Invariant Sublinear Expectations

Yongsheng Song

REVIEW 2 major objections 5 minor 10 references

Invariant Sublinear Expectations

T0 review · 2 major / 5 minor · reviewed 2026-08-12 · deepseek-v4-flash

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

Pith's one-line read For a continuous invariant sublinear expectation that is strongly T-ergodic, the family of periodic components stabilizes: there is some d such that Θ^(l) ⊆ Θ^(d) for every l, and with a periodic decomposition the whole expectation has a…

desk verdict The periodic decomposition is a real contribution, but Theorem 4.6's proof as written does not establish the claimed contradiction, so the finite-period corollary rests on an unstated argument. read the letter →

arxiv 2411.14177 v1 pith:RJAIYNVK submitted 2024-11-21 math.PR

Yongsheng Song This is my paper · ORCID

classification math.PR MSC 28A1228D0537A30

keywords invariantsublinearexpectationperiodicdecompositionstrongergodicityupperprobabilityergodictheoremextremepointsBanach-MazurlimitT-invariant

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 establishes a structural decomposition for T-invariant sublinear expectations, i.e. upper expectations defined as suprema over a set of probability measures. Each component E^(d), built from time averages along T^d, is shown to be periodic: there is a finite p_d with E^(d)[f] = E^(d)[f∘$T^{{p_d}}$] for all f. The main result is that a continuous, strongly T-ergodic invariant sublinear expectation has a stabilizing top component, so whenever the expectation admits a periodic decomposition, it has a finite period p_E. From this, the paper derives an ergodic theorem: along the p_E-step dynamics, the time means converge to the upper expectation E[f] almost surely under at least one probability from the defining set Θ. This answers a natural question left open by earlier ergodic-capacity results, where time means were only bounded by E[f] rather than attaining it.

What carries the argument

The workhorse is the family E^(d)[f] = lim_{n→∞} (1/n) E[∑_{k=0}^{n-1} f∘$T^{{kd}}$] of subadditive time averages, with Θ^(d) the set of T^d-invariant probabilities dominated by E. These components satisfy a gcd lattice law — if Θ^(l) ⊆ Θ^(d) then Θ^(l) = Θ^(gcd(l,d)) (Lemma 3.3) — which forces the periods p_d to divide each other along divisibility chains. The stabilization proof uses the mutual singularity of extreme points: if ext Θ^(d) ⊆ co(ext Θ^(l)), then the supports of the extreme measures form disjoint sets, and a strictly increasing chain of cardinalities would produce infinitely many disjoint measurable sets of positive upper probability, contradicting continuity.

What would settle it

Construct a continuous, strongly T-ergodic invariant sublinear expectation on a countable state space with a periodic decomposition whose component periods p_d grow unboundedly along a divisibility chain; Theorem 4.6 and Corollary 4.8 predict this is impossible, so finding such an example would refute the central theorem.

Watch

Extended reading notes

Core claim

The central discovery is the stabilization theorem (Theorem 4.6): given a continuous sublinear expectation E = sup_{P∈Θ} E_P on B_b(Ω) that is strongly T-ergodic, there exists a single d ∈ N such that Θ^(l) ⊆ Θ^(d) for every l ∈ N. Combined with the periodic decomposition E[f] = sup_{d∈N} E^(d)[f], this forces E itself to be periodic with a finite period p_E, and every component Θ^(d) to be the convex hull of finitely many $T^{{p_d}}$-ergodic probabilities. The paper then proves the limit of the p_E-step time means attains the upper expectation: for each bounded f there is a P_f ∈ Θ with lim_{n→∞} (1/n)∑_{k=0}^{n-1} f∘$T^{{kp_E}}$ = E[f], P_f-a.s.

Load-bearing premise

The proof that the component chain stabilizes assumes that the extreme probabilities chosen at successive levels are mutually singular, so their supports form countably many disjoint measurable sets; if this singular-support separation fails, the contradiction with continuity no longer follows.

Editorial extensions

If this is right

If E has a periodic decomposition and is continuous strongly T-ergodic, then E = E^(p_E) and Θ^(d) = Θ^(gcd(d,p_E)); the whole structure is determined by finitely many T^{p_E}-ergodic probabilities.
For each bounded f, the p_E-step ergodic average converges P_f-a.s. to E[f] for some P_f ∈ Θ, so the upper expectation is exactly attainable as a long-run time mean.
For i.i.d. sequences under a regular sublinear expectation, the component periods satisfy p_d = d whenever E is not linear, recovering the known strong law of large numbers as a special case.
If Ω is countable, every continuous T-invariant sublinear expectation has a periodic decomposition (Example 5.2), so the finite-period and attainment results apply whenever strong ergodicity holds.

Reading between the lines

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

The stabilization theorem suggests a general dichotomy for continuous invariant sublinear expectations: either the component chain Θ^(1) ⊆ Θ^(2) ⊆ ... stabilizes at a finite level, or the upper probability fails continuity by admitting infinitely many singular extreme measures.
One testable extension is whether the attainment result can be strengthened to simultaneous attainment: a single P ∈ Θ making the p_E-step time means converge to E[f] for every f in a dense subspace, rather than a separate P_f for each f.
The gcd lattice structure (Lemma 3.3) points to an analogous period-group structure for continuous-time flows; under an R-action, strongly ergodic invariant sublinear expectations might have a group of periods either {0} or a discrete lattice.
A concrete computational check would be to compute component periods for finite-state Markov chains with a sublinear expectation and see whether the stabilized period p_E matches the gcd of the periods of the ergodic classes, as the theorems predict.

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

Here's my take. The paper has one genuinely good idea and one genuine hole. The good idea is the periodic decomposition E = sup_d E^(d) in Section 3: each component is periodic, and the divisibility properties of periods (Theorem 3.6) are clean and convincing. Example 5.3 is also worth having—it gives a continuous, strongly ergodic ISE on the circle with no periodic decomposition, which proves the decomposition condition in Corollary 4.8 isn't vacuous.

The hole is Theorem 4.6. The proof asserts that a strictly increasing chain of card(ext Θ^(n_k)) contradicts continuity by Remark 2.1. To invoke Remark 2.1 you need disjoint sets A_k with V(A_k) bounded below by a positive constant. The manuscript never constructs them. And the natural construction doesn't work: a T-ergodic measure can split into two T^2-ergodic components that both overlap the original measure, so a 'new' extreme point at a later level need not be singular to the old ones. The paper gives no replacement argument. This matters because Corollary 4.8 and the advertised finite-period result sit directly on this theorem. The abstract overstates further by omitting the periodic-decomposition assumption from the finite-period claim.

I don't think the paper is beyond repair. The theorem is likely true—continuity probably does rule out infinite splitting—and the rest of the paper is solid. The gap is a missing proof step, not a wrong idea. Sections 2 and 3 are written carefully. The use of Sheng-Song [8] as a black box is normal cumulative work, not circular.

Recommendation: send it to a referee, but ask specifically for the proof of Theorem 4.6 to be completed and the abstract corrected. The paper deserves referee time; it just should not be accepted as is.

Referee Report

2 major / 5 minor

Summary. The paper studies T-invariant sublinear expectations E = sup_{P in Θ} E_P. It first decomposes E into components E^(d) obtained by Cesàro averaging along T^d, proves that each E^(d) is periodic with period p_d dividing d, and derives divisibility relations among the periods. For continuous sublinear expectations on (Ω, B_b(Ω)), it introduces strong T-ergodicity and proves Theorem 4.6, which asserts that the chain Θ^(d) stabilizes in the sense that Θ^(l) ⊂ Θ^(d) for some d and every l. As a consequence, Corollary 4.8 shows that a strongly T-ergodic continuous sublinear expectation with a periodic decomposition has a finite period p_E, and that the p_E-step time means are bounded above and below by E[f] and -E[-f] for every P in Θ, with some P_f attaining E[f] a.s. The paper ends with three examples: an i.i.d. sequence on R^∞, a countable state-space example, and a rotation example without a periodic decomposition.

Significance. If the proof gap identified below is repaired, the paper gives a clean structural result: under strong ergodicity and a periodic decomposition, the family of invariant sublinear components collapses to a single finite-period component, and an ergodic theorem attains the upper expectation. This is a natural and worthwhile extension of the Sheng–Song characterization of continuous ergodic capacities, and the divisibility lemmas in Section 3 are useful tools. The paper is honest about its use of [8] as a black box, and the new arguments are not circular. The examples are concrete and help delineate the boundary of the class of sublinear expectations admitting periodic decompositions.

major comments (2)

[Section 4, Theorem 4.6] The final step of the proof is incomplete. The strict increase card(ext Θ^(n_k)) < card(ext Θ^(n_{k+1})) for all k does not by itself contradict continuity. To invoke Remark 2.1, one must exhibit disjoint measurable sets A_k with V(A_k) ≥ ε > 0 for all k. This requires choosing P_k ∈ ext Θ^(n_k) \ ext Θ^(n_{k-1}), proving that P_k is mutually singular with all previously chosen P_j, and then applying the standard lemma that countably many pairwise singular probability measures admit pairwise disjoint full-measure sets. The manuscript states none of these steps; without them the contradiction is an assertion rather than a proof. The missing lemma is standard and the gap appears repairable, but because Corollary 4.8's finite-period conclusion rests on this stabilization, the construction should be written out explicitly.
[Section 4, Theorem 4.6, extreme-point paragraph] The assertion that the sets A_j = {i : α_j^i > 0} are disjoint because the probabilities in ext Θ(k), k = d, l, are mutually singular is too terse and, read literally, asks for cross-singularity between ext Θ(d) and ext Θ(l), which is not generally true: a T^{p_d}-ergodic measure and a T^{p_l}-ergodic measure need not be mutually singular. The needed argument is that the Q_i are pairwise singular, and that if two P_j shared a Q_i with positive coefficient, then both P_j and P_j' would assign positive mass to every set of Q_i-measure one, contradicting their mutual singularity; hence the A_j are disjoint. Moreover, because each P_j is extreme, each A_j is a singleton, so ext Θ(d) ⊂ ext Θ(l). This nested-extreme-point fact is also needed to make the sequence of new extreme measures in the final contradiction mutually singular with all previous ones. Please supply this argument.

minor comments (5)

[Throughout] There are numerous typos and misspellings: "Genenrally", "Theroem", "probablity", "euqal", "domonated", "contradition", "fucntion", "divider" for gcd, "invairant", "consenquently". A careful proofreading pass is needed.
[Corollary 4.8, Eq. (4.2)] Equation (4.2) writes the time mean with a plain "lim" even though the inequality is between -E[-f] and E[f]. Since P ∈ Θ = Θ^(p_E), every P is T^{p_E}-invariant and the usual pointwise ergodic theorem gives P-a.s. convergence, so the notation is defensible, but the reason should be stated; alternatively, replace "lim" by liminf/limsup and note that the limit exists a.s.
[Proposition 4.2] The proof uses a partition {E_k}_{k=1}^m ⊂ I with P_i(E_i) = 1. This follows from the standard lemma that finitely many pairwise singular probability measures admit pairwise disjoint sets of full measure; please state or cite this fact so the argument is self-contained.
[Example 5.3] The phrase "root of unitary" should read "root of unity". Also, the proof of continuity of V relies on the uniform integrability of the translates φ ∘ T^{-k}; a brief justification would help the reader.
[References] Reference [1] lists the second author as "Boeder"; the correct name is "Border".

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the imported Theorem 2.3 from [8] is published external support, and the new derivation does not assume its own conclusion.

full rationale

The derivation chain is self-contained relative to its stated assumptions. Proposition 3.2 defines E^(d) directly from E and proves periodicity; Theorem 3.6 and Corollary 3.4 are internal algebraic consequences. Theorem 4.6 imports from Sheng–Song [8] (same author as the present paper) the fact that a continuous ergodic upper probability has finitely many ergodic extreme points and is their convex hull. This is a published external theorem used as a black box, not an assumption of the present theorem; citing it is cumulative research, not circular reasoning. The remaining steps of Theorem 4.6 do not redefine a fitted quantity as a prediction: the increasing-cardinality argument is meant to contradict continuity via disjoint sets with V(A_n) >= epsilon. The manuscript omits the standard lemma that countably many pairwise singular measures admit pairwise disjoint full-measure sets, and it does not explicitly construct cross-level singular extreme measures; this is a rigor gap in the written proof, not a circular reduction, because the theorem's conclusion is not assumed in its hypotheses. Corollary 4.8's ergodic limit (4.3) follows from Birkhoff's ergodic theorem once the finite convex-hull structure is granted. No equation in the paper is shown to be equal to its own input by construction, and no fitted parameter is relabeled as a prediction. Therefore no significant circularity is present.

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

The central results rely on the stated regularity assumptions (continuity, strong ergodicity) and on classical ergodic theory plus the authors' earlier theorem [8]. No numerical parameters are fitted and no new entities are postulated.

assumptions (6)

domain assumption Θ = {P ∈ M(Ω) | E_P[f] ≤ E[f] for f ∈ H}
Section 2 defines the representing set of probabilities as all probabilities dominated by E. This duality is standard in sublinear expectation theory and is used throughout to translate between the expectation and its set of measures.
domain assumption Continuity of the upper probability V(A) = sup_{P∈Θ} P(A) from above
Section 2, used in Section 4. Crucially invoked in Theorem 4.6 via Remark 2.1 to rule out infinitely many disjoint sets with uniformly positive V. Without continuity the main finiteness theorem may fail.
ad hoc to paper Strong T-ergodicity (Definition 4.5): for each d ∈ N, E^{(d)} is strongly ergodic
Introduced specifically for this paper. It is a strong hypothesis and is the key assumption of Theorem 4.6 and Corollary 4.8.
standard math Theorem 2.3 from Sheng-Song [8] on continuous ergodic capacities
Prior published result by the same research group (PAMS 2024), used in Propositions 4.2 and 4.4 to characterize invariant probabilities as convex hulls of ergodic ones. Treated as independent support.
standard math Existence of Banach-Mazur limits
Appendix 6, used in Lemma 4.1 and Example 5.2 to construct invariant probabilities from time averages.
standard math Birkhoff's ergodic theorem, mutual singularity of distinct ergodic measures, Poincaré recurrence
Classical ergodic theory results invoked in Examples 5.1 and 5.2 and Proposition 4.4.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Invariant Sublinear Expectations." pith.science (2026). https://pith.science/paper/RJAIYNVK

@misc{pith2026241114177,
  author       = {Pith},
  title        = {Pith review of: Invariant Sublinear Expectations},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/RJAIYNVK}},
  note         = {Machine review of arXiv:2411.14177}
}

abstract

We first give a decomposition for a $T$-invariant sublinear expectation $\mathbb{E}=\sup_{P\in\Theta}\mathrm{E}_P$, and show that each component $\mathbb{E}^{(d)}=\sup_{P\in\Theta^{(d)}}\mathrm{E}_P$ of the decomposition has a finite period $p_d\in\mathbb{N}$, i.e., \[\mathbb{E}^{(d)}\left[f-f\circ T^{p_d}\right]=0, \quad f\in\mathcal{H}.\] Then we prove that a continuous invariant sublinear expectation that is strongly ergodic has a finite period $p_{\mathbb{E}}$, and each component $\Theta^{(d)}$ of its periodic decomposition is the convex hull of a finite set of $T^{p_d}$-ergodic probabilities. As an application of the characterization, we prove an ergodicity result which shows that the limit of the $p_{\mathbb{E}}$-step time means achieves the upper expectation.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

10 extracted references · 9 canonical work pages

[8]

Sheng and Y

Y. Sheng and Y. Song: Continuous ergodic capacities. Proceedings of the American Math- ematical Society, 152(9), (2024), 3893–3898

work page 2024
[1]

Aliprantis, Boeder.: Inﬁnite Dimensional Analysis, 3r d ed., Springer, Berlin, 2006

work page 2006
[2]

Cerreia–Vioglio, F

S. Cerreia–Vioglio, F. Maccheroni and M. Marinacci.: Er godic theorems for lower proba- bilities. Proceedings of the American Mathematical Society, 144(8), (2016), 3381-3396

work page 2016
[3]

Chen: Strong laws of large numbers for sublinear expec tations

Z. Chen: Strong laws of large numbers for sublinear expec tations. Sci China Math. 59,(2016), 945-954

work page 2016
[4]

Z. Chen, W. Huang, P. Wu,: Extension of The Strong Law of La rge Numbers for Capacities. Mathematical Control And Related Fields. 9(1) (2019), 175-190

work page 2019
[5]

C. Feng, P. Wu, and H. Zhao: Ergodicity of invariant capac ities. Stochastic Process. Appl. 130(8)(2020), 5037-5059

work page 2020
[6]

Feng, and H

C. Feng, and H. Zhao: Ergodicity of sublinear Markovian s emigroups. SIAM J. Math. Anal. 53(5) (2021), 5646-5681

work page 2021
[7]

Peng: A New Central Limit Theorem under Sublinear Expe ctations

S. Peng: A New Central Limit Theorem under Sublinear Expe ctations. arXiv:0803.2656v1 [math.PR]

arXiv

Show all 10 references

[9]

Song: A strong law of large numbers under sublinear exp ectations

Y. Song: A strong law of large numbers under sublinear exp ectations. Probability, Uncer- tainty and Quantitative Risk. 8(3), (2023), 333-350

2023
[10]

Zhang: The suﬃcient and necessary conditions of the s trong law of large numbers under the sub-linear expectations

L. Zhang: The suﬃcient and necessary conditions of the s trong law of large numbers under the sub-linear expectations. Acta. Math. Sin.-English Ser. 39, (2023), 2283–2315. 12

2023

Pith tools

Reviewed August 12, 2026 · model on record in the stance chip above.

[1] [8]

Sheng and Y

Y. Sheng and Y. Song: Continuous ergodic capacities. Proceedings of the American Math- ematical Society, 152(9), (2024), 3893–3898

work page 2024

[2] [1]

Aliprantis, Boeder.: Inﬁnite Dimensional Analysis, 3r d ed., Springer, Berlin, 2006

work page 2006

[3] [2]

Cerreia–Vioglio, F

S. Cerreia–Vioglio, F. Maccheroni and M. Marinacci.: Er godic theorems for lower proba- bilities. Proceedings of the American Mathematical Society, 144(8), (2016), 3381-3396

work page 2016

[4] [3]

Chen: Strong laws of large numbers for sublinear expec tations

Z. Chen: Strong laws of large numbers for sublinear expec tations. Sci China Math. 59,(2016), 945-954

work page 2016

[5] [4]

Z. Chen, W. Huang, P. Wu,: Extension of The Strong Law of La rge Numbers for Capacities. Mathematical Control And Related Fields. 9(1) (2019), 175-190

work page 2019

[6] [5]

C. Feng, P. Wu, and H. Zhao: Ergodicity of invariant capac ities. Stochastic Process. Appl. 130(8)(2020), 5037-5059

work page 2020

[7] [6]

Feng, and H

C. Feng, and H. Zhao: Ergodicity of sublinear Markovian s emigroups. SIAM J. Math. Anal. 53(5) (2021), 5646-5681

work page 2021

[8] [7]

Peng: A New Central Limit Theorem under Sublinear Expe ctations

S. Peng: A New Central Limit Theorem under Sublinear Expe ctations. arXiv:0803.2656v1 [math.PR]

arXiv