{"id":"55df46f8-4a3f-4855-a88a-95332bedbfb9","arxiv_id":"1908.03186","paper_version":4,"verdict":"ACCEPT","confidence":"MODERATE","novelty_score":8.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"A generalized Young measure comes from A-free measures exactly when it satisfies Jensen-type inequalities for all A-quasiconvex integrands and its concentration part lies in the wave cone.","lead":"This paper establishes a complete mathematical characterization of all possible oscillation and concentration limits (generalized Young measures) of sequences of solutions to linear differential constraints with constant rank. It provides the key duality tool for variational problems at the borderline L1 energy scale, with applications to failure of rigidity and compactness.","discovery_kind":"unification","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Sufficiency rests on an unproved relaxed relaxation theorem (Remark A.1/Theorem A.1); if its lower bound fails for signed integrands, the Hahn--Banach contradictions in Section 9 collapse.","rationale":"The paper is a serious analytic proof of a natural characterization, and the overall architecture is coherent. The reader's weakest assumption, constant rank, is explicitly stated and honestly discussed in Section 1.2; I do not regard that as an internal flaw. However, the sufficiency of the main theorem passes through Propositions 9.2 and 9.3, and both use the relaxation theorem with an integrand that is only known to lie in E and may be signed. The stated Theorem A.1 requires f >= 0 plus Lipschitz regularity; the relaxed version in Remark A.1 is asserted with a sketch, and the lower-bound step is the part where the original positivity hypothesis is most likely doing real work. Since the paper does not supply a complete proof of that relaxed lower bound, there is a genuine, load-bearing missing justification. This does not mean the theorem is false; it means acceptance should be conditional on completing or independently verifying that auxiliary result. A concrete analytical check is to re-derive the lower semicontinuity bound for signed f under assumptions (A)--(C), and then re-run the separation steps. If the check passes, the central claim is very likely sound; if it fails, the sufficiency proof would need substantial repair.","tokens_in":66151,"tokens_out":15695,"duration_ms":182685,"concrete_test":"Write out a complete proof of Theorem A.1 under exactly the relaxed assumptions (A)--(C) of Remark A.1 for signed f in E(Omega;W), with |mu|(partial Omega)=0, and verify in particular the lower semicontinuity inequality G[mu] >= integral Q_A f(x,mu_ac) dx + integral (Q_A f)^#(x,g_mu) d|mu_s| for sequences u_j with u_j L^d weak-* converging to mu and A u_j -> 0 in W^{-k,q}. Then repeat Steps 3 of Propositions 9.2 and 9.3 with f=(tilde f_H)^epsilon to confirm the separation contradiction. If the lower bound cannot be proved, exhibit a signed f (for instance f = -dist(.,K)) and an operator A for which the claimed relaxation identity is violated; such a counterexample would invalidate the Hahn--Banach argument in Section 9.","verdict_should_be":"CONDITIONAL","load_bearing_attack":"The central theorem is Theorem 1.1. Its sufficiency is reduced in Section 9.4 to Propositions 9.2 and 9.3, whose proofs proceed by Hahn--Banach separation: assume the target Young measure lies outside the A-free class, extract a separating integrand f_H, form f = (tilde f_H)^epsilon, invoke the relaxation result Theorem A.1 to build a recovery sequence, and then derive a contradiction from the assumed Jensen-type inequalities. The separating integrand f_H is only known to lie in E(Omega;W) and need not be nonnegative, while Theorem A.1 as stated requires f >= 0 and Lipschitz dependence in z. The paper substitutes a relaxed version in Remark A.1 under assumptions (A)--(C), but only sketches the changes. The critical lower bound, namely G[mu] >= integral Q_A f(x,mu_ac) dx + integral (Q_A f)^#(x,g_mu) d|mu_s|, is exactly the point where the positivity of f was used in [6] to prevent negative boundary concentration, and Remark A.1 asserts that |mu|(partial Omega)=0 fixes this without giving a proof. If the relaxed lower bound fails, or if it requires additional regularity of Q_A f not guaranteed by Propositions 4.6 and 4.8, then Propositions 9.2 and 9.3 do not establish the contradiction and the sufficiency direction of Theorem 1.1 is not fully supported. This is the most load-bearing unproved input: the convexity and localization architecture of Section 9 depends on it. The constant rank assumption identified by the reader is an explicit hypothesis, not an internal gap.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper characterizes generalized Young measures generated by A-free measures and by B-gradients for linear homogeneous constant-rank operators of arbitrary order. The main results are Theorem 1.1 (duality with A-quasiconvex integrands via Jensen-type inequalities and a wave-cone support condition), Theorem 1.2 (local tangent characterization), and Theorems 1.3 and 1.6 (area-strict approximation of A-free and B-gradient measures). Theorem 1.5 gives the B-gradient analogue. The proof architecture in Sections 7–9 uses a Helmholtz-type decomposition of generating sequences into B-gradients plus a lower-order A-free part, a convexity result for Y_{A,0}(mu, Omega), Hahn–Banach separation against A-quasiconvex integrands, and a relaxation result in Appendix A. The applications in Section 3 show failure of L1-compensated compactness and flexibility of divergence-free Young measures.","tokens_in":66493,"tokens_out":6539,"duration_ms":71373,"significance":"Assuming the relaxation input in Appendix A is fully justified, the paper solves a natural and long-standing problem: it extends the Kinderlehrer–Pedregal/Fonseca–Müller program and the BV-gradient characterization of Kristensen–Rindler to arbitrary constant-rank operators, covering both oscillation and concentration. The local tangent characterization and the area-density theorems are independent contributions with clear applications. The constant-rank hypothesis is stated explicitly and is not hidden; Section 1.2 even discusses its failure. The paper is honest about which steps are delegated to earlier work and which are new; nevertheless, one delegated and partly asserted step is load-bearing and currently unsupported.","major_comments":[{"comment":"The sufficiency direction of Theorem 1.1 rests on applying the relaxation Theorem A.1 to the integrand f = (tilde f_H)^epsilon, where f_H arises from Hahn–Banach separation in (95) and (100) and is not known to be nonnegative. As stated, Theorem A.1 requires f >= 0 and Lipschitz dependence in z. The relaxed version in Remark A.1 replaces this by assumptions (A)–(C), but the critical lower bound G[mu] >= integral Q_A f(x, mu_ac) dx + integral (Q_A f)^#(x, g_mu) d|mu_s| is only asserted. Item 7 says that positivity was used in [6] to prevent negative boundary concentration and that assumption (C) dispenses with this, but no proof is given. Since this lower bound is used in (97)–(98) to contradict the separation inequality, this is a load-bearing missing argument. A revision should either prove the lower bound for signed integrands under (C) or state and prove a complete relaxation theorem under the hypotheses actually used.","section":"Appendix A, Remark A.1; Eqs. (97)–(98)"},{"comment":"The integrand f_H separating the putative Young measure from Y_sing_A(mu) or Y_reg_A(P_0) is an arbitrary element of E(Q;W) and may be signed. The paper uses Lemma 9.1 to infer finiteness of Q_A tilde f_H, but the proof is delegated verbatim to [12, Lemma 5.5], even though [12] was proved under additional assumptions (first-order operators and a Morrey-type bound). If the verbatim transfer is correct, that should be stated explicitly and checked; if it is not, Propositions 4.6 and 4.8 do not suffice to make the relaxation upper bound applicable. The same pattern appears in Proposition 9.3, where the inequality Q_A tilde f_H(0) >= s_H is used without an independent proof that the signed relaxation lower bound holds for tilde f_H. This is not a presentation point: the contradictions in Step 3 of Propositions 9.2 and 9.3 collapse if the signed relaxation statement in Remark A.1 is invalid.","section":"Sections 9.2–9.3, Lemma 9.1 and Eqs. (95)/(100)"}],"minor_comments":[{"comment":"The numbering after condition (iv) is off: the second condition labelled (iv) should be (v) (the L1(U) integrability of x -> <|q|, nu_x>), since (iv) already refers to local integrability.","section":"Definition 1.1"},{"comment":"The text has a typo: 'Avj = 0 in the sese of distributions on Omega' should read 'in the sense of distributions on Omega'.","section":"Section 3.1, Example 3.1"},{"comment":"The expression 'mu_ac(u)' should presumably be 'mu_ac(y)' in the argument of Q_A f; as written it is unclear.","section":"Section 9.2, Eq. (98)"},{"comment":"In the proof, 'Auj = 0' should be 'Awj = 0' to match the notation of the sequence being constructed.","section":"Section 3.2, Lemma 3.2"}],"recommendation":"major_revision","confidential_remarks":"This is a strong paper with a clear research program, and I see no indication that the main characterization is false. The only substantive obstacle I see is the missing proof of the relaxed relaxation lower bound for signed integrands in Remark A.1; everything else is either carefully proved or reasonably delegated. If the author supplies a complete proof of that lower bound, or a precise reference where it is proved in the required generality, I would support acceptance."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The paper does what the title says: it removes the first-order and Morrey-bound restrictions from Báia–Matias–Santos and gives dual and tangent/local characterizations plus area-strict density theorems for general constant-rank A-free and B-gradient Young measures. That is a real step forward, and the proof is not just a rehash of the gradient or symmetric-gradient cases. The convexity proof for YA,0(µ,Ω), the separation arguments, and the Besicovitch gluing are substantial and, as far as I can see, correctly assembled. I also give credit for the applications: the failure of L1-compactness examples are informative and correctly tied to the unconstrained singular part.\n\nMy main reservation is exactly where the stress-test points: the sufficiency direction of Theorem 1.1 relies on Propositions 9.2 and 9.3, and those rely on the relaxation theorem in Appendix A. The stated Theorem A.1 assumes f ≥ 0, but the separating integrands in Section 9 are not nonnegative. Remark A.1 asserts the result extends to f ∈ E(Ω;W) under (A)–(C), and item 7 says positivity was only used to stop negative boundary concentration, with assumption (c) removing that need. That is plausible, but it is a sketch, not a proof, and the lower bound G[µ] ≥ ∫Q_A f dx + ∫(Q_A f)# d|µ_s| is the one place where signed integrands can genuinely bite. If that bound fails, the contradictions in Propositions 9.2 and 9.3 collapse. I do not see an actual error, but I would want a referee to check this point carefully before recommending acceptance. It is a load-bearing step, not a cosmetic gap.\n\nOther soft spots are minor: the constant-rank hypothesis is essential but explicit, and the citation pattern leans on the author's own prior work, though those results are published and independently checkable. The separation of singular cases in Theorem 9.1 is intricate; I did not verify every estimate, but the structure is coherent.\n\nWho is this for? Anyone working in compensated compactness, L1-relaxation, or microstructure under PDE constraints. It is a serious paper that deserves real refereeing. My recommendation: send it to a knowledgeable referee, and ask that referee specifically to check Remark A.1 and the lower-bound claim for signed integrands.","headline":"A genuinely new and largely convincing characterization of A-free generalized Young measures for arbitrary-order constant-rank operators; the main caveat is that a load-bearing relaxation result is only sketched for the signed integrands actually used in the Hahn–Banach argument.","tokens_in":67032,"tokens_out":1195,"would_cite":true,"duration_ms":16676,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["49J45","49Q15","46G10","35B05"],"pacs":[],"model":"deepseek-v4-flash","headline":"This paper characterizes generalized Young measures generated by A-free measures and their B-gradient counterparts, under constant rank, by a barycenter condition, a Jensen-type inequality against A-quasiconvex integrands, and a wave-cone…","keywords":["A-free measure","generalized Young measure","constant rank operator","compensated compactness","concentration","oscillation","two-state problem","PDE constraint"],"falsifier":"The constant-rank assumption is testable through the diagonal-gradient operator $\\mathcal A(w_1,w_2)=(\\partial_2w_1,\\partial_1w_2)$, which violates the constant-rank condition: if one can exhibit a triple $(\\nu,\\lambda,\\nu^\\infty)$ satisfying conditions (i)-(iii) of Theorem 1.1 but not generated by any $\\mathcal A$-free sequence, the characterization does not extend beyond constant rank, while a proof that no such triple exists would indicate the assumption is only technical.","tokens_in":65930,"feed_emoji":"📐","tokens_out":7937,"duration_ms":77853,"temperature":0.7,"pith_summary":"The paper proves a complete characterization of generalized Young measures generated by sequences of $\\mathcal A$-free measures, for linear homogeneous operators satisfying the constant rank property. A generalized Young measure records both the oscillations and the concentrations of a weakly convergent sequence; the characterization says the record is valid exactly when three conditions hold: the barycenter is itself an $\\mathcal A$-free measure, a Jensen-type inequality holds against every $\\mathcal A$-quasiconvex integrand with linear growth, and the singular concentration directions lie in the span of the wave cone. The same theorem has a counterpart for $\\mathcal B$-gradient measures, where $\\mathcal A$-quasiconvexity is replaced by $\\mathcal B$-gradient quasiconvexity. This closes the $p=1$ gap left after earlier characterizations for gradients and symmetric gradients, and it supplies the missing duality framework behind compensated compactness when mass concentration is allowed.","feed_headline":"Three conditions classify A-free Young measures","feed_subtitle":"Duality with A-quasiconvex integrands decides which oscillations and concentrations are possible","key_machinery":"The load-bearing machinery has three parts. The constant-rank property, $\\mathrm{rank}\\,\\mathcal A(\\xi)=r$ for all $\\xi\\neq0$, makes the orthogonal projection onto $(\\ker\\mathcal A(\\xi))^\\perp$ an analytic homogeneous multiplier, so the $\\mathcal A$-representative of a measure obeys Sobolev estimates of Fonseca-M\\\"uller type and can be localized with commutator errors. The wave cone $\\Lambda_{\\mathcal A}=\\bigcup_{\\xi\\neq0}\\ker\\mathcal A(\\xi)$ and its span $W_{\\mathcal A}$ encode which Fourier amplitudes can oscillate or concentrate inside the constraint $\\mathcal A\\mu=0$; at singular points the concentration directions are forced into $W_{\\mathcal A}$. The third ingredient is the exact potential theorem for constant-rank operators, $\\mathrm{Im}\\,\\mathcal B(\\xi)=\\ker\\mathcal A(\\xi)$, which turns $\\mathcal A$-free objects into $\\mathcal B$-gradients after removing a compact commutator error, and thereby transfers the characterization from $\\mathcal A$-free to $\\mathcal B$-gradient Young measures.","core_discovery":"Under the constant-rank assumption, a triple $(\\nu,\\lambda,\\nu^\\infty)$ with no boundary concentration is a generalized $\\mathcal A$-free Young measure if and only if there is $\\mu\\in M(\\Omega;W)$ with $\\mathcal A\\mu=0$ and $\\mu=\\langle\\mathrm{id},\\nu\\rangle L^d+\\langle\\mathrm{id},\\nu^\\infty\\rangle\\lambda$, the inequality $h(\\mu_{\\mathrm{ac}}(x))\\le\\langle h,\\nu_x\\rangle+\\langle h^\\#,\\nu^\\infty_x\\rangle\\lambda_{\\mathrm{ac}}(x)$ holds almost everywhere for every upper semicontinuous $\\mathcal A$-quasiconvex $h$ with linear growth, and $\\mathrm{supp}(\\nu^\\infty_x)\\subset W_{\\mathcal A}:=\\mathrm{span}\\,\\Lambda_{\\mathcal A}$ at $\\lambda^s$-almost every $x$. The same statement with $\\mathcal B$-gradient quasiconvexity and barycenter $\\mathcal B u$ characterizes generalized $\\mathcal B$-gradient Young measures. The proof obtains this by a local tangent-measure criterion, an area-strict approximation theorem for arbitrary bounded open sets, and a convexity result for the class of $\\mathcal A$-free Young measures with fixed barycenter, followed by a Hahn-Banach separation argument.","pith_inferences":["If Theorem 1.1 is pushed one step further, the tangent-cone criterion of Theorem 1.2 suggests the same three conditions may survive up to the boundary once tangent concentrations on $\\partial\\Omega$ are controlled; the paper only treats $\\lambda(\\partial\\Omega)=0$.","The area-strict approximation theorem implies that relaxation formulas of the type proved for gradients should hold on arbitrary bounded open domains without star-shapedness; this is a consequence of Theorem 1.3 that the paper does not state as a separate relaxation result.","One can test the sharpness of the constant-rank assumption by looking for a non-constant-rank operator where conditions (i)-(iii) hold but generation fails; the diagonal-gradient example in Section 1.2 is the natural first candidate.","The two-state failure suggests that in the presence of concentrations, the relevant invariant is the span of the wave cone rather than the absence of wave-cone connections; a three-state variant with concentrations along a wave-cone-free segment would probe this distinction."],"forward_implications":["Every bounded $\\mathcal A$-free measure on a bounded open set is the area-strict limit of smooth $\\mathcal A$-free functions, with no star-shapedness or Lipschitz boundary condition on the domain.","For full-rank elliptic $\\mathcal A$, the only generalized $\\mathcal A$-free Young measures are elementary triples $(\\delta_w,0,q)$ with $\\mathcal Aw=0$; there is no room for oscillations or concentrations.","At the potential level, a generalized $\\mathcal B$-gradient Young measure is characterized by $\\mathcal B$-gradient quasiconvexity and a barycenter $\\mathcal B u$, so $\\mathcal B$-gradient and $\\mathcal A$-free Young measures differ only by their barycenter structure under the exactness identity.","$L^1$-compensated compactness fails once concentrations are allowed: $\\mathcal A$-free sequences can have values approaching a set away from wave-cone connections while failing weak $L^1$ convergence and equi-integrability, as in the two-state problem.","Divergence-free generalized Young measures are constrained only by the divergence-free barycenter condition, because every convex integrand is div-quasiconvex."],"supporting_citations":[{"why":"Defines A-quasiconvexity and supplies the Fonseca-M\\\"uller projection estimates that the paper extends to generalized measures.","marker":"[28]"},{"why":"Proves the exact potential identity Im B(\\xi)=ker A(\\xi) used to pass from A-free to B-gradient statements.","marker":"[57]"},{"why":"Gives the structure theorem for A-free measures that pins down singular concentration directions.","marker":"[22]"},{"why":"Establishes the p=1 generalized Young measure characterization for gradients that this paper generalizes.","marker":"[42]"},{"why":"Supplies the earlier partial A-free characterization and the weak-star closedness lemma reused here.","marker":"[12]"},{"why":"Provides the rigidity for positively homogeneous rank-one convex functions used for the singular-part inequalities.","marker":"[39]"},{"why":"Introduced generalized Young measures as triples (nu,lambda,nu^infty), the formalism of the paper.","marker":"[25]"},{"why":"Extends generalized Young measure theory and the non-uniform integrability criterion used in applications.","marker":"[4]"}],"fun_headline_variants":["A-free Young measures fully characterized by duality","Constant-rank proof: duality fixes A-free Young measures","L1 estimates fail: A-free Young measures show rigidity","Area-strict approximation gives A-free Young measure test","Triple condition classifies all A-free Young measures"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is the constant-rank property, that the rank of $\\mathcal A(\\xi)$ is the same for every nonzero direction $\\xi$; if the rank jumps with direction, as for the diagonal-gradient operator described in Section 1.2, the projection and potential theorems that carry the proof are not available and the characterization is not claimed.","fun_headline_variants_meta":{"raw":{"variants":["A-free Young measures fully characterized by duality","Constant-rank proof: duality fixes A-free Young measures","L1 estimates fail: A-free Young measures show rigidity","Area-strict approximation gives A-free Young measure test","Triple condition classifies all A-free Young measures"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000198,"raw_usage":{"total_tokens":1429,"prompt_tokens":1070,"completion_tokens":359,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":686,"completion_tokens_details":{"reasoning_tokens":284}},"tokens_in":686,"tokens_out":359,"duration_ms":3700,"temperature":1.0,"reasoning_tokens":284,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T14:20:57.554043+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"The constant-rank assumption is testable through the diagonal-gradient operator $\\mathcal A(w_1,w_2)=(\\partial_2w_1,\\partial_1w_2)$, which violates the constant-rank condition: if one can exhibit a triple $(\\nu,\\lambda,\\nu^\\infty)$ satisfying conditions (i)-(iii) of Theorem 1.1 but not generated by any $\\mathcal A$-free sequence, the characterization does not extend beyond constant rank, while a proof that no such triple exists would indicate the assumption is only technical.","supporting_citations":[{"cited_title":"Fonseca and S","cited_arxiv_id":null,"evidence_quote":"Defines A-quasiconvexity and supplies the Fonseca-M\\\"uller projection estimates that the paper extends to generalized measures."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Proves the exact potential identity Im B(\\xi)=ker A(\\xi) used to pass from A-free to B-gradient statements."},{"cited_title":"De Philippis and F","cited_arxiv_id":null,"evidence_quote":"Gives the structure theorem for A-free measures that pins down singular concentration directions."},{"cited_title":"Kristensen and F","cited_arxiv_id":null,"evidence_quote":"Establishes the p=1 generalized Young measure characterization for gradients that this paper generalizes."},{"cited_title":"Ba ´ ıa, J","cited_arxiv_id":null,"evidence_quote":"Supplies the earlier partial A-free characterization and the weak-star closedness lemma reused here."},{"cited_title":"Kirchheim and J","cited_arxiv_id":null,"evidence_quote":"Provides the rigidity for positively homogeneous rank-one convex functions used for the singular-part inequalities."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Introduced generalized Young measures as triples (nu,lambda,nu^infty), the formalism of the paper."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Extends generalized Young measure theory and the non-uniform integrability criterion used in applications."}],"review_version":1}