Correlation matrix tool for error diagnostics in QEC experiments

Identification and mitigation of nonconventional errors such as leakage and cross-talk in repetition and surface code experiments is essential to achieve exponential suppression of logical errors with increasing the code distance. In this talk, we introduce an error-diagnostic tool that allows us to characterize long-range as well as long-time errors on the error graph caused by, e.g., cross-talk or leakage to non-computational states. The probability p_ij of an error involving arbitrary nodes i and j of the error graph is extracted from correlation of the error detection events at these nodes. The matrix p_ij can be used to identify particular error mechanisms and their strengths. In addition, these probabilities can provide accurate edge weights for minimum-weight-perfect-matching decoders.