Lower bounds on the complexity of branching programs

Okol'nishnikova E.A.
Sobolev Institute of Mathematics of Siberian Branch of Russian Academy of Sciences
(Novosibirsk)

The problem of finding nontrivial lower bounds on the complexity of "well-defined" Boolean functions is one of the most important problems in the mathematical theory of complexity. In the paper the computation of characteristic functions of Reed-Muller codes by nondeterministic branching programs is considered. It is known that almost all Boolean functions can't be computed by nondeterministic branching program of the size less than , where is a constant. But the best known lower bounds on the complexity for nondeterministic branching programs is the bound , obtained by Pudlak [4] by the use of Nechiporuk method. From the result of Razborov [2] for switch-rectifier circuits follows the bound on the complexity of some symmetric Boolean functions for nondeterministic branching programs.

In [1] the bound on the complexity of characteristic functions of BCH-codes with a code distance for deterministic branching programs was obtained. In the paper we use a modification of a method from [1] to obtain nonlinear lower bounds on the complexity of characteristic functions of Reed-Muller codes for nondeterministic branching programs. Deterministic branching programs are a special case of nondeterministic branching programs. Therefore all obtained bounds are valid for deterministic branching programs too.

We use the programs with restrictions on the number of occurance of each variable in each path (namely, branching read--times only programs) for obtaining lower bounds for programs without restrictions.

By denote the complexity of a Boolean function for nondeterministic branching programs.

Theorem 1 Let be a Boolean function and , , be a constant. If for each subset of variables , , , there exists a substitution of constants in the set , such that the complexity of the function for nondeterministic branching read--times only programs is not less than , where is a growing function, then .

Let , and be natural numbers, such that . We introduce the following denotation:

(1)
(2)
(3)

Let us consider all possible representations of the function as

(4)

where $X_1 $, and are nonintersecting sets; ; , and , , .

By we denote the minimum number of disjunctive terms in representation (4).

Theorem 2 Let , be a sequence of Boolean functions, be a growing function, and , be a constant. If for any subset of variables , , , there exists a substitution of constants in the set $X_0 $, and natural numbers , , , such that , and , calculated by formulas (1) - (3) are positive, then


By we denote the maximum number of "ones" of a Boolean function that belong to - faces of the cube.

Theorem 3 Let , , be a sequence of Boolean functions, be an growing function, and , , be a constant. Then the following relation is valid for the complexity of for nondeterministic branching programs without restrictions


where , .

By denote a Reed-Muller code with a code distance [3]. A total number of variables of the characteristic function of this code is equal to . In order to obtain the main result of the paper we also use the generalized Hamming weights for Read-Muller codes considered by Wei [5].

Theorem 4 Let as , . Then


This research was supported by the Russian Foundation for Basic Research (Grant 00-01-00874) and the Federal program "Integration" (Project AO-110).

Bibliography

  1. E.A.Okol'nishnikova (1991) Lower bounds on complexity for the realization of characteristic functions of binary codes by binary programs (in Russian), Metody Diskret. Anal. 51, 61-83 (see also: Siberian Adv. Math. 3 (1993), No. 1, 152-166).
  2. Razborov A. A. Lower bounds on the complexity of symmetric Boolean functions for switching-rectifier circuits // Matemat. zametki -- 1990. -- V. 48, No. 6. -- P. 79-90.
  3. MacWilliams F.J., Sloane N.J.F. The theory of Error-Correcting Codes, Amsterdam: North-Holland, 1977.
  4. Pudlak P. The hierarchy of Boolean circuits // Comput. Artificial Intelligence -- 1987. -- V. 6, No. 5. -- P. 449-468.
  5. Wei V.K. Generalized Hamming weights for linear codes. // IEEE Transactions on information theory, V. 37, No. 5, 1991, pp. 1412-1418.


Ваши комментарии
[SBRAS]
[Головная страница]
[Конференции]
[СО РАН]

© 2001, Сибирское отделение Российской академии наук, Новосибирск
© 2001, Объединенный институт информатики СО РАН, Новосибирск
© 2001, Институт вычислительных технологий СО РАН, Новосибирск
© 2001, Институт систем информатики СО РАН, Новосибирск
© 2001, Институт математики СО РАН, Новосибирск
© 2001, Институт цитологии и генетики СО РАН, Новосибирск
© 2001, Институт вычислительной математики и математической геофизики СО РАН, Новосибирск
© 2001, Новосибирский государственный университет
Дата последней модификации Wednesday, 03-Oct-2001 17:17:43 NOVST