{"id":"fb374949-bc8e-4c47-b5f8-5bf0e6d8f83b","arxiv_id":"1908.05579","paper_version":3,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"On trees with finite linear branches, there exist universal harmonic functions whose boundary martingale sequences are dense in the space of measurable functions, and frequently universal ones visit every open set with positive lower density.","lead":"This paper proves that on many infinite tree-shaped graphs, there exist harmonic functions whose boundary behavior can approximate, in a strong repeated sense, any measurable function. Such universal behavior is usually studied for trigonometric series and holomorphic functions; here it is built on trees.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Theorem 3.4 proof gap: the stated extension lemma yields convergence of f^*_{r_k}, not frequent universality; the induction does not force f^*_{r_k} near the prescribed h^*_{\\ell(k)}.","rationale":"The reader's conditionality is well placed, but the precise weak point is not merely missing details in the branching estimates; it is an internal gap in the induction of Theorem 3.4. The stated extension lemma is too weak and, if applied literally, forces the constructed martingale to converge along the subsequence (r_k), which would preclude frequent universality. The intended construction is recoverable: Theorem 3.1's method actually yields the stronger extension lemma with an arbitrary prescribed target, and the branching-at-every-vertex hypothesis is exactly what is needed to make the exceptional arcs have total measure at most 2^{-ℓ(k)}. This is a correction to the proof, not a refutation of the result. I credit the paper for the explicit construction in Theorem 3.1 and the G_delta argument, which appear sound, and for the clear statement of the branching hypotheses. The concrete test above would settle whether the stronger lemma holds; if it does, the paper's main claim stands after a revised proof. No issue was found with the transfer to very regular operators in Section 4, assuming the bijection HP ↔ HQ preserves restrictions to each C_n, which it does.","tokens_in":13892,"tokens_out":21251,"duration_ms":185474,"concrete_test":"Re-derive Theorem 3.4 with the stronger extension lemma (replace the lemma on page 9 by: for every h on B_n harmonic in B_{n-1} and every φ ∈ K_{n+m}, there exists h' = h on B_n with (h')^*_{n+m} ∈ B(φ^†, 2^{-m})), and verify that the induction with n = r_{k-1}, m = ℓ(k), φ = h^*_{\\ell(k)} yields f^*_{r_k} ∈ B(h^*_{\\ell(k)}, 2^{-ℓ(k)}) for every k. If the stronger lemma fails under the branching hypothesis, Theorem 3.4's conclusion is unsupported; if it holds, the stated proof needs only a correction, not a new result.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The most load-bearing concern is the proof of Theorem 3.4. The paper claims that the argument from Theorem 3.1 gives an extension lemma: any h on B_n harmonic in B_{n-1} extends to B_{n+m} with h^*_{n+m} ∈ B(h^*_n, 2^{-m}). This lemma, applied in the induction with n = r_{k-1}, m = ℓ(k), only yields f^*_{r_k} close to f^*_{r_{k-1}}. By the triangle inequality, the sequence f^*_{r_k} is then Cauchy in measure and converges to a limit, so it cannot be frequently universal. The stated conclusion f^*_{r_k} ∈ B(h^*_{\\ell(k)}, 2^{-ℓ(k)}) for a fixed universal h does not follow. The necessary lemma is stronger: for every h on B_n harmonic in B_{n-1} and every K_{n+m}-measurable φ, there should exist an extension h' agreeing with h on B_n with (h')^*_{n+m} ∈ B(φ^†, 2^{-m}). This is what the construction in Theorem 3.1 actually provides (set φ^† on all but one subarc per A_n-atom and adjust the exceptional value to preserve the conditional expectation). Since ℓ(k) ≤ r_{k-1} for large k, φ = h^*_{\\ell(k)} is A_{r_{k-1}}-measurable and the corrected induction succeeds. As written, the proof of the paper's central frequent-density theorem is invalid, although the intended argument is salvageable.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The manuscript studies harmonic functions on infinite locally finite trees with respect to nearest-neighbor transition operators. It identifies P-harmonic functions with boundary martingales via the Poisson transform, and proves several universality results. Theorem 3.1 states that, under the condition that every linear branch is finite, for any infinite sequence of radii n, the set U(n) of harmonic functions whose circle restrictions along n are dense in the space of measurable functions on the boundary is a dense G_delta subset of H_Q, and Theorem 3.2 adds algebraic genericity. Theorem 3.4, the main positive result, asserts that if every vertex has at least two descendants, then the set FU of frequently universal harmonic functions is dense. Theorem 3.6 claims that the set XU of functions whose visits to every open set have upper density 1 is a dense G_delta disjoint from FU, and Section 4 extends the results to very regular transient transition operators.","tokens_in":14188,"tokens_out":23387,"duration_ms":213714,"significance":"If correct, these results would provide a valuable discrete analogue of Menshov's universal trigonometric series and of frequent universality for holomorphic functions, with constructive proofs based on tree geometry and Baire category arguments. The frequent-density statement (Theorem 3.4) is the central new claim, and the paper also offers algebraic genericity and a meagerness complement. The approach via forward-only operators and boundary martingales is elegant. However, the present version contains two serious proof gaps: the induction in Theorem 3.4 uses an extension lemma too weak to force approximation of a prescribed universal target, and the Baire category representation of XU in Theorem 3.6 does not match the definition of XU. These gaps affect the main existence theorem and the meagerness theorem, respectively, and consequently also the transfer to very regular operators in Section 4.","major_comments":[{"comment":"The proof claims that the argument in the last part of Theorem 3.1 yields an extension lemma of the form: any h on B_n harmonic in B_{n-1} extends to B_{n+m} with h^*_{n+m} in B(h^*_n, 2^{-m}). Even if this lemma is granted, the inductive application with n = r_{k-1} and m = ell(k) yields f^*_{r_k} in B(f^*_{r_{k-1}}, 2^{-ell(k)}), not the asserted f^*_{r_k} in B(h^*_{ell(k)}, 2^{-ell(k)}). The sequence f^*_{r_k} is therefore merely Cauchy in measure and need not follow the prescribed universal function h; the conclusion of frequent universality does not follow. The construction in Theorem 3.1 actually appears to support a stronger lemma in which an arbitrary A_{n+m}-measurable target replaces h^*_n, and the proof should state and prove this stronger lemma explicitly. It should also verify the measurability condition ell(k) <= r_{k-1} before applying the lemma, and it should show how, for each open set O, a fixed level m_0 with h^*_{m_0} in O together with Lemma 3.3 yields positive lower density of the hitting set.","section":"Section 3, proof of Theorem 3.4"},{"comment":"Theorem 3.6 defines XU as the set of h such that N(h,O) = {n : h^*_n in O} has upper density 1 for every non-empty open O. However, the sets E(j,m,n) in the proof are defined using the condition that there exist q > (1 - 1/m)n indices k_1 <= ... <= k_q <= n with f^*_{r_{k_l}} in O_j. Thus the intersection in (3.9) characterizes functions whose subsequence along the sparse set {r_k} visits O_j with upper density 1, not functions whose full sequence h^*_n does. Since the set {r_k} has positive but not full density in N, the equality (3.9) is false. The subsequent density argument, which refers to 'the same argument of Theorem 3.1', also appears to construct approximations only at the radii r_k and not at all indices n. The stated result that FU is meager in H_Q is therefore not established by the given proof.","section":"Section 3, Theorem 3.6, Eq. (3.9)"},{"comment":"Corollary 4.1 asserts that Theorems 3.1, 3.2, 3.4 and 3.6 hold for every very regular transition operator. Since the proofs of Theorems 3.4 and 3.6 have the gaps described above, Corollary 4.1 inherits those gaps. In addition, the transfer itself is justified in a few sentences ('It follows easily that the results from Section 3 transfer without changes'); for a result that is part of the paper's advertised scope, the proof should spell out how the Dirichlet-problem construction of Theorem 3.1 is performed for a general very regular P, using the uniform bound U(v^-,v) <= 1 - epsilon from [10] to replace condition (3.3).","section":"Section 4, Corollary 4.1"}],"minor_comments":[{"comment":"In Definition 2 and in the proof of Theorem 3.1, the space M of measurable functions with metric d_nu should be described as the space of equivalence classes modulo nu-a.e. equality; otherwise d_nu is only a pseudometric.","section":"Section 2.3 and proof of Theorem 3.1"},{"comment":"The phrase 'Choose k > log2 s' should read k > log_2 s, and the condition should be 2^{-k} < 1/s, which is a minor typographical issue.","section":"Section 3, proof of Theorem 3.1"},{"comment":"The proof uses the symbol N both for the variable in quantifiers and for the set N(h,O), which makes the argument harder to follow; a different notation for the set of hitting indices would improve readability.","section":"Section 3, Theorem 3.6"},{"comment":"The manuscript contains several typographical errors (e.g., 'measu rable' in the abstract, 'of of' in Theorem 3.6) and would benefit from a careful proofreading pass.","section":"Throughout"},{"comment":"The informal statement that the uniform decay condition is 'equivalent to an exponential decay of the hitting probability' would be clearer as a formal lemma quantifying the rate of decay obtained from the very regular bounds.","section":"Section 4"}],"recommendation":"major_revision","confidential_remarks":"The manuscript contains original ideas and the intended construction is plausible; the gap in Theorem 3.4 is likely repairable by stating and proving the stronger extension lemma that the construction in Theorem 3.1 actually supports. The issue in Theorem 3.6 is more substantial, since the Baire category representation does not match the definition of XU; the authors may need to correct the definition or develop a genuinely different proof. Given that the central existence theorem is affected, a major revision is appropriate. The paper also draws heavily on the authors' previous work [1]; this is legitimate, but the new proof should be sufficiently self-contained for the key extension step."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Punchline: the paper is a genuine contribution to universality on trees, but the main frequent-universality theorem (Theorem 3.4) has a proof gap as written. The gap is repairable, so this is a revise-and-resubmit, not a reject.\n\nWhat's actually new: full proofs of the dense G_delta and algebraic genericity results announced in [1], plus the frequently universal formulation with positive lower density, the meager FU vs XU dichotomy, and a transfer to very regular transition operators. The core construction in Theorem 3.1 is explicit and correct: solve the Dirichlet problem on balls, pick descendant paths with enough branching so the exceptional arcs have small measure, and use Baire category. The arithmetic lemma (Lemma 3.3) that builds the subsequence r_k is neat.\n\nSoft spots:\n\n1. Theorem 3.4 proof. The stated extension lemma only gives h*_{n+m} close to h*_n. Applied inductively, that makes h*_{r_k} Cauchy in measure, so the limit cannot be frequently universal. To get h*_{r_k} close to h*_{ell(k)} for a fixed dense sequence, you need the stronger lemma: any harmonic h on B_n extends to B_{n+m} with h*_{n+m} close to any prescribed K_{n+m}-measurable function. The construction in Theorem 3.1 actually provides this, because you can set the prescribed values on all but one subarc per direction and adjust the exceptional value to preserve harmonicity. Since ell(k) ≤ r_{k-1} eventually, h*_{ell(k)} is measurable at the required level, so the intended induction goes through. But the paper never states this stronger lemma, and the proof as written does not logically establish the theorem.\n\n2. Corollary 4.1 (transfer to very regular operators) is too compressed. “It follows easily” is not enough for a claim that all of Section 3 transfers; a referee should ask for the Dirichlet extension argument spelled out with backward transitions.\n\n3. Remark 3.5 (bounded linear branches) gives only a sketch, not a proof. If the claim is kept, it needs details.\n\n4. Minor typo: in the Theorem 3.6 proof, “open sets in C” should be “open sets in M.”\n\nWho it's for: people working on universality, harmonic analysis on trees, and boundary martingales. The self-citation to [1] is appropriate; this is the promised full version. The paper deserves a serious referee. I'd send it out and require a rewritten Theorem 3.4 with the stronger extension lemma, plus expansion of Section 4.","headline":"Real contribution to universality on trees, but the proof of Theorem 3.4 has a repairable gap: the stated extension lemma is too weak, though the paper's own construction supplies the needed stronger version.","tokens_in":14713,"tokens_out":8119,"would_cite":false,"duration_ms":75281,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["05C05","31A20","60J45"],"pacs":[],"model":"deepseek-v4-flash","headline":"The paper proves that on trees with no infinite linear branches, for each prescribed sequence of levels, the harmonic functions whose restrictions at those levels are dense among measurable boundary functions form a dense Gδ set, and the…","keywords":["harmonic functions on trees","boundary martingales","universal functions","frequent universality","transient transition operators","Poisson transform","tree boundary","universal approximation in measure"],"falsifier":"A direct test of the dichotomy: take a tree that is a one-sided infinite chain, with a forward-only transition operator that moves only forward. Every harmonic function is constant along the chain, so the restrictions to circles are constant sequences and cannot be dense in the measurable functions on the boundary; hence U is empty, exactly as the theorem asserts. For frequent universality, construct a tree by inserting into every edge of a binary tree a linear chain of length $m_n$ growing without bound along the levels $n$; the boundary martingale then repeats values $m_n$ times in a row, so check whether the lower density of visits to any fixed open set drops to zero, which would confirm the role of the bounded-branch-length hypothesis in Theorem 3.4.","tokens_in":1940,"feed_emoji":"🌳","tokens_out":2705,"duration_ms":133222,"temperature":0.7,"pith_summary":"This paper establishes a discrete analogue of universal trigonometric approximation for harmonic functions on infinite trees. It shows that whenever the tree has no infinite linear branch, for any prescribed sequence of levels, the harmonic functions whose restrictions to those levels—viewed as locally constant functions on the boundary—approximate every measurable boundary function in measure form a dense Gδ set. If every vertex has enough descendants, the functions whose associated boundary martingale visits every open set of measurable functions with positive lower density are also dense. All these conclusions extend from the special forward-only transition operators to every very regular transient nearest-neighbor operator. A sympathetic reader would care because it shows that a single harmonic function on a tree can encode, in subsequences of its boundary values, the whole space of measurable functions, a universality phenomenon previously known for trigonometric and holomorphic series.","feed_headline":"Martingales on tree boundaries hit every measurable function","feed_subtitle":"A dense, typical family of harmonic functions has boundary values whose subsequences converge in measure to any target.","key_machinery":"Central machinery is the boundary-martingale representation of harmonic functions. A probability measure $\\nu$ on the boundary $\\Omega$ defines projections $\\pi_n f$ that average $f$ over the arcs subtended by vertices of length $n$; these projections form a martingale, and the quotients $q(u,w)=\\nu(I(w))/\\nu(I(u))$ define a forward-only transition operator $Q$ whose harmonic functions are exactly the Poisson transforms of such martingales. The load-bearing estimate is the arc-shrinking inequality: from any vertex with no infinite linear branch, one can find a descendant path through branching vertices along which each chosen arc has measure at most half its parent's, so values at a sufficiently deep level approximate any target locally constant function within an arbitrarily small measure error. A harmonic-extension step then matches a prescribed function on an earlier ball by choosing a single exceptional value at the end of each path, and a category argument turns the dense approximants into a dense $G_\\delta$ of universal functions. For frequent universality, an arithmetic lemma on indices $r_k$ whose levels with a given 2-adic valuation have strictly positive lower density supplies the schedule along which the martingale visits every open set with positive lower density.","core_discovery":"On a rooted tree whose linear branches all have finite length, fix any forward-only transition operator Q, so that the random walk never moves toward the root. The paper proves that the set U(n) of Q-harmonic functions h whose level-n restrictions, lifted to locally constant functions on the boundary $\\Omega$, form a dense sequence in the space of measurable functions on $\\Omega$ is a dense $G_\\delta$ subset of $H_Q$ for every sequence of levels n; and if the tree has an infinite linear branch, U is empty. When every vertex has at least two forward descendants, the set FU of harmonic functions whose boundary martingales visit every non-empty open set of measurable functions with positive lower density is dense in $H_Q$, while the set XU of functions whose visits have upper density one is a dense $G_\\delta$ disjoint from FU, so FU is meager. The same statements hold for every very regular transient nearest-neighbor transition operator, that is, one with transition probabilities bounded below by $\\delta>0$ and backward probabilities bounded above by $1/2-\\delta$.","pith_inferences":["The same arc-shrinking mechanism is quantitative: replacing the factor $1/2$ by any $\\lambda<1$ should preserve all theorems, so the results should extend to transient operators whose Green kernel decays uniformly along branching paths, a larger class than the very regular operators named in Section 4.","A clean test suggested by Remark 3.5: on a tree obtained by inserting linear chains of length $m_n$ growing unboundedly at level $n$, universality should still hold while frequent universality should fail, because the boundary martingale repeats values for increasingly long stretches and the lower density of visits to any open set should drop to zero.","Read through the martingale representation, the same projection construction could be transplanted to any totally disconnected compact space with a tree-like filtration, for instance the $p$-adic integers, giving universal harmonic functions there as well."],"forward_implications":["On any tree whose linear branches are all finite, universality is typical: for every choice of levels, the universal harmonic functions form a dense Gδ set, so almost every harmonic function in the category sense has a dense boundary-martingale orbit.","The same dense-Gδ conclusion holds for every very regular transient nearest-neighbor operator, so the phenomenon is independent of the precise transition probabilities provided they satisfy uniform lower and upper bounds.","Frequent universality, meaning visits with positive lower density, is dense but topologically negligible: the stronger upper-density-one set XU is residual and disjoint from FU, so the two notions of frequent behavior are genuinely different.","There is a dense vector subspace of HQ consisting, apart from zero, entirely of universal functions, giving algebraic genericity in addition to topological genericity.","If a tree contains one infinite linear branch, no universal harmonic function exists for forward-only operators, since harmonic functions are constant along such a branch."],"supporting_citations":[{"why":"The earlier companion paper that supplies the ramified-set density and frequent-density results for harmonic functions on trees, and announces the universality statements proved here.","marker":"[1]"},{"why":"Establishes the Poisson-transform and boundary-martingale representation of harmonic functions on trees on which the whole construction rests.","marker":"[5]"},{"why":"Provides the uniform hitting-probability estimates for very regular transition operators that let Section 4 transfer the results beyond forward-only operators.","marker":"[10]"},{"why":"The original universal trigonometric series whose approximation notion is transplanted to tree boundaries.","marker":"[13]"},{"why":"Supplies the abstract disjointness argument between positive-lower-density and upper-density-one universal sets used in Theorem 3.6.","marker":"[12]"},{"why":"Constructs the full family of transient nearest-neighbor operators with a prescribed hitting distribution, allowing the extension to very regular operators.","marker":"[15]"}],"fun_headline_variants":["Tree harmonic functions whose martingales hit every measurable function","Dense set of tree harmonic functions with universal boundary martingales","On trees, a dense family of harmonic functions has universal martingale limits","Universal martingale limits on trees hit every measurable function","Harmonic functions on trees with universal boundary martingale convergence"],"cache_read_input_tokens":16768,"weakest_assumption_plain":"The construction assumes that from every vertex one can descend through enough genuinely branching vertices to make boundary arcs shrink by a factor of at least two; this fails exactly on trees with an infinite linear branch, and frequent universality needs the stronger uniformity that every linear branch has bounded length.","fun_headline_variants_meta":{"raw":{"variants":["Tree harmonic functions whose martingales hit every measurable function","Dense set of tree harmonic functions with universal boundary martingales","On trees, a dense family of harmonic functions has universal martingale limits","Universal martingale limits on trees hit every measurable function","Harmonic functions on trees with universal boundary martingale convergence"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000571,"raw_usage":{"total_tokens":2651,"prompt_tokens":847,"completion_tokens":1804,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":463,"completion_tokens_details":{"reasoning_tokens":1719}},"tokens_in":463,"tokens_out":1804,"duration_ms":14591,"temperature":1.0,"reasoning_tokens":1719,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T13:09:32.196913+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"A direct test of the dichotomy: take a tree that is a one-sided infinite chain, with a forward-only transition operator that moves only forward. Every harmonic function is constant along the chain, so the restrictions to circles are constant sequences and cannot be dense in the measurable functions on the boundary; hence U is empty, exactly as the theorem asserts. For frequent universality, construct a tree by inserting into every edge of a binary tree a linear chain of length $m_n$ growing without bound along the levels $n$; the boundary martingale then repeats values $m_n$ times in a row, so check whether the lower density of visits to any fixed open set drops to zero, which would confirm the role of the bounded-branch-length hypothesis in Theorem 3.4.","supporting_citations":[{"cited_title":"Abakumov, V","cited_arxiv_id":null,"evidence_quote":"The earlier companion paper that supplies the ramified-set density and frequent-density results for harmonic functions on trees, and announces the universality statements proved here."},{"cited_title":"Cartier, Fonctions harmoniques sur un arbre , Symp","cited_arxiv_id":null,"evidence_quote":"Establishes the Poisson-transform and boundary-martingale representation of harmonic functions on trees on which the whole construction rests."},{"cited_title":"Kor´ anyi, M","cited_arxiv_id":null,"evidence_quote":"Provides the uniform hitting-probability estimates for very regular transition operators that let Section 4 transfer the results beyond forward-only operators."},{"cited_title":"Menshov, Sur les S´ eries Trigonom´ etriques Universelles, Comptes Rendus (Dok- lady) de l’Acad` emie des Sciences de l’URSS, XLIX (2), (1945), 79–82","cited_arxiv_id":null,"evidence_quote":"The original universal trigonometric series whose approximation notion is transplanted to tree boundaries."},{"cited_title":"Kyrezi, V","cited_arxiv_id":null,"evidence_quote":"Supplies the abstract disjointness argument between positive-lower-density and upper-density-one universal sets used in Theorem 3.6."},{"cited_title":"On the duality between jump processes on ultrametric spaces and random walks on trees","cited_arxiv_id":"1211.7216","evidence_quote":"Constructs the full family of transient nearest-neighbor operators with a prescribed hitting distribution, allowing the extension to very regular operators."}],"review_version":1}