Comments on a 2026 stochastic subgradient method for VLSI placement: Exact formula checks and questions about the stochastic convergence argument

This independent technical note examines Yue, Dai and Yu, “An efficient stochastic subgradient method for the global placement problem in very large-scale integration circuits,” Journal of Global Optimization 95 (2026), 863–892, DOI: 10.1007/s10898-026-01619-4. It corrects a displayed ReLU identity, identifies an undefined component-wise quotient under a literal reading of the adaptive-penalty rules, and discusses discrepancies between the printed algorithm and analyzed recurrence. It examines filtration and conditional-expectation issues and provides a two-cell example refuting a covariance-sign inference in the convergence proof. This is not a counterexample to the main convergence theorem, and it does not establish that the numerical experiments are invalid. This deposit preserves the original note dated 3 November 2026 and its accompanying LaTeX source and exact-arithmetic Python script, initially prepared for private scholarly correspondence and now made public. References to private circulation in the PDF describe its original context. No private correspondence is included. This is an independent technical note, not a journal-issued corrigendum or an endorsement by the original authors. AI assistance and responsibility: An OpenAI language model was used for English improvement and numerical checks, and assisted with drafting and mathematical analysis, including development of counterexamples and draft arguments. The author bears full responsibility for the construction and verification of counterexamples, lemmas and propositions, all proofs and conclusions, and the final wording and mathematical correctness of the manuscript.