Sergii Strelchuk

We revisit the problem of characterising the complexity of Quantum PAC learning, as introduced by Bshouty and Jackson [1]. Several quantum advantages have been demonstrated in this setting, however, none are generic: they apply to particular concept classes and typically only work when the distribution that generates the data is known. In the general case, it was recently shown by Arunachalam and de Wolf [2] that quantum PAC learners can only achieve constant factor advantages over classical PAC learners. We show that with a natural extension of the definition of quantum PAC learning used by Arunachalam and de Wolf, we can achieve a generic advantage in quantum learning. To be precise, for any concept class C of VC dimension d, we show there is an (ϵ,δ)-quantum PAC learner with sample complexity O(1/√ϵ[d+log(1/δ)]log^9(1/ϵ)). Up to polylogarithmic factors, this is a square root improvement over the classical learning sample complexity. We show the tightness of our result by proving an Ω(d/√ϵ) lower bound that matches our upper bound up to polylogarithmic factors. [1] SIAM J. Comput. 1998, 28, 1136-1153 [2] JMLR, 19 (2018) 1-36

Magic states, pivotal for universal quantum computation via classically simulable Clifford gates, often undergo decomposition into resourceless stabilizer states, facilitating simulation through classical means. This approach yields three operationally significant metrics: stabilizer rank, fidelity, and extent. We extend these simulation methods to encompass matchgate circuits (MGCs), and define equivalent metrics for this setting. We begin with an investigation into the algebraic constraints defining Gaussian states, marking the first explicit characterisation of these states. The explicit description of Gaussian states is pivotal to our methods for tackling all the simulation tasks. Central to our inquiry is the concept of Gaussian rank – a pivotal metric defining the minimum terms required for decomposing a quantum state into Gaussian constituents. This metric holds paramount significance in determining the runtime of rank-based simulations for MGCs featuring magic state inputs. The absence of low-rank decompositions presents a computational hurdle, thereby prompting a deeper examination of fermionic magic states. We find that the Gaussian rank of 2 instances of our canonical magic state is 4 under symmetry-restricted decompositions. Additionally, our numerical analysis suggests the absence of low-rank decompositions for 2 or 3 copies of this magic state. Further, we explore the Gaussian extent, a convex metric offering an upper bound on the rank. We prove the Gaussian extent's multiplicative behaviour on 4-qubit systems, along with initial strides towards proving its sub-multiplicative nature in general settings. One important result in that direction we present is an upper bound on the Gaussian fidelity of generic states.

Port-based teleportation (PBT) is a form of quantum teleportation in which no corrective unitary is required on the part of the receiver. Two primary regimes exist - deterministic PBT in which teleportation is always successful, but is imperfect, and probabilistic PBT, in which teleportation succeeds with probability less than one, but teleportation is perfect upon a success. Two further regimes exist within each of these in which the resource state used for the teleportation is fixed to a maximally entangled state, or free to be optimised. Recently, works resolved the long-standing problem of efficiently implementing port-based teleportation, tackling the two deterministic cases for qudits. Here, we provide algorithms in all four regimes for qubits. Emphasis is placed on the practicality of these algorithms, where we give polynomial improvements in the known gate complexity for PBT, as well as an exponential improvement in the required number of ancillas (albeit in separate protocols). Our approach to the implementation of the square-root measurement in PBT can be directly generalised to other highly symmetric state ensembles. For certain families of states, such a framework yields efficient algorithms in the case that the Petz recovery algorithm for the square-root measurement runs in exponential time.

We give a strengthening as well as a generalization of an inequality for the quantum conditional mutual information of a tripartite quantum state recently proved by Fawzi and Renner, connecting it with the ability to reconstruct the state from its bipartite reductions. We provide three alternative and simplified proofs ranging from quantum state redistribution via duality of semidefinite programming to elementary properties of pinching maps and the operator logarithm.