Fast differentiable evolution of quantum states under Gaussian transformations

TL;DR

This paper introduces a faster, differentiable algorithm for simulating Gaussian quantum state evolution, significantly reducing computational complexity and enabling gradient-based optimization for quantum circuit design.

Contribution

A novel algorithm that computes Gaussian state evolution more efficiently without generating the full transformation matrix, facilitating optimization tasks.

Findings

Reduces computational complexity from exponential to more manageable levels.

Enables gradient-based optimization of quantum circuits involving Gaussian transformations.

Achieves up to tenfold speedup in circuit optimization benchmarks.

Abstract

In a recent work we presented a recursive algorithm to compute the matrix elements of a generic Gaussian transformation in the photon-number basis. Its purpose was to evolve a quantum state by building the transformation matrix and subsequently computing the matrix-vector product. Here we present a faster algorithm that computes the final state without having to generate the full transformation matrix first. With this algorithm we bring the time complexity of computing the Gaussian evolution of an $N$-dimensional $M$-mode state from $O(MN^{2M})$ to $O(M(N^2/2)^M)$, which is an exponential improvement in the number of modes. In the special case of high squeezing, the evolved state can be approximated with complexity $O(MN^{M})$. Our new algorithm is differentiable, which means we can use it in conjunction with gradient-based optimizers for circuit optimization tasks. We benchmark our algorithm by optimizing circuits to produce single photons, Gottesman-Kitaev-Preskill states and NOON states, showing that it is up to one order of magnitude faster than the state of the art.