On the Power of Semidefinite Programming Hierarchies
description
Transcript of On the Power of Semidefinite Programming Hierarchies
![Page 1: On the Power of Semidefinite Programming Hierarchies](https://reader035.fdocuments.in/reader035/viewer/2022081520/56816784550346895ddc9980/html5/thumbnails/1.jpg)
On the Power of Semidefinite Programming Hierarchies
Prasad RaghavendraGeorgia Institute of Technology,
Atlanta, GA.
David SteurerMicrosoft Research New England
Cambridge, MA
![Page 2: On the Power of Semidefinite Programming Hierarchies](https://reader035.fdocuments.in/reader035/viewer/2022081520/56816784550346895ddc9980/html5/thumbnails/2.jpg)
Overview
• Background and Motivation
• Introduction to SDP Hierarchies (Lasserre SDP hierarchy)
• Rounding SDP hierarchies via Global Correlation.
BREAK
• Graph Spectrum and Small-Set Expansion.
• Sum of Squares Proofs.
![Page 3: On the Power of Semidefinite Programming Hierarchies](https://reader035.fdocuments.in/reader035/viewer/2022081520/56816784550346895ddc9980/html5/thumbnails/3.jpg)
Graph Spectrum&
Small-Set Expansion
![Page 4: On the Power of Semidefinite Programming Hierarchies](https://reader035.fdocuments.in/reader035/viewer/2022081520/56816784550346895ddc9980/html5/thumbnails/4.jpg)
Question: If is a -small set expander,how many eigenvalues larger than can the adjancency matrix
have?
= ¿𝑒𝑑𝑔𝑒𝑠 𝑙𝑒𝑎𝑣𝑖𝑛𝑔𝑆𝑑∨𝑆∨¿
vertex set SA random neighbor of a random vertex in
is outside of with probability
A Question in Spectral Graph Theory
d-regular graph G
Def (Small Set Expander): A regular graph G is a -small set expander
if for every set
Let be the normalized adjacency matrix of the d-regular graph , (all entries are )
has eigenvalues (say )
![Page 5: On the Power of Semidefinite Programming Hierarchies](https://reader035.fdocuments.in/reader035/viewer/2022081520/56816784550346895ddc9980/html5/thumbnails/5.jpg)
A Question in Spectral Graph Theory
Def (Small Set Expander): A regular graph G is a -small set
expander if for every set
Question: If is a -small set expander, How many eigenvalues larger than can the normalized adjancency matrix have?
Cheeger’s inequality: every large eigenvalue a sparse cut in
() ( cut of sparsity )
d-regular graph G
Intuitively,How many sparse cuts can a graph have without having a
unbalanced sparse cut?
![Page 6: On the Power of Semidefinite Programming Hierarchies](https://reader035.fdocuments.in/reader035/viewer/2022081520/56816784550346895ddc9980/html5/thumbnails/6.jpg)
Def (Small Set Expander): A regular graph G is a -small set
expander if for every set
Question: If is a -small set expander, How many eigenvalues larger than can the normalized adjacency matrix have?
Significance of the Question
of eigenvalues of graph G that are
[Barak-Raghavendra-Steurer][Guruswami-Sinop]
-round Lasserre SDP solves on graphs with low threshold rank .
Graph has small non-expanding sets,
decompose in to smaller pieces and solve each piece.
If UGC is true, there must be hard instances of Unique Games that
a) Have high threshold rank b) Are Small-Set Expanders.
![Page 7: On the Power of Semidefinite Programming Hierarchies](https://reader035.fdocuments.in/reader035/viewer/2022081520/56816784550346895ddc9980/html5/thumbnails/7.jpg)
Def (Small Set Expander): A regular graph G is a -small set
expander if for every set
Question: If is a -small set expander, How many eigenvalues larger than can the normalized adjacency matrix have?
Answer to the above question (a function of )
Significance of the Question
[Arora-Barak-Steurer 2010]
There is a -time algorithm for GAP-SMALL-SET-EXPANSION problem – a problem closely related to UNIQUE GAMES.
Gap Small Set Expansion Problem (GapSSE)Given a graph G and constants ,
Is OR ?where minimum expansion of sets of size
[Arora-Barak-Steurer 2010] A subexponential-time algorithm for GAP-SMALL-SET-EXPANSION problem
At the time,Best known lower bound for
(GAP-SMALL-SET-EXPANSION problem could be solved in quasipolynomial time.)
![Page 8: On the Power of Semidefinite Programming Hierarchies](https://reader035.fdocuments.in/reader035/viewer/2022081520/56816784550346895ddc9980/html5/thumbnails/8.jpg)
[Barak-Gopalan-Hastad-Meka-Raghavendra-Steurer]
For all small constant
There exists a graph (the Short Code Graph) that is a small set expander with eigenvalues
for some depending on
Short Code Graph
• Led to new gadgets for hardness reductions (derandomized Majority is Stablest) and new SDP integrality gaps.
• An interesting mix of techniques from, Discrete Fourier analysis, Locally Testable Codes and Derandomization.
[BGHMRS 11] [ABS]
![Page 9: On the Power of Semidefinite Programming Hierarchies](https://reader035.fdocuments.in/reader035/viewer/2022081520/56816784550346895ddc9980/html5/thumbnails/9.jpg)
Long Code Graph
• Eigenvectors
• Small Set Expansion
Short Code Graph
Overview
![Page 10: On the Power of Semidefinite Programming Hierarchies](https://reader035.fdocuments.in/reader035/viewer/2022081520/56816784550346895ddc9980/html5/thumbnails/10.jpg)
The Long Code Graph aka Noisy Hypercube
Noise Graph:
Vertices:
Edges: Connect every pair of points in hypercube separated by a Hamming distance of
n dimensional hypercube : {-1,1}n
x
y
and differ in coordinates.
Dictator cuts: Cuts parallel to the Axis (given by )
The dictator cuts yield sparse cuts in graph
1
1 1
1
-1 -1
-1
Eigenvectors are functions on
![Page 11: On the Power of Semidefinite Programming Hierarchies](https://reader035.fdocuments.in/reader035/viewer/2022081520/56816784550346895ddc9980/html5/thumbnails/11.jpg)
Sparsity of Dictator Cuts
n dimensional hypercube
Connect every pair of vertices in hypercube separated by Hamming distance of
Dictator cuts: -eigenvectors with eigenvalues for graph (Number of vertices N = , so #of eigenvalues = )
1
1
1
1
-1 -1
-1
![Page 12: On the Power of Semidefinite Programming Hierarchies](https://reader035.fdocuments.in/reader035/viewer/2022081520/56816784550346895ddc9980/html5/thumbnails/12.jpg)
Eigenfunctions for Noisy Hypercube Graph
Eigenfunction Eigenvalue
………………………
Degree d multilinear polynomials
Eigenfunctions for the Noisy hypercube graph are multilinear polynomials of fixed degree!
(Noisy hypercube is a Cayley graph on , therefore its eigen functions are characters of the group )
![Page 13: On the Power of Semidefinite Programming Hierarchies](https://reader035.fdocuments.in/reader035/viewer/2022081520/56816784550346895ddc9980/html5/thumbnails/13.jpg)
Suppose indicator function of a small non-expanding set.
S has -fraction of vertices
Fraction of edges inside = =
If then, at least a -fraction of edges are inside So,
(f is close to the span of eigenvectors of G with eigenvalue )
Small-Set Expansion (SSE)
regular graph with vertex set , parameter Given:
𝑆
Conclusion:Indicator function of a small non-expanding set f = is a• sparse vector • close to the span of the large eigenvectors of G
![Page 14: On the Power of Semidefinite Programming Hierarchies](https://reader035.fdocuments.in/reader035/viewer/2022081520/56816784550346895ddc9980/html5/thumbnails/14.jpg)
HypercontractivityDefinition: (Hypercontractivity)
A subspace is hypercontractive if for all
Projector in to the subspace also called hypercontractive.
Hypercontractivity implies Small-Set Expansion projector into span of eigenvectors of with eigenvalue
is hypercontractive
No sparse vector in span of top eigenvectors of G
No small non-expanding set in . (G is a small set expander)
(No-Sparse-Vectors)Roughly, No sparse vectors in a hypercontractive subspace
because, w is -sparse “”
![Page 15: On the Power of Semidefinite Programming Hierarchies](https://reader035.fdocuments.in/reader035/viewer/2022081520/56816784550346895ddc9980/html5/thumbnails/15.jpg)
Hypercontractivity for Noisy Hypercube
Top eigenfunctions of noisy hypercube are low degree polynomials.
(Hypercontractivity of Low Degree Polynomials)For a degree multilinear polynomial on ,
(Noisy Hypercube is a Small-Set Expander)For constant the noisy hypercube is a small-set expander.
Moreover, the noisy hypercube has vertices and eigenvalues larger than .
![Page 16: On the Power of Semidefinite Programming Hierarchies](https://reader035.fdocuments.in/reader035/viewer/2022081520/56816784550346895ddc9980/html5/thumbnails/16.jpg)
[Barak-Gopalan-Hastad-Meka-Raghavendra-Steurer 2009]
For all small constant
There exists a graph (the Short Code Graph) that is a small set expander with eigenvalues
for some depending on
Short Code Graph
![Page 17: On the Power of Semidefinite Programming Hierarchies](https://reader035.fdocuments.in/reader035/viewer/2022081520/56816784550346895ddc9980/html5/thumbnails/17.jpg)
Noise Graph:
Vertices:
Edges: Connect every pair of points in hypercube separated by a Hamming distance of
x 1
-1
Short Code Graph
Has sparse cuts, but vertices -- too many vertices!
Idea: Pick a subset of vertices of the long code graph, and their induced subgraph.1. The dictator cuts still yield -sparse cuts 2. The subgraph is a small-set expander!
If we reduce to then, the number of eigenvalues will be Choice: Reed Muller Codewords of large constant degree.
![Page 18: On the Power of Semidefinite Programming Hierarchies](https://reader035.fdocuments.in/reader035/viewer/2022081520/56816784550346895ddc9980/html5/thumbnails/18.jpg)
Sparsity of Dictator Cuts
n dimensional hypercube
Connect every pair of vertices in hypercube separated by Hamming distance of
Easy: Pretty much for any reasonable subset of vertices, dictators will be sparse cuts.
1
1
1
1
-1 -1
-1
![Page 19: On the Power of Semidefinite Programming Hierarchies](https://reader035.fdocuments.in/reader035/viewer/2022081520/56816784550346895ddc9980/html5/thumbnails/19.jpg)
Preserving Small Set Expansion
Top eigenfunctions of noisy hypercube are low degree polynomials.
(Hypercontractivity of Low Degree Polynomials)For a degree multilinear polynomial on ,
(Noisy Hypercube is a Small-Set Expander)For constant the noisy hypercube is a small-set expander.
+
![Page 20: On the Power of Semidefinite Programming Hierarchies](https://reader035.fdocuments.in/reader035/viewer/2022081520/56816784550346895ddc9980/html5/thumbnails/20.jpg)
Preserving Small Set Expansion(Hypercontractivity of Low Degree Polynomials)For a degree multilinear polynomial on ,
For a degree polynomial
By hypercontractivity over hypercube,
We picked a subset and so we want,
is degree d, so and are degree .
If S is a 4d-wise independent set then,
![Page 21: On the Power of Semidefinite Programming Hierarchies](https://reader035.fdocuments.in/reader035/viewer/2022081520/56816784550346895ddc9980/html5/thumbnails/21.jpg)
Preserving Small Set Expansion
Top eigenfunctions of noisy hypercube are low degree polynomials.
We Want:Only top eigenfunctions on the subgraph of noisy hypercube
are also low degree polynomials.
Connected to local-testability of the dual of the underlying code S!
We appeal to local testability result of Reed-Muller codes[Bhattacharya-Kopparty-Schoenebeck-Sudan-Zuckermann]
![Page 22: On the Power of Semidefinite Programming Hierarchies](https://reader035.fdocuments.in/reader035/viewer/2022081520/56816784550346895ddc9980/html5/thumbnails/22.jpg)
Applications of Short Code
[Barak-Gopalan-Hastad-Meka-Raghavendra-Steurer 2009]
`Majority is Stablest’ theorem holds for the short code.
-- More efficient gadgets for hardness reductions.
-- Stronger integrality gaps for SDP relaxations.
Recap:
``How many large eigenvalues can a small set expander have?’’
-- A graph construction led to better hardness gadgets and SDP integrality gaps.
[Kane-Meka]
A SDP gap with triangle inequalities for BALANCED SEPARATOR
![Page 23: On the Power of Semidefinite Programming Hierarchies](https://reader035.fdocuments.in/reader035/viewer/2022081520/56816784550346895ddc9980/html5/thumbnails/23.jpg)
On the Power of Sum-of-Squares Proof
![Page 24: On the Power of Semidefinite Programming Hierarchies](https://reader035.fdocuments.in/reader035/viewer/2022081520/56816784550346895ddc9980/html5/thumbnails/24.jpg)
Level-8 SoS relaxation refutes UG instances based on long-code and short-code graphs
Result:
SoS hierarchy is a natural candidate algorithm for refuting UGC
Should try to prove that this algorithm fails on some instances
Only candidate instances were based on long-code or short-code graph
We don’t know any instances on which this algorithm could potentially fail!
![Page 25: On the Power of Semidefinite Programming Hierarchies](https://reader035.fdocuments.in/reader035/viewer/2022081520/56816784550346895ddc9980/html5/thumbnails/25.jpg)
Show in this proof system that no assignments for these instances exist
How to prove it? (rounding algorithm?)
Interpret dual as proof system
Level-8 SoS relaxation refutes UG instances based on long-code and short-code graphs
Result:
qualitative difference to other hierarchies: basis independence
Try to lift this proof to the proof system
We already know “regular” proof of this fact! (soundness proof)
![Page 26: On the Power of Semidefinite Programming Hierarchies](https://reader035.fdocuments.in/reader035/viewer/2022081520/56816784550346895ddc9980/html5/thumbnails/26.jpg)
Sum-of-Squares Proof System
𝑃1 (𝑧 )≥0
𝑃𝑚 (𝑧 )≥0
Axioms
…
derive
𝑄 (𝑧 )≤𝑐
(informal)
Rules
Polynomial operations“Positivstellensatz” [Stengel’74]
for any polynomial
Intermediate polynomials have bounded degree
( bounded-degree polynomials)
(c.f. bounded-width resolution, but basis independent)
![Page 27: On the Power of Semidefinite Programming Hierarchies](https://reader035.fdocuments.in/reader035/viewer/2022081520/56816784550346895ddc9980/html5/thumbnails/27.jpg)
1−𝑧=𝑧−𝑧 2+(1− 𝑧 )2
≥ 𝑧−𝑧 2
Axiom: Derive: 𝑧 ≤1
(non-negativity of squares)
≥0 (axiom)
Example
![Page 28: On the Power of Semidefinite Programming Hierarchies](https://reader035.fdocuments.in/reader035/viewer/2022081520/56816784550346895ddc9980/html5/thumbnails/28.jpg)
Non-serious issues:
Serious issues:
Cauchy–Schwarz / Hölder
Hypercontractivity
Invariance Principle
Influence decoding
can use variant of inductive proof,works in Fourier basis
typically uses bump functions, but for UG, polynomials suffice
Components of soundness proof (for known UG instances)
![Page 29: On the Power of Semidefinite Programming Hierarchies](https://reader035.fdocuments.in/reader035/viewer/2022081520/56816784550346895ddc9980/html5/thumbnails/29.jpg)
projector into span of eigenfunctions of with eigenvalue
long-code graph where
is a sum of squares
SoS proof of hypercontractivity:
![Page 30: On the Power of Semidefinite Programming Hierarchies](https://reader035.fdocuments.in/reader035/viewer/2022081520/56816784550346895ddc9980/html5/thumbnails/30.jpg)
projector into span of eigenfunctions of with eigenvalue
long-code graph where
‖𝑃𝑓 ‖44≼ 2𝑂 (1 /𝜀 ) ‖ 𝑓 ‖24
For long-code graph, projects into Fourier polynomials with degree
Stronger ind. Hyp.:
where is a generic degree- Fourier polynomialand is a generic degree- Fourier polynomial
difference is sum of squaresSoS proof of hypercontractivity:
𝐄 𝑓 2= ∑𝑆 ,|𝑆|≤𝑑
𝑓 𝑆2
![Page 31: On the Power of Semidefinite Programming Hierarchies](https://reader035.fdocuments.in/reader035/viewer/2022081520/56816784550346895ddc9980/html5/thumbnails/31.jpg)
projector into span of eigenfunctions of with eigenvalue
long-code graph where
‖𝑃𝑓 ‖44≼ 2𝑂 (1 /𝜀 ) ‖ 𝑓 ‖24
For long-code graph, projects into Fourier polynomials with degree
Stronger ind. Hyp.:
where is a generic degree- Fourier polynomialand is a generic degree- Fourier polynomial
Write and (degrees of smaller than )
𝐄 𝑓 2𝑔2=𝐄 𝑓 02𝑔02+𝐄 𝑓 12𝑔02+𝐄 𝑓 02𝑔12+𝐄 𝑓 12𝑔12+4𝐄 𝑓 0❑ 𝑓 1❑𝑔0❑𝑔1❑
…
( (ind. hyp.)
SoS proof of hypercontractivity:
![Page 32: On the Power of Semidefinite Programming Hierarchies](https://reader035.fdocuments.in/reader035/viewer/2022081520/56816784550346895ddc9980/html5/thumbnails/32.jpg)
Open Questions
Does 8 rounds of Lasserre hierarchy disprove UGC?
Can we make the short code, any shorter? (applications to hardness gadgets)
Subexponential time algorithms for MaxCut or Vertex Cover(beating the current ratios)
Thanks!