This is largely an expository paper on the general theory of synchronous parallel computation. The models of parallel computers discussed include uniform circuit families, alternating Turing machines, conglomerates, vector machines, and par­ allel random access machines. A classification of these models indicates the need for still more; so “aggregates” and “hardware modification machines” are intro­ duced. The resources sequential time, space, parallel time, circuit size and depth, hardware size etc., are discussed and interrelated. Work in progress at Toronto is mentioned and basic open questions are listed.

An historical overview of computational complexity is presented. Emphasis is on the fundamental issues of defining the intrinsic computational complexity of a problem and proving upper and lower bounds on the complexity of problems. Probabilistic and parallel computation are discussed. This is the second Turing Award lecture on Computational Complexity. The first was given by Michael Rabin in 1976. 1 In reading Rabin’s excellent article [Rab77] now, one of the things that strikes me is how much activity there has been in the field since. In this brief overview I want to mention what to me are the most important and interesting results since the subject began in about 1960. In such a large field the choice of topics is inevitably somewhat personal; however, I hope to include papers which, by any standards, are fundamental.

We introduce the tree evaluation problem , show that it is in LogDCFL (and hence in P ), and study its branching program complexity in the hope of eventually proving a superlogarithmic space lower bound. The input to the problem is a rooted, balanced d -ary tree of height h , whose internal nodes are labeled with d -ary functions on [ k ] = {1,..., k }, and whose leaves are labeled with elements of [ k ]. Each node obtains a value in [ k ] equal to its d -ary function applied to the values of its d children. The output is the value of the root. We show that the standard black pebbling algorithm applied to the binary tree of height h yields a deterministic k -way branching program with O ( k h ) states solving this problem, and we prove that this upper bound is tight for h = 2 and h = 3. We introduce a simple semantic restriction called thrifty on k -way branching programs solving tree evaluation problems and show that the same state bound of Θ ( k h ) is tight for all h ≥ 2 for deterministic thrifty programs. We introduce fractional pebbling for trees and show that this yields nondeterministic thrifty programs with Θ ( k h/2+1 ) states solving the Boolean problem “determine whether the root has value 1”, and prove that this bound is tight for h = 2,3,4. We also prove that this same bound is tight for unrestricted nondeterministic k -way branching programs solving the Boolean problem for h = 2,3.

It is shown that any recognition problem solved by a polynomial time-bounded nondeterministic Turing machine can be “reduced” to the problem of determining whether a given propositional formula is a tautology. Here “reduced” means, roughly speaking, that the first problem can be solved deterministically in polynomial time provided an oracle is available for solving the second. From this notion of reducible, polynomial degrees of difficulty are defined, and it is shown that the problem of determining tautologyhood has the same polynomial degree as the problem of determining whether the first of two given graphs is isomorphic to a subgraph of the second. Other examples are discussed. A method of measuring the complexity of proof procedures for the predicate calculus is introduced and discussed.

Aiming to provide weak as possible axiomatic assumptions in which one can develop basic linear algebra, we give a uniform and integral version of the short propositional proofs for the determinant identities demonstrated over $GF(2)$ in Hrubes-Tzameret [SICOMP'15]. Specifically, we show that the multiplicativity of the determinant function and the Cayley-Hamilton theorem over the integers are provable in the bounded arithmetic theory $\mathbf{VNC}^2$; the latter is a first-order theory corresponding to the complexity class $\mathbf{NC}^2$ consisting of problems solvable by uniform families of polynomial-size circuits and $O(\log ^2 n)$-depth. This also establishes the existence of uniform polynomial-size $\mathbf{NC}^2$-Frege proofs of the basic determinant identities over the integers (previous propositional proofs hold only over the two element field).

We give a uniform and integral version of the short propositional proofs for the determinant identities demonstrated over GF(2) in Hrubeš-Tzameret [9]. Specifically, we show that the multiplicativity of the determinant function over the integers is provable in the bounded arithmetic theory VNC2, which is a first-order theory corresponding to the complexity class NC2. This also establishes the existence of uniform polynomial-size and O(log2n)-depth Circuit-Frege (equivalently, Extended Frege) proofs over the integers, of the basic determinant identities (previous proofs hold only over GF(2)). In doing so, we give uniform NC2-algorithms for homogenizing algebraic circuits, balancing algebraic circuits (given as input an upper bound on the syntactic-degree of the circuit), and converting circuits with divisions into circuits with a single division gate—all (Σ1B-) definable in VNC2. This also implies an NC2-algorithm for evaluating algebraic circuits of any depth.

Relations vs. functions If F is a class of functions, then RF , the F -relations, is the class of relations whose characteristic function is in F . Conversely, if R is a class of relations, then it is possible to associate a class F of functions with R by saying f ∈ F iff first the relation y = f (x) is in R, and second f is bounded by a function from some appropriate class. For example, in case R is the class of constructive arithmetic relations the appropriate class of bounding functions turns out to be the class of polynomials. If we start with a class F of functions which includes the function x = y and is closed under substitution and limited minimalization, then the class of functions associated with RF is again precisely F , provided the class of bounding functions is chosen to be cofinal (see below) with F . On the other hand, suppose we start with a class R of relations which includes the identity relation and is closed under explicit transformation and the Boolean operations. If we pass to the associated class F of functions using any bounding class which includes the constant function I, then RF is precisely R.

Existing definitions of the relativizations of NC 1, L and NL do not preserve the inclusions NC 1 ⊆ L, NL ⊆ AC 1. We start by giving the first definitions that preserve them. Here for L and NL we define their relativizations using Wilson’s stack oracle model, but limit the height of the stack to a constant (instead of log(n)). We show that the collapse of any two classes in AC 0(m), TC 0, NC 1, L, NL implies the collapse of their relativizations. Next we develop theories that characterize the relativizations of subclasses of P by modifying theories previously defined by the second two authors. A function is provably total in a theory iff it is in the corresponding relativized class. Finally we exhibit an oracle α that makes AC k (α) a proper hierarchy. This strengthens and clarifies the separations of the relativized theories in [Takeuti, 1995]. The idea is that a circuit whose nested depth of oracle gates is bounded by k cannot compute correctly the (k + 1) compositions of every oracle function.

We prove exponential lower bounds on the size of semantic read-once 3-ary nondeterministic branching programs. Prior to our result the best that was known was for D-ary branching programs with |D| >= 2^{13}.

Monotone span programs are a linear-algebraic model of computation which were introduced by Karchmer and Wigderson in 1993 [1]. They are known to be equivalent to linear secret sharing schemes, and have various applications in complexity theory and cryptography. Lower bounds for monotone span programs have been difficult to obtain because they use non-monotone operations to compute monotone functions; in fact, the best known lower bounds are quasipolynomial for a function in (nonmonotone) P [2]. A fundamental open problem is to prove exponential lower bounds on monotone span program size for any explicit function.We resolve this open problem by giving exponential lower bounds on monotone span program size for a function in monotone P. This also implies the first exponential lower bounds for linear secret sharing schemes. Our result is obtained by proving exponential lower bounds using Razborov's rank method [3], a measure that is strong enough to prove lower bounds for many monotone models. As corollaries we obtain new proofs of exponential lower bounds for monotone formula size, monotone switching network size, and the first lower bounds for monotone comparator circuit size for a function in monotone P. We also obtain new polynomial degree lower bounds for Nullstellensatz refutations using an interpolation theorem of Pudlak and Sgall [4]. Finally, we obtain quasipolynomial lower bounds on the rank measure for the st-connectivity function, implying tight bounds for st-connectivity in all of the computational models mentioned above.

Scientists from various fields met at the Tata Institute of Fundamental Research on 2 January 2013 to discuss Alan Turing's legacy. A panel consisting of the following made initial statements around which the discussion was conducted.

In 1990, Subramanian [1990] defined the complexity class CC as the set of problems log-space reducible to the comparator circuit value problem (CCV). He and Mayr showed that NL ⊆ CC ⊆ P, and proved that in addition to CCV several other problems are complete for CC, including the stable marriage problem, and finding the lexicographically first maximal matching in a bipartite graph. Although the class has not received much attention since then, we are interested in CC because we conjecture that it is incomparable with the parallel class NC which also satisfies NL ⊆ NC ⊆ P, and note that this conjecture implies that none of the CC-complete problems has an efficient polylog time parallel algorithm. We provide evidence for our conjecture by giving oracle settings in which relativized CC and relativized NC are incomparable. We give several alternative definitions of CC, including (among others) the class of problems computed by uniform polynomial-size families of comparator circuits supplied with copies of the input and its negation, the class of problems AC0-reducible to Ccv, and the class of problems computed by uniform AC0 circuits with AXccv gates. We also give a machine model for CC, which corresponds to its characterization as log-space uniform polynomial-size families of comparator circuits. These various characterizations show that CC is a robust class. Our techniques also show that the corresponding function class FCC is closed under composition. The main technical tool we employ is universal comparator circuits. Other results include a simpler proof of NL ⊆ CC, a more careful analysis showing the lexicographically first maximal matching problem and its variants are CC-complete under AC0 many-one reductions, and an explanation of the relation between the Gale--Shapley algorithm and Subramanian’s algorithm for stable marriage. This article continues the previous work of Cook et al. [2011], which focused on Cook-Nguyen style uniform proof complexity, answering several open questions raised in that article.

As of today, we do not know if even one among the above sequence of containments is strict. In fact, it is open if AC(6)=PH ! In this lecture by Steve we learn about an attempt aimed at separating P from NL using a specific problem called the Tree Evaluation Problem. Most of the contents of the talk are derived from the work “Pebbles and Branching Programs for tree evaluation” [1] by Steve Cook, Pierre McKenzie, Dustin Wehr, Mark Braverman and Rahul Santhanam. The problem chosen in this program is inspired by Michael Taitslin’s FOCS 2005 submission, an attempt to separate NL and P.

This paper is a contribution to our understanding of the relationship between uniform and nonuniform proof complexity. The latter studies the lengths of proofs in various propositional proof systems such as Frege and bounded-depth Frege systems, and the former studies the strength of the corresponding logical theories such as VNC1 and V0 in [Cook/Nguyen, 2010]. A superpolynomial lower bound on the length of proofs in a propositional proof system for a family of tautologies expressing a result like the pigeonhole principle implies that the result is not provable in the theory associated with the propositional proof system. We define a new class of bounded arithmetic theories n^epsilon-ioV^\infinity for epsilon < 1 and show that they correspond to complexity classes AltTime(O(1),O(n^epsilon)), uniform classes of subexponential-size bounded-depth circuits DepthSize(O(1),2^O(n^epsilon)). To accomplish this we introduce the novel idea of using types to control the amount of composition in our bounded arithmetic theories. This allows our theories to capture complexity classes that have weaker closure properties and are not closed under composition. We show that the proofs of Sigma^B_0-theorems in our theories translate to subexponential-size bounded-depth Frege proofs. We use these theories to formalize the complexity theory result that problems in uniform NC1 circuits can be computed by uniform subexponential bounded-depth circuits in [Allender/Koucky, 2010]. We prove that our theories contain a variation of the theory VNC1 for the complexity class NC1. We formalize Buss's proof in [Buss, 1993] that the (unbalanced) Boolean Formula Evaluation problem is in NC1 and use it to prove the soundness of Frege systems. As a corollary, we obtain an alternative proof of [Filmus et al, ICALP, 2011] that polynomial-size Frege proofs can be simulated by subexponential-size bounded-depth Frege proofs. Our results can be extended to theories corresponding to other nice complexity classes inside NTimeSpace(n^O(1), n^o(1)) such as NL. This is achieved by essentially formalizing the containment NTimeSpace(n^O(1), n^o(1)) \subseteq AltTime(O(1), O(n^epsilon)) for all epsilon > 0.

An approximate computation of a Boolean function by a circuit or switching network is a computation in which the function is computed correctly on the majority of the inputs (rather than on all inputs). Besides being interesting in their own right, lower bounds for approximate computation have proved useful in many sub areas of complexity theory, such as cryptography and derandomization. Lower bounds for approximate computation are also known as correlation bounds or average case hardness. In this paper, we obtain the first average case monotone depth lower bounds for a function in monotone P. We tolerate errors that are asymptotically the best possible for monotone circuits. Specifically, we prove average case exponential lower bounds on the size of monotone switching networks for the GEN function. As a corollary, we separate the monotone NC hierarchy in the case of errors -- a result which was previously only known for exact computations. Our proof extends and simplifies the Fourier analytic technique due to Potechin, and further developed by Chan and Potechin. As a corollary of our main lower bound, we prove that the communication complexity approach for monotone depth lower bounds does not naturally generalize to the average case setting.

Using Jerábek's framework for probabilistic reasoning, we formalize the correctness of two fundamental RNC^2 algorithms for bipartite perfect matching within the theory VPV for polytime reasoning. The first algorithm is for testing if a bipartite graph has a perfect matching, and is based on the Schwartz-Zippel Lemma for polynomial identity testing applied to the Edmonds polynomial of the graph. The second algorithm, due to Mulmuley, Vazirani and Vazirani, is for finding a perfect matching, where the key ingredient of this algorithm is the Isolating Lemma.

Kushilevitz [1989] initiated the study of information-theoretic privacy within the context of communication complexity. Unfortunately, it has been shown that most interesting functions are not privately computable [Kushilevitz 1989, Brandt and Sandholm 2008]. The unattainability of perfect privacy for many functions motivated the study of approximate privacy . Feigenbaum et al. [2010a, 2010b] define notions of worst-case as well as average-case approximate privacy and present several interesting upper bounds as well as some open problems for further study. In this article, we obtain asymptotically tight bounds on the trade-offs between both the worst-case and average-case approximate privacy of protocols and their communication cost for Vickrey auctions. Further, we relate the notion of average-case approximate privacy to other measures based on information cost of protocols. This enables us to prove exponential lower bounds on the subjective approximate privacy of protocols for computing the Intersection function, independent of its communication cost. This proves a conjecture of Feigenbaum et al. [2010a].