multi-way spectral partitioning and higher-order cheeger inequalities
DESCRIPTION
James R. Lee. University of Washington. multi-way spectral partitioning and higher-order Cheeger inequalities. Shayan Oveis Gharan. Luca Trevisan. Stanford University. S i. Consider a d -regular graph G = ( V , E ). Define the normalized Laplacian : . - PowerPoint PPT PresentationTRANSCRIPT
![Page 1: multi-way spectral partitioning and higher-order Cheeger inequalities](https://reader034.vdocument.in/reader034/viewer/2022051518/56816252550346895dd29b00/html5/thumbnails/1.jpg)
multi-way spectral partitioningand higher-order Cheeger inequalities
University of WashingtonJames R. Lee
Stanford UniversityLuca Trevisan
Shayan Oveis Gharan
Si
![Page 2: multi-way spectral partitioning and higher-order Cheeger inequalities](https://reader034.vdocument.in/reader034/viewer/2022051518/56816252550346895dd29b00/html5/thumbnails/2.jpg)
spectral theory of the Laplacian
Consider a d-regular graph G = (V, E).
kf (x) ¡ f (y)k1 =
Define the normalized Laplacian: (where A is the adjacency matrix of G)
L is positive semi-definite with spectrum
Fact: is disconnected.
¸2 = minf ? 1
hf ;Lf ihf ; f i = min
f ? 1
1d
Xu» v
jf (u) ¡ f (v)j2X
u2Vf (u)2
![Page 3: multi-way spectral partitioning and higher-order Cheeger inequalities](https://reader034.vdocument.in/reader034/viewer/2022051518/56816252550346895dd29b00/html5/thumbnails/3.jpg)
Cheeger’s inequality
Expansion: For a subset S µ V, define
E(S) = set of edges with one endpoint in S.
S
Theorem [Cheeger70, Alon-Milman85, Sinclair-Jerrum89]: ¸22 · ½G (2) ·
p2̧ 2
½G (k) = minfmaxÁ(Si ) : S1;S2; : : : ;Sk µ V disjointgk-way expansion constant:
S2
S3
S4
is disconnected.
![Page 4: multi-way spectral partitioning and higher-order Cheeger inequalities](https://reader034.vdocument.in/reader034/viewer/2022051518/56816252550346895dd29b00/html5/thumbnails/4.jpg)
Miclo’s conjecture
has at least k connected components.f1, f2, ..., fk : V R first k (orthonormal) eigenfunctions
spectral embedding: F : V Rk
Higher-order Cheeger Conjecture [Miclo 08]:
¸k2 · ½G (k) · C(k)
p¸k
for some C(k) depending only on k.
For every graph G and k 2 N, we have
![Page 5: multi-way spectral partitioning and higher-order Cheeger inequalities](https://reader034.vdocument.in/reader034/viewer/2022051518/56816252550346895dd29b00/html5/thumbnails/5.jpg)
our results
¸k2 · ½G (k) · O(k2)
p¸k
Theorem: For every graph G and k 2 N, we have
½G (k) = minfmaxÁ(Si ) : S1;S2; : : : ;Sk µ V disjointg
½G (k) · O(p
¸2k logk)Also,- actually, can put for any ±>0- tight up to this (1+±) factor
- proved independently by [Louis-Raghavendra-Tetali-Vempala 11]
If G is planar (or more generally, excludes a fixed minor), then½G (k) · O(
p¸2k)
![Page 6: multi-way spectral partitioning and higher-order Cheeger inequalities](https://reader034.vdocument.in/reader034/viewer/2022051518/56816252550346895dd29b00/html5/thumbnails/6.jpg)
small set expansion
Á(S) · O(p
¸k logk)
Corollary: For every graph G and k 2 N, there is a subset of vertices S such that and,
Previous bounds:jSj · np
k Á(S) · O(p
¸k logk)and[Louis-Raghavendra-Tetali-Vempala 11]
jSj · nk0:01 Á(S) · O(
p¸k logk n)and
[Arora-Barak-Steurer 10]
![Page 7: multi-way spectral partitioning and higher-order Cheeger inequalities](https://reader034.vdocument.in/reader034/viewer/2022051518/56816252550346895dd29b00/html5/thumbnails/7.jpg)
proof
½G (k) · O(k2)p
¸k
Theorem: For every graph G and k 2 N, we have
![Page 8: multi-way spectral partitioning and higher-order Cheeger inequalities](https://reader034.vdocument.in/reader034/viewer/2022051518/56816252550346895dd29b00/html5/thumbnails/8.jpg)
proof
½G (k) · kO(1)p ¸k
Theorem: For every graph G and k 2 N, we have
![Page 9: multi-way spectral partitioning and higher-order Cheeger inequalities](https://reader034.vdocument.in/reader034/viewer/2022051518/56816252550346895dd29b00/html5/thumbnails/9.jpg)
Dirichlet Cheeger inequality
For a mapping , define the Rayleigh quotient:
R(F ) =
1d
Xu» v
kF (u) ¡ F (v)k2
X
u2VkF (u)k2
Lemma: For any mapping , there exists a subset
such that:
![Page 10: multi-way spectral partitioning and higher-order Cheeger inequalities](https://reader034.vdocument.in/reader034/viewer/2022051518/56816252550346895dd29b00/html5/thumbnails/10.jpg)
Miclo’s disjoint support conjecture
Conjecture [Miclo 08]:For every graph G and k 2 N, there exist disjointly supportedfunctions so that for i=1, 2, ..., k,Ã1;Ã2; : : : ;Ãk : V ! R
Localizing eigenfunctions:
Rk
![Page 11: multi-way spectral partitioning and higher-order Cheeger inequalities](https://reader034.vdocument.in/reader034/viewer/2022051518/56816252550346895dd29b00/html5/thumbnails/11.jpg)
isotropy and spreading
Isotropy: For every unit vector x 2 Rk
M =
0BBB@
f 1f 2...
f k
1CCCA
Total mass:
![Page 12: multi-way spectral partitioning and higher-order Cheeger inequalities](https://reader034.vdocument.in/reader034/viewer/2022051518/56816252550346895dd29b00/html5/thumbnails/12.jpg)
radial projection
Define the radial projection distance on V by,
dF (u;v) =°°°°
F (u)kF (u)k ¡ F (v)
kF (v)k
°°°°
Fact:
Isotropy gives: For every subset S µ V, diam(S;dF ) · 1
2 =)X
v2SkF (v)k2 · 2
kX
v2VkF (v)k2
![Page 13: multi-way spectral partitioning and higher-order Cheeger inequalities](https://reader034.vdocument.in/reader034/viewer/2022051518/56816252550346895dd29b00/html5/thumbnails/13.jpg)
smooth localization
Want to find k regions S1, S2, ..., Sk µ V such that,mass:
separation:Then define by,Ãi : V ! Rk
so that are disjointly supported and
Si "(k)=2 Sj
![Page 14: multi-way spectral partitioning and higher-order Cheeger inequalities](https://reader034.vdocument.in/reader034/viewer/2022051518/56816252550346895dd29b00/html5/thumbnails/14.jpg)
smooth localization
Want to find k regions S1, S2, ..., Sk µ V such that,mass:
Claim:
Then define by,Ãi : V ! Rk
so that are disjointly supported and
separation:
![Page 15: multi-way spectral partitioning and higher-order Cheeger inequalities](https://reader034.vdocument.in/reader034/viewer/2022051518/56816252550346895dd29b00/html5/thumbnails/15.jpg)
disjoint regions
Want to find k regions S1, S2, ..., Sk µ V such that,mass:
separation:
For every subset S µ V, diam(S;dF ) · 1
2 =)X
v2SkF (v)k2 · 2
kX
v2VkF (v)k2
How to break into subsets?Randomly...
![Page 16: multi-way spectral partitioning and higher-order Cheeger inequalities](https://reader034.vdocument.in/reader034/viewer/2022051518/56816252550346895dd29b00/html5/thumbnails/16.jpg)
random partitioning
Rk
![Page 17: multi-way spectral partitioning and higher-order Cheeger inequalities](https://reader034.vdocument.in/reader034/viewer/2022051518/56816252550346895dd29b00/html5/thumbnails/17.jpg)
random partitioning
Rk spreading )
separation
![Page 18: multi-way spectral partitioning and higher-order Cheeger inequalities](https://reader034.vdocument.in/reader034/viewer/2022051518/56816252550346895dd29b00/html5/thumbnails/18.jpg)
smooth localization
We found k regions S1, S2, ..., Sk µ V such that,mass:
separation:
Si Sj
![Page 19: multi-way spectral partitioning and higher-order Cheeger inequalities](https://reader034.vdocument.in/reader034/viewer/2022051518/56816252550346895dd29b00/html5/thumbnails/19.jpg)
improved quantitative bounds
- Take only best k/2 regions (gains a factor of k)
½G (k) · O(p
¸2k logk)
- Before partitioning, take a random projection into O(log k) dimensionsRecall the spreading property: For every subset S µ V,
diam(S;dF ) · 12 =)
X
v2SkF (v)k2 · 2
kX
v2VkF (v)k2
- With respect to a random ball in d dimensions...
u vu
v
![Page 20: multi-way spectral partitioning and higher-order Cheeger inequalities](https://reader034.vdocument.in/reader034/viewer/2022051518/56816252550346895dd29b00/html5/thumbnails/20.jpg)
k-way spectral partitioning algorithm
ALGORITHM:1) Compute the spectral embedding
2) Partition the vertices according to the radial projection
3) Peform a “Cheeger sweep” on each piece of the partition
![Page 21: multi-way spectral partitioning and higher-order Cheeger inequalities](https://reader034.vdocument.in/reader034/viewer/2022051518/56816252550346895dd29b00/html5/thumbnails/21.jpg)
planar graphs: spectral + intrinsic geometry
2) Partition the vertices according to the radial projection
For planar graphs, we consider the induced shortest-path metricon G, where an edge {u,v} has length dF(u,v).
Now we can analyze the shortest-path geometry using [Klein-Plotkin-Rao 93]
![Page 22: multi-way spectral partitioning and higher-order Cheeger inequalities](https://reader034.vdocument.in/reader034/viewer/2022051518/56816252550346895dd29b00/html5/thumbnails/22.jpg)
planar graphs: spectral + intrinsic geometry
2) Partition the vertices according to the radial projection
[Jordan-Ng-Weiss 01]
![Page 23: multi-way spectral partitioning and higher-order Cheeger inequalities](https://reader034.vdocument.in/reader034/viewer/2022051518/56816252550346895dd29b00/html5/thumbnails/23.jpg)
open questions
1) Can this be made ?2) Can [Arora-Barak-Steurer] be done geometrically?
We use k eigenvectors, find ³ k sets, lose .What about using eigenvectors to find sets?
3) Small set expansion problemThere is a subset S µ V with and
Tight for (noisy hypercubes)Tight for (short code graph) [Barak-Gopalan-Hastad-Meka-Raghvendra-Steurer 11]