A probabilistic quantum algorithm for Lyapunov equations and matrix inversion

We present a probabilistic quantum algorithm for preparing mixed states which, in expectation, are proportional to the solutions of Lyapunov equations -- linear matrix equations ubiquitous in the analysis of classical and quantum dynamical systems. Building on previous results by Zhang et al., arXiv:2304.04526, at each step the algorithm either returns the current state, applies a trace non-increasing completely positive map, or restarts depending on the outcomes of a biased coin flip and an ancilla measurement. We introduce a deterministic stopping rule which leads to an efficient algorithm with a bounded expected number of calls to a block-encoding and a state preparation circuit representing the two input matrices of the Lyapunov equations. We also consider approximating the normalized inverse of a positive definite matrix $A$ with condition number $κ$ up to trace distance error $ε$. For this special case the algorithm requires, in expectation, at most $\lceil κ\ln(1/ε) \rceil+1$ calls to a block-encoding of $\sqrt{A/\|A\|}$. This matches the optimal query complexity in $κ$ and $ε$ of the related, but distinct, quantum linear system solvers. In its most general form, the algorithm generates mixed states which approximate matrix-valued weighted sums and integrals.