Repeat-Until-Success: Non-deterministic decomposition of single-qubit unitaries

We present a decomposition technique that uses non-deterministic circuits to approximate an arbitrary single-qubit unitary to within distance $ε$ and requires significantly fewer non-Clifford gates than existing techniques. We develop "Repeat-Until-Success" (RUS) circuits and characterize unitaries that can be exactly represented as an RUS circuit. Our RUS circuits operate by conditioning on a given measurement outcome and using only a small number of non-Clifford gates and ancilla qubits. We construct an algorithm based on RUS circuits that approximates an arbitrary single-qubit $Z$-axis rotation to within distance $ε$, where the number of $T$ gates scales as $1.26\log_2(1/ε) - 3.53$, an improvement of roughly three-fold over state-of-the-art techniques. We then extend our algorithm and show that a scaling of $2.4\log_2(1/ε) - 3.28$ can be achieved for arbitrary unitaries and a small range of $ε$, which is roughly twice as good as optimal deterministic decomposition methods.