arxiv: 2604.03852 · v1 · submitted 2026-04-04 · 🧮 math.FA

Recognition: 2 theorem links

· Lean Theorem

A Functional-Analytic Framework for Nonlinear Adaptive Memory: Hierarchical Kernels, State-Dependent Sensitivity, and Memory-Dependent Functionals

Jiahao Jiang

Authors on Pith no claims yet

Pith reviewed 2026-05-13 16:51 UTC · model grok-4.3

classification 🧮 math.FA

keywords nonlinear adaptive memorymemory kernelsadaptive sensitivity functionsmemory-dependent functionalsfunctional analysisdiscontinuous functionssupremum norm comparison

0 comments

The pith

A memory-dependent functional S strictly contains the continuous functions and exceeds the supremum norm when maxima occur in the interior.

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

The paper develops a three-layer functional-analytic framework for nonlinear adaptive memory. Memory kernels are classified into admissible, regular, and generalized classes. Adaptive sensitivity functions Lambda are defined with a history-based construction that satisfies a Lipschitz condition. The resulting functional S_{kappa, Lambda}(f) defines a space M that properly contains C(I), includes discontinuous functions such as indicator functions of subintervals, and satisfies S(f) > ||f||_inf for functions whose maximum is attained inside the interval.

Core claim

The central claim is that the adaptive memory-dependent functional S_{kappa, Lambda}(f) = sup_{t in I} (|f(t)| + integral_0^t Lambda(s, f(s)) kappa(t-s) |f(s)| ds) generates a space M_{kappa, Lambda}(I) in which C(I) is strictly contained, discontinuous functions belong to the space, and the strict inequality S(f) > ||f||_inf holds whenever the maximum of |f| is interior, demonstrating a nontrivial memory contribution beyond the classical supremum norm.

What carries the argument

The adaptive memory-dependent functional S_{kappa, Lambda}(f) that augments the pointwise absolute value with a weighted integral of past states using state-dependent sensitivity Lambda and a memory kernel kappa.

If this is right

Discontinuous functions such as indicator functions of subintervals belong to M, capturing abrupt changes like on-off switching in nonlinear systems.
A strict inequality S(f) > ||f||_inf holds when the maximum of |f| is attained in the interior of the interval.
The framework establishes absolute convergence, measurability, uniform boundedness, positive definiteness, and direct comparison with the classical supremum norm.
The Lipschitz estimate ||Lambda_f - Lambda_g||_inf <= L_Lambda ||f - g||_inf holds for the constructed sensitivity functions.

Where Pith is reading between the lines

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

The space M may provide a natural setting for analyzing stability of systems whose responses include abrupt switches or state-dependent memory.
The explicit construction of Lambda from accumulated deviations could be used to derive bounds on solution growth in integral equations with adaptive memory.
Because M properly contains C(I), norm estimates derived from S could be applied to a wider class of signals than classical uniform norms allow.

Load-bearing premise

Adaptive sensitivity functions Lambda satisfy natural conditions that permit a continuous interpolation between instantaneous response and history-dependent sensitivity via historical deviation accumulation.

What would settle it

An explicit computation for a continuous function attaining its maximum in the interior of I where S equals the supremum norm would falsify the strict inequality claim.

read the original abstract

This work develops a systematic functional-analytic framework for nonlinear adaptive memory, where the influence of past events depends on both elapsed time and the state values along a trajectory. The framework comprises three hierarchical layers. First, memory kernels are classified into mathematically admissible, regular (uniformly bounded, normalized, Lipschitz), and generalized (bounded variation, possibly sign-changing) classes. Second, adaptive sensitivity functions Lambda(s, f(s)) are introduced, satisfying natural conditions; a concrete construction based on historical deviation accumulation interpolates continuously between instantaneous response and history-dependent sensitivity, with an explicit Lipschitz estimate ||Lambda_f - Lambda_g||_inf <= L_Lambda ||f - g||_inf. Third, an adaptive memory-dependent functional S_{kappa, Lambda}(f) = sup_{t in I} (|f(t)| + integral_0^t Lambda(s, f(s)) kappa(t-s) |f(s)| ds) and the associated set M_{kappa, Lambda}(I) = {f : S_{kappa, Lambda}(f) < infinity} are constructed. Fundamental properties of the framework are established, including absolute convergence, measurability, uniform boundedness, positive definiteness, and comparison with the classical supremum norm. It is shown that C(I) is strictly contained in M_{kappa, Lambda}(I), with discontinuous functions (e.g., indicator functions of subintervals) belonging to the set -- capturing abrupt signal changes such as on-off switching in nonlinear systems. When the maximum of |f| is attained in the interior of the interval, a strict inequality S_{kappa, Lambda}(f) > ||f||_inf is proved, demonstrating the nontrivial contribution of the memory component.

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

The paper sets up a clean three-layer framework for memory functionals that includes discontinuous functions and shows S exceeding the sup norm, but the results track closely from the definitions with no external validation or examples.

read the letter

The main point is a hierarchical construction for nonlinear adaptive memory: kernels split into admissible, regular (bounded, normalized, Lipschitz), and generalized classes; an adaptive Lambda built from historical deviation accumulation with an explicit Lipschitz bound; and the functional S as the sup of |f(t)| plus the weighted integral against kappa. They prove C(I) sits strictly inside M, with indicators belonging to M, and S(f) > ||f||_inf when the max of |f| is attained inside the interval. That containment and inequality follow once kappa is integrable and Lambda meets the boundedness conditions, which is straightforward but useful for capturing on-off type jumps in systems. The Lambda construction itself is concrete and gives a continuous interpolation between instantaneous and history-dependent response. The paper does a decent job keeping the definitions consistent and listing the basic properties like absolute convergence, measurability, and uniform boundedness. The stress-test note is right that there is no internal gap between the hypotheses and the stated conclusions. Where it is softer is the absence of any worked example, numerical check, or link to an existing applied model; everything stays at the level of setting up the objects and verifying the immediate consequences. The claims about positive definiteness and comparison with the classical norm are asserted without sketches here, so it is hard to judge how much extra work is hidden in the details. The engagement with prior memory-kernel literature is light, which is fine for a new classification but leaves the novelty feeling mostly organizational. This is for functional analysts or modelers who want a norm-like object that handles abrupt changes in history-dependent nonlinear equations. A reader already thinking about integral operators with state-dependent weights could borrow the Lambda idea or the M set. It deserves peer review because the framework is coherent on its own terms and the central claims are checkable; referees could push for examples and fuller derivations without the paper being incoherent.

Referee Report

1 major / 3 minor

Summary. The manuscript develops a three-layer functional-analytic framework for nonlinear adaptive memory. It first classifies memory kernels kappa into admissible, regular (uniformly bounded, normalized, Lipschitz), and generalized (bounded variation) classes. It then introduces adaptive sensitivity functions Lambda(s, f(s)) satisfying natural boundedness and Lipschitz conditions, with a concrete construction via historical deviation accumulation that interpolates between instantaneous and history-dependent responses. Finally, it defines the memory-dependent functional S_{kappa, Lambda}(f) = sup_{t in I} (|f(t)| + integral_0^t Lambda(s, f(s)) kappa(t-s) |f(s)| ds) and the space M_{kappa, Lambda}(I) = {f : S(f) < infinity}. The paper claims to establish absolute convergence, measurability, uniform boundedness, positive definiteness, and comparison with the sup norm; it proves the strict inclusion C(I) subset M_{kappa, Lambda}(I) (with discontinuous functions such as interval indicators in M) and the strict inequality S(f) > ||f||_infty whenever the maximum of |f| is attained in the interior of I.

Significance. If the stated properties and inclusions are rigorously verified, the framework offers a systematic extension of the classical supremum norm that incorporates state-dependent memory effects, potentially useful for analyzing nonlinear systems with abrupt transitions or history-dependent sensitivities. The hierarchical kernel classification and explicit Lipschitz estimate on Lambda provide concrete tools that could support further development in functional analysis or applications such as adaptive control, though the overall significance hinges on the completeness of the derivations in the full manuscript.

major comments (1)

[§4] §4, Theorem on strict containment and interior-max inequality: the claim that S_{kappa, Lambda}(f) > ||f||_infty for interior maxima follows from positivity and support conditions on kappa and Lambda, but the manuscript must explicitly verify that the integral term is strictly positive under the admissible-class hypotheses for every such f; without this step the strict inequality is not load-bearing.

minor comments (3)

[§2] Notation for the interval I and the classes (admissible/regular/generalized) is introduced without a consolidated table; adding one would improve readability.
[§3.1] The concrete construction of Lambda via historical deviation accumulation is given with an explicit Lipschitz constant, but the dependence of L_Lambda on the deviation parameter is not stated; this should be clarified.
[Abstract] The abstract asserts 'absolute convergence' and 'positive definiteness' without cross-references to the relevant propositions; these should be linked in the text.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for the careful reading, positive overall assessment, and the specific suggestion regarding the strict inequality in §4. The comment is well-taken and we will incorporate an explicit verification step in the revised manuscript.

read point-by-point responses

Referee: [§4] §4, Theorem on strict containment and interior-max inequality: the claim that S_{kappa, Lambda}(f) > ||f||_infty for interior maxima follows from positivity and support conditions on kappa and Lambda, but the manuscript must explicitly verify that the integral term is strictly positive under the admissible-class hypotheses for every such f; without this step the strict inequality is not load-bearing.

Authors: We agree that the strict inequality S(f) > ||f||_infty requires an explicit demonstration that the integral term is strictly positive. In the current proof we invoke the nonnegativity of admissible kernels kappa together with the fact that Lambda(s,f(s)) is bounded below by a positive constant on a left neighborhood of any interior maximum point (by the historical-deviation construction of Lambda). Because admissible kernels are normalized, nonnegative, and have positive measure support on every interval (0,δ), the integral over that neighborhood is strictly positive. To make this step fully load-bearing and transparent, we will add a short auxiliary lemma in the revised §4 that isolates and proves the strict positivity of the integral term under precisely the admissible-class hypotheses. This will be inserted immediately before the statement of the interior-max theorem. revision: yes

Circularity Check

0 steps flagged

No significant circularity

full rationale

The paper constructs a functional-analytic framework by defining three hierarchical layers—admissible/regular/generalized memory kernels, adaptive sensitivity functions Lambda satisfying explicit boundedness and Lipschitz conditions, and the memory-dependent functional S_{kappa, Lambda}(f) together with the set M_{kappa, Lambda}(I)—then derives the claimed inclusions and strict inequalities directly from those definitions. Continuous functions lie in M because they are bounded on compact I; indicator functions belong to M but not C(I) because the integral term remains finite for any bounded measurable f once kappa is integrable; the strict inequality S > ||f||_inf when the maximum is interior follows from the positivity and support conditions on the kernel and the historical-deviation construction of Lambda. No step reduces a prediction to a fitted parameter, invokes a self-citation as load-bearing, or renames a known result; all conclusions are immediate consequences of the stated hypotheses with no internal gap.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 3 invented entities

The framework rests on definitions of kernel classes and natural conditions for Lambda without external benchmarks or independent evidence for the new entities; no free parameters are explicitly fitted in the abstract.

axioms (1)

domain assumption Lambda(s, f(s)) satisfies natural conditions including the Lipschitz estimate ||Lambda_f - Lambda_g||_inf <= L_Lambda ||f - g||_inf
Invoked to ensure the sensitivity functions are well-behaved and to support the construction interpolating between instantaneous and history-dependent responses.

invented entities (3)

adaptive sensitivity function Lambda(s, f(s)) no independent evidence
purpose: To encode state-dependent memory influence that varies with historical deviation accumulation
Newly introduced construction in the second layer of the framework.
memory-dependent functional S_{kappa, Lambda}(f) no independent evidence
purpose: To define a size measure that incorporates both current value and integrated memory kernel effects
Central new object defined as the sup over t of |f(t)| plus the integral term.
set M_{kappa, Lambda}(I) no independent evidence
purpose: To collect functions with finite S, strictly containing C(I) and including discontinuous signals
Defined as the set where S is finite, used to demonstrate the framework's reach beyond classical spaces.

pith-pipeline@v0.9.0 · 5617 in / 1456 out tokens · 52251 ms · 2026-05-13T16:51:49.742817+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 Jcost_pos_of_ne_one; washburn_uniqueness_aczel echoes

?

echoes
ECHOES: this paper passage has the same mathematical shape or conceptual pattern as the Recognition theorem, but is not a direct formal dependency.

S_{κ,Λ}(f) = sup_{t∈I} (|f(t)| + ∫_0^t Λ(s,f(s)) κ(t-s) |f(s)| ds) and M_{κ,Λ}(I) = {f : S < ∞}; C(I) ⊂ M strictly, with S(f) > ||f||_∞ when max attained interior
IndisputableMonolith/Foundation/AlphaCoordinateFixation.lean costAlphaLog_high_calibrated_iff refines

?

refines
Relation between the paper passage and the cited Recognition theorem.

Λ_f satisfies ||Λ_f - Λ_g||_∞ ≤ L_Λ ||f-g||_∞ and positivity at zero; historical deviation accumulation interpolates instantaneous vs history-dependent sensitivity

What do these tags mean?

matches: The paper's claim is directly supported by a theorem in the formal canon.
supports: The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends: The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses: The paper appears to rely on the theorem as machinery.
contradicts: The paper's claim conflicts with a theorem or certificate in the canon.
unclear: Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

48 extracted references · 48 canonical work pages

[1]

Cortázar, F

C. Cortázar, F. Quirós, N. Wolanski,A heat equation with memory: Large-time behavior, J. Funct. Anal.281(2021), 109174. 40

work page 2021
[2]

Cortázar, F

C. Cortázar, F. Quirós, N. Wolanski,Asymptotic profiles for inhomogeneous heat equations with memory, Math. Ann.389(2024), 3705–3746

work page 2024
[3]

J. Wang, Q. Ma, W. Zhou,Attractor of the nonclassical diffusion equation with memory on time- dependent space, AIMS Math.8(2023), 14820–14841

work page 2023
[4]

N.Mori,DissipativestructureandglobalexistenceincriticalspaceforTimoshenkosystemofmemory type, J. Differ. Equ.265(2018), 1627–1653

work page 2018
[5]

Y. Qin, X. Pan,Global existence, asymptotic behavior and uniform attractors for a non-autonomous TimoshenkosystemofthermoelasticityoftypeIIIwithatime-varyingdelay,J.Math.Anal.Appl.484 (2020), 123672

work page 2020
[6]

Baeumer, M

B. Baeumer, M. Geissert, M. Kovács,Existence, uniqueness and regularity for a class of semilinear stochastic Volterra equations with multiplicative noise, J. Differ. Equ.258(2015), 535–554

work page 2015
[7]

B.S.H.Kashkaria,M.I.Syam,Evolutionarycomputationalintelligenceinsolvingaclassofnonlinear Volterra–Fredholm integro-differential equations, J. Comput. Appl. Math.311(2017), 314–323

work page 2017
[8]

Z. L. Lu,A posteriori error estimates of fully discrete finite-element schemes for nonlinear parabolic integro-differential optimal control problems, Adv. Differ. Equ.2014(2014), 15

work page 2014
[9]

W. Feng, S. X. Yang, H. X. Wu,Robust stability analysis of neutral-type hybrid bidirectional asso- ciative memory neural networks with time-varying delays, Abstr. Appl. Anal.2014(2014), 560861

work page 2014
[10]

S. L. Manu, S. Shikaa, T. Richard, E. P. Dovi,Mathematical model for prediction of Tuberculosis in Nigeriausinghybridfractionaldifferentialequationsandartificialneuralnetworkmethods, Franklin Open11(2025), 100248

work page 2025
[11]

Anal.25(2022), 1426–1458

M.D.Ruiz-Medina,SpectralanalysisofmultifractionalLRDfunctionaltimeseries,Fract.Calc.Appl. Anal.25(2022), 1426–1458

work page 2022
[12]

Wulkow, P

N. Wulkow, P. Koltai, C. Schütte,Memory-based reduced modelling and data-based estimation of opinion spreading, J. Nonlinear Sci.31(2021), 19

work page 2021
[13]

D (2025), 134939

A.Korepanov,J.Leppänen,Improvedpolynomialratesofmemorylossfornonstationaryintermittent dynamical systems, Phys. D (2025), 134939

work page 2025
[14]

Alberti, G

G. Alberti, G. Crippa, A. L. Mazzucato,Exponential self-similar mixing by incompressible flows, J. Amer. Math. Soc.32(2019), 445–490

work page 2019
[15]

J. F. Brock, N. M. Dunfield,Norms on the cohomology of hyperbolic 3-manifolds, Invent. Math.210 (2017), 531–558

work page 2017
[16]

K. D. Schmidt,A general Jordan decomposition, Arch. Math.38(1982), 556–564

work page 1982
[17]

Notes63(1996), 61–71

A.S.Leonov,Onthetotalvariationforfunctionsofseveralvariablesandamultidimensionalanalog of Helly’s selection principle, Math. Notes63(1996), 61–71

work page 1996
[18]

V.V.Chistyakov,Y.V.Tretyachenko,Mapsofseveralvariablesoffinitetotalvariation.II.E.Helly-type pointwise selection principles, J. Math. Anal. Appl.369(2010), 82–93

work page 2010
[19]

V.Ene,OnthedecompositiontheoremsofLebesgueandJordan,RealAnal.Exchange23(1)(1998/99), 313–324. 41

work page 1998
[20]

Kantorovitz,A Jordan decomposition for operators in Banach space, Trans

S. Kantorovitz,A Jordan decomposition for operators in Banach space, Trans. Am. Math. Soc.120 (1965), 526–550

work page 1965
[21]

Saminger-Platz, B

S. Saminger-Platz, B. De Baets, H. De Meyer,A Generalization of the Mulholland Inequality for Continuous Archimedean t-Norms, J. Math. Anal. Appl.345(2008), 607–614

work page 2008
[22]

Davvaz, V

B. Davvaz, V. Leoreanu-Fotea,Applications of interval valued fuzzy n-ary polygroups with respect to t-norms (t-conorms), Comput. Math. Appl.57(2009), 1413–1424

work page 2009
[23]

Peiris, V

V. Peiris, V. Roshchina, N. Sukhorukova,Artificial neural networks with uniform norm-based loss functions, Adv. Comput. Math.50(2024), 31

work page 2024
[24]

N.E.Yudin,AdaptiveGauss–Newtonmethodforsolvingsystemsofnonlinearequations,in: Doklady Mathematics, Vol.104(2021), 293–296

work page 2021
[25]

Kitkuan, P

D. Kitkuan, P. Kumam, V. Berinde, A. Padcharoen,Adaptive algorithm for solving the SCFPP of demicontractive operators without a priori knowledge of operator norms, Ann. Univ. Ovidius Math. Ser.27(2019), 153–175

work page 2019
[26]

Sunthrayuth, K

P. Sunthrayuth, K. Muangchoo, P. Cholamjiak, W. Nithiarayaphaks,Inertial self-adaptive algorithm with two different inertial factors for solving split feasibility problems in Banach spaces, The J. Anal. (2025), 1–33

work page 2025
[27]

D. Tian, L. Shi, R. Chen,Iterative algorithm for solving the multiple-sets split equality problem with split self-adaptive step size in Hilbert spaces, J. Inequal. Appl.2016(2016), 34

work page 2016
[28]

Jailoka, S

P. Jailoka, S. Suantai,On split fixed point problems for multi-valued mappings and designing a self-adaptive method, Results Math.76(2021), 133

work page 2021
[29]

Karmitsa, K

N. Karmitsa, K. Joki, A. Airola, T. Pahikkala,Limited memory bundle DC algorithm for sparse pairwise kernel learning, J. Glob. Optim.92(2025), 55–85

work page 2025
[30]

T. Linß, G. Radojev,Maximum-norm a posteriori error bounds for an extrapolated upwind scheme applied to a singularly perturbed convection-diffusion problem, Mediterr. J. Math.21(2024), 161

work page 2024
[31]

A.Esser,A.Mukherjee,S.Sarkar,Memory-efficientattacksonsmallLWEkeys,J.Cryptol.37(2024), 36

work page 2024
[32]

Torregrosa, V

S. Torregrosa, V. Champaney, A. Ammar, V. Herbert, F. Chinesta,Physics-based active learning for design space exploration and surrogate construction for multiparametric optimization, Commun. Appl. Math. Comput.6(2024), 1899–1923

work page 2024
[33]

M. E. Belouafi, M. Beggas, N. E. H. Nesba,Uniform convergence of multigrid methods for elliptic quasi-variational inequalities and its implementation, Commun. Math. Appl.14(2023), 633

work page 2023
[34]

Hintermüller, M

M. Hintermüller, M. Hinze, R. H. W. Hoppe,Weak-duality based adaptive finite element methods for PDE-constrained optimization with pointwise gradient state-constraints, J. Comput. Math. (2012), 101–123

work page 2012
[35]

R.Blanquero,E.Carrizosa,N.Gómez-Vargas,Oncontextualinversemultiobjectiveproblems,Eur.J. Oper. Res. (2025)

work page 2025
[36]

T.C.F.Cheng,C.K.Ing,S.H.Yu,Inversemomentboundsforsampleautocovariancematricesbased on detrended time series and their applications, Linear Algebra Appl.473(2015), 180–201. 42

work page 2015
[37]

G. Chen, G. Fang,Probabilistic adaptive width of a multivariate sobolev space equipped with a gaussian measure, Constr. Approx.24(2006), 245–262

work page 2006
[38]

Girosi,Some extensions of radial basis functions and their applications in artificial intelligence, Comput

F. Girosi,Some extensions of radial basis functions and their applications in artificial intelligence, Comput. Math. Appl.24(1992), 61–80

work page 1992
[39]

Math.45(2024), 5156–5171

A.R.Nurutdinov,Bio-inspiredneuralnetworkarchitectureofembodiedintelligence, LobachevskiiJ. Math.45(2024), 5156–5171

work page 2024
[40]

M. E. Cornejo, J. Medina, F. J. Ocaña,Theories, models and bases of attribute implications in multi-adjoint concept lattices with hedges: ME Cornejo et al., Comput. Appl. Math.45(2026), 30

work page 2026
[41]

I.Tsuda,T.Namiki,Somecommentsontherelationshipbetweentheratefunctioninthelargedeviation principle and the Kullback–Leibler divergence: toward the interpretation of neural estimation of mutual information, Jpn. J. Ind. Appl. Math.43(2026), 7

work page 2026
[42]

Allana, R

S. Allana, R. Dara, X. Lin, P. Xiong,Towards integration of privacy enhancing technologies in explainable artificial intelligence, Knowl.-Based Syst. (2025), 115235

work page 2025
[43]

Palitta, V

D. Palitta, V. Simoncini,Computationally enhanced projection methods for symmetric Sylvester and Lyapunov matrix equations, J. Comput. Appl. Math.330(2018), 648–659

work page 2018
[44]

Shamrai,Analysis of perturbations of singular values in concatenated matrices, Ukrainian Math

M. Shamrai,Analysis of perturbations of singular values in concatenated matrices, Ukrainian Math. J. (2025), 1–14

work page 2025
[45]

J. H. Jiang, B. Miao,A study of anomalous stochastic processes via generalizing fractional calculus, Chaos35(2025)

work page 2025
[46]

B.Dong,D.H.Lv,K.Y.Xi,J.H.Li,D.H.Yu,Adaptivefrequencyevolutiondecompositioncombined with improved fluctuation-based dispersion entropy for muscle fatigue characterization, IEEE Trans. Instrum. Meas. (2026)

work page 2026
[47]

J. H. Jiang,Towards a mathematical theory of adaptive memory: from time-varying to responsive fractional Brownian motion, arXiv preprint arXiv:2512.10057 (2025)

work page arXiv 2025
[48]

J. H. Jiang,Foundations and fundamental properties of a two-parameter memory-weighted velocity operator, arXiv preprint arXiv:2601.05122 (2026). 43

work page arXiv 2026