New Quantum Algorithms Deliver Speed-Ups Without Sacrificing Predictability

Researchers have begun to systematically investigate pseudo-deterministic quantum algorithms, a novel class of quantum computation that consistently yields a canonical solution with high probability. Hugo Aaronson and Tom Gur from the University of Cambridge, working with Jiawei Li from UT Austin, present compelling evidence of their potential and limitations within the query complexity model. Their findings, detailed in a new paper, demonstrate significant complexity separations, including a problem where pseudo-deterministic quantum algorithms require substantially more queries than their classical randomised counterparts. This work is particularly significant as it establishes both the advantages, an exponential speed-up for certain problems, and the boundaries, a quintic advantage over deterministic algorithms, of this emerging computational paradigm, potentially reshaping our understanding of quantum algorithmic power.

Problems currently intractable for even the most powerful computers could yield to a new class of quantum algorithms. These ‘pseudo-deterministic’ quantum methods find correct answers with high probability, offering speed-ups for specific calculations. Initial results demonstrate an exponential advantage over classical approaches for certain problems, such as Quantum-Locked Estimation. Meanwhile, remaining within a quintic limit for general computations.

Scientists have begun a systematic investigation into pseudo-deterministic quantum algorithms, representing a unique intersection between the power of quantum mechanics and the reliability of deterministic computation. Recent work has focused on the query complexity model, revealing surprising separations in what these algorithms can achieve compared to classical counterparts.

Understanding these differences has implications for the development of more dependable quantum technologies and a deeper understanding of the fundamental limits of computation. To establish clear boundaries between what quantum and classical algorithms can accomplish, even with pseudo-determinism, presents a considerable challenge. Scientists have identified a specific problem, termed Avoid One Encrypted String, where classical randomised algorithms perform with limited efficiency. But any pseudo-deterministic quantum algorithm requires a substantially greater number of queries to reach a solution.

This disparity suggests that quantum mechanics can offer an advantage even when predictability is central. A newly defined problem, Quantum-Locked Estimation, demonstrates an exponential speed-up for pseudo-deterministic quantum algorithms over classical pseudo-deterministic approaches, while maintaining reasonable query complexity for randomised algorithms.

Investigations reveal that for any given search problem, pseudo-deterministic quantum algorithms can achieve, at most, a quintic advantage over purely deterministic algorithms. A broad range of quantum search problems, including Grover search, element distinctness, and graph collision, can be adapted to operate in a pseudo-deterministic manner with minimal additional computational cost.

These findings suggest a subtle relationship between quantum speed-ups and the requirement for predictable outputs, potentially linking quantum pseudo-determinism to the generation of certified randomness. For instance, the Avoid One Encrypted String problem presents a scenario where classical randomised algorithms succeed with constant probability through random sampling.

However, a pseudo-deterministic quantum algorithm faces a more demanding task, needing to consistently avoid a hidden string across multiple runs — effectively requiring it to extract information about the string itself. This highlights a critical distinction: while quantum algorithms often rely on probabilistic outcomes, pseudo-determinism demands a consistent, predictable result. Forcing the algorithm to overcome inherent randomness. Beyond these separations, The project also introduces the concept of “canonization”, and where a quantum algorithm stabilizes a probable solution from a randomised algorithm into a definitive, predictable output.

Constructing the Avoid One Encrypted String problem to probe algorithmic limitations

The project into pseudo-deterministic quantum computation presented in this effort underpins a 72-qubit superconducting processor. Initially, researchers focused on establishing query complexity separations, employing techniques designed to address the unique challenges posed by pseudo-determinism, where algorithms output a correct solution with high probability.

Well-established lower bound techniques, such as the polynomial method, proved inadequate, necessitating the development of new approaches. As a result, The effort prioritised constructing problems specifically tailored to highlight the limitations of pseudo-deterministic algorithms, central to which was designing the ‘Avoid One Encrypted String’ (AOES) problem, leveraging the ability of randomised algorithms to generate random strings.

Here, this problem consists of multiple instances of the XOR problem, encoding a secret string, and challenges algorithms to output any string differing from this secret. Once the AOES problem was defined, a randomised reduction from XOR was implemented, demonstrating that a zero-error quantum algorithm solving AOES with a certain query complexity could solve XOR more efficiently.

To further explore the capabilities of pseudo-deterministic quantum algorithms, the team formulated the ‘Quantum-Locked Estimation’ (QL-Estimation) problem. Here, the input combines an instance of Simon’s problem, a well-known quantum algorithm benchmark, with an instance of the Hamming problem, which asks for an estimate of the number of 1s in a string.

Instead of seeking a general solution, the QL-Estimation problem asks for an estimate of the Hamming weight that is already known to be a valid solution to Simon’s problem. In turn, a classical randomised algorithm estimates the Hamming weight, while a quantum algorithm directly calculates the solution encoded within Simon’s problem, effectively canonising a particular estimate.

For both AOES and QL-Estimation, The team carefully analysed the query complexity — assessing how many queries to the input are needed to solve the problem. By carefully constructing these problems and analysing their query complexity, and the effort aimed to delineate the boundaries between what is classically and quantumly achievable with pseudo-deterministic algorithms. Also, a class of quantum search problems, including search, element distinctness. Triangle finding, can be rendered pseudo-deterministic with minimal overhead.

Maximal quantum speed-up for avoiding a hidden string in encrypted XOR gadgets

Initial analysis reveals a clear separation between classical randomised query complexity and quantum pseudo-deterministic query complexity for the Avoid One Encrypted String problem, achieving a complexity of O(1) versus Ω(N). Meanwhile, this establishes a maximal separation, exceeding previous bounds demonstrated by the Find1 problem which had a pseudo-deterministic quantum lower bound of Ω(√N).

The AOES problem involves encoding multiple XOR gadgets that encrypt a hidden string. At the same time, a classical randomised algorithm succeeds with constant probability through random sampling. Conversely, any pseudo-deterministic algorithm must consistently avoid the hidden string, necessitating the learning of at least one bit of information. Here, the project extends beyond demonstrating limitations, also exhibiting the power of quantum pseudo-determinism.

Quantum pseudo-deterministic algorithms achieve an exponential speed-up over classical pseudo-deterministic algorithms on Quantum-Locked Estimation, while maintaining a randomised query complexity of O(log N). In turn, this advantage stems from “canonization”, where a quantum algorithm stabilizes a probable solution from a randomised algorithm into a definitive, canonical solution.

Specifically, a quantum pseudo-deterministic algorithm with O(log N) query complexity canonizes a randomised classical algorithm also with O(log N) complexity. Whereas any classical pseudo-deterministic algorithm requires Ω(√N) queries. Fundamental properties of quantum pseudo-determinism were also explored. Across total problems, pseudo-deterministic quantum algorithms cannot exceed deterministic algorithms by more than a quintic factor — as demonstrated by the theorem stating D(R) = O(psQ(R)5 log(N)).

At the same time, this contrasts with classical pseudo-deterministic algorithms, which experience a quartic blow-up, represented by D(R) = O(psR(R)4 log(N)) — the proof relies on a new lemma that provides an upper bound on the number of possible outputs from a pseudo-deterministic quantum algorithm. A step that is trivial in the classical case. The Find1 problem remains complete for pseudo-deterministic classical query algorithms, and by extension, for pseudo-deterministic quantum query algorithms, and also, a class of quantum search problems, including search, element distinctness, triangle finding, k-sum. Graph collision, can be rendered pseudo-deterministic with minimal overhead, suggesting broad applicability of the techniques developed.

Pseudo-deterministic quantum computation decisively outperforms randomised classical approaches

Scientists have long sought to define the limits of what quantum computers can achieve beyond classical capabilities, and recent work offers a sharper understanding of this boundary. To establish clear separations between quantum and classical computational power has proved elusive, hampered by the difficulty of designing problems where quantum speedups are guaranteed and demonstrably distinct from clever classical algorithms.

This new research tackles this challenge by focusing on ‘pseudo-deterministic’ quantum computation. A model where quantum algorithms consistently yield the correct answer with high probability. Defining the advantages of this specific quantum model is not straightforward. By constructing problems specifically tailored to expose the strengths of pseudo-deterministic quantum systems, researchers have demonstrated a clear separation from classical randomised computation.

They’ve identified scenarios where quantum algorithms can solve problems with far fewer queries than their classical counterparts. Meanwhile, remaining within the bounds of pseudo-determinism. The extent of this advantage is limited. For general problems, the speedup is capped at a quintic factor compared to deterministic classical algorithms. The implications extend beyond theoretical computer science.

While a quintic advantage isn’t a revolution, understanding these limits is vital for guiding the development of practical quantum applications. Unlike earlier work relying on highly structured problems, these separations hold for more general cases, offering a more realistic assessment of quantum potential. A key limitation remains the reliance on query complexity as a measure of computational cost.