SLIDE 1 DDG 07 GENERALIZED CURVATURES J.M. Morvan Institut Camille Jordan Universit Claude Bernard Lyon 1, with ...
- V. Borrelli,
- D. Cohen-Steiner,
- B. Thibert.
1
SLIDE 2 ... An Introduction ... "In what sense do two sets have to be close to each other, in order to guarantee that their curvature measures are close to each
- ther ?"
- J. Milnor, (1994)
... Enigmatic question ...
- What are the curvature measures of a
set ?
- What does mean : close to each other ?
2
SLIDE 3 ... TWO QUESTIONS ...
- Does a geometric property dened on the
class of smooth objects have an analo- gous on the class of discrete objects?
- Dene a frame in which one can compare
geometric properties of a smooth object and the corresponding geometric proper- ties of a discrete object close to it .
3
SLIDE 4 ... EXAMPLE ...
- On a smooth curve, one can dene its
length l, its curvature k, its torsion t ...
- On a smooth surface, one can dene its
area A, its Gauss curvature G, its mean curvature H, its second fundamental form h, the lines of curvatures ... Can one dene these invariants on discrete
- bjects like polyhedron for instance ?
The main problem is to nd GOOD DEFINITIONS
4
SLIDE 5
5
SLIDE 6 HISTORY
- 1. The convex case, Steiner, 1840
- 2. The smooth case, Weyl, 1939
- 3. The introduction of geometric measure,
Federer, 1958
- 4. The introduction of cohomology, Wint-
gen, 1978
- 5. The recent works, (Zahle, Fu, etc... 1980−
2007)
6
SLIDE 7 THE CURVATURES OF A SMOOTH HYPERSURFACE M : (oriented) smooth hypersurface of the (oriented) Euclidean space EN, < . >, ˜ ∇.
- Let ξ be the unit normal vector eld to
M.
- The Weingarten tensor (shape operator)
A dened on M : A(X) = −˜ ∇Xξ. (1)
- The second fundamental form h of M :
< A(X), Y >= h(X, Y ). (2)
(λ1, ..., λi, ..., λ(N−1))
- f M at m, and the principal vector elds,
7
SLIDE 8 Denition Let M be a smooth hypersurface
- f EN. For every , k = 0, ..., (N −1), the k−th-
elementary symmetric function of the princi- pal curvatures of M Ξk = {λi1, ..., λik} is called the k−th-mean curvature of M. Remark : One has det (I + tA) =
k=N−1
Ξktk, (3)
- Ξ0 = 1;
- Ξ1 is the trace of h, (H = 1
nX1 is called
the mean curvature of M).
- ΞN−1 is the determinant of h, (usually
denoted by G), called the Gauss curva- ture of M.
SLIDE 9 CURVATURE MEASURES OF A SMOOTH HYPERSURFACE Let U be a domain of M. ϕk(U) =
is called the k-th Lipschitz-Killing curva- ture of M.
- ϕ0(U) is the classical (N − 1)-volume of
U;
- ϕN(U) is the total Gauss curvature of U.
8
SLIDE 10
Go back to the problem : How to dene these curvature measures for discrete subsets of EN ? A IMPORTANT TOOL : THE VOLUME By computing the volume of some tubular neighborhoods of a subset X, we can exhibit some geometric invariants of X...
9
SLIDE 11 THE CONVEX CASE
- K convex body,
- Kǫ = {m ∈ EN, d(m, K) ≤ ǫ},
ε
A
- ∂K: convex hypersurface,
- ∂Kǫ: parallel hypersurface at distance ǫ.
10
SLIDE 12 THE CONVEX CASE... STEINER FORMULA. Theorem - Let K be a convex body. Then, V (Kǫ) =
N
Φk(K)ǫk, ∀ǫ ≥ 0. AN ADDITIVITY PROPERTY Theorem: If K1, K2 an K1 ∪ K2 are convex, then ∀k,
Φk(K1 ∪ K2) = Φk(K1) + Φk(K2) − Φk(K1 ∩ K2).
CONTINUITY OF THE Φk IN THE CONVEX WORLD Theorem: If a sequence of Kn tends to K in the Hausdor sense, then ∀k, lim
n→∞ Φk(Kn) = Φk(K).
12
SLIDE 13 Question: Compute explicitly the Φ′
ks ?
We will test on
- smooth objects,
- polyhedra.
13
SLIDE 14 THE CONVEX CASE... STEINER FORMULA FOR SMOOTH BODIES Steiner formula can be stated as follows : VolN(Kǫ) =
k=N
Φk(K)ǫk, with Φk(K) = CNϕk(K) = CN
In particular, if K is a compact convex body in E3 with smooth boundary, then Vol3(Kǫ) = Vol3(K)+A(∂K)ǫ+(
3(
where H, (resp. G), denotes the mean cur- vature, (resp. Gauss curvature) of ∂K.
14
SLIDE 15 THE CONVEX CASE... STEINER FORMULA FOR POLYHEDRA We give the following Denition 1 Let σl be a l-dimensional face
- f a k-simplex σk, (l < k).
Let q ∈ int(σl). The following notions are independent of q.
- 1. The normal cone C⊥(σl, σk) to σl :
C⊥(σl, σk) = {x ∈ int(σk) : qx ∈ σl⊥}.
- 2. The basis of the normal cone C⊥(σl, σk)
to σl is L(σl, σk) = C⊥(σl, σk) ∩ Sk−l−1, where
Sk−l−1 is the unit sphere centered
at q.
15
SLIDE 16 3
q
σ e
- 3. The internal dihedral angle
- (σl, σk) is the
measure of L(σl, σk). The normalized internal dihedral angle (σl, σk) : (σl, σk) =
sk−l−1 .
- 4. The external dihedral angle
- (σl, σk)∗ is
the measure of the subset of
Sk−l−1 ob-
tained by intersecting Sk−l−1 with the half- lines whose origin is q, and making an an- gle greater than π
2 with the interior of σk.
SLIDE 17 t v
The normalized external dihedral angle (σl, σk)∗ : (σl, σk)∗ =
sk−l−1 .
SLIDE 18 THE CONVEX CASE... STEINER FORMULA FOR POLYHEDRA Theorem - Let K be a convex body of EN whose boundary ∂K is a polyhedron. Then, VolN(Kǫ) =
k=N
Φk(K)ǫk, with Φk(K) =
VolN−k(σN−k)(σN−k, σN)∗. where
- σN−k denotes a generic (N − k)-face of
∂K,
- (σN−k, σN)∗ denotes the normalized diedral
external angle,
16
SLIDE 19 THE CONVEX CASE... STEINER FORMULA FOR POLYHEDRA IN E3
Vol3(Kǫ) = Vol3(K)+A(∂K)ǫ+(
∠(a)l(a))ǫ2+ 4
3πǫ3. Remember the smooth case : Vol3(Kǫ) = Vol3(K)+A(∂K)ǫ+(
Hda)ǫ2+1 3(
Gda)ǫ3,
17
SLIDE 20 Example : One dene the global mean curvature
- of a smooth compact convexe surface S
by Φ2(S) =
- S Hda,
- of a compact convex polyhedron P by:
Φ2(P) =
∠(a).l(a).
Then
Φ2(A∪B) = Φ2(A)+Φ2(B)−Φ2(A∩B).
- 2. Moreover, if Pn tends to S in the Haus-
dor sense then lim
∠(an).l(an) =
18
SLIDE 21 HOWEVER, ALL THIS THEORY FAILS FOR NON CONVEX SUBSETS BECAUSE :
- THE ADDITIVITY FORMULA IS NOT
SATISFIED ;
- THE CONTINUITY PROPERTY IS
NOT SATISFIED.
19
SLIDE 22
... A CLASSICAL PROBLEM ... NON CONTINUITY OF THE AREA The situation is completely dierent for sur- faces and their areas.... Let M be a C∞ sur- face of E3. One can dene its area A(M). It satises A(M1∪M2) = A(M1)+A(M2)−A(M1∩M2). Let P be a polyhedron. One can dene its area which satises the same property. What about the continuity property?
20
SLIDE 23
THE LANTERN OF SCHWARZ
21
SLIDE 24 THE LANTERN OF SCHWARZ Let C be a cylinder of height l and radius R. Let Pn,N be the lantern with N slices and 2n triangles on each slice. The sequence Pn,n2. The area of Pn,n2 sat- ises: A(Pn,n2) → 2πR
4 + l2, Then: A(Pn,n2) → A′ = A(C). However δ(Pn,n2, C) → 0 (where δ is Hausdor distance).
22
SLIDE 25
NON ADDITIVITY OF Vol Kǫ Vol (I1 ∪ I2)ǫ = Vol (I1)ǫ + Vol (I2)ǫ − Vol (I1 ∩ I2)ǫ.
r p q
23
SLIDE 26
... ANOTHER POINT OF VIEW ... THE NORMAL CYCLE Let M smooth hypersurface in EN. The good object to study is the Gauss map. G : M → EN × SN−1 ⊂ EN × EN ≃ TEN m → G(m) = (m, ξ) CRUCIAL FACT : integrating particular dierential forms of EN × SN−1 on G(U) gives all the Lipschitz curvatures of U !!
24
SLIDE 27 INVARIANT (N − 1)-FORMS ON EN × SN−1 Let G the group of rigid motions of EN. G acts on EN × SN−1 in a natural way. ω ∈ ΛN−1(EN × SN−1) is invariant (by G) if φ∗(ω) = ω, for all φ ∈ G. One can build a basis ω0, ..., ωN of the space
- f invariant dierential forms on EN × SN−1.
25
SLIDE 28
INVARIANT 2-FORMS ON E3 × S2 The vector space of invariant 2-forms on E3× S2 has dimension 4. It is spent by
ωA = y1dx2 ∧ dx3 + y2dx3 ∧ dx1 + y3dx1 ∧ dx2. ωG = y1dy2 ∧ dy3 + y2dy3 ∧ dy1 + y3dy1 ∧ dy2. ωH = y1(dx2 ∧ dy3 + dy2 ∧ dx3)+ y2(dx3 ∧ dy1 + dy3 ∧ dx1)+ y3(dx1 ∧ dy2 + dy1 ∧ dx2). Ω = (y2
2 + y2 3)dx1 ∧ dy1 − y1y2dx1 ∧ dy2 − y1y3dx1 ∧ dy3
−y1y2dx2 ∧ dy1 + (y2
1 + y2 3)dx2 ∧ dy2n − y2y3dx2 ∧ dy3
−y1y3dx1 ∧ dy3 − y2y3dx3 ∧ dy2 + (y2
2 + y2 3)dx3 ∧ dy3.
26
SLIDE 29 M smooth surface of E3
- G(U) ωA = A(U);
- G(U) ωG =
- U G;
- G(U) ωH =
- U H;
- G(U) Ω = 0.
27
SLIDE 30 THE NORMAL CYCLE... (the case N=3) X → N(X) "Theorem" - To "each" compact sub- set X of EN, one can associate canonically an unique interesting rectiable Legendrian (N − 1)-current N(X) of EN × SN−1. More-
- ver, the map N satises the following addi-
tivity formula : If A and B are two compact subsets of E3 then, N(A ∪ B) = N(A) + N(B) − N(A ∩ B).
28
SLIDE 31
THE NORMAL CYCLE... Example ... The normal cycle of a surface If M is a smooth surface, N(M) = {(m, v) ∈ E3 × S2 : v ∈ T ⊥
m(M)}.
(N(M) is the normal bundle in the usual sense).
29
SLIDE 32
THE NORMAL CYCLE ... The case of a triangulated polyhedron of E3 ... Let P be any triangulated polyhedron of E3. One denes N(P) by decomposing P in a union of simplices and using the additivity formula N(A ∪ B) = N(A) + N(B) − N(A ∩ B). If σ is a k-simplex, N(σ) is simply the unit normal cone of σ.
r p q
30
SLIDE 33
THE NORMAL CYCLE ... CURVATURE MEASURES OF A SUBSET X ... Let U ⊂ X ⊂ EN. Φk(U) =< N(X), χ(U × E3)ωk > .
31
SLIDE 34
SECOND FUNDAMENTAL MEASURE...in E3 Let X, Y be two constant vectors of E3, and (p, ξ) un point de E3 × E3. We build the 2- form on E3 × E3 :
hX,Y
(p,ξ) = (ξ × X) ∧ Y.
32
SLIDE 35
SECOND FUNDAMENTAL MEASURE...in E3 Let K be a subset of E3, and U ⊂ X. Denition
hX,Y
K
(U) =< N(K), χU×E3hX,Y
K
> Theorem The map E3 × E3 → R (X, Y ) → hX,Y
K
(U) is bilinear and symmetric.
33
SLIDE 36 SECOND FUNDAMENTAL MESURE... For a smooth surface S,
hX,Y (U) =
hS(prTSX, prTSY )dvS,
where hS is the usual second fundamental form of S. Let T be a triangulated surface,
hX,Y
T
(U) =
∠(e)l(e ∩ U) <
e, X >< e, Y > .
34
SLIDE 37
35
SLIDE 38 AN APPROXIMATION RESULT "In what sense do two sets have to be close to each other, in order to guarantee that their curvature measures are close to each
- ther ?"
- J. Milnor, (1994)
A possible answer : Let K ⊂ EN and K′ ⊂
EN such that N(K) and N(K′) are close as
- currents. Then the corresponding curvature
measures are close. For instance : if K is smooth, U ⊂ K, U′ = pr U, Theorem -
|Φk
K(U) − Φk K′(U′)| ≤
C(δU + αU)[2(1 + ||hU)||) 1 − δU||hU|| ]N−1,
where C is a constant depending on the mass
- f the normal cycles of K and K′.
36
SLIDE 39
THANK YOU
37