Description
Accurate computation of chemical energies is a challenging problem for classical computers, that quantum computers are expected to be able to accelerate. While the argument for advantage in the fault-tolerant regime is relatively clear, the required circuit depths are well beyond current hardware capabilities. Recently, focus on near-term techniques has shifted to quantum-centric supercomputing techniques. In particular, sample-based quantum diagonalisation methods have been gaining traction due to their resilience to hardware noise and the fact that they do not require optimisation of circuit parameters. However, to achieve precision results on difficult chemical, huge sample sets in the tens to hundreds of millions are needed and techniques to extend the sample subspace without placing additional burden on the quantum processor are required. We have modified the sample-based approach with a Lanczos-cluster expansion framework that includes contributions from beyond the subspace sampled from the quantum hardware. This results in orders of magnitude improvement in the number of samples needed to achieve a given accuracy, which we demonstrate for molecular nitrogen using 52 qubits (and 5 ancillas) on IBM quantum hardware. Unlike other methods that expand the subspace - such as ext-SQD, TrimSQD or HI-VQE - the Lanczos-cluster framework is applied after the diagonalisation is performed and so can be integrated trivially with these methods to further improve the sample-efficiency.
| I am the presenting author | Yes |
|---|