{"id":"8962e120-5262-4e97-8c0b-c1fc1156deaf","arxiv_id":"2608.07248","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"Mirror descent iterates converge to a Karush-Kuhn-Tucker point for a broad class of nonconvex problems with entropy and power kernels, via a metric-flattening reparameterization.","lead":"This paper proves that mirror descent converges to a valid solution, called a KKT point, even when the iterates approach the edge of the allowed region. It does so by reparameterizing the problem to remove the boundary blow-up and then applying a classical convergence argument.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Theorem 4.1's KKT conclusion is imported from [16, Prop 5.3], an external unpublished result whose hypotheses are not verified here; if it fails for diminishing stepsizes, the central claim is unsupported.","rationale":"I re-derived the key estimates in §4.2: the conformal circuit bound (Lemma 4.8), the reparameterized relative-error bound (Prop 4.11), and the KL finite-length argument (Lemma 4.12) are internally consistent. The coordinatewise examples in §5 satisfy the stated conditions, and the curvature bound κ<∞ is finite for all three kernel families on compact feasible sets. The only place where the proof hands off to an external result is the last step of Theorem 4.1: from convergence of x_k and 0∈∂E(z⋆) the paper concludes KKT via [16, Prop 5.3]. This is genuinely load-bearing because the reparameterized subdifferential inclusion at a boundary point does not automatically correspond to the original KKT system. The concern is not that the result is false—the cited preprint may well prove it—but that the present manuscript does not supply the verification, and the theorem's advertised flexibility (α_k→0) makes the missing hypothesis check nontrivial. Because this is a completeness/correctness-risk issue rather than a discovered counterexample, I keep the reader's CONDITIONAL verdict and recommend the specific verification above.","tokens_in":22150,"tokens_out":23236,"duration_ms":199467,"concrete_test":"Load arXiv:2507.15264v3, locate Proposition 5.3, and (i) verify that its proof is complete and self-contained; (ii) check whether its hypotheses are satisfied under Theorem 4.1's assumptions, in particular whether it allows stepsizes bounded below only by 0 with ∑α_k=∞ and no bound on ∥∇ϕ(x_k)∥. If the proposition requires α_k bounded away from zero or bounded dual gradients, then Theorem 4.1's diminishing-stepsize claim is not established, and the paper must either prove the missing statement or restrict the stepsize schedule.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The reparameterization argument in §4 establishes finite length of S(x_k), convergence to z⋆∈N, and 0∈∂E(z⋆) through Lemma 4.12. It does not by itself give KKT stationarity of x⋆=S^{-1}(z⋆): at a boundary point, ∂E(z⋆) can be strictly larger than the set of vectors D(S^{-1})(z⋆)^T(∇f(x⋆)+A^Tλ) for λ∈R^m, because the limiting subdifferential includes normal-cone contributions from the constraint A S^{-1}(z)=b. The final sentence of the proof of Theorem 4.1 simply invokes [16, Proposition 5.3], a result stated in the authors' earlier preprint arXiv:2507.15264v3 and not proved in this manuscript. The proposition is asserted to imply that any convergent mirror descent sequence with positive nonsummable stepsizes has a KKT limit. Theorem 4.1 explicitly permits α_k→0, and it is not checked here whether Proposition 5.3 covers such diminishing stepsizes or requires additional hypotheses (e.g., boundedness of {∇φ(x_k)}, or square-summability). If Proposition 5.3 does not apply, the KKT conclusion of Theorem 4.1 fails even though all the paper's new estimates are correct. The definable-boundary-extension assumption, by contrast, is a stated scope restriction, not a correctness risk.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper studies mirror descent for smooth nonconvex problems with linear constraints and separable Legendre kernels, aiming at convergence to boundary KKT points. It introduces a coordinatewise metric-flattening reparameterization S, assumes a definable boundary extension, and proves finite length of the reparameterized sequence S(x_k), convergence of x_k, and stationarity 0 ∈ ∂E(z*) for the reparameterized objective E. The main tool is a KL finite-length argument fed by a uniform, boundary-independent bound on the Lagrangian gradient obtained through a conformal circuit decomposition. The paper then concludes KKT stationarity of the original limit by invoking [16, Proposition 5.3], rather than by a self-contained proof. Applications are given for Shannon entropy, Fermi–Dirac entropy, and power kernels, together with coordinatewise convergence rates.","tokens_in":22426,"tokens_out":14396,"duration_ms":128610,"significance":"If the KKT step is made fully supported, this is a meaningful advance: it gives a broad positive result on boundary KKT convergence for nonconvex mirror descent, in contrast to the counterexample in [18], and the circuit-based uniform bound on the Lagrangian gradient (Lemma 4.8 and Proposition 4.11) is an elegant and apparently correct technical contribution. The metric-flattening perspective and the explicit coordinatewise rate examples are also useful. The main caveat is that the central KKT conclusion currently rests on an unpublished external proposition, so the significance is conditional on that step being verified or proved.","major_comments":[{"comment":"The KKT conclusion of Theorem 4.1 is not derived from the paper's own estimates. The proof establishes finite length of S(x_k), convergence to z* ∈ N, and 0 ∈ ∂E(z*), and then the final sentence invokes [16, Proposition 5.3]. This is load-bearing: equation (18) and the identity in (19) are only valid for z ∈ N∘, and at a boundary point the limiting subdifferential of E can contain extra normal directions coming from the box constraints on z. Therefore 0 ∈ ∂E(z*) does not by itself imply existence of λ* satisfying (2). The manuscript must either prove the needed statement from [16, Proposition 5.3] or give a direct argument for the KKT condition.","section":"Section 4.2.3, proof of Theorem 4.1"},{"comment":"The paper asserts without proof that [16, Proposition 5.3] implies every convergent mirror descent sequence with positive nonsummable stepsizes has a KKT limit. Since [16] is an unpublished preprint and the proposition is not stated here, its hypotheses are not checked. In particular, Theorem 4.1 explicitly allows α_k → 0, and the paper's own comparison with [10] emphasizes that its stepsize condition is more flexible than a lower-bound condition; whether [16, Proposition 5.3] covers the diminishing-stepsize case must be verified or proved.","section":"Section 1.1, statement on benign behavior under sequence convergence"},{"comment":"The proof of Theorem 4.3 also relies on [16, Proposition 3.1] for global existence and uniqueness of the mirror flow in X∘. This is another external result from the authors' earlier preprint that is not proved in this manuscript. If the flow theorem is meant to be a contribution of the paper, the relevant existence statement should be stated as a hypothesis or proved in an appendix.","section":"Appendix B, proof of Theorem 4.3"}],"minor_comments":[{"comment":"The definition of a definable boundary extension includes compactness of N, but the text immediately says compactness is not essential and may be replaced by boundedness of the generated sequence. This should be reconciled, for example by stating the compactness as a standing assumption in the main theorems rather than as part of the definition.","section":"Section 3, Definition 3.1"},{"comment":"Proposition 3.3(iii) states power kernels for p ∈ (1,2), while Section 5.3 uses p_i ∈ [1,2). The case p = 1 is already covered by the Shannon entropy example, but the ranges should be aligned or the p = 1 case explicitly referenced so that the reader is not left to infer the extension.","section":"Section 5.3 vs Proposition 3.3"},{"comment":"In the proof of Lemma 4.12, the phrase 'Since each limiting subdifferential is closed in finite dimensions' would be clearer as 'Since each ∂E(z_{k_j}) is closed in finite dimensions'; the current wording is slightly confusing. This is a presentation issue only.","section":"Section 4.2.3, Lemma 4.12"}],"recommendation":"major_revision","confidential_remarks":"The main substantive issue is the dependence of the central KKT claim on [16, Proposition 5.3], an unpublished result of the same authors. I recommend asking the authors to include a proof or a precise statement and verification of that proposition, or to replace the final step with a direct argument. The reparameterization and circuit-based estimates appear sound and are the paper's strongest original contributions. The comparison with [10] is fair and informative."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Here's the short version: the paper proves finite length of S(x_k) and KKT convergence for nonconvex mirror descent with separable Legendre kernels, under a curvature condition (kappa<infinity) and a definable boundary extension of the metric-flattening map. The genuinely new piece is the uniform control of ||nabla f(x_k)+A^T lambda_{k+1}||_infinity via conformal circuit decomposition. That argument is sound and is the heart of the paper. The KL-based finite-length proof in reparameterized variables is standard but clean, and the coordinatewise convergence rates for Shannon, Fermi-Dirac, and power kernels are a nice extra.\n\nThe soft spot is the final step of Theorem 4.1. The paper concludes that x* is a KKT point by citing [16, Proposition 5.3], a result from the authors' own earlier preprint that is not proved or even precisely stated in this manuscript. The reparameterized argument yields 0 in dE(z*). At a boundary point of N, the limiting subdifferential can be larger than the set of vectors D(S^{-1}(z*))^T(nabla f(x*)+A^T lambda), so the KKT conclusion does not follow from the new estimates alone. The cited proposition is doing load-bearing work, and the paper doesn't verify its hypotheses, especially for diminishing stepsizes, which are explicitly allowed. If that proposition is correct and applicable, the theorem holds; if not, the main claim is unsupported. This is fixable, but it is a genuine external dependency.\n\nThe novelty overlap with independent preprint [10] is disclosed and handled honestly—the two works share the same reparameterization map, but this paper's conformal circuit bound and allowance of diminishing stepsizes are distinct. The definable boundary-extension condition excludes the p=2 log-barrier kernel; that is a clearly stated scope limit, not a hidden flaw.\n\nOverall the mathematics I checked is correct; Lemma 4.8 and Proposition 4.11 are the real contributions and they hold up. The paper deserves a serious referee. The main revision ask should be: make the KKT step self-contained, either by proving the needed version of Proposition 5.3 or by adding a direct argument from 0 in dE(z*) to the KKT system. After that, I'd be comfortable with it.\n\nRecommendation: send it to peer review, with the expectation that the authors close the dependency gap.","headline":"Solid paper with a genuinely useful uniform Lagrangian-gradient bound, but the final KKT conclusion leans on an unproved earlier result of the authors that needs to be made self-contained.","tokens_in":22964,"tokens_out":3875,"would_cite":true,"duration_ms":31764,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["90C26","90C30","49J52"],"pacs":[],"model":"deepseek-v4-flash","headline":"This paper proves that nonconvex mirror descent with a Legendre kernel converges to a boundary KKT point whenever the metric-flattening reparameterization admits a definable continuous boundary extension.","keywords":["mirror descent","boundary KKT stationarity","Legendre kernel","metric-flattening reparameterization","Kurdyka–Łojasiewicz inequality","conformal circuit decomposition","definable functions","Bregman divergence"],"falsifier":"Find a definable $C^1$ objective, a compact linearly constrained set, and one of the three allowed kernels (Shannon entropy, Fermi–Dirac entropy, or a power kernel with $1\\le p<2$) such that mirror descent with $\\bar\\alpha L<1$ and $\\sum\\alpha_k=\\infty$ accumulates at a boundary point violating the KKT inequalities; that would refute Theorem 4.1. A simpler probe is to run the same experiment with the logarithmic-barrier kernel $h''(t)=t^{-2}$, which fails the boundary-extension condition; if non-KKT accumulation appears there, the excluded case is genuinely different rather than a technicality.","tokens_in":21918,"feed_emoji":"🎯","tokens_out":9170,"duration_ms":71606,"temperature":0.7,"pith_summary":"Mirror descent is a workhorse for optimization over simplices and polytopes, but when the solution lies on the boundary of the feasible set the Legendre kernel's gradient blows up and the usual stationarity argument breaks. This paper closes that gap for smooth nonconvex problems with linear constraints by changing variables: it reparameterizes each coordinate so that the Hessian metric becomes the identity, and proves that if the inverse of this map extends continuously and definably to the boundary, then the reparameterized iterates have finite length and converge. The original iterates then converge to a point satisfying the Karush–Kuhn–Tucker conditions. This matters because a recent counterexample showed mirror descent can otherwise accumulate at non-KKT boundary points, so the boundary extension condition is what separates guaranteed stationarity from pathological winding.","feed_headline":"Boundary KKT convergence proven for mirror descent","feed_subtitle":"Metric-flattening reparameterization gives finite-length iterates and restores stationarity at the boundary.","key_machinery":"The engine is the metric-flattening reparameterization $S_i(x_i)=\\int_{\\bar x_i}^{x_i}\\sqrt{h_i''(u)}\\,du$, chosen so that the Hessian metric of the separable Legendre kernel becomes the identity, $D(S^{-1})^T\\nabla^2\\phi(S^{-1})D(S^{-1})=I$, and therefore cannot degenerate at the boundary. Its inverse must extend continuously and definably to the closure $N=\\mathrm{cl}\\,S(X^\\circ)$; under that extension the reparameterized objective $E(z)=f(S^{-1}(z))$ is definable and satisfies the Kurdyka–Łojasiewicz property, so mirror descent becomes a Euclidean-type descent on $E$. The other essential piece is the conformal circuit decomposition of the mirror-step displacement, which yields a uniform bound on the Lagrangian gradient $\\|\\nabla f(x)+A^T\\lambda\\|_\\infty\\le\\Gamma$ independent of distance to the boundary; this uniform bound lets the standard KL finite-length argument run all the way to the boundary.","core_discovery":"On the paper's own terms, the central claim is Theorem 4.1: under a finite curvature bound on the kernel ($\\kappa<\\infty$), a definable boundary extension of the coordinatewise metric-flattening map $S$, and an objective that is $C^1$, relatively smooth, and definable, mirror descent with stepsizes $0<\\alpha_k\\le\\bar\\alpha$, $\\bar\\alpha L<1$, and $\\sum\\alpha_k=\\infty$ yields reparameterized iterates $z_k=S(x_k)$ with $\\sum_k\\|z_{k+1}-z_k\\|<\\infty$. Consequently $z_k\\to z^\\star$, the original iterates converge to $x^\\star=S^{-1}(z^\\star)$, and $x^\\star$ is a KKT point of the linearly constrained problem. The proof applies the Kurdyka–Łojasiewicz inequality to the reparameterized objective after establishing a boundary-uniform bound on the Lagrangian gradient via a conformal circuit decomposition.","pith_inferences":["The definable boundary-extension condition is essentially an integrability statement: $\\int\\sqrt{h_i''}$ must be finite so that $S_i$ reaches the boundary, which is why the logarithmic barrier (the $p=2$ power kernel, where $h''(t)=t^{-2}$) is excluded; one testable extension is to replace definability by tameness or semialgebraic regularity and see whether finite length survives.","The conformal-circuit bound suggests a transferable recipe for other Bregman-type methods (proximal point, proximal gradient, ADMM): stationarity at boundary limits can be certified through the elementary-vector structure of the constraint set rather than by taming the kernel's gradient blow-up.","A practical monitoring rule follows from the coordinatewise rates: if one tracks $S(x_k)$, residuals in boundary coordinates are expected to vanish faster than interior residuals, giving a principled stopping criterion.","Because the framework uses ordinary KL on the reparameterized objective, any definable objective with a known KL exponent yields explicit rates; it would be natural to test whether the rates are sharp by constructing objectives whose reparameterized KL exponent reaches the edge of the allowed range."],"forward_implications":["For Shannon entropy, Fermi–Dirac entropy, and power kernels on a compact polyhedron, Theorem 4.1 turns mirror descent with diminishing nonsummable stepsizes into a guaranteed boundary-KKT method even when the limit lies on a face.","The reparameterized sequence has finite length, so the original iterates form a Cauchy sequence and converge to a single point rather than merely clustering.","Coordinatewise rates follow from the KL exponent: coordinates whose limit is on the boundary converge faster (squared rate for Shannon and Fermi–Dirac, rate $q_i=2/(2-p_i)$ for power kernels) than interior coordinates.","The same reparameterization proves that mirror-flow trajectories have finite length in reparameterized variables and converge to a KKT point.","The stepsize condition $\\sum\\alpha_k=\\infty$ with $\\alpha_k\\le\\bar\\alpha<1/L$ permits stepsizes tending to zero, a flexibility not available in some related iterate-convergence frameworks."],"supporting_citations":[{"why":"Counterexample showing mirror descent can accumulate at non-KKT boundary points; this is the obstruction the paper's conditions must rule out.","marker":"[18]"},{"why":"Supplies Proposition 5.3, which converts sequence convergence of mirror descent into a KKT limit, and earlier results on spurious stationary points and mirror flow.","marker":"[16]"},{"why":"Provides the standard Kurdyka–Łojasiewicz finite-length argument that Lemma 4.12 adapts.","marker":"[3]"},{"why":"Gives the KL inequality for subanalytic and definable functions, used to obtain the KL property of the reparameterized objective.","marker":"[5]"},{"why":"Defines o-minimal definability and establishes that definable functions are stratifiable and KL, used throughout the extension condition.","marker":"[6]"},{"why":"Supplies the conformal circuit decomposition of vectors in ker A used to bound the Lagrangian gradient uniformly.","marker":"[33]"},{"why":"Provides relative smoothness and the descent lemma (Lemma 2.6) that gives sufficient decrease in the reparameterized variables.","marker":"[4]"},{"why":"Supplies the limiting subdifferential calculus used to identify the subdifferential of the reparameterized objective.","marker":"[35]"}],"fun_headline_variants":["Reparameterized mirror descent reaches boundary KKT","Finite-length iterates prove boundary KKT for mirror descent","Mirror descent boundary KKT via metric flattening","Definable reparameterization secures KKT at boundary"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The argument rests on the inverse of the metric-flattening map extending continuously and definably to the boundary of the reparameterized feasible set, a condition that excludes common kernels such as the logarithmic barrier; unless that extension exists the finite-length argument cannot start, and the KKT conclusion additionally inherits a cited proposition from the authors' earlier preprint that is not proved in this manuscript.","fun_headline_variants_meta":{"raw":{"variants":["Reparameterized mirror descent reaches boundary KKT","Finite-length iterates prove boundary KKT for mirror descent","Mirror descent boundary KKT via metric flattening","Definable reparameterization secures KKT at boundary"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000675,"raw_usage":{"total_tokens":3077,"prompt_tokens":953,"completion_tokens":2124,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":569,"completion_tokens_details":{"reasoning_tokens":2057}},"tokens_in":569,"tokens_out":2124,"duration_ms":14109,"temperature":1.0,"reasoning_tokens":2057,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-11T04:14:14.497096+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Find a definable $C^1$ objective, a compact linearly constrained set, and one of the three allowed kernels (Shannon entropy, Fermi–Dirac entropy, or a power kernel with $1\\le p<2$) such that mirror descent with $\\bar\\alpha L<1$ and $\\sum\\alpha_k=\\infty$ accumulates at a boundary point violating the KKT inequalities; that would refute Theorem 4.1. A simpler probe is to run the same experiment with the logarithmic-barrier kernel $h''(t)=t^{-2}$, which fails the boundary-extension condition; if non-KKT accumulation appears there, the excluded case is genuinely different rather than a technicality.","supporting_citations":[{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Provides the standard Kurdyka–Łojasiewicz finite-length argument that Lemma 4.12 adapts."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Gives the KL inequality for subanalytic and definable functions, used to obtain the KL property of the reparameterized objective."},{"cited_title":"Bauschke, J´ erˆ ome Bolte, and Marc Teboulle","cited_arxiv_id":null,"evidence_quote":"Provides relative smoothness and the descent lemma (Lemma 2.6) that gives sufficient decrease in the reparameterized variables."},{"cited_title":"Boyd, and Peter W","cited_arxiv_id":null,"evidence_quote":"Supplies the limiting subdifferential calculus used to identify the subdifferential of the reparameterized objective."}],"review_version":2}