Title: Recursive Variational Quantum Compiling

Abstract: Variational quantum compiling algorithms \cite{qaqc} aim to compress quantum circuits into parameterized ansatzes utilizing a hybrid quantum-classical approach. In this article a variant of VQC named the recursive variational quantum compiling algorithm (RVQC) is proposed. In contrast with existing VQC algorithms, where one is required to execute the full target circuit in order to compile it, RVQC works by dividing the target circuit into $N$ parts and recursively compresses one part at a time into the parameterized ansatz. Since smaller parts are being compressed at a time, the circuit depth required to implement the algorithm is not dependent on the depth of the target circuit, but on the depth of the ansatz and the depth of each of the $N$ parts. We are hence able to compress the target circuit by executing circuits of smaller depths than the target circuit itself. To show that this approach has an edge over the standard VQC method for sufficiently large circuit depths, RVQC was compared with VQC on a noise model of the IBM Santiago device with the goal of compressing a randomly generated five-qubit circuit of depth 5000. It was shown that VQC was not able to converge within 500 epochs of training. On the other hand, RVQC was able to converge to a fidelity ($0.88$) within a total of 500 epochs when splitting the target circuit into $N = 5$ parts.