Signed-weight L log L-Luxemburg discrepancy

Determine whether the curse of dimensionality holds for the L log L-Luxemburg discrepancy when signed quadrature weights are allowed, thereby resolving the corresponding discrepancy problem without the nonnegativity restriction.

Background

The paper proves that the normalized L log L-Luxemburg discrepancy generated by the Young function Φ(s)=s log(1+s) suffers from the curse of dimensionality for arbitrary nonnegative quadrature weights, and consequently also for equal-weight quasi-Monte Carlo rules. The proof relies essentially on representing the normalized counting function as a positive mixture of anchored-box atoms and applying a subadditivity estimate to that positive mixture.

Allowing signed weights invalidates the positivity-based argument used in the proof. The authors explicitly identify the extension to signed weights as unresolved, without asserting whether the curse of dimensionality persists or fails in that setting.

References

The nonnegativity assumption is essential to the proof in Section~\ref{sec:proof}: it permits a positive-mixture representation of the normalized counting function and the subadditivity estimate used in Lemma~\ref{lem:mixture}. The corresponding problem for signed weights remains open.

— The $L\log L$-Luxemburg Discrepancy for Nonnegative Weights Suffers from the Curse of Dimensionality  (2610.21717 - Dick et al., 15 Oct 2026) in Section 2, immediately following Theorem 2.1