Metric functionals and weak convergence

Armando W. Guti\'errez; Olavi Nevanlinna

REVIEW 3 major objections 6 minor 25 references

Metric functionals and weak convergence

T0 review · 3 major / 6 minor · reviewed 2026-08-07 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read A metric-only definition of weak convergence matches classical weak convergence on bounded sequences in normed spaces.

desk verdict A clean metric-only notion of weak convergence that matches classical weak convergence for bounded sequences, with one imported-result gap to fix. read the letter →

arxiv 2506.04154 v1 pith:6MIQO5G6 submitted 2025-06-04 math.FA math.MG

Armando W. Gutiérrez , Olavi Nevanlinna This is my paper

classification math.FAmath.MG MSC 46B2046B10

keywords metricfunctionalsweakconvergencespaceshorofunctionsnormedlinearboundedsequencescompactificationBusemann

verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

The paper introduces a definition of weak convergence that makes sense in any metric space, with no linear structure needed: a sequence $x_n$ converges $d$-weakly to $z$ if $\liminf_{n\to\infty} h(x_n)\ge h(z)$ for every metric functional $h$ on the space. Metric functionals are the pointwise limits of distance-to-point functions $d(x,w)-d(o,w)$ built from a fixed basepoint $o$. The central claim, proved as Theorem 1.3, is that for bounded sequences in a normed linear space with its norm-induced metric, this new $d$-weak convergence is exactly the classical weak convergence defined by continuous linear functionals. If true, this gives a fully metric notion of weak convergence that agrees with the usual one wherever the usual one is defined, and it supplies a candidate weak notion for arbitrary metric spaces. The paper further proves uniqueness of $d$-weak limits in normed spaces, upgrades $d$-weak convergence to strong convergence in $\ell^1$ and in spaces with compact closed balls, and conjectures that unbounded sequences never converge $d$-weakly.

What carries the argument

The central object is the metric compactification $X^{\diamondsuit}$: the closure, in the topology of pointwise convergence, of the set of internal functionals $h_w(x)=d(x,w)-d(o,w)$ for a fixed basepoint $o$. Its elements, the metric functionals, are all $1$-Lipschitz and generalize Busemann functionals to arbitrary metrics. The load-bearing identity in the proof is that every extreme point of the dual unit ball of any normed space is one of these metric functionals; combining that inclusion with the classical theorem that bounded sequences are weakly convergent once they converge on all extreme points of the dual ball yields the equivalence in Theorem 1.3.

What would settle it

Check whether the inclusion of extreme points into metric functionals actually holds on a normed space with unknown horofunction structure, such as $\ell^{\infty}$ or $c_0$: identify the extreme points of the dual unit ball and verify that each is a pointwise limit of functions $x\mapsto \|x-w\|-\|w\|$, and simultaneously test whether the bounded, non-weakly-convergent sequence of standard unit vectors $e_n$ in $\ell^{\infty}$ satisfies the $d$-weak inequality against every metric functional.

Watch

Extended reading notes

Core claim

The central discovery is that the inequality $\liminf_{n\to\infty} h(x_n)\ge h(z)$ for every metric functional $h$ characterizes weak convergence for bounded sequences in all normed linear spaces. Metric functionals on a normed space include all extreme points of the dual unit ball, so the metric test is at least as strong as the linear test; conversely, a convex-separation argument shows that ordinary weak convergence forces the metric inequality for every metric functional. The equivalence is what makes $d$-weak convergence a genuine extension rather than a competing definition: on the class where classical weak convergence already exists, the two notions coincide, while the new definition also applies to metric spaces with no linear structure.

Load-bearing premise

The equivalence proof depends on the imported theorem that every extreme point of the dual unit ball of any normed space is a metric functional; if that inclusion fails, the direction from $d$-weak convergence to weak convergence in Theorem 1.3 does not follow from the stated argument.

Editorial extensions

If this is right

For every normed linear space, a bounded sequence converges $d$-weakly if and only if it converges weakly, so the metric definition is a faithful substitute in all classical settings.
In $\ell^1$ and in normed spaces whose closed balls are compact, $d$-weak convergence forces strong convergence, so the new notion is strictly stronger than weak convergence there.
Whenever a sequence converges $d$-weakly to two points, those points coincide in normed spaces, and the set of all $d$-weak limits of a fixed sequence in a $W$-convex metric space is $W$-convex and closed.
If the conjecture on unbounded sequences is correct, then every $d$-weakly convergent sequence in any normed linear space is bounded, so the agreement with weak convergence extends to all sequences, not just bounded ones.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

The definition makes 'weak convergence' a purely metric concept, so it could be used to define a weak topology on any metric space; a natural next step is to study whether this topology is Hausdorff or compact in interesting classes such as CAT(0) spaces.
Because the notion depends on the metric itself, choosing an equivalent but different norm can change which sequences converge $d$-weakly; this suggests the definition is a property of the metric geometry, not of the linear topology.
The proof's reliance on the extreme-point inclusion suggests a route to test the unboundedness conjecture: if a normed space admits an extreme point of the dual ball that is not a metric functional, the equivalence might fail there, so the conjecture could be checked in spaces where explicit horofunction formulas are incomplete.
The paper's $\ell^1$ argument using the gliding hump technique could be adapted to prove strong convergence from $d$-weak convergence in other non-reflexive spaces with explicit metric functionals, such as $c_0$ with a suitable norm.

Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Desk Editor's Note

read the letter

Colleague,

The short version: this is a clean, honest note that does what it says. Gutierrez and Nevanlinna define d-weak convergence by testing against all metric functionals on a metric space, and prove that for bounded sequences in a normed space this agrees with ordinary weak convergence. The definition itself is a natural combination of existing horofunction ideas, but I don't know of another place where exactly this notion is isolated and the equivalence theorem proved. The paper is worth a serious referee.

What it does well: the weak-to-d-weak direction in Theorem 1.3 is a neat separation argument; the converse uses Walsh's result that extreme points of the dual unit ball are metric functionals and Rainwater's theorem. The examples are informative, especially the discrete metric case and the closed unit ball of ℓ2 and ℓ1, and the paper is explicit about what it cannot prove: unbounded sequences are left as Conjecture 1.5, with partial results for ℓ1, C[0,1], and strictly convex duals. That honesty is welcome.

The soft spot is the imported machinery. Theorem 1.3 is stated for all normed linear spaces, including incomplete ones, but the proof relies on Walsh's Corollary 3.5 and Rainwater's theorem, which are typically formulated for Banach spaces. The inclusion E(B_X*) ⊂ X^♢ is not proved in the text, and the hypotheses are not stated. If Walsh's theorem needs completeness, then a density or completion argument is required to transfer it to the normed space; the paper doesn't give one. I don't think this is fatal—likely the results extend by standard arguments—but a referee should ask the authors to spell out the statements they are importing and, if needed, add the completion step. Also note that the d-weak-to-weak implication only gives convergence at extreme points and then calls on Rainwater; if the space is incomplete, Rainwater's theorem may need a Banach-envelope argument. This is a fixable presentation issue, not a counterexample.

The citation pattern looks fine. Self-citations are to the authors' own prior work on metric compactifications, used for formulas and a Busemann representation, not for the main equivalence. No data, code, or fitting is involved; the results are proof-based.

Bottom line: a solid functional-analytic note with a genuinely new definition and a correct-looking main theorem. It deserves peer review, and I'd ask for a revision that makes the Walsh/Rainwater dependency explicit and complete. I'd be happy to see it accepted after that.

Referee Report

3 major / 6 minor

Summary. The paper introduces a notion of weak convergence in arbitrary metric spaces, called d-weak convergence, defined by the condition liminf_n h(x_n) ≥ h(z) for every metric functional h on the space. The main result, Theorem 1.3, states that for bounded sequences in a normed linear space with the induced metric, d-weak convergence coincides with classical weak convergence. The paper also proves uniqueness of d-weak limits (Theorem 1.2), conditions under which unbounded sequences cannot converge d-weakly (Theorem 1.4), stability under taking linear combinations with strongly convergent sequences (Theorem 1.6), and closure/convexity properties of the set of d-weak limits in W-convex spaces (Theorem 1.7). Several examples illustrate the behavior of the notion in discrete spaces, ℓ_p spaces, and closed balls.

Significance. If the main theorem is correct, the paper gives a genuinely metric notion of weak convergence that extends the classical one for bounded sequences in normed spaces. The proof of Theorem 1.3 combines a self-contained convex-separation argument in one direction with two imported tools in the other: Walsh's identification of extreme points of the dual unit ball as metric functionals and Rainwater's theorem. The paper is concise and the central definition is natural. The authors are explicit about the reliance on Walsh's theorem, which is a strength; however, the statement of that theorem is not given, and the paper's claims that depend on it are not fully verified for incomplete normed spaces. The paper also contains a false assertion in the proof of Theorem 1.4 for C[0,1] and false claims in Proposition 3.6 about discrete metric spaces. These issues do not necessarily invalidate the main theorem, but they must be corrected before the paper can be accepted.

major comments (3)

[§4 (proofs of Theorems 1.2, 1.3, 1.4)] The proofs of Theorems 1.2, 1.3, and 1.4 invoke the inclusion E(B_{X*}) ⊂ X^♢ as [Wal18, Corollary 3.5] without stating the result or its hypotheses. The paper applies it to arbitrary normed linear spaces, which may be incomplete and nonseparable. If Walsh's corollary is proved for Banach spaces under the norm metric, the transfer to the normed space X requires an argument (for instance, via the completion of X and the equality of duals) that is not supplied. The same proof of Theorem 1.3 uses Rainwater's theorem, usually formulated for Banach spaces; the reduction to the completion should also be made explicit. Please state the exact versions used and verify that they apply to X.
[§4, proof of Theorem 1.4 (C[0,1] case)] The proof assumes that an unbounded sequence in C[0,1] has a subsequence with |f_{n_i}(τ)|→∞ for some fixed τ∈[0,1]. This is false: the continuous functions f_n(t)=n max(0,1−n|t−1/n|) satisfy ∥f_n∥_∞=n yet converge to 0 pointwise, so no such τ exists. Please replace the argument with one based on the explicit formulas for metric functionals on C[0,1] from [Wal18, Theorem 5.1] or an additional compactness step.
[§3, Proposition 3.6] Items 3 and 4 do not follow from the stated proof and are in fact false. For the discrete metric on an infinite set, take the sequence (2,3,2,4,2,5,...) in N with basepoint 1. The internal functional h_3 satisfies h_3(2)=0, h_3(3)=−1, and h_3(x)=0 otherwise, so its limit inferior along the sequence is −1, while h_3(2)=0; hence the sequence does not converge d-weakly to 2, although 2 is the only point occurring infinitely often. This contradicts item 3, and a similar argument contradicts item 4. The correct behavior is that a sequence in a discrete metric space converges d-weakly if and only if it is eventually constant; please correct the proposition accordingly.

minor comments (6)

[Introduction, first paragraph] The phrase "inallmetric spaces" should read "in all metric spaces."
[§2.2, paragraph after (2.1)] “We equipped the product space R^X with the topology of pointwise convergence” should be “We equip the product space R^X with the topology of pointwise convergence.”
[§2.3, proof of Proposition 2.4] The compactness argument is elliptical: the diagonal subsequence converges pointwise on the countable dense set, and one must show that the limit point h obtained from compactness agrees with that pointwise limit before the epsilon argument applies.
[§4, proof of Theorem 1.3] The convexity of the set Y = {y: h*(y) ≤ A+ε} follows from Proposition 2.8 (every metric functional is W-convex, and a normed space is W-convex with the usual affine combination); this should be cited explicitly.
[§4, proof of Theorem 1.6 and Lemma 4.1] In Lemma 4.1, the limit defining η(x) is asserted without comment; it exists because the displayed identity shows that |s|(‖x−z_α‖−‖z_α‖) converges as the difference of two convergent nets.
[§4, proof of Theorem 1.3] The typo "funtional" should be corrected to "functional."

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: d-weak convergence is independently defined and the main equivalence rests on external Walsh/Rainwater theorems, not on self-citations.

full rationale

The paper's central claim, Theorem 1.3, is an equivalence between a newly defined notion, d-weak convergence (Definition 1.1: liminf h(x_n) ≥ h(z) for all metric functionals h), and classical weak convergence. The definition makes no reference to the dual space or to weak convergence, so the equivalence is not true by definition. The weak-to-d-weak direction is proved internally by separating a point from a closed convex sublevel set of a metric functional; the only non-internal ingredients are standard separation theorems and the fact that metric functionals are convex, which is proved in Proposition 2.8. The d-weak-to-weak direction uses two external results: Walsh [Wal18, Cor. 3.5], that extreme points of the dual unit ball are metric functionals, and Rainwater's theorem [Rai63]. These are independent mathematical theorems, not assumptions that already contain the target equivalence, and no self-citation carries the proof of Theorem 1.3. Author self-citations ([Gut19a], [Gut19b], [GN24]) appear in supporting examples and auxiliary formulas; for instance, Proposition 3.12 uses the published explicit formulas for l1 metric functionals from [Gut19b], a result external to the present target claim. The remaining concern, whether Walsh's and Rainwater's hypotheses are verified for incomplete normed linear spaces, is a correctness or verification issue rather than a circularity. No fitted parameters, no ansatz smuggled in via citation, and no renaming of a known result are used to force the conclusion. The paper is accordingly non-circular.

Assumptions & free parameters 0 free parameters · 6 assumptions · 0 invented entities

The central equivalence rests on standard separation and Rainwater theorems plus Walsh's theorem that extreme points of the dual unit ball lie among metric functionals. No free parameters or invented entities appear. Some examples depend on the first author's earlier explicit formulas for ell_p metric functionals, which are cited rather than reproved.

assumptions (6)

domain assumption Extreme points of the dual unit ball are metric functionals (Walsh, Corollary 3.5).
Used in Theorems 1.2, 1.3, and 1.4 to convert d-weak convergence into convergence against linear functionals. This is an external published result, not proved in the paper.
standard math Rainwater's theorem: a bounded sequence in a Banach space converges weakly iff it converges on the extreme points of the dual unit ball.
Used in Theorem 1.3 to conclude weak convergence from d-weak convergence on extreme points.
standard math Separation theorem for closed convex sets in normed spaces.
Used in the weak-to-d-weak direction of Theorem 1.3 to derive a linear functional that contradicts weak convergence.
standard math Krein-Milman theorem.
Used in Theorem 1.2 to extend equality on extreme points of the dual ball to all continuous linear functionals.
domain assumption Explicit formulas for all metric functionals on infinite-dimensional ell_p spaces [Gut19b, Theorem 3.6].
Used in Proposition 3.12 and Examples 3.16-3.18. This is the first author's earlier published work, cited rather than reproved.
domain assumption Point evaluations and their negatives are metric functionals on C[0,1] [Wal18, Theorem 5.1].
Used in Theorem 1.4 to rule out unbounded d-weakly convergent sequences in C[0,1].

how reviews work

0 comments

Cite this review

Pith. "Pith review of Metric functionals and weak convergence." pith.science (2026). https://pith.science/paper/6MIQO5G6

@misc{pith2026250604154,
  author       = {Pith},
  title        = {Pith review of: Metric functionals and weak convergence},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/6MIQO5G6}},
  note         = {Machine review of arXiv:2506.04154}
}

read the original abstract

We introduce a notion of weak convergence in arbitrary metric spaces. Metric functionals are key in our analysis: weak convergence of sequences in a given metric space is tested against all the metric functionals defined on said space. When restricted to bounded sequences in normed linear spaces, we prove that our notion of weak convergence agrees with the standard one.

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

25 extracted references · 25 canonical work pages

[1]

Deep limits and a cut-off phenomenon for neural networks

Benny Avelin and Anders Karlsson. Deep limits and a cut-off phenomenon for neural networks. J. Mach. Learn. Res. , 23(1), January 2022

work page 2022
[2]

Old and new challenges in Hadamard spaces

Miroslav Ba c \'a k. Old and new challenges in Hadamard spaces. Jpn. J. Math. (3) , 18(2):115--168, 2023

work page 2023
[3]

Introduction to Banach spaces and their geometry , volume 68 of North-Holland Math

Bernard Beauzamy. Introduction to Banach spaces and their geometry , volume 68 of North-Holland Math. Stud. Elsevier, Amsterdam, 1982

work page 1982
[4]

Horofunctions and metric compactification of noncompact Hermitian symmetric spaces

Cho-Ho Chu, Mar \' a Cueto-Avellaneda, and Bas Lemmens. Horofunctions and metric compactification of noncompact Hermitian symmetric spaces. Ann. Mat. Pura Appl. (4) , 203(4):1703--1751, 2024

work page 2024
[5]

Isabel Garrido, Jesus A Jaramillo, and Sebastian Tapia-Garc \'i a

Aris Daniilidis, M. Isabel Garrido, Jesus A Jaramillo, and Sebastian Tapia-Garc \'i a. Horofunction extension and metric compactifications , November 2024. Preprint hal-04786684

work page 2024
[6]

Convex geodesic bicombings and hyperbolicity

Dominic Descombes and Urs Lang. Convex geodesic bicombings and hyperbolicity. Geom. Dedicata , 177:367--384, 2015

work page 2015
[7]

CAT $(k)$ -spaces, weak convergence and fixed points

Rafa Esp \' nola and Aurora Fern \'a ndez-Le \'o n. CAT $(k)$ -spaces, weak convergence and fixed points. J. Math. Anal. Appl. , 353(1):410--427, 2009

work page 2009
[8]

Subadditive and multiplicative ergodic theorems

S \'e bastien Gou \"e zel and Anders Karlsson. Subadditive and multiplicative ergodic theorems. J. Eur. Math. Soc. (JEMS) , 22(6):1893--1915, 2020

work page 1915

Show all 25 references

[9]

Guti \'e rrez and Anders Karlsson

Armando W. Guti \'e rrez and Anders Karlsson. Comments on the cosmic convergence of nonexpansive maps. J. Fixed Point Theory Appl. , 23(4):10, 2021. Id/No 59

2021
[10]

Guti \'e rrez and Olavi Nevanlinna

Armando W. Guti \'e rrez and Olavi Nevanlinna. Subinvariant metric functionals for nonexpansive mappings. Preprint arXiv :2407.04234 [math. FA ] (2024), 2024

2024 arXiv
[11]

Guti \'e rrez

Armando W. Guti \'e rrez. Metric compactification of Banach spaces . Doctoral thesis, Aalto University, 2019

2019
[12]

Guti \'e rrez

Armando W. Guti \'e rrez. On the metric compactification of infinite-dimensional $ _p$ spaces. Can. Math. Bull. , 62(3):491--507, 2019

2019
[13]

A maximin characterisation of the escape rate of non-expansive mappings in metrically convex spaces

St \'e phane Gaubert and Guillaume Vigeral. A maximin characterisation of the escape rate of non-expansive mappings in metrically convex spaces. Math. Proc. Camb. Philos. Soc. , 152(2):341--363, 2012

2012
[14]

Hahn- Banach for metric functionals and horofunctions

Anders Karlsson. Hahn- Banach for metric functionals and horofunctions. J. Funct. Anal. , 281(2):17, 2021. Id/No 109030

2021
[15]

Elements of a metric spectral theory

Anders Karlsson. Elements of a metric spectral theory. In Dynamics, geometry, number theory. The impact of Margulis on modern mathematics , pages 276--300. Chicago, IL: The University of Chicago Press, 2022

2022
[16]

A metric fixed point theorem and some of its applications

Anders Karlsson. A metric fixed point theorem and some of its applications. Geom. Funct. Anal. , 34(2):486--511, 2024

2024
[17]

W. A. Kirk and B. Panyanak. A concept of convergence in geodesic spaces. Nonlinear Anal., Theory Methods Appl., Ser. A, Theory Methods , 68(12):3689--3696, 2008

2008
[18]

Remarks on some fixed point theorems

Teck-Cheong Lim. Remarks on some fixed point theorems. Proc. Am. Math. Soc. , 60:179--182, 1977

1977
[19]

Nonexpansive maps with surjective displacement

Brian Lins. Nonexpansive maps with surjective displacement. J. Fixed Point Theory Appl. , 24(1):28, 2022. Id/No 3

2022
[20]

Detecting fixed points of nonexpansive maps by illuminating the unit ball

Bas Lemmens, Brian Lins, and Roger Nussbaum. Detecting fixed points of nonexpansive maps by illuminating the unit ball. Isr. J. Math. , 224:231--262, 2018

2018
[21]

Weak topology on CAT (0) spaces

Alexander Lytchak and Anton Petrunin. Weak topology on CAT (0) spaces. Isr. J. Math. , 255(2):763--781, 2023

2023
[22]

Lipschitz- Free Spaces : A Topometric Approach and Group Actions

Michael Megrelishvili. Lipschitz- Free Spaces : A Topometric Approach and Group Actions . Preprint arXiv :2408.15208 [math. FA ] (2024), 2024

2024
[23]

Weak convergence of bounded sequences

John Rainwater. Weak convergence of bounded sequences. Proc. Am. Math. Soc. , 14:999, 1963

1963
[24]

A convexity in metric space and nonexpansive mappings

Wataru Takahashi. A convexity in metric space and nonexpansive mappings. I . K \=o dai Math. Semin. Rep. , 22:142--149, 1970

1970
[25]

Hilbert and Thompson geometries isometric to infinite-dimensional Banach spaces

Cormac Walsh. Hilbert and Thompson geometries isometric to infinite-dimensional Banach spaces. Ann. Inst. Fourier , 68(5):1831--1877, 2018

2018

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.

[1] [1]

Deep limits and a cut-off phenomenon for neural networks

Benny Avelin and Anders Karlsson. Deep limits and a cut-off phenomenon for neural networks. J. Mach. Learn. Res. , 23(1), January 2022

work page 2022

[2] [2]

Old and new challenges in Hadamard spaces

Miroslav Ba c \'a k. Old and new challenges in Hadamard spaces. Jpn. J. Math. (3) , 18(2):115--168, 2023

work page 2023

[3] [3]

Introduction to Banach spaces and their geometry , volume 68 of North-Holland Math

Bernard Beauzamy. Introduction to Banach spaces and their geometry , volume 68 of North-Holland Math. Stud. Elsevier, Amsterdam, 1982

work page 1982

[4] [4]

Horofunctions and metric compactification of noncompact Hermitian symmetric spaces

Cho-Ho Chu, Mar \' a Cueto-Avellaneda, and Bas Lemmens. Horofunctions and metric compactification of noncompact Hermitian symmetric spaces. Ann. Mat. Pura Appl. (4) , 203(4):1703--1751, 2024

work page 2024

[5] [5]

Isabel Garrido, Jesus A Jaramillo, and Sebastian Tapia-Garc \'i a

Aris Daniilidis, M. Isabel Garrido, Jesus A Jaramillo, and Sebastian Tapia-Garc \'i a. Horofunction extension and metric compactifications , November 2024. Preprint hal-04786684

work page 2024

[6] [6]

Convex geodesic bicombings and hyperbolicity

Dominic Descombes and Urs Lang. Convex geodesic bicombings and hyperbolicity. Geom. Dedicata , 177:367--384, 2015

work page 2015

[7] [7]

CAT \((k)\) -spaces, weak convergence and fixed points

Rafa Esp \' nola and Aurora Fern \'a ndez-Le \'o n. CAT \((k)\) -spaces, weak convergence and fixed points. J. Math. Anal. Appl. , 353(1):410--427, 2009

work page 2009

[8] [8]

Subadditive and multiplicative ergodic theorems

S \'e bastien Gou \"e zel and Anders Karlsson. Subadditive and multiplicative ergodic theorems. J. Eur. Math. Soc. (JEMS) , 22(6):1893--1915, 2020

work page 1915