• Aucun résultat trouvé

Witness Complex

N/A
N/A
Protected

Academic year: 2022

Partager "Witness Complex"

Copied!
32
0
0

Texte intégral

(1)

Witness Complex

Jean-Daniel Boissonnat DataShape, INRIA

http://www-sop.inria.fr/datashape

(2)

Delaunay triangulations

Finite set of pointsP∈Rd

σ ∈DT(P) ⇔ ∃cσ : kcσ−pk ≤ kcσ−qk ∀p∈σ and ∀q∈P

d

(3)

Motivations

Define Delaunay-like complexes in any metric space Walk around the curse of dimensionality

I The combinatorial complexity of DT depends exponentially on the ambient dimensiond

I The algebraic complexity depends ond

p0

p1 p2

p4

insphere(p0, . . . ,pd+1) =orient(ˆp0, . . . ,ˆpd+1)

=sign

1 . . . 1

p0 . . . pd+1

p20 . . . p2d+1

(4)

Motivations

Define Delaunay-like complexes in any metric space Walk around the curse of dimensionality

I The combinatorial complexity of DT depends exponentially on the ambient dimensiond

I The algebraic complexity depends ond

p0

p4

insphere(p0, . . . ,pd+1) =orient(ˆp0, . . . ,ˆpd+1)

=sign

1 . . . 1

p . . . p

(5)

Witness Complex

[de Silva]

La finite set of points (landmarks) vertices of the complex

W a dense sample (witnesses) pseudo circumcenters

The Witness Complex

Wit(L, W)

w

Definition

2Wit(L, W) () 8⌧ 9w2W with

d(w, p)d(w, q) 8p2 8q2L\

L:Landmarks (black dots); the vertices of the complex W :Witnesses (blue dots)

R. Dyer (INRIA) Del(L,M) =Wit(L, W) Assisi, EuroCG 2012 2 / 10

Letσbe a (abstract) simplex with vertices inL, and letwW. We say thatwis awitnessofσif

kwpk ≤ kwqk pσ and qL\σ

Thewitness complexWit(L,W)is the complex consisting of all simplexes σsuch that for any simplexτσ,τ has a witness inW

(6)

Witness Complex

[de Silva]

La finite set of points (landmarks) vertices of the complex

W a dense sample (witnesses) pseudo circumcenters

The Witness Complex

Wit(L, W)

w

Definition

2Wit(L, W) () 8⌧ 9w2W with

d(w, p)d(w, q) 8p2 8q2L\

L:Landmarks (black dots); the vertices of the complex W :Witnesses (blue dots)

Letσbe a (abstract) simplex with vertices inL, and letwW. We say thatwis awitnessofσif

kwpk ≤ kwqk pσ and qL\σ

(7)

Witness Complex

[de Silva]

La finite set of points (landmarks) vertices of the complex

W a dense sample (witnesses) pseudo circumcenters

The Witness Complex

Wit(L, W)

w

Definition

2Wit(L, W) () 8⌧ 9w2W with

d(w, p)d(w, q) 8p2 8q2L\

L:Landmarks (black dots); the vertices of the complex W :Witnesses (blue dots)

R. Dyer (INRIA) Del(L,M) =Wit(L, W) Assisi, EuroCG 2012 2 / 10

Letσbe a (abstract) simplex with vertices inL, and letwW. We say thatwis awitnessofσif

kwpk ≤ kwqk pσ and qL\σ

Thewitness complexWit(L,W)is the complex consisting of all simplexes σsuch that for any simplexτσ,τ has a witness inW

(8)

Easy consequences of the definition

The witness complex can be defined for any metric space and, in particular, for discrete metric spaces

IfW0⊆W, thenWit(L,W0)⊆Wit(L,W) Del(L)⊆Wit(L,Rd)

(9)

Easy consequences of the definition

The witness complex can be defined for any metric space and, in particular, for discrete metric spaces

IfW0⊆W, thenWit(L,W0)⊆Wit(L,W) Del(L)⊆Wit(L,Rd)

(10)

Easy consequences of the definition

The witness complex can be defined for any metric space and, in particular, for discrete metric spaces

IfW0⊆W, thenWit(L,W0)⊆Wit(L,W) Del(L)⊆Wit(L,Rd)

(11)

Construction of the k-skeleton of Wit(L, W )

Input: L,W, and∀w∈Wthe listN(w) = (p0(w), ...,pk(w)) of theknearest landmarks ofw, sorted by distance fromw Wit(L,W) :=∅

W0 :=W

fori=0, ...,kdo foreachw∈W0 do

ifthei-simplexσ(w) = (p0(w),p2(w), ...,pi(w))6∈Wit(L,W) then

ifthe(i−1)-faces ofσ(w)are all inWit(L,W)then addσ(w)toWit(L,W)

else

W0 :=W0\ {w}

Output: the witness complexWit(L,W)

(12)

Construction of the k-skeleton of the witness complexes

Constructing the listsN(w):O(|W|log|W|+k|W|)[Callahan & Kosaraju 1995]

Construction ofWit(L,W):O (|Witk(L,W)|+|W|)k2 log|L|

(using ST)

|Witk(L,W)≤k|W|

Algebraic complexity : comparisons of (squared) distances : degree 2 Implementation and experimental results : see the Gudhi library !

(13)

Delaunay and Witness complexes

(14)

Identity of Del

|

(L) and Wit(L, R

d

)

[de Silva 2008]

Theorem : Wit(L,W)⊆Wit(L,Rd) =Del(L) Remarks

I Faces of all dimensions have to be witnessed

I Wit(L,W)isembeddedinRd ifLis in general position wrt spheres

(15)

Identity of Del

|

(L) and Wit(L, R

d

)

[de Silva 2008]

Theorem : Wit(L,W)⊆Wit(L,Rd) =Del(L) Remarks

I Faces of all dimensions have to be witnessed

I Wit(L,W)isembeddedinRd ifLis in general position wrt spheres

(16)

Identity of Del

|

(L) and Wit(L, R

d

)

[de Silva 2008]

Theorem : Wit(L,W)⊆Wit(L,Rd) =Del(L) Remarks

I Faces of all dimensions have to be witnessed

I Wit(L,W)isembeddedinRd ifLis in general position wrt spheres

(17)

Proof of de Silva’s theorem

τ = [p0, ...,pk]is ak-simplex ofWit(L)witnessed by a ballBτ (i.e.Bτ L=τ) We prove thatτDel(L)by induction onk

Clearly true fork=0

Bσ

c

w

Bτ

τ σ

Hyp.: true for k0k1 B:=Bτ

σ:=∂Bτ, l:=|σ|

//σDel(L)by the hyp.

whilel+1=dimσ <kdo Bthe ball centered on[cw]s.t.

- σ∂B,

- Bwitnessesτ - |∂Bτ|=l+1

(Bwitnessesτ)∂B) τDel(L)

(18)

Proof of de Silva’s theorem

τ = [p0, ...,pk]is ak-simplex ofWit(L)witnessed by a ballBτ (i.e.Bτ L=τ) We prove thatτDel(L)by induction onk

Clearly true fork=0

Bσ

c

w

Bτ

τ σ

Hyp.: true for k0k1 B:=Bτ

σ:=∂Bτ, l:=|σ|

//σDel(L)by the hyp.

whilel+1=dimσ <kdo Bthe ball centered on[cw]s.t.

- σ∂B,

- Bwitnessesτ

(19)

Case of sampled domains : Wit(L, W) 6 = Del(L)

W a finite set of points⊂Ω⊂Rd

Wit(L,W)6=Del(L), even ifWis a dense sample ofΩ

a b

Vor(a, b)

[ab]Wit(L,W) ⇔ ∃pW, Vor2(a,b)W 6=

(20)

Protection

δ

cσ

δ-protectionWe say that a Delaunay simplexσ⊂Lisδ-protected if

(21)

Protection implies Wit(L, W) = Del(L)

Lemma LetΩbe a convex subset ofRdand letWandLbe two finite sets of points inΩ. IfW isε-dense inΩand if all simplices (of all dimensions) ofDel|Ω(L)areδ-protected withδ≥2ε, then

Wit(L,W)) =Del|Ω(L)

Proof It suffices to proveWit(L,W)⊇Del|Ω(L)

∀σ ∈Del|Ω(L),∃c∈Ωsuch that

∀p∈σ, ∀q∈L\σ, kc−pk ≤ kc−qk −δ

W ε-dense inΩ ⇒ ∃w∈Wsuch thatkw−ck ≤ε

∀p∈σ, ∀q∈L\σ kw−pk ≤ kw−ck+ (kc−qk −δ)

≤ kw−qk+2kw−ck −δ

≤ kw−qk+2ε−δ

(22)

Good links

A simplicial complexK is ak-pseudomanifold complexwithout boundary if

1 Kis a purek-complex

2 every(k1)-simplex is the face of exactly twok-simplices

We say that a complexK⊂ d = d/ d with vertex setLhas

(23)

Good links

A simplicial complexK is ak-pseudomanifold complexwithout boundary if

1 Kis a purek-complex

2 every(k1)-simplex is the face of exactly twok-simplices

We say that a complexK⊂Td =Rd/Zd with vertex setLhas good linksif

∀p∈L, link(p,K)is a(d−1)-pseudomanifold

(24)

Good links implies Wit(L, W ) = Del(L)

Lemma

IfKis a triangulation ofTd andK0 ⊆K a simplicial complex with the same vertex set

then K0 =K ⇔ K0has good links Corollary

If all vertices ofWit(L,W)havegood links,Wit(L,W) =Del(L)

(25)

Good links implies Wit(L, W ) = Del(L)

Lemma

IfKis a triangulation ofTd andK0 ⊆K a simplicial complex with the same vertex set

then K0 =K ⇔ K0has good links Corollary

If all vertices ofWit(L,W)havegood links,Wit(L,W) =Del(L)

(26)

Good links implies Wit(L, W ) = Del(L)

Lemma

IfKis a triangulation ofTd andK0 ⊆K a simplicial complex with the same vertex set

then K0 =K ⇔ K0has good links Corollary

If all vertices ofWit(L,W)havegood links,Wit(L,W) =Del(L)

(27)

Turning witness complexes into Delaunay complexes

Input: L,W,ρ(perturbation radius) Init :L0 :=L; computeWit(L0,W)

while a vertexp0 ofWit(L0,W)has a bad linkdo perturbp0 and the points ofI(p0)

updateWit(L0,W)

Output: Wit(L0,W) =Del(L0)

Analysis : adapt Moser-Tardos algorithmic version of the Lov ´asz Local Lemma (last lecture)

(28)

Main result

Under the condition µ

4 ≥ρ≥ 24dε

¯

µJ where J−1=

2

¯ µ

O(d2)

the algorithm terminates and returnsWit(L0,W) =Del(L0)

The expected complexity islinear in|L|

(29)

Main result

Under the condition µ

4 ≥ρ≥ 24dε

¯

µJ where J−1=

2

¯ µ

O(d2)

the algorithm terminates and returnsWit(L0,W) =Del(L0)

The expected complexity islinear in|L|

(30)

Relaxed witness complex

Alpha-witnessLetσ be a simplex with vertices inL. We say that a pointw∈W is anα-witness ofσ if

kw−pk ≤ kw−qk+α ∀p∈σ and ∀q∈L\σ Alpha-relaxed witness complexTheα-relaxed witness complex Witα(L,W)is the maximal simplicial complex with vertex setLwhose simplices have anα-witness inW

Wit0(L,W) =Wit(L,W)

Filtration : α≤β ⇒ Witα(L,W)⊆Witβ(L,W)

(31)

Construction

Ai ={z∈L: mi ≤d(w,z)≤mi+α}

m5

z0 z1 z2 z3 z4 z5 z6 z7 z8 z9 z10

{z4, z5, z6, z7, z8} 3

!

w

m3 m3+ρ A3

Fori≤j+1,wα-witnesses all thej-simplices that contain

{z0,· · ·,zi−1}and a(j+1−i)-subset ofAi(provided|Ai| ≥j+1−i) Allj-simplices that areα-witnessed byware obtained this way, and exactly once, wheniranges from0toj+1

(32)

Interleaving complexes

LemmaAssume thatWisε-dense inΩand letα≥2ε. Then Wit(L,W)⊆Del|Ω(L)⊆Witα(L,W)

Proof

σ : ad-simplex ofDel(L), cσ its circumcenter W ε-dense inΩ ∃w∈W s.t. kcσ−wk ≤ε For anyp∈σandq∈L\σ, we then have

∀p∈σ and q∈L\σ kw−pk ≤ kcσ−pk+kcσ−wk

≤ kcσ−qk+kcσ−wk

≤ kw−qk+2kcσ−wk

Références

Documents relatifs

(13) Mardochée dit pour répondre à Esther “Ne pense pas en toi-même être sauvée étant dans la maison du roi au contraire de tous les Juifs (14) car si tu te tais vraiment

Mais l'Esprit souffle où il veut… Dans chacune de ses paroles, dans l'embrouillamini de ses feuilles de notes déliquescentes, dans le sourire provocant qu'il

Mathilde Imer, jeune militante écologiste engagée dans la naissance de ce projet et membre de son comité de gouvernance, Cyril Dion, cinéaste engagé pour la cause climatique qui

Bien que les cantons précités continuent de s'opposer au pr inc ipe d'une révis ion qu' ils jugent inopportune dans la conjoncture économique présente et motivée

Le TF donne raison à l'autorité fédérale quant à la conclusion les autorisations sont annulées; son raisonnement diffère pourtant de celui de l'OFJ seule la

Jean Natali écrit : « Jean m’avait demandé d’en écrire la préface et je rappelai cette séance de l’Académie de Chirurgie du 29 avril 1929 où, après qu’un chirurgien

The lambda-calculus is more directly connected to functional languages (Lisp, Scheme, SML, Ocaml, Haskell), but its type theory inspired many other languages (Java, C#, Scala)..

Caml is the version of Robin Milner’s ML language developed at INRIA by Xavier Leroy et al.. Types in ML are polymorphic with a powerful type