{"id":"6438a9fd-2c61-4909-b47d-f08db4a43a8c","arxiv_id":"2507.15264","paper_version":3,"verdict":"CONDITIONAL","confidence":"HIGH","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"A Riemannian subgradient differential inclusion unifies Hessian barrier and mirror descent methods and explains their spurious stationary points as stable equilibria outside the true stationary set.","lead":"This paper builds a continuous-time flow for optimization over the intersection of a convex open set and a smooth manifold, and shows that two known algorithms, Hessian barrier and mirror descent, are discrete versions of the same flow. It explains a known failure mode, spurious stationary points, as stable equilibria of this flow that are not true optima, and gives conditions under which trajectories and stochastic iterates avoid them.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Condition (18), which transfers the flow's escape and avoidance results to constrained mirror descent, is unverified—only the unconstrained case is checked—so the paper's explanation of mirror descent's spurious points is not established for the algorithm that exhibits them.","rationale":"The reader's weakest assumption—that the discrete schemes track the continuous flow accurately enough—is the right locus, and I concur. The Hessian barrier method is not the weak side: for an affine manifold the identity retraction satisfies Assumption 5.1.6 with zero error, so Theorem 5.1 for that classical setting is essentially the continuous theory plus standard stochastic approximation. The weak side is the mirror descent scheme of §5.2, where the constrained update is labeled 'heuristic', verifying (18) is admitted to be 'generally difficult', and the only verified instance (Example 5.1) is the unconstrained entropy kernel with no Aᵀy term. Yet the spurious-stationarity phenomenon being explained (from [19]) lives precisely in the constrained case, where near ∂C the preconditioner degenerates and the equality constraint couples the escape direction through y. If (18) fails there, the interpolated mirror descent process is not a perturbed solution of (7), so the escape and avoidance theorems do not transfer and the claim that mirror descent's pathological convergence is a property of the continuous flow is unsupported at the algorithmic level. Two corroborating gaps strengthen rather than change the verdict: the nonlinear extension in §4.1 contains the unedited phrase 'has some issues', and Theorem 4.1(i)'s proof invokes a constant lower bound ẋ > δ even though the dynamics give ẋ = −H⁻¹s with H⁻¹ → 0 at the boundary; the correct proportional bound ẋ ≥ δx likely still yields escape via Gronwall, so this is repairable. Credit where due: Proposition 3.2 (equivalence of (7) and (8)) is a clean formal derivation; Proposition 3.6's identification of [19]'s fixed points with the stable set S is coherent; and a sanity check of the complementarity argument on the simplex example f(x) = −x₁ with x₁ + x₂ = 1 confirms the algebra of Theorem 4.1(ii) is internally consistent. The paper has no experiments and no code, but as a theory paper the issue is not the absence of numerics; it is the absence of a proof or feasible verification of the bridge in the regime where the central application lives. The proposed analytic and numerical checks on a [19] instance would settle whether the concern lands. Pending that, the CONDITIONAL verdict remains appropriate.","tokens_in":27439,"tokens_out":24731,"duration_ms":271479,"concrete_test":"In the canonical constrained setting of [19] (C = {x > 0}, affine M = {Ax = b}, entropy kernel), run two checks. (1) Analytic: for the two-dimensional simplex M = {x₁ + x₂ = 1}, compute y(x,d,η) and x₊(x,d,η) in closed form, expand in η, and evaluate the limit in condition (18); this either verifies or refutes (18) in the archetypal constrained case. (2) Numerical: on a concrete [19] instance with a known spurious point x̄ ∈ S∖Ω, integrate the flow (7) and run the constrained mirror descent (17) from the same x₀. If the flow escapes x̄ while the iterates converge to it, condition (18) fails and the discrete transfer is invalid; if the flow also converges to x̄, the escape theorems are contradicted; if both escape, the bridge is restored in that instance.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The paper's headline application is the constrained mirror descent spurious-point phenomenon of [19] (§3.1.1, Proposition 3.6). All continuous-flow conclusions (Proposition 3.7, Theorems 4.1 and 4.2) transfer to the discrete mirror descent scheme only through condition (18) in §5.2, which requires the implicit Bregman step to differ from the explicit Euler step x − η∇²φ(x)^{-1}(d − Aᵀy(x,d,η)) by o(η‖∇²φ(x)^{-1}g‖), uniformly in x and d. Section 5.2 explicitly proposes a 'heuristic Riemannian mirror descent scheme', and the authors state that verifying (18) 'is generally difficult'; Example 5.1 verifies it only in the unconstrained case M = Rⁿ, where the Aᵀy term is absent. The constrained case is exactly the regime of the phenomenon being explained: near the boundary of C the preconditioner degenerates (for entropy, ∇²φ^{-1} = diag(x)), and the dual variable y couples the escape direction to the equality constraint, so condition (18) is precisely where boundary behavior could prevent the interpolated process from being a perturbed solution. Without (18), Benaim's stochastic approximation theorem does not apply, and none of the escape, avoidance, or stationarity conclusions are known to hold for the mirror descent iterates, so the claimed explanation of spurious stationarity rests on an unverified premise. Corroborating gaps, second-order relative to this missing bridge: §4.1's nonlinear extension literally contains 'has some issues', and Theorem 4.1(i)'s proof assumes a constant lower bound ẋ > δ although the flow gives ẋ = −H⁻¹s with H⁻¹ → 0 at the boundary (only ẋ ≥ δx holds; the conclusion is likely repairable via Gronwall).","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper studies a nonsmooth, nonconvex optimization problem over the intersection of the closure of an open convex set C and a smooth manifold M. It introduces the differential inclusion (7), ẋ(t) ∈ -P_{T_{x(t)}M} ∇²φ(x(t))^{-1} ∂f(x(t)), and interprets this flow as a unified continuous-time model for both the Hessian-barrier method and mirror descent. The central claim is that the 'spurious stationary points' observed in both algorithms are exactly the points in the stable set S = {x : 0 ∈ P_{T_xM}∇²φ(x)^{-1}∂f(x)} that are not true stationary points Ω = {x : 0 ∈ ∂f(x) + N_M(x) + N_C(x)}. The paper develops finite-time escape results near S \\ Ω, provides two conditions (complementarity and isolation) under which trajectories avoid S \\ Ω, proposes a random perturbation strategy, and derives two discrete interior-point algorithms whose convergence is analyzed through stochastic approximation. The authors are explicit that several steps are heuristic or incomplete, including the verification of condition (18) for constrained mirror descent and the nonlinear extension of Theorem 4.1.","tokens_in":27711,"tokens_out":8742,"duration_ms":92964,"significance":"If the main claims were fully established, the paper would supply a valuable unifying explanation for a known pathological behavior: the spurious-stationary-point phenomenon of mirror descent and Hessian-barrier methods would be traced to the stable set of a single continuous flow. The conceptual link between the extended mirror-descent mapping of [19] and the stable set S is attractive, and the paper is commendably honest about its limitations. However, the load-bearing bridge from the continuous flow to constrained mirror descent is verified only in the unconstrained case, and several central proofs—most notably Theorem 4.1(i), Proposition 3.4, and the nonlinear extension in §4.1—contain gaps or are self-flagged as problematic. The significance is therefore conditional: the paper provides a promising framework and several correct-looking continuous-flow arguments, but not yet a fully supported explanation of the constrained mirror-descent phenomenon that motivates the title and abstract.","major_comments":[{"comment":"The paper's explanation of the spurious-stationary-point phenomenon in constrained mirror descent rests entirely on condition (18), which would make the interpolated mirror-descent process a perturbed solution of the differential inclusion (7). The authors state that verifying (18) 'is generally difficult' and verify it only in the unconstrained case M = Rⁿ, where the Aᵀy term is absent. The constrained case is exactly the regime of the phenomenon being explained: near the boundary of C the preconditioner ∇²φ(x)^{-1} degenerates for typical barriers such as entropy, and the dual variable y couples the escape direction to the equality constraint. Without (18), the stochastic-approximation theorem of Benaim et al. is not applicable, and none of the escape, avoidance, or stationarity conclusions of Sections 3 and 4 are known to hold for the mirror-descent iterates. The claimed unification with the constrained mirror-descent spurious-point results of [19] is therefore unsupported as written.","section":"§5.2, Eq. (18), Example 5.1"},{"comment":"The proof of Theorem 4.1(i) asserts that when s(x) + x < 0 at a spurious point, there is a neighborhood in which ẋ > δ > 0. For the entropy kernel, H(x)^{-1} = diag(x) and the flow components satisfy ẋ_i = -x_i s_i(x), so as x_i → 0 the velocity tends to 0 even when s_i(x) ≤ -δ. The uniform linear lower bound is false. The proof's case analysis also does not exclude the possibility that the trajectory exits a neighborhood and later re-enters it; the text says 'we can argue in case 1' but case 1 does not establish that re-entry is impossible. The stronger 'never subsequentially converge' claim therefore does not follow from the given argument. A Gronwall-type log-coordinate argument as in Proposition 3.7 may repair the claim, but as written the proof is invalid.","section":"§4.1, Theorem 4.1(i)"},{"comment":"The paragraph extending Theorem 4.1 to nonlinear constraints C = {x : g(x) ≤ 0} contains the sentence 'the complementarity condition s(x) + x ≥ 0 ... has some issues', followed by an assertion that the repelling mechanism works 'analogous to the linear case' without a proof. Since the theorem statement explicitly claims the nonlinear extension, the text itself flags a missing proof. This is not a minor presentation issue: the nonlinear constraint setting is one of the stated contributions of the paper. The extension must either be proved in full or the theorem must be restricted to the linear case and the abstract and introduction adjusted accordingly.","section":"§4.1, nonlinear extension"},{"comment":"The proof of the inclusion NC(x) ⊂ Null((∇²φ(x))^{-1}) is not valid as written. The proof defines a set S of limits ∇²φ(x_k)e_k → d with e_k → 0 and shows that v ∈ span(∂∞φ(x)) can be represented as such a limit, but it never derives that H(x)^{-1}v = 0; the set S is not identified with the nullspace of H(x)^{-1}. In addition, the equality span(∂∞φ(x)) = NC(x) is asserted rather than proved for a general open convex C. Proposition 3.4 underlies the stable-set characterization used in Proposition 3.5, Proposition 3.6, and Theorem 4.1, so this gap is load-bearing rather than cosmetic.","section":"§3.1, Proposition 3.4"},{"comment":"The random-perturbation argument is not fully justified. The proof of Lemma 4.1 applies Morse-Sard theory to the map F(x, λ), but the required differentiability is not established: φ is only assumed C², so ∇²φ(x)^{-1}∇f(x) need not be C¹, and C¹ Morse-Sard is false in general for maps Rᵈ → Rᵈ. Even if local uniqueness of solutions follows from the implicit function theorem, that only shows the stable set S(u,v) is discrete; it does not by itself rule out accumulation of isolated points within a bounded region, and no boundedness of S(u,v) is shown. Since the perturbation strategy is the paper's answer for the generic case, this argument needs a careful repair or a precise set of additional hypotheses.","section":"§4.2, Lemma 4.1 and Theorem 4.3"}],"minor_comments":[{"comment":"There are numerous typos and grammatical errors, including 'unifily', 'stationsrt point', 'supercoecive', 'Gronwalls inequality', and 'separate theorem' for 'separation theorem'. The abstract also says 'strict complementarity conditions' while the actual sufficient condition in Theorem 4.1 is the non-strict inequality s(x) + x ≥ 0 on the ω-limit set.","section":"Abstract and throughout"},{"comment":"The proof of Proposition 2.1 contains a sign error: the KKT condition '0 = H(z)^{-1}(d(z) + Aᵀy)' should presumably be '0 = H(z)^{-1}(d(z) - Aᵀy)', and the expansion of the projection term omits a term. This is fixable but should be corrected for clarity.","section":"§2.3, Proposition 2.1 proof"},{"comment":"The equivalence between (7) and (8) is central to the mirror-descent interpretation, but its proof is omitted with 'similar to the proof of Proposition 2.1'. Given that Proposition 2.1 itself has sign issues, the authors should provide a complete proof of Proposition 3.3 or a precise reference.","section":"§3.1, Proposition 3.3"},{"comment":"Equation (16) has a typo: the displayed chain ends with 'xk = −ηk(vk + ξ̃k)' but should be 'xk+1 = xk − ηk(vk + ξ̃k)'. This makes the algebra in that paragraph confusing.","section":"§5.1, Eq. (16)"},{"comment":"The proposed 'heuristic Riemannian mirror descent scheme' (R-Breg) is presented with only a brief discussion and is not analyzed beyond the unverified condition (18). If the paper keeps this algorithm, it should either be clearly labeled as a proposal open for future work or provided with some form of justification; otherwise the reader may expect a convergence theorem that is not supplied.","section":"§5.2, heuristic algorithm"}],"recommendation":"major_revision","confidential_remarks":"The paper is honest about its gaps, and the continuous-flow part has several interesting ideas, including the stable-set characterization and the repelling phenomenon. However, the headline application to constrained mirror descent is not established because condition (18) is only verified in the unconstrained case, and the nonlinear extension of the main avoidance theorem is explicitly flagged by the authors as having issues. With a substantial revision that closes the discrete-to-continuous bridge or substantially narrows the claims, the paper could be suitable for publication; in its current form the gap between the advertised results and the proven statements is too large for acceptance. I also note that the paper relies on the authors' own prior work [25] for the unconstrained unification; the incremental novelty over [25] and [19] should be made explicit."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Read it. The continuous-flow core is genuinely useful and, as far as I can tell, mostly correct. Writing the constrained update as ẋ ∈ -P_{T_xM}∇²φ(x)^{-1}∂f(x) and identifying the spurious points of Hessian-barrier and mirror descent with S\\Ω is a real unification, not just notation. Proposition 3.6, connecting the fixed points of [19]'s extended mirror map to the stable set, is the cleanest part of the paper. The finite-time escape argument (Proposition 3.7) and the isolation/complementarity avoidance results in Section 4 are meaningful, and the random perturbation scheme via Sard is a reasonable way to get a generically isolated stable set. This is more than a repackaging of [1] and [19].\n\nThe soft spots are real, and the paper basically confesses to the two largest. Section 4.1's nonlinear extension literally contains 'has some issues' at the point where complementarity is supposed to imply s(x) ∈ -N_C(x); as written, that part is a sketch. Theorem 4.1(i) uses a uniform lower bound ẋ > δ to prove repulsion, but near the boundary the flow gives ẋ_i = -x_i s_i, so the bound is only proportional to x_i. I think the conclusion is repairable (exponential escape via Gronwall), but the proof as written is not correct.\n\nThe bigger problem is the bridge to mirror descent. All the discrete mirror-descent conclusions flow through condition (18), and the authors verify it only in the unconstrained entropy case. That is exactly the case without the A^T y dual coupling and without boundary degeneracy of ∇²φ^{-1} = diag(x). The constrained case is where the spurious-point phenomenon lives. So the paper convincingly explains the continuous trajectory, and it proposes a heuristic Riemannian mirror descent, but it does not establish that the constrained mirror descent iterates of [19] track this flow. The Hessian-barrier side is on firmer ground, because Assumption 5.1.6 is at least given content by the self-concordant barrier proposition that keeps iterates feasible.\n\nThe perturbation results (Lemma 4.1, Theorem 4.3) are fine, and the lack of experiments is not a flaw for a theory paper. But the paper's own central claim about explaining spurious points of mirror descent is conditional on an unverified premise. That needs to be either proven for a nontrivial constrained kernel or explicitly downgraded to a conjecture/heuristic.\n\nFor whom: people working on interior-point dynamics, mirror descent pathology, and stochastic approximation. It deserves a serious referee, but not acceptance as-is. Send it to review with the message that condition (18) is the make-or-break point: verify it in a constrained case or scale back the discrete mirror-descent claims.","headline":"The continuous-flow unification and stable-set reading of spurious points are solid and useful, but the paper's own headline application to constrained mirror descent rests on an unverified condition, so the discrete claims are not yet established.","tokens_in":28346,"tokens_out":3188,"would_cite":true,"duration_ms":35310,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["90C26","49J52","65K10","34A60","37N40"],"pacs":[],"model":"deepseek-v4-flash","headline":"One continuous flow explains why mirror descent and Hessian barrier methods can stop at spurious stationary points, and provides conditions that rule those stops out.","keywords":["mirror descent","Hessian barrier","Riemannian subgradient flow","differential inclusion","spurious stationary points","stochastic approximation","nonsmooth nonconvex optimization","interior point methods"],"falsifier":"Run the flow (7) for a bounded $C^2$ function $f$ with affine constraints $Ax = b$ and the entropy kernel, starting near a spurious point $x^* \\in S \\setminus \\Omega$ that satisfies $s(x^*) + x^* \\ge 0$; if the trajectory accumulates at $x^*$, or a mirror descent run with step size $\\eta_k = 1/k$ accumulates there, then Theorem 4.1(ii) and Proposition 3.7 would be contradicted. A concrete experimental signature would be $x_i(t)$ failing to grow exponentially while the trajectory remains indefinitely in a neighborhood of $x^*$, contrary to the Gronwall-based escape bound.","tokens_in":2282,"feed_emoji":"🎯","tokens_out":9938,"duration_ms":144895,"temperature":0.7,"pith_summary":"This paper studies a nonsmooth nonconvex optimization problem whose feasible set is the intersection of the closure of an open convex set $C$ with a smooth manifold $M$. It introduces a Riemannian subgradient flow, $\\dot{x}(t) \\in -P_{T_{x(t)}M}\\nabla^2\\varphi(x(t))^{-1}\\partial f(x(t))$, where the metric comes from the Hessian of a barrier function $\\varphi$ for $C$. The paper claims that both the Hessian barrier method and the mirror descent scheme are discrete approximations of this single flow, so their well-known deficient convergence behavior is a property of the continuous dynamics, not an artifact of discretization. The spurious stationary points reported for both methods are exactly the stable equilibria $S$ of the flow that are not true stationary points $\\Omega$ of the original problem. The paper proves that under a complementarity condition or an isolation condition, trajectories avoid $S \\setminus \\Omega$ and converge to $\\Omega$; without such conditions, a random perturbation of the objective and constraints makes the stable set isolated, yielding approximate stationarity of the original problem.","feed_headline":"One flow exposes why mirror descent hits false stationary points","feed_subtitle":"Both methods trace the same interior-point flow; regularity conditions push trajectories past its non-stationary equilibria.","key_machinery":"The central object is the interior Riemannian subgradient flow (7), with $\\nabla^2\\varphi$ inducing a Riemannian metric on $C$ and $P_{T_xM}$ the metric projection onto the tangent space of $M$. Its equivalent dual form (8), $\\frac{d}{dt}\\nabla\\varphi(x(t)) \\in -\\partial f(x(t)) + N_{x(t)}M$, is what makes the mirror descent connection explicit: discretizing (8) with a Bregman step gives the mirror descent update, while discretizing (7) with a retraction gives the Hessian barrier method. The load-bearing distinction is between the flow's stable set $S$ and the optimization problem's stationary set $\\Omega$; the argument uses $f$ itself as a Lyapunov function, a separating-hyperplane construction on the active constraints to produce a repelling direction at points of $S \\setminus \\Omega$, and stochastic approximation theory to transfer trajectory statements to discrete iterates.","core_discovery":"On its own terms, the paper establishes that the differential inclusion (7), equivalently written as $\\frac{d}{dt}\\nabla\\varphi(x(t)) \\in -\\partial f(x(t)) + N_{x(t)}M$, is the continuous object underlying both the Hessian barrier method and mirror descent: the former is a retraction-based discretization of the projected Riemannian subgradient, and the latter is a Bregman discretization of the dual form of the same flow. The stable set $S = \\{x : 0 \\in P_{T_xM}\\nabla^2\\varphi(x)^{-1}\\partial f(x)\\}$ can be strictly larger than the true stationary set $\\Omega = \\{x : 0 \\in \\partial f(x) + N_M(x) + N_C(x)\\}$; the difference $S \\setminus \\Omega$ is precisely where the spurious stationary points of mirror descent live, and the paper identifies them with fixed points of the extended Bregman update map. It shows that every trajectory of the flow exits a neighborhood of any point in $S \\setminus \\Omega$ in finite time, so a convergent trajectory must land in $\\Omega$. If the complementarity condition $s(x) + x \\ge 0$ holds on the $\\omega$-limit set, or if $S \\cap \\partial C$ has only isolated points, then the escape dynamics preclude subsequential convergence to spurious points. Absent such regularity, generically perturbing the objective to $f + \\langle \\nabla\\varphi, v\\rangle$ and the constraint to $c + u$ makes the stable set have no cluster points by a Morse--Sard argument, so the flow and its discrete approximations converge to stationary points of the perturbed problem, which are nearly stationary for the original problem.","pith_inferences":["This stable-set criterion suggests a screening test for any new Bregman or barrier scheme: compute $S \\setminus \\Omega$ for the kernel and constraint geometry; if it is nonempty, spurious limits or slow escape should be expected unless a retraction condition like Assumption 5.1.6 or condition (18) holds.","The escape mechanism is quantitative: the Gronwall argument makes a coordinate grow exponentially while the trajectory remains near a spurious point, so Proposition 3.7 could be turned into a numerical diagnostic that measures exponential repulsion to certify whether a near-stationary iterate is spurious.","If the paper's transfer is correct, the honest failure mode of mirror descent is geometric rather than due to step sizes: any kernel whose Hessian inverse degenerates at the boundary enlarges the stable set, and comparing kernels by the size of $S \\setminus \\Omega$ could become a principled kernel-design criterion.","The perturbation strategy is effectively a random smoothing of the manifold, and for structured cones such as the positive semidefinite cone it preserves membership of the slack variable in the dual cone, hinting at practical regularized interior-point solvers for conic programs."],"forward_implications":["Any trajectory of (7) that converges must converge to a true stationary point in $\\Omega$, because spurious points in $S \\setminus \\Omega$ are escaped in finite time.","When the complementarity condition $s(x) + x \\ge 0$ holds on the $\\omega$-limit set, or when $S \\cap \\partial C$ consists of isolated points, the flow and the associated discrete interior-point methods subsequentially converge to $\\Omega$ instead of spurious points.","Without those conditions, perturbing $f$ to $f + \\langle \\nabla\\varphi, v\\rangle$ and $c$ to $c + u$ makes the stable set isolated for almost all small $(u,v)$, so the trajectory reaches a stationary point of the perturbed problem, which is approximately stationary for the original one.","The proposed discretizations are interior point methods whose iterates remain in $M \\cap C$, and their convergence inherits the continuous flow's conclusions whenever the retraction error scaled by step size vanishes uniformly.","The spurious stationary points of mirror descent correspond exactly to fixed points of the extended Bregman update map, so the paper transfers that phenomenon from a property of one algorithm to a property of the unifying flow."],"supporting_citations":[{"why":"Supplies the original Hessian Riemannian gradient flow for linearly constrained smooth convex problems, which the paper extends to nonsmooth objectives and general manifolds.","marker":"[1]"},{"why":"Provides the stochastic approximation and differential inclusion framework that connects discrete iterates to trajectories of the flow.","marker":"[5]"},{"why":"Provides a Sard-type result for stratifiable functions used to show that random perturbation makes the stable set isolated.","marker":"[9]"},{"why":"Introduces the Hessian barrier algorithm, one of the two discrete methods that the paper identifies as approximations of the same flow.","marker":"[13]"},{"why":"Documents spurious stationary points of mirror descent; the paper proves these coincide with the stable-set points outside the true stationary set.","marker":"[19]"},{"why":"Supplies the path-differentiable stochastic subgradient convergence theory and Lyapunov arguments that the paper adapts to its setting.","marker":"[23]"},{"why":"Links mirror descent flow to gradient flow on a reparameterized model, which the paper uses to interpret the observed non-implication of stationarity as spurious.","marker":"[46]"},{"why":"Provides the Dikin ellipsoid containment property for self-concordant barriers, used to keep iterates strictly inside the feasible set.","marker":"[50]"}],"fun_headline_variants":["One flow unifies Hessian barrier and mirror descent","Spurious stationary points traced to stable equilibria of the flow","How mirror descent's false stops arise from a single interior flow","Perturbation rescues mirror descent from spurious stationary points","Escaping spurious points: the flow's stable equilibria and how to avoid them"],"cache_read_input_tokens":30208,"weakest_assumption_plain":"The load-bearing premise is that each discrete method tracks the continuous flow accurately: the retraction error in Assumption 5.1.6, and the analogous condition (18) for mirror descent, must vanish uniformly as the step size goes to zero, and the paper itself notes that no retraction can satisfy this in a one-dimensional interval example and that verifying (18) beyond an entropy kernel is difficult.","fun_headline_variants_meta":{"raw":{"variants":["One flow unifies Hessian barrier and mirror descent","Spurious stationary points traced to stable equilibria of the flow","How mirror descent's false stops arise from a single interior flow","Perturbation rescues mirror descent from spurious stationary points","Escaping spurious points: the flow's stable equilibria and how to avoid them"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000318,"raw_usage":{"total_tokens":1923,"prompt_tokens":1199,"completion_tokens":724,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":815,"completion_tokens_details":{"reasoning_tokens":634}},"tokens_in":815,"tokens_out":724,"duration_ms":7752,"temperature":1.0,"reasoning_tokens":634,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-06T15:38:06.583440+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Run the flow (7) for a bounded $C^2$ function $f$ with affine constraints $Ax = b$ and the entropy kernel, starting near a spurious point $x^* \\in S \\setminus \\Omega$ that satisfies $s(x^*) + x^* \\ge 0$; if the trajectory accumulates at $x^*$, or a mirror descent run with step size $\\eta_k = 1/k$ accumulates there, then Theorem 4.1(ii) and Proposition 3.7 would be contradicted. A concrete experimental signature would be $x_i(t)$ failing to grow exponentially while the trajectory remains indefinitely in a neighborhood of $x^*$, contrary to the Gronwall-based escape bound.","supporting_citations":[{"cited_title":"Alvarez, J","cited_arxiv_id":null,"evidence_quote":"Supplies the original Hessian Riemannian gradient flow for linearly constrained smooth convex problems, which the paper extends to nonsmooth objectives and general manifolds."},{"cited_title":"Bena ¨ ım, J","cited_arxiv_id":null,"evidence_quote":"Provides the stochastic approximation and differential inclusion framework that connects discrete iterates to trajectories of the flow."},{"cited_title":"Bolte, A","cited_arxiv_id":null,"evidence_quote":"Provides a Sard-type result for stratifiable functions used to show that random perturbation makes the stable set isolated."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Introduces the Hessian barrier algorithm, one of the two discrete methods that the paper identifies as approximations of the same flow."},{"cited_title":"Davis, D","cited_arxiv_id":null,"evidence_quote":"Supplies the path-differentiable stochastic subgradient convergence theory and Lyapunov arguments that the paper adapts to its setting."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Links mirror descent flow to gradient flow on a reparameterized model, which the paper uses to interpret the observed non-implication of stationarity as spurious."},{"cited_title":"Nesterov et al","cited_arxiv_id":null,"evidence_quote":"Provides the Dikin ellipsoid containment property for self-concordant barriers, used to keep iterates strictly inside the feasible set."}],"review_version":1}