only on
the respective pair of boundaries. Therefore, the weighting functions w a ( u, v)
and Wb(11., v) take value 1 on one pair of the boundaries and zero on another.
The synthesized patch is given by
s(u, v) = wa(u, v)sa(u, v) + wb(u, V)Sb(u, v), (14.15)

wa(u, v) + wb(u, v) = 1, °
~ wa(u, v), wb(u, v) ~ 1. (14.16)
The weighting functions wa(u, v) and wb(u, v) must satisfy the following condi-
tions on the boundaries:
wa(O, v) = w°(1, v) wb(u, O) wb(u, 1) 1,

= wa(u,l) wb(l, v)
= =
}
wa(u, 0) wb(O,v) 0,
(14.17)
w~(O,v) = w~(l, v) = w~(u,O) = w~(u, 1) 0,
w~( 11,0) w~(u, 1) w~(O, v) w~(1, v) = 0.
14.3 Independent Boundary Conditions 237
With these weighting functions, we have for i, j = 0,1,

s(i, v) = sa(i, v), s(u,j) = Sb(U,j), }
(14.18)
s,,(i, v) = s~(i, v), sv(u,j) = s~(u, j).
These are the values we want to have. One form of the weighting functions which
satisfy the conditions (14.16) and (14.17) is given by
a v2 (1 - V)2
w (u,v) = u 2 (1- U)2 + v 2 (1- V)2' (14.19)
b u 2 (1 - U)2
w (u,v)= u 2 (1-u}2+v 2 (1-vP' (14.20)
Disturbance of smoothness by these weighting functions is a little larger than

that of other methods described in the subsequent sections.
14.3.2 Two-Valued Twist Vectors and Floating Inner Control Points

The twist vectors at the corners of a patch influence the cross-boundary tangent
distribution on each of its boundaries. If the twist vectors are not indepen-
dent, externally defined cross-boundary tangents are not independent. Then it
is difficult to construct a net of smooth connecting Coons' patches. If the twist
vectors are taken null, patch connection becomes easy, but the synthesized shape
is usually unsatisfactory, because of many flat regions on the surface.
A method is devised to make the twist vectors independent on each boundary
of a patch. This is to replace s"v( i, j)'s by the following expressions for i, j = 0,1
(v - j)suv(i,j) + (u - i)suv(i,j)
(14.21)
(u - i) + (v - j)
This expression was given by A. Gregory [2] . The expression (14.21) takes a
different value according to whether u = i or v = j. For a patch expression in
Bezier form of 3 x 3 degree, the same effect can be attained when its inner four
control points P1+i,1+J for i, j = 0,1 are replaced by
(v - j)pj'+i,1+j + (u - i)P~+i ,1+j

(14.22)
( U - i) + (v - j)
When parameter u takes value u = i or v takes value v = j , eq. (14.22) becomes

P~+i,1+J or P~+i,1+j' that is, the location of a control point is different. See
Fig. 14.2. For instance, the cross-boundary tangent on the bounda.ry v = 0 is
given by
sv(u, O) = 3(1- u + EU)3 ao , (14.23)

where
aO = Pal - Poo
(14.24)
a2 P~l - P20
Figure 14.2 Double inner control

points
The cross-boundary tangent on the boundary u = 0 is given by

s .. (O,v) = 3(1- v + FV)3bo, (14.25)
where
(14.26)
Comparing the tangent control vectors (14.24) and (14.26), we know there is
no coupling between them. So we can make the independent cross-boundary
tangents on the boundaries v = 0 and u = O. This variant of Gregory's method
was introduced by H.Chiyokura [6].
14.4 Correction of Cross-Boundary Tangent Vectors
To treat the connection of a patch with its surrounding ones, we have to increase
the degree of freedom of the patch to be connected. For this purpose we use
a compensating patch which gives the cross-boundary derivative vectors to be
added to the original ones to satisfy externally defined conditions [20][21].
The compensating patch gives null positional vectors on all the boundaries
and arbitrary cross-boundary tangential and higher derivative vectors. With a
composite patch of the original and compensating ones, we can satisfy all its
boundary conditions. We use techniques similar to those for connection of a
three-sided patch with its adjacent patches already fixed.
As for the connection in C(2), we treat it in the next section, since the
technique applied is similar, but the expression is more complicated.
14.4.1 Connection of Four-Sided Patches

Since it is easy to define a B patch Sb( u, v) which satisfies only given boundary
curves , we must add a correction patch SC( u, v) to match the given cross-boundary
tangent vectors. Here the superscript b shows a base patch ana another one C
indicat es a compensating patch. Let the patch Sb( u, v) be expressed by
(14.27)
One form of the correction patches is given by
SC(u, v) = Q(u, v)v(l- v)u(l- u). (14.28)
Let the vector function Q( u, v) have a form:
Q(u, v) = (14.29)
T I (l - u)v(l - v) + T2u(1 - u)(l - v) + T3uv(1- v) + T 4 vu(1 - u)
{u(l - u) + v(l - v)P
where Ti is a cross-boundary tangent on one boundary i. Equation (14.28)
vanishes on all the boundaries, but its first derivative gives independent values
on each boundary of the patch:
s~(O,v) = T}, s~(u, 0) = T 2 , (14.30)

s~(l,v) = T 3 , s~(u, 1) = T 4 • (14.31)
Since TI can be a function of v, we assume the following form:
(14.32)
where ti, i = 0,1,2,3, are the control vectors of the cross-boundary tangent on
the boundary u = O. From equations (14.27) and (14.32), the cross-boundary
tangent on u = 0 of the patch to be connected is given by the composite vectors:
su(O,v) s~(O, v) + s~(O, v) = 3(1 - v + Fv)3(aoo + to), (14.33)

where aOj = PIj - POj' j = 0, 1,2,3.
The cross-boundary tangent can take arbitrary form so long as it does not vi-
olate the geometric constraints at the corners of the patch. For instance, at each
node of boundary curve meshes of patches, a tangent vector of a compensating
patch must be in the tangential planes defined there. The cross-boundary tan-
gents on the other boundaries take forms simila.r to the above . If in eq. (14.32),
the control ta.ngents to and t3 are zero, that is, at both ends of the boundary, there
is no need of tangent compensation there, and the denominator of eq. (14.29) can
be reduced to {u(l- u) + v(l- vninstead of {u(l- u) + v(l- v)P.
Example 1. Modification of connection of two patches from C(O) to C(l)
Figure 14.3(a) shows two connecting Bezier patches of degree 2 x 2, which are
symmetrical with respect to the vertical plane containing their common boundary
curve. Therefore, the respective cross-boundary tangent vectors make acute
angles on the common boundary.. The section curves of both patches cut by
(a) Two conn~ted Bezier patches (b) Section curves of shape (a)
(c) Compensated patch (d) Section curves of shape (c)
Figure 14.3 Example of tangent vector correction
the parallel vertical planes which are orthogonal to the plane of symmetry are
shown in Fig. 14.3(b).
We add compensating patches to both patches. The form of the compen-
sating tangent is T I = (1 - v Fto + 2v( 1 - V)tl + v2 t 2 • The tangent control
vectors of the compensating patches are determined so as to make the composite
cross-boundary ta.ngent control vectors of the two patches the same magnitude
and opposite directions:
t, (Pli - p~i)/2 - aoi,

t: = -(Pli - p~i)/2 - ~i' i = 0, 1,2. }
(14.34)
,,
/~
" ~, '
~~
..
(a) Net of control points (b) Contour curves
Figure 14.4 Rounding of a convex corner
From eq, (14,33), the composite cross-boundary tangents are
Then on the common boundary, their connection become e(l), Figure 14,3(c)
show the shape of composite patches and Fig, 14,3( d) is the section curve sim-
ilar to Fig, 14,3(b), We can notice clearly the rounding effects caused by the
compensating patches,
Example 2, Rounding of a convex corner
This is a similar case to Sect, 13.4,1. But use of the compensating patches makes
the rounding process simple and flexible, A net of the control points of three
Bezier patches B}, B 2 , B3 of degree three is shown in Fig, 14.4( a), Let the symbols
used for corresponding control points both sides of the boundary be the same as
used in Chap, 13. In the central triangular region where three patches meet at
P03' their common four control points, point P03 and three points P02 indicated
by the small circles, are in a plane to give the e(1) connection.
The locations of the set of control points (POO,PIO ,POI ,Pll) along the con-
necting boundaries of the patches can easily be arranged to make control vectors
of the cross-boundary tangent such as (PIO - Poo) and (Pll - POI) equal to the
corresponding vectors in the connecting patch.
But determining the locations of the control points P12 to guarantee the
e ( l) connection is difficult. So, we use the technique of compensating patches.
The form of the compensating tangent is TI = V3t 3 + 3v(1 - V)2t z which is
applied to both sides of the boundaries to make the control vectors of the cross-
boundary tangents equal and opposite. In Fig. 14.4(b), the contour curves of the
(a) Net and original shape
(b) Contour curves
Figure 14.5 Rounding of a convex-concave corner
compensated patch are shown for checking its smoothness.

Example 3. Rounding of a convex-conca,ve corner
The shape of this type is usually difficult to make with simple connection of
patches. One example is shown in Sect. 13.4.2. To show the ability of our method,
we use two Bezier patches of degree two for the base patches and the two compen-
sating patches whose control tangent vectors are of the form T 1 = v 2 t. Figure
14.5( a.) is the contour curve pattern for C(O) connection of the patches, while
Fig. 14.5(b) is that for connection with the compensating patches to attain C(1).
The contour curves of the former have shaTp corners, but in the latter the sharp
corners become rounded.
Exa.mple 4. Rounding of hexagonal region

An interlaced convex and concave hexagonal region, treated in Sect. 13.5, can
be synthesized from six patches of degree two with their compensating patches.
The locations of the control points of the basic patches are shown in Fig. 14.6(a)
in solid lines. The tangent vectors of the compensating patch is given by the
form Tl = v 2 t. The normal vectors along the their connecting boundaTY before
and after the compensation are shown in Fig. 14.6(b) and (c). The compensation
(a) Locations of nets of the base patches (b) Normal vectors along the boundary
connected in 0(0) before compensation
(c) Normal vectors after compensation (d) Contour curve pattern of hexagonal region
Figure 14.6 Rounding of hexagonal region
makes the normal vectors equal. The contour curves of the region are shown in
Fig. 14.6(d). Dotted curves are parameter constant curves.
14.4.2 Evaluation and Comparison of Methods
Owing to addition of a compensation patch SC( U, v) for the C(1) connection,

the shape of the original patch is modified together with the slope along its
boundary curves. To investigate its effect on the shape, we draw a contour map
(a) Four sided patch compensation (b) Triangular patch compensation
Figure 14.7 Shape modification by compensating patch
for a. compensating pat.ch SC( u, v), whose base patch is a Bezier of degree two:
( 14.36)
Let the compensating cross-boundary vector on u

and all other T;'s be zero, On the boundary u = °=the°becross-boundary
T 1 = (1 - + FV)2t
v
tangent
o
is given by
s,,(O, v) s~(O, v) + s~(O, v) = 2(1- v + Fv?(aoo + to/2),

where aoo = aOl = ao2 = (0,5,0,0), ( 14.37)
to = tl = t2 = (0 , 0, I),
The height of the pat.ch is
z k . s( u, v) = k . Sb( u, v) + k· SC(u, v)
(1 - v? + 2v(1 - v) + v 2 x u(l- u)v(l- V )2
(14.38)
{u(l - u) + v(l- v)P
°
The contour curves of the above equation are shown in Fig. 14.7(a), Though
the slope on the boundary u = is modified from zero degrees to 45 degrees,
its effect on shape of the patch is at most 5% of the boundary length and its
variation is smooth. In Fig. 14.7(b) , a similar diagram for a triangular patch for
slope correction on it.s boundary w = is shown. °
For the example of Fig. 14. 7( a), Chiyokura.'s method cannot be applied, be-
cause it. cannot modify the tangent direction at the ends of a boundary curve.
So, we give the same condition of the cross-boundary tangent. on one bounda.ry of
a connecting pat.ch t.o compare the effects of both methods, Figures 14,8( a) and
(b) are similar cont.our diagrams of the shapes generated by the same correction
(a) Method of Sect. 14.4.1
(b) Chiyokura's method
Figure 14.8 Comparison of the methods for the same cross-boundary tangent
of the cross-boundary tangent on one boundary of a Bezier patch of 3 x 3 degree.

The shape expression by the method of this section is
9u(1- u)2v(l- v)3

z = {u(l-
~~--~~~--~~
u) + v(l- v)p·
(14.39)
That. by Chiyokura's method is
(14.40)
As for the influence on the shape by the correction, the disturbance of this
method can be cont.rolled by the magnitude of the compensating vectors.
One of the important features of the method is the correction ahility of
the cross-boundary tangent even at its boundary ends. For such a case, we
examine an example to determine which one of two connecting patches should
be compensat.ed. In Fig. 14.9(a), at the top vertex of the connecting patches, we
want to have two composite control vectors of the tangent for each patches, which
are collinear and equal with opposite sign. Let the original two control tangent
vectors be a, a l and 6a = a - a /. By adding two compensating vectors t = f-l6a
and t l = (1 - f-l)6a to a and a l respectively, we have the equal and opposite
control tangent vectors. The value of f-l varies from a to 1. In Fig. 14.5(b) the
shape for f-l = 0.5 is shown and in Fig. 14.9(b) the case of J.l = 1.0. In the former,
both patches sha.re the same compensa.tion, while in the latter only one patch
shares the compensa.tion. In this case J.l = 0.5 gives naturally good result, because
the base patches aj:e symmetrical. For general cases, more study is needed for
sharing compensation and the magnitudes of the compensating vectors to attain
desirable shapes.
14.4.3 Three-Sided Patches

The compensating patch technique applied for a four-sided patch is also appli-
cable for a triangular patch. Since the connection of triangular patches is less
flexible, this method is more useful in this case than that of four-sided patches.
We treat here the connection of a triangular patch and a four-sided patch. For
the. expression of a triangular patch, the operator form is introduced in Cha.p. 15
such as
(14.41)
where u, v and ware parameters having the relation u + v + w = 1. A tri-
angular pat.ch which connects with a four-sided patch consists of a base patch
Sb( u, v, w) which satisfy at least the given boundary curves and a compensa.tion
pat.ch SC( u, v, w) which gives null position vectors and non-zero independent tan-
gent vectors on its three boundaries. By adjusting independent tangent vectors
of t.he compensat.ion pa.tch, we can connect it with four-sided patches smoothly.
As for t.he detail of the triangular pa.tch and its derivatives on the boundaries,
refer to Chap. 15, we use only the results for the connection. For the base patch
we use eq. (14.41). For the compensation patch, we assume the following form:
SC(u, v, w) = Q(u, v, w)uvw, (14.42)
The vector function Q takes the following forms, for a case of zero tangent values
a.t all its vertices,
(14.43)
~a =a - a'
t = - J1.~a Compensating
t' = (1 - J1.)~a vector
( a) Compensating and compensated vectors
(b) Case of It = 1.0
Figure 14.9 Effects of sharing of compensation
a,n d for a case of non-zero tangent vectors at the vertices,
Q = T1uv + T 2 vw + T3 WU • (14.44)
(uv + vw + wu)2
On t.he boundary w = 0, derivatives of triangular patches with respect to u are:
s~('U, 0)
V, s~(u, v, 0) = -T1uv for eq. (14.43), (14.45)
s~(u, v, 0) = s~(u, v, 0) = -Tl for eq. (14.44). (14.46)
By using t.he independent tangent vectors T i , i = 1,2,3, we can adjust the

boundary condit.ions for connection.
Example 5. A t.riangular patch surrounded by a four-sided patch array
Using a compensation patch as given by eq. (14.43), we embed a triangular patch
in a net of spline surfaces. Figure 14.10(a.) is a, control net in the center of which
is a, t.riangular region. Figure 14.10(b) shows connected patches and Fig. 14.10(c)
(a) A control net in the center of which is a triangular region
(b) Connected patches (c) Contour pattern
Figure 14.10 A triangular patch surrounded by a four-sided patch array
its contour curves. The boundaries of e(O) connection are corrected to give e(1)
connection.
Let a four-sided pa.tch which is expressed by
s'(u, v) = (1- u + Eu?(1- v + Fv?p~o (14.47)
connect with a three-sided patch which is given by
s(u, v, w) = (u + Ev + FW)2pOO (14.4S)
on their common boundary curve s'(l, v) = s(l- v, v, 0). Locations of their con-
trol points on both sides of the boundary are schematically shown in Fig. 14.11,
The cross-boundary tangent of the four-sided patch s'(u, v) on the boundary
u = 1 is
s~(1, v) == 2(1 - v + Fv?b~, (14.49)
PH
P 02
P OI Shaded region is
a triangular patch
Figure 14.11 Connection of a triangular patch with a four-sided patch
where
The cross-boundary tangent t of the triangular patch s( u, v, w) on the bound-

ary w = 0 consists of the tangent vector from the base triangular patch and that
generated from the compensation patch. The former can be given by an expres-
sion (1 - 1I)s~(1 - v, v, 0) + vs~(l - v , v, 0), where
s~(1-v,v,O) = 2(1-V+EV)(1-F)poo=2(1-V+EV)bo,} (14.50)

s~(1- v, v, 0) = 2(1 - 11 + Ev)(E - F)poo = 2(1 - v + Evco,
and t.he latter is given by s~( u, v, 0) + s~( u, v, 0) from eq. (14.43), which becomes
-2v(1-v)T I . Consequently, the cross-boundary tangent ofthe triangular patch
is given by
(1 - u)s~(u, v, 0) + us~(u, v , 0) + s~( u, v, 0) + s~(u, v , 0) =

2(1 - v)2b o + 4v(1 - v)(b 1 + co/2 - Td2) + 2V2Cl ' (14.51)
Comparing eq. (14.51) and eq. (14.49), we can determine Tl to make all the
coefficients of (1 - v)2, v( 1 - v) and v2 proportional to those control vectors of
another patch t.o be connected in C(Jl. The result is shown in Fig. 14.10(b) and
(c) .
When t.he control tangent vectors of the triangular patch and the connecting
four-sided patch are not colinear, we have to use the compensating patch given
by eq.(14.44) and TI = 2(1-v+vF)2t o. Then, the right-hand side ofeq. (14.51)
is replaced by
(14.52)
This must be equal to the cross-boundary tangent of the four-sided patch, which
is given by eq. (14.49), and we can determine the control tangents to, tl and t 2.
14.5 Case of C(2) Connection

For the C(2) connection, a compensating patch must provide not only arbitra,r y
second derivative vectors independently, but also arbitrary first derivative vectors
on some or all of the boundaries of a compensating patch. The technique for
synthesizing compensating patches is systematically given in the following.
14.5.1 Four-Sided Patches

Let a base patch be
(14.53)
and its compensating patch be
SC(u, v) = Q(u, v)v(l- v)u(l- u), (14.54)
in which a vector function Q( u, v) is given by
A B
Q(u,v) = + , ( 14.55)
{u(l- u) + v(l- v))3 {u(l- u) + v(l- v)}4
where
A (Tl + Rl'u){(l - u)v(l - V)}2 + (T2 + R2v){u(1- u)(l- V)}2

+(T3 + R3(1 - u)){uv(l - V)}2 + (T4 + R 4(1- v)){vu(l - U)}2
B = (-3/4)[Td(1- u)v(l- V)}3 + T2{u(1- u)(l- V)}3
+T3{uv(1 - v)}3 + Tdu(l- u)vn.
From equations. (14.54) and (14.55), we obtain the first and the second cross-
boundary derivative vectors for 'U, = 0:
(14.56)
where T 1 and Rl are the cross-bounda,ry first and second derivative vectors.
They are independently given. We obtain similar results for the other boundaries.
So we can make their connection in C(2) even when the original patch connection
is C(O).
Example 6. Compensation for C(2) from C(O) connected pa.tch
Figure 14.12(a) shows C(O) connected patches of degree 2 x 2. Figure 14.12(b) is
its silhouette pattern without the compensation; its loci are discontinuous along
14.5 Case of e(2) Connection 251
(a) e(O) con'neeted patch (b) Silhouette pattern

without compensation
--- ---- ....

-"i-- __ . ___ • .". _______ __ ,_. __ ___ ___ •••
-"j-----
.. --------;
j
-~- ~~.-
~ i j j 1 ;
: i j f
I / i !
-'-. -~. "-.

-.--.~
\]
(c) Silhouette pattern with (d) Silhouette pattern wi th

e(l) compensation e(2) compensation
Figure 14.12 Compensation for e(2) from e(O) conneeted patch
the connected boundary. In Fig. 14.12(c) the silhouette pattern with the added
compensating patches for e(l) are shown; the loci become continuous, but have
corners on the boundary. In the same original patch with added compensation
patches both for e(1)and e(2), Fig. 14.12(d) shows that the discontinuity and
the sharp corners of the loci on the boundary, which appeared in (b) and (c),
disappear and that these compensation effects are local near the boundary. The
compensating effect can be changed by the magnitudes of the compensating
vectors.
14.5.2 Three-Sided Patches

To obtain independent compensating derivative vectors of the first and the second
order on each boundary of a three-sided patch, we introduce t.he compensating
patch SC( u , v, w) of the same form as eq. (14.44), but with a different vector
function Q(u,v,w) which is
A B
Q(u,v,w) (uv + vw + wup - (3/4)-;-(U-V-+-VW-+-W---'U):-:-4' ( 14.57)
where A (Tl + R 1w)(UV)2 + (T2 + R 2u)(vw)2 + (T3 + R3V)(WU?,

B Tl(UV)3 + T 2(vW)3 + T3(WU?
On the boundary w = 0, we have
(14.58)
By determining t.he forms of the compensating derivative vectors Tl and RlJ

we can make the composite derivative vectors to match the given boundary
conditions for t.hem. For the other boundaries, similar results are obtained.
When the power of the denominators of Q is reduced by one, the compen-
sating derivative vectors eq. (14.58) become T 1 uv/4and R1uv . When there is
no need for compensation at the vertices of the patch, the vector function Q of
this type ca.n be used.
The above stated method of constructing compensation patches can be ex-
tended to cases of higher than second-degree continuity on the patch boundaries,
because its form has a systematic structure.
15 Triangular Surface Patches and
Their Connection
15.1 Introduction
In Chap. 13, we treated non-four-sided patches, each of which was synthesized
from four-sided Bezier patches, and described their uses in special regions of
shapes. In this chapter, first we explain a triangular Bezier patch defined by
linear interpolations in barycentric coordinates, whose expressions in operator
form are similar to that of Bezier patches and their natural extension to rational
form. Then we tre'at methods of connection of triangula.r patches. Since there is
no simple equivalent of a spline net as in the four-sided patch connection [10][19]'
we cannot control a local shape of connected triangular patches easily. But there
are number of connection patterns, which are explained [12].
However, synthesizing a shape from only triangular patches of Bezier type is
not easy. Accordingly, we introduce the tangent compensating patch of triangular
form to increase the freedom of connection with other triangular patches. A case
of a triangular patch connected with a surrounding four-sided patch net has
already been treated in the previous chapter. In Sect. 15.6, we describe division
and degree elevation of a triangular patch, but they are not so effective as in the
case of four-sided patches.
15.2 Operator Form of a Triangular Patch

15.2.1 Triangular Bezier Patches
Just like in a Bezier surface patch, by using the shift operators E and F, a control
point Pij of a. triangular patch is conveniently expressed as Ei Fjpoo for the purpose
of simple manipulation. Refer to Fig.I5.I. A triangular plane s(u, v, w; 1), whose
three vertices are Poo, PlO and POI' is expressed in the barycentric coordinates
(u,v,w) by
s(u, v, w; 1) = uPoo + VPlO + wPOl = (u + vE + wF)poo, (15.1)
whereu+v+w=I, O::;(u,v,w)::;l.
Two of the three parameters u, v and ware independent. Values of u, v and
w which determine a point in the triangle can be considered proportional to
the distances from the three edges to that point. In the parameter space, the
triangle is equilateral, and u+ v +w is proportional to its area, so by normalizing
it we have eq. (15.1). If the value of one of the parameters (u, v, w) is zero,.
the expression represents one of Its edges. That with two zero values of the
254 15 Triangular Surface Patches and Their Connection
8(u. v. w;3} P03
2j1l· · · · · · · ··· ···· ···· · · · .lQP" ";2\0
POD P20 P30
Figure 15.1 Triangular plane and barycentric coordinates
parameters corresponds to one of it.s vertices. For inst.ance, U = °corresponds

to the edge PlOP01" which is
8(0, V, W; 1) = vP10 + wPOl = (1- W)P10 + WP01' (15.2)
and for u = v = 0, then W = 1,
8(0,0, w; 1) = WPOl = POl' (15.3)
Generalizing eq.(15.1), we define a triangular patch of degree n in the following

form:
8(u,v,w;n) = (u+ vE + wFtpOQ, (15.4)

u+v+w=l.
For W = 0, eq.(15.4) becomes
8(U, v, 0; n) = (u + vE)npoo = (1- v + vEtpoo' (15.5)
This ia a Bezier curve of degree n, whose control points are
POQ PlO ... PnO'
Similarly, for the boundary curve of v = 0, its control points are
POQ POl ... POn'
For u = 0, eq.(15.4) becomes
8(0, v, w; n) = (1- W + wFE-l)nEnpQQ = (1- W + wFE-l)n pnO ' (15.6)
Since the operator FE- l changes an index (ij) into ((i - 1)(j + 1)), its control
points are
PnQ P(n-l)l ... POn'
15.2 Operator Form of a Triangular Patch 255
Figure 15.2 Relation of a triangu-

Paa lar patch of degree n and triangular
S OO( u. v. w ; 2 ) patches of degree n - 1
These control points determine the shapes of the patch boundaries. The other
control points Pij with a < i + j < n determine the interior shape. The number
of all the control points is (1/2)(n + 2)(n + 1).
From eq.(15.4), we can interpret that a point of a triangular patch of degree
n is on a triangular patch of degree one with its three vertices placed on the
three mutually shifted triangular patches of degree n - 1. See Fig. 15.2. Let
Sij( u, v, W; n) be a patch whose lower-left-most control point be Pij, then the
shift operators E and F can operate on Sij( u, v, W; n). Hence, eq.(15.4) can be
transformed into
v, W; n)
SOO( U, (15.7)
= (u + vE + wFtpoo = (u + vE + wF)soo(u, v, W; n - 1)
= usoo(u, v, W; n - 1) + VSlO(U, v, W; n - 1) + WSOl(U, v, W; n - 1).
15.2.2 Rational Triangular Patches

We assign a weight Wij to a control point Pij and instead of parameter u, v and
w, we use l' • 11I00/woo(r, s , t; 1) , s · WlO/wOO(r, s, t; 1) and t . woI/woo(r, s, t; 1) ,
where
woo(r, s, t; 1) = r· Woo + s . WlO + t . W01 = (r + sE + tF)woo. (15.8)
So we have a normalizing condition just like u+v+w = 1 in the previous section,

r . woo/woo(r, s, t; 1) + s . WIO/wOO(r, s, t; 1) + t . woI/woo(r, s, t; 1) = 1. (15.9)
Then the rational equivalent of eq.(15.1) is
(r + sE + tF)(wooPoo)
Soo (r, s, t; 1) = -'-----:----...:.-'--;;,.:.=....::.::..:... (15.10)
woo(r, s, t; 1)
The interpretation of the above equation is that a point soo(r, s, t;,I) on a rational
triangular patch of degree one is the rational interpolating point of three points
of degree zero.
soo(r, s, t; 0) = Poo , slO(r, s, t; 0) = PlO , s01(r, s, t; 0) = POI' (15.11)
Generalizing woo(r, s, t; 1) , which is given by eq.(15.9), we have
woo(r, s, t ; n) = (r + sE + tFtwoo = (r + sE + tF)woo(r, s, t; n - 1) . (15.12)
Simila.rly, the general form of eq.(I5.IO) is
.) _ (r + sE + tFt(wooPoo)
sao (r, s, t, n - ( ) . (15.13)
woor,s,t ;n
This is also obtained from the following recurrence formula:
.) _ (r+sE+tF)(woo(r,s,t;n-I)soo(r,s ,t;n-I»
sao (r, s,t , n - () . (15.14)
Woo r, s, t; n
The interpretation of the above equation is that a point soo(r, s, t; n) on a rational
triangular patch of degree n is a rational interpolating point of three adjacent
rational triangular patches soo(r, s , t; n - 1), slO(r, s, t; n - 1), s01(r, s, t; n - 1)
of degree n - 1 with respective weights woo(r, s , t; n - 1), wlO(r, s, t; n - 1) and
WOI (r , s, t; n - 1) of degree n - 1. Since a conic is expressed by the rational Bezier
of degree two, the boundary curves of a triangular patch can be made circular
arcS of the same radius and the same center. But it is possible, but not simple,
to construct a part of the spherical surface by rational triangular patches of the
same size [16].
15.2.3 Tangents on Patch Boundaries

On the surface of a triangular patch, there are three kinds of constant parameter
curves, that is, U = canst , v = canst and w = canst. Accordingly, we can set
up three basic vectors which are not independent. For simplicity, by replacing it
dependent variable , for instance, w with 1 - u - v we make a patch expression
with the usual two independent variables:
s(u, v, 1 - u - v) = {u(l- F) + v (E - F) + F}poo

Differentiating this expression with respect to u or v, we obtain tangent directions
s"or Sv along the parameter curves v = canst or u = canst on the patch. Refer
to Fig. 15.1. On one boundary w =0, they are
s" = n(I- v + vEt- I (I - F)poo = n(I - v + vEt-Ib oo , (15 .15)

Sv n(I - v + vE)n-l(E - F)poo = n(l- v + vE)n-Icoo , (15 .16)
15.3 C(l) Connection of Triangular Patches 257
where
boo = (1 - F)poo = Poo - Pal' (1.5.17)
Coo = (E-F)PoO=PlO-P01'
The directions of these tangent vectors are not along the boundary w = O. A
cross-boundary tangent, which we denote f, is given by
f (1 - >.)s" + >.sv
n(1 - v + vEt- l {(1 - >')b oo + >.COO} (15.18)
= n(1 - v + vE)n-l(b oo + >'aoo),
where aoo = (E - l)poo = Coo - boo, and>' is an arbitrary constant between 0

and 1.
A tangent vector along the boundary w = 0, which we denote t, is obtained
from differentiatioR of the boundary curve s(1 - v, v, 0: n) by v,
(15.19)
Derivatives of higher degree are similarly obtained, for instance, on the boundary
w= 0,
s"" n(n - 1)(1 - v + vE)n-2(1 - F?poo (15.20)

= n(n - 1)(1 - v + vE)n-2(1_ F)b oo = boo - ball
Svv n(n - 1)(1 - v + vE)n-2(E - F?poo (15.21)
n(n - 1)(1 - v + vEt- 2(E - F)coo = ClO - Cal,
s"v n(n - 1)(1 - v + VE)"-2(E - F)(1 - F)poo (15.22)
= n(n - 1)(1 - v + VE)"-2(E - F)b oo = b lO - bOl = Coo - COl'
They are used for connection in C(2).
15.3 C(1) Connection of Triangular Patches
Connection of two triangular patches along a. boundary in C(1) is not difficult,

but it is difficult to cover a certain area by connected triangular patches, because
there is no simple aut.omatic control mechanism to attain globally a specified
connection condition such as an S net. We have to connect a patch one by one
to sat.isfy the C(J) condition. Refer to Fig. 1.5.3. Let s~o( u, v, w; n) be another
pat.ch connected on the boundary w = 0 with sao. Its cross-boundary tangent f'
is given similarly by
t' = (1 - p)s~ + ps~ = n(1 - v + VE)"-l(b~o + paoo). (15.23)
where II is an arbit.rary constant b-etween 0 and 1.

Q------I-----~ P (j+I) O
(a) Connection of patches (b) Control quadrilateral
Figure 15.3 Connection of patches and control quadrilateral
v v,
If at an arbitrary point on the common boundary, three tangent vectors t, t
and t are coplanar, then two patches Sao and s~o are ,continuous with a common
tangent plane. This condition of coplanarity is
[t , t' , t] = o. (15.24)
For this to be satisfied, the following conditions must hold:

v v,
(i) t = I't, (15.25)
or (ii) t = at + /1t', (15.26)
where a, /1 and I' a,re appropriate constants.
1. Substituting equations (15.19) and (15.23) into eq.(15.25), we obtain
(1 - v + vEt- 1 (b oo + Aaoo) = 1'(1 - v + vEt-l(b~o + paoo). (15.27)
Developing both sides by terms (1- v)iv n - 1- i , i = 0 ... n -1, and equating
their coefficients, we get
i=O ... (n-l). (15 .28)
Refer to Figures 15.3 and 15.1. The simplest case to satisfy the above
condition is to take I' = -1, A = p. Then each quadrilateral (Pia, Pil'
P(i+1)o, P:l' Pia) is in a plane and its diagonal PdP:l divides the other
diagonal PiOP(i+l)O in the same ratio for all i. We call this quadrilateral
(Pia, Pil' P(i+l)O, P:l) the control quadrilateral. A pair of grey triangles in
Fig. 15.3 makes the control quadrilateral.
15.3 a(l) Connection of Triangular Patches 259
(a) Connection of patches (b) Contour curves
'igure 15.4 Connected triangular patches and contour curves
2. From equations (15.19), (15.23) and (15.26), we obtain
i=O . .. (n-l), (15.29)
where O!, j3 and '"Yare constants independent of i. The connection of two

patches in t.his pattern is shown in Fig. 15.4 together with their contour
curves.
If PiO and O!, j3, '"Yare given, P:o is determined from eq.(15 .29). Two control
points Pil and P:l and an edge aiO determine a control quadrilateral.
If O!, j3, '"Yare all equal to 1, all the control quadrilaterals become paral-
lelepipeds.
For n = 2 and '"Y = -1, eq.(15.29) determines relations among the magni-
tudes ofthe vectors boo, b~o, aoo and b 10 , b~o, alO' Denoting their magnitudes
by boo, b~o, 0.00 and blO , b~o, 0.10 , we have the following equations.
O! sineo • boo + j3 sin eo . b~o = 0, 1

O! C?See O • bb oo + j3
j3 c.oseeo ·b~~O = 00. 00 , . (15.30)
O!sm 1· 10+ sm 1· 10= ,
O! cos e1 • blO + j3 cos e1 • b~o = 0.10,
where ei is an angle between (-b iO ) and aiO, i = 0, l.

We can choose two vectors arbitrarily among boo, b~o , b lO , and b~o ' The
remaining two vectors are determined from the above equations.
rhough various shapes can be constructed from the triangular patches connected
me by one so as to satisfy the above conditions, the lack of a. simple mechanism
)f automatic adjust.ment of locations of the control points makes general use of
;riangular patches difficult .
Figure 15.5 Special connection of triangular patches and their contour curves
Figure 15.6 Surface synthesized from triangular patches
Examples. From eq.(15.30) , if boo, b~o , aoo and eo are given, two unknown con-
stant.s cy and f3 are obtained. With these cy and f3 and given e1 and alQ, two
unknown 610 and b~o are determined from eq.(15.30) . Using these data, we can
const.ruct the C(1) connected triangular patches such as shown in Figures 15.5
and 15.6. In their central regions, five or six patches are connected. In Fig. 15.5
a hatched area of pent.agonal shape and in Fig. 15.6 the hexagonal areas must be
planar. Furt.her connection can be made outwards t.o satisfy equations (15.30)
and (15.30). Figures 15.5(b) and 15.6(b) are shapes of synt.hesized surfaces with
t.heir contour patterns .
15.4 Arbitrary Connection of Three-Sided Patches 261
Figure 15.7 Control triangle and connected patches
If 'Y = 0, t.hen bia and b: a must be on a straight line and the ratios of
t.heir magnitudes a.re constant. In this case, a control quadrila.teral reduces to a
triangle. See Fig. 105.7.
15.4 Arbitrary Connection of Three-Sided Patches

The connection of a three-sided patch surrounded by four-sided patches was
treated in t.he previous Sect. 14.3 with use of the cross-boundary tangent com-
pensat.ing patch. Without the compensating patch, the shape synthesis by simple
connection of three-sided patches is not easy, because the conditions imposed on
the control quadrilaterals gathering around all the common vertices of the patches
of a synt.hesized surfaces are difficult to satisfy. Accordingly, for easy synthesis of
a shape, we need to introduce the compensating patches of the cross-boundary
tangents to satisfy the conditions easily.
A triangular patch s( u, v, w; 2) is considered consisting of a base patch
Sb(u , v, w; 2) and a compensating patch SC(u, v, w):
s(u,v,w;2) = sb(u,v,w;2)+sC(u,v,w), (105.31)
where sb(u,v,w;2) = (u+vE+wF)2 Paa ,
SC(v.,v,w) = Quvw.
The coefficient. Q is a function of u, v, wand given by
Q = T1uv + T 2 vw + T3 WU ( 15.32)
(uv + vw + wuF
The compensating patch SC( u, v, w) does not affect the boundary shape of the
patch, but the derivative takes one of the relative vectors Tl or T2 or T3 on
their boundaries w = 0 or u = 0 or v = 0 . From eq.(15.19), the cross-boundary

tangent of t.he base pat.ch Sb( u, v, w; 2) on w = 0 is given by
£b = (1 - A)S" + ASv = 2(1 - v + vE){(I- A)b oo + ACoo}

2(1- v + vE){boo + A800}. (15.33)
The cross-bounda.ry t.angent of the compensat.ing patch sC(u,v,w) on w = 0 is

given by
(15.34)
because s~ = -Tl = s~.

Let the compensat.ing vector Tl be composed of 2(1 - v)d o and 2vd l :
Tl = -2(1 - v + vE)d o. (15.35)
The cross-boundary tangent £ of the synthesized patch on the boundary w = 0

becomes
- -b -c
t=t +t =2(I-v+vE)(boo+do+Aaoo). ( 15.36)
Since t.he cross-boundary tangent of the other connecting patch on the same
boundary is given by
e = 2(1 - v + vE)(b~o + d: + A800) , ( 15.37)
we can make two vectors (15.36) and (15.37) proportional by adjust.ing the com-
pensat.ing cont.rol vectors do and d l . With this technique, we can construct
shapes composed of smoothly connecting triangular patches. The degree of the
patches is not restricted to two. An example of a synthesized surface is shown
in Fig. 15.8. Four t.riangular patches are connected in C(O) around a point. With
addit.ion of t.he compensating patches, their boundary regions are rounded to give
the C(l) connection. The shape of the surface can also be modified by magnitudes
of t.he cont.rol vectors do, d l and d~, d~.
15.5 Division of a Triangular Patch

Since a triangle can be divided into four small equilateral t.riangles, the cor-
responding t.riangular patch can be divided into four triangular patches. See
Fig. 15.9. Therefore the number of control points of the patch increases, but
t.here is no increase of freedom so long as the degree of the connection conditions
is lowered. Since t.here is no such simple equivalent net as a spline net, sha.pe
cont.rol by division is difficult. But by using the connection technique described
in t.he previous section we can replace a divided triangular patch by another
triangular patch wit.h smooth connection.
15.5 Division of a Triangular Patch 263
(a) Connected patches and parameter constant curves
(b) Contours before compensation (c) Contours after compensation
Figure 15.8 Synthesized surface with compensating patches
Figure 15.9 Division of a triangular

patch of degree two: location of new
con trol points
Letting r , s, t be new parameters for the parameter space oHhe central small
triangle in Fig. 15.9, the relations between r, s, t and u, v, ware given by
u = (1 - r)/2, v = (1 - s)/2, w = (l-t)/2. (15.38)
The following relation naturally holds:
r+s+t=1. (15.39)
Substituting these u, v and w into eq.(15.4), we get
s(r, s, t; n) = (1/2t{1 - r + (1 - s)E + (1 - t)F}npoo

= (r + sEa + tFa)np~o, ( 15.40)
where Ea and Fa are the shift operators for the first and the second subscripts
of the new control points pt. Substituting 1 - s - t for r in eq.(15.40) and
differentiating it t times with respect to sand j times with respect to t and
setting s = 0, t = 0, we obtain the relations between the old and the new control
points Pij and pi) :
(15.41)
Application of the above equation is treated below: Next, the control points of
the divided smaU triangle at the left corner are to be obtained. Referring to
Fig. 15.9, since the relation between the new parameters r, s , t and the old ones
u, v, ware
u = r/2, v = (s + 1)/2, w = t/2 , (15.42)
the equation corresponding to eq.( 15.40) is
s(r, s, t; n) = (1/2)n{r + (1 + s)E + tF}npoo

= (r + sE b + tFbtpgo . ( 15.43)
where Eb and Fb are the shift operators for the first and the second subscripts of
the new control points prj.
By the same met.hod as previously used, we obtain the equation which determines
the relation bet.ween the control points of the new and the old patches :
(15.44)
From equations (15.41) and (15.44) with the given values of n, i and j , we ca.n
obtain the relation between the original control points Pij a.nd the new control
points pij of the divided triangle patch, and the relation between Pij and prj.
Since a new control points pi can be transformed in the following way:
pi) = (Ea)i(Fa)ip~o = {(P - 1) + 1r{(F a - 1) + l})p~o ( 15.45)

= ECkC~(Ea - l)k'(Fa - l)mp~o, k = 0··· i, m = 0··· j,
15.6 Elevation of Degree 265
and t.he right-hand side of the above equation can be substituted by the left-hand
side of eq.( 1.5.41), we obtain desired rela.tion between the new and the old control
points. Similarly we obtain the relation between p? and Pi'
Example. Division for n = 2.
Let the cont.rol point.s before the division be
( 15.46)
The locations of the new control points of the center triangle are given by
Poo (1/4)(E + F?poo = (1/4)(p2o + 2Pll + P02) ,
Plo = {1/4)(F 2 + E + F + EF)poo = (1/4)(Po2 + PlO + POI + Pll)'
POI = (1/4)(E 2 + E + F + EF)POO = (1/4)(p2o + PlO + POI + Pll)' ( 15.47)
P~l = (1/4)(1 + E + F + EF)poo = (1/4)(poo + PlO + POI + Pll)'
P20 (1/4)(1 + F)2poo = (1/4)(poo + 2POI + P02) ,
P02 (1/4)(1 + E)2pOO = (1/4)(poo + 2PIO + P20)'
The above equations show that the new control points are located at each cent.roid
of four point.s of t.he old ones. The control point of the divided sma.ll triangle
patch at t.he left corner are
pgo = POO'
pto = (1/2)(1 + E)poo = (1/2)(poo + PlO),
pgl (1/2)(1 + F)poo = (1/2)(poo + POI),
(15.48)
P~l (1/4)(1 + E + F + EF)poo = (1/4)(poo + PlO + POI + Pll),
p~o = (1/4)(1 + E)2poo = {l/4)(poo + 2PIO + P20),
pg2 = (1/4)(1 + F)2pOO = {1/4)(poo + 2POI + PoJ·
The control point.s of the divided patches a.t the right corner and the upper corner
have the similar relations. Geometrical interpretation of the locations of the new
cont.rol point.s given by equations (15.47) or (15.48) with respect to the old ones
is easy.
15.6 Elevation of Degree

It is possible t.o raise t.he degree of a triangular patch without changing its shape.
The method is similar to that of the four-sided patch: multiplying a factor
(u+ v + w) with (u + vE + wF)npoo, which represents a triangular patch of degree
n, and comparing to (u + VE' + wF/)n+lp~o, we obtain the locations of the new
cont.rol point.s P:j in terms of the old control points Pi),
For n = 2 we obt.ain the following expressions:
p~o Poo, p~o = P20' P~3 P02,
p~o (1/3)(p20 + 2PIO), P~2 (1/3)(Po2 + 2POI),
p~o {l/3)(poo + 2PIO), P~2 (1/3)(Po2 + 2Pll), (15.49)
P~l = (1/3)(poo + 2POI)' P~l = (1/3)(p2o + 2Pll)'
P~l = (1/3)(PIO + POI + Pll)'
16 Surface Intersections
16.1 Introduction
Interference calculation of surfaces is an important basic procedure in CAD/CAM
technology. It appears in various processes such as set operations in solid mod-
elling, surface modelling, cutter path calculation for NC machining, and calcu-
lations in robot motion. Efficiency and robustness of its solutions are required
especially for those of free-form surfaces, because the degree of intersection curves
becomes so high that their analytical methods are generally hopeless and the nu-
merical methods become unstable where the normal vectors of interfering surfaces
on their intersecting curves become nearly parallel.
Recently, a great many mathematical studies on these problems have been
made , but they seem still too theoretical to be used by engineers for their prac-
tical applications. A basic textbook [10] on these topics has appeared recently,
but we explain our methods of intersection calculation in this section, because
they are easier as well as practically proved [13].
Interference of quadrics has been treated in Chap. 5 using pencils of two
quadrics and conics, but in this chapter it is included in a case of the intersection
of surfaces of implicit forms. First we treat a case of the intersection between
a plane and a free-form surface as a basis of the intersection of surfaces which
are represented in parametric forms or implicit forms . Since the contour curves
of a shape, which are its section curves by equidistant cutting planes, are very
useful for understanding the shape of a curved surface without its real model ,
methods of producing the contour curves of free-form surfaces must be efficient
and accurate. Accordingly first we treat the intersection of a curved surface and
a plane, then we extend the methods to the intersection of curved surfaces.
To trace them we apply numerical methods for solving ordinary differential
equations with automatic stepsize adjustment, because this provides good results
without requiring attention to determine the stepsizes to the next point [14][15].
Selection of stepsizes has always been a problem for tracing curves whether the
method of differential equation is adopted or not. Since near the singular points
of the interfering surfaces the behavior of the curve to be traced becomes un-
predictable and in other regions a large stepsize may be applied, control of the
stepsizes is needed for efficient and robust calculation. If we adopt the meth-
ods of differential equation solving with quality control, we are freed from this
problem. The other important problem is to find the initial starting points of
the curves and all the intersection loops. Since the intersection may become a
contact point or a very small loop, they may be missed if an appropriate method
268 16 Surface Intersections
to find them is not given. We discuss this problem and show a practical method
to solve it.
In regions where the normal vectors of interfering surfaces become nearly
parallel wit.h each other, the basic differential equation for their intersection curve
becomes invalid. We explain a method of treating these regions and discuss
the behavior of the intersection curves there. In this chapter, we also treat
intersection curves of offset surfaces, because the problem is practically important
and similar solving methods can be applied to this problem.
Some of the methods of differential equation solving for our purpose are
compared and evaluated in the Appendix.
16.2 Intersection of a Curved Surface and a Plane

Intersection curves of a plane and a free-form surface are important, because
they are fundamental to the interference calculation and frequently required in
other problems. For example, as we cannot understand and evaluate a three-
dimensional free-form shape only by its two-dimensional projections, we make
drawings of the intersection curves of the surface with equi-distant parallel planes
placed in various directions for visualizing its detailed shape. Since obtaining the
intersection between curved surfaces frequently becomes a rather complicated
unstable problem in numerical analysis, we introduce a common cutting plane
for both surfaces to determine a common intersecting point of the two intersection
curves of the surfaces and the plane. Or we frequently replace a curved surface by
its tangential plane or a polyhedron, in which case we need the cutting procedure
of the surface and faces of the polyhedron.
Let nf be the unit normal vector of a plane and d its distance from the
origin, then an equation of the plane is given by r . nf = d (refer to eq. (3.11)).
Let a surface patch be denoted by s( u, v). Since points of the plane and those
of the surface pat.ch are coincident on the intersection curve, that is, s( u, v) = r,
eliminating r from both equations we have
s(u,v) ·nf = d. (16.1 )
This is the equation of the intersection curves. Fixing a value of u or vasa

constant, we can obtain the corresponding v or u of a point on the intersection
by solving eq. (16.1). For patches of the first or the second degree, the equation
to solve being linear or quadratic, we obtain its analytic solution easily. But
for a patch of higher than the second degree, we have to turn to the numerical
methods of solving high-degree algebraic equations. If only a few points are to
be obtained, the finding roots of higher-degree polynomials (usually of degree
three to five) is not a problem, but calculating all the points by this method is
not desirable. Methods of tracing .the curve from a starting point are required.
In regions where the surface normal vector is nearly parallel to that of the cut-
ting plane, the solution becomes sensitive to numerical errors. Such cases occur
frequently in practical problems, so this problem is treated in Sect. 16.6.
When the contour curves of a surface which has many peaks are to be ob-
tained, a sophisticated method such as the extremum search curves (refer to
Sect. 7.3.3) is recommended to determine the starting points for tracing the con-
tour curves. The extremum search curves are loci of points on the surface where
its basic vectors are parallel to the plane. During its tracing, the distance from
the cutting plane is calculated to determine an intersecting point, where one of
the intersection curves starts. Since the extremum search curves determine all
the peaks, the bottom points and the saddle points as well as the paths between
these points, we can find all the starting points of the loops on the same level.
For a simle surface, a standard method for tracing a contour curve is:
1. Make sure that there is a possibility of intersection between a plane F and

a patch s( u, v) by using the convex hull property of the patch. If there is
the possibility then perform the next steps.
2. Determine intersection points of the plane and the boundaries of the patch.
An even number of points must be obtained including the points correspond-
ing to double roots.
3. If the intersection points exist, then from one of the points, trace the in-
tersection curve until it reaches another intersection point on the boundary.
Determine all other intersection curve segments corresponding to the re-
maining intersection points on the boundaries.
4.' When there is no new starting point on the boundaries, divide the patch
into smaller ones by making new boundaries and test the possibility of in-
tersection with them, and repeat the above procedure.
Since the intersection points between a plane F and a parametric curve s( u, v)

for u or v = canst are obtained from solution of an algebraic equation (16.1)
for u or v = canst, those of the boundaries of the patch and the plane are
easily determined using the Newton-Raphson method or as roots ofthe algebraic
equation.
16.2.2 A Practical Method of Obtaining a Point on

an Intersection Curve
When it is difficult to solve eq. (16.1), for instance in case of an offset surface, to
obtain a point on the section curve we use the following iteration method to reach
the true point on it. Refer to Fig. 16.1. Let the initial point be Po = s( uo, va)
and the tangential plane there be Ft and at the point Po make a normal plane
Fn which is also orthogonal to the plane F. An intersection point Pl of the three
planes F, Ft and Fn, which is 'given by eq. (16.11), is the next approximate point.
270 16 Surface In tersections
Figure 16.1 A model of obtaining

a point on an intersection of a curved
surface and a plane
Since this point is in the tangential plane at Po, a vector PI - Po can be

expressed by the basic vectors of the surface at Po. We have
8"du + 8v dv = PI - Po = 6p, ( 16.2)
where 8" and 8 v are the basic vectors at Po . By making the inner product with a
vector 8 v X nj and that with a vector 8" x nj on both sides of the above equation,
we obtain respectively
du- [6p,8 v ,nj] dv = _ [6p, 8", nj] (16.3)

- [8", 8 v , nj] , [8", 8v , nj]
Then considering 8( Uo + du , Vo + dv) a new Po , we iterate the above procedure
until 6p becomes sufficiently small. Usually the convergence is very good and
a point on the intersection curve is determined. When the procedure does not
converge, a t.angent plane at the intermediate point p and the plane F becomes
pa.rallel. In such a case, we can examine the shape of the patch around a point
q which is determined on an intersection curve of 8 with the plane F slightly
moved in its normal direction.
A similar but more complicated process is used for a point on the intersection
of two curved surfaces.
16.2.3 Curve Tracing by Differential Equation Solving

After determining an intersection point , we start the tracing of an intersection
curve from t.here. The convent.ional method to determine the next point on the
intersection is to take an approximate point for 8(Uo + 6u , vo + 6v) on the
tangent line of the intersection curve at the initial point. The tangent direction
is given by the vector product of the normal vectors of the plane and the patch.
We determine the distance between the approximate and the initial points by
estimating t.he curvat.ure of t.he intersection curve there (see Sect : 6.3) or we take
a sufficiently small value for it. Since this approximate point is represented by
two expressions, P = Po + Ls on the plane and q = s(u + Lu, v + Lv) on
the patch, from these two points a point on the intersection is obtained by the
method described in the previous section. Repeating the same process, we obtain
successive points of the intersection curve.
Instead of the conventional procedure explained above, we use well estab-
lished differential equation solvers with adaptive stepsize to optimize the accuracy
of the results and the calculation speed as well as t.o standardize the procedure
for obtaining the intersection curves of all kinds of surface expressions, in which
a. plane is a special case. In application of numerical methods of differential
equation solvers for initial-value problems, initial values have to be determined
a.nd accumulation of truncation errors has to be kept within a given tolerance.
But both problems are satisfactorily solved in our case.
By ta.king the differential of eq. (16.1), we have
¢>udu + ¢>vdv = 0, (16.4 )

where ¢>(u,v)=nj's,
Considering eq. (16.4) the differential equation of a section curve, we can trace it
by using one of the established numerical methods for solving initial value prob-
lems. Usually eq. (16.4) is transformed into the appropriate forms by introducing
a function f(u,v):
(16.5)
Then we have
dujds = -f(u, v)<pv, dvjds = f(u, v)<Pu, (16.6)

where s is an independent varia.ble representing the arc length of the curve and
E, F, G are the fundamental magnitudes of the first order of the surface. Since
we obtain values of u and v as functions of s, we can calculate the section curve
s(u(s), v(s)).
Where the normal vectors of the surface and the plane become nearly parallel,
the tracing cannot be continued, because the functions <Pu and ¢>v become very
small. Then the behavior of the curve must be estimated from the second-
order expansion of the surface at that point. This is explained in Sect. 6.7.
However, if one uses a suitable numerical method of solving differential equations
with adaptive stepsize control, even in a region where the difference of an angle
between the normal vectors of the two surfaces becomes as small as 10- 10 radian,
the intersection curve can be traced without using the second-order terms. A
practical approach for obtaining the solution is discussed in the Appendix.
Example. Figure 16.2 shows contour curves of a. surface together with its ex-
tremum search curves. There are eight extremum points: three peaks, one bot-
tom point and four saddle points: Many examples of the contour curves are
Figure 16 .2 Contour curves and

ex tremum search curves
found in this book t.o indicate t.he shapes of surfaces, or to check the smoothness
of connection of patches. They are the results of this method.
16.3 Points on Intersection of Two Curved Surfaces

Determining the intersection curves of two surfaces is not so simple as in the
case of a curved surface and a plane, because we have to solve three nonlinear
simultaneous equations of four variables instead of solving one nonlinear equation
oftwo variables such as eq. (16.1). A slight change ofthe relative positions of two
intersecting surfaces may cause large a variation of the shape and the characters
of their int.ersection curves, and singular points on the curve frequently appear
in practical cases. We can determine aU the singular points of the intersection
curves in advance, but their calculation is not easy and we treat the problem in
Sect. 16.3.3.
Intersection of two quadric surfaces, each expressed by an implicit equation
such as f(x, y, z) = 0, has been treated theoretically in detail in Chap. 4, but
the intersection of parametric surfaces equal to or higher than the second de-
gree, which is practically important, cannot be treated with their implicit forms,
because the degree of an algebraic equation of intersection curves is too high to
treat with current numerical calculation techniques. For example, the degree of
the equation of the intersection for parametric surfaces of degree two is 64 and
that of degree three is 324. We are also required to obtain the intersection curves
of surfaces which are not expressed by closed-form equations, such as offset sur-
-faces. Accordingly, we need some new ideas to solve this probl~m theoretically
and pra.ctically.
Since differential equation solvers with adaptive stepsize can be applied to
produce the intersection curves of various types of surfaces with good results, an
important problem of applying them is to find all the initial starting points for all
intersection loops, which in extreme cases become contact points. In obtaining
the intersection curves of a plane and a. free-form surface, the extremum search
curves of the patch can be used successfully for determining the starting points
and the loops. Curves similar to the extremum search curves can be applied for
determining their initial starting points and all the loops of the intersection curve
of two surface patches . This is explained in Sect. 16.3.3. But first, we explain
a method which gives a point on an intersection curve from two points, one on
each respective surface.
To avoid confusion with the symbols used in calculating interference of two
surface patches, we use different symbols for the parameters in the expressions
of the patches r(w, t) and s(u, v). Suppose we have two points Po and qo , one of
which is on each respective patch and near the probable intersection curve. Let
their paTameter values there be Wo, to and Uo, va. Then we have
Po = r(wo,to) and qo = s(uo , vo). (16.7)
Starting from the above points, we finally have to find a point on the intersection
curve with parameters which satisfy the equation
s(u, v) = r(w, t). (16.8)
The Newton- Raphson method solving nonlinear equations can be applied

when the first approximate points are near the intersection curve. But this
usually does not give good results unless favorable initial values are supplied. So
we modify it to a more geometrically oriented method which gives a point on the
curve.
16.3.2 Method with an Auxiliary Plane

A more reliable method of obtaining a point on the intersection curve of two
surfaces is to introduce an auxiliary plane with which the two surfaces intersect.
Moreover, for the first approximation, we replace the surfaces by their tangential
planes at points near the probable intersection curve. Then we use an auxiliary
plane which intersects with both tangential planes orthogonally.
Refer to Fig. 16.3. With the following definitions and procedures, we de-
termine a point on the intersection curve of the two patches from Po and qo.
1. Let the tangential plane and the unit normal vector at Po of r( Wo, to) be
Fp and n p , and those at qo of s(uo, va) be Fq and n q . Symbols dp and dq
Figure 16.3 A model of obtaining a

poin t on an intersection of two curved
surfaces
denote t.he dist.ances from the origin to the tangential planes Fp and Fq.
Since Po = r( Wo, to) and qo = s( uo, va), the distances are given by
(16.9)
2. Make a plane Fn which passes Po and is orthogonal to both Fp and Fq. Let
Un be t.he unit normal of the plane Fn , whose distance from the origin is dn .
Then we have
Up x U q
Un = 1
dn = Un' r(w, t). (16.10)
Up x u q I'
Let p be the int.ersection point of the three planes Fp, Fq and Fn- It is given
by
(16.11)
(Refer to eq. (3.21) of Chap. 3.) It is an approximate point on the intersection

of r( w, t) and s( u, v).
3. We are to express the difference vectors oPo = P - Po and oqo = p - qo by
the basic vectors r w , 1't and su, Sv at Po and qo' We define vectors tv." tt, s"
and s" which are orthogonal to the basic vectors of the respective surfaces
and their normah:
(16.12)
Since the following incremental relations can be assumed
1'w ow + rt ot = oPo, } (16.13)

su ou + sv ov = oqo,
we obtain the increments of the paramet.ers by using the above-defined vec-

t.ors
(16.14)
(16.15)
Then the updated initial points are given by

Po = r(wo + ow, to + ot) qo = s(uo + ou, Vo + ov). (16.16)
4. If Ipo - qol is not within the given tolerance, the above process from (1)
is repeated unt.il IPo - qol becomes sufficiently small. Convergence of this
met.hod is usually very good, even when the initial points are not near.
'When convergent results are not obtained, information on the nearness of
the two patches can be extracted.
16.3.3 Initial Starting Points and Critical Contact Points

16.3.3.1 Detection of Intersection Loops
Since the method stated above is very effective for detection of an intersection
loop, if one is given several pairs of initial points, pract.ically all the intersection
loops ca.n be detected. Figure 16.4 shows an example obtained by this method.
But if one wants to make the procedure more systematic, the following procedure
is recommended:
1. At first we fix a reference vector nf towards which both intersecting sur-
faces are directed, that is, the rear sides of both surfaces can be seen from
that direction. Determine the extremum search curves cPu = 0 and cPv = 0
for each of the interfering patches rand s. Regions separated by these
curves have monotonic gradients. Moreover, if possible, determine the points
cPuu = cPvv = 0 on the patch s, which we call the max-gradient points. There
is one max-gradient point in each region. Otherwise we select an arbitrary
point. in a region in place of the max-gradient point of the region. These
points are taken as Po.
2. For each Po we fix a point qo on another patch r, which is arbitrary, but not
very far from Po and preferably in a region of large difference of slope from
that at Po. We can exclude regions which surely do not interfere each other,
beca.use we can determine approximately a space in which no interference
exists by t.he data of the extremum search curves of both surfaces.
3. From a pair Po and qo, we can obtain a point on an intersection loop , which
is traced from this point. Using all the pair points we can obtain all the
intersection curves of the two patches. Usually most of points thus obtained
are on the loops already determined.
Intersection curve
Figure 16.4 Intersection of two free-form surfaces. Contour curves are shown
4. If we reverse the role of the surfa.ces sand r and perform the same procedure,
we can catch all the intersection loops more surely, although usually this may
be omitted.
Example. Figure 16.5(a) shows intersection curve patterns of two surface patches
at ~ine relative positions. The shapes of two patches sand r are the same as
shown in Fig. 16.2, but r is rotated by 90 degrees anti-clockwise relative to sand
separated by constant heights. Shapes of the intersection curves vary gradually
or rather a.bruptly as the height changes . At certain heights, new loops are
generated or disappear or one loop separates into two. These curves are obtained
by the process stated above. The z value (height) of the patch r is changed from
h = -1.0 to 1.0 with 6h = 0.25.
Around h = 0 the intersection pattern changes rather abruptly. Figure
16.5(b),(c),(d) are the intersection curves at h = 0, h = 0.001 and h = -0.001.
Figure 16.6 is the enlarged portion at the central part. At h = -0.775 X 10- 4
there is a branch point.
To show t.he effectiveness of the method, we explain a critical case. Figure
16.7 shows the intersection curves at the critical position of the two surfaces.
This position is a little lower than that of the case h = -0.5 of Fig. 16.5: the
left small loop which is seen in the case h = -0.5 reduces to a point at h = -0.65
whose location is indicated by a small dot and the upper-right intersection curve
is shown by a t.hick line. To test how the method of Sect. 16.3.2 can work well
in t.his critical case, we set the point pairs of Po and qo and see if we can obtain
a. convergence point. We move the location of Po to various points on s to
determine allowable domains on s:
( a) (b)
(c) (d)
Figure 16.5 Variation of intersection curve pattern by displacement of one surface: (a) nine
positions from h = -1.0 to h = 1.0 and oh = 0.25, (b) h = 0, (c) h = 0.001, (d) h = -0.001
Figure 16.6 Enlarged patterns of intersection curves around h = -0.775 X 10- 4
R(w , t)
Intersection
curve
S(u,u)
o : Locations of extremum gradient

point on R(w, t)
• : Locations of peak or saddle point
on S( w, t)
Figure 16.7 Example of very small intersection loop

lntersection
A sm all loop (L) - +-V-n-r"rrl
curve (C)
Extremum
sea.rch curves
* Fixed initial point

o Points converging to L
• Points converging to C
x Points not converging
Figure 16.8 Map of starting points in one surface from Fig. 16.7.
Refer t.o Fig. 16.8. A max-gradient point in one region of r is selected as qo,
wh<?se projected location is indicated by a small star marker and the other point
Po is select.ed at va.rious points on the patch s. A blank small circle indicates
locations of t.he point Po converging to a very small intersection loop indicated
by a small dot. and each of black small disk mark points converges to a point on
the upper-right. int.ersection curve. A cross mark point shows a point which goes
out of t.he boundary of the patch. In the figure the extremum search curves of
the patch s which indicate boundaries of different slopes are also shown.
Even in such a. crit.ical case as this, the allowable area for convergence to the
crit.ical point. is large. There are cases of convergence where the distance between
the initial point. pair is great and the dista.nce between the final convergence point
and t.he init.ia.l point. pair is large. But the convergence is mainly determined by
the distance of the pair from the converging point and also the relative slope
difference of the t.angent planes at the st.art.ing points.
16.3.3.2 Critical Points
So long as numerical computation is involved, we cannot find very small inter-

section loops or a cont.act point of two surfaces which are out.side the precision of
the comput.at.ion process adopted. But if we translate one of the surfaces relative
to the other in a near-normal direCtion or make a small offset surface of one of
280 16 S unace Intersections
them, then we can detect an intersection loop, from which we can 'determine com-
mon normal points of t.he both surfaces and approximate shapes of the surfaces
around these point.s. The method is similar to that explained in Sect. 16.5.
When applying t.he method of obtaining an intersection point, monitoring
the dist.ance between two points p and q gives information on the critical point
without using the offset or displaced surfaces. Figure 16.9(a) shows the distance
and the number of iterations. Figure 16.9(b) shows examples of routes to the
convergence points. For a usual convergence to an intersection curve, a few itera-
tions suffice, but for critically separated points, the number of iterations increases
without convergence. When there is an intersecting point, the convergence to a
final point is very rapid, whereas in a case of near contact, but really separat-
ing, after rapid decrese of the distance of the pair points the distance fluctuates
without converging to a final points. The mean value of the distance fluctuation
is an approximate minimum distance between two surfaces in this region. The
above stated techniques are applied for detection of critical contact points.
The following techniques are also applicable for detection of small loops of
the intersection curves.
(i) Observing the pattern of the intersection curves of one surface and a set of
further surfaces, each shifted from the initial position by translation, we
can estimate critical cases and also existence 6f the isolated loops of the
intersection. See Fig. 16.5.
(ii) Since the contour curves of both surfaces are easily obtained, we can test
interferences of the two sets of the contour curves, and this gives all the
. intersection curves of both curved surface. See Fig. 16.2.
If one wants to detect theoretically all the intersection loops including the
critical points, a kind of extremum search curves can be used. For intersection
curves of a patch r and parallel planes F with their normal nf, the extremum
search curves, which are given by
nf . Su = 0, and nf . Sv = 0, (16.17)
can be used to search for all the starting points for the loops of the intersection
curves and the singular points at each of which one of the loops becomes a point
or has a branch point. Similar extremum sea.rch curves can be considered for a
set of the intersection curves of a surface S and offset surfaces r+ of r with va.rious
offset values e, or for a set of intersection curves of the surface s and the translated
surfaces of r along a vector a. We call here the extremum search curves for the
intersection loops stated above the extended extremum search curves. When the
surface r is a plane, the extended extremum search curves become the ordinary
extremum search curves. Equations of the intersection curves of the above stated
surfaces are given by
s = r+ = r + I1T • e, or s = r + a . e, (16.18)
~
=
'0
c-
0.00015
Converging
o
~
=
<l> 0.00010 Not converging
/
<l>
~
<l>
.D
<l>
=
U
j 0.00005
is'"
Number of iteration
(a)
Critically
not contacting
point pair
Converged
point pair
(b)
Figure 16.9 Convergence in a critical case: (a) distance and number of iterations, (b) routes
to a converged point and a critically not contacting point pair
where n T is the normal of r( w, t) and a is the given direction.

The surface s is considered to be covered by the intersection curves. A certain
intersection loop may converge to a point(s) or have a branch point(s) at a certain
offset value or displacement value e. At these points the normal directions of the
surface s coincide with normals of the other surfaces. Accordingly, these singular
points are determined by the following two equations which are the orthogonal
conditions of the basic vectors of s with the normal vector n T of the other surface
r:
n T
• s" = 0, nT • Sv = o. (16.19)
Each of the above equations represents the extended extremum search curves,
and their intersecting points give locations of the singular points.
Since the offset value e or the displacement value e is the parameter of these
extremum search curves, we can determine the values of e at the singular points
and the starting points of the intersection curves of sand r with the offset zero
and the displacement zero. The method provides all the intersection loops, but
tracing the curves given by eq. (16.19) is not easy, because the normal n T is a
function of the parameters wand t, which are also determined as functions of
u and v through eq. (16.18). We obtain simultaneous equations with respect to
the arc length for these parameters and the other parameter e for the offset or
the displacement. We have tried this method and get the expected result, but
compared wit.h the previous method this one does not seem practical, because
the computation load is too much.
16.4 Intersection Curves Described by Differential

Equations
16.4.1 Both Surfaces with Parametric Expressions
Since the intersection curves are usually space curves, they have non-zero torsion,
accordingly when expanded they have terms higher than third degree of the arc
length. Derivation of higher-degree derivatives along the intersection curve from
the two intersecting surfaces of parametric expressions are complicated, but if
we apply the Runge-Kutta process (refer to the Appendix), we can estimate
the values of the higher-degree derivatives by the weighted average of values of
the first derivatives at a few sampled intervals. So we introduce the differential
equations for describing the intersection curves and use reliable and efficient
methods of numerical integration of the differential equations. The methods with
automatic adjustment of stepsize work well for our problems. Hence, we explain
these methods in this section. Surface expressions may both be of parametric
form or one of them may be parametric and the other may be of implicit form,
or both may be of implicit form . .
16.4 Intersection Curves Described by Differential Equations 283
On the intersection curve of two parametric expressions, the following equa-

tion must hold:
s(u, v) = r(w, t). (16.20)
We now assume that a point on the intersection of two surface patches s( u, v)
and r( w, t) has been determined. From there we start the curve tracing. The
intersection curve on each surface must have the same tangent vector on their
intersection, so we have from eq. (16.20),
(16.21)
Multiplying the normal nl of s(u,v) and n2 of r(w,t) with eq. (16.21), we have
two equations:
(nz . s,J(du/ds) + (nz· sv)(dv/ds) = 0, (16.22)

. (nl . sw)(dw/ds) + (nl . td(dt/ds) = O. (16.23)
From the square of eq. (16.21) an arc element ds of the intersection curve is given
by
(16.24)
where E;, F i , G i , for i = 1,2 are the fundamental magnitudes of the first order of
the patches s(u,v) and r(w,t).
From equations (16.22), (16.23) and (16.24), we obtain simultaneous differ-
en~.ial equations for the intersection curve of two surfaces s( u, v) and r( w, t):
du dv
ds = -¢>l(U,V,w,t)/t, ds = ¢>l(U,V,w,t)gl (16.25)
dw dt
ds = -¢>2(U, v, w, t)h, ds = ¢>2(U, v, w, t)g2' (16.26)
where
¢>l( U, v, w, t) = ±{ElR - 2Fdlgl + Glgn- l / Z, }

(16.27)
¢>z( U, v, w, t) ±{Ezfi - 2Fzhgz + G 2gD- 1 / 2 ,
11 = (n2 . sv), gl = (n2' StL), }

(16.28)
fz = (nl . 1't), g2 = (nl . rw).
If we adopt one of the suitable numerical methods of solving differential equa-
tions which have automatic error control and adaptive stepsize, we can trace an
intersection curve of two surfaces without special attention to determining the
stepsize. The steps usually ends at a boundary or returns to the starting point.
If the candidate staTting points still remain, the trace starts there again. During
the integration process, errors can be checked against eq. (16.20), but this check-
ing can be omitted for efficiency, because usually accumulated truncation errors
of this int.egrat.ion can be kept within a given tolerance and the curve returns to
the int.ial point. t.o complete a loop.
If during marching on a.n intersection curve to be traced, normal vectors of
the two surface becomes almost parallel, the integration program halts giving
a warning of vanishing values of terms Ii and gi. This indicates the possible
existence of a. singular point. Then intersection curves around a. possible singular
point are searched using the second-order terms of eq. (16.20). The method is
described in the next section.
16.4.2 Surfaces with Implicit and Parametric Expressions

There are cases where one or both of the intersecting surfa.ces are expressed in
implicit forms which are difficult to transform to parametric expressions. Their
intersection curves can also be traced with the method of differential equation
solving.
Let. one of the surface be given by I(x, y, z) = 0 and the other by s('11, v). On
their intersection curve we have the relations:
x = SX = (i . s), y = sY = (j . s), z = SZ = (k· s), (16.29)
where sX, sY and SZ are x,y , z components ofs('11,v) .

Differentiating the above relat.ions with respect to the arc length of the in-
tersection curve,
(16 .30)
dx
ds
dy
ds
dz
Substituting the above three equations into eq. (16.30), we have
(fx S"x + I yS"Y + f zS"Z) d'ds11 + (fx S"x + f yS"Y + f zS"Z) dv

ds
-
- O. (16.31)
Since Ix , Iy and Iz are expressed by x , y and z, which are also functions of'11 and
v by eq. (16.29) , eq. (16.31) is the differential equation of an intersection curve
expressed in '11. and v. Hence it is solved in a similar way as equations (16.4) ,
(16.6) and (16.5).
16.4 Intersection Curves Described by Differential Equations 285
Intersection curve
Figure 16.10 Intersection of a free-

form surface and an ellipsoid (implicit
expression), both of which are displayed
by their silhouette curves
6.4.3 Both Surfaces with Implicit Expressions

Vhen the ot her surface is also expressed by an implicit form 9 (x, y, z) = 0 as
,ell as eq. (16.30),' we have
dx dy dz
gx ds + gy ds + gz ds = O. (16.32)
"rom equations (16.30) and (16.32)' we determine a ratio (dx/ds : dy/ds : dz/ds)
uch as
dx
1jJ(x, y, z)(fygz - gyfz),
ds
dy
= 1jJ(x, y, z)(fzgx - gzfx), (16.33)
ds
dz
= 1jJ(x, y, z)(fxgy - gxfy),
ds
vhere 1jJ(x, y, z) is the common factor, which is obtained from the normalizing
:ondition,
(dx? + (d Y )2 + (dz?
ds ds ds
= 1jJ(x, y, Z)2{(fygz - gyfz? + (fzgx - gzfx)2 + (fxgy - gxfy?}
= 1. (16.34)
~qua.tions (16.33) together with eq. (16.34) are the differential equations of the
nt.ersection curve, which can be traced by the numerical methods.
8xample 1. Figure 16.4. Intersection of two free-form surfaces of degree 4 x 4.
::::ont.our curves are also shown.
8xample 2. Figure 16.10. A free-form surface and an ellipsoid with three axes
)f different. length.
Example 3. Figure 16.11. Two surfaces of implicit forms: an ellipsoid and a
lyperboloid.
Figure 16.11 Intersection of surfaces

all implicit expressions: an ellipsoid and
a hyperboloid, both of which are dis-
played by their silhouette curves
16.5 Intersection Near a Probable Singular Point

If two surfaces share a common normal vector and a common point, their inter-
section curves may pass the point, or the point may be isolated. But if the two
surfaces have nearly parallel normal vectors near a coincident point on the two
surfaces, there are various pattern of the intersection curves. We treat this kind
of problem here.
The method we use for tracing an intersection curve can be applied to a
point p on the curve, where normal vectors nl and n2 of the intersecting surfaces
are nearly parallel, as small an angle as 10-6 radian or less, though at a point
of equal normal direction, its differential equations (16.22) and (16.23) vanish.
Accordingly, we search around the point p = s( u, v) = r( w, s) for two points
s(u + du,v + dv) and r(w + dW,t + dt) whose normal vectors are coincident.
Instead of the not.ations for normals nl and n2, we change them to n S and n S for
clarit.y. Let their common unit normal be nand t.he distance between the two
surfaces along t.he normal be a small value 0. Then we have the relation:
( 16.35)
dn s = n - n S , dn T = n - nT, 6no = n T
- nS. (16.36)
So long as we can assume that each normal vector varies continuously, the loca-
tions of the desired normal direction are considered very nea.r to the last point of
the int.ersection curve. For a surface region with very small Gaussian curvature
this assumpt.ion does not always hold (refer to eq. (6.36), then in such a region
t.here is no critical point on intersection curves. We can proceed the integration
process with smaller tolerance. Except such a case we ha.ve the equations:
(16.37)
From equations (16.36) and (16.37),
6no = dn s - dn T = n~du + n~dv - n~dw - n;dt. (16.38)
Multiplying n r or n S with eq. (16.35) and neglecting a small value 15 • dn s , we

obt.ain
n r . sudu + nr. sv dv = -15, }
(16.39)
n
S • rwdw + n'· rtdt = 15.
Multiplying n' x n r with eq. (16.35), we have
[n$, nr, su]du + [n', nr, sv]dv

= [n$,nr,rw]dw + [n$,nT,rt]dt - [n$, nT, n]15. (16.40)
Since the last term of the above equation is small of second order, it can be
neglected.
Multiplying n T or n' with eq. (16.38) and considering orthogonality condi-
tions n S • dn' = n T . dn r = 0, we obtain
n T . n~du + n T . n~dv = E, }
(16.41)
n$ . n~dw + n S • n;dt = E '
where E= n T. 6no = -n$ ·6no.

Solving equations (16.39) and (16.41), we have du, dv and dw, dt as functions
of 15 and E:
(16.42)
where
Dl = {_(nT. n~)(nT . sv) + (nT . n~)(nT . sun, }

( 16.43)
D2 = {-(n$· rt)(n$ . n:-.,) + (n$ . rw)(n$ . n;)}.
Substituting the above relations into eq. (16.40), we obtain an equation for 15
A15 = BE, (16.44)
where
A = D2 { -[n$, nT, su](nT . n~) + [n nT, sv](nT . n~)}

S ,
- Dd[n$, nT, rw](n' . nD - [n', nT, rt](n' . n~)}

+D 1 D2 [n', nT, n], (16.45)
B = Dd _[n S , nT, rw](n' . rt) + [n" nT, rtl(n' . rw)}
+D2{[n"nT,su](nT ·sv) - [n',11T,s,,](11T,su)}. (16.46)
From eq. (16.44) t.he distance 8 between two surfaces at the oommon normal
points, and t.hen from eq. (16.42), du, dv, dw, dt are determined. The last term of
eq. (16.45) must be neglected initially and if an iteration is needed for accuracy
enhancement the t.erm is added. Thus the common normal points s( u+du, v+dv)
and r{w+dw, t+dt) of the two surfaces are obt.ained. In the a.bove derivation, we
assume that the distance to a common normal is so small that the Weingarten's
formulas hold.
'When t.he common normal exist.s, the surface shapes around t.hese point.s are
expressed, using t.heir respective principal directions and the common normal as
their coordinat.e systems (x, y, z) and (~, T), 0, by
z = (1I:1X2 + 1I:2y 2)/2, }

( = (k1e + k2T)2)/2, (16.47)
z = (+ 8,
where 11:1,11:2 and ki, k2 are their respective pairs of principal curvatures. Rotate
t.he coordinates syst.em (~, T), 0 by an angle ¢, which is the angle between the x
and the ~ axes, around its ( axis to eliminate ~ and T):
~ = x cos ¢ + y sin ¢, T) = -x sin ¢ + Y cos ¢. (16.48)
Then we have an equat.ion for x and y:
(16.49)
where
](1 = k1 cos 2 ¢ + k2 sin 2 ¢, }
](2 = k1 cos 2(¢ + 90) + k2 sin2(¢ + 90), (16.50)
h = (k2-k1)cos¢sin¢.
The const.ant.s ](1 and ](2 are the orthogonal normal curvat.ures of r( w, t) in the
principal directions of s(u, v) at the common normal points.
Equation (16.49), which represent.s the intersection curve of t.he two surfaces,
is a quadratic curve, whose discriminant D is given by
D = (11:1 - ](1)(11:2 - ](2) - h2

+ k1k2 - 11:1](2 - 11:2](1
11:111:2 (16.51 )
](; + ](~ - 4(I{:"](;" - 11:1](1 - 11:2](2),
where ](g and ](m are the Gaussian curvat.ure and the mean curvat.ure at the
point., t.heir superscript.s discriminating the respective surfaces.
We can classify t.he shape of t.he curve according to equations (16.49) and
(16.51).
If D "# 0, it has a cent.er at the origin. Rotating the coordinate system
(x, y, z) around t.he z axis by an angle
(16.52)
we t.ransform eq. (16.49) to
(16.53)
where the coefficients ).1 and ).2 are roots of the characteristic equation for
eq. (16.49) (refer t.o Sect. 1.9):
).2 _ (1):1 + 1>:2 - K1 - K 2 )A + D = 0,
Using equations (16.50) and the mean curvatures K:", K;", the above equation
is changed to
).2 _ 2(J<:r, - K:n). + D = O. (16.54)
The shape of t.he int.ersection curve is classified according to the root of the above
equation:
• 0> 0:
(K:r, - K:n) > 0, D>0 ellipse, ( 16.55)

(J<:" - K:n) < 0, D>0 imaginary ellipse, (16.56)
D<0 hyperbola. ( 16.57)
If (1):1 - K1 )(K2 - K 2 ) ::; 0, the intersection curve is surely a hyperbola, as

easily inferred from their geometric configuration near the origin. This case
is naturally included in the above case of D < O.
• 0 = 0:
D< 0 two lines, (16.58)

D > 0 and (K:r, - K:n) > 0 one contact point, (16.59)
D=O a parabola or two lines. (16.60)
A similar analysis can be applied for the intersection curve of surfaces, one or
bot.h of which is (are) expressed by implicit formes).
Note t.hat in calculating eq. (16.44), quantities of the denominators and the
numerators aTe very small, and repeated calculations for approximation are
needed to reduce the truncation errors.
Example. Figure 16.12 shows int.ersection curve around a common normal vector
of t.wo surfaces, whose distance is 0 = 8 X 10- 7 . Figure 16.12(a) is global view of
the int.ersection curves, which are hyperbolic in t.he very near region ofthe center.
To observe t.he shapes of the two surfaces around the points of common normal,
the contour curves made by two cutting planes, whose normals are the same as
the common one, are shown in Fig. 16.12(b). The distance of the two cutting
planes is 0.0002, where the length of an edge of the patches is of order 1, and the
figure shows clearly the triple intersecting points of the three surfaces. We can
know t.hat the principal curvatures of two surfaces are almost equal, because the
shapes of t.he cont.our curves of the two surfaces are elliptical of similar size.
Intersection curve
(a) (b)
Figure 16.12 Shape of intersection curves near a critical point: (a) global view of intersection
curves, (b) contour curves around a common normal point and intersection curves
16.6 Intersection of Offset Surfaces

16.6.1 Intersection with a Pla11e
The intersection curves of an offset surface and equidistant planes are needed
when the center of a ball of a ball end-mill is controlled to move on them for
machining of the original surface.
Let s+(u,v) be an offset surface of a part surface s(u,v) to be machined by
a ball-end mill of radius r,
s+ (u, v) = s( u, v) + n( u, v) . r, (16.61)
where 11( U, v) is the unit normal vector of the part surface at u and v.
If the center of the ball-end mill moves on the offset surface, it cuts the part
surface. Let the unit normal vector of the plane, whose distance from the origin
is d, be 11f . Then from eq. (16.1), we have
(16.62)
After the starting point of the center is determined, the intersection curve can
be traced using equations (16.5) and(16.6), in which ¢11. and ¢v are given by
¢11. = st· 11f = (S11. + 1111.· r)· 11f, }

(16.63)
¢v = s: ·11f = (sv + 11v • r) ·11f '
16.6 Intersection of Offset Surfaces 291
Center of a rolling ball
Locus of contact poin t
Offset surface
Figure 16 .13 Intersection of an offset curve and parallel planes. Locus of contact point of a
rolli ng ball is also sh own
where the derivatives n" and nv of the normal of s( u, v) can be calculated by

using Weingarten's formulas (refer to equations (6.28) and (7.37)- (7.39)) . Figure
16.13 shows the intersection curves and the loci of the contact point of the ball.
In this case the radius of the ball is too large, and though the loci of the center of
the ball are smooth, the corresponding loci of the contact points on the original
surface show complicated behavior around regions of large curvatures.
16.6.2 Intersection of Two Offset Surfaces

For smooth blending of two intersecting surfaces, an envelop surface of a ball
rolling in contact with the two surfaces is used. A locus of the center of the ball
is det.ermined by the intersection of two offset surfaces , whose distance to the
original surfaces is equal to the ra.dius of the ball.
16.6.2.1 Two Parametric Surfaces
In this case equations (16.22) and (16.23) become

(n2· s~)(du/ds+) + (112· s~)(dv/ds+) = 0, (16.64)
(nl . r~)(dw/ds+) + (nl . ri)(dt/ds+) = 0, (16.65)
where ds+ is the line element of the intersection curve, a.n d there is a relation
(dS+)2 = E+ldu2+2F+ldUdV+G+ldv2, }
(16.66)
(dS+) 2 = E+ 2dw 2 + 2F+ 2dwdt + G+ 2dt 2.
The symbols with a superscript + correspond to those of the original surfaces.
They are given by equations (7.37)-- (7.39) . Equations (16 .64) and (16.65) are
/
Center of a rolling ball ........._--................ Offset surface
// "'.'"''
..,/ / ...: ~;- .......
,J"' ............- ..'
,/
Original surface
(a)
Original surface
Offset surface
(b)
Figure 16.14 (al Intersection curves of two offset surfaces, (b) intersecting point of three
offset surfaces
the different.ial equations of the locus of the center of a rolling ball contacting
the two surfaces; t.hey can be traced numericaUy by solving t.he above differential
equations.
Examples. Figure 16.14( a) shows the int.ersect.ion curves of the two offset surfaces
and (b) shows t.he intersecting p'oint of the three offset surfaces.
16.6.2.2 A Parametric Surface and a Surface of Implicit Form
Let p be a point on a surface 51 whose equation is f(x, y, z ) = 0 and III a unit

normal there. A point q offset from p by a distance r is given by
q = p+ lli' r. ( 16.67)
Let s+(u, v) be an offset parametric surface. On its intersection curve with q an

equation s+ (u , v) = q = p + III • r holds, from which we obtain its line element
vector:
ds+ = s+ "du + s+ "dv = dq = dp + dll l . r. (16.68)
Multiplying the normal vector III of 51 with both sides of the above equation,
we have
(lll . s~)du + (lll . s~)dv = III • dq = O. (16.69)
Since III is proportional to (Jx : fy : fz), we can determine the ratio (du : dv) as
a function of u, v, x, y and z.
Multiplying the normal vector ll2 of s+ with both sides of eq. (16.68), we
obtain
ll2 . (dp + dll l . r) = O. (16.70)
Since panel III are the functions of x,y and z, eq.(16.70) is transformed to
where, for instance, llix is the derivative of III with respect to x.

Moreover we have the relation
df(x, y, z) = fxdx + fydy + fzdz = O. (16.72)
From equations (16.71) and (16.72), we obtain a ratio (dx : dy : dz ). As a

normalizing condition, we take the line element ds+ of the intersection curve
which is given by
ds+ 2 = E+ du 2 + 2F+ dudv + C+ dv 2 =

(i + rlll x?dx 2 + (j + rllly?dy2 + (k + rlllz?dz2
+2(i + rlllx)(j + rllly)dxdy
+2(j + rllly)(k + rlll.)dydz
+2(k + rlllz)(i + rlllx)dzdx. (16.73)
From the above equation and the ratios (du : dv) and (dx : dv : dz ), we can
determine (du/ds+) , (dv/ds+), (dx/ds+), (dy/ds+)and (dz/ds+) as functions of
u, v, x , y and z. We can trace the intersection curve by solving these simultaneous
differential equations.
Figure 16.15 shows an example of the intersection of offset. surfaces: one is a
pa.r ametric and t.he other is an implicit expression.
Center of a rolling ball
Figure 16.15 Intersection of offset

surfaces, one a free-form and the other
an implicit expression
16.6.2.3 Two Sl~rfaces with Implicit Expressions
°
A similar method can be applied for tracing the intersection curve of two implicit
surfaces 51: f(x,y, z ) = and 52 :g(u,v,w) = 0, in which symbols u, v and w
are used instead of x, y and z for discriminating purpose only. A point q on an
intersection curve of the offset surfaces is given by
(16.74)
°
where Pi and ni ,( i = 1,2) are a point and its unit normal vector on the original
surface 51 or 52' Considering ni·dpi = and ni ·dni = 0, we obtain the relations:
n2dp1 + n2dn1 . r = 0, df = 0, }
n1dp2 + n1dn2 . r = 0, dg = °. (16.75)
From the above equations, we obtain the ratios (dx : dy : dz) and (du : dv : dw).
The line element of the intersection curve is given from eq. (16.74) by
ds+ 2 = (i + rn1x)2dx2 + (j + rll1y?dy2 + (k + rll1z)2dz 2

+2(i + rn1x)(j + rn1y)dxdy
+2(j + rn1y)(k + rn1Jdydz
+2(k + rn1z)(i + rn1x)dzdx,
= (i + rll2x?dx 2 + (j + rn2y)2dy2 + (k + rn2J 2dz 2
+2(i + rll2x)(j + rn2y)dxdy
+2(j + rn2y)(k + rn2 z)dydz
+2(k + rll2 z)(i + rn2x)dzdx.
(16.76)
From the above equations, six simultaneous differential equations of the first
order with six dependent variables, whose left-hand sides are (dx/ds+), (dy/ds+'),
Envelope surface
Figure 16.16 Intersection of two offset surfaces, two implicit expressions
(d z/ds+) and (dv/ds+), (dv/ds+), (dw/ds+), are determined. By solving them

we can obt.ain t.he intersection curve and the contacting curves on the original
surfaces.
To solve these differential equations, we need initial conditions or a starting
point. Since t.he start.ing point is on the intersection curve, a. method similar to
t.hat. given in Sect. 16.3.2 can be applied.
Figure 16.16 shows an example of the intersection of offset surfaces of the
two original surfaces with implicit expressions.
17 Applications of the Theories in Industry
17.1 Introduction
Various theories and methods explained in the previous chapters were originally
developed to meet the practical requirements of Japanese industries. In this
chapter their application examples are described: one is style design and stamp-
ing die design and manufacture in the integrated CAD/CAM system of a large
motor car company. The other is also style design, of small audio devices in a
large electronic appliance company.
In these systems", computer-internal models or descriptions of the design ob-
jects were needed in place of engineering drawings. Accordingly, we first confirm
the conventional roles of engineering drawings in design and manufacturing and
then explain the reasons for replacing engineering drawings by the internal mod-
els. Next an example of this replacement is given from the motor car industry.
Since readers of this book may not be experts on the cat industry, we explain con-
cisely the conventional processes of style design of a car body and manufacture
of its stamping dies and then we show how the internal models are constructed,
manipulated and used in a new system with the underlying philosophy of the
conversion from the conventional system.
The other example is the design and manufacture of free-form injection-mold
products for in-the-ear headphones in audio devices. In this case the shape of
the products is a very important factor to attract customers together with their
performance. In both cases their engineering drawings are not enough to convey
information for manufacturing and to perform prior evaluat.ion of the products.
Comparing the scale and complexity of the products, the design and manufacture
of motor cars include far more difficult factors than those of electric appliances,
but the latter must be also excellent in design and cost, because the performance
of the product is almost the same as that of similar competitive products on the
market. Anyway there are common factors in design methods in both industries.
In this chapter we treat mainly the case of motor-car style design and stamp-
ing die manufacture and then touch on the case of the electric appliance. As the
scale of the latter is small, it may be easy for understanding.
17.2 Engineering Drawings and Geometric Models
Engineering design is a kind of information-generating activity for manufadur-

ing an object which performs specified tasks and satisfies given aims. For this
purpose, the designers have to set Clearly the specification of the requirements
298 17 Applications of the Theories in Industry
for the aims, and then develop it to specify every detail of the object, to de-
termine components and materials to be used. And at the same time they are
required to optimize the conflicting conditions for performance, reliability, cost
and marketability within the given constraints of time, manpower and money.
Manufacturing activities include preparing materials and components, decid-
ing machines and their operations for production of the objects and assembling
the components which are specified by the design process. They include gen-
erating control information for manufacturing and performing the instructions
thus produced and finally inspecting whether the manufactured objects satisfy
the specifications defined in the early stages. Of course these activities are under
the constraints of cost and production time.
In design and manufacturing before the advent of computers in these fields,
information generation and storage and tra.nsfer were mainly performed through
engineering drawings. Since their roles were and still are very important, fre-
quently the design activity was considered equivalent to the making of engineer-
ing drawings.
Design engineers represent their design intention and designed objects mainly
by engineering drawings, in which materials to be used, shapes of designed ob-
jects, and various information on meeting the specifications given to the designed
products are described. Through the process of making engineering drawings,
designers put their various ideas in order and select the best ones to optimize the
conflicting conditions. Completed engineering drawings are media to be com-
municated correctly to other people and they must carry basic information for
manufacturing and inspection. There are various types of engineering drawings
according to their uses and they are gua.ranteed to have no errors and no contra-
diction. Through them, other people not only understand the designed objects,
but also they can manufacture the real objects the drawings specify.
These engineering drawings are produced by persons who understand the
objects to be manufactured and have their images in their minds and know the
rules for how to express the objects. Persons who can read them not only know
the rules of the drawings but have the ability to construct in their minds the
images of the objects the drawings represent.
Even today the computer cannot understand engineering drawings or pro-
duce them automatically in the way that expert designers can. However, if
the computer can produce, according to our instructions, information which the
engineering drawings contain and can process its meanings, that information
is equivalent to our explicit knowledge obtained from the engineering drawings.
Without formation of this type of information, the computer is merely a machine
for simple data processing in design activities.
The source of this explicit information which the engineering drawings hold
is the geometric models or product models which are produced in the computer
by relevant commands given by designers who can interact with the models and
extract necessary information from the models. If such internal models can be
17.3 Examples of Integration 299
constructed, the computer has the possibility to support all the design activities
and control t.he manufacturing processes without human intervention for data.
transfer and conversion.
In the construction of this kind of model, to fully exert the designers' skill
and t.alent, t.heir way of thinking and practices must not be greatly altered so long
as more convenient and helpful alternatives cannot be offered to them. Usually
this is the most difficult requirement for a computer system designed to accept
experts' methods directly.
For advanced CAD/CAM systems which expert designers can use willingly,
the human interface with the systems is very important. The interface means
not only the physical interface, but also the mental interface between the de-
signers' ment.al model of the objects to be designed and the comput.er's int.ernal
information model of the object to be constructed. In design and manufactur-
ing activities, a great part of this internal information model is related to the
geometric model of the object to be designed and manufactured.
One of the main objectives of this book is to present theories and methods
of free-form surfaces which are easily and intuitively understandable to practical
engineers, some of whom are users of the systems or their developers. Another
objective is to show that an excellent user interface is an essential component of
a successful advanced CAD/CAM system.
17.3 Examples of Integration

As the integration of CAD and CAM is ideal, but difficult to construct, we show
one' real example in this chapter and explain how it was planned and how it
has been developed. The example is taken from car body development processes
in a car manufacturing company in Japan. For ease of understanding, first the
conventional processes and then the converted ones are explained.
17.3.1 Conventional processes

Before the advent ofint.egrated CAD/CAM, in conventional processes of car body
development there were repeated transfers of information from one drawing of
the previous process to further drawings or real models of current processes and
from one real model produced in the previous process to further drawings or
models of t.he current processes. These tasks required considerable manpower
and degraded t.he accuracy of the data. Figure 17.1 shows the convent.ional car
body engineering processes, each of which is explained in t.he following.
- Style design
Style designers produce full-scale clay models based on their conceptual
sketches. After these models are authorized by the Board of Directors, the
models are measured to obtain 3D digitized data, which are used for produc-
tion of full-sized external drawings. They are used as the basis of the external
(I) S'yle Design

ItJea Sketch
II~~I
Clay Model+
I~ I
Ex ternal D rawing •
IltMtHll·ffi I
12)80 dy
oeSlgn
5""c'"'. ' l I
M.ilS ler Dr awing
IIJrum-~ I
Pro duc,ion Drawing "
I ~m~
(3) 5 l.;Impl" !IOle
OeslQfl ~ L
Ole Dlawi ng
_ 1.lml
I ~ S)
§3 I
14) M .ster Model l
S tan lping Oie Design
l 't
1~
MaSler Model Checker M orjel
1 ~
-"!d -
1- I
S tamping Die
~ Fixture
J
I ~ I 1 ~\~!Cll
I
Figure 17.1 Conventional design and manufacturing processes of car body and its stamping
die (courtesy of Toyota Motor Corp.)
17.3 Examples of Integration 301
shape of the cars. Drawings for interior components are produced in the same
manner.
- Body structural design
Based on the full-sized external drawings and the base plan of the structure,
the structural design of the car is investigated. A full-sized master drawing is
made, based on which product drawings of each part of the body are produced.
They contain all the information needed for their manufacture.
- Stamping die design
Based on the drawings produced in the previous processes, the face shape of
the stamping die as well as the structure of the die are designed. The die-face
design includes special shapes called blank-holder surfaces, step draws and
draw beads. They perform important roles during the stamping action of the
pressing machine to deform a steel sheet, which is set in the cavity of the die,
into the designed shape without surface distortion or fracture of the sheet.
- Master model arid stamping die manufacture
The master model which becomes the standard for the following manufactur-
ing processes is produced based on the previously defined shapes expressed
by the drawings. The stamping dies are produced and checked against the
master model. Their manufacture requires special skills and manpower, and
correction of the die-face shape is frequently needed following a try-out with
the manufactured die. This takes very much time and effort.
- Checkers and jig manufacture
Copies of the master model are made for the sub-contractors of making welding
fixture jigs and checkers.
In this example, since the relevant information was transferred by analog

copying and analog checking, and data were interpreted by persons. Therefore,
the manpower and time required in copying and interpreting were great and the
management of data accuracy was a serious problem. After the introduction of
computers, this data transfer could not be abolished because the computers were
used for calculation and making drawings by instruction and needed separate
data for each independent task. There were no common databases or internal
models accessed from various tasks.
17.3.2 New Integrated Processes

Refer to Fig. 17.2 which shows the structure of an integrated CAD/CAM system
developed to eliminate the above stated analog data storage and transfer through
engineering drawings and the manufacture of analog real models. Simple con-
version of the conventional system explained in the previous subsection to the
conceptually new system was impossible, because construction of the indispens-
able internal model which replaces engineering drawings was quite different from
making the engineering drawings and clay models, which are directly connected
to the skills and intuition of designers and modelers. Since the clay models are
I Planning l
----- ------ --------------------t-- -------- ----- ------- -------- - -,,,
~-~- ,
:l§j I Styling design f.---.
L. ____ _
Proto-type
! Structural
manufacturing ~
Body structural design ~
r-----
, -,, ,, analysis
,, !
, :I Stamping die design ~
l_~~~~~~~ ~ ~~~~~~~~~~~~~~~~~~~~~~~~~~ ~ ~~~~~~ ~~ --,,
,
Stamping die IStamping die machining I
!
I Stamping die finishing I
manufacturing
,
L. _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ ___ __ 1
Figure 17.2 Integrated CAD/CAM system replacing a conventional system (courtesy of

Toy ota Motor Corp.)
easily understood by others , and features of the shape which the designers want
are built into the clay models by collaboration of the designers and the mod-
elers, their checking a.nd evaluation are appa.rently easy. Accordingly, from the
standpoint of the style designers, use of the internal model must be as easy as
that of the engineering drawings and the clay models in their construction and
evaluation. And all the complexity of the expression and handling of the internal
models should be hidden and transparent to them.
In the new system shown in Fig.17.2 the most remarkable characteristic is the
use of t.he internal models, which become the databases, and interactive graphic
terminals, bot.h of which replace the engineering drawings and the clay models.
In style design , the designer's conceptual shapes depicted on drawing sheets are
supplied to the system as input data, which are manipulat.ed interactively by the
designers to const.ruct t.he internal model of the car body shape instead of making
a clay model. On the other hand a cla.y model is produced by the machine using
the internal model data and is used to evaluate the overall style design produced
from the internal model. The body structure design subsystem uses the shape
data for its design base and the designers create structures and product drawings
with interadive display terminals using the body structure data base. Large scale
FEM calculations are performed in a dedicated system to which input data are
supplied from this subsystem.
In the stamping die design subsystem, the necessary information is supplied
from the above stated subsystems and from its own database, and the die-face de-
sign is carried out by the designer at an interadive terminal observing simulated
17.4 Style Design System 303
stamping processes. In the master model and the stamping die manufacture sub-
system, the va.rious real master models are replaced by the internal model, from
which required shape information can be accessed freely to control machines
producing stamping dies and checking their accuracy. Also from the internal
model, various real master models and fixtures are manufactured for supply to
the sub-cont.ractor part makers where conventional manufacturing methods are
still used.
The important breakthrough of the system is the successful construction
and use of style design and die-face design subsystems. Of course, the body
structural design subsystem is the largest and central one, but this is the powerful
extension of a CAD system for ma.king engineering drawings. So, we explain in
the following sections the style design and die-face design subsystems which a.re
more characteristic of the advanced CAD/CAM systems.
17.4 Style Design System

17.4.1 Importance of Shape Design
Different design processes for the shapes of products have been developed in
various fields of engineering. When the aesthet.ic aspect of the products is a main
factor for their value as commercial products, specialists called style designers
have the leading role in the shape design process. They have special talents and
characters, which are different from those of engineers who engage in the design
of functions or structures of the products. For example, in motor cars, their
exterior and interior design are very attractive factors for their users together
wit.h t.he ca.r's performance. Accordingly, their style designers have been working
as professional specialists and their tasks have not been treated directly by the
comput.er. And most of them think of themselves as artists and that their works
are among those least a.daptable to computerization.
However, it has been recognized at the Toyota motor car company in Japan
that this pa.rt is one of the bottlenecks to manufacturing high quality products
a.nd to attaining better productivity in the whole process of new car development.
There, t.he st.yle design process and evaluation methods for design qualit.y were
analyzed with the colla.borat.ion of the style designers . And with the result of
this analysis an a.dvanced interactive style design system has been developed,
which the style designers have been pleased to use. The style designers feel that
with this syst.em their ability is enhanced by making various design alternatives
available to t.hem and by decreasing the time they consume in their routine work
such as drawing and clay model making.
17.4.2 Two Aspects of Style Design

We give a concise check list of motor car style design. In it there are two aspects:
motor cars may be seen as engineering products and also as aesthetic objects.
Items of these aspects are:

- Aestlletic a.spect
Newness. Level of class in feeling. Proportion and balance of style.
Composition of surfaces.
View of surfaces; high light, shading, feel of volume.
- Engineering product aspect
Ease of use, sense of comfort.
Structure of body, strength and rigidity, aerodynamic characteristics, new
mechanisms, new materials.
Law and regulations.
Manufacturability (stamping, assembling and surface processing).
Cost.
The design tasks proceed with these va.rious factors in both aspects being con-
sidered. The style design processes consist of
(a) the process of converting from images to real shapes,
(b) the process of adapting to the requirements from the engineering product
aspect.
(c) the process of checking and evaluating the created shapes.
These processes are iterative ones, such as creating idea sketches and rendering
drawings, making various engineering drawings and producing clay models of one-
fift.h size, and returning to the previous stages as a result of the checking processes
in bet.ween. Fina.lly, these processes conclude with complete clay models of real
sizl':. So far, up to this stage in the design processes, computers have not been
used, because t.he tasks were considered to be a realm for the sensitive talents of
the specialists of style and model making.
17.4.3 Computer-Aided Style Design

To promote and rationalize the tasks performed by the style designers and also
to produce the precise data for the models used in the other sections, computers
have to be introduced in this design area from the very beginning of shape
design. Manual drawings and clay models have to be replaced by interactively
constructed models which can be displayed on CRT with appropriate conversion
from the internal description. The constructed internal models must be evaluated
and corrected from both aspects, as aesthetic objects and engineering products.
If the checks are passed, then the models provide the fundamental data for the
succeeding design tasks, and finally they can replace the solid master models
used in the manufacturing stages.
The following items have been realized in the system to attain the above
stated objectives.
(1) Easy input. method for three-dimensional curve segments from designers'
plan and elevation drawings. Conversion to 3D curve segment expressions.
(2) Evaluation of quality of constructed curves and their modification.

(3) Systematized methods of making surfaces which correspond to the manual
procedures for making clay models.
(4) Evaluation and correction methods for defined surface segments.
(5) Various display techniques and object manipulation techniques.
(6) High-quality display and high-speed response time of the system.
(7) High-speed NC machining for a real-sized clay model from its internal model
data and preparation of its engineering drawings.
The surface construction process is shown by the example in Fig. 17.3.
(a) shows input data of character lines of the orthogonal side views. These
charader lines are converted to 3D curves in the system.
(b) shows a cross-section shape which a designer want to make,
(c) shows construction of a side window using the data of the cross-section.
(d) shows the base surfaces of the cabin constructed from the above data.
(e) shows the trimmed base surfaces. With the base surfaces the object is ex-
pressed roughly. Then the boundary regions of the base surface are modified
by the approach surfaces and the abrupt change of curvature is rounded by
the fillet surfaces.
(f) and (g) show the construction of approach surfaces and fillet surfaces and
their trimming.
(h) shows the final state of the designed surface.
As shown above, the base, approach and fillet surfaces are hierarchical classes of
surfaces.
In the procedures for defining surfaces, which is the item (3), the types of
the surfaces which have the specified highlight features are selected besides their
classes. Wit.h this systematization, designers can express their intentions clearly
and can converge quickly to their intended styles with the highlight features.
In items (2), (4) and (5), by the functions for evaluation of curves and sur-
faces, designers can evaluate the quality of their designed curves and surfaces
objectively and with the functions for shape control of the system they can mod-
ify the shapes without violating their intended features.
In item (7), finally, the model is approved by the Board of Directors of the
company. Then the information on the model is transferred and used in the
various departments of design and manufacturing as the shape database . This
is the outline of the computer-aided style design.
17.4.3.1 Input of Simplified Drawings
The style designers and modelers of motor cars use orthogonal projection draw-
ings to make a solid model of a car with clay, in which the main character lines
and vertical sectional curves of the car body are depicted. They make these
drawings from their various ideaS and rough sketches of the car they want to
( I 1
(a) (b)
(d)
(f)
(g) (h)
Figure 17.3 Example of body surface design by CAD: (a) input of character lines, (b) input
of section curves, (c) design ofside window part, (d) base surfaces of a cabin, (e) after trimming
of base surfaces, (f) design of approach and fillet surfaces, (g) after their trimming, (h) final
state of design (courtesy of Toyota Motor Corp.)
design. These are not precise drawings in a strict sense, but they contain the
main features of the designers' images of the car. If these curves are to be in-
put to the computer for style design, the characters of the curves have to be
transferred and if any geometric contradictions exist in the drawings, they have
to be removed. So far, to input a curve to the computer, the curve is traced
as a point sequence by a person using a cursor and a digitizing tablet. From
these data, a curve-fitting method of least squares is applied to obtain a smooth
curve, but with this approach a desired result which conveys the character of the
original curve is not always obtained. To improve the situation, a new method
was developed.
Since designers create curve segments whose curvature profile varies mono-
tonically by using appropriate curved rulers, besides least-square fitting, infor-
mation on the curved rulers used by a designer at certain stages of drawing can
be transferred to the computer. These curved rulers have been systematically
prepared or sometimes newly created by designers according to their needs. The
ruler information consists of its identifier, its used side, and its location on the
drawing sheet. From this information, the mathematical expression of a single
curve segment can be determined in the computer, and from the geometrical
properties of the pair of projected curves, the corresponding space curve is ob-
tained by a calculation which was explained in Sect. 5.5. Curve segments are con-
nected to make their curvature distribution as smooth as possible; see Sect. 10.5.
Methods of modifying input curves, without violating the original characters of
the curve, are provided. They consist of the following functions:
(a.) Shape unmodified: translation, rotation, sliding, extension.
(b) Shape modified: change of tangent angles, change of tangent magnitudes.
(c) Shape transformed: one end-point moving with the tangent free, one end-
point moving with the tangent constrained, transforming in offset direction.
The basic data are manipulated, evaluated and modified by the designers
using interactive graphic displays to construct a wire-frame model of the car.
17.4.3.2 Classes and Types of Surfaces in Style Design of Motor Cars

To facilitate construction of a high-quality surface, surface segments used in a
car body are classified by their methods of creation and the locations at which
they a.re used. The types of surfaces are called (quasi) cylindrical, conical and
spherical or symbolically +, Hand 0 types by analogy with their generating
elements. Each one can be used in various places as a class of base or approach
or fillet surfaces.
The cylindrical ( + type) surface is created by sweeping a certain (variable)
curve, which called a basis curve, along one guide curve. The conical ( H type)
surface uses two guide curves along which a basis curve moves. For the spherical
( 0 type) surface, four basis curves which make closed area move toward the
center of the surface with blending of the sweeping surfaces. These construction
308 17 Applications of the Theories in Ind ustry
Method of generator Highlight pattern
<II
a.
~
OJ)
e
OJ)
.....
<II
a.
~
N ,
\Q
<II
a.
~
,
Figure 17.4 Types of surfaces and their highlight patterns and their generating methods
(courtesy of Toyota Motor Corp.)
processes correspond to features of the highlight lines of desired surface. See

Fig. 17.4.
While the base surface determines the skeleton of the whole shape, the ap-
proach surface modifies the parts of t.he base surface to make varia.tions in the
surface being designed and to make transient regions within the base surface.
The fillet surfaces are rounding parts along the connection with the other sur-
faces to reduce abrupt changes of curvature. With these standardized surface
construction methods, together with connection, interfering and trimming of
surfaces, a basic internal model is constructed. At this stage, the quality of de-
signed surfaces is evaluated by the highlight check. And t.hen det.ailed surface
design follows . The number of the commands which create and process curves
and surfaces including their evaluation is as many as t.hree hundred. They are
classified int.o nineteen categories shown in Table 17.1.
17.4.3.3 Evaluation of Curve and Surface Quality
The next important function in a style CAD system is evaluation of the quality
of the constructed curves and surfaces. In the manual process, the curve quality
is checked by style designers observing from various directions the shape of the
Table 17.1 Nineteen categories of commands
Category Related items

1. digitizer input point sequence, straight line, circle/circular arc,
tangent, etc.
2. point coordinate values, drafter input, point on line,
point on surface, intersection of two lines ,
line and surface, etc.
3. line one, two points, angle , tangent, normal, etc.
4. curve circle, circular arc, curved ruler, fitting, etc.
5. adjust.ment connection, correction, generating line,
integration, area definition, etc.
6. 2D to 3D two projection curves, family curves, additional curve,
rotation, etc.
7. line modify extension, rotation, slide, end-point displacement ,
offset, etc.
8. control control polygon or net manipulation.
9. check connection, tangent, curvature, highlight ,
hard point, etc.
10. surface plane , patch generation of three kinds, sweeping,
rotation , symmetrical surface, etc.
11. section intersection, projection, etc.
12. trim, delete surface, line, figure, etc.
13. mat.h.op. coordinate values, distance , angle, area, etc.
14. std. parts tire, lamp, etc.
15. drawings block registration , layout, size, line attributes, etc.
16. NC data. tools, ma.chining area, interference, cutter location path,
cutter location cont.rol, etc.
17. data. t.ra.ns. to and from other systems.
18. system
19. help/t.est
curve drawn in a la.rge scale on a sheet. The style designers' keen observing
eyes ca.n point. out. regions of a curve as good, uneven, ext.ending, bulged and so
on. To int.erpret. t.heir words , logarithmic values of radii of curvature along the
inspected curve were plotted and the regions where the designers pointed were
found to correspond to non-smoothly varying regions of the plotted curve. This
plot. is now called the test of profile or distribution of the radius of curvature.
The surface quality is checked in the manual process by the designers ob-
serving a real surface of the clay model, but with the comput.er met.hod , a CRT
surfa.ce is too small and has inadequate shape and its display quality is not. good
enough to evaluate the quality of designed lines and surfaces in space.
Even if the expressions of the surfaces to be designed satisfy conditions on

patch connection and constraints specified on the shape, the preliminary designed
surfa.ces do not always have characteristics the designers want to have. As their
mathematical expressions are not.hing but applications of interpolation formu-
las, it is not necessarily guaranteed t.ha.t the expressions generate the favorable
properties all over the designed surfaces.
On t.he ot.her hand, the favorable properties themselves are not clearly de-
fined mathematic any, and they depend on the designers' subjective or intuitive
recognition and int.erpretation. There are no general methodologies for t.he eval-
uation of t.he aest.hetic aspect. of shapes , but for the manual design of free-form
surfaces of motor cars, a set of designers' and the modelers' empirical rules fot
const.ruct.ing surfaces of good quality seems to have been implicitly established.
Accordingly, to introduce a CAD system in this field, it must have feat.ures t.he
experts in t.his field can accept.
A CAD system' must provide its users not only with various surface definition
methods , but also with surface quality evaluation techniques. Otherwise users
cannot proceed t.heir design process further with confidence. Since t.he current
manual design met.hods and evaluat.ion crit.eria cannot be applied directly to
t.heir computerization, it is important to systematize the procedures the style
designers engage and to clarify what they observe .for t.heir evaluation of the
qualit.y of surface in terms of engineering and mathematical formulations. Then
an acceptable CAD system for shape design can be constructed.
One possible met.hod of checking the quality of a designed surface is t.o sim-
ulate the highlight of the surface. The highlight check for a manually made clay
model is as follows: a model with smoothly finished surfaces is placed in a. room
wit.h many paTanel fluorescent lamps attached to its ceiling and wans. Images of
these pa.rallellamps on the surfaces are observed to check where disorder occurs.
As precise simulat.ion of this process is very difficult and t.ime-consuming
even wit.h a very high-speed computer and the purpose is to check t.he qualit.y
of a. surface and to locate regions of low qualit.y, a similar but simplified method
may be applied. The assumption is that the observing eye is at infinity, the line
of sight is paranel and silhouett.e curves of the object surface are obtained, and
then changing the direction of the parallel lines of sight, we obtain the silhouette
patt.ern whose equation is given in Sect. 7.3. This pattern of the silhouett.e loci
is called t.he highlight. pattern in the factory. If a. real highlight pattern is to be
calculated , this is also possible, but with high cost of calculation. See Chap. 7.
Even wit.h t.his simplified assumption, t.he calculation is rat.her complicated
and the loci are very sensitive to the variation of the direction of the normal vector
on each point on t.he surface. Disorder of the silhouette pattern corresponds to
irregularity of t.he curva.ture distribution on the surface. On a boundary of e(l)
connection, t.he highlight patt.ern has break points and for e(O) connection it has
discontinuity of the highlight lines.
Another method of shape evaluat.ion is to observe the radius of curvature
Figure 17.5 Profile ,o f radius of curvature of section curves and highlight pattern of a surface
(courtesy of Toyota Motor Corp.)
profiles of curves on the surface or section curves of the surface. Irregularity of the
profiles indicates the existence of unfavorable regions on the surfa.ce. The radius
of curvat.ure distribution check can be performed in less time than the highlight
check, so it is used frequently during the design stage. This test for a curve is
rather severe and it corresponds well to the inspection test by the observing eyes
of proficient designers. However, the curvature profile check is not sufficient for
the surface quality evaluation. Figure 17.5 shows both the curvature profile and
the highlight distribut.ion. The curvature profile along the longitudinal sections
aTe deemed good, but the highlight pattern has wavy disturbances. The surface
construction must be modified. Figure 17.6 shows highlight lines of a model on
a display and Fig. 17.7 is a clay model made from the same data as in Fig. 17.6
and shows the real highlight test. Figure 17.8 is a drawing of a. complet.ed body
generated from an int.ernal model. Figure 17.9 is a photograph of the real car
developed from the int.ernal model.
Usually, graphic display with color shading is not so effective, because

t.he purpose of the shading display is to make pictures of good quality, but not
to show the qualit.y of the surface of the products. But recently, with greatly
increased computing power, it has become possible to evaluate viewing effects of
color paint.s on a car under various environmental light conditions. In order to
do so, accurat.e shape data of a geometric model and data of real light reflection
propert.ies of paints on metal sheets are supplied for ray tracing calculation.
Though it. takes rather a long time for the calculation, the evaluation of a color
simulat.ion of a car and that of color paints is possible long before the real model
is produced. Figure 17.10 is an example of a display of this kind.
Figure 17.6 CRT display of highlight pattern of a car (courtesy of Toyota Motor Corp.)
17.4.4 Die-Face Design System

A blank steel sheet is deformed into the final shape of a smooth outer panel
of a ca.r by the stamping press. A sectional view of a stamping die and stamping
action is schematically shown in Fig. 17.11. First a sheet of metal is placed on
the lower die and the periphery of the sheet is clamped and then the stamping
punch presses the sheet to deform it to the final shape. Refer to Fig.17.12. Since
manufacture of a product by press forming requires a number of processes such
as drawing (elongation of a steel plate by pressing), trimming, flanging, and
so on, and the forming is a complicated process, the designer must be able to
visualize all stages, from the blank steel sheet to the final product. Deformation
of the sheet is produced by pressing the stamping punch, but the quality of
deformed sheet is strongly affected by the shape of the holding surface and the
surrounding part called the 'step draw' around the shape of the product part
surface of the die. See Fig. 17.12. These two surfaces are called the die-face and
its determination in the stamping die design is critical for the production of a
car body surface of good quality.
A theoretical analysis of press deformation which is applicable to the real
problem is very difficult, because it has to treat a 3D elasto-plastic large defor-
mation problem, but with simplified FEM analysis the initial form of the steel
plate at its boundary holding state- is approximately determined to give the least
Figure 17.7 Highlight test of a clay model (courtesy of Toyota Motor Corp.)
Figure 17.8 Line drawing from an internal model( courtesy of Toyota Motor Corp.)
unnatural deformation by the initial holding.

Since a considerable amount of knowledge on the behavior of steel sheet
during stamping concerning the occurrence of fracture, surface distortion and
other imperfections has been accumulated over the years, the designer of a die-
face wit.h this knowledge can evaluate the quality of its shape appropriately if he
can visualize t.he deforming state of the sheet during pressing and calculate the
indices of it.s local deformation. Accordingly, Computer Aided Die-face Design
resembles Computer Aided Style Design in the format.ion of surfaces and the
Figure 17.9 Photog,aph of the real car of Fig. 17.8 (courtesy of Toyota Motor Corp.)
need to evaluate them, though the evaluation criteria are quite different and
the sources of input base data are different.. From the body style and structure
databases, the shape information of the final product is obtained which defines
the shape of the central region of the die. Around the central region, the die-face
has to be determined to guarantee to deform the steel sheet, which is set in the
cavity ohhe die, into the predefined shape by the stamping. Refer to Fig. 17.12.
The design procedures are:
(1) The optimum direction of pressing is decided.
(2} The surrounding shape consisting of step-draw parts to prevent fracture of
the sheet is designed.
(3) The clamping surface part to prevent wrinkles of the sheet during stamping
action is designed.
(4) These surfaces are evaluated by calculating the indices which relate to local
deformation of the sheet during the stamping action. If the criterion is not
satisfied, then return to (2).
(5) Draw beads are designed and the shape of the material sheet is determined.
The indices calculation in item (4) consist of that of geometric displacement
during the stamping punch movement, which is precisely obtained, and that of
material deformation, which is performed by using FEM on the approximated
shape. Figure 17 .13( a) shows the ini tial blank steel sheet sha.pe held by the holder
around the peripheral region, and (b) shows diagram of variation of drawing
depth during the press punching from the initial clamped position of the steel
sheet. The shape of the die-face affects this distribution, from which values of the
indices to estimate formability are obtained. Just like the style CAD subsystem,
the die-face CAD subsystem has the following capabilities:
- Smooth reception of data from the other subsystem to construct an internal
model for designing the die-face.
Figure 17.10 Shading display for evaluation of color paint on a car, (courtesy of Toyota
Motor Corp.)
...
Sheet meta l after forming
Forming
Figure 17.11 Sections of a stamping die (courtesy of Toyota Motor Corp.)

c.:>
.....
0>
Part shape Tipping

examination position Step draw
.....
..J
~f~ >
Body data ¢(~)c:::) 'S
base
"",,'
~
0'
::.
00
2,
...
ff
o-j
=-
"Q
(5'
00
5'
....
::.
0-
"~
.:<
Drawing
CAM Evaluation of press

system forming severity
Blank shape
Figure 17.12 Design processes of a die-face (courtesy of To yola Motor Corp.)

17.5 CAD ICAM of Free-Form Injection-Mold Products 317
Blank shape
on die face
Blank holder surface
(a) Blank shape in die cavity (b) Spreading behavior of

punch contact area
Figure 17.13 Evaluation of die-face design (courtesy of Toyota Motor Corp.)
- Interactive creation, alteration of die-face shapes.

- Generation of displays of various diagrams and charts for evaluating the forma-
bility of t.he steel sheet which is set in the die cavity.
Following the success of this system, real models made of plaster to investigate
the die-face shapes in the conventional methods have been eliminated and the
number of real t.ry-outs of stamping using the newly manufactured dies has de-
creased considerably. Manufacturing and checking measurements of the stamping
dies of car bodies and those of the fixtures for welding and inspecting are also
controlled by the computer using the information in the internal model which
are constructed in the previous stages. With this integration of CAD and CAM,
the productivity and quality of new car development have been enhanced greatly
in the company.
17.5 CAD/CAM of Free-Form Injection-Mold Products

In commercial products such as audio and video equipment, conventional design
procedures can easily produce simple shapes, but currently industrial designers
have to design aesthetically pleasing free-form shapes which have great commer-
cial appeal even in a highly competitive and technically well-developed market.
The conventional procedures for design and manufacture of aesthetically
pleasing products take a long time. Because mock-ups, molding dies and proto-
types of the products are made based on two-dimensional drawings, each of t.hese
steps results in a slightly different version of the designers' ideas. Accordingly,
negot.iations between the designers and makers of mock-ups and molding dies
take a long time to make them remodel or modify the objects until the designer
are satisfied. But since most current commercial CAD/CAM systems are more
concerned wit.h t.he productivity of products than their aesthetic aspects, they
are rarely helpful.
Sony's CAD/CAM system called FRESDAM is the one which is aimed at

a,e sthetic aspect.s of products. Compared to Toyota's CAD/CAM system for
car body development, this is very small, but. the concept is the same. From
designers' idea sketches to the final product.s, for instance, to dies for injection
molding, interactive design of free-form surfaces is supported wit.h an engineering
work st.ation. Since t.he number of pat.ches used is not large and each shape of the
patches used covers a relatively large area and its connection with it.s adjacent.
patches must be smoot.h, the system uses a kind of Bezier patch expression with
abundant cont.rol points such as
s(u, v) = (1- u + EuY(1- v + FV)3 pOO
The inner control points have the form:
_ u2 (1 - u).2pii + 2u(1 - u)v(l- V)qij + v 2(1 - v)2 p f)

Pij- {u(l-u)+v(l-v)J2 '
i,j = 1,2.
The control points qij are used for modifying the shape of a patch wit.hout affect-
ing its boundary conditions of C(1) connection. The control point.s pi) and pi)
are used for modifying the cross-boundary ta,ngents in u and v directions inde-
pendently. The a,bove expression is a modification of that explained in Sect. 14.4.
Tool path calculation for a ball-end mill is performed by intersection of the
offset surface of the original patches with t.he t.ool driving planes. Calculation for
collision detection and avoidance, compensation for heat shrinkage of the material
used, and allowance for the electrode gap in an electro-discha.rge ma,chine must
all be performed. And since separately produced molded parts of a, product must
be glued on their parting surfaces to become a complete product, det.erminat.ion
of the part.ing curves and parting surfaces becomes indispensable. The parting
curve on the surface is a space curve which has the same geometric characterist.ic
as a silhouette curve from a viewpoint at infinity.
The design and manufacturing process are shown in Fig. 17.14(a)-(f) for an
in-the-ear headphone with a large resonant. tube and a nice fit to an ear: (a)
shows the designer's final idea sketches with rendering, (b) is t.he defined free-
form shape, (c) is it.s shading display for evaluation, (d) is a photograph of
the produced model at five times actual size, (e) shows the electrodes of the
discharge machine, and (f) is a shading display for a case for the device, which
is also aesthetically designed.
Since this example is on a small scale, all the information is generated, stored
and t.ransferred without converting to engineering drawings. Evaluation of de-
signed objects is performed on CRT in a sha,ding display.
17.5 CAD/CAM of Free-Form Injection-Mold Products 319
(a)
(b)
Figure 17.14 Design and manufacturing of in-the-ear head-phone: (a) final idea sketch, (b)
model output, (c) shading display, (d) machined mod el for evaluation, (e) produced electrodes,
(f) display for its case design (courtesy 'of Sony Corp.)
(c)
(d)
Figure 17.14(contillued)
17.5 CAD ICAM of Free-Form Injection-Mold Products 321
(e)
(f)
FiglU'e 17.14 (continued)

Appendix Numerical Methods of
Differential Equation Solving
A.I Introduction
In this book, we have used the numerical methods of solving differential equations
for tracing curves on surfaces which show their special features or intersection
with other surfaces. Though some of these surfaces are represented by implicit
equations, we convert them into the corresponding differential equations to trace
the curves, because-it is easier to trace a curve with step-by-step solution of its
differential equation than to calculate a great many of the coordinate values of
discrete points on the curve by solving its nonlinea.r equations. Since usually these
curves are determined uniquely by given geometric conditions, their tracing by
the differential equations does not become unstable, so long as truncation errors
do not accumulate. In rare cases at singular points on the curves the validity
of the differential equations fails. In such cases, we have to take the omitted
higher-order terms of the differential equation into consideration.
There are several differential equation solvers available for our problems.
Among them, those methods equipped with adaptive stepsize control show good
results. This enables us not only to avoid the problem of determining the inte-
gration stepsizes, but also provides the optimum ones for tracing the curves. We
explain the methods we have used in order to give an understanding of their prin-
ciples and uses. We do not intend to introduce their detailed theoretical aspects,
which, if needed, should be consulted in those references treating numerical meth-
ods of computation [14][15][16]. Since the equations for the intersection curves
have been given, we can check the accuracy of the results obtained from the
integration process at any stage of the curve tracing. But we can keep expected
errors of the integration process within the given tolerance, hence we need not
perform extra checks of accuracy in the real space using the original equations.
The tracing curve passes through the points on the intersection curve, which are
obtained at the initial stage, and the loop is closed.
By our experience on problems treated in Chap. 7 and Chap. 16, the fourth-
order Runge-Kutta method with adaptive stepsize control is most reliable though
its speed is low. It is faster than the Runge-Kutta,-Fehlberg method, which uses
terms up to the fifth order, but the Bulirsch-Stoer method is good and relatively
fast. Predictor-corrector methods are fast, but frequently they fail in critical
regions. If such a method is equipped with a mechanism of variable stepsize
control, then together with the Runge-Kutta method it gives good results and
is the fastest, but not always reliable. The Bulirsch-Stoer method with rational
324 Appendix Numerical Methods of Differential Equation Solving
extrapolation works well, though that with polynomial extrapolation is a little

faster and less reliable.
We explain these methods in this chapter. For simplicity, we treat a differ-
ential equation with a single dependent variable, but the solving methods can
easily be extended to cases of multiple dependent variables.
Given a differential equation of the form
dy
dt = f(y, t) (A.I)
with an initial condit.ion of the value Yi at the start t = ti, our objective is to
deduce the value Yn at t = ti + H as a good estimate of y(ti + H) with a given
tolerance.
A.2 Adaptiye Runge-Kutta Method

A.2.1 Runge-Kutta Step
In t.his case, we set H = hand Yn = Yi+1 at t = ti+l = ti + h and we obtain Yi+l
by the following procedure:
at t = ti, t; + h/2, and t; + h , calculate k 1 , k2, k3 and k4 by the following formulas:
kl = hf(t;, Yi) }
k2 = hf(t; + h/2, Y; + kd2)
(A.2)
k3 = hf(t; + h/2, Y; + k2/2)
k4 = hf(t; + h, Yi + k3)'
Then t.he value Yi+l is given by
(A.3)
Four evaluat.ions of the right-hand side of the differential equation (A.I) are
required per step and the error term is O(h5). This formula gives good results
for most different.ial equations if a proper stepsize is chosen. Since it is difficult
t.o give a proper st.epsize in advance, we are liable to take a smaller stepsize.
This brings long computation time, but still the fixed stepsize does not always
guarantee precise results in critical cases. Accordingly, adapt.ive stepsize methods
are preferable.
A.2.2 Runge-Kutta with Quality Control

This is a stepsize doubling method. Let us integrat.e the same equation with
stepsizes 2h and h using the Runge-Kutta step a.nd call the results Ya and Yb.
Then we can assume that two equations:
y(t + 2h) = Ya + (2h)5¢ + O(h6) }

(A.4)
y(t + 2h) .= Yb + 2h 5¢ + O(h6)
A.2 Adaptive Runge-Kutta Method 325
hold for the correct value y( t + 2h), where 4> is considered a constant independent
of the stepsize adopted so long as h is small. From the above t.wo, we know that
t.he difference I::::. = Yb - Ya of the two results is proportional to hS :
(A.5)
We use t.his relat.ion for adjusting the stepsize of the Runge-Kutta integration.
Let 1::::.0 denot.e the desired accuracy wit.h the st.epsize ho and 1::::.1 be a value
obt.ained wit.h a stepsize hI for the same equation. There is a relation which is
obtained from eq.(A.5):
(A.6)
This can be used to adjust hI for the next try of the integration: if 1::::.1 is larger
than 1::::.0 in magnitude in the first try, the value of the right-hand side indicates
how much the stepsrze should be decreased in the second try of the same interval,
and if it is smaller than 1::::. 0, the right-hand side value shows the degree of increase
of the stepsize in the next section.
The magnitude of 1::::.0 for the desired accuracy depends of course on the
problems we treat and the maximum magnitude of Y and h. Therefore, for
safety, wit.h introduction of a safety factor S(;::;; 0.95), eq.{A.6} is changed to a
slightly different crit.erion:
In the examples used in this book, value of 1::::.0 was taken as 10-6 rv 10-8
or E = 1::::. 0/ h = 10- 4 rv 10- 6 , and 1::::.1 was the maximum one chosen among the
differences I::::. for several dependent variables.
A.2.3 Runge-Kutta-Fehlberg Method

Both the fourth-order and the fift.h-order Runge-Kutta formulas are applied to
approximate Y in the given interval of t. The number of subintervals is increased
step-by-step until the fractional difference between the results of the fourth-
order and the fifth-order formulas (which give a truncation error h S and h6 ,
respecti vely) in the same interval is less than the specified tolerance. This method
gives good results, but compared to the Runge-Kutta method with adaptive
stepsize control it is slow and its accuracy seems the same order as that of the
adaptive Runge-Kutta method for our problems.
The procedure is given below. Att =ti, t,+h/4, t,+3h/8, ti+12h/13, t,+h
and ti + h / 2, ca.lculate k 1 , k2 , k 3 , k 4 , ks and k6 by the following formulas:
kl = hf(t" y,), .
k2 = hf(ti + h/4, Yi + kl/4),

k3 hf(ti + 3h/8, Yi + 3kl/32 + 9k2/32),
k4 hf(ti + 12h/13, Yi + 1932kl/2197 -
7200k 2/2197 + 7296k3/2197),
ks hf((ti + h, Yi + 429kl/216 - 8k2 +
3680k3/513 - 845k 4/4104),
k6 hf(ti + h/2, Yi - 8kd27 + 2k2 -
3544k3/2565 + 1859k 4 - llk s /40).
Then calculate r for a.n error est.imate,
(A.7)
if r ::; t, the value Yi+1 is given by
(A.8)
For the next stepsize h', calculate the following criterion:
q = 0.84( ~ )1/4. (A.9)

r
If q ::; 0.1, then h' = O.lh, if q 2:: 4, then h' = 4h, otherwise h' = qh.
A.3 Variable Stepsize Predictor-Corrector Method

Start.ing from (ti-3, Yi-3) for the next three points, the Runge-Kutta method with
a. stepsize h is used to obtain (Yi-2, Yi-1, y;). After these points, the following
explicit Adams-Bashforth formula for determining a predicted next value Yi+1:
Yi+1 = Yi + (h/24){55f(ti, Yi) - 59f(ti-1, Yi-1) + 37f(ti-2, Yi-2)

- 9f(ti-3, Yi-3)}
(A.10)
is applied, then t.he following implicit Adams-Moulton formula for its corrected
value:
Yi+1 = Yi + (h/24){9f(ti+ll Yi+1) + 19f(ti, Yi) - 5f(ti-ll Yi-1) + f(t i-2, Yi-2)}
(A.11)
is used for a more precise Yi+1' This method is fast compared to the Runge-
Kutta method and the accuracy is the same. Accordingly if it is reasonably
certain that the differential equation is well-behaved, this is one of the most
desirable methods. But when a curve to be traced has regions of great variation
of curvature and torsion, the tracing often fails.
Accordingly, if the method with adaptive stepsize control is introduced, the
situation is greatly improved, but since the predictor-corrector method always
A.4 Bulirsch-Stoer Method 327
requires four starting points, the change of stepsize must be kept 'minimal. As a
consequence, we use it only when the estimated truncation error
_ Iy(ti+l) - Yi+ll '" I . _- I~ (A.12)

e- h '" Y.+l Yi+l 270h
becomes smaller than (E/10) or greater than E, that is,
(E / 10) > e or e > E.

The stepsize h is changed to qh, where
q = 1.5( hE _ )1/4. (A.13)

IYi+l - Yi+ll
If q > 4 or q < 0.1, set q = 4 or q = 0.1 to prevent abrupt change of the interval.
If h becomes smaller than a. predefined minimum value hm' switch the inte-
gra.tion process to the Runge-Kutta method with stepsize hm' proceed for several
steps and then return to the predictor-corrector method.
A.4 Bulirsch-Stoer Method

A.4.1 Principle of the Method
This usually gives good results with automatic adjustment of stepsizes and its
speed is relatively high. After many trials of various differential equation solvers,
we have relied on this method for application in geometric computation.
For the integration procedure of this solver, the modified midpoint method is
applied, whose error is a funct.ion of even powers of the stepsize h. Since in this
method integration is performed several times for the same interval (t, t + H)
with different stepsizes, we use an index i to indicate its integration number.
Hence the symbol ti used in the previous sections is abandoned hereafter. Let
the correct value of Y at t + H be y(t + H) and the obtained value there be Yn
after n = H / h steps starting from Yo = y(t). The error at t = t + H is given by
Yn - y(t + H) = L: cxj h2) , (A.14)

)=1
where cx) are coefficients independent of the stepsize h. Let WI, W2, W3' •• be
values of the int.egration at t = t + H obtained by the modified midpoint met.hod
with t.he st.epsizes hI, h2' h3 ••• (hI> h2 > h3 •• .). We consider these values
wI , W2, W3 ••• to be a function of the square of the stepsize: let this function pass
through the points (hi, wd, (h~, W2), (h~, W3) • ". The estimate of y(ti + H) is
considered as the ext.rapolated value of this function at stepsize zero.
This is the principle of extrapolation methods of differential equation solvers.
The Bulirsch-Stoer met.hod adopts a rational extrapolation, for it gives usually
better results than a polynomial extrapolation does .
A.4.2 Outline of the Procedures
Next. we explain the procedures of the method. We use the st.epsizes of decreasing
order in a section between t and t + H. The section is divided into eleven sets
of subsections whose numbers are (2,4,6,8,12,16,24,32,48,64,96), each one of
which is indicat.ed by the index k.
1. For each st.epsize hi, we calculate the integration value Wi and produce a
current extrapolated estimate y at h = 0 using the current and previously
obtained values Wi'S.
2. If the difference between the current estimate and the last obtained estimate
becomes negligible, further decrease of t.he substeps is unnecessary. When
t.he est.imates do not converge even at the eleventh st.epsize, the length H of
this sect.ion is made a quarter and we st.art again from the beginning.
3. When the estimat.e converges at an int.ermediate step, the length of the next
section should be adjust.ed to keep t.he index value for subsections nearly
constant. Let. t.he desirable index of subsections be kd and that of the current
stepsize be kc' If kc = kd - 1, the next section length is extended by 1.2
t.imes and if kc = kp, it is shortened by 0.95 times.
4. Otherwise it is given by the formula H x (number of subsections of the index
kd -l)/(number of current subsections). The index kd = 7 is said to be good.
A.4.3 Integration Procedure
In' t.he Bulirsch-Stoer method, the modified midpoint method is used for the
integration procedure, because its error of integration is expressed by a function
of the square of the stepsize h and the calculation is simple. The integration
procedure between t and t + H is given by:
h
= H/n, )
~;t~ hf(t, zo),
Zo
Zl = (A.15)
Zi+l = Z.-l + 2hf(t + ih, z.) for i = 1 ... n - 1,
y(t + H) ;::::: Yn = (1/2){zn + Zn-l + hf(t + H, zn)}·
There are n+ 1 evaluations of f(t, y) in t.he interval H which comprises n sub-

divisions, whereas the Runge-Kutta met.hod requires 4n evaluations. In Bulirsch-
Stoer, the number of evaluations is at most 73 compared to 96 in Runge-Kutt.a
for the same interval H. But since in Bulirsch-Stoer H is adjust.ed automatically
t.o be t.aken much larger, it is more suitable for fast tracing of curves. When H
is t.oo large for dra.wing a curve, the int.ermediate points can be obt.ained from
(A.16)
AA Bulirsch-Stoer Method 329
A.4.4 Polynomial Extrapolation

Polynomial and rational interpolations and extrapolations are treated in Sect. 8.2.
Our object here is to estimate a value Y at x = 0 from a given sequence of points
(Xl, yd, (X2 ' Y2),'" (Xn' Yn), {Xl> X2 > ... > Xn > O}, by extrapolation, where
terms Xi represent. the square of the sub-interval in the integration processes and
terms Yi the corresponding result values of the integration. Use of polynomial
extrapolation does not seem to provide good results because of its oscillatory
behavior, but. usually it gives good results in non-critical cases. The reason may
be an extrapolat.ion of a very short distance. Calculation methods of obtaining
extrapolated values by polynomial or rational functions are almost the same,
of course, the polynomial calculation is faster. We first explain the method of
polynomial functions, because it is easy to understand and can be extended t.o
the case of the rational functions.
Let P;(i+1) .. .(i+m)' represent the extrapolated value of Y at X = 0 from a given
sequence of m + 1 points (Xi, Yi), (Xi+l> Yi+1),'" (Xi + m, Yi + m) and let it be
called the value on level m . We want to obtain P12 ...n for the est.imate of Y at
X = 0, but. at the same time we want to make n as small as possible wit.hin a
given tolerance of the estimated value y.
We assume that from YI, .•. Yn-l, an estimated value Y = P1...n-1 has already
been determined and now a new point Yn at Xn has been obt.ained. Since we can
make two estimates P 23 ...n from the sequence of Y values (Y2' " Yn), and P12. .. n
from (YI .. , Yn), if their difference is smaller than the tolerance, we take P I2 ... n
as the estimate of Y at X = 0, otherwise we proceed to X n +1 for the next finer
sub-interval and obtain the integration result Yn+l. Then make P 23 ... n ,(n+1) and
P12.~,(n+1) ' and t.est if it is final.
To make the above stated process systematic, at first we define the differences
between Pi(i+1) ... (i+m) and its lower neighboring ones:
Ci,m Pi ... (i+m) - Pi .. .(i+m-l), }

Di,m Pi ... (i+m) - P(i+1) ... (i+m) , (A.17)
Ci,o P; = Yi, Di,o = Pi = Yi ·
We make two t.riangular tables whose entries contain Ci,m or D i,m' The first
index i indicates t.he i-th row of the tables of the difference and the second index
m the m-t.h column of the tables. According to the following formulas:
Xi( Ci+l,m - Di ,m) }

(X i - Xi+m+l) ,
(A .18)
Xi+m+I(Ci+1,m - Di,m) ,
Di,m+l =
(Xi - Xi+m+1)
entries of the tables are filled step by step when a new value Yi for Xi is added
in the entries (i, 0) of both tables.
The above tables are constructed a.n d the intermediate estimates are determined
in the following way.
Table A.1
YI = PI CI,I C I ,2 C I ,3 CI,4
Y2 = P2 C 2,1 C2,2 C2,3
Y3 = P3 C 3,1 C3,2 (A .19)
Y4 = P4 C4,l
Ys = Ps
Table A.2
YI = PI Dl,l D1,2 D I,3 D I,4

Y2 = P2 D2,1 D2,2 D2,3
Y3 = P3 D3,1 D3,2 (A.20)
Y4 = P4 D4,1
Ys = Ps
1. Let PI (= yd already have been obtained. Then P2( = Y2) is added. DI,I and
Cl,l are calculated and stored, but Cl,l is not used. They are given from
equations (A.17) and (A.18) by
= p p _ XI P2 - PI) _ XI(C2,O - DI,o) }

12 - I - Xl - X2) - (XI - X2) ,
(A.2I)
p _ p _ X2(P2 - PI) _ X2(C2,O - DI,o)
12 2- (XI- X2) - (XI- X2)
An estimated value at level two is obtained from
(A.22)
2. P3 is added. The level-one values C2 ,1 and D 2 ,1 are calculated and st.ored

using formulas similar to the above. From D I ,l and C2 ,1, new entries C I ,2
and D1,2 of level t.wo a.re calculat.ed by using (A.18) and stored,
(A.23)
An est.imated value at level t.hree is P123 = P23 +D I ,2. Since P23 = P3+D2,1,
we have
(A.24)
AA Bulirsch-Stoer Method 331
3. P4 is added. C3 ,l and D3 ,l are calculated and stored. From D 2 ,l and C3 ,l,

the level-two values C2 ,2 and D 2 ,2 are obtained. From D 1 ,2 and C 2 ,2 the final
ent.ries C 1,3 and D 1,3 of level t.hree are filled. We obt.ain
(A.25)
4. P 5is added. C 4 ,l and D 4 ,l are obtained. From D 3 ,l and C 4 ,l, C 3 ,2 and D 3 ,2

are det.ermined. From D 2 ,2 and C3 ,2, C 2 ,3 and D 2 ,3 are obtained. From C 1 ,3
and D2 ,3, t.he final entries C 1 ,4 and D 1 ,4 are filled.
(A.26)
Generally the estimate of level n is given by
P12'" n = Pn + Dn-1,1 + D n - 2,2 + ... + D1,n-l. (A.27)
The last term of t.he estimate of level n is considered the last correction
value for its final estimate. If D1,n-l is smaller t.han a. given tolerance, the
est.imate value of y is given by P12 . .. n.
A.4.5 Rational Extrapolation
The Bulirsch-St.oer algorit.hm uses a rational function for t.he interpolation in-
stead of the polynomial one .
. Let Ri(i+ll ..(i+m), which is t.he rational function equivalent of Pi(i+l)".(i+m), be
the value obt.ained by extrapolation at x = O. We want to find it by constructing
two triangular tables similar to Tables A.I and Tables A.2 of the polynomial case.
Let us define the differences of adja.cent rational functions similarly to eq.(A.17):
Ci,m = Ri".(i+m) - Ri".(i+m-l), }

Di,m = R;" .(i+m) - R(i+l).,,(i+m), (A.28)
Ci,o = Ri = Yi, Di,o = Ri = Yi'
Instead of eq.{A.I8), we have the recurrence formulas for Ci,m and Di,m:
(A.29)
Then we can const.ruct tables similar to Tables A.I and A.2 each time a new pair
(Xi, y;) is obt.ained. Similarly to eq.{A.27), the estimate R 12 .. .n of the level n is
given by
R 12 ".n = Rn + Dn-1 ,1 + D n- 2 ,2 + .,. + D1,n-l ' (A.30)
A.5 Examples and Evaluation

Various figures of curves shown in Chapters 7 and 16 are results obtained from
different.ial equation solvers explained in this Appendix. We add t.wo examples
for quantit.ative evaluat.ion of some of the methods.
Example 1. Contour curves of a very flat surface which is synt.hesized from
3 x 3 pat.ches and has a flat. line segment cont.acting with a horizontal plane
are shown in Fig. A.1(b). The lengt.h of the cont.act line segment is 0.5. Its
contour curves were produced by intersection with the horizontal cutting planes
whose heights are 2 x 10- 8 , 2 X 10- 7 , 2 X 10- 6 , ' ··2 X 10- 2 , 2 X 10- 1 . They
are shown in Fig. A.1(a). The differential equation method was used for tracing
contour curves. The Runge-I<utta method with adaptive stepsize control gave
good results: a cont.our curve whose height is as sma.ll as 2 x 10- 8 was obtained.
With a value smaller t.han this, t.he procedure did not work with double precision
calculat.ion.
Example 2. Cont.our curves of a surface which has a subtle flat part are shown
in Fig. A.2, where the global shape is shown by the contour curves with each
height step 0.01 of the cutting plane. In Fig. A.3, the height st.ep is 0.0002 and
its shape around the peak is shown, whereas in Fig. AA it is 0.00001, in which
clea.rly a. local hump and two st.raight lines can be observed.
In tracing the curves near the critical points, a predictor-corrector method
fails in the early stage, and so does Runge-Kutta without adaptive st.epsize. The
Bulirsch-Stoer met.hod with the polynomial extrapolation fails in the last critical
stage. Bulirsch-Stoer with rational ext.rapolation is very fast. in the non-critical
region, but. becomes very slow at the critical region and past it, while Runge-
I< utt.a with adaptive st.epsize is not fast, but at the critical region it is faster than
Bulirsch-Stoer with rat.ional extrapolation and successful. Runge-I<ut.ta-Fehlberg
is slow hut succeeds.
Figure A.5 shows variation of the st.epsize in the Bulirsch-Stoer method,
though there occur very large st.epsizes, the calculated point.s are accurate. The
smooth curves are drawn by extracting the intermediate integration values given
by eq.(A.16) in t.he Bulirsch-Stoer method. We can recognize in these examples
that. t.he t.runcat.ion errors are very small. Ot.herwise, we cannot discriminate the
int.ersection curves wit.h very small height. differences.
(a)
Boundary of
3 x 3 patch
Control point
(b)
Figure A.1 Contour curves at very small heights obtained by Runge-Kutta method with
quality control: (a) contour curves, (b) locations of control points of connected patches
Extremum sea.rch
Figure A.2 Global shape with contour curves and extremum search curves
Figure A.3 Magnified figure around

the critical region. Contour curves are
shown
Figm'e A.4 More magnified ngure of

the contour curves
One calculating
Figure A.5 Example of automatic

stepsize adjustment by Bulirsch-Stoer
method
Bibliography
Since a comprehensive bibliography in the field of Geometric modelling and

CAD/CAM is to be included in Kimura's book (see Preface), only literature
and references related to the contents of this book are shown separately for each
part. An asterisk indicates the papers written in Japanese.
Abbreviat.ions for Japanese journals and academic societ.ies cited in this book
are as follows:
- IPS Japan: the Information Processing Societ.y of Ja.pan. Its publications are:
- J. IPS Japan: ' Journal of the Informat.ion Processing Society of Japan
(in Japanese).
- Trans. IPS Japan: Transactions of the Information Processing Society of
Ja.pan (in Japanese).
- J. InL Process.: Journal of Informat.ion Processing (in English)
- Informat.ion Processing in Japan. Annual publication before 1978 (in English).
- Proc. -th Annual Convention IPS Japan (in Japanese).
- IPSJ SIG report.s (in Japanese).
- Proc. IPSJ SIG Symposium (in Japanese).
Journals of ot.her academic societies.
- J. ICSJ: Journal of t.he Societ.y of Instrument and Control Engineers of Japan

(in Japanese).
- J.JSPE: Journal of the Japan Society of Precision Engineering (in Japa.nese).
- J.JSME: Journal of the Japan Society of Mechanical Engineers (in Japanese).
- Trans.JSME: Transadions of the JSME (in Japanese).
Preface
[1] Farin, G.: Curves and surfaces for computer aided geometric design: a
practical guide. 2nd ed. San Diego CA: Academic Press 1990
[2] Hoffmann, C.M.: Geometric and solid modeling. San Mateo CA: Morgan
Kaufmann 1989
[3] Hosaka, M.: MARS-l. Seat Reservation System of Japanese National
Railways. In: History Committee of IPS Japan (ed.): *History of Early
Japanese Computers. Tokyo: Ohmsha 1985, pp. 155-172
[4] Hosaka, M.: *Digit.al differential analyzer. Hit.achi Hyoron (J. of R&D of
Hitachi Ltd.) 45(10): 61-65, 1963
Hosaka, M.: *Generation of figures by DDA. J.ICSJ 3(7): 422- 452, 1965
Hosaka., M.: *On generation, storage and processing of graphics. J .IPS
338 Bi bliogr aphy
Japan 6(3): 129-139, 1965. English abstract: Information Processing in

Japan 6: 29-40, 1967
[5] Sutherland, I.E.: Sketchpad - A man machine graphical communication
system. In: Proc. SJCC. 1963, pp. 329-346
Coons, S.: An Outline of the requirements for the computer aided design .
In: Proc. SJCC: 1963, pp.299- 304
[6] Hosaka, M.: *Man-machine interaction and computer display. Monthly
Memoranda of IPS Japan 22, Feb. 1967
Hosaka, M.: *Design methods by computer assistance. J.JSME
71(590): 118-127, 1967
[7] Hosaka, M.: *Theory and design of free-form surface. J.lPS Japan 8(2):
65-72, 1967. English abstract: Information Processing in Japan 7: 54-61,
1968
Hosaka, M.:Design of free-form surface and computer. J.JSPE 34(1): 4-10,
1968 .
Hosaka, M.: Theory of curve and surface synthesis and their smooth fitting.
J.lPS 10(3): 121-131, 1969 (English abstract, Information Processing in
Japan 9: 60-68, 1970)
[8] Ishimatu, Y. et al.: Computer-aided generation of sculptured surfaces.
J. Numerical Control: 33- 39, April, 1974
[9] Matsushita, T ., Hosaka, M.: *Shape processing by a new interactive
language GIL. In: Proc. 11th annual convention IPS Japan: 65-66, 1970
Kimura, F., Hosaka, M.: *On processing of solid body. In: Proc.13th annual
convention IPS Japan: 271-272, 1972
. Kimura, F., Hosaka, M.: *Extension of interactive language GIL. In: Proc
14th annual convention IPS Japan: 275-276, 1973
Kimura, F., Hosaka, M.: *Input-output control of GIL. In: Proc.14th an-
nual convention IPS Japan: 277-278, 1973
[10] Braid, I. Lang, C. A.: Computer-aided design of mechanical components
with volume building blocks. In: Proc. Prolamat '73, Amsterdam:
North- Holland 1973
Engli, M. E.: Language for 3D graphics application. In: A. Gunther et
al.(eds.) : Proc. European Computing symp., Amsterdam: North-Holland ,
1973
[11] Hosaka, M., Kimura, F. et al.: A unified method for processing polyhedra.
In: Rosenfeld, J .R. (ed.): Proc.lFIP Congress '74, Amsterdam: North-
Holland 1974, pp.768-772
Hosaka, M. , Kimura, F . et al.: GIL- A Language for Interactive Graphics,
In: Proc.2nd US-Japan computer conference: Montvale NJ: AFIPS Press
1975, pp.546-550
Hosaka, M., Kimura, F., Kakishita, N.: A software system for computer-
aided activities. In: J.J. Allan (ed.): CAD Systems. Proc.IFIP working
conference: Amsterdam: North-Holland 1977, pp.169-198
Bibliography 339
[12] Hosaka, M.: *Computer Graphics: Tokyo: Sangyo-Tosho Jan. 1974, Many
references for Hosaka's research work during 1963-1972 are included
[13] Hosaka, M., Kimura, F.: An interactive geometric design system with
handwriting input. In: Gilchrist , B.(ed.): Proc.IFIP Congress '77: Ams-
terdam: North-Holland 1977, pp. 167-172
[14] Kimura, F.: Geomap-III: Desining solids with free-from surfaces. IEEE
Computer Graphics and Applications 4: 58-72, 1984
[15] Hosaka, M., Kuroda, M.: *Generation of curves and surfaces in CAD, J. IPS
Japan 17(12): 1120-1127, 1976. English abstra.ct, Information Processing
in Japan (17): 75- 79, 1976
[16] Hosaka, M., Kimura, F.: Synthesis method of curves and surfaces in
interactive CAD. In: Proc. Intn'l conference on interactive technology (IEEE
78 CH 289-8C) 1978, pp. 151- 156
[17] Hosaka, M., Kimura, F.: A mathematical theory and design methodolgy for
free-form shapes. J. IPS Japan 13(3): 1-12, 1980
[18] Ohara, M.: CAD/CAM at Toyota Motor Corporation. In: Kitagawa, T.
(ed): JARECT, Computer science and technologies 18, Tokyo: Ohmsha and
North-Holland 1988, pp.191-210
[19] Hosaka, M., Kimura, F.: Interactive input methods for free-form shape
design. In: Sata, T ., Warman, E. (eds.): Proe. IFIP working confer-
ence on man-machine communication, Amsterdam: North-Holland 1981,
pp.103-115
[20] Hosaka, M.: *Theory of shape connection. In: Proe. symp. Graphics and
CAD '84 IPS Japan 1984, pp.51- 62
Hosaka, M.: *Intersction and connection of surfaces 1. In: Proc. symp.
Graphics and CAD '87 IPS Japan 1987, pp.57-66
Hosaka, M.: *Intersection and connection of surfaces 2. In: Proc. symp.
Graphics and CAD '88 IPS Japan 1988, pp. 121-132
Chapters 1-4
[1] Hosaka, M., Kimura, F.: A unified method of processing polyhedra. In:
Rosenfeld, J.R. (ed.): Information Processing'7 4 Proc. IFIP Congress'7 4,
Amsterdam: North-Holland 1974, pp.768-772
[2] Hosaka, M., Kimura, F.: An interactive geometric design system with hand-
writing input. In: Gilchrist, B. (ed.): Information Processing '77 Proe. IFIP
Congress '77, Amsterdam: North-Holland 1977, pp.167-172
[3] Levin, J.: A parametric algorithm for drawing pictures of solid objects of
quadric surfaces. Comm. ACM (10): 555-563, 1976
[4] Levin, J.: Mathematical models for detecting the intersection of quadric
surfaces. Computer Graphics and Image Processing 11: 73-87, 1979
[5] Sarraga, R.F.: Algebraic methods for intersections of quadric surfaces in
GMSOLID. Computer Vision, Graphics and Image Processing 22: 222-328,
1983
340 Bi bliograpllY
[6] Mantyla, M.: An introduction to Solid Modeling. Rockville >MD: Computer

Science Press Inc. 1988
[7] Miller, J .R.: Analysis of quadric-surface based solid models. IEEE Com-
puter Graphics and Applications 1: 28-41, 1988
[8] Saitoh, T.: *Interference and display of quadrics. In: Proc. JSPE '90 JSIP
1990, pp.l037-1038
Chapters 5-8
[1] Coxeter, H.S.M.: Introduction to geometry. 2nd edition, John Wiley 1969
[2] Hosaka, M.: *Theory of curve and surface synthesis and their smooth fitting.
J. IPS Japan 10(3): 121-131, 1969 (English abstract, Information Processing
in Japan 9: 60-68,1969)
[3] Enomoto, H. et al.: *Computer experiment on global properties of structure
lines of images using graphic display and its consideration). J. IPS Japan
17(7): 641-649, 1976
[4] Kajiya, J.T.: Ray tracing parametric patches. Computer Graphics
(Proc. Siggraph '82) 16(3): 224-2.54, 1984
[.5] Sederberg, T.W., Anderson, D.C., Goldman, R.N.: Implicit representation
of pa.r ametric curves and surfaces. Computer Vision, Graphics and Image
Processing 28: 72-84, 1984
[6] Goldman, R.N., Sederberg, T.W., Anderson, D.C.: Vector elimination:
a. technique for implicitization, inversion and intersection of plana.r
parametric rational polynomial curves. Computer Aided Geometric Design
1(4): 327-3.56, 1984
[7] Sederberg, T.W.: Planar piecewise algebraic curves. Computer Aided
Geometric Design 1(3): 241-2.55, 1984
[8] Poeschle, R.: Detecting surface irregularities using isophotes. Computer
Aided Geometric Design 1(2): 163-168, 1984
[9] Tiller, W., Hanson, E.G.: Offsets of two-dimensional profiles. IEEE
Computer Graphics and Applications 9: 36-46, 1984
[10] Satterfield, S.D., Rogers, D.F.: A procedure for generating contour lines
from a B spline surface. IEEE Computer Graphics and Applications
4: 71-75, 1985
[11] Fa.rouki, R., Rajan, V.: Exact offset procedures for simple solids. Computer
[12] Beck, J.M., Farouki, R.T., Hind, J.K.: Surface analysis methods. IEEE
Computer Graphics and Applications 6(12): 18-36, 1986
[13] Klok, F.: Two moving coordinate frames for sweeping along 3D trajectory.
Computer Aided Geometric Design 3(3): 217-229, 1986
[14] Farouki, R.: The approximation of non-degenerate offset surface. Computer
[1.5] Rossignac, J.R. , Requicha, A.A.G.: Offsetting operations in solid modelling.
Bibliography 341
[16] Stoer, J., Bulirsch, R.: Introduction to numerical analysis. New York:
Springer-Verlag 1980
[17] Press, W. H., Flannery, B.P., Teukolsky, S.A., Vetterling W.T.: Numerical
Recipies in C Cambridge: Cambridge University Press 1988
[18] Higashi, M., Kushimoto, T., Hosaka, M.: On formulation and display
for visualizing features and evaluating quality of free-form surfaces. In:
Vandoni, C.E., Duce, D.A.(eds): Proc. Eurographics '90 1990, pp.299-309
[19] Love, A.E.H.: Treatise on the mathematical theory of elasticity. Cambridge:
Cambridge University Press 1934, pp.401-410
Chapters 9-12
[1] Hosaka, M.: *Theory and design of free-form surface. J. IPS Japan 8(2):
65-72, 1967
[2] Hosaka, M.: · Synthesis and smoothing of free-form curves and surfaces.
J. IPS Japan 10(3): 121-131, 1969
[3] Forrest, A.: Interactive interpolation and approximation by Bezier
polynomials. Computer J. 15: 71-79, 1972
[4] Riesenfeld, R.F.: Non-uniform spline curves. In: Proc. 2nd USA-Japan com-
puter conference IPS Japan 1975, pp. 551-555
[5] Hosaka, M., Kuroda, M.: *Generation of curves and surfaces in CAD. J. IPS
Japan 17(12): 1120-1127, 1976. (English abstract, Information Processing
in Japan (17): 75-79, 1976)
[6] Hosaka, M., Kimura, F.: Synthesis methods of curves and surfaces in
interactive CAD. In: Proc. intn'l conference on interactive techniques in
CAD, IEEE 1978, pp. 151-156
[7] Faux, I.D., Pratt, M.J.: Computational geometry for design and
manufacture. London: Ellis Horwood 1979
[8] Hosaka, M., Kimura, F.: A theory and methods for three dimensional
free-form shape construction. J. Inf. Process 3(3): 140-151, 1980
[9] Hosaka, M., Kimura, F.: Interactive input methods for free-form shape
design. In: Sata,T., Warman, E. (eds.) Proc.IFIP working conference on
man-machine communication. IFIP 1980, pp.l03-115
[10] Boehm, W.: Generating the Bezier points of B-splines. Computer Aided
Design 13(6): 365-366, 1981
[11] Barsky B.A., Beatty, J.C.: Local control of bias and tension in Beta-splines.
Proc. Siggraph '83 17(3): 193-218, 1983
[12] Hosaka, M.: Theory of shape connection. In: Proc. symp. Graphic and
CAD '84 IPS Japan 1984, pp.51-62
[13] Boehm, W., Farin, G., Kahmann, J.: A survey of curve and surface in
computer aided geometric design. Computer Aided Geometric Design 1( 1):
1-60, 1984
342 Bibliography
[14] Varady, T., Pratt, M.J.: Design techniques for the definition' of solid objects
with free-form geometry. Computer Aided Geometric Design 1(3): 207-225,
1984
[15] Boehm, W.: Curvature continuous curves and surfaces. Computer Aided
Geometric Design 2(2): 313- 323, 1985
[16] Piegl, L.: The sphere as a rational Bezier surface. Computer Aided
Geometric Design 3(1): 45-52, 1986
[17] Piegl, L.: Infinite control points - a method for representing surface of
revolution using boundary data. IEEE Comput.er Graphics and Applications
3: 45-55, 1987
[18] Piegl, L.: Interactive data interpolat.ion by rational Bezier curves, IEEE
Comput.er Graphics and Applications 4: 45-53, 1987
[19] Sa.rraga, R.F.: G 1 interpolation of generally unrestricted cubic Bezier curves.
Computer Aided Geometric Design 4(1-2) : 23-40, 1987
[20] Boehm, W.: Rational geometric splines. Computer Aided Geometric Design
4(1-2) : 67-78, 1987
[21] Farin, G., Rein, G. et 31.: Fairing cubic B-spline curves. Computer Aided
Geometric Design 4(1-2): 91-104, 1987
[22] Bartels, R. H., Beatty, J .C, Barsky, B.A.: An introduction to splines for
use in computer graphics and geometric modeling. San Mateo CA: Morgan
Kaufman 1987
[23] Farouki , R., Rajan, V.: On the numerical condition of polynomials in
Berstein form. Computer Aided Geometric Design 4(3): 191-216, 1987
[24] Hagen, H., Schultz, G.: Automatic smoothing wit.h geometric surface
. patches. Computer Aided Geometric Design 4(3): 231- 236, 1987
[25] Meier, H., Nowacki, H.: Interpolating curves with gradual change in
curvature. Computer Aided Geometric Design 4(4): 297-306,1987
[26] Farouki, R., Rajan, V.: Algorit.hm for polynomials in Berstein form.
[27] Hosaka, M.: *Intersection and connection of surfaces-2. In: Proc. Symp.
graphics and CAD IPS Japan 1988, pp.121-132
[28] Higashi, M., Kaneko, K., Hosaka, M.: Generation of high-quality curve and
surface with smoot.hly varying curvature. In: Duce, D.A., Jancene, P. (eds.):
Proc. Eurographics '88 Amsterdam: North-Holland, 1988, pp.79-92
[29] Fergson, D.R., et al.: Surface shape cont.rol using constrained optimization
on the B-spline representation. Computer Aided Geometric Design 5, 1989
[30] Piegl, L.: Modifying the shape of rational B-splines. 1: curves. Computer
Aided Design 21(8): 509-518, 1989
[31] Piegl, L. : Modifying the shape of rational B-splines. 2: surfaces. Computer
Aided Design 21(9): 538-545, 1889
[32] Dokken, T. et a1.: The role of NURBS in geometric modeling and
CAD/CAM . In: Krause, F-L., Jansen, H. (eds.): Advanced geometric
modeling in engineering application, Berlin: APZ 1989, pp.95-102
Bibliography 343
L33] Farin, G.: Curves and surfaces for computer aided geometric design: a
practical guide. 2nd ed. San Diego CA: Academic Press 1990
[34] Saitoh, T., Hosaka, M.: *On the extended rational Bezier curve and its use
for curve fitting methods. Trans.lPS Japan 31(1): 33-41, 1990
[35] Saitoh, T. Hosaka, M.: *High quality outline font by the extended rational
Bezier curve. Trans. IPS Japan 31{ 4): 562-570, 1990
Chapters 13-15
[1] Coons S.A.: Surfaces for computer aided design of space forms. MIT Project
MAC TR-41 1967
[2] Gregory J .A.: Smooth interpolation without twist constaints. In: Burnhill,
R.E., Riesenfeld, R.F.{eds.): Computer aided geometrric design, New York:
Academic Press 1974, pp.71-87
[3] Sabin, M.A.: "The use of piecewise forms for the numerical representation of
shape. Computer and Automation Inst. Hungarian Academy of Sci. 1977
[4] Barnhill, R.E., Brown, J.H., Klucewicz I.M.: A new twist in computer aided
geometric design. Computer Graphics and Image Processing 8: 78-91, 1978
[5] Higashi, M. et. al.: An interactive CAD systems for construction of shapes
with high quality surface. In: Warman, E.A. fed.): Proc. CAPE '83 IFIP
1983, pp.371- 390
[6] Chiyokura, H., Kimura, F.: Design of solids with free-form surfaces.
Computer Graphics 17(3): 289-298, 1983
[7] Hosaka, M., Kimura, F.: Non-four-sided patch expressions with control
points. Computer Aided Geometric Design 1(1): 75-86, 1984
[8] Charrot, P., Gregory, J .A.: A pentagonal surface patch for CAGD.
[9] Herron, G.: Smooth closed surface with triangular interpolants. Computer
[10] Farin G. Triangular Bernstein-Bezier patches. Computer Aided Geometric
Design 3(2): 83-127, 1986
[11] Sabin, M. A.: Some negtive results in N sided patches. Computer Aided
Design 18(1): 38-44, 1986
[12] Chiyokura, H.: Localized surface interpolation method for irregular meshes.
In: Kunii, T. (ed.): Computer Graphics. Berlin: Springer-Verlag, 1986
[13] Holmstroem, L.: Piecewise quadric blending of implicitly defined surface.
Computer Aided Geometric Design 4: 171-189, 1987
[14] Shirman, L.A., Sequin, C.H.: Local surface interpolation with Bezier
patches. Computer Aided Geometric Design 4{ 4): 279-295, 1987
[15] van Wijk J.J.: Bicubic patch for approximating non-rectangular meshes.
[16] Farin, G. et al.: The octant of a sphere as a non-degenerate triangular patch.
344 Bibliography
[17] Storry, D.J.T., Ball, A.A.: Design of an n-sided surface patch from Hermit.e
boundary data. Computer Aided Geometric Design 6(2): 111-120, 1989
[18] Kushimot.o, T., Hosaka, M.: *Design met.hods of surface in non-four-sided
regions. 1, J.JSPE 55(08): 55-60, 1989. 2, J.JSPE 55(11): 119-124, 1989
[19] Kushimot.o, T., Hosa.ka, M.: *Connection of triangular patches and it.s
application. J.JSPE 55(10): 67-72, 1989
[20] Sait.oh, T., Hosaka, M.: Int.erpolating curve networks with new blending
pat.ches. In: Vandoni, C.E., Duce, D.A. (eds.): Proc. Eurographics '90,
pp.137-146
[21] Hosaka, M.: *New solution of connection problem in free-form surfaces.
J. IPS Japan 31(5): 612-622,1990
[22] Peters, J.: Local cubic and bicubic C 1 surface interpola.tion with linearly
varying boundary normal. Computer Aided Geometric Design 7: 499-516,
1990
Chapter 16 and Appendix

[1] Phillips, M.B., Odel, G.M.: An algorithm for locating and displaying
t.he int.ersect.ion of two arbitrary surfaces. IEEE Computer Graphics and
Applications 9: 48-55, 1984
[2] Dokken, T.: Finding intersections of B-spline represent.ed geometies using
recursive subdivision techniques. Computer Aided Geometric Design 2(1-3):
37-48, 1985
[3] Pfeifer, H-U.: Met.hods used for intersecting geometrical entities in the GPM
. module for volume geometry. Computer Aided Design 17(7): 311-318, 1985
[4] Hought.on, E.G.: Implement.ation of a divide-and-conquer method for
int.ersection of parametic surfaces. Computer cAided Geometric Design
2(1-3): 173-183, 1985
[5] Miller, JcR.: Sculptured surfaces in solid models: issues and alt.ernative
approaches. IEEE Computer Graphics and Applications 12: 37-48, 1986
[6] Barnhill, R.E., Farin, G., Jordan, M., Piper, R.R.: Surface/suface
intersection. Computer Aided Geomet.ric Design 4(1-2): 3-16, 1987
[7] Bajaj, C.L., Hoffmann, C.M., Lynch, R.E.: Tracing surface intersections.
Comput.er Aided Geometric Design 5(4): 295-307, 1988
[8] Hosaka, M.: *Connect.ion and interference of surfaces. In: Proc. symp. on
graphics and CAD '87: IPS Japan 1988, pp. 57-66
[9] Hosaka, M., Kushimoto, T., Gonda, H.: *Connection and interference of
surfaces. In: Proc. symp. on graphics and CAD '88 IPS Japan: 1988,
pp.121-131
[10] Hoffmann, C.M.: Geomet.ric and solid modeling, San Mateo CA: Morgan
Kaufmann 1989
Bi bliograpllY 345
[11] Higashi, M, Mori, T., Hosaka, M.: Int.erference calculation of surfaces

based on t.heir geometric properties. In: Kimura, F., Rolst.adas,
A.( eds.): Comput.er Applications in Production and Engineering CAPE '89
: Amsterdam: Elsevier 1989, pp.275-282
[12] Aziz, M.A., Bata, R., Bhat. S.: Bezier surface/surface intersection IEEE
Comput.er Graphics and Applications 1: 50-58, 1990
[13] Hosaka., H., Kushimoto, T.:*New methods of intersection tracing of curves
surfaces. In: Proc. symp. on design JSPE: 1990, pp. 1-3, pp.7-9
[14] Stoer, J., Bulirsch, R.: Introduction to numerical a.nalysis. New York:
Springer-Verlag 1980
[15] Press, W.H., Flannery, B.P., Teukolsky, S. A., Vetterling, W.T.: Numerical
recipies in C. Ca.mbridge: Cambridge University Press 1988
[16] Burden, R.L ., Fa.ires, J.D.: Numerica.l analysis-4th edition. PWS-KENT,
1989
Chapter 17
[1] Sakai, Y., Kuranaga, Y.: Development of CADDETT system. In:
Proc.ISATA'81, Croydon: Automotive Automation Ltd. 1981
[2] Takahashi, A., Inoue, M., Okamaoto, 1., Mori, T.,: Stamping die CAD at
Toyota.. In: Proc.CAMP '83 Berlin: AMK , 1983, pp.623-636
[3] Higashi, M., Kohzen, I., Nagasaka, J.: An interactive CAD system for
construction of shapes with high-quality surface. In: Warman, E.A. (ed.):
Proc. CAPE '83 Amsterdam: North-Holland 1983, pp.371-390
[4] . Nagasaka., J., Higashi, M. Sa.nnokyou, H., Aoyama., N.: High speed
machining system for styling design. In: Proc. ISATA 84, Automotive Au-
tomation Lt.d., Croydon: 1984, pp. 1083-1098
[.5] Higashi, M., Mori T., Taniguchi, H., Yoshimi, J.: Geometric modelling
for efficient evaluation of press forming severity. In: Wang, M.M., Fa.ng,
F.C.(eds.): Proc. TMS-AIME symposium on computer modelling for sheet
metal forming process, Detroit: AIME, 1985, pp.21-35
[6] Takahashi, A., Okamoto, 1., Hiramatsu, T.,Yamada, N.: Evaluation meth-
ods of press forming severity in CAD application. In: Wang, M.M., Fang,
F.C.(eds.): Proc. TMS-AIME symposium on computer modelling of sheet
meta.l forming process, Detroit: AIME, 1985, pp.37-50
[7] Araki, H., Kawa.i, M., Amano, Y., Nonaka, M.: Repla.cement of physical
models by accurate numerical models for ma.nufacturing st.amping dies in
t.he integrated CAD/CAM system. In: Crest.in, J.P., MacWaters, J.F.(eds.)
Proc. Prolamat: Amsterdam: North-Holland 1985, pp.47-60
[8] Araki, H., Niwa, S, Makino, T., Yoshinaka, T.: Computer aided finishing
system for stamping die manufacturing. In: Proc. CAMP '85, Berlin: AMK,
1985, pp.203-208
[9] Kobayashi, S.: Calculation of free-form surface intersection and its
application to evaluation of mass-property. In: Proc. Intn'l conf. computers
346 Bibliography
in engineering. ASME: 1985, pp.325-330

[10] Izumi, M., Kobayashi, S.: Development of a solid modeller for FEM data
generation. In: Proc. ISATA '85, Croydon: Automotive Automation Ltd.
1985, pp. 535-570
[11] K uragano T. et a1.: TheFRESDAM system for generating tool paths for
ma.nufact.uring aesthetically pleasing injection mold products. In: Kimura,
F., Roistadas, A. (eds.) Proc. CAPE'89, Amsterdam: Elsevier 1989,
pp.411-418
Subject Index
B point locations, general expression 177 Chiyokura's method 238

B spline basis functions 180 clay models 299,301
BO,l 165 compensation
ball end mill 290 connection of four-sided patches 238
barycentric coordinates 253 connection of three-sided patches 246
beta spline 156 computer aided style design 304
Bezier patch 133 conics 45
division and degree elevation 137 classification 47
geometric properties 136 equation 45
Bezier intersection 48
B polygon 121 pencil 48
both-end-truncated B polygons 148 transformation 45
CDP B(2)-1 ,0
146 connection
connection of B curve segments 142 connection defining polygon 146
continuity condition of connection 131 compensating patch 238
continuity conditions 126 compensation for C(2) 250
control polygon 117 evaluation and comparison of methods 243
curves and control points 117 floating inner control points 237
derivatives 122 general C(l) connection of B patches 213
different expressions 121 patch connection from C(O) to C(1) 239
division of curve segment 128 patches by blending 233
elevation of degree 131 pentagonal region: general case 232
, geometric properties 126 three or five-Sided patch in regular patch
hodograph 122 nets 217
incomplete B polygons 147 three-sided patch in C(2) 252
operator form 118 three-sided patch in S net 246
original form 138 triangular region: general case 231
scale ratio 131 two-valued twist vectors 237
weighting functions 118 contour orthogonal curves 88
binormal vector 57 control of curvature distribution 158
blending by weighted sum 236 control points 117
Bulirsch-Stoer location symbol 176
method 327 convex edge 37
integration procedure 328 convex hull 118
outline of procedures 328 convex polyhedron 37
polynomial extrapolation 329 Coons' patch 233
principle of the method 327 coordinate transformation matrix 1 15
rational extrapolation 331 coordinate transformation matrix 2 17
c(n-l) connection 144 coordinate transformation 11
CAD/CAM offree-form injection-mold prod- coordinate
ucts 317 construction of a coordinate system 7
canonical perspective 184 independent 118
car developmentf 299 oblique coordinate system 7
CDP 146 correction of cross-boundary tangent vectors
Ceva's theorem 34,231 238
348 Subject index
curvature profile 157 independent boundary conditions 236

curvature 55,60 input of simplified drawings 305
center 57 integrated CAD/CAM 301
radius 57 interface 299
curve interference of polyhedra 41
differential geometry 55 interpolation
expressions with parameter t 59 and extrapolation with conics 204
fitting by conics 210 by parametric curves 113
with minimum energy 111 numerical calculation 104
deBoor 180 rational function 107
derivation of B polygons from an S polygon intersection curve 22
170 differential equations 282
die-face 312 intersection
design system 312 a curved surface and a plane 268
design 301 both surfaces with implicit expressions
differential equation solvers 285
examples and evaluation of 332 both surfaces with parametric expressions
dihedral angle 37 282
distance between a conic and a point 209 critical contact points 275
division of S polygon 174 critical points 279
division of a triangular patch 262 curve tracing 270
Dupin's indicatrix 77 detection of intersection loops 275
elastic beam analogy 115 initial starting points 275
elastic energy 115 line and plane 35
elastica 115 method with an auxiliary plane 273
elimination process of coefficients 64 near a probable singular point 286
engineering drawings and geometric models offset surfaces 290
297 points on intersection 272
Euler's formula for normal curvature 77 surfaces with implicit and parametric ex-
Euler-Poincare formula 44 pressions 284
Eulerian relation 39,42 three planes 38
evaluation two offset surfaces 291
curvature distribution check 311 Lagrange's formula 103
curve and surface quality 308 line 31
highlight check 310 segment 31
extremum search curves 282 lines of curvature 73
FEM calculations 302 intersection 31
FRESDAM 318 local control of shape 145
fundamental magnitudes of the first order local operations 42
68 local shape of a curve 59
fundamental magnitudes of the second order local shape of a surface 76
70 location symbols
Gaussian curvature 72 application 177
Gaussian curvature, another form 74 formulas 178
generator lines 51 magnitude of a vector 3
geometric constraints calculation 22 master drawing 301
geometric continuity condition G(2) 154 master model 301
Gregory, A 237 matrix 9
Hermite interpolation method 110 arithmetic operations 10
implicit expression of a parametric curve 63 eigenvector 14
incomplete B polygons • B~{, B(j(n) 147 orthogonal matrix 13,14
incomplete S polygon Si(n) 149 principal directions and eigenvalues 14
Subject index 349
product of matrices 10 rational Bezier

transposed matrix 12 curves 182
mean curvature 72 canonical perspective 184
Menelaus' theorem 33 ,70 division of a rational polygon 205
Menelaus extension of a curve segment 206
division 33 weight of a control point 204
edge 168 of degree two, classification by weight 201
edge, level of 175 rational
Meunier's theorem 70 B polygons from rational S polygon 193
mirror Bezier nets 197
displacement matrix 29 Bezier patches 191
movement by four mirrors 27 canonical perspective 184
rotation around an 25 circular arcs 203
translation 25 conversion to an implicit form 199
modellers 301 derivatives at ends of a segment 190
movement of a body by reflection 24 dividing ratio 182
movement division and degree elevation 189
rotation 18 G(2) connection 196
successive rotations 19 sphere and surface of revolution 204
translation 18 spline nets 197
Newton-Raphson method 273 splines 193
non-regular connection of B patches 213 surface of revolution 204
normal curvature 69 triangular patch 255
offset intersection virtual origin 182
parametric and implicit surfaces 293 weight of zero 188
two implicit surfaces 294 weight 182
two parametric surfaces 291 reduced S Polygons 170
offset surface 97 the first kind 171
fundamental magnitudes 100 the second kind 172
Gaussian and mean curvatures 99 Rodrigues' formula 75,98
offset rounding
Intersection with a Plane 290 a convex corner 241
Rodrigues' formula 95 a convex-concave corner 242
osculating plane 56 convex region 217
pencil 53 convex-concave mixed region 1 220
plane equation 35 hexagonal region 242
polyhedron 37 with a rolling ball 228
geometric characteristics 39,40 Ruled Surfaces 102
local operation 42 Runge-Kutta
topological net 42 Runge-Kutta-Fehlberg method 325
topological relations 37 adaptive Runge-Kutta method 324
predictor-corrector method, variable step size Runge-Kutta step 324
326 with quality control 324
principal curvatures of surface 72 S polygon 148
principal directions 73 8i - 1 ,i 167
principal normal vector 56 scale factor list 143
product of matrices 11 scale ratio 131,142
product of vectors 3 Serret-Frenet's formula 9,58
quadrics 45,49 shift operator 120
classification 51 skew shifting 169
coordinate transformation 49 spline
intersection 53 both-end-truncated S polygon 149
350 Subject index
conditions of C(i) connection 143 tensor product surface 12

dividing points 168 unit normal vector 69
di viding ratio list 148,166 tangent 55
division of S net 164 along-boundary tangent 214
increase of vertices 152 cross-boundary tangent 214
insertion of vertex 174 unit tangent vector 55
locating B points 148 three-term simultaneous equations 114
main scale factor 149 torsion 57,60
main scale 167 transformation
nets 161 calculation of transformation matrix 16
reduced-truncated S polygon 149 movement and coordinate transformation
sub-edge 148 18
tensor product surface 161 transposition 2
unit S polygon 148 triangular
stamping die design 301 arbitrary connection 261
stretch factor 155 C(1) connection 257
style design system 303 control points of the divided small triangle
style design, two aspects 303 264
surfaces in style design, calasses and types control quadrilateral 258
307 elevation of degree 265
surface operator form 253
basic theory 67 patch of degree n 255
basic vectors 68 surface 253
Boolean sum 236 tangent on patch boundary 256
characteristic curves 83,74,95 truncated B polygon 148
contour curves 88 twist vectors 235
equi-brightness curves 88 umbilics 79
equi-gradient Curves 88 unit S polygon 167
extremum Search Curves 85 unit vectors 1
fu.ndamental magnitudes 68 user interface 299
gradient extremum curves 95 vector 1
highlight curves 92 absolute value 3
lines of curvature 73,84 coplanar 5
loci of these extremum curvature 97 differentiation of vectors 8
locus of zero Gaussian curvature 96 displacement vector 2
normal curvature 69 inner product 3
normal section 69 position vector 2,15
principal curvatures 71 scalar triple product 4
principal directions 73 triple products 4
ridge-valley curves 95 unit vector 3
shape around a point 76 vector product 3
silhouette curve 91 vector triple product 5
silhouette pattern 91 Weingarten's equations 74
Springer Series
Computer Graphics - Systems and Applications
Former Subseries of SYMBOLIC COMPUTATION
J.L.Encarnarrao, R.Schuster, E.Voge (eds.):

Product Data Interfaces in CAD/CAM Applications.
Design, Implementation and Experiences.
IX, 270 pages, 147 figs., 1986
URembold, RDillmann (eds.):

Computer-Aided Design and Manufacturing.
Methods and Tools. Second, revised and enlarged edition.
XIV; 458 pages, 304 figs., 1986
G. Enderle, K. Kansy, G. Pfaff:

Computer Graphics Programming. GKS - The Graphics
Standard. Second, revised and enlarged edition.
XXIII, 651 pages, 100 figs., 1987
Y. Shirai:
Three-Dimensional Computer Vision.
XII, 297 pages, 313 figs., 1987
D.B.Arnold, P.RBono:
CGM and CGI. Metafile and Interface Standards
for Computer Graphics.
XXIII, 279 pages, 103 figs., 1988
J. L. Encarnarrao, P. C. Lockemann (eds.):

Engineering Databases. XII, 229 pages, 152 figs., 1990
P. Wisskirchen:
Object-Oriented Graphics. From GKS and PRIGS
to Object-Oriented Systems .
XIII, 236 pages, 83 figs., 1990
J. L. Encarnarrao, R Lindner, E. G. Schlechtendahl:

Computer Aided Design. Fundamentals and System
Architectures. Second, revised and extended edition.
XII, 432 pages, 240 figs., 1990
H. Hagen, D. Roller (eds.):
Geometric Modeling. Methods and Applications.
VII, 286 pages, 140 figs., 1991
T. Yagiu:
Modeling Design Objects and Processes
X, 327 pages, 82 figs. , 1991
M.Hosaka:
Modeling of Curves and Surfaces in CAD/CAM.
XXI, 350 pages, 90 figs ., 1992

(Mamoru Hosaka (Auth.) ) Modeling of Curves and Su

Uploaded by

Copyright:

Available Formats

(Mamoru Hosaka (Auth.) ) Modeling of Curves and Su

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

(Mamoru Hosaka (Auth.) ) Modeling of Curves and Su

Uploaded by

Copyright:

Available Formats

Computer Graphics -

Systems and Applications

Managing Editor: J. L. Encarna~ao

© Springer-Verlag Berlin Heidelberg 1992

'IYPesetting: Camera ready by author

1 Aims and Features of This Book

2 Structure and Contents of the Book

afterwards. A topology of faces, edges and vertices of a solid body is explained

Since a rational Bezier of degree two is converted to an implicit form which is

3 Short Historical Note

chine for interactive designing would be completed in five years. Stimulated

Tokyo, August 1991 Mamoru Hosaka

1 Excerpts from Vector and Matrix Theory 1

2 Coordinate Transformations and Displacements 15

3 Lines, Planes and Polyhedra . . . . . . . . . . . . . 31

4 Conics and Quadrics 45

6 Basic Theory of Surfaces 67

7 Advanced Applications of Theory of Surfaces 79

7.3.1 General Remarks . 83

8 Curves Through Given Points, Interpolation and

9 Bezier Curves and Control Points 117

10 Connection of Bezier Curves and Relation to Spline

11 Connection of Bezier Patches and Geometry of Spline

12 Rational Bezier and Spline Expressions 181

13 ' N on-regular Connections of Four-Sided Patches and

14 Connections of Patches by Blending 233

15 Triangular Surface Patches and Their Connection 253

16 Surface Intersections 267

16.6.2.3 Two Surfaces with Implicit Expressions 294

17 Applications of the Theories in Industry 297

Appendix. Numerical Methods of Differential Equation Solving 323

Subject Index 347

1.2 Notations of Vectors and Vector Arithmetic

i = (1,0,0), j = (0, 1,0), k = (0,0, I), (1.1)

and if we let components of an arbitrary vector a be ax, ay and az , then a is

c = ca = (ca x, cay, ca z ), (1.3)

Figure 1.1 Graphic representation of vectors and their arithmetic

To express a vector, its components are written horizontally and enclosed by a

The superscript T indicates transposition.

1.3 Product of Vectors

a .b = lal . Ibl cos e. (1.12)

An inner product a . a, which is written a 2 , is the same as the square of the

1.3.2 Vector Product

Icl = lal·lbllsinel· (1.15)

b. When the order of multiplication is reversed, the direction of the resulting

By the two orthogonal conditions a . c=O and b . c=O, we have

where 'Y is a constant independent of a and b. To determine the value of the

1.4 Triple Products

Figure 1.2 Scalar triple product

[a, b, c] = [b, c, a] = [c, a, b] = -[b, a, c]. (1.2.3 )

1.4.2 Vector Triple Product

(axb)xc Figure 1.3 Vector triple product

(axb) xC=Aa+f.lb, (1.24)

1.4.3 Application Examples

Example 1. Construction of a Coordinate System

Then writing ( = n, ~ = it, we make iJ orthogonal to ( and ~:

(a x b) x a = (a· a)b - (b· a)a = a 2b - (b· a) a,

so the expression for iJ is

The orthogonal unit vectors ~, f] and ( thus determined make an orthogonal