We show an Ω(√n/T) lower bound for the space required by any unidirectional constant-error randomized T-pass streaming algorithm that recognizes whether an expression over two types of parenthesis is well parenthesized. This proves a conjecture due to Magniez, Mathieu, and Nayak (2009) and rigorously establishes that bidirectional streams are exponentially more efficient in space usage as compared with unidirectional ones. We obtain the lower bound by analyzing the information that is necessarily revealed by the players about their respective inputs in a two-party communication protocol for a variant of the index function, namely augmented index. We show that in any communication protocol that computes this function correctly with constant error on the uniform distribution (a “hard” distribution), either Alice reveals Ω(n) information about her n-bit input, or Bob reveals Ω(1) information about his (logn)-bit input, even when the inputs are drawn from an “easy” distribution, the uniform distribution over inputs that evaluate to 0. The information cost tradeoff is obtained by a novel application of the conceptually simple and familiar ideas, such as average encoding and the cut-and-paste property, of randomized protocols. Motivated by recent examples of exponential savings in space by streaming quantum algorithms, we also study quantum protocols for augmented index. Defining an appropriate notion of information cost for quantum protocols involves a delicate balancing act between its applicability and the ease with which we can analyze it. We define a notion of quantum information cost, which reflects some of the nonintuitive properties of quantum information. We show that in quantum protocols that compute the augmented index function correctly with constant error on the uniform distribution, either Alice reveals Ω(n/t) information about her n-bit input, or Bob reveals Ω(1/t) information about his (log n)-bit input, where t is the number of messages in the protocol, even when the inputs are drawn from the abovementioned easy distribution. While this tradeoff demonstrates the strength of our proof techniques, it does not lead to a space lower bound for checking parentheses. We leave such an implication for quantum streaming algorithms as an intriguing open question.
We show an Ω(√n/T) lower bound for the space required by any unidirectional constant-error randomized T-pass streaming algorithm that recognizes whether an expression over two types of parenthesis is well parenthesized. This proves a conjecture due to Magniez, Mathieu, and Nayak (2009) and rigorously establishes that bidirectional streams are exponentially more efficient in space usage as compared with unidirectional ones. We obtain the lower bound by analyzing the information that is necessarily revealed by the players about their respective inputs in a two-party communication protocol for a variant of the index function, namely augmented index. We show that in any communication protocol that computes this function correctly with constant error on the uniform distribution (a “hard” distribution), either Alice reveals Ω(n) information about her n-bit input, or Bob reveals Ω(1) information about his (logn)-bit input, even when the inputs are drawn from an “easy” distribution, the uniform distribution over inputs that evaluate to 0. The information cost tradeoff is obtained by a novel application of the conceptually simple and familiar ideas, such as average encoding and the cut-and-paste property, of randomized protocols. Motivated by recent examples of exponential savings in space by streaming quantum algorithms, we also study quantum protocols for augmented index. Defining an appropriate notion of information cost for quantum protocols involves a delicate balancing act between its applicability and the ease with which we can analyze it. We define a notion of quantum information cost, which reflects some of the nonintuitive properties of quantum information. We show that in quantum protocols that compute the augmented index function correctly with constant error on the uniform distribution, either Alice reveals Ω(n/t) information about her n-bit input, or Bob reveals Ω(1/t) information about his (log n)-bit input, where t is the number of messages in the protocol, even when the inputs are drawn from the abovementioned easy distribution. While this tradeoff demonstrates the strength of our proof techniques, it does not lead to a space lower bound for checking parentheses. We leave such an implication for quantum streaming algorithms as an intriguing open question.
We show how two recently developed quantum information theoretic tools can be applied to obtain lower bounds on quantum information complexity. We also develop new tools with potential for broader applicability, and use them to establish a lower bound on the quantum information complexity for the Augmented Index function on an easy distribution. This approach allows us to handle superpositions rather than distributions over inputs, the main technical challenge faced previously. By providing a quantum generalization of the argument of Jain and Nayak [IEEE TIT'14], we leverage this to obtain a lower bound on the space complexity of multi-pass, unidirectional quantum streaming algorithms for the DYCK(2) language.
We show two unconditional quantum space advantages in the random-order streaming model. First, we show the Yamakawa-Zhandry Code Intersection problem admits exponential quantum advantage in the streaming model when its inputs are streamed in random order. This means quantum computers exhibit exponential space advantage even when simply receiving (x,f(x)) pairs for a uniformly random function f in a uniformly random order. Notably, our classical space lower bound does not follow from a communication lower bound. Instead, we utilize density-restoring partitions from communication complexity directly with potential-function analysis in the style used to prove memory-sample tradeoffs for learning. The resulting argument is distinct from both lines of work.
Second, we use our framework to show quantum space advantage for the Optimal Polynomial Intersection (OPI) problem in certain regimes via a streaming version of the Decoded Quantum Interferometry algorithm. In particular, we show that for degree d and n evaluation points, attaining 1/2+Ω(d/n) fraction of satisfied OPI constraints via streaming requires Ω(n) classical bits of memory but only O(dlogn) qubits. This yields provable quantum advantage in a "low-rate" regime when the number of evaluation points is much larger than the degree, with a space advantage that can be as large as exponential in certain parameter settings.
Computational complexity is the field that studies the computational resources needed by algorithms to correctly solve computational problems. In this thesis, we derive lower bounds on the computational complexity of different problems for different classes of algorithms. The computational resources we consider are the running time, the memory usage and the input query complexity; i.e., the number of positions in the input that the algorithm reads. While deriving upper bounds involves constructing algorithms that run efficiently in their computational resources, lower bounds for a computational problem establish limits on how fast or space-efficient any algorithm solving that problem can be. We consider two main themes of lower bounds: time-space tradeoffs and query complexity lower bounds. Time-space tradeoffs define a relation between the running time and space of any algorithm solving a certain problems. They give a complete picture of the hardness of a problem by deriving a lower bound on one resource when the other is limited. Query complexity measures the depth of a decision tree computing the problem. We derive several lower bounds and upper bounds for different problems on different types of classical computational models and on quantum computers, where the queries to the input are represented by quantum operations operating on the qubits of the algorithm: 1) We derive the largest time-space tradeoff known for a randomized algorithm solving an explicit problem. We consider the pointer jumping problem, also known as the outdegree-1 graph accessibility problem and derive a T in Ω(n log(n/S) log log(n/S)) time-space tradeoff for any randomized oblivious algorithm computing it. Oblivious algorithms have the property that the sequence of operations performed is independent of the value of the input. We derive a similar lower bound for randomized oblivious algorithms with boolean inputs. 2) Frequency moments and order statistics are attributes computed on a sequence of numbers and are extremely relevant in various data analysis settings. We consider computing several copies of these functions on overlapping sequences of numbers by considering a window of a fixed length and sliding it over the input; these versions are called sliding-window versions of these functions. We give hardness separations between the sliding-window versions of these statistics. On one hand, we give a TS in Ω(n 2 ) lower bounds for any algorithm computing sliding-window frequency moments. To complement our lower bound, we give a TS in O(n 2 log c n) comparison-based algorithm for sliding-window frequency moments. On the other hand we describe a T 2 S in O(n 3 log c n) algorithm for sliding-window element distinctness. For order statistics, we derive a similar lower bound of TS in Ω(n 2 ) for any algorithm computing a sliding-widow version of the median, while for the sliding window version of maximum (and minimum), we construct an O(n log n)-time, logarithmic space algorithm. 3) We study the increase in complexity on quantum computers when considering sliding-window frequency moments. We show that quantum computers do not require the same increase in complexity as their classical counterparts by constructing a quantum algorithm for sliding-window version of the zero th frequency moments running in time O(n 3/2 ) and logarithmic space, thus beating the classical lower bound for this problem. 4) In the area of error-correcting codes, we study the query complexity of error detection and correction algorithms. We prove that linear codes over large alphabets that are small in size and concentrated in terms of their codeword weights have testers and correctors that require only a constant number of queries to a word received across a noisy channel. Such codes are called locally-testable and self-correctable. As random codes of small size have the property that their codeword weights are concentrated in a small range of values, they are also locally-testable and self-correctable with high probability. 5) For linear codes with information-efficient encoding, we study the tradeoff between the time and space of their encoders, and their error-correction property. We derive properties on generator matrices of asymptotically good linear codes that guarantee an optimal time-space tradeoff of TS in Ω(n 2 ) for any algorithm encoding them. Moreover, we prove a first step in an approach to establish these properties for all generator matrices of such codes. 6) To compare classical computers to their quantum counterparts, we consider the query complexity of quantum algorithms computing frequency moments and the onto function that takes as input a function and determines whether it is surjective. We prove that any quantum computer computing the onto function or the parity of the zero th frequency moment requires a linear number of queries to the input. Since the onto function with boolean inputs has a simple polynomial-size, constant-depth circuit, we obtain an almost linear lower bound on the query complexity of AC 0 , a special class of circuits that have polynomial size, constant depth and unbounded fan-in for their gates.
Finding the singularity of a matrix is a basic problem in linear algebra. Chu and Schnitger [SC95] first considered this problem in the communication complexity model, in which Alice holds the first half of the matrix and Bob holds the other half. They proved that the deterministic communication complexity is Omega(n^2 log p) for an n by n matrix over the finite field F_p. Then, Clarkson and Woodruff [CW09] introduced the singularity problem to the streaming model. They proposed a randomized one pass streaming algorithm that uses O(k^2 log n) space to decide if the rank of a matrix is k, and proved an Omega(k^2) lower bound for randomized one-way protocols in the communication complexity model. We prove that the randomized/quantum communication complexity of the singularity problem over F_p is Omega(n^2 log p), which implies the same space lower bound for randomized streaming algorithms, even for a constant number of passes. The proof uses the framework by Lee and Shraibman [LS09], but we choose Fourier coefficients as the witness for the dual approximate norm of the communication matrix. Moreover, we use Fourier analysis to show the same randomized/quantum lower bound when deciding if the determinant of a non-singular matrix is a or b for non-zero a and b.
We derive lower bounds for tradeoffs between the communication C and space S for communicating circuits. The first such bound applies to quantum circuits. If for any function f with image Z the multicolor discrepancy of the communication matrix of f is 1/2^d, then any bounded error quantum protocol with space S, in which Alice receives some l inputs, Bob r inputs, and they compute f(x_i,y_j) for the lr pairs of inputs (x_i,y_j) needs communication C=Ω(lrd \log |Z|/S). In particular, n\times n-matrix multiplication over a finite field F requires C=Θ(n^3\log^2 |F|/S). We then turn to randomized bounded error protocols, and derive the bound C=Ω(n^3/S^2) for Boolean matrix multiplication, utilizing a new direct product result for the one-sided rectangle lower bound on randomized communication complexity. This implies a separation between quantum and randomized protocols.
We define a new notion of information cost for quantum protocols, and a corresponding notion of quantum information complexity for bipartite quantum channels, and then investigate the properties of such quantities. These are the fully quantum generalizations of the analogous quantities for bipartite classical functions that have found many applications recently, in particular for proving communication complexity lower bounds. Our definition is strongly tied to the quantum state redistribution task. Previous attempts have been made to define such a quantity for quantum protocols, with particular applications in mind; our notion differs from these in many respects. First, it directly provides a lower bound on the quantum communication cost, independent of the number of rounds of the underlying protocol. Secondly, we provide an operational interpretation for quantum information complexity: we show that it is exactly equal to the amortized quantum communication complexity of a bipartite channel on a given state. This generalizes a result of Braverman and Rao to quantum protocols, and even strengthens the classical result in a bounded round scenario. Also, this provides an analogue of the Schumacher source compression theorem for interactive quantum protocols, and answers a question raised by Braverman. We also discuss some potential applications to quantum communication complexity lower bounds by specializing our definition for classical functions and inputs. Building on work of Jain, Radhakrishnan and Sen, we provide new evidence suggesting that the bounded round quantum communication complexity of the disjointness function is Ω(n/M + M), for M-message protocols. This would match the best known upper bound.