Title: Domain%20Theoretic%20Models%20of%20Geometry%20and%20Differential%20Calculus
1Domain Theoretic Models of Geometry and
Differential Calculus
Abbas Edalat Imperial College http//www.doc.i
c.ac.uk/ae
Joint work with Andre Lieutier
Dassault Systemes
With contributions from Ali Khanban, Marko
Krznaric
ISDT 2001 Chengdu, China
2Computational Model for Classical Spaces
- A research project since 1993
- Reconstruct basic mathematics!
- Embed classical spaces into the set of maximal
elements of suitable domains
3Computational Model for Classical Spaces
- Previous Applications
- Fractal Geometry
- New algorithms for computing fractal objects
- Measure Integration Theory
- The Generalized Riemann Integral
- Exact Real Arithmetic
- C-library based on linear fractional
transformations for Computation of Elementary
Functions
4Part 1 A Domain-Theoretic Model of Geometry
- To develop a Computable model for Geometry and
Solid Modelling, so that
- the model is mathematically sound, realistic
- the basic building blocks are computable
- it bridges theory and practice.
5Why do we need a data type for solids?
- Answer To develop robust algorithms!
- Lack of a proper data type and use of real RAM
in which comparison of real numbers is decidable
give unreliable programs in practice!
6The Intersection of two lines
- With floating point arithmetic, find the point P
of the intersection L1 ? L2. Then
min_dist(P, L1) gt 0, min_dist(P,
L2) gt 0.
7The Convex Hull Algorithm
With floating point we can get
8The Convex Hull Algorithm
A, B C nearly collinear
With floating point we can get (i) AC,
or
9The Convex Hull Algorithm
A, B C nearly collinear
With floating point we can get (i) AC,
or (ii) just AB, or
10The Convex Hull Algorithm
A, B C nearly collinear
With floating point we can get (i) AC,
or (ii) just AB, or (iii) just BC, or
11The Convex Hull Algorithm
A, B C nearly collinear
With floating point we can get (i) AC,
or (ii) just AB, or (iii) just BC, or (iv) none
of them.
The quest for robust algorithms is the most
fundamental unresolved problem in solid modelling
and computational geometry.
12A Fundamental Problem in Topology and Geometry
- Subset A ? X topological space.Membership
predicate ?A X ? tt, ff - is continuous iff A is both open and closed.
- In particular, for A ? Rn, A ? ?, A ? Rn ?A
Rn ? tt, ff is not continuous. - Most engineering is done, however, in Rn.
13Non-computability of the Membership Predicate
- There is discontinuity at the boundary of the
set.
False
True
14Non-computable Operations in Classical CG SM
- ?A Rn ? tt, ff not continuous means it is not
computable, even for simple objects like
A0,1n. - x ? A is not decidable even for simple objects
for A 0,?) ? R, we just have the
undecidability of x ? 0. - The Boolean operation ? is not continuous, hence
noncomputable, wrt the natural notion of topology
on subsets? C(Rn) ? C(Rn) ? C(Rn), where
C(Rn) is compact subsets with the Hausdorff
metric.
15Intersection of two 3D cubes
16Intersection of two 3D cubes
17Intersection of two 3D cubes
18This is Really Ironical!
- Topology and geometry have been developed to
study continuous functions and transformations on
spaces. - The membership predicate and the binary operation
for ? are the fundamental building blocks of
topology and geometry. - Yet, these fundamental functions are not
continuous in classical topology and geometry.
19Elements of a Computable Topology/Geometry
- The membership predicate ?A X ? tt, ff fails
to be continuous on ?A, the boundary of A. - For any open or closed set A, the predicate x ?
?A is non-observable, like x 0.
- ?A is now a continuous function.
20Elements of a Computable Topology/Geometry
- Note that ?A?B iff int Aint B int
Acint Bc, i.e. sets with the same
interior and exterior have the same membership
predicate. - We now change our view In analogy with classical
set theory where every set is completely
determined by its membership predicate, we define
a (partial) solid object to be given by any
continuous map - f X ? tt, ff ?
- Thenf 1tt is open its called the interior
of the object. f 1ff is open its called
the exterior of the object.
21Partial Solid Objects
- We have now introduced partial solid objects,
since X \ (f 1tt ? f
1ff)
may have non-empty interior. - We partially order the continuous functionsf, g
X ? tt, ff ? f ? g ? ?x ? X . f(x) ?
g(x) - f ? g ? f 1tt ? g 1tt f 1ff
? g 1ffTherefore, f ? g means g has more
information about an idealized real solid object.
22The Geometric (Solid) Domain of X
- The geometric (solid) domain S (X) of X is the
poset (X ? tt, ff ?, ? ) - S(X) is isomorphic to the poset SO(X) of pairs of
disjoint open sets (O1,O2) ordered componentwise
by inclusion
23Properties of the Geometric (Solid) Domain
- Theorem For a second countable locally compact
Hausdorff space X (e.g. Rn), S(X) is bounded
complete and ?continuous with (U1, U2) ltlt (V1,
V2) iff the closures of U1 and U2 are compact
subsets of V1 and V2 respectively.
24Examples
- A x?R2 ? x 1 ? 1, 2represented in the
model by(int A, int Ac) ( x ? x lt 1, R2 \
A )is a classical (but non-regular) solid
object.
25Boolean operations and predicates
- Theorem All these operations are Scott
continuous and preserve classical solid objects. -
26Subset Inclusion
- Subset inclusion is Scott continuous.
27General Minkowski operator
- For smoothing out sharp corners of objects.
- SbRn (A, B) ? SRn Bc is bounded ?(Ø,Ø).
- All real solids are represented in SbRn.
- Define _?_ SRn ? SbRn ? SRn
((A,B) , (C,D)) ? (A ? C , (Bc ? Dc)c) where
A ? C ac a? A, c? C - Theorem _?_ is Scott continuous.
28An effectively given solid domain
- The geometric domain SX can be given effective
structure for any locally compact second
countable Hausdorff space, e.g. Rn, Sn, Tn,
0,1n. - Consider XRn. The set of pairs of disjoint open
rational polyhedra of the form K (L1 , L2) ,
with L1 ? L2 ?, gives a basis for SX. - Let Kn (p1 ( K n ) , p2 ( K n) ) be an
enumeration of this basis. -
- (A, B) is a computable partial solid object if
there exists a total recursive function ßN?N
such that -
-
(A , B) ( ?n p1 ( K ß(n) ) , ?n p2 (
K ß(n) ) )
29Computing a Solid Object
- In this model, a solid object is represented by
its interior and exterior.
-
- The interior and the exterior
- are approximated by two
- nested sequence of rational polyhedra.
30Computable Operations on the Solid Domain
- F (SX)n ? SX or F (SX)n ? tt,
ff ? - is computable if it takes computable sequences
of partial solid objects to computable sequences. - Theorem All the basic Boolean operations and
predicates are computable wrt any effective
enumeration of either the partial rational
polyhedra or the partial dyadic voxel sets.
31Quantative Measure of Convergence
- In our present model for computable solids,
there is no quantitative measure for the
convergence of the basis elements to a computable
solid. - Example The minimum distance from say a
rational point to a computable solid is
semi-computable, but not computable. - We will enrich the notion of domain-theoretic
computability to include a quantitative measure
of convergence.
32Hausdorff Computability
- We strengthen the notion of a computable solid by
using the Hausdorff distance d between compact
sets in Rn. - d(C,D) min r C ? Dr D ? Cr
where Dr x ? y ? D.
x-y ? r -
33Hausdorff computability
- Two solid objects which have a small Hausdorff
distance from each other are visually close. - The Hausdorff distance gives a natural
quantitative measure for approximation of solid
objects. - However, the intersection or union of two
Hausdorff computable solid objects may fail to be
Hausdorff computable. - Examples of such failure are nontrivial to
construct.
34Boolean Intersection is not Hausdorff computable
is Hausdorff computable.
However Q?(0,1 ? 0) r,1 ? 0 ? R2is
not Hausdorffcomputable.
35Lebesgue Computability
- (A , B) ? S k, kd is Lebesgue computable iff
there exists an effective chain K ß(n) of basis
elements with ß N?N a total recursive
functions such that - (A , B) ( ?n p1 ( K ß(n) ) , ?n
p2 ( K ß(n) ) ) - µ(A) - µ(p1 ( K ß(n) ) ) lt 1/2 n µ(B)
- µ(p2 ( K ß(n) ) ) lt 1/2 n - A function is Lebesgue computable if it
preserves Lebesgue computable sequences. - Theorem Boolean operations are Lebesgue
computable. -
36Hausdorff and Lebesgue computability
- Hausdorff computable ? Lebesgue
computableComplement of a Cantor set with
Lebesgue measure 1 r with r lim rn left
computable but non-computable real.
- At stage n remove 2n open mid-intervals of length
sn/2n.
37Hausdorff and Lebesgue computability
- Lebesgue computable ? Hausdorff computable
- Let 0 lt rn ? Q with rn ? r, left
computable, non-computable 0 lt r lt 1.
38Hausdorff and Lebesgue Computable Objects
- Hausdorff computable ? Lebesgue computable
- Lebesgue computable ? Hausdorff computable
- Theorem A regular solid object is computable
iff it is Hausdorff computable. - However A computable regular solid object may
not be Lebesgue computable.
39Conclusion
- Our model satisfies
- A well-defined notion of computability
- Reflects the observable properties of geometric
objects - Is closed under basic operations
- Captures regular and non-regular sets
- Supports a methodology for designing robust
algorithms
40Part 2 Data-types for Computational Geometry
and Systems of Linear Equations
- Voronoi Diagram or the Post Office problem
- The Partial Circle through three partial points
- Partial Lines and Projective Geometry
the intersection of hyper-planes,
the solution of systems of
linear equations
41The Outer Convex Hull Algorithm
42The Inner Convex Hull Algorithm
43The Convex Hull Algorithm
44The Convex Hull Algorithm
45The Convex Hull map
- Let Hm (R2)m ? C(R2) be the classical convex
Hull map, with C(R2) the set of compact subsets
of R2, with the Hausdorff metric. - Let (IR2, ? ) be the domain of rectangles in R2.
- For x(T1,T2,,Tm)?(IR2)m, define
- Cm (IR2)m ? SR2,Cm(x)(Im(x),Em(x))
with - Em(x)?Hm(y) y?(R2)m, yi?Ti, 1 ? I ? m
- Im(x) ?Hm(y) y?(R2)m, yi?Ti, 1 ? I ? m
46The Convex Hull is Computable!
- Proposition Em(x)(H4m((Ti1,Ti2,Ti3,Ti4))1?i?m)c
Im(x)Int(?Hm((Tin))1?i?m)
n1,2,3,4). - Theorem The map Cm (IR2)m ? SR2 is Scott
continuous, Hausdorff and Lebesgue computable. - Complexity
- Em(x) is O(m log m).
- Im(x) is also O(m log m).
- We have precisely the complexity of the
classical convex hull algorithm in R2 and R3.
47Voronoi Diagrams
- We are given a finite number of points in the
plane.
- Divide the plane into components closest to
these points.
- This problem is equivalent to the Delaunay
triangulation of the points - (1) Triangulate the set of given points so
that the interior of the circumference circles do
not contain any of the given points.
(2) Draw the perpendicular bisectors of
the edges of the triangles.
48Voronoi Diagram Partial Circles
- The centre of the circle through the three
vertices of a triangle is the intersection of the
perpendicular bisectors of the three edges of the
triangle.
- The partial circle of three partial points in the
plane is obtained by considering the Partial
Perpendicular Bisector of two partial points in
the plane.
49Partial Perpendicular Bisector of Two Partial
Points
50PPBs for Three Partial Points
51Partial Circles
Each partial circle is defined by its interior
and exterior. The exterior (interior) consists of
all those points of the plane which are outside
(inside) all circles passing through any three
points in the three rectangles.
The exterior is the union of the interiors of the
three red circles.
The Interior is the intersection of the interiors
of the three blue circles.
52Partial Circles
With more exact partial points, the boundaries of
the interior and exterior of the partial circle
get closer to each other.
53Partial Circles
- The limit of the area between the interior and
exterior of the partial circle, and the Hausdorff
distance between their boundaries, is zero.
- We get a Scott continuous map C (IR2)3?SR2
- We obtain a robust Voronoi algorithm which is m
log m on average.
54Lines and hyperplanes
- One can define a Scott continuous function which,
given any two partial points, gives the partial
line through them as a partial solid object -
L (IR2)2 ? SR2 (x,y) ? (?,A)
- However, if x and y intersect then A ?, i.e. no
information is provided. - Moreover, the intersection of two partial lines
will not be a partial point as one would desire. - We would like to model a partial line as a set of
lines rather than a set of points.
55Projective Lines and hyperplanes
- By going to the projective real space we can take
lines and hyper-planes as primitives on an equal
footing with points. - By going further to a domain model of the
projective real space we get a computational
model for projective geometry and also a
data-type for systems of linear equation. - The d-dimensional real projective space Pd is
- 1. the set of one dimensional linear
subspaces of Rd1, or equivalently, - 2. the quotient of Rd1-0 under the
colinearity relation X
Y if ??. X ?Y, or equivalently, - 3. the quotient of the sphere Sd by
identifying anti-podal points X Y iff XY or X
-Y. - The one dimensional linear subspaces of Rd1
intersect the hyperplane Xd1 1 at an affine
subspace or, when the intersection is empty, at a
point at infinity. -
56The Real Projective Space
X
X
o
-8
8
-X
57The Orthogonality Relation Projective Duality
- The orthogonality of vectors in Rd1 induces an
orthogonality relation on Pd we write p ? q - O(p) q ?Pd p ? q is a d-1 dimensional
projective subspace of Pd . - Conversely, for any d-1 dimensional projective
subspace S, there exists p ?Pd with SO(p). - Therefore the set of points and the set of d-1
dimensional projective subspace of Pd are in
one-to-one correspondence with each other. - We have the projective duality
- q ? O(p) iff p ? q iff p ? O(q)
58- The set ? O(pi) i1,2,,d is the
intersection of d hyper-planes It
can be a single point or a hyper-plane, but, in
contrast to classical geometry, it cannot be the
empty.
- But O(p) is, by duality, also the set of
hyper-planes containing p
- Therefore, ? O(pi) i1,2,,d is, by
duality, also the non-empty set of hyper-planes
containing all pi (i1,2,,d ).
59A Domain-theoretic Model for Projective Geometry
- Proposition. For any comact subset A? Pd the
set
O(A) q ?Pd ?p ? A. p ? q ? Pd is
compact. - Let UPd be the upper space of Pd, i.e.the set of
non-empty subsets of Pd ordered by reverse
inclusion. - Proposition. The map O UPd ?Upd is Scott
continuous. - Theorem. (Data Type for Linear Systems) L
(UPd)d ? UPd defined by L(A1,A2,.,Ad)
? O(Ak) k1,,d is Scott
continuous and computable. - L computes the intersection of d hyper-planes.
- L also computes, by duality, the hyper-plane
through d points. - Remark. The corresponding classical operation is
not continuous - The intersection of two lines in R2 can be a
single point, a line or the empty set. - The intersection of two projective lines P2 can
be a projective point or a projective line.
60Part 3 A Domain-Theoretic Model for
Differential Calculus
- Interval derivative of a function
- Fundamental Theorem of Calculus
- Domain of C1 functions
- Higher interval derivatives Domains for Ck and
C8 - Picards Theorem
A data-type for differential
equations
61Non-smooth Mathematics
Smooth Mathematics
- Geometry
- Differential Topology
- Manifolds
- Dynamical Systems
- Mathematical Physics
- .
- .
- All based on differential
calculus
- Set Theory
- Logic
- Algebra
- Point-set Topology
- Graph Theory
- Model Theory
- .
- .
62The Domain of Intervals
- (I0,1, ?) is a bounded complete dcpo ?i?I ai
?i?I ai - a b ? ao ? b
- (I0,1, ?) is ?-continuousBasis q, q q lt
q q, q ? Q - Scott topology has basis ?a b ? I0,1 a
b, a ? I0,1 - x ? x 0,1 ? I0,1Topological embedding
?
x
x
0,1
I 0,1
63Continuous Functions
- f 0,1 ? R, f ? Co0,1,has continuous
extensionIf I0,1 ? IR a ? f(a)
f(x) x ? a - Scott continuous maps I0,1 ? IR withf ? g ?
?x ? I0,1 . f(x) ? g(x)is another bounded
complete ?-continuous dcpo. - ? Co0,1 ? (I0,1 ? IR), with f ? Ifis a
topological embedding. - Image (?) ? Max(I0,1 ? IR) a proper subset.
64Step Functions
- We will write a?b as (a, b) on this page.
- Finite lubs of consistent step functions (simple
functions) ?1?i?n(ai,bi), with ai, bi rational
intervals, give a basis for I0,1 ? IR - The finite family A ? I0,1 ? IRis consistent
iff?B ? A . Con(?1B) ? Con?(?2B)
65Step Functions-An Example
R
b3
a1
a3
b1
b2
a2
0
1
66Refining the Step Functions
R
b3
a1
a3
b1
b2
a2
0
1
67Operations in Interval Arithmetic
68Interval Derivative
iff ?x1,x2 a. b(x1 x2) ? If (x1) If
(x2)
iff ?x1,x2 a. ?y1,y2 ? IR, x1?y1 ? If
x2?y2 ? If
b(x1 x2) ? y1 y2
69Interval Derivative
y1
If(x2)
y2
If(x1)
0
1
x2
x1
a
70Definition of Interval Derivative
- f I0,1 ? IR has an interval derivativeb ? IR
in a ? I0,1 if ?x1, x2 a, ? y1,y2 ? IR x1?y1,
x2?y2 ? f b(x1 x2) ? y1 y2. - The collection of all such f is called the tie of
a with b, denoted ?(a,b).
71 Proposition.
- Note that
- f ? g ? (f ? ?(a,b) ? g ? ?(a,b)) i.e. ?(a,b)
?upper I0,1 ? IR - If f g c, c ? R, thenf ? ?(a,b) ? g ?
?(a,b)
72 Proposition.
- ?(a1,b1) ? ?(a2,b2) iff a2 ? a1 b1 ? b2
- ?ni1 ?(ai,bi) ? ? iff ? ai?bi?ni1 is
consistent. - ?i?I ?(ai,bi) ? ? iff ?J?f I . ?i?J ?(ai,bi) ?
?(non trivial) - In fact ?(a,b) behaves exactly like a?b.
73The Interval Derivative
74Examples
75The Interval Derivative Operator
- (I0,1 ? IR) ? (I0,1 ? IR)is
monotone but not continuous. Note that the
classical operator is not continuous either. - (a?b) ?x . ?
- is not linear! For f x ? x I0,1
? IR g x ?
x I0,1 ? IR - (fg) ? ?x . (1 1) ?x
. 0
76The Differential and Integral Operators
- We construct a domain for C1 functions.
- Let L1 ? P(I0,1 ? IR), ( P the power set
functor) with x ? L1 iff x
is the non-empty intersection of a family of
ties x ?i?I ?(ai,bi) ? ?
i.e. x is the collection of functions
satisfying a given set of differential
properties. - Theorem. (L1, ?) is a dcpo and the differential
operator - D (L1, ?) ? (I0,1 ? IR)
x ? ??(a,b)?x a?bis an isomorphism
with the integral operator as its inverse - ? (I0,1 ? IR) ? (L1, ?)
- f ? ?a?b?f ?(a,b)
77Fundamental Theorem of Calculus
- For f, g ? C10,1, define fg if f g c, for
some c ? R. - There is a unique map which makes the following
two diagrams commute
78Fundamental Theorem of Calculus
79The Domain of C1 Functions with C1 Norm
- Define the consistency relationCon ? (I0,1 ?
IR) ? (I0,1 ? IR) with(f,g) ? Con if
(?f) ? (? g) ? ? - (?i ai?bi, ?j cj?dj) ? Con iff we have an open
channel
- (f,g) ? Con iff L ?G . It is decidable on the
basis.
- The Update is Up(f,g) (fg , g) where
fg ? (?f) ? (? g)
80The Domain of C1 Functions with C1 Norm
- Lemma. Con ? (I0,1 ? IR)2 isScott closed.
- Theorem.D1 (f,g) ? (I0,1?IR)2 (f,g) ?
Conis a bounded complete ?continuous dcpo.
- Theorem.??? C10,1 ? C00,1 ? (I0,1 ?
IR)2restricts to give a topological embedding
D1c ? D1 (with C1 norm)
(with Scott topology)
81Consistency Test for (f,g)
- Also define L(x) supy?O?Dom(f)(f (y)
d(x,y)) and G(x)
infy?O?Dom(f)(f (y) d(x,y))
82Consistency Test
- Theorem. (f, g) ? Con iff ?x ? O. L(x) ? G(x).
For (f, g) (?1?i?n ai?bi, ?1?j?m cj?dj) - the rational endpoints of ai and cj induce a
partition X x0 lt x1 lt x2 lt lt xk of O. - Proposition. For arbitrary x ? O, there isp,
where 0 ? p ? k, with L(x)
f (xp) d(x,xp). - Similarly for G(x).
83Higher Interval Derivative
84Higher Interval Derivative
- For f I0,1 ? IR we define I0,1 ?
IR by - Then ?f ? ?2(a,b) a?b
- Define L20,1 ? P(I0,1 ? IR) by x ? L20,1
iff x ? ?2(ai,bi) ? ? - Then L20,1 ? L10,1 is an isomorphism.
?i?I ?2(ai,bi) ? ? i?I ?1(ai,bi)
D
85Domains of C 2and C k functions
- D2 (f0,f1,f2) ? (I0,1?IR)3 ?f0 ? ?f1 ?
?? f2? ?
- Theorem. ????? restricts to give a topological
embedding D2c ? D2
86Picards Theorem
87Picards Solution Reformulated
88Strong Consistency
- Define (f,g) to be strongly consistent,
(f,g)?SCon, if - ?h?g . (f,g)?Con
- We define s(x) supy?O?Dom(f) (f (y)
d(x,y)) and t(x)
infy?O?Dom(f) (f (y) d(x,y)) - Theorem. (f,g) is strongly consistent iff
- ?x ? O. s(x), t(x) ? f(x)
- If (f, g) (?1?i?n ai?bi, ?1?j?m cj?dj), then
there exists z in the partition X withs(x) f
(xz) d(x,xp). Similarly for t(x). - On basis elements, strong consistency is
decidable.
89A Data Type for Differential Equations
- Let F I0,1 ? IR ? IR and?F (I0,1 ? IR)2
? (I0,1 ? IR)2 (f,g) ? (
f , ?T . F(T,f(T)) ) - Let Up D1 ? D1, with Up(f , g) ? ( fg , g )
where fg ? ( ?f ? ?g ) - DF (f,g) ? D1 (f,g) ? SCon, (f,g) ?
?F(f,g) is a dcpo. - For any initial (f0,g0) ? DF, the continuous map
P Up o ?F DF ?
?(f0,g0) ? DF ? ?(f0,g0)
has a least fixed point. - If FIf for a map f0,1 ? R which satisfies the
Lipschitz property of Picards theorem, then in
the limit the domain-theoretic solution tends to
the classical solution.
90Picards Theorem (Example)
F is approximated by a sequence of step
functions, F1, F2,
.
t
We take gi0 Fi and a sequence of rectangles
ai?bi, representing initial condition (1/2,9/8)
?i ai?bi,
F
t
91Picards Theorem (Example)
hi fixed point of P wrt Fi with initial
condition (fi0,gi0)
.
At each stage we find L and G with
hi(x)L(x),G(x)
92Picards Theorem (Example)
hi fixed point of P wrt Fi with initial
condition (fi0,gi0)
.
At each stage we find L and G with
hi(x)L(x),G(x)
93Picards Theorem (Example)
hi fixed point of P wrt Fi with initial
condition (fi0,gi0)
.
At each stage we find L and G with
hi(x)L(x),G(x)
L and G tend to the exact solutionx ? x2/2 1
94Further Work
- Differential Calculus with Several Variables
- Analytic Functions
- Implicit and Inverse function Theorems
- Reconstruct Geometry and Smooth Mathematics with
Domain Theory
95THE ENDhttp//www.doc.ic.ac.uk/ae