Forward-reverse error mitigation algorithm for quantum annealers

We propose a novel way to try to improve ground-state sampling statistics on quantum annealers with no cost in ancilla qubits—“forward-reverse error mitigation” (FREM) sampling. FREM starts by partitioning a Hamiltonian such that H = H_F + H_R and proceeds by forward annealing H_F while backward annealing H_R. While there are no strict requirements on how H should be partitioned, one should have a good approximation of the ground-state of H projected onto the qubits in HR for the reverse anneal. We study the efficacy of FREM using numerical simulations. In particular, our simulation is modelled after the annealing processes on a D-Wave 2000Q, and we use it to compare the ground-state sampling success of forward, reverse, and FREM annealing by comparing their Kullback-Leibler divergence with respect to direct diagonalization. Overall, this work provides an interesting new method to attempt to mitigate errors on near-term quantum annealers with limited qubit numbers.