REVIEW 4 minor
Sharp Lower Bounds on the Haraux Function Beyond Reflexivity
T0 review · 0 major / 4 minor · reviewed 2026-08-15 · deepseek-v4-flash
Pith's one-line read The sharp 1/2 Haraux lower bound holds for type (NI) monotone operators on any real Banach space.
desk verdict A small, honest paper that answers Combettes–Mayrand by moving the sharp 1/2 Haraux bound to type (NI) operators on arbitrary Banach spaces; the internal proof is sound, and the only risky hinge is a cited equivalence the authors do not prove. read the letter →
The pith
A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.
The reading
What carries the argument
The load-bearing object is the weighted residual $R_\gamma(a,a^*) = \|a\|^2/(2\gamma) + (\gamma/2)\|a^*\|^2_* + \langle a,a^*\rangle$, together with the exact local identity $h_A(x,u^*;y,y^*) + R_\gamma(y-x,y^*-u^*) = \tfrac12 D^2_\gamma((x,u^*),(y,y^*))$. This identity decomposes the half squared graph displacement into a Haraux coupling and a quasidensity mismatch for every graph point. The zero-infimum step is supplied by Lemma 2.1, which scales the operator by $\gamma$ and applies the type-(NI)-quasidensity equivalence to conclude that the residual infimum vanishes. The defect $\delta_{A,\gamma}(x,u^*) = \inf_{(y,y^*)\in\mathrm{gra}\,A} R_\gamma(y-x,y^*-u^*)$ packages this residual, and Corollary 3.1 shows that $\delta=0$ is equivalent to a Haraux-side asymptotic exhaustion condition and yields the defect-corrected bound.
What would settle it
A concrete refutation would be a maximally monotone operator of type (NI) on a nonreflexive Banach space with a point $(x,u^*)$ and weight $\gamma>0$ for which the weighted residual infimum $\delta_{A,\gamma}(x,u^*)$ is positive, since Lemma 2.1 asserts it is always zero; equivalently, any pair with $H_A(x,u^*) < \tfrac12 d^2_{\mathrm{gra}\,A,\gamma}(x,u^*) - \varepsilon$ would overturn Theorem 3.1. One could search for this by evaluating the residual infimum on the explicit $c_0$ operator $A=0$ at shifted targets and checking whether the computed value ever exceeds a positive $\varepsilon$.
Extended reading notes
Core claim
The central claim is Theorem 3.1: for every maximally monotone operator $A$ of type (NI) on an arbitrary real Banach space $X$, every $(x,u^*)\in X\times X^*$ and every $\gamma>0$, $H_A(x,u^*) \ge \tfrac12 d^2_{\mathrm{gra}\,A,\gamma}(x,u^*)$, and $\tfrac12$ is the best uniform constant. The proof does not produce a point where the graph-distance infimum, the residual infimum, or the Haraux supremum is attained. Instead it uses the exact identity $h_A(x,u^*;y,y^*) + R_\gamma(y-x,y^*-u^*) = \tfrac12 D^2_\gamma((x,u^*),(y,y^*))$, valid for every graph point. Because type (NI) is equivalent to quasidensity for maximally monotone operators, the infimum of the residual $R_\gamma$ is zero, and $\varepsilon$-level certificates give the bound by taking $\varepsilon\to 0$. The paper also derives a defect-corrected bound for arbitrary operators with nonempty graph and shows that the zero-defect regime characterizes type (NI) within the maximally monotone class.
Load-bearing premise
The load-bearing premise is the published equivalence, for maximally monotone operators on arbitrary real Banach spaces, between type (NI) and quasidensity: Lemma 2.1 uses it to conclude that the weighted residual infimum is zero, and that zero-infimum step is what turns the local decomposition into the global sharp bound. If the equivalence fails in this generality, the proof of Theorem 3.1 collapses.
Editorial extensions
If this is right
- For every weight $\gamma>0$, $\tfrac12$ is the best uniform constant in the Haraux graph-distance bound for maximally monotone operators of type (NI) on arbitrary real Banach spaces.
- Reflexivity is demoted from a geometric hypothesis to an attainment device: reflexive operators admit exact certificates, but the sharp bound only needs approximate certificates.
- Every proper lower semicontinuous convex function $\varphi$ on an arbitrary Banach space satisfies $\varphi(x)+\varphi^*(u^*) - \langle x,u^*\rangle \ge \tfrac12 d^2_{\mathrm{gra}\,\partial\varphi,\gamma}(x,u^*)$, by the Fenchel\,–\,Young bridge.
- The earlier $\tfrac14$ nonreflexive estimate is not an intrinsic obstruction; it is superseded for the type (NI) class.
- The defect-corrected inequality gives a quantitative lower bound even for operators with nonempty graph that are not type (NI).
Reading between the lines
- For a maximally monotone operator that is not type (NI), the defect $\delta_{A,\gamma}(x,u^*)$ is positive at some targets, so the decomposition suggests a quantitative hierarchy $H_A \ge \tfrac12 d^2 - \delta$ in which computing or bounding the defect becomes a natural numerical target.
- The same identity may transfer to other graph-displacement energies built from different gauges or moduli, producing analogous sharp constants tied to the gauge's modulus rather than to the Hilbertian square norm.
- In optimization over nonreflexive spaces, Corollary 4.2 gives a checkable lower bound on the Fenchel\,–\,Young gap that could certify when a candidate pair is far from optimality.
- A natural stress test is to seek a maximally monotone type (NI) operator on $\ell^\infty$ or another nonreflexive space where the old $\tfrac14$ bound is attained; the theorem says none exists, so any apparent counterexample would expose an error in the type-(NI)-quasidensity equivalence or in the scaling step.
Signed reviews
Editorial analysis
A structured set of objections, weighed in public.
Referee Report
Summary. The paper proves a sharp lower bound for the Haraux function of maximally monotone operators of type (NI) on arbitrary real Banach spaces. Theorem 3.1 states that for every such operator A, every (x,u*) in X times X*, and every gamma > 0, one has H_A(x,u*) >= (1/2) d^2_{gra A, gamma}(x,u*), and that 1/2 is the best uniform constant. The proof is built on an exact local decomposition, Proposition 3.1, expressing the sum of the Haraux contribution and a weighted residual as half the squared weighted graph displacement. Lemma 2.1 supplies the key zero-infimum property of the residual via the published equivalence between type (NI) and quasidensity. Corollary 3.1 records a defect-corrected bound and a Haraux-side characterization of type (NI). Example 4.1 proves sharpness for every weight, Example 4.2 separates exact-certificate attainment from zero defect on c0, and Corollary 4.2 transfers the bound to Fenchel-Young gaps of lower semicontinuous convex functions.
Significance. If the cited type (NI)-quasidensity equivalence is valid in the stated generality, the paper fully answers the question raised by Combettes and Mayrand: the sharp 1/2 constant survives beyond reflexivity, without any attainment assumption. The main strengths are the exact algebraic energy decomposition, the circulation-free and explicit epsilon arguments, the clean separation of the three logical levels in Remark 4.1, and the concrete c0 example showing that approximate certificates can exist where exact ones do not. The authors are also careful to state that the proof does not assert equality between the Haraux function and the squared graph distance. The only nonlocal input is the cited equivalence from Simons [16]; this is a standard published result, not an internal gap, but because the main theorem rests on it, the manuscript should state its precise scope explicitly. Overall, this is a focused, correct, and useful contribution to monotone operator theory.
minor comments (4)
- [Definition 2.3 / Lemma 2.1] The nonnegativity of R_gamma is proved in (23) under the hypotheses of Lemma 2.1, but Definition 2.3 defines delta_{A,gamma} for an arbitrary operator with nonempty graph and refers to (23). Since the norm inequality R_gamma(a,a*) >= (1/2)(||a||/sqrt(gamma) - sqrt(gamma)||a*||_*)^2 holds for all (a,a*) independently of monotonicity or type (NI), the nonnegativity statement should be moved before Definition 2.3 and stated in that generality.
- [Lemma 2.1, Eq. (22)] The zero-infimum step is the single load-bearing external input: it applies the type (NI)-quasidensity equivalence of [16] to the scaled operator B. The authors should state the equivalence explicitly as a named known theorem or a clearly identified premise, and confirm that it holds for maximally monotone operators on arbitrary real Banach spaces. As written, the main theorem is conditional on the exact scope of [16].
- [Example 4.2] The display defining z^{(N)} contains a typographical artifact in the underbrace; it should read z^{(N)}=(1,...,1,0,0,...) with N leading ones.
- [Corollary 4.2] The citation chain should be made fully explicit: Corollary 4.2 invokes Theorem 3.1 after using [17] for quasidensity of the subdifferential, and the quasidensity-to-type-(NI) step depends on the equivalence quoted in Lemma 2.1. A one-sentence reminder would make the logic easier for readers to verify.
Circularity Check
No significant circularity: the proof is self-contained given the independently cited type (NI)-quasidensity equivalence.
full rationale
The derivation chain is not circular. Proposition 3.1 is an exact algebraic identity obtained by expanding definitions and canceling the two duality terms; it uses neither monotonicity nor the desired bound. Corollary 3.1 derives the defect inequality (34) from Proposition 3.1 by a standard epsilon argument over the graph. Lemma 2.1 constructs the scaled operator B, verifies maximal monotonicity and type (NI) directly, and then invokes the published type (NI)-quasidensity equivalence [16] to obtain the zero residual infimum (22). Theorem 3.1 follows by selecting a graph point with residual below epsilon and letting epsilon tend to zero; it requires no attained residual, no attained graph distance, and no reflexivity. The sharpness proof in Example 4.1 is a direct calculation for A = gamma^{-1} I in Hilbert space, and Example 4.2 is a direct computation on c0 separating zero defect from exact certificate attainment. The paper explicitly states that Definition 2.3 only packages the translated weighted quasidensity infimum, so there is no hidden renaming of an input as a prediction. No fitted constants, no fitted inputs called predictions, and no load-bearing self-citations appear; the author's self-citations [8,9] are contextual and unrelated to the main theorem. The one external premise, the equivalence between type (NI) and quasidensity for maximally monotone operators on arbitrary real Banach spaces, is cited to independent prior work (Simons [16]) rather than established by assuming the conclusion. Reliance on that external theorem is normal mathematical practice, not circularity, and no internal step assumes the sharp bound in order to prove it.
Assumptions & free parameters
assumptions (3)
- domain assumption For every maximally monotone operator, type (NI) is equivalent to quasidensity (Simons).
- domain assumption Subdifferentials of proper lsc convex functions are maximally monotone and quasidense (hence type (NI)) in arbitrary Banach spaces.
- standard math Standard convex analytic facts: q*_gamma(b*) = gamma/2 ||b*||^2_*, dq_gamma = (1/gamma)J, and the duality and Fenchel-Young identities.
Cite this review
Pith. "Pith review of Sharp Lower Bounds on the Haraux Function Beyond Reflexivity." pith.science (2026). https://pith.science/paper/QE4GV3AA
@misc{pith2026260813139,
author = {Pith},
title = {Pith review of: Sharp Lower Bounds on the Haraux Function Beyond Reflexivity},
year = {2026},
howpublished = {\url{https://pith.science/paper/QE4GV3AA}},
note = {Machine review of arXiv:2608.13139}
}
abstract
We prove that the sharp $\frac{1}{2}$ lower bound for the Haraux function holds for every maximally monotone operator of type~(NI) on an arbitrary real Banach space. This resolves the nonreflexive extension raised by the recent reflexive result. We establish an exact decomposition at each graph point, where the local contribution to the Haraux function and a nonnegative residual together equal $\frac{1}{2}$ times the weighted squared displacement. Since the equivalence between type~(NI) and quasidensity provides graph points whose residuals tend to zero, this decomposition also yields the sharp bound without requiring a graph point at which the residual vanishes. Moreover, for every operator with a nonempty graph, this decomposition yields a lower bound involving the residual infimum. For maximally monotone operators, this decomposition also yields a new characterization of type~(NI) in terms of the Haraux function. Finally, on $c_0$, we give a maximally monotone operator of type~(NI) for which the residual infimum is zero at some target but is not attained. This shows that the existence of a graph point at which the residual vanishes is strictly stronger than the vanishing of the residual infimum required in our proof.
Reviewed August 15, 2026 · model on record in the stance chip above.
Discussion (0). Continue with ORCID to comment.