Core Concepts
Quantum error correction can be significantly improved by dynamically adapting the decoding graph weights to account for drifted noise and correlated errors on real quantum hardware.
Abstract
The content discusses two key challenges in quantum error correction (QEC) using the popular Minimum-Weight-Perfect-Matching (MWPM) decoder:
Noise Drift: Noise in real quantum systems can drift over time due to various factors, leading to a mismatch between the noise model used by the MWPM decoder and the actual noise characteristics on the hardware. This mismatch can severely degrade the performance of the decoder.
Noise Correlation: The MWPM decoder assumes independence between errors, but in reality, errors can be correlated, especially in codes like surface code and honeycomb code. Overlooking these correlations can also impact the decoder's accuracy.
To address these challenges, the authors propose Decoding Graph Re-weighting (DGR), a two-part approach:
Alignment Re-weighting:
Maintains an occurrence tracer to track the frequency of edges in the decoding graph matchings over previous trials.
Uses this edge occurrence statistics to dynamically update the weights in the decoding graph, aligning them with the actual noise on the hardware.
Correlation Re-weighting:
Maintains a correlation tracer to track the co-occurrence of edge pairs in the decoding graph matchings.
Leverages this correlation information to adjust the weights of related edges, explicitly modeling the noise correlations.
The authors evaluate DGR on surface code and honeycomb code under various noise models and mismatch scenarios. DGR achieves significant improvements, reducing the logical error rate by 3.6x on average and up to 5000x in the worst-case mismatch scenario, compared to the traditional MWPM decoder.
Stats
Noise in real quantum systems can drift by over 15x within 100 seconds and 1000x over a year.
The majority of edges in the decoding graph have more than one correlated edge.
Quotes
"Noise in quantum systems is known to undergo significant drift over time, which can arise from various sources."
"The MWPM decoder has the assumption of independence between all the edges in the decoding graph. However, in reality, the edges are correlated."