Low Depth Virtual Distillation of Quantum Circuits by Deterministic Circuit Decomposition

Abstract: Virtual distillation (VD) using measurements of multiple copies of a quantum circuit have recently been proposed as a method of noise mitigation of expectation values. Circuit decompositions known as B gates were found only for single qubit expectation values however practical calculations require multi-qubit expectation values which cannot be corrected with B gates. We discover low depth circuit decompositions for multi-qubit expectation values by combining multiple projections to recover the correct measurement statistics or expectation values. Our method adds linear entangling gates with number of qubits, but requires extra measurements. Furthermore, in applications to find ground states such as the variational quantum eigensolver (VQE) algorithm, the variational principle is required which states the energy cannot go below the ground state energy. We discover that the variational principle is violated if noise is higher on single expectation values than multi-qubit which renders VQE useless. We show this occurs when using B gates and is preserved if using our low depth decomposition on all expectation values. We perform demonstration on real devices and demonstrate our decomposition can mitigate real experimental noise in VQE for the H$_2$ molecule with a two qubit tapered mapping, H$_3$ with three qubits, and H$_2$ with four qubits. Our decomposition provides a way to perform duplicate circuit virtual distillation on real devices at significantly lower depth and for arbitrary observables.