Sharp Asymptotic Minimaxity of the Gavrilov-Benjamini-Sarkar Step-Down Testing Procedure in Sparse Gaussian Sequence Models
Abstract
False discovery rate control does not by itself guarantee optimal recovery of sparse signals: a procedure may make few false discoveries while missing an unnecessarily large fraction of the alternatives. We investigate whether the classical Gavrilov--Benjamini--Sarkar (GBS) step-down procedure also achieves the sharp frequentist minimax limits for sparse recovery in independent Gaussian sequence models. Under a beta-min separation at the critical sparse detection boundary, GBS attains the exact asymptotic risk constant Φ̅(b) for normalized Hamming loss and combined FDP+FNP loss. The Hamming conclusion remains valid when the number of signals is only bounded above by the nominal sparsity sn. For heterogeneous signal magnitudes, the corresponding risk under either loss equals the sharp benchmark Λn(a)+o(1), without assuming that Λn(a) converges; this is additive, rather than necessarily relative, sharpness. We assume sn→∞, n/sn→∞, and a vanishing nominal level αn satisfying log(1/αn)=o(√(log(n/sn))). Unlike step-up rules, a step-down procedure must satisfy every preceding ordered p-value constraint before reaching a given rank. Our analysis handles this sequential obstruction directly: a finite-sample shifted order-statistic bound makes false positives negligible, while new uniform crossing arguments show that the rejection path reaches the optimal fraction of signals. The results extend the range of testing architectures known to attain the sharp minimax benchmarks, without requiring localization of a single rejection threshold.
Related
- Asymptotic Bayes Optimality Under Sparsity of the Gavrilov-Benjamini-Sarkar Step-Down Testing Procedure
- How Much Can Gaussian Dependence Inflate the Benjamini-Hochberg Procedure's FDR?
- The Benjamini-Hochberg procedure can fail to control the FDR for correlated two-sided Gaussian tests.
- A Non-Asymptotic Gaussian Likelihood Theory for Long-Memory Toeplitz Models
- Non-asymptotic bounds for the average singular value of a complex Gaussian matrix
BibTeX
@misc{ghosh2026sharp,
title = {{Sharp Asymptotic Minimaxity of the Gavrilov-Benjamini-Sarkar Step-Down Testing Procedure in Sparse Gaussian Sequence Models}},
author = {Prasenjit Ghosh},
howpublished = {arXiv},
year = {2026},
month = oct,
doi = {10.48550/arxiv.2607.05893},
eprint = {2607.05893},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2607.05893},
}
RIS
TY - UNPB
TI - Sharp Asymptotic Minimaxity of the Gavrilov-Benjamini-Sarkar Step-Down Testing Procedure in Sparse Gaussian Sequence Models
AU - Prasenjit Ghosh
PY - 2026
DA - 2026/10/24
PB - arXiv
DO - 10.48550/arxiv.2607.05893
UR - https://arxiv.org/abs/2607.05893
AB - False discovery rate control does not by itself guarantee optimal recovery of sparse signals: a procedure may make few false discoveries while missing an unnecessarily large fraction of the alternatives. We investigate whether the classical Gavrilov--Benjamini--Sarkar (GBS) step-down procedure also achieves the sharp frequentist minimax limits for sparse recovery in independent Gaussian sequence models. Under a beta-min separation at the critical sparse detection boundary, GBS attains the exact asymptotic risk constant Φ̅(b) for normalized Hamming loss and combined FDP+FNP loss. The Hamming conclusion remains valid when the number of signals is only bounded above by the nominal sparsity s_n. For heterogeneous signal magnitudes, the corresponding risk under either loss equals the sharp benchmark Λ_n(a)+o(1), without assuming that Λ_n(a) converges; this is additive, rather than necessarily relative, sharpness. We assume s_n→∞, n/s_n→∞, and a vanishing nominal level α_n satisfying log(1/α_n)=o(√(log(n/s_n))). Unlike step-up rules, a step-down procedure must satisfy every preceding ordered p-value constraint before reaching a given rank. Our analysis handles this sequential obstruction directly: a finite-sample shifted order-statistic bound makes false positives negligible, while new uniform crossing arguments show that the rejection path reaches the optimal fraction of signals. The results extend the range of testing architectures known to attain the sharp minimax benchmarks, without requiring localization of a single rejection threshold.
ER -