2203.14527
Distributed Finite-Sum Constrained Optimization subject to Nonlinearity on the Node Dynamics
Mohammadreza Doostmohammadian, Maria Vrakopoulou, Alireza Aghasi, Themistoklis Charalambous
incompletemedium confidenceCounterexample detected
- Category
- Not specified
- Journal tier
- Note/Short/Other
- Processed
- Sep 28, 2025, 12:56 AM
- arXiv Links
- Abstract ↗PDF ↗
Audit review
The paper correctly states feasibility/sum-invariance for odd nonlinearities with bidirectional links and identifies exact convergence under sector-bounded nonlinearities and only practical convergence under uniform quantization. However, it asserts that the residual F is non-increasing for any sign-preserving nonlinearity without explicitly requiring oddness or a stepsize restriction; this is false in discrete time (a simple 2-node counterexample shows F can increase). The model’s solution pinpoints the missing oddness and stepsize conditions, provides a rigorous vector/incidence formulation, and derives a conservative but sound linear-rate bound under sector conditions, plus an O(δ) neighborhood result for uniform quantization, aligning with the paper’s goals while fixing its gaps. Key statements referenced: update laws and feasibility conditions (9–12, odd implies feasibility), the sign-preserving residual claim and sector-rate formula (13), and the qualitative uniform-quantizer conclusion.
Referee report (LaTeX)
\textbf{Recommendation:} major revisions \textbf{Journal Tier:} note/short/other \textbf{Justification:} The work addresses a relevant and practical class of nonlinearities in distributed optimization, combining actuation/communication effects with convex resource allocation. Yet several core statements—most notably discrete-time descent under sign-preserving nonlinearities—are stated without essential hypotheses (oddness, stepsize). Proofs are largely deferred and some constants are ambiguous. With clarified assumptions, corrected statements, and tighter presentation, the contribution could serve as a useful overview.