Sei sulla pagina 1di 46

Qualitative Smooth Shape in the Plane

Ellis D. Cooper
xtalv1@netropolis.net
March 3, 2012
Abstract
When you look at a shape in space, say a sculpture, from one position you
only see one part of it. Once you see every part, you can imagine the whole
sculpture. And, if you can imagine the whole sculpture, you can predict how it
would look from any position. These inverse experiences depend less on precise
measurements than on qualitative shape. What types of mathematics might
make precise similar observations in the presumably easier case of qualitative
shape in the plane?
Keywords
ambient isotopy, anorthoscopic perception, Category Theory, Knot Theory,
Morse Theory, phenomenology, qualitative shape
0
Contents
1 Introduction 2
1.1 Phenomenology . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.2 Psychology of Visual Perception . . . . . . . . . . . . . . . . . . 3
1.3 Optical Character Recognition . . . . . . . . . . . . . . . . . . . 3
2 Part I. Partial View of Opaque Shape from Near Viewpoint 4
2.1 Points, Vectors and Located Vectors . . . . . . . . . . . . . . . . 4
2.2 Curve, Circle, and Shape . . . . . . . . . . . . . . . . . . . . . . 6
2.3 Near Viewpoint and Partial View . . . . . . . . . . . . . . . . . . 8
2.4 Visual Event and Abstract Partial View . . . . . . . . . . . . . . 10
2.5 Aspect Graph . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
3 Part II. Total View of Transparent Shape from Far Viewpoint 17
3.1 Orientation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
3.2 Inecta and Extrema . . . . . . . . . . . . . . . . . . . . . . . . . 18
3.3 The Cell Complex Induced by a Far Viewpoint . . . . . . . . . . 19
3.4 Vertically Ordered Graded Poset . . . . . . . . . . . . . . . . . . 20
4 Part III. Approximation and Experimentation 28
4.1 Bezier Shape . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
4.2 Screenshots of Scultoro . . . . . . . . . . . . . . . . . . . . . . . . 30
5 Part IV. An Imperfect Analogy: Knots and Shapes 32
5.1 Homotopy, Isotopy, and Ambient Isotopy . . . . . . . . . . . . . 34
5.2 Analogy Table with Imperfections . . . . . . . . . . . . . . . . . 38
6 Part V. Conclusion and Problems 39
Bibliography 41
Index 43
1
1 Introduction
1.1 Phenomenology
We may speak of the evidence that the identity is given only if a
continuous transition from the one perception to the other is guar-
anteed in the unity of experience. The unity of the object demon-
strates itself only in the unity of the synthesis continually joining
the manifold perceptions, and this continuous synthesis must lie at
the foundation in order for the logical synthesis, that of identica-
tion, to produce the evident givenness of the identity of the objects
appearing in the various perceptions. The perceptions must be in-
tegrated as phases into the synthesis, and we see that precisely only
if we carry out the synthesis.
This important fact holds generally. In our case, it means that
an identical and unchanged spatial body demonstrates itself as such
only in a kinetic series of perceptions, which continually brings to
appearance the various sides of that thing. The body must rotate or
be displaced, or I must move, move my eyes, my Body, in order to
see it from all around, and at the same time I must keep approaching
it and receding. Or. nally. both I and the thing must move. That
is how the state of aairs is expressed from the standpoint of the
appearing thing [13].
I make no pretense of understanding the principle of phenomenological
reduction introduced by Edmund Husserl for his general philosophy of phe-
nomenology. I do know that his work includes mathematical diagrams per-
taining to the lawfulness of the modications of appearances in the case of
qualitative change (e.g., [13], p.232). I have challenged myself to see what I
can say on the topic of qualitative shape in terms of some basic modern mathe-
matical technologies. By no means have I a mathematical theory of qualitative
shape, with calculations and reasoning to prove theorems. I do have geomet-
rical constructions and formal denitions of what seem to me to be relevant
concepts. And I do make technical claims for which the evidence is provided by
numerous diagrams. Thus, this document is less of a research paper and more
of a preliminary research proposal.
Due diligence leads me to conclude that there is a gap in the mathematical
literature between the study of topological spaces that are considered equivalent
if they are homeomorphic, and the study of rigid shapes entirely determined by
the distances between pairs of distinct points. The hierarchy of geometries cod-
ied in terms of the group of automorphisms of a space as in the Erlanger
Program of Felix Klein
1
includes, between topology with its oppy homeomor-
phisms and Euclidean Geometry with its rigid isometries, many intermediaries
such as projective geometry, algebraic geometry, and equiane geometry. I do
1
Weisstein, Eric W. Erlanger Program. From MathWorld A Wolfram Web Resource.
http://mathworld.wolfram.com/ErlangerProgram.html
2
not nd in that range a geometry of qualitative shape, which is a concept
that remains to be dened. To be sure, the idea is not ignored by scientists and
mathematicians. For decades Jan J. Koenderink has studied conceptually and
mathematically the psychology and philosophy of perception, computer vision
and ecological physics [15][16]. Also, Dierential topology, and specically
Morse Theory, provide a suitable setting for formalizing and solving several
problems related to shape analysis [4]. And there are generative theories
of shape, such as [19] in which features of shapes correspond to symmetry
groups, and addition of features corresponds to group extensions. However, to
my knowledge only one study directly targets the fundamental question of den-
ing qualitative shape and that is in the context of Geographical Information
Systems that simulate the mental processes of human beings [7].
1.2 Psychology of Visual Perception
A purpose of this proposal is to formalize some questions about integration of
partial views of a shape to form a total perception of it. The phenomena of
anorthoscopic perception also relate to integration of partial information, but
that which is obtained by moving a viewing slit across the eld of view.
Our purpose was to investigate the contribution of pursuit eye
movements to the integration of gures moving behind a slit (anortho-
scopic perception). Outline shapes were moved across a simulated
slit (without visible borders) on a CRT screen. Eye movements were
monitored with a dual-Purkinje image eye-tracker. The slit width
was adjusted so that there were spontaneous transitions between pe-
riods in which observers saw a horizontally moving shape and periods
in which they saw only vertically moving contour segments. Subjects
reported transitions between these alternative percepts. On half the
trials the slit was retinally stabilised to eliminate pursuit- dependent
retinal painting. During periods of gure perception, there was often
low-amplitude spontaneous pursuit of the horizontal gure motion.
Lower-amplitude pursuit was also some- times observed when no
gure was reported. Irrespective of gure perception, pursuit ampli-
tudes were larger with stabilisation. However, the total percentage
of time that a shape was seen, and the duration and frequency of
episodes of gure perception, were not signicantly dierent during
stabilised and non-stabilised viewing. Our ndings suggest that, al-
though anorthoscopic gure percepts can elicit spontaneous pursuit
under free-viewing conditions, this pursuit and the pattern it pro-
duces do not contribute to the formation or maintenance of those
percepts [10].
1.3 Optical Character Recognition
My interest over thirty ve years ago leading to this work centered on the prob-
lem of optical character recognition by computers. Nowadays this problem is
3
largely solved, except perhaps for persistent diculties with general handwrit-
ing. But for printed text in newspapers, books and journals one may use a
scanner and commercially available OCR software to obtain a digital version of
the material in a matter of minutes. Back in the day, however, pondering this
problem I imagined a tiny spark of intelligence, a pointhuman, descending by
parachute into the middle of a letter on a page, challenged to determine the
identity of the letter. The assumptions are that the letter is a shape with empty
interior and opaque boundary, and that (a) the pointhuman is mobile in any
direction, (b) it can look in any one direction at a time and determine whether
a far point of the boundary is occluded by a near portion of the boundary,
and (c) it has unlimited calculating and reasoning ability, including unlimited
memory. Of course, (d) the pointhuman must also have some sort of table of
shapes against which to compare the information it gathers and integrates by
exploring the interior of the letter [8]. One thing led to another, and I imagined
a pointhuman landing, not inside a letter, but outside an island in the ocean
of an alien planet, now challenged not to recognize but to cognize the shape
of the island. That thought led to questions such as, what is the minimum
number of exterior viewpoints from which the collected partial views of the is-
land would be sucient to claim knowledge of its entire shape? For a circular
island that would be 3. And then I was thinking about similar explorers in
three-dimensional space, and noting that a minimum of 4 viewpoints would be
necessary to view the entire surface of a spherical planet. What would be the
minimum number to see the entire surface of a doughnut-shaped world? In
any case, the minimum number of comprehensive partial views should be an
invariant of qualitative shape.
2 Part I. Partial View of Opaque Shape from
Near Viewpoint
2.1 Points, Vectors and Located Vectors
Denition 1. The plane is mathematically represented by the set R
2
. An
element A = (x, y) of R
2
is called a point. every point A = (x, y) corresponds
to a vector

A =

(x, y), and the space of vectors is denoted by

R
2
, hence there
is a 1-1 correspondence R
2

R
2
: A

A between points and vectors. R
2
is an unstructured set of ordered pairs, whereas

R
2
is a 2-dimensional vector
space dened by the usual operations of component-wise addition and scalar
multiplication.
The translate of a point A = (u, v) along a vector B =

(x, y) is the point


A+

B := (u +x, v +y). See Fig.(1).


Translation of points along vectors is a map R
2

R
2
+
R
2
that satises the
4
(0, 0)
(x, y)

(x, y)
(u, v)
(u, v) +

(x, y)
Figure 1: A point, its vector, and a point translated by its vector. The vector
of a point translates the origin (0, 0) to the point.
following equations:
A = (0, 0) +

A; (1)
A + (

B +

C) = (A +

B) +

C. (2)
Furthermore, for any two points A, B there exists a unique vector

C such that
B = A+

C [30].
Denition 2. The unique vector that translates a point A to a point B is
denoted by B A

R
2
. This means there exists a map R
2
R
2

R
2
such
that A+ (B A) = B.
A located vector is a pair (A,

C), that is to say, an element of R


2

R
2
.
Observation 1. Located vectors are the morphisms of a category [21][28] whose
objects are the points, where dom(A,

C) := A, cod(A,

C) := A+

C, the identity
morphism of A is (A, (0, 0)), and the composition of (A,

C) followed by (A +

C,

D) is (A,

C +

D). See Fig.(2).


Denition 3. For a located vector (A,

C) the ray Ray(A,

C) R
2
with
vertex A in the direction of

C is the set of points of the form A+s

C where
0 s R.
For any two points A and B there are intervals of points dened as follows.
The closed interval [A, B] R
2
is the set of points of the form A+s(B A)
where 0 s 1; the half-open interval [A, B) is the set of points of the form
A + s(B A) where 0 s < 1; and (A, B], (A, B) are dened analogously,
although the latter is ambiguously either an open interval, or just an ordered
pair, so context must be consulted to clarify intention. See Fig.(3).
5
A

D
Figure 2: Addition of vectors is a composition of morphisms.
Ray(A,

C)
A
C
D
E
[D, E)
Figure 3: Intervals and a ray.
2.2 Curve, Circle, and Shape
Denition 4. The (unit) circle S
1
is the 1-dimensional submanifold of the
plane R
2
with underlying set { (x, y) | x
2
+ y
2
= 1 }. If R
2
is considered to be
the underlying set of the complex plane C then also S
1
= { e
i
| R}. A unit
vector

u :=

e
i

S
1
is called a direction, see Fig.(4). Every vector

A has
magnitude

and direction, namely



A

.
Denition 5. A shape is a smooth embedding S
1
K
R
2
of the circle in the
plane. Smooth means K is a dierentiable map of manifolds, so there is a
derivative tangent vector

t =

t (K, x) at each point of the image of K in
the plane. Embedding means K is an injective map, i.e., distinct points of
S
1
map to distinct points of R
2
. It is convenient for the letter K to denote
ambiguously the map and its image Im K = K(S
1
) R
2
.
Observation 2. K smooth implies that at each point x of K there exists a
6

e
i
1 + 0i
0 + 1i

e
i
Figure 4: The complex rotator.
unique tangent line L(K, x) consisting of all points of the form x + s

t (K, x)
for s R. See Fig.(5).
x

t (K, x)
y
L(y)
Figure 5: A tangent line and a unit tangent vector.
Denition 6. A curve (also called path) in the plane is a continuous map
[0, r]

R
2
dened on a closed interval [0, r] R. See Fig.(6). Curve is
simple if it is an injective map, and closed if the end (r) equals the start
(0). The notation x

y signies that path starts at x and ends at y.
Recall the Jordan Curve Theorem (rst proved satisfactorily by Oswald Ve-
blen) which asserts that for any simple closed curve the plane is partitioned into
its bounded interior, the curve itself, and its unbounded exterior [3][24][25][31],
R
2
= In K K Ex K .
Observation 3. Some might prefer to say is a parametrization of its image
([0, r]), and that image is the curve. The advantage of dening the curve to
7
be the map is that the natural orientation of R (from 0 to 1) induces a direction
along the curve [29]. Consequently, there exists a category with objects the points
of R
2
and morphisms the paths x

y. The constant path on [0, 0] with value x
is the identity morphism of x, and composition is dened by following one path
and then the next [6].
x
y

Figure 6: A curve.
2.3 Near Viewpoint and Partial View
Denition 7. Let K be a shape. A point v Ex K is called a near viewpoint
of K. A sight-interval from near viewpoint v to K is a closed interval [v, x]
where x K and [v, x) Ex K. Note that If [v, x] is a sight-interval and
w [v, x] then [w, x] is also a sight-interval. See Fig.(7).
v
x
K
Ex K In K
[v, x]
Figure 7: A sight-interval.
Denition 8. The partial view of K from near viewpoint v is the subset of
points x of K such that there exists a sight-interval from v to x. See Fig.(8).
8
v
x
1
x
2 x
3
x
4
x
5
x
6
K
Figure 8: Partial view. Little open circles at points x
3
, x
5
represent occlusion
of those points by x
2
, x
4
, respectively.
Denition 9. Let K be a shape and L a line in the plane. If L meets K
and x L K then either L is tangent to K at x so L = L(K, x), or not,
in which case L crosses K at x. Likewise, if v is a near viewpoint of K and
R = Ray(v,

u ) meets K at x where

u

S
1
is a direction, then either R is
tangent to K or crosses K at x.
Claim 1. For any shape K, near viewpoint v, and direction

u , if R :=
Ray(v,

u ) and n = #(R K), then either n = 0, or there exists a sequence


of points (x
1
, . . . , x
n
) the contact list such that x
i
K for 1 i n,
[v, x
1
] is a sight-interval from v to K, and

u is the direction of x
i+1
x
i
, for
1 i < n. Of course, R is either tangent to or crosses K at each contact
point x
i
. Moreover, for every sight-interval [v, x] there exists a unique ray R
with vertex v such that x is the rst contact point of R with K. Thus, the partial
view of K from v is the set of rst contact points of rays with vertex v that meet
K. See Fig.(9).
v
x
1
x
2
x
3
x
4
Figure 9: Contact list.
9
2.4 Visual Event and Abstract Partial View
Denition 10. Let First(K, v,

u ) be the partially dened function of

u that is
equal to the rst contact point of Ray(v,

u ) with K if it exists, and equal to


otherwise. The partial view of K from v is the set of values where First(K, v,

u )
is dened.
Claim 2. Let

u := e
i
be a direction and

u

:= e
i()
,

u
+
:= e
i(+)
be
innitely near directions, that is, > 0 is an innitesimal.
2
If Ray(v,

u ) is
tangent to K at First(K, v,

u ), then exactly one of the following holds:


First(K, v,

u
+
) = and First(K, v,

) First(K, v,

u ) (3)
First(K, v,

u
+
) First(K, v,

u ) and First(K, v,

) = (4)
First(K, v,

u
+
) / First(K, v,

u ) and First(K, v,

) First(K, v,

u ) (5)
First(K, v,

u
+
) First(K, v,

u ) and First(K, v,

) / First(K, v,

u ) (6)
First(K, v,

u
+
) First(K, v,

u ) and First(K, v,

) First(K, v,

u ) (7)
where is the relation innitely near, which means equal standard parts,
and the / / case is impossible since K is continuous. See Fig.(10).
Denition 11. In case Equ.(3) the partial view drops to innity from the
right; in case Equ.(4) it drops to innity from the left; in case Equ.(5) it
drops back from the right; in case Equ.(6) it drops back from the left;
and, last, in case Equ.(7) the partial view sees an inection.
Claim 3. Let K be a shape, v a near viewpoint, and

u a direction. Let R :=
Ray(v,

u ). Then
A First(K, v,

u ) = so R K = ; otherwise, let x := First(K, v,

u ) K. Then either
B =: X R crosses K at x; or R is tangent to K at x, and either
C =: S x is an inection of K; or
D =: C K drops to innity from the right at x; or
E =: C K drops to innity from the left at x; or
F =: C K drops back from the right at x; or
G =: C K drops back from the left at x.
Observation 4. Consequent to this claim, there exists a function associated to
K and v assigning to each direction

u one of the values A, B, C, D, E, F, G. As
a mnemonic aid, a special symbol is associated with each of these letters.
Claim 4. Values , X, S, C, C , C, and C occur at most in a nite number
of directions, and on an open interval separating two of those directions, the
value is constantly one of or X.
2
Mikhail G. Katz and David Tall, Tension between Intuitive Innitesimals and Formal
Mathematical Analysis, http://arxiv.org/ftp/arxiv/papers/1110/1110.5747.pdf
10
(C) ( C )
(C) ( C )
(Z) (S)

u
+
u

u

Figure 10: Contact classication.


Observation 5. A direction with value just means that there is no sight-
interval from the near viewpoint to the shape. Value X means there is a sight-
interval along a ray that immediately crosses the shape. Otherwise, either there
is a sight-interval along a ray tangent to the shape at an inection, S, or there
is a sight-interval along a horizon of the shape, Cor C , or a sight-interval
tangent to the shape at an extremum that occludes a point of the shape further
away, Cor C . See Fig.(8).
Denition 12. The values S, C, C , C, or C are called visual events, and
the values , Xare called visual non-events.
Observation 6. As a direction at angle rotates from = 0 counter-clockwise,
there exists a periodic sequence of visual events separated by constant or
Xvalues along open intervals. There may or may not be a visual event at = 0.
If there is, then there is a nal open interval as approaches 2 before the
sequence repeats. If there is not, then there is an open interval between the last
visual event before = 2 and the rst visual event after = 0. Either way,
there exists a unique sequence of symbols , X, S, C, C , C, or C in one
period of the sequence as varies from 0 to 2. Any occurrence of signies an
interval of directions in which no point of the shape is in sight. However, distinct
occurrences of Xrepresent intervals in which distinct portions of the partial view
are visible.
Denition 13. The nite sequence of visual events and non-events associated
with directions at v rotating counter-clockwise from

1 +i0 is called the abstract


partial view of K at v.
Two near viewpoints v, w for which there exists a path v

w such that
the abstract partial view is invariant up to a cyclic permutation are abstractly
equivalent near viewpoints, for which the notation is v

= w. Any two near
viewpoints of a circle are abstractly equivalent, with abstract partial view C C.
11
2.5 Aspect Graph
Claim 5. Abstract equivalence of near viewpoints is an equivalence relation on
Ex K, and there exists a nite number of abstract equivalence classes of near
viewpoints. An equivalence class is called an aspect of the shape. Each aspect is
a (possibly unbounded) connected region of the plane. The boundary of an aspect
may or may not contain a subset of the points of the shape. The boundary of
an aspect is composed of sets of points its components of one or more of
the following kinds:
a continuous portion of the shape itself; (8)
a ray contained in the exterior of the shape; (9)
a tangent ray with vertex on the shape, and not meeting its interior; (10)
an interval [x, y), x, y K, and [x, y) In K = ; (11)
an interval [x, z], x K on the shape, y Ex K, and [x, y] In K = . (12)
Denition 14. Two aspects that share a common boundary component are
called contiguous.
Claim 6. Two contiguous aspects share exactly one boundary component. The
shared boundary component of two contiguous aspects is a subset of exactly one
of them.
C
Z C
C
C
C
Figure 11: Abstract partial view.
Referring to Fig.(12), examples of abstract partial views of an opaque smooth
12
x
1
x
2
x
3
x
4
x
5
x
6
x
7
v
1
v
2
v
3
v
4
v
5
v
6
v
7
v
8

u
1

u
2
y
1
y
2
Figure 12: An opaque shape, near viewpoints, and aspectual boundary compo-
nents.
shape K are as follows:
v
1
C XC (13)
v
2
XC C (14)
v
3
C XC (15)
v
4
C XC (16)
v
5
C X
1
C X
2
C X
3
C (17)
v
6
X
1
C C X
2
C (18)
v
7
XC C (19)
v
8
X C (20)
Likewise, examples of boundary components of aspects are as follows:
[x
1
, x
2
] closed interval between points of K, once tangent to K (21)
[x
3
, x
4
] likewise (22)
Ray(x
5
,

u
1
) ray tangent to K (23)
[x
6
, y
1
] closed interval from point of K to an external point (24)
Ray(y
1
,

u
2
) ray entirely external to K (25)
13
Denition 15. The directed graph whose vertices are the aspects of a shape,
and in which there exists an edge from one aspect to another provided they are
contiguous and the second contains their shared boundary component, is called
the aspect graph of the shape [26][27].
Observation 7. A line L R
2
may or may not meet a shape K. If L meets
K then L may or may not be tangent to K. If K has a attened portion then
there exists a line L tangent to K at an innite number of points. See Fig.(13).
L
1
L
2
L
3
Figure 13: A shape that is not rounded.
Denition 16. For line L and shape K let Tn(L, K) denote the number of
points at which L is tangent to K, and dene Tn(L, K) = if LK = , and
Tn(L, K) = + if L is tangent to K at an innite number of points. Otherwise,
Tn(L, K) is a non-negative integer, and Tn(L, K) = 0 means LK = but L is
not tangent to K at any point. Say K is a rounded shape if Tn(L, K) < +
for every line L.
Henceforth, assume all shapes are rounded.
Claim 7. If Tn(L, K) = 0 then #(L K) is a positive even integer.
Evidence 1. K is bounded so there exist oppositely directed rays contained in
L that do not meet K. At every point of L K there exists a crossing between
Ex K and In K. But the extreme end points of L K have Ex K on opposite
sides.
2
The circle is a trivial rounded smooth shape. First in complexity beyond
the circle is the bean, which serves throughout this proposal as the generic
example of a rounded smooth shape.
[C]onsider an oval with a dent. The dent will involve a bitan-
gent, two inection points, and a vertex (of the opposite curvature
to the major part). The conguration is not arbitrary, either, for
14
instance, the pair of inections must lie between the points of tan-
gency of the bitangent. Such feature clusters are certain to be
important in practice, yet they dont gure in the literature. One
way to get a handle on this is to view the dent as a result of some
operation on an oval. When you put the dent in you must pass
through a stage where the oval develops a at point, and the features
are seen to originate from an explosion of the at point into inec-
tion points, etc. Since the features were once contained in a single
point, they really belong together! Similar analyses can be done in
surface theory. The authors consider any surface as the end result of
the deformation of an initially simple surface such as an ovoid. As
the surface develops, feature clusters are created, and by noting
the history one may build a hierarchical order of features. It is like
noting the changes put by a sculptor on an originally shapeless blob
of clay that made it into a simile of a human face [16].
Observation 8. The bean shape also appears in diverse contexts such as Morse
Theory [5] and computer vision.
3
C X
1
C X
2
C
X C C
X
1
CX
2
C C
C X C
Figure 14: The bean and its abstract partial views.
3
Daniel Keren, Advanced Topics in Computer Vision: 3D vision, Chapter 23, Aspect
Graphs, http://cs.haifa.ac.il/
~
dkeren/acv/index.html
15
C X
1
C X
2
C

N
N
N
N
N
N
N
N
N
N
N
XC C

p
p
p
p
p
p
p
p
p
p
p

N
N
N
N
N
N
N
N
N
N
N
C XC
X
1
CX
2
C C

p
p
p
p
p
p
p
p
p
p
p
Figure 15: The abstract opaque shape of the bean.
Observation 9. Every shape has an abstract opaque shape specied by an
algebraic structure, namely a directed graph with dots labeled by abstract partial
views. For example, the abstract opaque shape of the bean is given by the diagram
in Fig.(15).
16
3 Part II. Total View of Transparent Shape from
Far Viewpoint
3.1 Orientation
Denition 17. Given a line L in the plane, an orienter of the line L is a
direction

w anchored at a point x of L, and a second orienter

w

at x

has the
same orientation if

w

=

w. Hence, any line L has exactly two orientations.
An oriented line is an ordered pair (L,

w) where L is a line and



w is the
direction of an orienter of L.
An orienter of the plane R
2
is an ordered pair (

u ,

v ) of orthogonal vec-
tors anchored at the same point. Two plane orienters have the same orientation
if one is a translate and rotation of the other. Hence, the plane has exactly two
orientations.
The equation e
i
+ 1 = 0, with rotator function e
i
of the complex
plane C suggests the natural orienter (1 + 0i, 0 + 1i) = (1, e
i/2
).
L
x
x

v
y

v

Figure 16: Orientation of the line and of the plane.
Henceforth, assume that the plane has the natural orientation.
Denition 18. Given an oriented line (L,

w) the natural orienter of the plane
may be translated and rotated until 1 + 0i coincides with

w. In that case,
0 + 1i ends up lying in exactly one of the two half-planes into which L splits
R
2
. Call that one the positive half-plane of (L,

w), and other its negative
half-plane.
Observation 10. Relative to the natural orientation, each oriented line (L,

w)
induces a partition of the plane, R
2
= H
+
(L,

w) L H

(L,

w) where
H
+
(L,

w) (respectively, H

(L,

w)) is the positive (respectively, negative) half-


plane associated with (L,

w).
For a shape S
1
K
R
2
, the derivative unit tangent vector

t (x) at x = K(e
i
)
is an orienter for the tangent line L(x), hence at each point x of a shape there
17
exists a natural positive half-plane H
+
(x) := H
+
(L(x),

t (x))) and negative


half-plane H

(x) := H

(L(x),

t (x))).
3.2 Inecta and Extrema
Claim 8. Let K be a rounded shape in the naturally oriented plane. For any
angle R and positive innitesimal > 0, if x = K(), b := K({ e
i
| <
}) (before), and a := K({ e
i
| < < + }) (after), then exactly one
of the following holds:
a, b H
+
(x); (26)
a H
+
(x) and b H

(x); (27)
a H

(x) and b H
+
(x); (28)
a, b H

(x). (29)
Denition 19. The point x K is a positive inection of K if a H
+
(x)
and b H

(x), and it is a negative inection if a H

(x) and b H
+
(x).
Otherwise, x is an extremum of K.
Claim 9. (1) For any rounded shape K, either a, b H
+
(x) for all extrema
x, or a, b H

(x) for all extrema x. (2) The inections of K alternate from


positive to negative and back.
Denition 20. For every shape S
1
K
R
2
, e
i
K(e
i
) there exists its re-
verse shape K dened by S
1
K
R
2
, e
i
K(e
i
).
Observation 11. The orientations of tangent lines of K are the reverses of
the orientations of the tangent lines of K, and so H
+
(x) and H

(x) are inter-


changed at all .
Observation 12. Every shape induces a smooth map S
1
S
1

R given by
(, ) := Re(e
i
K()), where is the parameter along K, and is the angle
of a direction. In other words, viewing a shape at direction angle is equivalent
to rotating the shape through angle and viewing it along the imaginary axis.
Then measures the height along the real axis of a point with parameter . The
result is essentially a smooth real-valued function on a 2-dimensional surface
the torus S
1
S
1
for each 1-dimensional shape, which is possibly of interest
to Morse theorists [5]. It seems that these functions are unique for each shape.
But there is the question, exactly which functions on the torus arise in this way
from shapes?
Denition 21. The set of all lines parallel to a given line in the plane is called a
parallel pencil [22]. A far viewpoint

v is the set of all oriented lines parallel


to a given line with the same direction. A line belonging to a far viewpoint is
called a sight-line, and a sight-line paired with one of its two directions is called
a sight-ray.
18
Observation 13. A parallel pencil corresponds to a point at innity. That
is, at innity the lines in a parallel pencil converge to the same point in both
directions. The set of all such points constitute the line at innity, which
together with the lines in the plane constitute the real projective plane [11], and
one might be tempted to adopt the terminology and theory of the real projec-
tive plane in this work on shape. However, that would be misleading, because
the direction of view can make a dierence (by duality see Claim (15) and
Observation (19)) even if along the same sight-lines.
For a rounded shape K, and for each sight-line L of a far viewpoint, Tn(L, K)
is if the sight-line does not meet K, 0 if L meets K but is nowhere tangent
to K, or is a positive integer that is the number of times L is tangent to K.
Denition 22. The set of sight-rays of the far viewpoint determined by di-
rection

u such that 0 < Tn(L, K) < is called the view of K in direction

u .
Observation 14. The contact list of a sight-ray in a view consists of a point
or points other than the rst one which are not directly visible. In this sense
the shape is considered to be transparent to a view.
Claim 10. No matter the shape or view, since the shape is bounded there always
exists at least two sight-lines in the view.
3.3 The Cell Complex Induced by a Far Viewpoint
Observation 15. The sight-lines of a view divide the shape into connected
subsets of the plane. Indeed, there exists a 2-dimensional cell complex [12] whose
1-dimensional skeleton is comprised of segments of sight-lines and portions of
the shape itself. As the direction of the view rotates, the cell complex changes
piecewise continuously. That is to say, 2-cells may change in shape without
change in boundary 1-cells, they may suddenly pop into being, and they may
just as suddenly wink out of existence. The same 2-cell may be the result of
wink-outs in more than one way.
Denition 23. A nite connected 1-dimensional cell complex is called a cycle-
like complex if every 0-cell joins exactly two 1-cells, and is a chain-like com-
plex if two 0-cells meet exactly one 1-cell each, and all other 0-cells join exactly
two 1-cells.
4
Claim 11. Every view of a shape K subdivides it into a cycle-like complex and
determines a dimension-preserving projection of C onto a chain-like complex.
See Fi.g(17).
4
-like must be appended to the terms cycle and chain because these words have
specic very well-known meanings in algebraic topology, and there must be no confusion
introduced by my appropriation of them, which is based merely on visual appearance and my
predilection for short suggestive words.
19
Figure 17: The cycle-to-chain projection of a view of a shape.
Denition 24. For shape K, a far viewpoint

u is a regular viewpoint if
no sight-line of the view of

u meets K at an inection nor at more than one
tangent point. Otherwise,

u is a critical viewpoint.
Claim 12. For every shape K and far viewpoint

u there exists a 2-dimensional
cell complex C
K
(

u ) with underlying set In K K such that


the 0-cells of C
K
(

u ) are the intersections of sight-lines of



u with K; (30)
the 1-cells of C
K
(

u ) are either segments of K, or intervals of intersections


of sight-lines of

u with In K K; (31)
K itself is a cycle-like complex, and is the boundary of C
K
(

u ). (32)
Claim 13. For every shape K there exists a regular viewpoint. There exists at
most a nite number of critical viewpoints. Any non-zero innitesimal variation
of a critical viewpoint is a regular viewpoint. Let M be the maximum number
of 2-cells among the associated cell complexes of the regular viewpoints. For an
arbitrary far viewpoint

u , let i be the number of inection points amongst the
sight-lines of

u , and let n
k
be the number of k-tangents amongst the sight-lines
of

u . Then n
k
= 0 for suciently large k, and the number M
2
(

u ) of 2-cells
in the cell complex associated with

u is
M
2
(

u ) = M 2i

k2
(n
k
1) . (33)
Claim 14. (1) If [
1
,
2
] is an interval of regular view angles, then C
K
(
1
) and
C
K
(
2
) are isomorphic cell complexes. (2) If 0 < R, and is the only
critical view angle in the range from to + , and both endpoints
are regular view angles, then C
K
is a deformation retract of C
K
( +) or vice
versa, and likewise it is a deformation retract of C
K
( ), or vice versa.
3.4 Vertically Ordered Graded Poset
Denition 25. A grading of a nite non-empty poset (P, <) is an order-
preserving surjection of (P, <) upon a totally ordered poset. A structure (P, <
20
, g) such that g is a grading of (P, <) is called a graded poset.
Observation 16. If P
g
X is a grading then it is just as well that X =
[n] := { 1, 2, . . . , n} for n := #X, where the ordering of X is 1 < 2 < < n.
With that understanding, g(x) is a natural number called the grade of x by g.
If the longest chain in (P, <) is of length n, then the Hasse diagram of (P, <)
automatically determines a grading where the least elements of (P, <) have grade
1, and the greatest elements have grade n.
Every nite poset has a trivial grading, namely any function with codomain
a singleton, but in general there are possibly several more interesting gradings.
Denition 26. The ber over i, 1 i n of a grading P
g
[n] is the inverse
image g
1
(i). A vertically ordered graded poset is a structure (P, <, g, )
where is a total ordering of each ber such that if x y in the ber over i,
and x < u, y < v for u, v in the ber over i + 1, then u v:
if x
<

u
y
<

v
then x
<

y
<

v
In the sequel, Vertically ordered graded poset is shortened to VOG poset,
and the whole structure is represented by (P, <, g, ), where (P, <) is the hor-
izontal poset, and for each i, 1 i n the vertically ordered ber at i is
(g
1
(i), ). Every VOG poset (P, <, g, ) has a dual VOG poset (P

, <

, g

) in which (P

, <

) is the reverse poset of (P, <), g

(x) := n g(x), and

is the reverse ordering of on each ber.


Claim 15. VOG posets are the objects of a category VOG in which the mor-
phisms are order-preserving morphisms of horizontal posets which commute with
the grading and preserve the vertical ordering. Moreover, dualization is a (co-
variant) functor
VOG

VOG .
21
A
B
C
D
E
F
G
H
I
J
K
L
(34)
D

e
E

j
J

g
G

B
B
B
B
B
B
B
B
F

k
K

N
N
N
N
N
N
h
H

i
I

p
p
p
p
p
p
A

b
C

c
C

l
L

p
p
p
p
p
p
(35)
A
A
AA
A1A
AA A
AA1A
I I A
2D1b
(36)
DI B
e2F c
E F C
jkl
J K L
gi
GI
h
H
(37)
22
A

P
N
T
M
U
Q
S
R
V
W
(38)
D

r
R

N
N
N
N
N
N
p
P

N
N
N
N
N
N
F

t
T

v
V

o
o
o
o
o
o
n
D

p
p
p
p
p
p
M

Q
Q
Q
Q
Q
Q
A

u
U

w
W

q
Q

{
{
{
{
{
{
{
{
(39)
A


A
1
A


A

A
1
A

I I A

(40)

D
1Ib

I B

F
c

vw
R T U
rtu
G

V W
pq
P Q
n
N
n

M
(41)
23
A

D
E

(42)
D

N
N
N
N
N
N
h

R
R
R
R
R
R
i

y
y
y
y
y
y
y
y
A

p
p
p
p
p
p
(43)
A


A
1
A


A

A
1
A

I I A

(44)

D
1Ib

I B

2
F
c

24
Observation 17. Diagrams (34,38,42) exhibit VOG posets associated with three
views of a smooth shape. The corresponding diagrams (35,39,43) are string
diagram representations constructed by a self-evident algorithm where 2-cells
become wires and 1-cells become boxes. The diagrams (36,40,44) are the
formal algebraic equivalents of the string diagrams.
Observation 18. These representations of VOG posets by diagrams of arrows
and tensor products are inspired by graphical notations encountered in the
context of Higher Category Theory. The Copy, Erase, Merge, and Ini-
tial diagrams associated with biproduct categories in Peter Selingers survey of
graphical languages for monoidal categories seem to t.
5
One may speculate that
the (nite) set of all 2-cells of all cell complexes determined by all far viewpoints
of a rounded smooth shape may be the objects of a biproduct category generated
by VOG poset morphisms.
5
http://www.mscs.dal.ca/
~
selinger/papers/graphical.pdf
25
view
sight-ray
vertically
ordered
graded
poset
moving retraction of 2-cell to 0-cell bitangent sight-ray
VOG poset morphism
moving homeomorphism of 2-cell
view rotation
inflection sightray moving retraction of 2-cell to 0-cell
moving retraction of 2-cell to 1-cell
a
x
b
c
d
e
z
y
a
b
c
d
e
z
a
b
c
d
v
e
u
z
u x
v
y
c
z
w
b
a
b
c
w c
b
Figure 18: For a bean shape, bitangent and inectional critical views in the
middle, with neighboring regular views to the left and the right.
26
Fig.(18) exhibits algebraic and topological structure associated with the
qualitative shape of a bean. Top center is the shape viewed along a sight-
ray meeting the shape at two tangent points a, e. Parallel sight-rays are tangent
at c, z, and intersect the shape at b, d. These rays subdivide the interior of the
shape together with the shape itself into a 2-dimensional cell complex comprised
of 0-cells a, b, c, d, e, z, 1-cells ab, bc, cd, de, ez, za, and 2-cells abz, bcdz, dez. This
view direction is critical, since slight rotation of the view direction clockwise
or counterclockwise dramatically alters the cell complex determined by tan-
gent sight-rays. Counterclockwise, the cell complex on the left consists of 0-
cells a

, x, b

, c

, d

, e

, z

, y, 1-cells a

x, x

, b

, c

, d

, e

, z

y, ya

, and 2-cells
a

xy, xb

y, b

, d

. Clockwise, the cell complex on the right is comprised


of 0-cells a

, b

, c

, d

, v, e

, u, z

, 1-cells a

, b

, c

, d

v, ve

, e

u, uz

, z

,
and 2-cells a

, a

, d

vuz

, ve

u. Similar cell complexes are associ-


ated with the view along an inection sight-ray, bottom center.
The 2-cells of each cell complex in Fig.(18) are the elements of a VOG poset,
indicated by red arrows. At lower right is the trivial 1-element VOG poset, which
represents the fact that the view direction from afar does not see any dierence
between the transparent bean shape and a circle. Then again, at lower left the
view direction from afar upon the transparent bean shape yields a 4 element
VOG poset.
The VOG posets associated with regular views innitesimally clockwise or
counterclockwise from a critical view are related by VOG poset morphisms as
indicated by green arrows.
Moreover, as view direction varies continuously away from a critical view
direction, cells change continuously. 0-cells move along the shape itself, hence
1-cells change within the interior of the shape, and consequently 2-cells change
via a continuous homeomorphism as well. Indeed, cells may come into existence
or disappear, but systematically. That is to say, for example, a 2-cell may retract
continuously to a moving 1-cell or even to a moving 0-cell. Such transformations
are indicated by purple arrows. A purple double arrrowhead signies moving
retraction onto a lower-dimensional cell, and a two-headed arrow is a moving
homeomorphism.
Denition 27. A smooth 2-dimensional variable cell-decomposition of
a space X is a structure (X
0
, X
1
, X
2
) such that X
i
is a nite set of smooth maps
D
i
[0, r]
a
X, and if
X
i
(t) := { D
i
( ,t)
X| a X
i
}, t [0, r], i = 0, 1, 2 (45)
(46)
then (X
0
(t), X
1
(t), X
2
(t)) is a cell-decomposition of X.
Claim 16. If a space X has a 2-dimensional cell-decomposition, then there
exists a category C(X) whose objects are cell-decompositions of X, and a mor-
phism from (U
0
, U
1
, U
2
) to (V
0
, V
1
, V
2
) is a smooth 2-dimensional variable cell-
decomposition (X
0
, X
1
, X
2
) such that X
i
(0) = U
i
and X
i
(r) = V
i
for i = 0, 1, 2.
27
Composition of variable cell-decompositions with intervals [0, r] and [0, s] is a
variable cell-decomposition with interval [0, r +s] by concatenation of maps, and
the identity morphism of (U
0
, U
1
, U
2
) consists of constant maps (cf. [6], p.78).
Observation 19. Given a rounded smooth shape K let X := In K K. Each
view direction from afar upon K considered to be transparent yields a cell-
decomposition of X. As the view direction rotates from 0 to 2, there exists
a nite circular diagram in C(X) in which the dots correspond to critical views
of K, and the arrows are smooth 2-dimensional variable cell-decompositions. As
suggested by Fig.(18), arrows may alternate in direction. Furthermore, for any
view direction, the VOG poset in the exact opposite direction is the dual VOG
poset.
Denition 28. For any rounded smooth shape K the circular diagram in C(X)
is the abstract transparent shape of K.
4 Part III. Approximation and Experimentation
4.1 Bezier Shape
The commercially available Adobe Illustrator application program has a mod-
estly named functionality called the Pen Tool. It is used to place and join a
sequence of anchor points. The rst fact about an anchor point is that it may
be selected and moved to an arbitrary position on the artboard. The amazing
fact about an anchor point is that it is an inection point on the curve joining
the anchor points in their sequence. This is amazing because these inection
points have handles which may be used to rotate, shrink, or expand the size
of the inection. An innite variety of open or (not necessarily simple) closed
curves may be constructed using the Pen Tool, see Fig.(19). Indeed, one may
claim that given any smooth curve and a tolerance not only on the position of
the points of the curve, but on the direction of the derivative at each point of
the curve, there exists a curve constructed by the Pen Tool which matches the
given curve within those tolerances. Complete mathematical programming code
for so-called Bezier curves is freely available.
6
The implication is that a nite
amount of data may approximate arbitrarily closely any given smooth curve.
There is a disturbing question of whether it makes sense to ask for the group
of transformations which preserve qualitative shape of Bezier shapes.
6
http://processingjs.nihongoresources.com/bezierinfo/
28
Figure 19: Closed curve constructed with the Pen Tool in Adobe Illustrator.
Some anchor points and handles are shown. (All the smooth shapes in this
proposal were created with Adobe Illustrator.)
29
4.2 Screenshots of Scultoro
A rst theorem in elementary Morse Theory is about a smooth real-valued map
on a smooth shape (a.k.a., smooth compact manifold). If the extrema of the
map are non-degenerate, that is, not like inecta, then the shape may be re-
constructed as a cell complex step-by-step from knowledge of the critical values
of the map [5].
Observation 20. Given a far viewpoint upon a transparent shape, suppose the
sight-ray through the center-of-gravity of the shape is singled out. With respect
to the standard orientation of the plane there exists a directed line orthogo-
nal to that sight-ray also through the center-of-gravity. These orthogonal rays
constitute a rectangular coordinate system, with sight-ray as abscissa and its
orthogonal ray as ordinate. Since the shape is actually a smooth map on the
circle to the plane, each image of a point on the circle determines a height from
the chosen sight-ray along the orthogonal. Hence, for each far viewpoint there
exists a smooth real-valued map on the shape. Since far viewpoints correspond
also to points on the circle, we have dened for the given shape a smooth map
on S
1
S
1
(cf. Observation (12)). One thing to do with this map, which I
claim is determined by the shape, and up to the choices made in its construction
determines the shape, is to apply Morse Theory to it. Another thing to do with
it is to graph it. Fig.(20) is a screenshot of my Mathematica program called
Scultoro which calculates and displays the map in this case for a circle shape.
One axis represents view direction from 0 to 2, a second axis represents the
parameter of the shape, and the third axis is the calculated height. As one would
expect, the graph is a sinusoidal wave.
Figure 20: The built-in circle shape invoked by the CRCL button.
Scultoro has a virtual button which enables a blue dot to behave like a
hammer, so that if the real mouse button is pressed while over the dot, then the
shape nearby becomes dented, as though struck with a hammer. The strength
of the blow is greater as the blue dot is moved closer to the shape. Fig.(21)
30
shows how the circle becomes a bean when it is banged by the hammer, and
the resulting crinkle in the wave is exhibited on the right. A more complex
Figure 21: Toggling the TRAK button to BANG enables the blue hammer to
bang the circle into a bean shape.
dented shape and its wave is exhibited in Fig.(22). The TRAK button converts
Figure 22: The built-in dented shape banged from a circle is invoked by the
DNTD button.
the mouse from a hammer banger to a free-hand drawing tool, as illustrated by
Fig.(23). Comparison of constructions based on a far viewpoint of the trans-
parent dented shape, and on a near viewpoint of the opaque dented shape, are
enabled by the DRWG button of Scultoro, as shown in Fig.(24).
31
Figure 23: Toggle to the TRAK button for free-hand shape drawing.
5 Part IV. An Imperfect Analogy: Knots and
Shapes
Aside. At the Joint Meeting of the AMS and MAA in Boston, January, 2012,
I had a conversation with magister Barry Mazur after his talk on mathemat-
ical plausibility, which emphasized examples from history, including multiple
references to Euler. With a seemingly false analogy between Knot Theory and
Shape Theory in the back of my mind, after his talk I waited my turn and said
to him, You spoke of analogy guiding mathematical inquiry. But what if the
analogy turns out to be a false analogy? Mazur: Then there is something
much deeper going on. For example Kroneckers liebster Jugendtraum. Pause.
Me: Did you say Eulers ... what? Mazur looking down his nose at me even
though I am taller than he said, No, Kronecker. His Jugendtraum. Dont
you know?
7
My personal liebster Altertraum (dearest old-age dream) is to relate ab-
stract algebraic structure of far views of a transparent shape to abstract alge-
braic structure of near views of the opaque shape, and thus better to understand
qualitative shape.
Fortunately, Barry Mazur distributed copies of rough notes for his talk,
Why is it plausible?.
But reasoning by analogy is, I think, the keystone: it is present
in much (perhaps all) daily mathematical thought, and is also of-
7
Triggered by an analogy articulated by David Mumford, Barry Mazur developed new ideas
relating Knot Theory (specically, the Alexander polynomial of a knot) to Number Theory.
See http://www.math.harvard.edu/
~
mazur/papers/alexander_polynomial.pdf.
32
Figure 24: At the left are sight-lines of an innitely far viewpoint, and a cycle-
like complex calculated for the freehand transparent shape. At the right is the
partial view of the opaque shape from a near viewpoint.
ten the inspiration behind some of the major long-range projects in
mathematics. And, Andre Weil was right when he said: nothing
also gives more pleasure to the researcher.
I add that the role of analogy in physics is equally important [23]. Mazur
mentions, among the grand analogies, the Langlands Program in Number The-
ory, and that [t]he analogy between number elds and function-elds of one
variable over nite elds is a more elementary, and older example. In a foot-
note, he writes as with all great analogies, its imperfections sparkle, raising
questions that may lead to future theories, far deeper than the ones we cur-
rently are at home with.
To suggest the inestimable importance of Knot Theory in mathematics and
physics I paraphrase the already brief history of Knot Theory recounted by
Michael Atiyah in [1]. Physicist Lord Kelvin in 1867 published the idea that
atoms are actually knotted vortex tubes of ether, based on the observations
that such ethereal knots would be stable, would come in many varieties the
chemical elements and would vibrate in ways to account for experimentally
measured spectral lines. In modern times, transmutation of elements would
correspond to cutting and recombination of knots. The idea was taken seriously
for a while, giving rise to long lists knot tables of knots represented by
diagrams.
But the idea was soon discarded by physicists and the study of knots became
merely an esoteric branch of pure mathematics, with some eorts devoted to
proving abstract conjectures about the classication of knots. Early on it was
realized that a knot, dened as a specic kind of subset of three-dimensional
33
space, might be deformed as though made of exible wire yet retain its
essentialqualitative nature even after deformation, so long as no portion of it is
forced to pass through another portion during the deformation. In other words,
it is not a knot per se that is the proper object of study, but its equivalence
class up-to-deformation.
In 1928 James W. Alexander published the rst algebraic structure asso-
ciated with each knot class, and in 1984 Vaughan F. R. Jones constructed a
polynomial for each knot class. Nothing is more algebraical than a polynomial,
and several polynomials have subsequently been invented for each knot class
[14][20]. Generally, the mathematical objective is to dene an algebraic struc-
ture for each knot class which uniquely identies the knot class: every knot class
should correspond to exactly one distinct structure. Such a structure should not
vary even if a knot class representative is deformed in space, so is called an in-
variant of the knot. The Alexander algebraic structure, the Jones polynomial,
and their children, are called knot invariants. It is a separate question to nd
means of accurately representing knots, which are intrinsically subsets of 3-
dimensional space, in terms of notations, called knot presentations, on the
2-dimensional plane of writing paper.
8
The surprise is that knots have re-appeared big-time in theoretical physics.
So much so, that already by 1989 there appeared a one thousand page collection
of papers on new developments in the theory of knots published not as a work
of mathematics, but in an advanced series on mathematical physics [17]. Indeed,
Knot Theory with related concepts, such as links, braids, and tangles is
entwined in a truly grand analogy between Category Theory, physics, topology,
and logic [2].
Aside. Perhaps I have knot envy analogous to the erstwhile physics envy
of social sciences since I recently asked myself whether knots in space are
intrinsically richer mathematical structures than smooth shapes in the plane,
and that therefore I should not harbor hopes that studying such shapes should
give rise to a particularly interesting theory. I console myself with the rejoinder
that mathematical riches arise from the invention of questions, and from the
diculty of answering the questions, and from how surprising, clever, or far-
reaching might be the answers. Simply, do the questions yield to breathtaking
feats of calculation and reasoning?
5.1 Homotopy, Isotopy, and Ambient Isotopy
A central idea in algebraic topology is continuous reshaping. In this con-
text a continuous map X
f
Y consists of an abstract shape X which via f
parametrizes the actual shape f(X) inside the ambient space Y . For example
in this work, the circle S
1
is the abstract shape, R
2
is the ambient space, and
a smooth shape is a smooth embedding S
1
K
R
2
. In the case of knots, S
1
is
8
Specication of the requirements for knot presentations are exactly stated, and several
historically important examples are reviewed, in Knots as processes: a new kind of invariant,
L. G. Meredith and David F. Snyder, http://128.84.158.119/abs/1009.2107.
34
also the abstract shape, but a knot is a continuous map S
1
K
R
3
. A three-
dimensional smooth shape would be a smooth embedding S
2
K
R
3
, where the
abstract shape is a 2-dimensional sphere, and so on. An abstract shape realized
in two possibly dierent actual shapes is modeled by a pair of maps,
X
f

Y (47)
Aside. The idea of simple things like the unit circle serving as a template
or prototype for shapes in an ambient space permeates algebraic topology. For
example, we have simple things like points, segments, triangles, and higher
dimensional triangles as abstract shapes mapping into spaces, and we have al-
gebraic tools such as simplicial homology calculated from these maps to study
properties of the ambient space. F. W. Lawvere discusses this idea in [18] with
the language X-shaped gure in Y , and he writes, An ancient principle of
mathematics holds that a gure is a locus of a varying element. (See Observa-
tion (3).)
On the other side, physics is shot through with maps from a space to a
relatively simpler structure such as the real or complex numbers. Namely,
measurements assign numbers to points of space, where space may, for example,
represent states of a physical system. A modern elaboration of this idea is
provided in the context of a proposal to radically revise the theory of quantum
gravity:
Our basic contention is that constructing a theory of physics is
equivalent to nding a representation in a topos of a certain formal
language that is attached to the system. Classical physics uses the
topos of sets. Other theories involve a dierent topos. For the types
of theory discussed in this proposal, a key goal is to represent any
physical quantity A with an arrow A

where

and
R

are two special objects (the state-object and quantity-value


object) in the appropriate topos,

[9].
Back down to Earth, Morse Theory begins with study of real-valued maps
dened on a shape, and in this proposal each shape determines such a map, as
discussed in Observation (20) about Scultoro.
The technical word for continuous reshaping is homotopy.
Denition 29. A homotopy from f to g in diagram (47) is a new diagram of
continuous maps of the form
X
( 1X 0 )
.t
t
t
t
f

A
A
A
X I
H

Y
X
( 1X 1 )
J
J
J
J
g

}
}
}
(48)
35
and this is abbreviated to the diagram
X
f

H
Y . (49)
If there exists at least one homotopy between coterminous maps f, g then
they are called homotopic maps. All the maps between two spaces are parti-
tioned into sets of mutually homotopic maps. In other words, homotopic to
is an equivalence relation on coterminous maps [6].
In a contractible space any two shapes are homotopic [12]. In particular,
R
2
or R
3
are contractible. In them a homotopy can shrink f(X) to a point
and then expand the point to g(X). In this way a homotopy can continuously
change any shape into any shape. So, homotopy is not a good candidate for
continuous re-shaping while preserving qualitative shape. A more restrictive
condition on a homotopy is that at all times t I the image of X in Y should
be homeomorphic to X.
Denition 30. A homotopy (49) is an isotopy if H ( 1
X
t ) is a homeomor-
phism at all times t I. Maps between two spaces fall into distinct mutually
isotopic sets, but these sets are more exclusive than the homotopy classes of
maps.
This is going in the right direction, but it has been observed that any knot
is isotopic to the circle. For, given a knot in space, pull a part of it outward to
the shape of a very large circular arc, and then continuously shrink the rest of
the knot down towards a single point. [I]t is easy to see that every suciently
smooth simple, closed curve can be deformed continuously into a circle without
ever acquiring a multiple point, even though the curve may be one which ought
to be regarded as knotted. In gure 1, for example, we have a series of self-
explanatory diagrams showing a trefoil knot in the process of being deformed
into a circle. It will be noticed that the curve acquires no multiple point even
at the moment it becomes a circle.
9
An even more restrictive idea is not that one shape is deformed to another
inside space, but that the whole space is deformed and thereby re-shapes one
shape to another.
9
J. W. Alenxander, Some problems in topology, http://www.mathunion.org/ICM/
ICM1932.1/Main/icm1932.1.0249.0257.ocr.pdf .
36
Denition 31. An ambient isotopy of f to g in (47) is a diagram pair
X
1Y

H
Y X
f
.~
~
~
~
~
~
~
g

@
@
@
@
@
@
@
Y
L

Y
(50)
where now H is an isotopy of the identity map of Y to the spatial homeomor-
phism L, and g = L f. Ambient isotopy is also an equivalence relation on
coterminous maps. Also, given an ambient isotopy (50), there exists an isotopy
H ( g 1
I
) between f and g.
The point of recapitulating these technical denitions is merely that Knot
Theory has in ambient isotopy a mature denition of what it means for two
knots to be qualitatively the same.
37
5.2 Analogy Table with Imperfections
Knot Shape
Analogy
(1) A knot is a map that embeds
the circle into the space.
A (2D smooth) shape is a
smooth map that embeds the
circle into the plane.
(2) A knot diagram is the image of
a regular knot projection of the
space upon the plane,
annotated with an ordered pair
of points of the circle to
indicate which is over and
which is under in the
diagram.
The dimension preserving
cellular map induced by a
regular far viewpoint of a
transparent shape totally
orders the 1-cells that project
upon the same 1-cell on the
line.
(3) At a crossing of a knot diagram
the inverse image has exactly 2
points.
Over a 1-cell on the line there
is an even number of 1-cells in
the inverse image.
(4) If a suciently small circle
surrounds each crossing, then
each circle intersects the knot
diagram in 2 2 points.
If a suciently small interval
surrounds an interior point of a
1-cell in the projection, then its
inverse image consists of n 2
intervals.
(5) The set of 2 2 small segments
at intersections are wired to
form the whole knot.
The set of n 2 small segments
corresponding to the choice of
1 interior point in each 1-cell of
the projection are connected to
form the whole shape.
(6) For every knot projection there
exists an innitely nearby
regular knot projection ([20],
p.22).
For every far viewpoint there
exists an innitely nearby
regular viewpoint.
(7) Every knot projection innitely
near to a regular knot
projection is a regular knot
projection.
Every far viewpoint innitely
near to a regular far viewpoint
is a regular far viewpoint.
38
Knot Shape
Imperfection
(8) A Reidemeister move of a knot
diagram is a local sliding
that is equivalent to a local
rotational distortion of the
space containing the knot
above the plane.
Certain moves of the 1-cells in
the projection corresponding to
a far viewpoint also correspond
to a change in the far
viewpoint, but not to a change
in the shape.
(9) Cutting a knot at two distinct
points yields two endpoints a, b
of one connected part, and two
endpoints c, d of the other.
Joining a to b and c to d yields
two new knots. Conversely,
given two knots, a new knot is
composed from them by
cutting one to yield endpoints
a, b and cutting the other to
yield endpoints c, d, then
joining a to c and b to d.
Cutting across a shape along a
non-tangential sight-line yields
two (not necessarily connected)
parts with multiple endpoints
at specic locations in the
plane together with specic
tangent directions at each
endpoint. These endpoints
may be paired o and
connected smoothly by
end-caps in at least one way
ultimately to yield two new
smooth shapes. Conversely,
two suitable gures in the
plane with matching data
involving locations and tangent
directions could be re-arranged
to join up and compose a new
smooth shape (cf. composition
of 1-dimensional tangles in ([2],
p.9)).
(10)
Knot Theory has a plethora of
combinatorial-algebraic
invariants or near-invariants of
knots-up-to-ambient-isotopy.
Although there exist algebraic
constructions associated with a
smooth shape, there does not
exist an independent denition
of qualitative equivalence of
shape.
6 Part V. Conclusion and Problems
Abstract algebraic structures are associated with near views of opaque shapes
and with far views of transparent shapes. There exist references to relevant
literature on phenomenology, psychology, computer vision, mathematics, and
mathematical physics. Formal denitions of concepts relating to shapes in the
plane are articulated, and several claims are made, with drawings and diagrams
oered in evidence.
One problem is to convert these claims and that evidence into theorems and
39
proofs. Another is to answer questions such as how to determine the minimum
number of near views that are adequate to see the entirety of a plane shape.
Observation (18) alludes to a biproduct category. If there exists a biproduct
category in which certain diagrams of chains of morphisms correspond to VOG
posets, then a problem is to translate the circle of VOG poset morphisms cor-
responding to the far views of a transparent shape into a circle of morphisms of
such chains.
If there exists a category of aspect graphs corresponding to the near views
of opaque shapes, and if there exists a category of circular diagrams of VOG
posets corresponding to the far views of transparent shapes, then there is the
problem of constructing a pair of adjoint functors between these categories. This
adjoint pair might answer the question implicit in the Abstract, which is, How
to formalize the inverse experiences going from mental model to partial views,
and back?.
As knots are included among links, which are multiple knots in space, so
shapes in the plane may collect into systems of shapes, like the British Isles.
Also, a shape may have more than one hole, so that its boundary consists of more
than one connected 1-dimensional complex. Then again, shapes may be nested,
as islands in lakes on islands in oceans. All these cases require generalizations
of constructions in this proposal based on near and far views of single opaque
and transparent shapes.
The most important problem is to solve the analogy, qualitative shape is
to shape as ambient isotopy is to knot for the unknown, qualitative shape.
Another way of putting this, is to observe that I have constructed candidates
(aspect graphs, VOG posets) for invariants, but there remains the question,
what is it that varies?
Perhaps a way forward is to create the category of Bezier shapes, in which
the objects are smooth shapes in the plane constructed by joining some nite
number of anchor points using the pen tool. The morphisms are nite sequences
of operations of adding or deleting anchor points, or adjustments of handles at
inecta. Maybe this Bezier category is to shapes in the plane as the category
of cell complexes is to the category of topological spaces, and is the ground in
which to grow the concept of qualitative shape, then to pick the low hanging
fruit.
Beyond the scope of this proposal is the project of generalizing from shapes
in the plane to shapes in space like sculptures, shapes which if sliced by planes
can sometimes lead back to smooth shapes in the plane.
40
References
[1] Michael Atiyah, The Geometry and Physics of Knots, Cambridge Univer-
sity Press, 1990.
[2] John C. Baez and Mike Stay, Physics, Topology, Logic and Computation:
A Rosetta Stone, March 2009.
[3] Nestor Bertoglio and Rolando Chuaqui, An elementary geometric nonstan-
dard proof of the Jordan curve theorem, Geometriae Dedicata 51 (1994),
no. 1, 1527.
[4] S. Biasotti, L. De Floriani, B. Falcidieno, P. Frosini, D. Giorgi, C. Landi,
L. Papaleo, and M. Spagnuolo, Describing shapes by geometrical-topological
properties of real functions, ACM Computing Surveys 40 (2008), no. 4.
[5] Raoul Bott, Morse theory indomitable, Publications Mathematiques de
LH.

E.S 68 (1988), 99114, Fig. (3) exhibits critical points of a function


dened on the circle in terms of a bean diagram.
[6] Ronald Brown, Topology and Groupoids, www.booksurge.com, 2006.
[7] Eliseo Clementini and Paolino Di Felice, A global framework for qualitative
shape description, GeoInformatica 1 (1997), 1127.
[8] Ellis D. Cooper, Point-humans: optical perception by people and computers,
Interface Age 2 (1977), no. 12.
[9] Andreas Doring and Chris Isham, What is a Thing?: Topos theory in
the foundations of physics, (2008), arXiv:0803.0417v1 [quant-ph] 4 March
2008.
[10] Rieger J. W. Fendrich and H-J Heinze, The contribution of eye movements
to anorthoscopic percepts under free-viewing conditions, Perception, vol. 33,
27th European Conference on Visual Perception, 2004, Abstract Supple-
ment.
[11] F.R.S. H. S. M. Coxeter, The Real Projective Plane, 2 ed., Cambridge At
The University Press, 1961.
[12] Allen Hatcher, Algebraic Topology, Cambridge University Press, 2002.
[13] Edmund Husserl, Thing and Space: Lectures of 1907, Kluwer Academic
Publishers, 2010, Translated by Richard Rojcewicz.
[14] Louis H. Kauman, On Knots, Princeton University Press, 1987.
[15] J. J. Koenderink and A. J. van Doorn, The shape of smooth objects and the
way contours end, Perception 11 (1982), 129137.
41
[16] Jan J. Koenderink, Two- and Three-Dimensional Patterns of the Face, The
Mathematical Intelligencer 22 (2000), no. 3, Review of book authored by
Peter W. Halliman, Gaile G. Gordon, A. L Yuille, Peter Giblin, and David
Mumford published by A. K. Peters, 1999.
[17] Toshitake Kohno (ed.), New Developments in the Theory of Knots, Ad-
vanced Series in Mathematical Physics, vol. 11, World Scientic, 1989.
[18] F. William Lawvere and Stephen Schanuel, Conceptual Mathematics, Cam-
bridge University Press, 1997.
[19] Michael Leyton, A Generative Theory of Shape, Springer-Verlag, 2001.
[20] Charles Livingston, Knot Theory, The Mathematical Association of Amer-
ica, 1993.
[21] Saunders Mac Lane, Categories for the Working Mathematician, Springer-
Verlag, New York, 1971.
[22] George E. Martin, The Foundations of Geometry and the Non-Euclidean
Plane, Springer-Verlag, 1975.
[23] Ciara Aisling Muldoon, Shall I Compare Thee to a Pressure Wave? Vi-
sualisation, Analogy, Insight and Communication in Physics, Doctor of
Philosophy, University of Bath, May 2006.
[24] Louis Narens, A nonstandard proof of the Jordan curve theorem, Pacic
Journal of Mathematics 36 (1971), no. 1, 219229.
[25] Lawrence Nestout, Two discrete forms of the Jordan curve theorem, Amer-
ican Mathematical Monthly 95 (1988), no. 4, 332336.
[26] N. Ahuja C. Dyer A. Pentland R. Jain O. Faugeras, J. Mundy and
K. Ikeuchi, Why Aspect Graphs are Not (Yet) Practical for Computer Vi-
sion, Workshop panel report, University of South Florida, Department of
Computer Science and Electrical Engineering, 1992.
[27] Robert D. Schienbaur, A Survey of Aspect Graphs, Technical Report TR-
CIS-2001-01, Polytechnic University, Department of Computer and Infor-
mation Science, 2001.
[28] Giandomenico Sica (ed.), What is Category Theory?, Polimetrica, Monza,
Italy, 2006.
[29] Hassler Whitney, On regular closed curves in the plane, Compositio Mathe-
matica 4 (1937), 276284, http://archive.numdam.org/ARCHIVE/CM/CM_
1937__4_/CM_1937__4__276_0/CM_1937__4__276_0.pdf.
[30] , Geometric Iintegration Theory, Princeton University Press, 1957,
Appendix I. Section 10, Ane spaces.
42
[31] Freek Wiedijk, Formal proof - getting started, Notices of the American
Mathematical Society 55 (2008), no. 11, 14081414, The Jordan Curve
Theorem was formally proved twice by independent researchers in 2005.
43
Index
liebster Altertraum, 32
adjoint pair, 40
ambient isotopy, 37
analogy, 32, 34, 40
false, 32
anorthoscopic perception, 0, 3
Bezier
category, 40
curve, 28
shape, 28, 40
category
aspect graph, 40
Bezier, 40
biproduct, 25, 40
cell-decomposition, 27
circular diagram, 40
located vector, 5
path, 8
vertically ordered graded poset, 21
Category Theory, 0
cell complex, 20
complex
1-dimensional, 40
cell, 19, 20, 25, 27, 30, 40
chain-like, 19
cycle-like, 19
number, 35
plane, 6, 17
computer vision, 3
contact
rst point, 10
list, 9, 19
continuous synthesis, 2
cross, 9
curve, 7
closed, 7
simple, 7
cyclic permutation, 11
deformation retract, 20
derivative tangent vector, 6
dimension preserving cellular map, 38
drops
back, 10
to innity, 10
embedding, 6
Erlanger Program, 2
Euclidean Geometry, 2
extremum, 18
ber, 21
graded poset, 21
grading, 20
group of transformations, 28
Hasse diagram, 21
homeomorphism, 2
homotopy, 35
horizontal poset, 21
inection
positive (negative), 18
invariant, 34
isotopy, 0, 36
Jordan Curve Theorem, 7
knot
class, 34
classication, 33
invariant, 34
presentation, 34
table, 33
Knot Theory, 0
Langlands Program, 33
line at innity, 19
links, 40
mental
model, 40
process, 3
Morse Theory, 0
44
optical character recognition, 3
oriented line, 17
positive (negative) half-plane, 17
orienter
of a line, 17
of the plane, 17
natural, 17
parallel pencil, 18
parametrization, 7
partial view, 4, 811, 40
abstract, 11, 12, 16
the bean, 15
integration, 3
perception, 2
phenomenology, 0, 2
plane, 4
point, 4
vector, 4
point
interval, 5
translate by a vector, 4
point at innity, 19
pointhuman, 4
poset
VOG, 25
psychology of perception, 3
qualitative change, 2
qualitative shape, 0
quantum gravity, 35
ray, 5
direction, 5
vertex, 5
real projective plane, 19
rigid isometry, 2
shape, 6, 11
abstract opaque, 16
abstract transparent, 28
aspect, 12
component, 12
contiguous, 12
aspect graph, 14
bean, 14
generative theory, 3
partial view of, 3
qualitative, 0, 2, 32
qualitative invariant, 4
reverse, 18
rigid, 2
rounded, 14
transparent, 19
unit tangent vector, 17
sight-
interval, 8, 9, 11
line, 18, 20, 39
ray, 18, 19, 27, 30
Theory
Category, 34
Higher Category, 25
Knot, 3234, 37, 39
Morse, 3, 15, 18, 30, 35
Number, 32, 33
Shape, 32
topological space, 2
total perception, 3
unit circle, 6
unity of experience, 2
variable cell-decomposition, 27
vector
located, 5
vertically ordered graded (VOG) poset,
21
view, 19
viewpoint
abstractly equivalent near, 11
exterior, 4
far, 18, 20, 25, 30, 38
near, 811
regular, 38
regular (critical), 20
visual
event, 11
perception, 3
45

Potrebbero piacerti anche