Some 3CNF properties are hard to test

被引:76
作者
Ben-Sasson, E [1 ]
Harsha, P
Raskhodnikova, S
机构
[1] Technion Israel Inst Technol, Dept Comp Sci, IL-32000 Haifa, Israel
[2] Harvard Univ, Dept Engn & Appl Sci, Cambridge, MA 02138 USA
[3] MIT, Comp Sci Lab, Cambridge, MA 02139 USA
[4] Microsoft Corp, Res, Mountain View, CA 94043 USA
[5] Weizmann Inst Sci, IL-76100 Rehovot, Israel
关键词
sublinear algorithms; lower bounds; property testing; CNF formulae; locally testable codes;
D O I
10.1137/S0097539704445445
中图分类号
TP301 [理论、方法];
学科分类号
081202 ;
摘要
For a Boolean formula phi on n variables, the associated property P-phi is the collection of n-bit strings that satisfy phi. We study the query complexity of tests that distinguish ( with high probability) between strings in P. and strings that are far from P. in Hamming distance. We prove that there are 3CNF formulae ( with O(n) clauses) such that testing for the associated property requires O( n) queries, even with adaptive tests. This contrasts with 2CNF formulae, whose associated properties are always testable with O(root n) queries [ E. Fischer et al., Monotonicity testing over general poset domains, in Proceedings of the 34th Annual ACM Symposium on Theory of Computing, ACM, New York, 2002, pp. 474 - 483]. Notice that for every negative instance (i.e., an assignment that does not satisfy.) there are three bit queries that witness this fact. Nevertheless, finding such a short witness requires reading a constant fraction of the input, even when the input is very far from satisfying the formula that is associated with the property. A property is linear if its elements form a linear space. We provide sufficient conditions for linear properties to be hard to test, and in the course of the proof include the following observations which are of independent interest: 1. In the context of testing for linear properties, adaptive two-sided error tests have no more power than nonadaptive one-sided error tests. Moreover, without loss of generality, any test for a linear property is a linear test. A linear test verifies that a portion of the input satisfies a set of linear constraints, which de. ne the property, and rejects if and only if it finds a falsified constraint. A linear test is by definition nonadaptive and, when applied to linear properties, has a one-sided error. 2. Random low density parity check codes ( which are known to have linear distance and constant rate) are not locally testable. In fact, testing such a code of length n requires Omega(n) queries.
引用
收藏
页码:1 / 21
页数:21
相关论文
共 23 条
[1]   Testing satistiability [J].
Alon, N ;
Shapira, A .
JOURNAL OF ALGORITHMS-COGNITION INFORMATICS AND LOGIC, 2003, 47 (02) :87-103
[2]   Regular languages are testable with a constant number of queries [J].
Alon, N ;
Krivelevich, M ;
Newman, I ;
Szegedy, M .
SIAM JOURNAL ON COMPUTING, 2001, 30 (06) :1842-1862
[3]   Efficient testing of large graphs [J].
Alon, N ;
Fischer, E ;
Krivelevich, M ;
Szegedy, M .
COMBINATORICA, 2000, 20 (04) :451-476
[4]  
Ben-Sasson E., 2003, P 35 ANN ACM S THEOR, P345
[5]  
Ben-Sasson Eli., 2003, P 35 ANN ACM S THEOR, P612, DOI 10.1145/780542.780631
[6]  
Ben-Sasson Eli, 2004, Proceedings of the Thirty-Sixth Annual ACM Symposium on Theory of Computing, P1, DOI DOI 10.1145/1007352.1007361
[7]   SELF-TESTING CORRECTING WITH APPLICATIONS TO NUMERICAL PROBLEMS [J].
BLUM, M ;
LUBY, M ;
RUBINFELD, R .
JOURNAL OF COMPUTER AND SYSTEM SCIENCES, 1993, 47 (03) :549-595
[8]  
Bogdanov A, 2002, ANN IEEE SYMP FOUND, P93, DOI 10.1109/SFCS.2002.1181886
[9]   MANY HARD EXAMPLES FOR RESOLUTION [J].
CHVATAL, V ;
SZEMEREDI, E .
JOURNAL OF THE ACM, 1988, 35 (04) :759-768
[10]   Testing graphs for colorability properties [J].
Fischer, E .
RANDOM STRUCTURES & ALGORITHMS, 2005, 26 (03) :289-309