## Graph Nonisomorphism Has Subexponential Size Proofs Unless The Polynomial-Time Hierarchy Collapses (1998)

### Cached

### Download Links

- [theory.lcs.mit.edu]
- [theory.lcs.mit.edu]
- [dimacs.rutgers.edu]
- [dimacs.rutgers.edu]
- DBLP

### Other Repositories/Bibliography

Venue: | SIAM Journal on Computing |

Citations: | 110 - 6 self |

### BibTeX

@ARTICLE{Klivans98graphnonisomorphism,

author = {Adam R. Klivans and Dieter Van Melkebeek},

title = {Graph Nonisomorphism Has Subexponential Size Proofs Unless The Polynomial-Time Hierarchy Collapses},

journal = {SIAM Journal on Computing},

year = {1998},

volume = {31},

pages = {659--667}

}

### Years of Citing Articles

### OpenURL

### Abstract

We establish hardness versus randomness trade-offs for a broad class of randomized procedures. In particular, we create efficient nondeterministic simulations of bounded round Arthur-Merlin games using a language in exponential time that cannot be decided by polynomial size oracle circuits with access to satisfiability. We show that every language with a bounded round Arthur-Merlin game has subexponential size membership proofs for infinitely many input lengths unless exponential time coincides with the third level of the polynomial-time hierarchy (and hence the polynomial-time hierarchy collapses). This provides the first strong evidence that graph nonisomorphism has subexponential size proofs. We set up a general framework for derandomization which encompasses more than the traditional model of randomized computation. For a randomized procedure to fit within this framework, we only require that for any fixed input the complexity of checking whether the procedure succeeds on a given ...

### Citations

2348 | Computational Complexity - Papadimitriou - 1994 |

1186 |
Probabilistic encryption
- Goldwasser, Micali
- 1984
(Show Context)
Citation Context ...e by Nisan and Wigderson [NW94]. It is an application of Yao's observation that distinguishability from the uniform distribution implies predictability [Yao82], and uses the so-called hybrid argument =-=[G-=-M84]. Proof: We proceed by contradiction. Assume that G n is not secure against B. This means that there exists an oracle circuit D of size at most n such that either Pr [D B (G n ()) = 1] Pr [D B (... |

650 | Queries and concept learning - Angluin - 1988 |

608 |
How to generate cryptographically strong sequences of pseudo-random bits
- Blum, Micali
- 1984
(Show Context)
Citation Context ...retic point of view, however, the merits of randomization remain unclear. Can we always eliminate the use of random bits without substantially increasing the need for other resources? Blum and Micali =-=[BM84]-=- and Yao [Yao82] gave a partial answer to this question. They realized that if some of the seemingly hard algorithmic problems really are computationally intractable then, in certain settings, randomn... |

517 |
Theory and application of trapdoor functions
- Yao
- 1982
(Show Context)
Citation Context ...view, however, the merits of randomization remain unclear. Can we always eliminate the use of random bits without substantially increasing the need for other resources? Blum and Micali [BM84] and Yao =-=[Yao82]-=- gave a partial answer to this question. They realized that if some of the seemingly hard algorithmic problems really are computationally intractable then, in certain settings, randomness cannot help ... |

379 | Proofs that yield nothing but their validity or all languages in NP have zero-knowledge proof systems - Goldreich, Micali, et al. - 1991 |

309 | Arthur-Merlin games: A randomized proof system, and a hierarchy of complexity classes - Babai, Moran - 1988 |

301 | Trading group theory for randomness
- Babai
- 1985
(Show Context)
Citation Context ...W94, BFNW93, IW97]. The above observations allow us to apply the classical hardness versus randomness results to various settings, in particular to the nondeterministic setting of Arthur-Merlin games =-=[Bab85]-=-. The class AM of languages with bounded round Arthur-Merlin games forms a randomized extension of NP. The most notable problem in AM not known to be in NP is Graph Nonisomorphism [GMW91, GS89, BM88].... |

289 | Hardness vs randomness
- Nisan, Wigderson
- 1994
(Show Context)
Citation Context ...this question. They realized that if some of the seemingly hard algorithmic problems really are computationally intractable then, in certain settings, randomness cannot help much. Nisan and Wigderson =-=[NW94-=-] established a range of hardness versus randomness trade-os. They showed how to use any language in exponential time that is nonuniformly hard in an average-case sense to construct a pseudo-random ge... |

212 |
NP is as easy as detecting unique solutions
- Valiant, Vazirani
- 1986
(Show Context)
Citation Context ...domized processes from different areas of theoretical computer science. ffl The Valiant-Vazirani random hashing procedure which prunes the number of satisfying assignments of a Boolean formula to one =-=[VV86]-=-. ffl Exact learning of circuits using equivalence queries and access to an NP oracle [BCG + 96]. ffl The construction of matrices with high rigidity [Val77]. ffl The construction of polynomial-size u... |

190 |
PP is as hard as the polynomial-time hierarchy
- Toda
- 1991
(Show Context)
Citation Context ...altime Turing reductions. If there is a Boolean function f 2 E such that C B f (n) 2 2sn) , then the polynomial-time hierarchy is contained in SPP. Corollary 5.7 is strongly related to Toda's Theorem =-=[Tod-=-91] that the polynomial-time hierarchy lies in BP P. Applying our derandomization technique directly to Toda's Theorem yields: Theorem 5.8 If there is a Boolean function f 2 E such that C SAT f (n) 2... |

189 | Pseudorandom generators for space-bounded computation
- Nisan
- 1992
(Show Context)
Citation Context ...liunas et al. [AKL + 79] showed that most sequences of length O(n 3 ) over f1; 2; : : : ; n 1g are universal traversal sequences for size n, but as of now the best explicit construction, due to Nisan =-=[Nis92]-=-, 16 yields universal traversal sequences of length n O(log n) . We give evidence supporting the belief that explicit universal traversal sequences of polynomial size can be generated eciently. A stra... |

184 | Private coins versus public coins in interactive proof systems, in: Randomness and Computation - Goldwasser, Sipser - 1989 |

130 | Pseudorandom generators without the XOR lemma
- Sudan, Trevisan, et al.
- 2001
(Show Context)
Citation Context ... random to any small B-oracle circuit, we need only assume the existence of a function that cannot be computed by small B-oracle circuits. The same holds for more recent constructions by Sudan et al. =-=[STV00]-=- that provide alternate proofs for some of the earlier results [NW94, BFNW93, IW97]. The above observations allow us to apply the classical hardness versus randomness results to various settings, in p... |

122 | Gap-definable counting classes
- Fenner, Fortnow, et al.
- 1994
(Show Context)
Citation Context ...r of satisfying assignments to the ith formula of the list produced by Theorem 5.2 on input . Then 1 Y i (1 h(; i)) (13) equals the characteristic function of SAT. By the closure properties of GapP [F=-=FK94-=-], (13) belongs to GapP. Similarly, we can conditionally derandomize the result by Toda and Ogiwara [TO92] that the polynomial-time hierarchy does not add power to GapP in a randomized setting. Apply... |

117 | Hard-core distributions for somewhat hard problems
- Impagliazzo
- 1995
(Show Context)
Citation Context ...e one by Sudan et al. [STV00] we used in the proof of Theorem 3.4. The original construction by Impagliazzo and Wigderson [IW97], building on previous work by Babai et al. [BFNW93] and by Impagliazzo =-=[Imp95]-=-, would do. Instead of going through the details of the latter constructions, we provide the details for the following weaker version of Theorem 5.17, for which the Sudan et al. construction used in o... |

114 | BPP has subexponential time simulations unless EXPTIME has publishable proofs
- Babai, Fortnow, et al.
- 1993
(Show Context)
Citation Context ...randomized decision algorithms under average-case hardness assumptions and even deterministic polynomial-time simulations under the strongest of their hypotheses. Babai, Fortnow, Nisan, and Wigderson =-=[BFNW93]-=- and Impagliazzo and Wigderson [IW97] relaxed the hardness condition from average-case to worst-case. As a corollary, they showed how to simulate every polynomial-time randomized decision algorithm de... |

105 | Learning when irrelevant attributes abound - Littlestone - 1987 |

104 |
Multiparty protocols, pseudorandom generators for logspace, and time-space trade-offs
- Babai, Nisan, et al.
- 1992
(Show Context)
Citation Context ...zed Turing machines with bounded two-sided error. Theorem 5.17 If there is a Boolean function f 2 DSPACE[n] that requires branching programs of size 2sn) then BPL = L. Along the lines of Babai et al. =-=[BNS92]-=-, the pseudo-random generators behind Theorem 5.17 let us conclude: Corollary 5.18 If there is a Boolean function f 2 DSPACE[n] that requires branching programs of size 2sn) then we can construct univ... |

99 |
Graph-theoretic arguments in low-level complexity LNCS
- Valiant
- 1997
(Show Context)
Citation Context ...e 2: Overview of pseudo-random generator constructions. Exact learning of circuits using equivalence queries and access to an NP oracle [BCG + 96]. The construction of matrices with high rigidity [V=-=al7-=-7]. The construction of polynomial-size universal traversal sequences [AKL + 79]. Elaborating on thesrst application, given a propositional formula , we can construct in subexponential time a collect... |

97 |
Two theorems on random polynomial time
- Adleman
- 1978
(Show Context)
Citation Context ...N 0 , there exists an index j such that for any position i, 1 6 i 6 N , M y j (i) outputs x i with high probability. { Each machine M j runs in time polynomial in log N and 1=. By Adleman's argument [=-=Adl78]-=- we can transform the randomized procedures in the second bullet of Theorem 3.15 into circuits. This yields the following corollary. Corollary 3.16 There exists a constant c such that for any N 2 N, x... |

92 | On the power of two-point based sampling
- Chor, Goldreich
- 1989
(Show Context)
Citation Context ... ; 2 a 1g at random in a pairwise independent way, and set S i to be the [(r i mod s ` ) + 1]-st subset of of size ` (say using the lexicographical order). Known constructions of such sample spaces [=-=CG89]-=- only need O(a) random bits and can generate the samples from the random bits in space O(a). Moreover, checking whether a given sample S 1 ; : : : ; S k forms an (m; `) design can be done within the s... |

75 | Canonical labeling of graphs - Babai, Luks - 1983 |

68 | Counting classes are at least as hard as the polynomial-time hierarchy - Toda, Ogiwara - 1992 |

67 | Derandomizing Arthur-Merlin Games using Hitting Sets
- Miltersen, Vinodchandran
- 1999
(Show Context)
Citation Context ... of two graphs on n vertices are of size 2 O( p n log n) , namely the transcripts of the deterministic graph isomorphism algorithm by Babai et al. [BL83, BKL83]. Recently, Miltersen and Vinodchandran =-=[MV99]-=- managed to relax the hardness assumption in the last line of Table 1. They showed that AM collapses to NP if there exists a Boolean function in NE \ coNE with exponential nondeterministic circuit com... |

54 |
Construction of extractors using pseudorandom generators
- Trevisan
- 1999
(Show Context)
Citation Context ...ical test, the computation of the predicate on which the generator is based reduces to the computation of the test. This idea lies at the heart of the breakthrough extractor constructions of Trevisan =-=[Tre99]-=-. In fact, Miltersen [Mil98] explicitly delineates the role of relativization there. He argues that every relativizable construction of a pseudo-random generator out of a hard Boolean function yields ... |

51 | Counting complexity
- Fortnow
- 1997
(Show Context)
Citation Context ...ion is a GapP function. The class PP contains all languages L for which there exists an f 2 GapP such that x 2 L i f(x) > 0. SPP contains both P and PP. We refer the reader to the survey by Fortnow [F=-=or97-=-] for background on these counting classes. Theorem 5.2 implies: Corollary 5.4 If there is a Boolean function f 2 E such that C SAT f (n) 2 2sn) , then NP is contained in SPP. Proof: Let h(; i) denote... |

46 | On Pseudorandomness and Resource-Bounded Measure
- Arvind, Köbler
- 1997
(Show Context)
Citation Context ... point out that if the hardness assumption in thesrst row of the weak version of Table 1 fails, then exponential time collapses to the third level of the polynomial-time hierarchy. Arvind and Kobler [=-=AK97]-=- obtained similar results to those in Table 1 using nondeterministic circuits, but needed average-case hardness assumptions instead of worst-case. As opposed to oracle circuits with access to SAT, non... |

40 |
C.H.: Symmetric space-bounded computation
- Lewis, Papadimitriou
- 1982
(Show Context)
Citation Context ... continue with the next symbol of . If we can construct universal traversal sequences in logspace, then we can solve undirected graph connectivity in logspace, and symmetric logspace equals logspace [=-=LP82]-=-. However, we do not know how to generate universal traversal sequences in logspace or even in polynomial time. Aleliunas et al. [AKL + 79] showed that most sequences of length O(n 3 ) over f1; 2; : :... |

39 | P=BPP unless E has subexponential circuits: derandomizing the XOR Lemma
- Impagliazzo, Wigderson
- 1997
(Show Context)
Citation Context ...erage-case hardness assumptions and even deterministic polynomial-time simulations under the strongest of their hypotheses. Babai, Fortnow, Nisan, and Wigderson [BFNW93] and Impagliazzo and Wigderson =-=[IW97]-=- relaxed the hardness condition from average-case to worst-case. As a corollary, they showed how to simulate every polynomial-time randomized decision algorithm deterministically in subexponential tim... |

36 |
Prediction preserving reducibility
- Pitt, Warmuth
- 1990
(Show Context)
Citation Context ...r provides a counterexample and the learner continues. A fundamental question is whether we can eciently learn Boolean circuits in this model. If P = NP, we can; if one-way functions exist, we cannot =-=[PW90]-=-. Without any complexity theoretic assumption, Bshouty et al. [BCG + 96] showed that access to an NP oracle and to a source of randomness suce to eciently learn Boolean circuits. We give evidence that... |

27 | Computational complexity and classification of finite simple groups - BABAI, KANTOR, et al. - 1983 |

24 | A note on matrix rigidity
- Friedman
- 1993
(Show Context)
Citation Context ...(n r) 2 and almost all matrices over asxedsniteseld have rigidity 3 n r) 2 = log n). The best known explicit constructions achieve rigiditysn 2 =r) over inniteselds [PV91, Raz] and n 2 r log( n r )) [=-=Fri90]-=- oversniteselds. These are not sucient to obtain circuit lower bounds using Theorem 5.11. Under a hardness assumption, our derandomization technique will give an explicit construction to which Theorem... |

22 | Isolation, matching and counting uniform and nonuniform upper bounds
- ALLENDER, REINHARDT, et al.
- 1999
(Show Context)
Citation Context ...ent approach which only needs O(s(n)) = O(log n) space. The construction goes along the lines of Impagliazzo and Wigderson [IW97]. Allender informed us that Wigderson showed him the same construction =-=[ARZ99-=-]. For any `(n) 2 (log n), it yields a (log n; `(n)) design of size n over f1; 2; : : : ; s(n)g for some positivesand s(n) 2 O(log n). Note that this design is smaller than the one obtained through th... |

21 | Constructions of near-optimal extractors using pseudo-random generators
- Trevisan
- 1999
(Show Context)
Citation Context ... generator, the computation of the predicate on which the generator is based reduces to the computation of the test. This idea lies at the heart of the breakthrough extractor construction of Trevisan =-=[Tre98]-=-. In [Mil98], Peter Bro Miltersen delineates the role of relativization in these constructions. Acknowledgments Wewant to express our sincere gratitude to Oded Goldreich, Madhu Sudan, Luca Trevisan, a... |

15 | Derandomizing Arthur-Merlin games under uniform assumptions
- Lu
- 2000
(Show Context)
Citation Context ... DTIME[t] DSPACE[t ]: (18) They concluded from their main results that every language in BPP can be decided in deterministic polynomial time for innitely many input lengths unless (18) holds. Lu [Lu=-=00]-=- noticed that the inclusion (18) even follows if E has B-oracle circuits of size 2 o(n) for some oracle B that is decidable in linear space. Therefore, setting B = SAT, the weak version of Theorem 3.1... |

11 |
propositional formulas represent nondeterministic computations
- Short
- 1988
(Show Context)
Citation Context ...h restricted access. 6 3.1 Proof of Lemma 3.1 Let L be a language satisfying (2) and (3), and consider L 0 = fhx; yi j y 2 f0; 1g t(jxj) and 9 z 2 f0; 1g t(jxj) : M(hx; y; zi) = 1g: By Cook's Theorem =-=[Coo88]-=-, since L 0 2 NTIME[n], there exists a circuit of size t 0 (n) 2 O(t(n) log 2 t(n)) that many-one reduces L 0 to SAT on inputs hx; yi with x 2 f0; 1g n and y 2 f0; 1g t(n) . For anysxed value of x 2 f... |

11 | On Rigid Matrices - Razborov - 1989 |

9 |
N ˜ P/qpoly-natural proofs
- Super-bits
- 1997
(Show Context)
Citation Context ...ized extension of NP. The most notable problem in AM not known to be in NP is Graph Nonisomorphism [GMW91, GS89, BM88]. Derandomizing AM requires security against nondeterministic adversaries. Rudich =-=[Rud97]-=- pointed out that pseudo-random generators in the traditional cryptographic setting where an adversary has more resources than the generator cannot hope to have this property because the adversary can... |

8 | Isolation, matching, and counting - Allender, Reinhardt - 1998 |

6 | Computation of rigidity of order n =r for one simple matrix - Pudlak, Vavrn - 1991 |

6 | Extracting all the randomness from a weakly random source - Vadhan - 1998 |

5 |
Recent advances towards proving P
- Clementi, Rolim, et al.
- 1998
(Show Context)
Citation Context ...programs of size 2sn) , then BPL = L, where BPL denotes the languages recognizable in randomized logspace with bounded two-sided error. This answers a question raised by Clementi, Rolim, and Trevisan =-=[CRT98]-=-. As a corollary to the proof, under the same hypothesis, we can generate universal traversal sequences in logspace. 1.1 Organization Section 2 introduces our notation. In Section 3 we generalize the ... |

2 | Computational complexity and the classi of simple groups - Babai, Kantor, et al. - 1983 |

2 | Derandomizing Arthur-Merlin games - Melkebeek - 1998 |

1 |
Relativizable pseudorandom generators and extractors. Comment to ECCC
- Miltersen
- 1998
(Show Context)
Citation Context ...f the predicate on which the generator is based reduces to the computation of the test. This idea lies at the heart of the breakthrough extractor constructions of Trevisan [Tre99]. In fact, Miltersen =-=[Mil98]-=- explicitly delineates the role of relativization there. He argues that every relativizable construction of a pseudo-random generator out of a hard Boolean function yields an extractor. We showed that... |

1 | Computation of rigidity of oerder n =r for one simple matrix - Pudlk, Vavn - 1991 |

1 | Computation of rigidity of oerder n =r for one simple matrix - Pudl'ak, Vavr'in - 1991 |