{"id":"4d4a8d6e-b14e-4d51-a242-ed98eaf57789","arxiv_id":"1908.09012","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"The Hajek-Renyi-Chow maximal inequality, submartingale convergence, and Chow's strong law of large numbers are extended to Lp-type Riesz spaces with conditional expectation operators.","lead":"This mathematics paper proves a version of the Hajek-Renyi-Chow maximal inequality for submartingales in Riesz spaces with conditional expectation operators, and derives a submartingale convergence theorem and a strong law of large numbers. A smart generalist might read it because it extends classical martingale tools from ordinary probability to vector-valued lattices, where stochastic processes are defined without a sample space.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Equation (6.7) invokes Jensen in the wrong direction, so Theorem 6.3 is not proved as printed; the error is localized and repairable.","rationale":"The reader's rationale correctly identifies equation (6.7) as a false Jensen bound, and I agree that Theorem 6.3 is not proved as printed. However, the reader's stated weakest_assumption is the dependence on [10, Corollary 4.5] for powers of submartingales, while my strongest concern is the concrete invalid inequality (6.7) in the proof of the p>2 strong law. These are related but not identical, so my agreement is partial. The concern is load-bearing because the abstract advertises a strong law of large numbers in Riesz spaces for all 1 < p < infinity, and the p>2 case rests on (6.7). The error is localized and there is a direct replacement inequality in the f-algebra, so a rejection is not warranted; the appropriate disposition remains conditional on correcting or replacing (6.7).","tokens_in":10153,"tokens_out":30336,"duration_ms":284380,"concrete_test":"Independently verify (6.7) in the scalar setting T1 = expectation, p = 4, with Y1,Y2 independent and P(Y = ±1) = P(Y = ±2) = 1/4: the left side equals E[(Y1^2+Y2^2)^2] = 2 E Y^4 + 2(E Y^2)^2 = 29.5, while the right side equals (E Y1^2 + E Y2^2)^2 = 25, so (6.7) fails. Then check whether the bound (6.9) follows from (6.6) using the direct pointwise inequality (sum |Y_i|^2)^{p/2} <= n^{p/2-1} sum |Y_i|^p in the f-algebra; if yes, Theorem 6.3 is salvageable with a localized correction, and the reader's conditional verdict should stand.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The p>2 strong law in Theorem 6.3 depends on bounding Z_n. After the Burkholder lower bound (6.6), the proof invokes Jensen's inequality of [10] to claim T1[(sum_{i=1}^n |Y_i|^2)^{p/2}] <= (sum_{i=1}^n T1|Y_i|^2)^{p/2} at equation (6.7). For p>2, the function s -> s^{p/2} is convex, so the Jensen inequality [10, Theorem 4.4] gives the reverse order: T1[(sum |Y_i|^2)^{p/2}] >= (T1[sum |Y_i|^2])^{p/2} = (sum T1|Y_i|^2)^{p/2}. Thus (6.7) is not a consequence of Jensen and is false in general. For example, in scalar probability with T1 = expectation, p = 4, and Y1,Y2 independent with P(Y = ±1) = P(Y = ±2) = 1/4, E[(Y1^2+Y2^2)^2] = 2 E Y^4 + 2(E Y^2)^2 = 17 + 12.5 = 29.5 while (E Y1^2 + E Y2^2)^2 = 25. Since (6.8) and (6.9) are derived from (6.7), the boundedness of Z_n, and hence Theorem 6.3, is not established as printed. The maximal inequality Theorem 4.2 and submartingale convergence Theorem 5.1 are not affected. A direct f-algebra inequality (sum |Y_i|^2)^{p/2} <= n^{p/2-1} sum |Y_i|^p would replace (6.7)+(6.8) and appears to recover (6.9), so the correct disposition is conditional rather than rejection.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper develops Riesz-space analogues of classical martingale inequalities. Theorem 4.2 establishes a Hajek-Renyi-Chow maximal inequality for submartingales in L1(T) using band projections onto the sets where the weighted process exceeds g. Theorem 5.1 gives an upcrossing-free weighted convergence theorem for non-negative submartingales in Lp(T). Section 6 applies these results to martingale difference sequences, obtaining Chow-type strong laws for 1 <= p <= 2 (Theorem 6.1 and Corollary 6.2) and for p > 2 (Theorem 6.3). Section 3 contains a Riesz-space Clarkson inequality for 1 <= p <= 2 and quotes a Riesz-space Burkholder inequality. The main technical novelty is the band-projection proof of the maximal inequality, which is then used as the engine for the convergence and strong-law theorems.","tokens_in":1757,"tokens_out":3111,"duration_ms":351761,"significance":"If the cited machinery from [10] is accepted, Theorem 4.2 is a genuine contribution: it is a clean, parameter-free transfer of the Hajek-Renyi-Chow maximal inequality to Riesz spaces, and it supports the submartingale convergence theorem and the strong laws that follow. The Clarkson inequality for Lp(T) in Section 3 is a useful independent result. The paper is written for specialists in Riesz-space stochastic processes and fits the journal's scope. The main limitation is that the proof of Theorem 6.3 as printed contains a genuine error in the Jensen step and a second unjustified bounding step. Both are localized and repairable, and the theorem itself appears to be true, but the p > 2 strong law is not proved until those repairs are made.","major_comments":[{"comment":"For p > 2 the map s -> s^{p/2} is convex, so Jensen's inequality in Riesz spaces ([10, Theorem 4.4]) gives T1[(sum_{i=1}^n |Y_i|^2)^{p/2}] >= (sum_{i=1}^n T1|Y_i|^2)^{p/2}, the reverse of the claimed (6.7). The displayed inequality is false in general: in scalar probability with p = 4 and independent Y1, Y2 having P(Y = ±1) = P(Y = ±2) = 1/4, E[(Y1^2+Y2^2)^2] = 29.5 > (E Y1^2 + E Y2^2)^2 = 25. Since (6.8) and (6.9) are derived from (6.7), the boundedness of Z_n, and hence the proof of Theorem 6.3, is not established as printed. The error is localized and repairable: the f-algebra inequality (sum_{i=1}^n |Y_i|^2)^{p/2} <= n^{p/2-1} sum_{i=1}^n |Y_i|^p yields (6.9) after applying T1, so the theorem's conclusion is recoverable.","section":"§6, Theorem 6.3 proof, Eq. (6.7)"},{"comment":"The step after (6.9) says that because T1|Xn|^p/n^p is order bounded by h, one has sum_{i=2}^{n-1}(1/i^p - 1/(i+1)^p)T1(|Xi|^p) <= p sum_{i=1}^{infty} i^{-p/2} h. This does not follow: the stated bound on T1|Xn|^p/n^p gives T1|Xi|^p <= i^p h, so the left side is at most p h sum_{i=2}^{n-1} 1/i, which is not bounded. A correct argument is available: with a_j = T1|Y_j|^p and R_i = i^{-(p/2+1)} sum_{j=1}^i a_j, (6.9) and the corrected (6.7) give T1|Xi|^p/i^{p+1} <= C R_i/i, and sum_i R_i/i converges because sum_i i^{-(p/2+2)} sum_{j=1}^i a_j = sum_j a_j sum_{i=j}^{infty} i^{-(p/2+2)} <= C sum_j a_j/j^{p/2+1}. Thus the proof needs revision at this step; it is another localized gap in Theorem 6.3.","section":"§6, Theorem 6.3 proof after (6.9)"}],"minor_comments":[{"comment":"The sentence immediately after (4.8) states QiTi(Y_{i+1}^+ - Y_i^+) <= Ti(Y_{i+1}^+ - Y_i^+), but the displayed estimate needs T1Qi <= T1 applied to the positive differences; please rephrase, since the conclusion is correct but the written composition is confusing.","section":"§4, proof of Theorem 4.2"},{"comment":"After (6.5), the order-convergent sum should be sum T1(|X_{i+1}|^p - |X_i|^p)/a_{i+1}^p rather than the displayed sum T_i(|X_{i+1}|^p - |X_i|^p)/a_{i+1}^p, because Theorem 5.1 is stated with T1; the two versions agree after applying T1 = T1T_i.","section":"§6, proof of Theorem 6.1"},{"comment":"The paper defines strict positivity of T but the blanket assumption after the extension to L1(T) does not state that T (and hence T1) is strictly positive; Theorem 5.1 relies on strict positivity of T1, so add this hypothesis explicitly.","section":"§1"},{"comment":"There is a typo in the statement: 'for all ∈ N' should be 'for all n ∈ N'.","section":"§3, Theorem 3.3"},{"comment":"The notation Lp(T1) in Corollary 6.2 and Theorem 6.3 is not introduced; earlier definitions use Lp(T). Please make the initial conditional expectation T1 explicit or use a consistent notation.","section":"§6"}],"recommendation":"major_revision","confidential_remarks":"The stress-test concern about Eq. (6.7) is valid and lands. There is an additional gap in the same proof at the bounding of the telescoping sum. Both issues are localized to Theorem 6.3 and do not affect Theorem 4.2 or Theorem 5.1. The theorem itself appears true and admits a fix; I expect a revised version to be publishable. The paper relies heavily on [10], but that is published independent work, so there is no circularity."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"One thing to know: Theorem 4.2, the Riesz-space Hajek-Renyi-Chow maximal inequality, is real and the proof is solid. The second thing: Theorem 6.3, the p>2 strong law, is not proved as printed because equation (6.7) applies Jensen in the wrong direction. The error is localized and easily repaired.\n\nThe paper does useful work. The maximal inequality for submartingales with a non-decreasing real sequence (a_i) and a band-projection product U_n is a natural extension, and the proof via telescoping plus the projection trick is clean. The Clarkson inequality for 1≤p≤2 is a nice Riesz-space variant, and the submartingale convergence theorem (Theorem 5.1) avoids upcrossing, which is a genuine method contribution. The authors also openly note that Theorem 5.1 follows from [14, Thm 3.5], which is the right kind of caveat; it doesn't undercut the new proof.\n\nThe soft spot is in Section 6. At (6.7), for p>2 the map s↦s^{p/2} is convex, so [10, Thm 4.4] gives T1[(∑|Y_i|^2)^{p/2}] ≥ (∑ T1|Y_i|^2)^{p/2}, not the reverse. The scalar counterexample with independent ±1,±2 variables shows the inequality as written is false. Since (6.8) and (6.9) depend on (6.7), the boundedness of Z_n is not established. This affects only Theorem 6.3; Theorem 6.1 and Corollary 6.2 don't use (6.7). The fix is straightforward: replace (6.7) with the f-algebra inequality (∑|Y_i|^2)^{p/2} ≤ n^{p/2-1}∑|Y_i|^p, then apply T1 and Hölder; you recover (6.9) with the same n-power, so the argument goes through. I'd call it a localized flaw, not a fundamental gap.\n\nMy bottom line: the paper deserves serious refereeing. The main inequality is a solid contribution to the Riesz-space martingale program, and the flaw in one corollary is the kind of thing a referee should catch. With the Section 6.3 repair, it's publishable in a specialist journal. If you want to see a clean band-projection argument, read Section 4; if you want to see a cautionary Jensen-direction error, look at (6.7).","headline":"The main HRC maximal inequality in Riesz spaces is solid and new, but Theorem 6.3 (p>2 strong law) contains a Jensen-direction error at equation (6.7) and is not proved as printed; the flaw is localized and easily repaired.","tokens_in":11083,"tokens_out":3027,"would_cite":false,"duration_ms":27523,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["46B40","60F15","60F25"],"pacs":[],"model":"deepseek-v4-flash","headline":"The paper proves a Riesz-space analogue of the Hájek–Rényi–Chow maximal inequality for submartingales and uses it to obtain a submartingale convergence theorem and Chow's strong law of large numbers.","keywords":["Riesz spaces","vector lattices","Hájek–Rényi–Chow maximal inequality","submartingale convergence","strong law of large numbers","conditional expectation operators","Clarkson's inequality"],"falsifier":"Construct a Dedekind complete Riesz space with weak order unit, a strictly positive conditional expectation $T$, and a non-negative submartingale $(X_n,T_n)$ in $L^p(T)$ for which $\\sum_{i=1}^\\infty T_1[(X_{i+1}^p-X_i^p)/a_{i+1}^p]$ converges in order but $X_n/a_n$ does not converge in order; one such example would refute Theorem 5.1. A more targeted check is to test the cited power-submartingale lemma in a concrete space such as $L^p(\\Omega,\\mathcal{F},\\mu)$ with conditional expectation $T$, since the convergence argument depends on that lemma.","tokens_in":50,"feed_emoji":"📈","tokens_out":18879,"duration_ms":272552,"temperature":0.7,"pith_summary":"Classical probability's Hájek–Rényi–Chow inequality bounds a submartingale's running maximum by its weighted increments. This paper proves the same kind of maximal inequality in Riesz spaces—vector lattices with a conditional expectation operator, where martingales and order convergence take the place of measure-theoretic expectations and almost-sure limits. The inequality is the load-bearing tool for two applications: a submartingale convergence theorem, in which a non-negative $L^p$ submartingale satisfies $X_n/a_n\\to 0$ in order whenever the weighted increments of $X_i^p$ are summable, and Chow's strong law of large numbers for martingale difference sequences in $L^p(T)$, $1<p<\\infty$. A Riesz-space version of Clarkson's inequality, $|x+y|^p+|x-y|^p\\le 2(|x|^p+|y|^p)$ for $1\\le p\\le 2$, is developed along the way.","feed_headline":"A classic maximal inequality is proved for Riesz-space submartingales","feed_subtitle":"It yields a submartingale convergence theorem and a strong law of large numbers in vector lattices.","key_machinery":"The load-bearing object is the family of band projections $P_{(g-Y_i/a_i)^+}$. In a Dedekind complete Riesz space with weak order unit, each positive element $h$ generates a band, and $P_h f=\\sup_n(f\\wedge n h)$ is the projection onto that band; $P_{(g-Y_i/a_i)^+}$ projects onto the band where the weighted submartingale has exceeded the level $g$ at time $i$, and the product $U_n$ projects onto the band of the running maximum. Lemma 4.1 telescopes the differences of these projections into the positive increments $Y_{i+1}^+-Y_i^+$, converting a statement about running maxima into one about sums of increments. Two supporting identities carry the applications: the Riesz-space Jensen machinery, by which $X_i^p$ is again a submartingale when $X_i$ is a non-negative submartingale in $L^p(T)$, and the Riesz-space Clarkson inequality $|x+y|^p+|x-y|^p\\le 2(|x|^p+|y|^p)$ for $1\\le p\\le 2$, which feeds the submartingale convergence theorem into the strong law.","core_discovery":"The central claim is Theorem 4.2: if $(Y_i,T_i)$ is a submartingale in $L^1(T)$, $(a_i)$ is a non-decreasing sequence of positive real weights, and $g\\in R(T_1)^+$, then $$T_1(I-U_n)g \\le \\frac{Y_1^+}{a_1}+\\sum_{i=1}^{n-1} T_1\\left[\\frac{Y_{i+1}^+-Y_i^+}{a_{i+1}}\\right],\\quad U_n=\\prod_{i=1}^n P_{(g-Y_i/a_i)^+}.$$ The left side measures the conditional expectation of the part of $g$ that remains above the running weighted maximum, while the right side is the weighted sum of positive increments of the submartingale. This is the Riesz-space form of the Hájek–Rényi–Chow maximal inequality; it specialises to Doob's maximal inequality and, through the power-submartingale route, yields an upcrossing-free proof that $X_n/a_n$ converges to zero in order under a weighted summability condition, plus Chow's strong law of large numbers for martingale difference sequences.","pith_inferences":["The same telescoping-of-band-projections argument is a natural template for convergence theorems in ordered vector spaces that lack an upcrossing theory; this is an extension the authors do not state.","The Riesz-space Clarkson inequality suggests that $L^p(T)$ for $1<p<2$ carries a lattice version of uniform convexity, which could open geometric questions about these spaces that the paper does not pursue.","One testable extension is to relax $g\\in R(T_1)^+$ in Theorem 4.2, since that assumption enters through the commutation of $T_1$ with the band projection $P_g$; if commutation holds more broadly, the maximal inequality would cover more general levels."],"forward_implications":["Doob's maximal inequality in Riesz spaces follows as a special case of Theorem 4.2.","For a non-negative $L^p$ submartingale with $\\sum_{i=1}^\\infty T_1[(X_{i+1}^p-X_i^p)/a_{i+1}^p]$ order convergent, the weighted process satisfies $X_n/a_n\\to 0$ in order, and the proof avoids upcrossing arguments.","For $1\\le p\\le 2$, a martingale difference sequence in $L^p(T)$ with $\\sum_{i=1}^\\infty T_1(|Y_i|^p/a_i^p)$ order convergent obeys $(1/a_n)\\sum_{i=1}^n Y_i\\to 0$ in order.","For $p>2$, the same strong law holds under the stronger moment condition $\\sum_{i=1}^\\infty T_1(|Y_i|^p/i^{1+p/2})<\\infty$ with $a_n=n$, and under a Hölder-type condition for general weights.","A Riesz-space Clarkson inequality, $|x+y|^p+|x-y|^p\\le 2(|x|^p+|y|^p)$ for $1\\le p\\le 2$, is established."],"supporting_citations":[{"why":"It supplies the Riesz-space Jensen inequality and the power-submartingale result (Corollary 4.5) needed to convert L^p submartingales into the processes used in Theorems 5.1 and 6.1.","marker":"[10]"},{"why":"It defines conditional expectation operators on Riesz spaces and proves that band projections commute with T1 for g in R(T1)+, a step used in the proof of Theorem 4.2.","marker":"[13]"},{"why":"It supplies the measure-theoretic Lemma 6.1.1 whose telescoping-series argument Lemma 4.1 generalises to vector lattices.","marker":"[6]"},{"why":"It constructs the L^p(T) spaces via functional calculus and gives the Hölder inequality for sums used in Corollary 6.2.","marker":"[3]"},{"why":"It provides the Riesz-space Burkholder inequalities used to bound T1|Xn|^p in the proof of the p>2 strong law.","marker":"[2]"},{"why":"It gives the classical martingale inequality and strong law whose Riesz-space analogue is the paper's target.","marker":"[4]"},{"why":"It gives the classical strong law for p>2 that Theorem 6.3 generalises.","marker":"[5]"},{"why":"It provides the upcrossing-based Riesz-space submartingale convergence theorem that the authors note could also prove Theorem 5.1.","marker":"[14]"}],"fun_headline_variants":["Hájek-Rényi-Chow inequality extends to Riesz spaces","Riesz-space maximal inequality drives strong law of large numbers","Submartingale convergence and SLLN in vector lattices via maximal inequality","Generalized maximal inequality for Riesz-space submartingales"],"cache_read_input_tokens":13056,"weakest_assumption_plain":"The chain from the maximal inequality to convergence and the strong law rests on the cited Riesz-space Jensen result that the $p$-th power of a non-negative submartingale is again a submartingale; if that power property fails in some $L^p(T)$, the weighted convergence theorem and the strong law would not follow from the maximal inequality.","fun_headline_variants_meta":{"raw":{"variants":["Hájek-Rényi-Chow inequality extends to Riesz spaces","Riesz-space maximal inequality drives strong law of large numbers","Submartingale convergence and SLLN in vector lattices via maximal inequality","Generalized maximal inequality for Riesz-space submartingales"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000607,"raw_usage":{"total_tokens":2790,"prompt_tokens":865,"completion_tokens":1925,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":481,"completion_tokens_details":{"reasoning_tokens":1850}},"tokens_in":481,"tokens_out":1925,"duration_ms":13754,"temperature":1.0,"reasoning_tokens":1850,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T11:26:40.764940+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Construct a Dedekind complete Riesz space with weak order unit, a strictly positive conditional expectation $T$, and a non-negative submartingale $(X_n,T_n)$ in $L^p(T)$ for which $\\sum_{i=1}^\\infty T_1[(X_{i+1}^p-X_i^p)/a_{i+1}^p]$ converges in order but $X_n/a_n$ does not converge in order; one such example would refute Theorem 5.1. A more targeted check is to test the cited power-submartingale lemma in a concrete space such as $L^p(\\Omega,\\mathcal{F},\\mu)$ with conditional expectation $T$, since the convergence argument depends on that lemma.","supporting_citations":[{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"It supplies the Riesz-space Jensen inequality and the power-submartingale result (Corollary 4.5) needed to convert L^p submartingales into the processes used in Theorems 5.1 and 6.1."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"It defines conditional expectation operators on Riesz spaces and proves that band projections commute with T1 for g in R(T1)+, a step used in the proof of Theorem 4.2."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"It supplies the measure-theoretic Lemma 6.1.1 whose telescoping-series argument Lemma 4.1 generalises to vector lattices."},{"cited_title":"Azouzi, M","cited_arxiv_id":null,"evidence_quote":"It constructs the L^p(T) spaces via functional calculus and gives the Hölder inequality for sums used in Corollary 6.2."},{"cited_title":"Azouzi, K","cited_arxiv_id":null,"evidence_quote":"It provides the Riesz-space Burkholder inequalities used to bound T1|Xn|^p in the proof of the p>2 strong law."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"It gives the classical martingale inequality and strong law whose Riesz-space analogue is the paper's target."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"It gives the classical strong law for p>2 that Theorem 6.3 generalises."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"It provides the upcrossing-based Riesz-space submartingale convergence theorem that the authors note could also prove Theorem 5.1."}],"review_version":1}