{"id":"e6aa5b99-9340-4585-9927-b2bae0f5ae53","arxiv_id":"1908.01895","paper_version":1,"verdict":"ACCEPT","confidence":"MODERATE","novelty_score":7.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"Noncommutative plurisubharmonic rational functions are exactly composites of convex rational functions with analytic noncommutative rational functions, with an explicit realization-based criterion.","lead":"A team of mathematicians proved that noncommutative plurisubharmonic rational functions, whose matrix Hessian is always positive semidefinite, are exactly the composites of a convex rational function with an analytic noncommutative rational function. They also gave a fast matrix test for recognizing them from their standard realizations.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Lemma 3.2 is the load-bearing point: Theorem 3.1's necessity and the Section 5 construction both depend on its spanning claim, yet the proof is a citation to a hermitian adaptation of [HMV06] and [BK13] that the manuscript does not supply.","rationale":"The reader's weakest_assumption matches mine. Lemma 3.2 is the only step where a genuinely noncommutative phenomenon—the existence of a small X and v making the resolvent produce a d-dimensional component space—is imported rather than proved. The rest of Theorem 3.1's necessity is linear algebra, the sufficiency direction is self-contained, and Section 5's construction is elaborate but built on the K-nonnegativity obtained from Theorem 3.1. I found no counterexample and no internal contradiction. The polynomial analog is known, and the rational extension is the claimed contribution; therefore the unverified spanning lemma is the single point on which the central claim should be checked. I recommend CONDITIONAL rather than REJECT because the imported lemmas are plausible and likely valid, but the manuscript does not contain enough detail to certify the hermitian adaptation. If an independent check confirms Lemma 3.2, or supplies a proof, the reader's ACCEPT stands.","tokens_in":21083,"tokens_out":24407,"duration_ms":265750,"concrete_test":"Independently re-derive Lemma 3.2 from minimality without citing HMV06: expand Δ(X)(c⊗v) as a formal power series in X and X*, show that minimality implies some d×d minor of the associated matrix is a nonzero nc rational function, and then invoke [BK13]'s local-global principle only in the form that applies to rational matrix entries. If the minor is identically zero on Bε for some ε, or if [BK13] requires polynomial entries and the rational-to-polynomial reduction is not supplied, then Theorem 3.1's necessity direction and the start of Section 5 are unsupported and the paper should be revised to include the missing argument.","verdict_should_be":"CONDITIONAL","load_bearing_attack":"Both main theorems inherit the risk carried by Lemma 3.2. In Proposition 3.4, plushness is used only through r↓(X,0)≥0, and the conclusion PKP≽0 requires that for the ε supplied by the free ball one can find X∈Bε and v such that z=Δ(X)(c⊗v) has d linearly independent components; Lemma 3.3 then turns this into a spanning statement for rngB⊗C^n. The same PSD conclusion is the input to the converse construction in Section 5, since that construction assumes rngB and rngB* are K-nonnegative by Theorem 3.1. Lemma 3.2 is not proved in the manuscript: it says to substitute x_j=y_j+iy'_j and apply a 'hermitian version' of [HMV06, Lemmas 7.2 and 7.4], justified only by the local-global principle in [BK13]. This is exactly the point where the paper's (x,x*) setting has to be reconciled with the symmetric/hermitian-variable results that Remark 1.7 promises but does not prove at the level of spanning lemmas. A secondary incompleteness appears in Lemma 5.3, whose second displayed identity is dismissed with 'We omit the details'; that identity is used in the moment-matching proof of Theorem 5.2. Neither point is circular or fraudulent, but both are unverified load-bearing steps.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper studies symmetric noncommutative (nc) rational functions in variables x_1,...,x_g and their adjoints, regular at 0, and defines a function to be plurisubharmonic (plush) if its nc complex Hessian is matrix positive semidefinite on an nc neighborhood of 0. The two main results are Theorem 1.3 and Theorem 1.4. Theorem 1.3 asserts that a symmetric nc rational function is plush in a neighborhood of 0 if and only if it can be written as f composed with q, where f is a convex nc rational function and q is an analytic nc rational mapping. Theorem 1.4 gives a computable criterion: for a minimal symmetric descriptor realization r(x,x*) = c*(K - Lambda_B(x) - Lambda_{B*}(x*))^{-1}c, the function r is plush if and only if PKP and P_* K P_* are positive semidefinite, where P and P_* are the projections onto the ranges of B and B*, respectively. The proof strategy is constructive: Section 2 reduces plushness to two separated estimates for the Hessian, Section 3 proves the realization criterion, Section 4 treats convex nc rational functions and proves the easy direction of Theorem 1.3, and Section 5 constructs f and q via formal power series and a finite-dimensional reduction.","tokens_in":1837,"tokens_out":6220,"duration_ms":86321,"significance":"If fully substantiated, these results are significant. Theorem 1.3 gives a clean structural characterization of plush nc rational functions, relating them to convex nc rational functions through analytic nc maps, and the proof is constructive. Theorem 1.4 provides a finite-dimensional, computable test for plushness in terms of a minimal realization, which is a practically useful criterion. The paper is careful in stating its dependence on realization theory and on prior results. However, the proof of the crucial spanning lemma, Lemma 3.2, is not self-contained and is delegated to a cited hermitian-variable version of results from [HMV06]; this lemma is used in the necessity direction of Theorem 3.1 and hence in the converse construction of Section 5. A second, smaller gap is the omitted proof of one identity in Lemma 5.3. These gaps do not appear to be circular or fraudulent, but they are load-bearing and should be repaired before the paper is accepted.","major_comments":[{"comment":"Lemma 3.2 is the central spanning lemma: it asserts that for a minimal realization and every epsilon > 0 there exist n, X in the free ball of radius epsilon, and v such that z = Delta(X)(c tensor v) has d linearly independent components. This lemma is used in Proposition 3.4 to prove the necessity direction of Theorem 3.1, and the converse construction in Section 5 relies on Theorem 3.1. The proof given in the manuscript, however, is limited to the sentence: substitute x_j = y_j + i y'_j and apply a hermitian version of [HMV06, Lemmas 7.2 and 7.4], justified by the local-global principle of [BK13]. This is not a proof in the manuscript; it is a citation to an adaptation that the paper does not state or verify. Since Remark 1.7 explicitly promises that results for symmetric variables transfer to hermitian variables but does not prove that transfer at the level of spanning lemmas, the necessity direction of the main characterization is not fully supported. Please provide a complete proof of Lemma 3.2, or a precise statement and proof of the hermitian-variable versions of the cited lemmas.","section":"3, Lemma 3.2"},{"comment":"Lemma 5.3 states two identities used to simplify alternating words in the moment-matching proof of Theorem 5.2. The first identity, B_l^* K [psi^* J psi] K B_j^* = B_l^* K B_j^*, is proved. The second identity, B_l K [psi^* J psi] K B_j^* = B_l K B_j^*, is dismissed with 'We omit the details.' This second identity is used in the simplification of the alternating products in equations (5.4)-(5.7) and is therefore essential to the conclusion that f(q(x)) equals the original realization r(x). The omission is not merely cosmetic. Please include a complete derivation, or at least a fully detailed sketch analogous to the proof of the first identity.","section":"5.2, Lemma 5.3"},{"comment":"The paper repeatedly applies theorems from the literature that are stated for symmetric variables or for hermitian variables, while the paper works with the (x,x*) formalism. Remark 1.7 acknowledges this and gives a heuristic justification via the change of variables x_j = y_j + i y'_j. This transfer is used at load-bearing points: Lemma 3.2 invokes a hermitian version of [HMV06, Lemmas 7.2 and 7.4], and Proposition 4.1 invokes [HMV06, Proposition 5.1] together with Remark 1.7. The manuscript does not supply the transfer proofs that Remark 1.7 promises. Because the characterization theorems depend on these transferred statements, the paper should either prove the relevant hermitian-variable versions or restructure the argument to avoid relying on them.","section":"Remark 1.7 and global transfer of results"}],"minor_comments":[{"comment":"The abstract lists the variables as x_1,...,x_g,x_1^*,...,x_g; the last symbol should presumably be x_g^* rather than x_g.","section":"Abstract"},{"comment":"The formal domain of the symmetric descriptor realization is described using 'J - Lambda_B(X) - Lambda_B(X)^*', but the realization in equation (1.5) uses the signature matrix K, not J. This appears to be a typo.","section":"Section 2.3, first paragraph"},{"comment":"In the proof, the notation R and R_* is used both for the subspaces rng B and rng B^* and, near the end, for the inclusions of these subspaces. This dual use of the same symbol is confusing; using distinct symbols for subspaces and inclusions would improve readability.","section":"Section 3, proof of Proposition 3.5"},{"comment":"The term 'convexotonic map' is used without definition or reference to a specific definition. A one-sentence definition or a pointer to the precise definition in [HKMV] would help the reader verify the claim made in the remark.","section":"Section 5.3.1, Remark 5.4"}],"recommendation":"major_revision","confidential_remarks":"The central content is promising and the paper is likely to be acceptable once the omitted spanning lemma and the omitted identity in Lemma 5.3 are supplied. The manuscript's own Remark 1.7 signals the main dependency, but the transfer from symmetric-variable results to the (x,x*) setting is not proved at the level where it is used. I do not see grounds for rejection, only for requiring the authors to fill these gaps. The paper is within the scope of the journal, and the constructive nature of Theorem 1.3 and the computable criterion in Theorem 1.4 are significant strengths."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Quick take: this is the right paper at the right time. The authors extend the known polynomial result (Greene, Helton, Vinnikov) to nc rational functions: a symmetric rational function is plush near 0 iff it is the composition of a convex rational function with an analytic rational map. They also give a checkable criterion on a minimal descriptor realization: PKP and P*KP* positive semidefinite. That is a clean, useful result, and the proof is constructive—Remark 1.5 notes it gives an algorithm.\n\nWhat is genuinely new is the rational case, including the finite-dimensional reduction in Section 5.3 that gets from infinitely many intermediate variables to finitely many, and the explicit formulas for f and q. The paper is well organized, and the main proof strategy is transparent: necessity from minimality plus a spanning argument, sufficiency from K-nonnegative subspace perturbation, and a careful moment-matching argument in the formal power series section.\n\nThe soft spot is exactly where the stress-test lands. Lemma 3.2 is load-bearing for both Theorem 3.1 necessity and the Section 5 converse: it asserts that in a minimal realization you can find X and v such that Δ(X)(c⊗v) has d independent components. The proof is a short paragraph saying to use a Hermitian version of [HMV06, Lemmas 7.2 and 7.4], justified by [BK13]. That is not a proof in this paper. The authors do flag in Remark 1.7 that they are moving between (x,x*) and Hermitian settings, but they do not actually produce the Hermitian adaptation of those spanning lemmas. If Lemma 3.2 fails, the whole characterization could fall apart. I do not think it fails—the cited results are likely adaptable—but a referee should verify this step. Similarly, Lemma 5.3 leaves one of two identities to the reader ('We omit the details'), and that identity is used in the moment matching for Theorem 5.2. That is a minor incompleteness by comparison.\n\nNo circularity, no invented entities. The citation pattern is appropriate: prior work by the same authors is cited where it is genuinely foundational (realization theory, local-global principle), and the new contribution is clearly separated from the polynomial case.\n\nBottom line: this deserves a serious referee. The main theorems are plausible and well motivated; the proof structure is sound modulo the imported spanning lemma. If I were refereeing, I would ask the authors to spell out Lemma 3.2 or provide the exact statement from [HMV06] with the Hermitian adaptation, and to complete Lemma 5.3. But I would not desk-reject; this is a meaningful advance in free analysis. I'd bring it to my reading group if anyone works on nc functions or convexity.","headline":"A constructive rational-function extension of the plush decomposition theorem; the main risk is a load-bearing spanning lemma that is cited rather than proved.","tokens_in":21909,"tokens_out":7078,"would_cite":true,"duration_ms":54409,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["47A56","46L07","32A99","46L89"],"pacs":[],"model":"deepseek-v4-flash","headline":"A noncommutative rational function is plurisubharmonic exactly when it factors as a convex rational function composed with an analytic rational function, and a two-line projection test on its minimal realization decides the property.","keywords":["plurisubharmonic","noncommutative rational function","descriptor realization","free analysis","complex Hessian","convex rational function","minimal realization","matrix positivity"],"falsifier":"Compute the complex Hessian numerically for a candidate minimal symmetric descriptor realization with a non-trivial signature $K$, testing the predicted equivalence between positivity of $PKP$ and $P_* K P_*$ and positivity of all Hessian blocks $r_\\downarrow(X,Y)[H]$ and $r_\\uparrow(X,Y)[H]$ on a dense grid of direct sums $X\\oplus Y$ with off-diagonal $H$; a single instance with PSD projected blocks but a negative Hessian block, or a plush $r$ with one projected block indefinite, would refute Theorem 1.4 and the constructive direction of Theorem 1.3.","tokens_in":20768,"feed_emoji":"📐","tokens_out":11722,"duration_ms":93655,"temperature":0.7,"pith_summary":"The paper establishes a structure theorem for plurisubharmonic (plush) noncommutative rational functions: a symmetric nc rational function regular at 0 is plush near 0 if and only if it is the composition of a convex nc rational function with an analytic rational mapping. This lifts to rational functions the known polynomial fact that plush polynomials split as convex composed with analytic. The factorization is constructive, built from a minimal descriptor realization via angular operators and finitely many intermediate variables. The paper also proves a computable test: plushness holds exactly when the two projected signature blocks $PKP$ and $P_* K P_*$ are positive semidefinite, for the projections onto the ranges of the two coefficient tuples. Testing plushness therefore reduces to checking a pair of matrix inequalities.","feed_headline":"Every plush rational map factors through a convex one","feed_subtitle":"Two positive-semidefinite projections on the minimal realization decide the property.","key_machinery":"The engine of the paper is the minimal symmetric descriptor realization $r(x,x^*)=c^*(K-\\Lambda_B(x)-\\Lambda_{B^*}(x^*))^{-1}c$, with signature matrix $K$, together with the two range projections $P$ onto $\\operatorname{rng} B$ and $P_*$ onto $\\operatorname{rng} B^*$. The central identity is the direct-sum decomposition of the complex Hessian: reading $r$ at $X\\oplus \\widetilde{X}$ in the block off-diagonal direction $H$ produces $\\operatorname{diag}(r_\\downarrow(X,\\widetilde{X})[H], r_\\uparrow(X,\\widetilde{X})[H])$, so plushness is equivalent to positivity of both $r_\\downarrow$ and $r_\\uparrow$ on free balls; these in turn reduce to $PKP\\succeq 0$ and $P_* K P_*\\succeq 0$. For the converse construction, maximal $K$-nonnegative subspaces containing the two ranges, parameterized by angular operators $\\rho,\\rho_*$, are used to build the convex factor $f$ and the analytic map $q$; a finite set of intermediate variables suffices once $\\{B_1,\\dots,B_g\\}$ is taken linearly independent.","core_discovery":"On its own terms, the paper's central claim is Theorem 1.3: for a symmetric nc rational function $r$ in $g$ variables, regular at 0, $r$ is plush in a neighborhood of 0 if and only if there are a positive integer $h$, a convex nc rational function $f$ in $h$ variables, and an analytic nc rational mapping $q$ with $r=f\\circ q$. The companion result, Theorem 1.4, gives a realization-level criterion: if $r$ is presented minimally as $c^*(K-\\Lambda_B(x)-\\Lambda_{B^*}(x^*))^{-1}c$, then $r$ is plush near 0 exactly when $PKP$ and $P_* K P_*$ are positive semidefinite, where $P$ and $P_*$ are the orthogonal projections onto $\\operatorname{rng} B$ and $\\operatorname{rng} B^*$. The sufficiency direction uses the $K$-nonnegativity of the two ranges to build a pairwise positive decomposition of the complex Hessian; the necessity direction uses the Hessian decomposition together with a spanning lemma securing that every direction in the ranges can be realized by an off-diagonal block. A parallel statement, Theorem 1.6, characterizes convexity near 0 by $QKQ$ being positive semidefinite for $Q$ the projection onto $\\operatorname{rng} B+\\operatorname{rng} B^*$.","pith_inferences":["Beyond the paper: the finite-variable reduction suggests an explicit bound on the intermediate dimension $h$ in terms of the realization size $d$ (roughly $d^2$ from the algebra generated by $KB_j$), which the paper does not state.","Beyond the paper: the necessity direction hangs on a spanning lemma imported from hermitian-variable realization theory; a self-contained proof, or a counterexample in the $x,x^*$ setting, would be the cleanest stress test of the characterization.","Beyond the paper: the formal power-series identity in Theorem 5.2 may hold for a wider class of noncommutative formal rational series, e.g., those with non-symmetric coefficients, suggesting a formal-language version of the factorization."],"forward_implications":["Plushness of an nc rational function is decidable: minimal realizations are computable, so Theorem 1.4 reduces the question to checking two positive-semidefiniteness conditions.","Free pseudoconvex sets defined by plush rational functions admit proper analytic nc rational maps onto convex free domains, connecting plurisubharmonicity to the linear matrix inequality world.","Convexity near 0 for symmetric nc rational functions has the same flavor of test: $QKQ\\succeq 0$ for the projection $Q$ onto $\\operatorname{rng} B+\\operatorname{rng} B^*$ (Theorem 1.6), and convexity implies plushness.","The class of plush rationals is closed under right composition with analytic rational maps, so the factorization $r=f\\circ q$ is not a one-off: every analytic reparameterization of a convex rational factor remains plush."],"supporting_citations":[{"why":"Supplies the minimal realization machinery and the spanning lemmas (7.2, 7.4) that Lemma 3.2 invokes for the necessity direction.","marker":"[HMV06]"},{"why":"Provides the observability/controllability characterization of minimality used in Proposition 2.4.","marker":"[BGM05]"},{"why":"Gives the angular-operator description of maximal Krein-space nonnegative subspaces that the explicit f and q construction relies on.","marker":"[And79]"},{"why":"Proves the local-global linear dependence principle used to extend the spanning lemma to the hermitian-variable setting.","marker":"[BK13]"},{"why":"Supplies realization theory for hermitian variables invoked in Remark 1.7 to transfer results to the (x,x*) setting.","marker":"[Vol18]"},{"why":"Establishes the polynomial precursor—plush polynomials are convex-composed-with-analytic—that Theorem 1.3 generalizes.","marker":"[Gre12]"},{"why":"Provides the convexotonic map identity (their Lemma 2.5) used to reduce the infinite intermediate-variable construction to finitely many variables.","marker":"[HKMV]"}],"fun_headline_variants":["Plush rational maps: convex modulo analytic","Plush rationals are convex after analytic","A simple projection test detects plush rational functions","Plush iff convex composition with analytic map","Minimal realization decides plush rationals"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The result assumes the spanning lemma: for a minimal realization one can always find a small block $X$ and vector $v$ for which the $d$ components of $\\Delta(X)(c\\otimes v)$ are linearly independent; the paper quotes this from prior Hermitian-variable work rather than proving it here.","fun_headline_variants_meta":{"raw":{"variants":["Plush rational maps: convex modulo analytic","Plush rationals are convex after analytic","A simple projection test detects plush rational functions","Plush iff convex composition with analytic map","Minimal realization decides plush rationals"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.001438,"raw_usage":{"total_tokens":5793,"prompt_tokens":938,"completion_tokens":4855,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":554,"completion_tokens_details":{"reasoning_tokens":4788}},"tokens_in":554,"tokens_out":4855,"duration_ms":34975,"temperature":1.0,"reasoning_tokens":4788,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T15:01:34.315722+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Compute the complex Hessian numerically for a candidate minimal symmetric descriptor realization with a non-trivial signature $K$, testing the predicted equivalence between positivity of $PKP$ and $P_* K P_*$ and positivity of all Hessian blocks $r_\\downarrow(X,Y)[H]$ and $r_\\uparrow(X,Y)[H]$ on a dense grid of direct sums $X\\oplus Y$ with off-diagonal $H$; a single instance with PSD projected blocks but a negative Hessian block, or a plush $r$ with one projected block indefinite, would refute Theorem 1.4 and the constructive direction of Theorem 1.3.","supporting_citations":[],"review_version":1}