Solving a Linear System of Equations on a Quantum Computer by Measurement

TL;DR

The paper introduces a measurement-based variational quantum algorithm capable of solving dense linear systems efficiently, overcoming limitations of previous methods in accuracy, matrix density, and eigenvalue encoding.

Contribution

It presents a novel measurement test algorithm that directly maximizes target fidelity without relying on Pauli decomposition, suitable for dense matrices.

Findings

Successfully simulated dense 16x16 matrices with non-zero determinants.

Achieves target fidelity with accuracy scaling as 1/N per measurements.

Overcomes previous variational algorithms' limitations regarding matrix density and condition number.

Abstract

We present a variational algorithm for fault tolerant quantum computing to solve a system of linear equations which directly maximises the parameters of the target fidelity. This so-called measurement test algorithm can be applied to any computational task with a solution that is represented as eigenvector of a self-adjoint matrix. The solution is prepared as state of a register in the quantum computer by a von Neumann measurement of a corresponding observable, which is implemented using the phase estimation algorithm. The probability to project the system thus into the unknown target state, which equals the target fidelity, is measured in terms of relative frequencies and iteratively optimised to read out the target state. The new algorithm overcomes three issues of previous variational quantum algorithms: i) It does not rely on a decomposition in terms of Pauli strings and therefore can compute eigenvectors of dense matrices. ii) The accuracy is not limited by the condition number $\kappa$ of the matrix, provided a logarithmic number ($O(\log\kappa)$) of qubits is used to encode the eigenvalues and iii) the target fidelity $F_T = 1-\epsilon$ can be reached with an accuracy $\epsilon$ that scales with $1/N$ for $N$ measurements per iteration. We demonstrate this by numerical simulations for dense random real-valued $16\times 16$ matrices with non-vanishing determinant.