Abstract
The RBC & UKQCD collaboration presents an update on the lattice determination of $\epsilon'$, the measure of direct CP-violation in kaon decays. We focus on recent developments in statistical techniques for estimating standard errors and goodness-of-fit metrics for large datasets with complex correlations in both temporal and molecular dynamics time. Key advancements include a block jackknife for stable error estimation and a recentered bootstrap method for determining null distributions without assuming normality or independence.
Introduction: The Challenge of CP-Violation on the Lattice
Direct CP-violation, parameterized by $\epsilon'$, is a fundamental measure of the asymmetry between matter and antimatter. Calculating this from first principles using Lattice QCD is a significant challenge. Recent efforts by the RBC & UKQCD collaboration faced two primary hurdles: a 5$\sigma$ discrepancy in pion-pion ($\pi\pi$) scattering phase shifts ($\delta_0$) compared to experimental predictions, and the breakdown of standard statistical tools as dataset sizes increased.
Statistical Refinements
To address these issues, the team introduced two key statistical upgrades:
Block Jackknife
Traditional binning methods to handle autocorrelations in Molecular Dynamics (MD) time often lead to "uncontrolled growth" in error estimates when dealing with large covariance matrices. The authors implemented a block jackknife approach. Instead of pre-averaging data, entire blocks of consecutive samples are removed during resampling. This prevents the "low eigenvalue" problem in large matrices that typically causes error bars to explode.
Recentered Bootstrap
Standard p-values rely on assumptions of normality that lattice data frequently violates. The authors developed a recentered bootstrap technique to calculate reliable goodness-of-fit metrics. By generating resampled ensembles and "recentering" them to satisfy the null hypothesis, they can empirically construct the null distribution of the test statistic ($q^2$) directly from the data. This method is free from the usual assumptions of independence and large-$n$ data.
Key Results and Discussion
The application of these techniques, combined with an increased statistics pool (from 216 to 1,438 configurations), has yielded significant improvements.
Resolving the Phase Shift Discrepancy
By introducing additional $\pi\pi$ operators and performing simultaneous fits, the team isolated the ground state more effectively. This approach appears to resolve the previous 5$\sigma$ discrepancy with dispersive predictions for the $I=0$ phase shift.
Error Stability
The block jackknife method successfully produced a stable "plateau" in the relative error. As shown in the figure below, the error estimates stabilize rather than continuing to grow, providing a much more reliable basis for precision physics.
Robust Goodness-of-Fit
The new bootstrap method provided a p-value of approximately 0.3 for $\pi\pi$ two-point fits. This is a substantial correction from the 0.04 or 0.1 values that standard $\chi^2$ or Hotelling $T^2$ distributions would have erroneously predicted.
Conclusion
The RBC & UKQCD collaboration has established a robust statistical framework for high-precision lattice calculations. These upgrades are critical for the final determination of $\epsilon'$, which serves as a sensitive probe for physics beyond the Standard Model. Future work will focus on quantifying finite-$n$ biases to further refine these measurements.
Summary updated