The History and Status of the P versus NP Question

Michael Sipser*
Department of Mathematics
Massachusetts Institute of Technology
Cambridge MA 02139

As long as a branch of science offers an abundance of problems,

so long it is alive; a lack of problems foreshadows extinction or
the cessation of independent development.

DAVID HILBERT, ilom a lecture delivered before the

International Congress of Mathematicians at Paris in 1900.

When?

JURIS HARTMANIS

1 Significance somewhere in a large space of possibilities. Comput-

ers may greatly speed up the search, but the extremely
The P versus NP question grew out of developments in large spaces that really do occur in cases of interest
mathematical logic and electronic technology during the would still require geologic time to search exhaustively
middle part of the twentieth century. It is now consid- even on the fastest machines imaginable. The only way
ered to be one of the most important problems in con- to solve such cases practically would be to find a method
temporary mathematics and theoretical computer sci- which avoided searching by brute force. Roughly speak-
ence. The literature on this subject is diverse but not ing, the P versus NP question asks whether, in general,
huge it is possible for the student to acquaint himself such a method exists.
with all of the important relevant work in a semester or
This question has attracted considerable attention.
two. In writing thw article, I have attempted to orga-
Its intuitive statement is simple and accessible to non-
nize and describe this Literature, including an occssiom-d
specialists, even those outside of science. By embracing
opinion about the most fruitful directions, but no tech-
issues in the foundations of mathematics as well as in the
nical details.
practice of computing, it gains something in character
In the first half of this century, work on the power of
beyond that of a mere puzzle, but rather with appar-
formal systems led to the formalization of the notion of
ently deeper significance and broader consequences.
algorithm and the realization that certain problems are
Church and Turing, in their work on Hilberts
algorithmically unsolvable. At around this time, fore-
entschiedungsproblem, have shown that testing whether
runners of the programmable computing machine were
an assertion has a proof is idgorithmically unsolvable.
beginning to appear. As mathematicians contemplated
The P versus NP question can be viewed as a more fini-
the practical capabilities and limitations of such devices,
tary version of the entschiedungsproblem. Suppose we
computational complexity theory emerged from the the
wish to test whether an assertion has a short proof. Of
ory of algorithmic unsolvability.
course this can be solved by a brute force algorithm, but
Early on, a particular type of computational task be-
can it be done efficiently?
came evident, where one is seeking an object which lies
The entschiedungsproblem does have practical im-
*Supported by NSF Grant 8912586 CCR and Air Force Con-
portance in addition to its philosophical signitlcance.
tract AFOSR S9-0271.
Permission to copy without fee all or part of thie material ie Mathematical proof is a codification of more generaJ
granted provided that the copiae ara not mada or distributed for human reasoning. An automatic theorem prover would
direct commercial advantsge, the ACM copyright notice and the have wide application within computer science, if it op
titla of the publication and ite date appear, and notice is givan Even though this is hope
erated efficiently enough.
that copying is by permission of the Association for Computing
Machinery. To copy otherwise, or to republish, requires a fee
and/or specific permission.
24th ANNUAL ACM STOC - 5/92/VICTORIA, B. C., CANADA
@1992 ACM 0.8$791.51 2.7/9210004/0603 ...$~ .5Q
603
less in general, there may be important special cases Neumann,l wrote about the issue in a remarkably mod-
which are solvable. It would be nice if Churchs or Tur- ern way, formulating it in terms of the time required
ings proofi gave us some information about where the on a Turing machine to test whether a formula in the
easier cases might lie. Unfortunately, their arguments predicate calculus has a proof of length n. He also ssked
rest on self-reference, a contrived phenomenon which how much we csn improve upon brute force, in general,
never appears spontaneously. This does not tell us what when solving combinatorial problems. It is not known
makes the problem hard in interesting cases. Conceiv- whether von Neumann, who was dying of caacer at the
ably, a proof that P is not equal to NP would be more time, thought about Godels letter or responded to it.
informative. Von Neumann does appear to have had some awareness
of the undesirability of brute force search. A couple of
years earlier he had written a paper giving an algorithm
2 History avoiding brute force search for the assignment problem
~053] .

Among the many equivalent formulations of the P ver- Edmonds [Ed65] gave the first lucid account of the

sus NP question, the following one is particularly simple issue of brute force search appearing in western litera-
ture, The main contribution of his paper was a polyn~
to describe.
mial time algorithm for the maximum matching prob-
Is them. an algorithm which, when presented with an lem. This algorithm itself still stands ss a beautiful ex-
undimct ed graph, determines whether it contains a com- ample of how brute force can sometimes be avoided, if
plete subgraph on hay its nodes, and whose running one is clever enough. Edmonds felt the need to explain
time is bounded above by a polynomial in the number why his was a result at all, when there is an obvious
of nodes? finite algorithm for solving the problem. In so doing he
discussed the problem of brute force search in general
By the term algorithm we mean a finite set of in- as an algorithmic method.
structions on any of a number of related models of com- In the 1950s, researchers in the Soviet Union were
putation, e.g., the multi-tape Turing math inc. The aware of the issue of brute force search, called pewbor
term P, for Polynomial time, refers to the class of lan- in Russian. Yablonski ~a59a] described the issue for
guages languages decidable on a Turing machine in time general problems and focused on the specific problem
bounded above by a polynomial. It is invariant over of constructing the smallest circuit for a given function.
reasonable computational models. NP, for Nondeter-
Curiously, in a later paper ~a59b] he obscured the issue
ministic Polynomial time, refers to the analogous class by claiming to prove that perebor is unavoidable for this
for nondeterministic Turing machines. NP consists of problem. That paper, whose resin theorem makes no
those languages where membership is verijiabie in poly- reference to the notion of algorithm, certainly did not
nomial time. The question of whether P ia equal to NP do as claimed, at least in the sense we would mean it
is equivalent to whether an NP-complete problem, such
today. Trakhtenbrot sheds some light on this murky
as the clique problem described above, can be solved in
situation in his survey of early Russian work on perebor
polynomial time. The books by Hopcroft and Unman [Tr84].
[HU79] and Garey and Johnson [GJ79] contain excellent
introductions to this subject.
2.2 P and NP

Several early papers, notably those of Rabin ~a60],

2.1 Brute Force Search
Hartmanis and Stearns [HS65], and Blum [B167], pro-
The idea of using brute force search to solve certain posed and developed the notion of measuring the com-
problems is certainly very old. In principle, many nat- plexity of a problem in terms of the number of steps
urally occurring problems may be solved in this way, required to solve it with an algorithm. Cobham [C064],
though if the search space is Isxge, this becomes obvi- Edmonds [Ed65], and Rabin [Ra66] suggested the class
ously impractical. People realized this difficulty in par- P ss a reasonable approximation to the class of realis-
ticular cases, and occasionally were able to find alterna- tically solvable problems,
tive methods. One of the early accomplishments of the The importance of the class NP stems from the large
theory of algorithms wss to recognize brute force search number of computational problems contained within.
as a phenomenon independent of any specitic problem. Cook [C071] and Levin ~e73] first defined this class and
A number of researchers, working separately, hit upon proved the existence of complete problems. Somewhat
this idea. 1A tr~at.jon of G6dels letter along with quotatims ~m @
Godel, in a recently discovered 1956 letter to von number of historical papers are included in an appendix.

604
earlier, Edmonds @d65b] captured the essence of the of languages efficiently computable in the presence of
notion of NP with his good characterization whereby that oracle, Baker, Gill, and Solovay [BGS75] demon-
an assistant can convince his supervisor of the solution strated an oracle relative to which P is equal to NP
to a problem. Since the solution may be either positive and another relative to which they differ. A step-by-
or negative in the case of a problem involving language step simulation of one machine by another, such as that
membership, Edmondss notion is generally associated used in a conventional diagonalization, still succeeds in
with the class NP n CONP. Karp [Ka72] demonstrated the presence of an oracle, as the simulator needs only
the surprisingly wide extent of NP-completeness when query the oracle whenever the simulated machine does.
he showed that many familiar problems have this prop- Any argument which relies only on step-by-step simula-
ert y. tion to collapse or separate P and NP would thus also
apply in the presence of any oracle. As this would con-
tradict the BGS theorem, step-by-step simulation will
2.3 Explicit Conjecture
not be sufficient to settle the question.
Cook first precisely formulated the P versus NP conjec- The above is a heuristic argument as the notion of
ture in his 1971 paper. More or less close approximw step-by-step simulation) is not precisely defined. In
tions to it appeared somewhat earlier in the papers of recent papers, Lund, Fortnow, Karloff, and Nisan, and
Edmonds, Levin, Yablonski, and in Godels letter, Shamir ~FKN90,Sh90] introduced a new type of sim-
ulation to prove equivalence for the complexity clssses
1P snd PSPACE. This simulation is not of the step-by-
3 Status step sort, but is rather indirect, and indeed does not
relativize to all oracles, as can be seen from the paper
3.1 Diagonalization and Relativization of Fortnow and Sipser [FS88]. Conceivably, an indi-

Rabin [Ra60] and Hartmanis and Stearns [HS65] proved rect simulation such se this could be combined with the

the existence of decidable problems with arbitrarily high diagonalization method to separate certain complexity

computational complexity. They used a classical diag- classes.

onalization, giving an algorithm which, by design, ac-
cepts a language differing from any language of low com- 3.2 Independence
plexity. This technique allows us to conclude, for ex-
ample, that there exist problems requiring exponential Following the celebrated work of Godel and Cohen
time. Problems constructed in this way are artificial, we know that certain mathematical questions, such as
having no meaning or significance outside of their role the Continuum Hypothesis, cannot be answered within
in this theorem. Meyer and Stockmeyer [MS72] used accepted formal systems of mathematical reasoning.
this theorem to demonstrate natural problems provably There has been speculation that the P versus NP ques-
outside of P by showing that they are complete for high tion may be unresolvable in a similar sense, This would
complexity classes. mean that if there were no polynomial time algorithm
A priori, it might seem possible to use diagonalization for the clique problem, we would never be able to prove
to prove that some problem in NP requires exponential the nonexistence. On the other hand if such an algo-
time. One would give a nondeterministic polynomial rithm were to exist, we would not be able to prove its
time Turing machine which, over the course of its in- performance. I will mention here some of the literature
puts, has an opportunity to run each of the determin- on independence, though I state at the outset my opin-
istic polynomial time Turing machines and arrange to ion that our current inability to resolve this question
accept a differing language. This simple argument fails derives from insufficient brain power, rather than any
because the nondeterministic machine, running in some lack of power in our formal systems.
fixed n~ time, is not able to simulate a deterministic In a sense, the work on relativization already suggests
machine whose time bound is a larger polynomial. Pos- a sort of limited independence for the P versus NP ques-
sibly there is a more clever nondeterministic simulation tion. One might be able to formulate an axiomatic sys-
of the deterministic machine which would allow the non- tem corresponding to pure recursive function theory, in
deterministic machine to carry out the diagonalization which the results of Baker, Gill, and Solovay presumably
process. The method of relativization, described below, would show that the question is unresolvable. However,
suggests that this will not be straightforward. this would be very far from establishing independence
In a relativized computation the machine is provided with respect to strong systems such as number theory
with a set, called an oracle, and the capability to deter- or set theory.
mine membership in this set without cost. For each or- In the direction of stronger independence, a number
acle, there is associated the relativized complexity class of papers have attempted to offer a beginning, notably:

605
[HH76] giving an oracle under which the P versus NP inference. Resolution is a proof system for showing that
question is independent of set theory and [Li78,DL79, formulas sxe unsatisfiable. The axioms are the clauses
Sa80,Bu86,CU88, BH91] establishing independence and of the formula. In the inference rule, if clauses (Z A a)
related results within weak fragments of number theory and (Z A /3) are already present, then the new clause
or related systems. While I do not profess expertise in (a A ~) maybe added. Unsatisfiability is proved when
this area, I doubt whether the results obtained so far in the empty clause ia derived. The complexity of the proof
this vein do more than raise the possibility of meaningful is the number of clauses added. Any unsatisfiable for-
independence. mula has a resolution proof. Haken, solving a question
that had been open for many years, proved that there
are formulas requiring exponential size proofs [Ha85].
3.3 Expressibility in Logic His work followed that of Tseitin [Ts62] who had solved
Another potential application of mathematical logic regular resolution, a further restricted form (see also
arises because it provides a different way of defining [Ga77]). Extended resolution, a generalized form, re-
familiar complexity classes. Fagin [Fa74] showed that mains open (see [C075]). Kozen [K077] considered other
NP equals the class of languages representing struc- proof systems and established lower bounds for them.
tures definable by X; second-order sentences. He and Proof systems are related to the class NP. If all
Jones and Selman [JS74] showed that the clam NEXP- unsatisfiable formulas had polynomial length resolu-
TIME equals the class of languages representing the col- tion proof% then it would easily follow that NP=coNP.
lections of cardinalities of universes of such structures. Hakens result proves a special case of the NP#coNP
Much earlier, Scholz [SC52] had considered such collec- conjecture, showing that one particular nondeterminia-
tions and asked whether they could be characterized tic algorithm is superpolynomial.
(see also Asser [As55]). Immerman [Im87] has extended
the correspondence with finite model theory to P, NL,
3.5 Circuit Complexity
and other complexity classes, using first-order calculus
adorned with various operators. Hope that this recod- Shannon [Sh49] proposed the size of Boolean circuits
ing into logic may help to settle questions in complexity as a measure of the complexity of functions. Savage
theory has been partly born out by Immermans dis- [Sa72] demonstrated a close relationship between cir-
covery ~m88] that NL=coNL, inspired by certain ob- cuit size snd time on a Turing machine. Circuits are
servations concerning first-order definability following an appealing model of computation for proving lower
an earlier weaker result of [LJK87]. Szelepcs6nyi [S288] bounds. Their relatively simple definition renders them
independently obtained NL=coNL at around the same more amenable to combinatorial analysis and allows for
time. natural variations and restrictions. This has been key
to achieving a number of important results. Mrmy re-
searchers consider circuit complexity to be the most vi-
3.4 Restricted Systems
able direction for resolving the P versus NP question.
The difficulty in proving lower bounds on complexity We will survey recent work in this area, with an eye
stems from the richness of the class of algorithms. One towards this goal. The survey paper of Boppana and
approach that allows for partial progress is to restrict Sipser [BS90] contains details of much of this work, de-
the class of algorithms considered. There is a substan- scribed in a readable fsshion. Dunne [Du88] and We-
tial literature on bounds in various restricted models. gener [We87] have recently written books on Boolean
These fall into two categories, which may be called nat- complexity.
ural models and handicapped models. In the former, a A circuit (or straight line program) over a basis (typ-
model is selected which allows only for operations spe- ically AND, OR and NOT) is a sequence of instructions,
cially tailored to the problem at hand. This includes each producing a function by applying an operation
sorting models, polynomial evaluation models, various from the bssis to previously obtained functions. Ini-
models oriented towards data structures, and others. tially, we start out with the functions naturally associ-
In the latter, we seek to maintain the generality of the ated with each of the input variables. If each function is
model, but weaken it sufficiently so that an interesting used at most once, then the circuit is called a formula.
lower bound is obtainable. Both approaches bear a con- By Savages theorem, any problem in P has a poly-
nection to separating complexity classes. We will treat nomial size family of circuits. Thus, to show that a
handicapped models shortly in the section on circuit problem is outside of P it would suffice to show that its
complexity. Here we consider restricted proof systems circuit complexity is superpolynornial. A simple count-
as a form of natural model. ing argument shows that most Boolean functions on n
A proof system is a collection of axioms and rules of variables require exponential circuit size [Sh49,Mu56].

606
The best lower bound we have to date for a problem in put in the form OR of small ANDS. This interchange
NP is 3n o(n), due to N. Blum [B184]. It does not allows adjacent levels of ORS to follow one another and
seem likely that the techniques used to get this result therefore to merge. A circuit may thus be simplified in-
will extend to significantly better bounds. ductively, while preserving the fact that it computes a
In [Va90], Valiant suggests that a direct attempt to parity function. Ajtais argument is also an induction.
prove lower bounds on the size of Boolean circuits may He showed that any set definable by a low depth circuit
not be the right approach. There are natural algebraic is well approximated by a set of a very special form, a
generalizations of certain NP-complete problems, such disjoint union of large cylinders. A cylinder is a maximal
as the Hamiltonian cycle problem, to an arbitrary ring. collection of strings which all agree on some collection
We may also generalize circuits over the bssis AND, XOR of variables. Clearly the parity function cannot be ap-
for such problems to arbitrary rings. Circuits which proximated in this way. His main combinatorial lemma
solve the generalized problem form a proper subclass of shows that the property of being well approximable is
the circuits which merely work correctly in the Boolean closed under polynomial union and intersection.
case. Aa such it may be essier to prove strong lower Yao ~a85] combined probabilistic restriction with
bounds in this restricted csse. Valiant points out that a type of approximation to give an exponential lower
our failure to do this even then argues poorly for our bound for constant depth parity circuits. H&tad [H&86]
chances in the Boolean case. He suggests directing our simplified and strengthened this bound by improving
energies toward solving the algebraic csae first. the core of the FSS argument using a type of analysis
There are merits to this line or reasoning. One of the of conditional probabilities developed by Yao.
impediments in the lower bounds area is a shortage of Sipser [Si83] showed that there are functions com-
problems of intermediate difficulty which lend insight putable by depth d linear size circuits and which require
into the harder problems. The algebraic generalizations superpolynomial size for depth d 1, by a technique very
may be important steps towards the Boolean goal. On similar to that used by FSS. Yao claimed without proof
the other hand, I do not believe, ss Valiant argues, that an exponential bound for this depth hierarchy. Using
the algebraic caze is prerequisite. True, the algebraic his simplified method, H&tad presented a proof of this
case is more restrictive, and hence easier.) The same result.
may be said for uniform models of computation ver- Razborov [Ra87] introduced a method of obtaining
sus nonuniform models. Nonuniform models, such ss lower bounds on the size of limited depth circuits over
circuits, are more powerful than uniform models, such a larger basis. He proved that the majority function re-
m Turing machines, because the algorithm used may quires exponential size for circuits having PARITY gates
change arbitrarily depending upon the length of the in- as well as AND, OR, and NOT gates. He showed that
put. Accordingly it may be more difficult to prove lower any set definable with a small, shallow circuit over this
bounds on these more powerful models. Nevertheless, basis is well approximated by a polynomial of low de-
the harder problem may allow one to see more eas- gree over the two element field, and argues that the
ily the heart of the matter, unobscured by unnecessary majority function cannot be well approximated in this
features. way. Barrington [Ba86a] extended this to circuits with
MODq gates for any prime q instead of only PARITY (i. e.,
MOD2) gates. Smolensky [Sm87] further simplified and
3.6 Bounded Depth Circuits
improved this, showing that computing the MODP func-
The next few sections treat handicapped variants of tion requires exponential size when computed by shal-
the circuit model. By placing various restrictions on low circuits having AND, OR, and MODq gates, for p and
structure, such as limiting the depth or restricting the q powers of different primes.
basis, it has been possible to obtain several strong
lower bounds. The first result of this kind is due to
3.7 Monotone Circuit Size
Furst, Saxe, and Sipser [FSS84], and independently Aj-
tai [Aj83], who proved that certain simple functions, A monotone circuit is one having AND and OR gates but
such as the parity function, require superpolynomial cir- no negations. A function is monotone if x ~ y implies
cuit size if the depth is held fixed. $(z) < ~(Y) under the usual Boolean ordering. It is
These two proofi of this theorem use different, though essy to see that monotone circuits compute exactly the
related, methods. Furst, Saxe, and Sipser introduced class of monotone functions.
the notion of probabilistic restnction, a randomly se- The first strong lower bound concerning this model
lected assignment to some of the variables. They showed is due to Razborov [Ra85a], who proved that the
that for any circuit of the form AND of small ORS, the clique function hss superpolynomial monotone complex-
circuit induced by a random restriction may likely be ity. Shortly thereafter Andreev [An85], using similar

607
methods, proved an exponential lower bound, further weaker than showing it is outside of P, but we are still
tightened by Alon and Boppana [AB87]. unable to do this for any language in NP.
Razborovs theorems on monotone and bounded Karchmer and Wigderson [KW88] gave a very nice
depth circuits, as well as the aforementioned proof of Aj- characterization of languages definable with such cir-
tai, rely upon a technique which has come to be called cuits, which may be useful in obtaining lower bounds
the approximation method. One of the difficulties in on depth. Let A be such a language. In the communi-
attempting to analyze the computation of a circuit (r* cation game for A, there are two players, one having a
stricted or general) for the purposes of proving lower string in A and the other having a string of the same
bounds is that the subfunctions computed by the cir- length not in A. The players communicate with each
cuit may be complicated and hard to grasp. In the ap- other to find a position where their two strings differ.
proximation method we show that the subfunctions are The minimum number of bits that they require to do
in a certain sense near functions which are much sim- this over all strings of length n is the complexity of the
pler. Take, for example Razborovs lower bound for the game. Karchmer and Wigderson show that this exactly
clique function. Consider any monotone circuit com- equals the minimum circuit depth necessary for this lan-
puting this function. The plan is to adjust the result guage.
of each of the AND and OR gates in such a way that A monotone variant of this game has played an im-
(1) each adjustment alters the result only slightly, while portant role in the discovery of lower bounds on mono-
(2) the adjusted output is far from the clique function. tone circuit depth. If one demands in addition that the
This means that there must have been many adjust- found position is 1 for the string in A, then the com-
ments and hence many gates. plexity of the game is the minimum monotone circuit
If general circuits computing monotone functions depth. This characterization of monotone depth com-
could be converted into monotone circuits with only a plexity led them to a Cl(logz n/ log log n) lower bound
polynomial increase in size then strong monotone lower on the monotone depth for the st-connectivity func-
bounds would yield separations of complexity classes in tion. This was subsequently improved to fl(logz n) by
general complexity theory. Razborov [Ra85b] showed Boppana ~S90] who avoided the communication game
that this is not true when he used the approximation by arguing directly on the monotone circuit (indepen-
to show that the that the problem of testing for a per- dently obtained by H&tad [unpublished]). Raz and
fect matching requires superpolynornial size monotone Wigderson [RW90] used the communication game to ob-
circuits. As this problem is in P [Ed65], we know that tain a linear depth lower bound for monotone circuits
it has polynomial size non-monotone circuits. Using a computing the matching function. Their proof uses a
simikar method, Tardos [Ta88] showed that there is a lower bound on the probabilistic communication com-
monotone problem in P which requires exponential size plexity of the set disjointness problem due to Kalyana-
monotone circuits. sundamm and Schnitger [KS87] (simplified by Razborov
There is a class of functions where the monotone and [Ra90a]). Because of this it seems that it will be diffi-
non-monotone complexities axe polynomially related. cult to phrsse the Raz-Wigderson argument without ap-
These are the slice functions, introduced by Berkowitz pealing to the communication game. Razborov [Ra90b]
[Be82]. A function ~ is a slice finction if for some k and Goldmann and Hiistad [GH89] obtained additiomd
the wdue of ~(iz) is O when the number of 1s in z is lower bounds using the communication game.
fewer than k, and the value is 1 when the number of 1s Grigni and Sipser [GS90] point out that many of
in x is more than k. He showed that any general cir- the results in monotone complexity can be viewed in
cuit computing a slice function may be converted into a terms of monotone analogs, mP, mNP, etc., of the fa-
monotone function by adding only a polynomial num- miliar complexity classes. Thus we already know that
ber of gates. Hence if one were able to prove a strong mP#mNP [Ra85a], mP# {P n mono} [Ra85b], and
lower bound on the monotone circuit complexity of a mNCl#mNL ~W88]. We use mono to designate the
slice function then this would show that P#NP. class of all monotone languages. Grigni and Sipser
demonstrate that mNL#m-coNL using the technique
of [KW88]. This shows the inherently nonmonotone
3.8 Monotone Circuit Depth
nature of the Imrnerman-Szelep cseny simulation. In
Circuits over the standard basis restricted to fan-in two a later paper [GS91], they strengthen the KW result
with O(log n) depth form an important subclass of the by separating mNCl from mL. There are a number of
polynomial size circuits. Such circuits axe equivalent in open questions in monotone complexity, for example,
power to polynomial size formulas. Proving that a lan- whether Barringtons [Ba86a] beautiful simulation of
guage is not definable with a circuit of this kind would log-depth circuits by polynomial-size, width 5, branch-
show that it is not in the class NCl. This would be ing programs, is inherently nonmonotone (see [GS90]).

6Q8
3.9 Active Research Directions cuits computing the parity function on n variables, is the
theorem that there are no countable-size, depth d, cir-
In this section we wilI examine a few directions which
cuits computing a parity function on w many variables.
appear to be plausible means of separating complex-
(A Boolean function on w many Boolean variables is a
ity classes. These are the approximation meihod, which
parity jknction if the output changes whenever any sin-
has been used successfully in analyzing restricted circuit
gle input changes.) The proof of the infinitary theorem
models; the function composition method, which may be
did precede and motivate the proof in the finite case.
useful to prove functions require large circuit depth; and
The two prootk are very similar in structure. A see;
an infinite analog, which has led to lower bounds in the
end, related application of this analogy appears in the
bounded depth circuit model.
constant-depth hierarchy theorem [Si83].
There is a possibility that the approximation method,
This same approach suggests a infinite analog to the
which has been used so successfully to obtain lower
class NP: the analytic sets (also called E; sets). The
bounds for monotone and bounded depth circuits, may
chissical theorem due to Lebesgue [Le05] stating that
apply in the general case a well. In [Ra89], Razborov
the there is an snalytic set whose complement is not
considers this question and gives both a positive and
analytic may thus be taken as an infinite analog to the
negative answer. He shows that for one formalization of
statement that NP#coNP. Lebesgues proof is a diag-
the method, it is too wesk to give interesting bounds.
onalization, and it does not seem likely that it has a
On the other hand a generalization of the method is, in
finite counterpart. In [Si84] I give a different proof of
principle, strong enough, though there are much greater
Lebesgues theorem. This new proof does not rest upon
technical difficulties when applying it. Roughly speak-
the notion of universal set, essential to the diagonaliza-
ing, these two versions of the approximation method
tion in Lebeagues proof. It offers more information of
differ in the way the class of approximating functions
a combinatorial nature that may be useful in the finite
are chosen. In the weaker version, the class is selected
case.
in advance and applies to all circuits. In the stronger
version the class depends upon the circuit.
Karchmer, Raz, and Wigderson [KRW91] proposed a
Acknowledgments
direction for investigating the NC1 versus P question.
Let B. denote the set of all Boolean functions on n
I wish to thank Mauricio Ksrchmer and Alexander
variables. Given .f c B. and g c Bm we define the
Razborov who read an earlier draft of this paper and
composition ~ o g : {O, I}nm - {O, 1} by
offered several suggestions and corrections.

f o 9(X, . ..AJ = f(9(xl), . ...9(Z))

I am grateful to Juris Hartmanis for providing me
with a copy of Godels letter and to Sorin Istrail and
where ii c {O, I}m for i = 1, ...}n. It is clear that Arthur S. Wensinger for its translation. It appears here
d(~ o g) $ d(j) + d(g). KRW conjecture that these with the permission of the Institute for Advanced Study.
two quantities are close to each other and argue that, if Finally, it is wonderful to thank my wife Ins, for
so, NCl #P. Edmonds, Impagliazzo, Rudich and Sgall putting up with the writing of th~ paper as our daugh-
[EIRS91] (and slightly later, Hbtad and Wigderson ter Rachel wss being born, and for other assistance at
[HW91] in a somewhat different way) proved a weak critical moments.
form of this conjecture.
In [Si81], I suggest a way to use idess from descrip-
tive set theory (Moechovakw @4080] has an excellent Appendix: Historical Quotes
text on this subject) to gain insight into certain prob-
lems in circuit complexity theory. This stems ftom a Shannon, discussing circuit design methods [Sh49]:
proposed analogy between polynomiality and countabil-
The problem of synthesizing non-series-parallel cir-
ity. A justification for this analogy is that uncountabil-
cuits is exceedingly ditlicult. It is even more dMicult to
ity and superpolynomialit y often have a common origin:
show that a circuit found in this way is the most eco-
the power set operation. In this way certain problems in
nomical one to realize a given function. The difficulty
circuit complexity have infinitary analogs, often essier
springs from the huge number of essentially different
to solve. When this occurs, it may lead to an approach
networks available and more particularly from the lack
in the finitsry case.
of a simple mathematical idiom for representing these
A successful application of this strategy occurred in
circuits.
proving the lower bound of Furst, Saxe, and myself,
mentioned earlier [FSS84]. The infinite analog to the von Neumann, presenting an algorithm for solving
theorem that there are no polynomial-size, depth d cir- the assignment problem fVo531:

609
 It turns out that this number [of steps required by Edmonds, a section marked Digression in a paper
the algorithm] is a moderate power of n, i.e., consider- giving a polynomial time algorithm for the mwmum
ably smaller than the obvious estimate n! mentioned matching problem fEd65]:
earlier. An explanation is due on the use of the words effi-
cient algorithm. First, what I present is a conceptual
Yablonski, discussing alternatives to perebor in de-
description of an algorithm and not a particular formal-
signing circuits /Ya59aJ
ized algorithm or code.
At present there is an extensive field of problems For practicrd purposes computational details are vi-
in cybernetics where the existence of certain objects tal. However, my purpose is only to show as attractively
or facts may be established quite trivially and, within as I can that there is an efficient algorithm. According
the limits of the classical definition of algorithms, com- to the dictionary, efficient means (adequate in oper-
pletely effectively, yet a solution is, in practice, often ation or performance. This is roughly the meaning I
impossible because of its cumbersome nature. Such, for want in the sense that it is conceivable for maximum
example, are some of the problems involving informa- matching to have no efficient algorithm. Perhaps a bet-
tion coding and the analysis and synthesis of circuits. ter word is good.
It is here that the necessity of making the classical def- I am claiming, as a mathematical result, the existence
inition an algorithm more precise naturally arises. It is of a good algorithm for finding a maximum cardinality
to be expected that this will, to a greater extent than matching in a graph.
at present, take into account the peculiarities of cer- There is an obvious finite algorithm, but that algo-
tain classes of problems, and may, possibly, lead to such rithm increases in difllculty exponentially with the size
developments in the concept of algorithm that dii7er- of the graph. It is by no means obvious whether or not
ent types of algorithms will not be comparable. At the there exists an algorithm whose difficulty increases only
moment it is too early to make predictions of how the algebraically with the size of the graph.
notion of an algorithm may be modified, for we have, ss The mathematical significance of this paper rests
yet, very little idea of how the various classes of prob- largely on the assumption that the two preceding sen-
lems should be speciiied, In the present article we at- tences have mathematical meaning. I am not prepared
tempt to explore the algorithmic difficulties arising in to set up the machinery necessary to give them formal
the solution of cybernetic problems which, while admit- meaning, nor is the present context appropriate for do-
ting of a trivial solution, on the basis of the clsssical ing this, but I should like to explain the idea a little fur-
definition of an algorithm, are not practically solvable ther informally. It may be that since one is customarily
because of the massive nature of that solution. concerned with existence, convergence, finiteness, and
so forth, one is not inclined to take seriously the ques-
Cobham, studying why certain problems, such a mu]-
tion of the existence of a better-than-finite algorithm.
triplication, are computationally more ditlicult tlmn oth-
The relative cost, in time or whatever, of the various
ers, such as addition [C064]:
applications of a particular algorithm is a fairly clear
Thus the process of adding m and n can be carried out notion, at least ss a natural phenomenon. Presumably,
in a number of steps bounded by a linear polynomial in the notion can be formalized. Here algorithm is used
i(m) and l(n). Similarly, we can multiply m and n in a in the strict sense to mean the idealization of some phys-
number of steps bounded by a quadratic polynomial in ical machinery which gives a definite output, consisting
i(m) tmd l(n). So, too, the number of steps involved in of cost plus the desired result, for each member of a
the extraction of square roots, calculation of quotients, specified domain of inputs, the individual problems.
etc., can be bounded by polynomials in the lengths of The problem-domain of applicability for an algorithm
the numbers involved, and this seems to be a property often suggests for itself possible measures of size for the
of simple function in general. This suggests that we individual problems for maximum matching, for ex-
consider the CISSS, which I will call X, of all functions ample, the number of edges or the number of vertices
having this property. in the graph. Once a messure of problem-size is chosen,
For several reasons the class L seems a natural one to we can define FA (N) to be the lesst upper bound on
consider. For one thing, if we formalize the above def- the coat of applying algorithm A to problems of size N.
inition relative to various general classes of computing When the measure of problem-sise is reasonable and
machines we seem always to end up with the same well- when the sizes sssume values arbitrarily large, an
defined class of functions. Thus we can give a mathe- asymptotic estimate of FA (N) (let us call it the order
matical characterization of L having some confidence of dificzdty of algowthm A) is theoretically important.
that it characterizes correctly our informally defined It cannot be rigged by making the algorithm artificially
class. difficult for smaller sizes. It is one criterion showing

610
how good the algorithm ia not merely in comparison which the supervisor crm require his assistant to search
with other given algorithms for the same class of prob- out along with a tilmum partition and which the su-
lems, but also on the whole how good in comparison pervisor can then use with ease to verify with mathe-
with itself. There are, of course, other equaUy valuable matical certainty that the partition is indeed minimum.
criteria. And in practice this one is rough, one reason Having a good characterization does not mean necessar-
being that the size of a problem which would ever be ily that there is a good algorithm. The assistant may
considered is bounded. have to kill himself with work to find the information
It is plausible to assume that any algorithm is equiva- and the partition.
lent, both in the problems to which it applies and in the
Rabin, in a survey paper on automata theory fRa66]:
costs of its applications, to a normal algorithm which
decomposes into elementaJ steps of certain prescribed We shall consider an algorithm to be practical if, for
types, so that the costs of the steps of all normal algo- automata with n states, it requires at most Cnb (k is
rithms are comparable. That is, we may use something a fixed integer and c a fixed constant) computational
like Churchs thesis in logic. Then, it is possible to ask: steps. This stipulation is, admittedly, both vague and
Doea there or does there not exist an algorithm of given arbitrary. We do not, in fact cannot, define what is
order of difficulty for a given class of problems? meant by a computational step, thus have no precise
One can find many classes of problems, besides maxi- and general measure for the complexity of algorithms.
mum matching and its generalizations, which have algo- Furthermore, there is no compelling reason to classify
rithms of exponential order but seemingly none better, algorithms requiring cn~ steps as practical.
An example known to organic chemists is that of de- Several points may be raised in defense of the above
ciding whether two gjven graphs are isomorphic. For stipulation. In every given algorithm the notion of a
practical purposes the difference between algebraic and computational step is quite obvious. Hence there is not
exponential order is often more crucial than the dtier- that much vagueness about the messure of complexity of
ence between finite and non-finite. existing aJgorithma. Another significant pragmatic fact
It would be unfortunate for any rigid criterion to in- is that all existing algorithms either require up to about
hibit the practical development of algorithms which are n4 steps or else require 2n or worse steps. Thus drawing
either not known or known not to conform nicely to the the line of practicality between algorithms requiring vak
criterion. Many of the beet algorithmic ideaa known steps and algorithms for which no such bound exists
today would suffer by such theoretical pedantry. In seems to be reasonable.
fact, an outstanding open question is, essentially: how
Cook, considering the complexity of theorem proving
good is a particular algorithm for linear programming,
[C071]. The class P is denoted by C:
the simplex method? And, on the other hand, many im-
portant algorithmic idess in electrical switching theory Furthermore, the theorems suggest that tautologies is
are obviously not good in our sense. a good candidate for an interesting set not in Z*, and I
However, if only to motivate the search for good, feel that it ia worth spending considerable effort trying
practical algorithms, it is important to realize that it to prove this conjecture. Such a proof would be a major
is mathematically sensible even to question their exis- breakthrough in complexity theory.
tence. For one thing the task can then be described in Levin, from Zhkhtenbrots survey on perebor [Le73,
terms of concrete conjectures. Z184]:
Fortunately, in the case of maximum matching the r~
This is the situation with so called exhaustive search
suits are positive. But possibly this favorable position
problems, including: the minimization of Boolean func-
is very seldom the case. Perhaps the twoness of edges
tions, the search for proofs of finite length, the deter-
makes the algebraic order for matching rather specisJ
mination of the isomorphism of graphs, etc. All of
for matching in comparison with the order of difficulty
these problems are solved by trivial algorithms entailing
for more general combinatorial extremum problems.
the sequentizd scanning of all possibilities. The operat-
Edmonds, discussing the matroid partition probiem ing time of the algorithm is, however, exponential, and
fEd65b]: mathematicians nurture the conviction that it is impos-
sible to find simpler algorithms.
We seek a good characterization of the minimum
number of independent sets into which the columns of Gi5del, in a letter to von Neumann. !fkamdated by
a matrix of II can be partitioned. As the criterion of Arthur S. Wensinger with Sorin Lstrai.1s guidance. The
good for the characterization we apply the princi- orignal letter is in the Manuscript Division of the Li-
ple of the absolute supervisor. The good characteriza- brary of Congress, Washington, D.C. The translator had
tion will describe certain information about the matrix access to Juris Hartmanis article @It?9]. Discussions

611
 with liartley Rogers and Philip Scowcroft provided con- from a generalization of the proof for the unsolvabil-
tributions to the trsdation: ity of the Entscheidungsproblem; 2.) p(n) N Kn (or
N Kn2) means, of course, simply that the number of
Dear Mr. von Neumann, steps vis-h-via dem Mossen Pro biereng can be reduced
from IV to log IV (or (log lV)2)Such strong reductions10
To my great regret I have heard about your illness. do indeed occur, however, in the case of other finite
The news reached me most unexpectedly. Morgenstern problems, e.g., in the csse of calculating a quadratic
had, to be sure, told me in the summer about some infir- residuell by means of repeated application of the law
mity you had been suffering, but at the time he thought of reciprocity. It would be interesting to know, for ex-
that no major significance was to be attached to it. I ample, what the situation is in the case of determining
hear that you have been undergoing radical treatment whether a number is a prime number, and in the csse of
in the psst several months, and I am happy2 that this finite combinatorial problems, how strongly12 in genend

has achieved the desired results and that things are now the number of steps via->vis the blossen Probieren can
going better for you. I hope and wish that your condi- be reduced.

tion continues to improve and that the latest medical

advances, if possible, can lead to a complete recovery. I wonder if you have heard that Posts problem (ss
to whether there are degrees of unsolvabtlty among the
Since I have hesrd that you are feeling stronger now, problems (3y)y@, z), where p is recursive) hss been
I tske the liberty3 of writing to you about a mathemat- solved in the positive by a very young man named
ical problem about which your views would interest me Richard Friedberg. The solution is very elegant13. Un-
greatly. It is evident that one can easily construct a fortunately, Friedberg does not want to pursue graduate
Turing machine which, for each formula F of the pred- work in mathematics, but in medicine (evidently under
icate calculus4 and for every natural number n, will al- the influence of his father).
low one to decide if F has a proof of length n [ length
= number of symbols ]. Let W(F, n) be the number of What is your opinion by the way, of the attempts
steps that the machine requires for that and let p(n) = recently in vogue again to bsse Analysis14 on rsmified
max~!l!(F, n). The question is, how fast does p(n) grow type-theory 15 ? You are probably aware that in con-
for an optimal machine. One can show that ~(n) z Kn. nection with this, Paul Lorenzen hss progressed ss far
If there actually were a machine with p(n) N Kn (or ss Lebesgues Messure Theory. But I believe that in
even onlys with w Kn2), this would have consequences important sspects of Analysis non-predicative methods
of the greatest magnitude6. That ia to say, it would of proof cannot be eliminated.
clearly indicate that, despite the unsolvability of the
Entscheidungsproblem, the mental effort of the math- I would be very happy to hear from you persomdly;
ematician in the case of yes-or-no questions could be and plesse let me know if I can do anything at all for
completely (G6dels footnote apart from the postulation of you.
axioms ) replaced by machines. One would indeed have
to simply select an n so large7 that, if the machine With best greetings and wishes, also to your wife16,
yields no result, there would then also be no reason Your very devoted,
to think further about the problem. Now, it seems to Kurt Godel
me, however, to be totally within the realm of possi-
bility that p(n) grows Slowly.s For 1.) it seems that P.S. I congratulate you heartily on the ...17
p(n) z Kn is the only estimate that can be derived
%.: the simply testing or trying out; exhaustive search
2to hear or brute force are the currently established terms
Sfit.: I &odd like to dow mYSSlf 10Verringerungen
4 de~ eW=en titione&&i&, lit.: of the narrower function- 11l%stsymbol
Calculue 12severely, radically; how much is not quite adequate as a
5It is not complet~ clear if the word is w: o~Y~ or mm: translation
now; however, onIy seems to fit the context better. 131t i5 worth ~mtig that this work was done by lWedberg
fJt~weite~$ suggets great breadth of r- or hnp~t, fit.: in his undergraduate senior honors thesis.
carrying power far into many fields or areas where it would have 14G~del ~e8 the E@sh word Analysis here, not the Germ=m
its eHect. term Anrdytik.
7or: large enough 15versweigte Typentheotie
8 G~del *WS -s0 lq~ wmt, Whk.h does nCItmean %ows
16an ihre Rau Gemahlin is an unusually polite and old-
so slowly; but more something like accordingly grows slowly? -fashioned formula, quite in accord with the formal tone of this
where accordingly means considering the type of growth of letter to a man clearly highly esteemed by G6deL
w(n) I have bem discussing? 17... ? The balance of the postscript is missing.

612
 A

-4
.

,-3
.-L.

613
 I
.

f-

-A

S. S.-*2?
J- s d_ f
.? f
$ =&

83
.~
---?.-

614
 [BS90] R. B. BOPPANA AND M. SIPSER, The com-
plexity of finite functions, Handbook of The-
[Aj83] M. AJTAI, Z~-formulae on finite structures, oretical Computer Science, Elsevier, 758-804,
Annals of Pun and Applied Logic 24, 1-48, 1990.
1983.
[BU86] S. R. BUSS, Bounded Arithmetic, Bibliopolis,
[AB871 N. ALON AND R. B. BOPPANA, The mono- 1986.
tone circuit complexity of Boolean functions,
Combinatorics 7:1, 1-22, 1987. [C064] A. COBHAM, The intrinsic computationzd dif-
ficulty of functions, Proc. of the 1964 Interna-
[An85] A. E. ANDREEV, On a method for obtaining tional Congwss for Logic, Methodology, and
lower bounds for the complexity of individual the Philosophy of Science, edited by Y. Bar-
monotone functions, Doklady Akademii iVauk Hillel, North-Holland, 24-30, 1964.
SSSR 282:5, 1033-1037 (in Russian). English
translation in Soviet Mathematics Doklady [CSV84] A. K. CHANDRA) L. J. STOCKMEYER, AND
31:3,530-534, 1985. U. VISIIKIN, Constant depth reducibility,
SIAM Journal on Computing 13:2, 423-439,
[As55] G. ASSER, Das Reprii.sentatenproblem im 1984.
Pri+dikatenkalkul der ersten Stufe mit iden-
titat, Zeitschrift fiir Matenaitische Logic und [C071] S. A. COOK, The complexity of theorem
Grundlagen der Mathematik, 1, 252-263, proving procedures, Proceedings of the 3rd
1955. Annual ACM Symposium on Theory of Com-
puting, 151-158, 1971.
[BGS75] T. P. BAKER, J, GILL, AND R. SOLOVAY,
Relativizations of the P=?NP question, SIAM [C075] S. A. CooK, Fessably constructive proofs
Journal on Computing 4:4,431-442, 1975. and the propositional calculus, Proceedings of
the 7th Annual ACM Symposium on Theory
[Ba86a] D. A. BARRINGTON, Bounded-width poly- of Computing, 8397, 1975.
nomial-size branching programs recognize ex-
actly those languages in IfC1, Journal of [CU88] S. A. COOK AND A. URQUHART, Functional
Computer and System Sciences .98, 150-164, interpretations of fessibly constructive arith-
1986. metic, Technical Report 210/88, Computer
Science Department, University of Toronto,
[Ba86a] D. A. BARRINGTON, A note on a theorem of 1988.
Rszborov, manuscript, 1986.
[DL79J R. E. DEMILLO AND R. J. LIPTON, The
[BH91] S. BEN-DAVID AND S. HALEVI, On the Inde- consistency of P=NP and related problems
pendence of P versus NP, Tech. Report #699, within fragments of number theory, Proceed-
Technion, 1991. ings of the llth ACM Symposium on Theory
of Computing, 153159, 1979.
[Be82] S. J. BERKOWITZ, On some relationships
between monotone and non-monotone cir- [DU88] P. E. DUNNE, The Complexity of Boolean
cuit complexity, Technical Report, Computer Networks, Academic Press, 1988.
Science Department, University of Toronto,
1982. [Ed65] J. EDMONDS, Paths, trees, and flowers,
Canadian Journal of Mathematics, 17, 449-
[B167] M. BLUM, A machine-independent theory of 467, 1965.
the complexity of recursive functions, Journal
of the ACM, 14, no. 2, 322336, 1967. [Ed65b] J. EDMONDS, Minimum partition of a ma-
troid into independent sets, Journal of Re-
[B184] N. BLUM, A Boolean function requiring 3n search of the National Bureau of Standards
network size, Theoretical Computer Science (B), 69,67-72,1965,
28, 337-345, 1984.
[EIRS91] J. EDMONDS, R. IMPAGLIAZZO, S. RUDICH,
[B086] R. B. BOPPANA, Threshold functions and J. SGALL, Communication complexity to-
bounded depth monotone circuits, Journal of wards lower bounds on circuit depth, Proceed-
Computer and System Sciences 32:2,222-229, ings of the 3$%d Annual Symposium on Foun-
1986. dations of Computer Science, 249-257, 1991.

615
[Fa74] R. FAGIN, Generalized first-order spectra and [H586] J. H~STAD, Computational Limitations for
polynomial-time recognizable sets, Complex- Small Depth Circu-its, MIT Press, 1986. -
ity of Computation, SIAM-AMS Proceedings
[HW91] J. HISTAD, A. WIGDEItSON, Composition of
7,43-73, 1974.
the Universal Relation, manuscript, 1991.
[FS88] L. FORTNOW AND M. SIPSER, Are there
[HU79] J. E. HOPCROFT AND J. D. ULLMAN, Intro-
interactive protocols for CGNP languages?
duction to Automata Theory, Languages, and
Information Processing Letters 28, 249-251,
Computation, Addison-Wesley, 1979.
1988.
~m87] N. IMMERMAN, Languages that capture com-
[FSS84] M. FURST, J. SAXE, AND M. SIPSER, Par-
plexity classes, SIAM Journal on Computing
ity, circuits and the polynomial time hierar-
16:4,760-778, 1987.
chy, Mathematical Systems Theoy 17, 1327,
1984. [Im88] N, IMMERMAN, Nondeterministic space is
closed under complement, SIAM Journal on
[Ga771 Z. GALIL, On the complexity of regular reso-
Computing, 935-938, 1988.
lution and the Davis-Putnam procedure, The-
oretical Computer Science 4, 23-46, 1977. [JS74] N. D. JONES AND A. L. SELMAN, Turing ma-
chines and the spectra of first-order formulas,
[GJ79] M. GAREY AND D. S. JOHNSON, Computers Journal of Symbolic Logic 39, 139-150, 1974.
and Intractability: a guide to the theoy of
NP-covnpleteness, Freeman, 1979. [KS87] B. KALYANASUNDARAM AND G. SCHNIT-
GER, The probabilistic communication com-
[GH89] M. Goldmann, J. Hiktad, A lower bound plexity of set intersection, Proceedings of the
for monotone clique using a communication Second Annual Conference on Structure in
game, manuscript, 1989. Complexity Theoy, 4149, 1987.

[GS90] M. GRIGNI AND M. SIPSER, Monotone [KRW91] M. KARCHMER, R. RAZ, A. WIGDERSON,

complexity, Proceedings of LMS workshop On proving super-logarithmic depth lower
on Boolean Function Complexity, Durham, bounds via the direct sum in communication
edited by M. S. Paterson, Cambridge Univer- complexity, Proceedings of the Sixth Annual
sity Press, 1990. Conference on Structure in Complexity The-
ory, 1991.
[GS91] M. GRIGNI AND M. SIPSER, Monotone sep-
aration of NC1 from logspace, Proceedings of [KW88] M. KARCHMER AND A. WIGDERSON, Mono-
the 6th Annual Symposium on Structure in tone circuits for connectivity require super-
Complexity Theory, 294-298, 1991. logarithmic depth, Proceedings of the 20th
Annual ACM Symposium on Theoy of Com-
[HS72] L. H. HARPER AND J. E. SAVAGE, The com-
puting, 539-550, 1988.
plexity of the marriage problem, Advances in
Mathematics 9, 299-312, 1972. [Ka72] R. M. KARP, Reducibility among combinat~
rial problems, Complexity of Computer Com-
[Hs89] J. HARTMANIS, Godel, von Neumann and putations, Plenum Press, 85-103, 1972.
the P=?NP Problem, Bulletin of the Euro-
pean Association for Theoretical Computer [K077] D. KOZEN, Lower bounds for natural proof
Science, 101-107, June 1989. systems, Proceedings of the 18th Annual Sym-
posium on Foundations of Computer Science,
[HH76] J. HARTMANIS AND J. HOPCROFT, Ind& 254-266, 1977.
pendence results in computer science, ACM
SIGACT News 8, no. 4, 13-24, 1976. [LJK87] K. J. LANGE, B. JENNER, B, KIRSIG, The
logarithmic hierarchy collapses: AX; = AII~,
[HS65] J. HARTMANIS AND R. E. STEARNS, On Proceedings of the l./th International Confer-
the computational complexity of ~gorithma, ence on Automata, Languages, and Program-
Transactions of the AMS 117,285-306, 1965. ming, 1987.

[Ha$5] A. HAKEN, The intractability of resolution, [Leo5] H. LEBESGUE, Sur les fonctions repr&en-
Theoretical Computer Science 39, 297-308, tables analytiquement, Journal de Math. 6
1985. serie 1, 139-216, 1905.

616
[Le73] L. LEVIN, Universal sequential search prob- [Ra85b] A, A. RAZBOROV, A lower bound on the
lems, Probl. Pemd. Inform. IX $?,1973; trans- monotone network complexity of the logical
lationin Problems of Information Trans 9,3, permanent, Matematicheskie Zametki 37:6,
265-266; corrected translation in llakhten- 887-900 (in Russian). English translation in
brot [TY84]. Mathematical Notes of the Academy of Sci-
ences of the USSR 37:6, 485-493, 1985.
[Li78] R. J. LIPTON, Model theoretic aspects of
complexity theory, Proceeding of the19th An- [Ra87j A. A. RAZBOROV, Lower bounds on the
nual Symposium on Foundations of Computer size of bounded depth networks over a com-
Science, 193-200, 1978. plete basis with logical addition, Matematich-
eskie Zametki 41:4, 598607 (in Russian). En-
[LFKN901 C. LUND, L. FORTNOW, H. KARLOFF, AND glish translation in Mathematical Notes of the
N. NISAN, Algebraic methods for interactive Academy of Sciences of the USSR 41:4,333-
proof systems, Proceedings of 22th Annual 338, 1987.
ACM Symposium on Theoy of Computing,
[Ra89] A. A. RAZBOROV, On the method of ap-
210, 1990; to appear in the Journal of the
proximations, Proceedings of the 21st Annual
ACM.
ACM Symposium on Theory of Computing,
~S72] A. R. MEYER, AND L. J. STOCKMEYER, 168-176, 1989.
The equivalence problem for regular expres-
[Ra90a] A. A. RAZBOROV, On the distributional com-
sions with squaring requires exponential time,
plexity of disjointness, Proceeding of the In-
Proceedings of the 13th Annual Symposium
ternational Conference on Automata, Lan-
on Switching and Automata Theoy, 125-129,
guages, and Computation, 1990; to appear in
1972.
Theoretical Compuier Science.

[M080] Y. N. MOSCHOVAKIS, Descriptive Set The-

[Ra90b] A. A. RAZBOROV, Applications of matrix
ory, North-Holland, 1980. methods for the theory of lower bounds in
computational complexity, Combinatorics 10,
[Mu56] D. E. MULLER, Complexity in electronic
81-93, 1990.
switching circuits, IRE !lhansactions on Elec-
tronic Computers 5, 15-19, 1956. [Sa72] J. E. SAVAGE, Computational work and time
on finite machines, Journal of the ACM 19:4,
[Pa76] M. S. PATERSON, An introduction to
660-674, 1972.
Boolean function complexity, Asi&isque 98-
39, 183-201, 1976. [ss80] V. Y. SAZANOV, A logical approach to
the problem P=NP? Mathematical Foun-
[Ra60] M. O. RABIN, Degree of dificutly of com- dations of Computer Science, Lecture notes
puting a function and a partial ordering of in Computer Science #88, 562575, 1980.
recursive sets, Tech. Rep. No. 2, Hebrew Uni-
versity, 1960. [SC52] H. SCHOLZ, Problem #1, Journal of Symbolic
Logic, 17, 160, 1952.
[Ra66] M. O. RABIN, Mathematical theory of au-
[Sh90] A. SHAMIR, IP=PSPACE, Proceedings of
tomata, Proceedings of 19th ACM Symposium
22th Annual ACM Symposium on Theory of
in Applied Mathematics, 153175, 1966.
Computing, 1115, 1990; to appear in the
[RW90] R. RAZ AND A. WIGDERSON, Monotone cir- Journal of the ACM.
cuits for matching require linear depth, Pro-
[Sh49] C. E. SHANNON, The synthesis of two-
ceedings of the 22nd Annual A CM Symposium
terminal switching circuits, Bell Systems
on Theory of Computing, 287292, 1990.
Technical Journal 28:1, 59-98, 1949.

[Ra$5a] A. A. RAZBOROV, A lower bound on the M. SIPSER, On polynomial vs. exponential

[Si81]
monotone network complexity of the logiczd growth, unpublished, 1981.
permanent, Matematicheskie Zametki 37:6,
887900 (in Russian). English translation in [Si83] M. SIPSER, Borel sets and circuit complexity,
Mathematical Notes of the Academy of Sci- Proceedings of 15th Annual ACM Symposium
ences of the USSR 37:6, 485493, 1985. on Theory of Computing, 6169, 1983.

617
 [Si84] M. SIPSER, A topological view of some prob-
lems in complexity theory, Colloquia Mathe-
matical Societatis J4nos Bolyai 44, 387391,
1984.

[Sm87] R. SMOLENSKY, Algebraic methods in the

theory of lower bounds for Boolean circuit
complexity, Proceedings of 19th Annual A CM
Symposium on Theoy of Computing, 77-82,
1987.

[s288] R. SZELEPCStiNYI, The method of forced

enumeration for nondeterministic automata,
Acts Informatica 26, 279-284, 1988.

[Ta88] E. TARDOS, The gap between monotone and

non-monotone circuit complexity is exponen-
tial, Combinatorics 8:1, 141-142, 1988.

[Tk84] B. A. TRAKHTENBROT, A survey of Russian

approaches to Perebor (brute-force search) al-
gorithms, Annals of the History of Computing
6, 110.4,384-400, 1984.

[Ts62] G. S. TSEITIN, On the complexity of deriva-

tion in propositional calculus, Studies in
Constructive Mathematics and Mathemati-
cal Logic, Part 2, Consultants Bureau, New
Youk-London, 115-125, 1962.

~a90J L. G. VALIANT, Why is Boolean complex-

ity theory dii%cult? Proceedings of LMS
workshop on Boolean Function Complexity,
Durham, edited by M. S. Paterson, Cam-
bridge University Press, 1990.

[V053] J. VON NEUMANN, A certain zero-sum two-

person game equivalent to the optimal assign-
ment problem, Contributions to the Theory of
Games 2,5-12, 1953.

~e87] I. WEGENER, The Complexity of Boolean

Functions, Wiley-Teubner, 1987.

~a59a] S. YABLONSKI, The algorithmic difficulties of

synthesizing minimal switching circuits, Prob-
lemy Kibernetiki 2, Moscow, Fizmatgiz, 75
121, 1959; translation in Problems of Cyber-
netics l?, Pergamon Press, 401457, 1961.

~a59b] S. YABLONSKI, On the impossibility of elim-

inating brute force search in solving some
problems of circuit theory, Doklady, AN SSSR
124,44-47, 1959.
~a85] A. C. YAO, Separating the polynomial-time
hierarchy by oracles, Proceedings of 26th An-
nual IEEE Symposium on Foundations of
Computer Science, 1-10, 1985.

618