How Many Directions Must a Truncated Diffusion Sampler Retain? Matching Bounds Under Power-Law Spectra
The research determines how many spectral directions ($K$) a truncated diffusion sampler needs to keep for good results when using power-law covariance spectra. It provides a formula for $K$ based on desired accuracy and resolution, showing that keeping too few directions causes errors, while keeping too many is computationally wasteful.