Dice Question Streamline Icon: https://streamlinehq.com

Interpretation of the range space condition for countable discrete domains

Investigate and characterize the range space condition Ran(T_{P0}^θ) induced by the Stein kernel K0 and the integral operator T_{P0}: L^2(P0)→L^2(P0) when the domain is countable (so that L^2(P0) is isomorphic to ℓ^2), and develop a detailed interpretation of the corresponding smoothness class for KSD-based goodness-of-fit testing on countably supported discrete spaces.

Information Square Streamline Icon: https://streamlinehq.com

Background

The paper’s minimax framework defines alternatives via the range space Ran(T_{P0}θ), capturing smoothness through the spectrum of the operator built from the Stein kernel and the null distribution. While several continuous-domain examples are analyzed, the authors point out that for countably supported discrete domains, L2(P0) becomes ℓ2 and the structure of Ran(T_{P0}θ) needs dedicated treatment.

This open problem seeks a thorough, domain-specific understanding of the range-space smoothness condition for countable discrete settings, which would clarify the alternative classes and guide KSD-based testing in such domains.

References

A more interesting scenario is when is countable, in which case L2(P_0) is isomorphic to the sequence space, ℓ2(). We leave a detailed study of interpreting the range space condition in this setting for future work.

Minimax Optimal Goodness-of-Fit Testing with Kernel Stein Discrepancy (2404.08278 - Hagrass et al., 12 Apr 2024) in Section 5 (Interpreting the class of alternatives, 𝒫) — end of section