arxiv: 2605.08491 · v1 · submitted 2026-05-08 · 🧮 math.OC · math.CA

Recognition: 2 theorem links

· Lean Theorem

Local Nonconvexity Indices for \(C^{1,1}\) Functions via Generalized Hessians

Marina Palaisti

Pith reviewed 2026-05-12 00:48 UTC · model grok-4.3

classification 🧮 math.OC math.CA

keywords local nonconvexity indexgeneralized HessianClarke subdifferentialC^{1,1} functionsspectral functionalnuclear normweak convexitysecond-order conditions

0 comments

The pith

Local nonconvexity indices for C^{1,1} functions arise by applying a spectral functional to their Clarke generalized Hessian sets.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper extends an existing index for the local lack of convexity, originally defined for twice continuously differentiable functions via the nuclear-norm distance of the Hessian from the positive semidefinite cone, to the broader class of C^{1,1} functions. It replaces the single Hessian matrix with the Clarke generalized Hessian set, which is the closed convex hull of all possible limiting Hessians obtained from nearby points of twice differentiability. This replacement produces an interval of index values rather than a single number, with the lower and upper endpoints indicating the smallest and largest amounts of second-order nonconvexity that can be detected at the given point. A sympathetic reader would care because this provides a practical scalar diagnostic for nonsmooth functions that appear in optimization and analysis, while preserving desirable properties such as vanishing on convex functions and agreeing with the smooth case.

Core claim

The central claim is that for a C^{1,1} function h at a point x, the local nonconvexity index is the set of values obtained by evaluating the spectral functional (nuclear-norm distance to the positive semidefinite cone) over the generalized Hessian set Hess(h;x). This yields an interval whose endpoints represent the least and greatest visible second-order nonconvexity at x. The construction reduces to the classical index when h is C^2, is zero for convex h, is orthogonally invariant, satisfies a subadditivity property for the upper endpoint under addition of functions, and is upper semicontinuous in the upper endpoint. It is also related to a pointwise weak-convexity curvature modulus.

What carries the argument

The Clarke-type generalized Hessian set Hess(h;x), the closed convex hull of limiting Hessians at sequences of twice-differentiable points approaching x, which replaces the classical Hessian and allows the spectral functional to be applied to a set, producing an interval index.

If this is right

The index reduces exactly to the original smooth-case index when the function is twice continuously differentiable.
It equals the zero interval when the function is convex.
The upper endpoint of the index is subadditive with respect to sums of functions.
The upper endpoint is upper semicontinuous as a function of the point x.
The upper endpoint is related to the curvature modulus of weak convexity at the point.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

This scalar interval could be approximated numerically by sampling limiting Hessians in computational optimization routines for C^{1,1} problems.
The approach suggests similar extensions might be possible for other nonsmooth classes using different generalized derivatives.
Applications in variational analysis could use these indices to locally adjust penalty terms or regularization based on detected nonconvexity ranges.

Load-bearing premise

The Clarke generalized Hessian set encodes the second-order nonconvexity information for C^{1,1} functions in essentially the same way the classical Hessian does for C^2 functions.

What would settle it

A C^{1,1} function that is convex at a point x but for which the upper endpoint of the index is strictly positive would falsify the vanishing property.

read the original abstract

Davydov, Moldavskaya, and Zitikis introduced local indices for quantifying the lack of convexity of a \(C^2\) function by measuring the nuclear-norm distance of its Hessian from the cone of positive semidefinite matrices. This paper develops a local analogue for functions of class \(C^{1,1}\). At a point \(x\), the classical Hessian is replaced by the Clarke-type generalized Hessian set \(\Hess(h;x)\), defined as the closed convex hull of limiting Hessians at nearby twice differentiability points. Evaluating the same spectral functional over \(\Hess(h;x)\) gives an interval-valued local nonconvexity index whose lower and upper endpoints represent, respectively, the least and greatest visible second-order nonconvexity at \(x\). The construction reduces to the original smooth index when \(h\in C^2\), vanishes for convex \(C^{1,1}\) functions, is invariant under orthogonal changes of variables, satisfies a subadditivity inequality for the upper endpoint under sums, and is upper semicontinuous in its upper endpoint. We also relate the upper endpoint to a pointwise weak-convexity curvature modulus and give explicit \(C^{1,1}\setminus C^2\) examples. The paper is deliberately local in scope: it proposes a scalar diagnostic extracted from generalized Hessian sets, not a replacement for the richer second-order variational theory of nonsmooth convexity.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

This paper cleanly extends the Davydov-Moldavskaya-Zitikis index to C^{1,1} functions by evaluating the nuclear-norm distance over the Clarke generalized Hessian set, and the extension looks internally consistent.

read the letter

The main takeaway is that the authors replace the classical Hessian with the Clarke generalized Hessian set (closed convex hull of limiting Hessians) and apply the same spectral functional to get an interval-valued local nonconvexity index. The lower and upper endpoints capture the range of second-order nonconvexity visible at a point for C^{1,1} functions. This is a direct technical move from the C^2 case and not just a rephrasing of prior work. They show it reduces correctly when the function is twice differentiable, vanishes on convex C^{1,1} functions, is orthogonally invariant, satisfies subadditivity on the upper endpoint for sums, and is upper semicontinuous in that endpoint. They also link the upper endpoint to a pointwise weak-convexity modulus and supply explicit C^{1,1} minus C^2 examples, which is helpful for seeing the behavior when Hessians fail to exist everywhere. The paper stays local and does not claim to supplant full second-order variational theory, which keeps the scope honest. The soft spots are modest. An interval rather than a scalar can feel less convenient for some algorithmic uses, though that follows from the set-valued input. Without the full derivations in front of me I cannot double-check every inequality, but the stress-test found no contradictions and the listed properties line up with the construction. No circularity or invented entities appear. This is for readers already working with generalized Hessians in nonsmooth optimization or local convexity diagnostics. Someone building tools or conditions in that niche will get a concrete scalar diagnostic to test or extend. It deserves a serious referee because the extension is well-defined, the properties are stated precisely, and the examples make the difference from the smooth case visible. I would send it to peer review.

Referee Report

2 major / 3 minor

Summary. The paper extends the Davydov-Moldavskaya-Zitikis local nonconvexity index, originally for C² functions via nuclear-norm distance of the Hessian to the PSD cone, to C^{1,1} functions. It replaces the Hessian with the Clarke generalized Hessian set Hess(h;x) (closed convex hull of limiting Hessians at nearby twice-differentiable points) and evaluates the same spectral functional over this set to produce an interval-valued index. The lower and upper endpoints quantify the least and greatest visible second-order nonconvexity at x. The authors claim reduction to the C² case, vanishing on convex C^{1,1} functions, orthogonal invariance, subadditivity of the upper endpoint under sums, upper semicontinuity of the upper endpoint, and a relation to a pointwise weak-convexity modulus, supported by explicit C^{1,1}∖C² examples. The scope is deliberately local and diagnostic.

Significance. If the properties hold, the construction supplies a concrete, interval-valued local diagnostic for second-order nonconvexity in the C^{1,1} setting that is consistent with the smooth case and respects basic convexity and invariance axioms. This could be useful in optimization and variational analysis as a scalar probe extracted from generalized Hessians, complementing richer nonsmooth second-order theory. The explicit examples and the link to the weak-convexity modulus are positive features that illustrate applicability beyond C².

major comments (2)

[Definition of the index (near the generalized Hessian set)] The definition of the index as the image of the nuclear-norm distance functional over the set Hess(h;x) (presumably the closed interval between min and max values) is load-bearing for the interval-valued claim and all subsequent properties. Please add an explicit statement (with reference to a proposition or lemma) confirming that the functional is continuous on the compact convex set Hess(h;x), so that the image is indeed a closed bounded interval.
[Subadditivity section] The subadditivity claim for the upper endpoint under sums of two C^{1,1} functions is a key property. The argument must be checked for whether it uses only the convex-hull definition or requires additional structure on the limiting Hessians; if the proof invokes a specific inequality for the nuclear norm on convex hulls, state the precise step and any equality cases.

minor comments (3)

[Abstract and introduction] In the abstract and introduction, the phrase 'evaluating the same spectral functional over Hess(h;x)' would benefit from one sentence clarifying whether the functional is applied pointwise to each matrix in the set and then the range is taken.
[Examples section] The examples in C^{1,1}∖C² are valuable; adding a short table or list summarizing the dimension, the form of the function, and the computed index interval for each would improve accessibility.
[Throughout] Notation for the generalized Hessian set is consistent in the abstract but should be checked for uniform use (Hess(h;x) vs. other variants) in all proofs and statements.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the positive evaluation and the precise comments on the definition of the index and the subadditivity property. We address each major comment below, indicating the revisions that will be incorporated in the next version of the manuscript.

read point-by-point responses

Referee: [Definition of the index (near the generalized Hessian set)] The definition of the index as the image of the nuclear-norm distance functional over the set Hess(h;x) (presumably the closed interval between min and max values) is load-bearing for the interval-valued claim and all subsequent properties. Please add an explicit statement (with reference to a proposition or lemma) confirming that the functional is continuous on the compact convex set Hess(h;x), so that the image is indeed a closed bounded interval.

Authors: We agree that an explicit confirmation is required. The functional is the nuclear-norm distance to the positive semidefinite cone, which is continuous because the nuclear norm is a continuous matrix norm and the distance function to the closed convex PSD cone is continuous. The set Hess(h;x) is compact (closed and bounded) by the standard properties of the Clarke generalized Hessian for C^{1,1} functions. The continuous image of a compact set is therefore compact, hence a closed bounded interval. We will add a short lemma (new Lemma 2.4) immediately after the definition of the index, stating this fact and referencing the continuity of the distance function together with the compactness of Hess(h;x) from Proposition 2.2. This lemma will be cited in all subsequent sections that rely on the interval-valued nature of the index. revision: yes
Referee: [Subadditivity section] The subadditivity claim for the upper endpoint under sums of two C^{1,1} functions is a key property. The argument must be checked for whether it uses only the convex-hull definition or requires additional structure on the limiting Hessians; if the proof invokes a specific inequality for the nuclear norm on convex hulls, state the precise step and any equality cases.

Authors: The proof relies exclusively on the convex-hull definition of the Clarke generalized Hessian. Specifically, the inclusion Hess(h_1 + h_2; x) ⊆ Hess(h_1; x) + Hess(h_2; x) follows directly because every limiting Hessian of the sum is the sum of limiting Hessians of each function, and the closed convex hull preserves the inclusion. The functional φ(M) = dist_{||·||_*}(M, PSD) satisfies φ(A + B) ≤ φ(A) + φ(B) by the triangle inequality for the nuclear norm (which is a norm) and the fact that the PSD cone is a closed convex cone; the distance therefore obeys the required subadditivity. No additional structure on the limiting Hessians is used beyond their existence in the definition. We will revise the subadditivity section to insert an explicit sentence stating this step and to add a remark on equality cases: equality holds, for instance, when the maximizing matrices in each set are simultaneously diagonalizable with the same eigenvectors (which includes the C^2 case). revision: yes

Circularity Check

0 steps flagged

No significant circularity

full rationale

The paper defines its interval-valued local nonconvexity index directly by evaluating the nuclear-norm distance functional (imported from the cited Davydov-Moldavskaya-Zitikis work on C^2 functions) over the Clarke generalized Hessian set Hess(h;x). This replacement is an explicit construction rather than a self-definition or fitted input; the reduction to the smooth case is presented only as a consistency check, not as a load-bearing derivation. No self-citations appear in the load-bearing steps, no parameters are fitted and relabeled as predictions, and no uniqueness theorems or ansatzes are smuggled via prior author work. The listed properties (vanishing on convex functions, orthogonal invariance, subadditivity of the upper endpoint, upper semicontinuity, and relation to weak-convexity modulus) are derived from the definition and standard properties of the generalized Hessian, making the derivation self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The construction rests on the standard definition of the Clarke generalized Hessian from nonsmooth analysis and on the nuclear-norm distance functional from the cited smooth case; no new free parameters, invented entities, or ad-hoc axioms are introduced.

axioms (1)

domain assumption The Clarke generalized Hessian set is well-defined for every C^{1,1} function as the closed convex hull of limiting Hessians at nearby twice-differentiability points.
This is the standard object from nonsmooth analysis invoked to replace the classical Hessian.

pith-pipeline@v0.9.0 · 5550 in / 1482 out tokens · 60114 ms · 2026-05-12T00:48:44.201435+00:00 · methodology

discussion (0)

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

IndisputableMonolith/Cost/FunctionalEquation.lean washburn_uniqueness_aczel unclear
Evaluating the same spectral functional over Hess(h;x) gives an interval-valued local nonconvexity index whose lower and upper endpoints represent, respectively, the least and greatest visible second-order nonconvexity at x.
IndisputableMonolith/Foundation/AbsoluteFloorClosure.lean absolute_floor_iff_bare_distinguishability unclear
H(h;x) := cl co L(h;x) ... Clarke generalized Jacobian of the locally Lipschitz mapping ∇h

Reference graph

Works this paper leans on

18 extracted references · 18 canonical work pages

[1]

Aubin and H

J.-P. Aubin and H. Frankowska,Set-Valued Analysis, Birkh¨ auser, Boston, 1990

work page 1990
[2]

F. H. Clarke, Generalized gradients and applications,Transactions of the American Mathematical Society, 205 (1975), 247–262

work page 1975
[3]

F. H. Clarke,Optimization and Nonsmooth Analysis, Classics in Applied Mathematics, Vol. 5, SIAM, Philadelphia, 1990

work page 1990
[4]

Y. A. Davydov, E. M. Moldavskaya, and R. Zitikis, Searching for and quantifying nonconvexity regions of functions,Lithuanian Mathematical Journal, 59(4) (2019), 507–518. doi:10.1007/s10986-019-09465-6

work page doi:10.1007/s10986-019-09465-6 2019
[5]

Davis and D

D. Davis and D. Drusvyatskiy, Stochastic model-based minimization of weakly convex functions,SIAM Journal on Optimization, 29(1) (2019), 207–239. doi:10.1137/18M1178244

work page doi:10.1137/18m1178244 2019
[6]

Deng and W

Q. Deng and W. Gao, A smooth approximation framework for weakly convex opti- mization, arXiv:2512.09720, 2025

work page arXiv 2025
[7]

Gfrerer, On second-order variational analysis of variational convexity of prox- regular functions,Set-Valued and Variational Analysis, 33 (2025), Article 8

H. Gfrerer, On second-order variational analysis of variational convexity of prox- regular functions,Set-Valued and Variational Analysis, 33 (2025), Article 8. doi:10.1007/s11228-025-00744-8

work page doi:10.1007/s11228-025-00744-8 2025
[8]

P. D. Khanh, B. S. Mordukhovich, and V. T. Phat, A generalized Newton method for subgradient systems,Mathematics of Operations Research, 48 (2023), 1811–1845. doi:10.1287/moor.2022.1320

work page doi:10.1287/moor.2022.1320 2023
[9]

P. D. Khanh, B. S. Mordukhovich, and V. T. Phat, Variational convexity of func- tions and variational sufficiency in optimization,SIAM Journal on Optimization, 33(2) (2023), 1121–1158. doi:10.1137/22M1519250

work page doi:10.1137/22m1519250 2023
[10]

P. D. Khanh, B. S. Mordukhovich, V. T. Phat, and D. B. Tran, Generalized damped Newton algorithms in nonsmooth optimization via second-order subdifferentials,Jour- nal of Global Optimization, 86 (2023), 93–122. doi:10.1007/s10898-022-01248-7

work page doi:10.1007/s10898-022-01248-7 2023
[11]

P. D. Khanh, B. S. Mordukhovich, V. T. Phat, and D. B. Tran, Globally convergent coderivative-based generalized Newton methods in nonsmooth optimization,Mathe- matical Programming, 205 (2024), 373–429. doi:10.1007/s10107-023-01980-2. 13

work page doi:10.1007/s10107-023-01980-2 2024
[12]

P. D. Khanh, V. V. H. Khoa, B. S. Mordukhovich, and V. T. Phat, Second-order subdifferential optimality conditions in nonsmooth optimization,SIAM Journal on Optimization, 35(2) (2025), 678–711. doi:10.1137/23M1627237

work page doi:10.1137/23m1627237 2025
[13]

P. D. Khanh, B. S. Mordukhovich, V. T. Phat, and L. D. Viet, Generalized twice differentiability and quadratic bundles in second-order variational analysis,Set-Valued and Variational Analysis, 34 (2026). doi:10.1007/s11228-026-00795-5

work page doi:10.1007/s11228-026-00795-5 2026
[14]

A. S. Lewis and S. Zhang, Partial smoothness, tilt stability, and generalized Hessians, SIAM Journal on Optimization, 23(1) (2013), 74–94. doi:10.1137/110852103

work page doi:10.1137/110852103 2013
[15]

P´ erez-Aros and E

P. P´ erez-Aros and E. Vilches, Moreau envelope of supremum functions with applica- tions to infinite and stochastic programming,SIAM Journal on Optimization, 31(3) (2021), 1635–1657. doi:10.1137/20M1373517

work page doi:10.1137/20m1373517 2021
[16]

R. A. Poliquin and R. T. Rockafellar, Generalized Hessian properties of regular- ized nonsmooth functions,SIAM Journal on Optimization, 6(4) (1996), 1121–1137. doi:10.1137/S1052623494279316

work page doi:10.1137/s1052623494279316 1996
[17]

R. T. Rockafellar,Convex Analysis, Princeton University Press, Princeton, 1970

work page 1970
[18]

R. T. Rockafellar and R. J.-B. Wets,Variational Analysis, Grundlehren der Mathe- matischen Wissenschaften, Vol. 317, Springer, Berlin, 1998. 14

work page 1998