Algorithm-Centric Error Mitigation*

We derive algorithm-specific error mitigation procedures for VQE. The core idea is to determine which part of a circuit contributes most to the final output error and then harden only these sub-circuits. Using simulation with gate level noise injection, our results indicate that VQE is most sensitive to errors on the source qubit of CNOT gates, and relatively insensitive to errors on any other qubits or single gates, irrespective of the noise distribution. We then devise an algorithm that attempts to increase the fidelity of qubits that appear as sources of CNOT gates. Given a circuit, we determine the qubits that appear in most CNOT gates (contribute most to error) and use ancilla qubits to mimic their operations (same state) and periodically switch between the original and ancilla qubit during circuit operation. Simulation results indicate that the technique is able to improve the quality of the solution from the quantum circuit. This is confirmed by experiments on the IBM hardware, where for a 4-qubit VQE with one ancilla qubit we observe fidelity improving from 46% to 52%.