Dynamical decoupling for arbitrary large-scale quantum algorithms

Dynamical decoupling (DD) is a method to suppress errors by repeatedly reversing the sense of error accumulation during idle delays. Ideally, DD suppresses dephasing errors on each qubit, and crosstalk errors on all pairs of qubits. The efficacy of crosstalk suppression depends jointly on DD sequences applied to both participating qubits. In a typical quantum circuit, delays are distributed asynchronously by the compiler; the qubits become idle or active at different times, and each qubit may in general be subjected to crosstalk from multiple neighbors at the same time. The error terms accumulate at different unknown rates and over various temporal intervals that may be fully or partially aligned. Given these complications, the optimal crosstalk-suppressing DD sequence is highly dependent on the specific circuit; it cannot be derived once and then applied naively to different circuits.

In this talk, we describe how ideal DD sequences may be derived efficiently for arbitrary input circuits. Our method scales linearly with the number of idle instructions, and remains tractable far beyond currently feasible circuit sizes. We show up to 9000x improvements in algorithmic performance using a correctly circuit-optimized DD embedding, compared to a default approach available in Qiskit.