{"id":"3829617d-77b6-4f56-b88a-6695c5778942","arxiv_id":"2412.18201","paper_version":1,"verdict":"REJECT","confidence":"MODERATE","novelty_score":3.0,"correctness_risk":"high","formal_verification":"none","parameter_count":0,"one_line_summary":"Optimal long-only portfolios in this framework are supported on a boundary set called the green frontier, with unused assets obeying a capital asset pricing inequality.","lead":"This paper gives a mathematical viewpoint on choosing a portfolio: the best portfolio is a kind of projection, and its holdings should sit on a special boundary of the available assets. The authors use this viewpoint to explain why optimal portfolios are often sparse, to reinterpret the capital asset pricing model, and to solve mazes with the same machinery.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Green topiary theorem is not established: Definition 6.1's frontier can be empty for generic K, and the cited Theorem 7.3 does not connect greedy convergence to support in that frontier.","rationale":"Read in good faith, the paper's central advertised result is the green topiary theorem: optimal portfolios are supported on a distinguished boundary. This is what would explain sparsity through maximum principles, and it is exactly the statement the abstract and Section 6 promote. However, the proof of Theorem 6.2 is a single sentence deferring to Theorem 7.3, which is an objective-improvement bound for greedy updates and contains no statement about green frontiers. The definition of Green(K) as an intersection over all open dense sets is strong enough that membership requires the relevant winning region W_x to be fat; the paper supplies no argument that support points of the topiary have this property. Consequently, the central claim is not established as stated. I partially disagree with the reader's choice of weakest assumption: the invisible index theorem's step topiary(K2)=psi+r can be repaired because the equality only needs to hold pointwise on K2, and integration over distributions supported on K1 subset K2 makes the substitution valid. The real vulnerability is Theorem 6.2. The valid parts of the paper, such as the capital asset pricing inequality and the update inequality, are correct elementary calculations and deserve credit, but they do not carry the main advertised conclusion. A finite-dimensional exact computation of Green(K) and the topiary support would settle whether Theorem 6.2 is true or false, and therefore whether the rejection is for missing proof or for a false claim.","tokens_in":15111,"tokens_out":12228,"duration_ms":122470,"concrete_test":"Take a finite-domain example, H=R^3 with K={1,2,3} and the standard kernel, and choose psi=(1,0.8,0). Compute the topiary p by maximizing psi·q - ||q||^2/2 over the simplex. For each i in supp(p), compute W_i={q in the simplex : margin_q(i) >= max_j margin_q(j)} explicitly and check whether W_i contains a relatively open subset of the simplex. By the criterion above, i is in Green(K) iff W_i has nonempty interior. If some i in supp(p) has empty interior, Theorem 6.2 fails. Repeat for random positive definite Gram matrices and random psi to search for a violation of the support-containment claim.","verdict_should_be":"UNCHANGED","load_bearing_attack":"Section 6 defines Green(K) as the intersection over all open dense subsets U of the space of embedded measures on K of M_U, where M_U consists of points x in K for which some mu in U maximizes the aesthetic margin. For a fixed x, let W_x be the set of all mu for which x is a maximizer of the margin. Because the margin is continuous in mu, W_x is closed, and x belongs to Green(K) iff W_x has nonempty interior: if W_x has empty interior, then the complement of W_x is an open dense U with x not in M_U; if W_x contains a nonempty open set, every open dense U meets that set, so x is in M_U. Thus the green frontier is exactly the set of points whose winning region W_x is fat. Theorem 6.2 asserts that topiary(K) is supported on Green(K), but its only proof is the sentence that it is an immediate consequence of the update inequality, Theorem 7.3. Theorem 7.3 bounds the increase of the aesthetic objective per greedy step; it never mentions Green(K) and gives no reason that each point in the support of topiary(K) has a W_x with nonempty interior. Without such an argument, the advertised explanation of sparsity via a distinguished boundary is unsupported, and Green(K) may even be empty for generic compact K. The invisible-index gap identified by the reader is secondary: pointwise equality of topiary(K2) with psi minus a constant on K2 suffices for the substitution in Theorem 5.1, since all competing measures are supported on K1 subset K2.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper studies the optimization problem O(μ)=∫ψ dμ - ½‖μ‖² over kernel-embedded measures supported on a compact set K. It defines the topiary as the maximizer, the aesthetic margin, and the topiaric index (the zero set of the margin). The main advertised results are the capital asset pricing inequality (Theorem 4.1), a Julia-Caratheodory type inequality (Theorem 4.2), the invisible index theorem (Theorem 5.1), and the Green topiary theorem (Theorem 6.2), which claims that an optimal measure can be chosen supported on the 'green frontier.' Sections 7 and 8 give greedy approximation results, finite constructability claims, and informal examples from maze solving and portfolio theory.","tokens_in":15462,"tokens_out":13748,"duration_ms":139491,"significance":"If established, the Green topiary theorem would give a maximum-principle explanation for sparsity in quadratic programs over kernel Hilbert spaces, with potential applications to long-only portfolio theory. The paper contains some correct variational observations: Lemma 3.2 and the subsequent first-order conditions are standard, Theorems 3.3, 4.1, and 4.2 are essentially valid KKT-type restatements, and the invisible index theorem is salvageable as a projection argument once 'topiaric index' is interpreted pointwise. However, the paper's central advertised result, Theorem 6.2, is asserted without a proof, the definition of Green(K) permits the set to be empty, and the admissible class of measures is never specified. The current manuscript therefore does not establish its main claims.","major_comments":[{"comment":"The green topiary theorem is asserted with no proof. The only justification is the sentence 'The green topiary theorem is an immediate consequence of the effectiveness of gradient ascent given by the update inequality, Theorem 7.3.' Theorem 7.3 bounds the per-step gain O(μ_t)-O(μ) in terms of the objective gap and a denominator; it says nothing about Green(K), nor about the winning regions W_x = {μ : ι_μ(x)=sup_K ι_μ} whose nonempty interior is exactly what membership in Green(K) requires. For a fixed x, x∈Green(K) iff W_x has nonempty interior: if W_x has empty interior, its complement is an open dense U with x∉M_U; if W_x contains an open set, every open dense U meets it. The manuscript gives no argument that any x∈K has a W_x with nonempty interior, and natural examples (for instance K=[0,1] with a smooth radial kernel and ψ equal to the embedded Lebesgue measure) suggest that the topiary can have full support while every W_x has empty interior, making Green(K)=∅. Thus the advertised boundary-support theorem is not established.","section":"Section 6, Theorem 6.2"},{"comment":"The admissible class of distributions is never defined. The paper says 'over all distributions' without specifying whether these are probability measures, positive measures of fixed total mass, or signed measures. The financial interpretation requires probability measures (long portfolios with weights summing to one), and Lemma 3.2 and Theorem 7.2 use convex combinations μ+t(δ_x-μ), so the feasible set must be convex and mass-preserving. Without this hypothesis the topiary need not exist or be meaningful; for example, if no mass constraint is imposed and ψ is bounded below, the objective can be made arbitrarily large by scaling. All later theorems inherit this ambiguity.","section":"Definition 3.1"},{"comment":"The proof's statement 'As K2 is an index topiary(K2)=ψ+r for some r∈R' is, taken literally, an equality in H and is generally false; what follows from K2 being a topiaric index is the pointwise identity ψ(x)-topiary(K2)(x)-r=0 for x∈K2. This pointwise identity is sufficient for the theorem, since any admissible μ on K1⊆K2 satisfies ∫topiary(K2)dμ = ∫(ψ+r)dμ. The proof should be rewritten to say this explicitly, rather than invoking an apparent equality of Hilbert-space elements that would require an unproved uniqueness-set property.","section":"Theorem 5.1"},{"comment":"The finite constructability results in Section 7.2 are not proved correctly. The 'Discussion theorem' (Theorem 7.8) says 'Take B to be the topiary of K∪{x}', but the topiary is a measure, not a subset of K; the statement requires a set B⊆K. Theorem 7.9 assumes without justification the existence of a proper subset K0 optimizing O(topiary(K0)) and then concludes from the update inequality that a higher objective on K0∪{x} contradicts maximality of topiary(K); a value higher than topiary(K0) does not itself contradict optimality on K. These claims need either corrected proofs or additional hypotheses.","section":"Theorems 7.8 and 7.9"}],"minor_comments":[{"comment":"The notation r_K is used before it is defined; state explicitly that r_K = ∫(ψ-μ)dμ at the point where it is introduced.","section":"Section 3"},{"comment":"The displayed chain 'ψ-r_K = μ = μ/‖μ‖²(∫ψdμ-r_K)' should be an inequality on K with equality exactly on the topiaric index; as printed the first equality is false for x outside the index.","section":"Theorem 4.1"},{"comment":"There are numerous typographical errors: 'manucript' (p.1), 'portolio' (§1.2), 'coavariance' (p.7), 'maxmimizing' (§8.5), 'greedn' (§7.1.1), and 'the the theme' (§6). These should be corrected in revision.","section":"Throughout"},{"comment":"Using U both for the collection of open dense sets and as the index in M_U is confusing; use a script letter such as \\mathcal{U} for the collection.","section":"Definition 6.1"},{"comment":"The statement 'such that ∥·∥_H ≤ ∥·∥_H' appears to have a typo; it should presumably compare the norm of the new Hilbert space with the original norm.","section":"Problem 3"},{"comment":"The maze-solving example is informal; no precise statement of the algorithm, the convergence guarantee, or the claimed path construction is given, so it is difficult to verify the advertised behavior.","section":"Section 8.1"}],"recommendation":"reject","confidential_remarks":"The main obstacle is Theorem 6.2, which is unproved and, on the present definition, may well be false because Green(K) can be empty. The reader's concern about Theorem 5.1 does not, in my reading, land as a fatal flaw: pointwise equality on K2 is enough for the substitution, so that theorem can be repaired by rewriting the proof. The undefined admissible class of measures in Definition 3.1 is a separate load-bearing gap. The paper's informal style is engaging, but the central mathematical claim is not supported."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Quick take: this paper repackages standard convex optimization facts in RKHS with new names, and its one advertised new result, the green topiary theorem, is not proved. The variational lemmas and the CAPM inequality are correct; the invisible index theorem is basically a projection restatement. The missing proof is load-bearing: Theorem 6.2 is asserted as an 'immediate consequence' of the greedy update inequality, which gives convergence rates but says nothing about support on the green frontier. The stress-test note is right that Green(K) can be empty for generic K—the 'unique maximizer region' for an interior point is often empty or has empty interior—so the theorem is not just unproved, it may be false as stated.\n\nWhat deserves credit: the greedy convergence result (Theorem 7.4) is correct, and the CAPM inequality (Theorem 4.1) is a clean KKT rearrangement. The Julia-Caratheodory inequality is a straightforward Cauchy-Schwarz bound. The invisible index theorem is a restatement of the objective, but the reader's concern about a uniqueness-set assumption is actually secondary: pointwise equality of topiary(K2) with ψ plus a constant on K2 is enough for the substitution, since all competing measures are supported on K1⊆K2. So that part holds.\n\nThe soft spot is exactly the green frontier. Definition 6.1 is almost purely topological; the only proof of Theorem 6.2 is a pointer to a convergence estimate that never mentions support. The example of maze-solving is fun but doesn't supply a proof. For a paper whose abstract promises a maximum-principle explanation of sparsity, this is a real problem.\n\nWho this is for: someone curious about whether a Shilov-boundary analog can explain sparsity in nonnegative quadratic programs. The framework and terminology might be useful, but the main theorem is not established. A serious referee should be sent this paper, because the question of whether Green(K) always supports the topiary is exactly the right question, and it's unresolved. I would not cite the green topiary theorem in my own work until it's proved; the rest is standard.","headline":"An entertaining but unproved central theorem: the green topiary theorem is asserted without a real proof, while the rest is mostly standard convex optimization repackaged.","tokens_in":15952,"tokens_out":4206,"would_cite":false,"duration_ms":39912,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["30C15","30C80","47B32","46E22","91G10","90C20"],"pacs":[],"model":"deepseek-v4-flash","headline":"For quadratic programs over reproducing kernel Hilbert spaces, the maximizing measure is supported on a small boundary set, the green frontier, and the capital asset pricing model holds exactly there.","keywords":["topiary","reproducing kernel Hilbert space","quadratic programming","kernel embedding of measures","green frontier","capital asset pricing inequality","long-only portfolio","maze solving"],"falsifier":"Take a small finite set $K_2$ whose kernel vectors are linearly dependent, choose a continuous $\\psi$ so that the aesthetic margin of its topiary vanishes on all of $K_2$ (making $K_2$ a topiaric index), and compute whether $\\psi+r$ equals $\\mathrm{topiary}(K_2)$ as an element of the Hilbert space; a nonzero difference would invalidate the identification step behind the invisible index theorem.","tokens_in":14875,"feed_emoji":"🌿","tokens_out":14020,"duration_ms":123912,"temperature":0.7,"pith_summary":"The paper establishes a maximum-principle explanation for sparsity in quadratic programs over reproducing kernel Hilbert spaces, which are function spaces where point evaluation is continuous. It proves that the maximizer of the aesthetic objective $O(\\mu)=\\int\\psi\\,d\\mu-\\|\\mu\\|^2/2$, called the topiary, can be supported on a small distinguished boundary, the green frontier, and that its support lies where a certain margin function vanishes on a marginal hypersurface. Read as portfolio theory, this says optimal long-only portfolios are sparse and boundary-supported, and it yields a capital asset pricing inequality: assets outside the topiaric index underperform the line set by the topiary, with equality only on the index. It also proves an invisible index theorem: if a larger universe contains a fully diversified topiaric index, then any restricted optimal portfolio is the best Hilbert-space approximation to that invisible index. The authors demonstrate the machinery by using the topiary's gradient to find a path through a maze, interpreting the path as the invisible index navigating around obstacles.","feed_headline":"Sparse portfolios live on the green frontier","feed_subtitle":"A theorem says long-only optimizers sit on a distinguished boundary, explaining sparse portfolios and capital asset pricing gaps.","key_machinery":"The load-bearing objects are the topiary, the aesthetic margin, and the green frontier. The topiary is the maximizer of $O(\\mu)=\\int\\psi\\,d\\mu-\\|\\mu\\|^2/2$ over probability measures on $K$; even when the carrier measure is not unique, its image in the reproducing kernel Hilbert space is unique. The aesthetic margin $\\iota_\\mu(x)=\\psi(x)-\\mu(x)-r_K$, with $r_K=\\int(\\psi-\\mu)\\,d\\mu$, is the optimality witness: it is zero on the support and nonpositive on $K$. The green frontier, $\\mathrm{Green}(K)=\\bigcap_{U\\in\\mathcal{U}}M_U$, is the intersection over open dense sets of embedded measures of the points where some margin attains its supremum; it plays the role of a distinguished boundary and, by the extreme-point theorem, lies inside the extreme points of $K$ under the kernel embedding. The Green topiary theorem says a measure supported on this frontier embeds to the topiary, and the invisible index theorem is carried by the identification $\\mathrm{topiary}(K_2)=\\psi+r$ for a topiaric index $K_2$, which converts restricted optimization into a best-approximation problem.","core_discovery":"The central claim is that sparsity in long-only quadratic optimization is forced by a boundary maximum principle. For a compact set $K$ and continuous function $\\psi$, the unique embedded topiary $\\mu$ has an aesthetic margin $\\iota_\\mu(x)=\\psi(x)-\\mu(x)-r_K$ that is nonpositive on $K$ and vanishes exactly on the topiaric index (the preimage of zero for this margin), the set where equality in the capital asset pricing inequality holds. The green frontier $\\mathrm{Green}(K)$ is constructed as the intersection over open dense sets of embedded measures of the points where some margin attains its supremum; in harmonic settings it lies on the boundary of $K$. The Green topiary theorem asserts that a measure supported on $\\mathrm{Green}(K)$ embeds to the same topiary as the global maximizer, so the support of an optimal portfolio lies in the intersection of the green frontier with a marginal hypersurface. The invisible index theorem then says that if $K_2$ is itself a topiaric index with topiary $\\psi+r$, the topiary of any smaller $K_1$ minimizes $\\|\\mu-\\mathrm{topiary}(K_2)\\|$ over measures on $K_1$, meaning the restricted portfolio is the best approximation of an invisible fully diversified index.","pith_inferences":["The same boundary-support mechanism should apply to any convex quadratic program with a kernel regularizer and a compact feasible set, so sparsity of optima is a generic geometric phenomenon rather than a special feature of finance.","One could test the invisible index interpretation empirically: compute the topiary on a large universe, remove its support, recompute on the remainder, and check whether the new topiary approaches the old one in reproducing-kernel-Hilbert-space norm; the theorem predicts convergence whenever the uniqueness assumption holds.","The maze-solving demonstration suggests a deterministic path-planning algorithm: the gradient of the aesthetic objective gives a curve from an interior point to the boundary that avoids a given obstacle, and the same construction could work for any compact obstacle set with a suitable kernel.","If the uniqueness assumption in the invisible index theorem fails, the finance narrative would need revision, but the green-frontier sparsity claim would survive; the two claims are separable and testable."],"forward_implications":["Optimal long-only portfolios are sparse: their support lies in the green frontier intersected with a marginal hypersurface, so the efficient allocation is describable by few assets.","Assets not on the topiaric index underperform the capital asset pricing line: $\\psi(x)-r_K\\leq \\beta(x)(\\int\\psi\\,d\\mu-r_K)$ with equality only on the index.","If a fully diversified invisible index exists in a larger universe, every restricted optimal portfolio is the best Hilbert-space approximation to it.","The greedy update algorithm converges in objective value at worst $O(1/n)$, so the sparse topiary is computable by simple incremental updates."],"supporting_citations":[{"why":"Supplies the reproducing kernel Hilbert space theory and the representation theorem for bounded linear functionals that define the kernel embedding of measures.","marker":"[29]"},{"why":"Supplies kernel embeddings of distributions and the learning context whose sparsity the paper says its results explain.","marker":"[33]"},{"why":"Supplies the modern portfolio theory and capital asset pricing model framework that the paper generalizes and reinterprets.","marker":"[18]"},{"why":"Supplies the quadratic optimization formulation for long-only portfolios that the topiary generalizes.","marker":"[25]"},{"why":"Supplies the capital asset pricing framework against which the topiaric beta and the inequality are compared.","marker":"[34]"},{"why":"Supplies the growth-rate rationale for maximizing mean minus variance over two, which motivates the aesthetic objective.","marker":"[22]"},{"why":"Supplies the critique of capital asset pricing model testability that motivates the invisible index theorem.","marker":"[31]"},{"why":"Supplies the support vector machine sparsity analogy used to frame the topiaric index as an outline of the set.","marker":"[30]"}],"fun_headline_variants":["Sparse portfolios force a boundary frontier theorem","Maximum principle pins optimal portfolios to frontier","Green frontier theorem explains sparse allocation","Why long-only optimizers sit on a distinguished edge"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The invisible index theorem assumes that once the margin vanishes on a compact set $K_2$, the embedded topiary equals the function $\\psi+r$ throughout the reproducing kernel Hilbert space; this uniqueness-of-frontier property is stated but not proved for general $K_2$, and the finance conclusion rests on it.","fun_headline_variants_meta":{"raw":{"variants":["Sparse portfolios force a boundary frontier theorem","Maximum principle pins optimal portfolios to frontier","Green frontier theorem explains sparse allocation","Why long-only optimizers sit on a distinguished edge"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000484,"raw_usage":{"total_tokens":2355,"prompt_tokens":878,"completion_tokens":1477,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":494,"completion_tokens_details":{"reasoning_tokens":1423}},"tokens_in":494,"tokens_out":1477,"duration_ms":12061,"temperature":1.0,"reasoning_tokens":1423,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-11T04:57:14.933686+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Take a small finite set $K_2$ whose kernel vectors are linearly dependent, choose a continuous $\\psi$ so that the aesthetic margin of its topiary vanishes on all of $K_2$ (making $K_2$ a topiaric index), and compute whether $\\psi+r$ equals $\\mathrm{topiary}(K_2)$ as an element of the Hilbert space; a nonzero difference would invalidate the identification step behind the invisible index theorem.","supporting_citations":[{"cited_title":"An introduction to the theory of repro- ducing kernel Hilbert spaces , volume 152","cited_arxiv_id":null,"evidence_quote":"Supplies the reproducing kernel Hilbert space theory and the representation theorem for bounded linear functionals that define the kernel embedding of measures."},{"cited_title":"Learning with kernels: support vector machines, regularization, optimization, and beyond","cited_arxiv_id":null,"evidence_quote":"Supplies kernel embeddings of distributions and the learning context whose sparsity the paper says its results explain."},{"cited_title":"Modern portfolio theory: Foundations, analysis, and new developments","cited_arxiv_id":null,"evidence_quote":"Supplies the modern portfolio theory and capital asset pricing model framework that the paper generalizes and reinterprets."},{"cited_title":"The optimization of a quadratic function subject to linear constraints","cited_arxiv_id":null,"evidence_quote":"Supplies the quadratic optimization formulation for long-only portfolios that the topiary generalizes."},{"cited_title":"Capital asset prices: A theory of market equilibrium under conditions of risk","cited_arxiv_id":null,"evidence_quote":"Supplies the capital asset pricing framework against which the topiaric beta and the inequality are compared."},{"cited_title":"A new interpretation of information rate","cited_arxiv_id":null,"evidence_quote":"Supplies the growth-rate rationale for maximizing mean minus variance over two, which motivates the aesthetic objective."},{"cited_title":"A critique of the asset pricing theory’s tests part i: On past and potential testability of the theory","cited_arxiv_id":null,"evidence_quote":"Supplies the critique of capital asset pricing model testability that motivates the invisible index theorem."},{"cited_title":"The mathematics of learning: Dealing with data","cited_arxiv_id":null,"evidence_quote":"Supplies the support vector machine sparsity analogy used to frame the topiaric index as an outline of the set."}],"review_version":1}