PCP Characterization of NP: Towards Polynomially Small ErrorProbability - PowerPoint PPT Presentation

1 / 22
About This Presentation
Title:

PCP Characterization of NP: Towards Polynomially Small ErrorProbability

Description:

PCP Characterization of NP: Towards Polynomially Small Error ... Conj[BGLR]-93: Thm[AS,ALMSS]-92: Gap-SAT[O(1) , f , 2-f ] for. f(n)=log n, =const 1/4 ... – PowerPoint PPT presentation

Number of Views:28
Avg rating:3.0/5.0
Slides: 23
Provided by: guyki
Category:

less

Transcript and Presenter's Notes

Title: PCP Characterization of NP: Towards Polynomially Small ErrorProbability


1
PCP Characterization of NP Towards Polynomially
Small Error-Probability
Irit Dinur
Eldar Fischer
Guy Kindler
Ran Raz
Shmuel Safra
2
The class NP
?L?
Membership in L is verifiable in polynomial time,
given a correct membership proof .
3
Cooks Theorem3-SAT?NPC
Input
LocalTests
VariableSet
4
?L
Gap-SAT
yes Local-tests are satisfiable no At most
? are satisfiable
5
Gap-SATD,V,?
BooleanFunctions
  • Given
  • ? A set of n local-tests over ?D variables.
  • Variable range size2V
  • Distinguish between
  • ? is satisfiable.
  • ? is not even ? satisfiable.

6
ThmAS,ALMSS-92
Gap-SATO(1) , 1 , 1/2 is NP-hard.
ConjBGLR-93
Gap-SATO(1) , f , 2-f for any f(n)?O(log
n)is NP-hard.
7
ThmRaSa-97
Gap-SATO(1) , f , 2-f for f(n)log? n,
?constlt1/4 is NP-hard.
Our Result
Gap-SATO(1) , f , 2-f for f(n)log? n, ?any
constlt1 is NP-hard.
8
Gap-QSD,F,?
  • Given
  • ? A set of n Quadratic-Equations over ?D
    variables.
  • Variables range in the field F.
  • Distinguish between
  • ? is satisfiable.
  • ? is not even ? satisfiable.

9
Main Theorem
Gap-QSO(1) , F , 2/F for F2log? n, ?any
constlt1 is NP-hard.
XY2
X2-YZ
ZXY
10
ThmGJ
Exact-QS is NP-hard (for any F).
Simple
ThmHPS-93
Gap-QSn, F, 2/F is NP-hard.
Our Goal
Gap-QSO(1) , F , 2/F
Constant numberof variables in each equation
11
Typical Proof Structure
Low Degree TestD,V,?
Gap-SATD,V,?
12
Low-Degree TestD,V,?
r,d-LDF (low-degree function)
A degree r polynomial PFd?F
Representation
A set of variables, fixing P induces an
assignment for each. Variable-range size is 2v.
Example
13
Low-Degree Test
  • A set of local-tests, each accessing D variables.
  • Local-tests can accept or reject.


LocalTests
RepresentationVariables
14
D,V, ?-test
? Assignment ? P1,,P1/?2
A randomly chosen ?(xi1,..,xiD) satisfies
P1
P8
P5
15
D,V, ?-test
? Assignment ? P1,,P1/?2
A randomly chosen ?(xi1,..,xiD) satisfies
P1
P8
P5
16
xi1,..,xiD agree with no Pi yet ? accepts lt ?
probability
17
History of Tests
  • BFL,BFLS,BLR,FGLSS
  • RuSu - r , logF , 1/2
  • ASa,ALMSS - O(1), logF , 1/2
  • RaSa - O(1), r2logF , F-c
  • ASu - O(1) , r logF , F-c
  • Our Test - O(1) , logF , F-c

? is exponentially small in the number of bits
accessed.
18
Typical Proof Structure
Low Degree TestD1 , V1 , ?1
CompositionRecursion
Gap-SATD1 , V1 , ?1
Low Degree TestD,V, ?
19
The RaSa Test
  • Representation Variable per point,
    Variable per plane.
  • Local-Tests Compare a point against a plane
    containing it.
  • Parameters 2 , r2logF , F-c

r,d
r,2
20
The Embedding Technique
Y
X
r,2
b,2logbr
b3
Y3Y3 Y9Y9 Y27Y27
X3X3 X9X9 X27X27
21
The Embedding
X12Y56(X3)1(X9)1(Y)2(Y27)2
Y3Y3 Y9Y9 Y27Y27
X3X3 X9X9 X27X27
(X3)1(X9)1(Y)2(Y27)2
22
Dimension Reduction
Degree Reduction
23
Future Work?
  • Full BGLR conjectureGap-SAT(O(1) , f , 2-f )
    for f(n)log n, is NP-hard.
  • Gap-QS analogue gap-QS(O(1) , F , 2/F)For
    F?n, is NP-hard.
  • Getting D2.
  • Showing hardness for other approximation problems
    by similar techniques - SSAT
Write a Comment
User Comments (0)
About PowerShow.com