{"id":"e6e46a9d-ec20-445a-a13e-b8ff6981d9fd","arxiv_id":"1908.05714","paper_version":1,"verdict":"CONDITIONAL","confidence":"HIGH","novelty_score":5.0,"correctness_risk":"low","formal_verification":"none","parameter_count":0,"one_line_summary":"Under the law of demand, a continuous demand mapping on a convex domain is injective if and only if it is not constant along any non-singleton line segment.","lead":"This paper gives a simple way to test whether a demand system is one-to-one: under a condition called the law of demand, you only need to check that the mapping is not flat along any straight line. The result gives economists a practical check for when demand can be inverted to recover preferences or market fundamentals.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Central Proposition 1 is sound; the printed Proposition 5 is false as stated and Example 3 has an arithmetic slip.","rationale":"The stress-test confirms the reader's high confidence in Proposition 1. The convexity of inverse images is the load-bearing step, and it is correct: for a,b with Q(a)=Q(b)=y and any z on the segment between them, monotonicity with a and b gives (Q(z)-y)·(b-a)=0. If Q(z)≠y, taking a small perturbation of a in the direction Q(z)-y and using continuity at a produces a violation of monotonicity, so Q(z)=y. Thus the equivalence (iii) => (i) is sound, and the reader is right that the law of demand is the genuine scope condition: it is exactly the monotonicity that makes preimages convex, and it can fail for non-quasilinear demand with income effects. That is a limitation of applicability, not an internal inconsistency.\n\nWhere the manuscript is actually defective is in an extension. Proposition 5's printed condition (iii) is about the wrong mapping and the wrong domain. The counterexample with Q equal to the identity and f(x,y)=(x,x) shows the printed equivalence is false; the proof's own logic indicates the condition should be stated for ~Q on T. Example 3 also contains an arithmetic error, though it does not change the conclusion. Both are local corrections. The central claim stands, so the reader's CONDITIONAL verdict is unchanged.","tokens_in":11624,"tokens_out":23781,"duration_ms":242748,"concrete_test":"Run the following check on Proposition 5 as printed: set T=U=R^2, Q(u)=u, f(x,y)=(x,x). Assumption 3 holds (Q satisfies the law of demand and is continuous; f is affine; U and T are open and convex). The printed condition (iii) holds because Q has no nondegenerate constant line segments, yet ~Q(x,y)=(x,x) is not injective. If this counterexample checks out, condition (iii) must be revised to refer to T and ~Q. Separately, recompute Example 3 at u=(1,2): Q = (20 - 80, -1 + 16) = (-60,15), not (-60,7); the inner product is -30, so the conclusion that the law of demand fails is unchanged.","verdict_should_be":"UNCHANGED","load_bearing_attack":"I find no defect that threatens the central equivalence in Proposition 1. Continuity plus the law of demand indeed makes every preimage convex (Lemma 1), and the three-way equivalence follows directly. The law of demand itself is a scope condition, not a hidden flaw: it is exactly monotonicity of Q, and demand systems with income effects can violate it, but within the assumption the argument is correct.\n\nTwo local corrections are needed. First, in Example 3, Q(1,2) is computed as (-60,7); the correct value is (20*1 + (-10)*8, -1*1 + 2*8) = (-60,15), giving an inner product of -30 rather than -46. The example still works, but the arithmetic is wrong. Second, and more substantively, the statement of Proposition 5 is false as printed. Condition (iii) is written as 'the only line segments in U along which Q is constant are points,' but the proposition concerns the composed map ~Q = Q∘f on T, and the condition must be about ~Q on T. As printed, take T=U=R^2, Q(u)=u, and f(x,y)=(x,x). Then ~Q(x,y)=(x,x), which is neither injective nor locally injective, yet Q has no nondegenerate constant line segments, so the printed condition (iii) holds. Thus the printed equivalence fails. The intended condition, 'the only line segments in T along which ~Q is constant are points,' is what the convex-preimage argument proves.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper develops necessary and sufficient conditions for global injectivity of a demand mapping Q: U ⊆ R^K → R^K under a monotonicity condition called the law of demand: (Q(u) − Q(ũ)) · (u − ũ) ≥ 0. Under continuity and openness/convexity of U, the paper's Proposition 1 shows that global injectivity, local injectivity, and the absence of non-degenerate line segments on which Q is constant are equivalent. This reduces a global injectivity check to a local or line-segment check. The paper also connects the law of demand to weak quasi-definiteness of the Jacobian, presents a directional-derivative characterization of injectivity (Proposition 3), compares the approach to Berry, Gandhi, and Haile's connected substitutes condition, discusses quasilinear utility and additive random utility models, and extends the line-segment criterion to settings where the law of demand holds after an affine or homeomorphic change of variables.","tokens_in":11902,"tokens_out":3459,"duration_ms":35117,"significance":"If the main results hold, the paper provides a clean, parameter-free criterion for injectivity in a class of demand systems allowing complementarity, complementing existing sufficient conditions such as connected substitutes. The central equivalence in Proposition 1 is sound and is a useful nondifferentiable counterpart to Gale and Nikaido's Jacobian-based univalence theorems. The paper also explicitly credits classical monotone-operator theory, and the main argument does not rely on circular reasoning or fitted parameters. The law-of-demand assumption is a genuine scope condition: for non-monotone maps such as Q(u)=u² on R, local injectivity does not imply global injectivity, so the line-segment test is valid only within the stated monotonicity framework. The paper's significance is moderate but appropriate for an econometrics journal, given the role of injectivity in identification.","major_comments":[{"comment":"The printed equivalence (iii) is false as stated. Condition (iii) says 'The only line segments in U along which Q is constant are points,' but the proposition concerns the composed map Q̃ = Q∘f on T, and the condition must be about Q̃ on T. A concrete counterexample is T = U = R², Q(u) = u, and f(x,y) = (x,x). Then Q̃(x,y) = (x,x) is neither injective nor locally injective, yet Q has no nondegenerate constant line segments, so the printed condition (iii) holds while (i) and (ii) fail. The intended condition, 'the only line segments in T along which Q̃ is constant are points,' is what the convex-preimage argument establishes and should replace the printed condition.","section":"Section 7, Proposition 5"},{"comment":"Example 3 contains an arithmetic error. For u = (0,0) and ũ = (1,2), the matrix product gives Q(ũ) = (20·1 + (−10)·8, −1·1 + 2·8) = (−60, 15), not (−60, 7). Consequently the displayed inner product is (−60, 15)·(1,2) = −30, not −46. The conclusion that the law of demand fails is unchanged, but the numbers should be corrected.","section":"Section 5, Example 3"},{"comment":"Corollary 2 is stated without proof, and it is load-bearing because Proposition 3 relies on the equivalence between (i) and (iii) from Corollary 2. The proof is likely the same line-segment argument as Corollary 1, but it should be written out. In Proposition 3, the proof of '(iii) iff (iv)' is compressed to a bare reference to the mean value theorem; since (iv) involves directional derivatives vanishing pointwise along the whole segment, the MVT argument should be spelled out, including the coordinate-wise application needed for a vector-valued demand mapping.","section":"Section 2, Corollary 2 and Section 3, Proposition 3"},{"comment":"Lemma 3 is cited to Rockafellar with the proof omitted. The result is used to draw a sharp equivalence between differentiability of C at Q(u) and singleton inverse images. Given that the manuscript's own contribution includes this lemma as a specialization, a proof or a precise pointer to the specific theorems in Rockafellar (Theorems 23.5 and 25.1) should be supplied, rather than 'proof omitted.'","section":"Section 6, Lemma 3"}],"minor_comments":[{"comment":"There are several typos: 'mapp ing' in the abstract, 'substitability' in Section 5, 'and and' in Proposition 3 and Corollary 2, and 'denoting denoting' in Example 4. These should be corrected.","section":"Abstract and throughout"},{"comment":"The notation '1tu P Au' is unusual; it would be clearer to write '1_{u ∈ A}' or explain that 1{·} is the indicator function in the text.","section":"Example 1"},{"comment":"The reference to 'Berry et al. [2013]' in Section 8 should be 'Berry, Gandhi, and Haile [2013]' for consistency with the reference list.","section":"Section 5"},{"comment":"The proof says 'by Proposition 3 because its directional derivatives are never zero,' but Proposition 3 appears later and the connection relies on condition (iv). Please add a forward reference or restate the relevant condition.","section":"Section 3, proof of Proposition 2"}],"recommendation":"minor_revision","confidential_remarks":"The central result (Proposition 1) is correct and the paper is within scope for an econometrics journal. The main issue is the erroneous Proposition 5 in Section 7; it is local and fixable, as is the arithmetic slip in Example 3. I do not see any threat to the paper's core contribution, so I recommend minor revision rather than rejection."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The central result is the real thing. Under continuity and the law of demand on an open convex domain, global injectivity, local injectivity, and the absence of non-degenerate constant line segments are equivalent. The proof is a clean application of convexity of preimages for monotone operators, and the paper is honest that Lemma 1 is classical. What is new and useful is the translation into an econometric injectivity test that allows complementarity, which the connected-substitutes route does not handle. The non-convex domain extension (Proposition 2) and the directional-derivative characterization (Proposition 3) are genuine additions. I also give credit for the clear comparison with Berry-Gandhi-Haile and the non-nesting examples; the citation pattern is appropriate and there is no circularity.\n\nThe paper has two local defects that a referee should catch. First, Proposition 5 as printed is false. Condition (iii) refers to line segments in U along which Q is constant, but the claim is about the composed map ~Q = Q∘f on T. With T=U=R^2, Q(u)=u, and f(x,y)=(x,x), the composed map is not injective yet Q has no non-degenerate constant segments, so the printed equivalence fails. The intended condition is about ~Q on T, and then the convex-preimage argument works; this is a misstatement rather than a deep flaw. Second, Example 3 computes Q(1,2) as (-60,7); the correct value is (-60,15), giving inner product -30 rather than -46. The example still goes through, but the arithmetic should be fixed. Also, a few proofs are compressed: Corollary 2 is asserted without proof, the (iii)⇔(iv) step in Proposition 3 leans on an unstated mean value theorem application, and Lemma 3 is cited with proof omitted. None of these threaten the main equivalence, but a refereed version should expand them.\n\nThe law of demand is a genuine scope condition—demand systems with strong income effects can violate it—but the paper states this clearly and uses it as an assumption rather than hiding it. Within that assumption, the argument is correct.\n\nThis is a useful paper for applied econometricians working on demand invertibility with complementarity, and for discrete-choice empiricists. It deserves serious refereeing: the central result is worth publishing, and the local errors are fixable. I would recommend conditional accept after a revision that corrects Proposition 5, fixes the example, and fleshes out the compressed proofs.","headline":"Sound and useful core equivalence—global injectivity under the law of demand reduces to checking no non-degenerate constant line segment—but Proposition 5 is misstated as printed and Example 3 has an arithmetic slip.","tokens_in":12402,"tokens_out":1899,"would_cite":true,"duration_ms":18726,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"A continuous demand mapping that satisfies the law of demand on an open convex domain is globally injective exactly when it is locally injective, which holds precisely when no non-degenerate line segment maps to a single point.","keywords":["injectivity","law of demand","demand mapping","global univalence","quasi-definite Jacobian","convex inverse images","connected substitutes","identification"],"falsifier":"Take a continuous mapping $Q:U\\subset\\mathbb{R}^2\\to\\mathbb{R}^2$ satisfying $(Q(u)-Q(\\tilde u))\\cdot(u-\\tilde u)\\ge 0$ on an open convex $U$, and search for a point $y$ whose preimage $Q^{-1}(y)$ contains two distinct points but not the segment joining them. Finding even one such example would falsify Lemma 1 and the equivalence between global injectivity and the line-segment condition.","tokens_in":11414,"feed_emoji":"📈","tokens_out":13883,"duration_ms":109914,"temperature":0.7,"pith_summary":"This paper shows that for a continuous demand mapping $Q:U\\subseteq\\mathbb{R}^K\\to\\mathbb{R}^K$ satisfying the law of demand, $(Q(u)-Q(\\tilde u))\\cdot(u-\\tilde u)\\ge 0$, on an open convex domain $U$, global injectivity is equivalent to local injectivity, and both are equivalent to the absence of any non-degenerate line segment on which $Q$ is constant. The result matters because many identification and estimation methods require the demand mapping to be one-to-one, and checking global injectivity directly is hard; the paper reduces it to a local or one-dimensional check. The proof relies on the fact that under the law of demand, the inverse image of any quantity vector is convex, so two inputs with the same output force the whole segment between them to have the same output. In the differentiable case, the paper obtains necessary and sufficient conditions in terms of directional derivatives, generalizing classical conditions for quasi-definite Jacobians.","feed_headline":"Global demand injectivity is just a local line-segment check","feed_subtitle":"For continuous demand maps obeying the law of demand, one-to-one checking reduces to finding no flat segment.","key_machinery":"The load-bearing object is the convexity of inverse images: for a continuous monotone map $Q$ on an open convex domain, the set $Q^{-1}(y)=\\{u\\in U:Q(u)=y\\}$ is convex (Lemma 1, a known result in monotone operator theory). This property carries the argument because if two distinct points map to the same $y$, convexity forces every point on the segment between them to map to $y$, creating a non-degenerate line segment of constancy. The law of demand provides the monotonicity that makes the preimage convex; differentiability then translates the 'no constant segment' condition into a directional-derivative condition stating that $Q'(u+\\lambda v,v)$ cannot be the zero vector for all $\\lambda$ in $[0,1]$.","core_discovery":"The central claim is Proposition 1: under Assumption 1, the following are equivalent: $Q$ is injective; $Q$ is locally injective; and the only line segments in $U$ along which $Q$ is constant are singleton points. The engine is Lemma 1, which states that every inverse image $Q^{-1}(y)$ is convex: if $Q(u)=Q(\\tilde u)$, then the line segment from $u$ to $\\tilde u$ lies in the inverse image. Consequently, a global collision between two distinct inputs implies a whole interval of collisions, so verifying injectivity reduces to checking that no such interval exists. For differentiable $Q$, the law of demand is exactly the weak quasi-definiteness of the Jacobian, and the paper derives a necessary and sufficient directional-derivative condition for injectivity that relaxes the usual Jacobian-invertibility requirement, along with extensions to restrictions of $Q$ to open or convex subsets and to demand systems that satisfy the law of demand after a homeomorphic or affine change of variables.","pith_inferences":["A natural extension is to turn the line-segment condition into a nonparametric empirical test: observing the same demand at two distinct utility vectors whose connecting segment lies in the domain would directly reject injectivity.","Because the law of demand is not ordinal, a monotone reparameterization can restore it, so the segment test might apply after a transformation even when the raw demand map fails the law of demand; the paper itself establishes local-global equivalence under homeomorphic changes and the full segment equivalence under affine changes.","In additive random utility models, aggregate choice probabilities inherit the law of demand from individual choices, so the segment test could be used to check whether the probability mapping is one-to-one in utility indices; the paper notes that injectivity depends on the distribution of heterogeneity."],"forward_implications":["For any continuous demand map satisfying the law of demand on an open convex domain, checking global injectivity is equivalent to checking that no non-degenerate line segment is mapped to a single point.","Local injectivity at every point implies global injectivity, so identification arguments can be built from local conditions on demand systems.","In the differentiable case, Jacobian invertibility is sufficient but not necessary for injectivity; the precise condition is that no directional derivative vanishes along a whole segment.","The equivalence carries over to restrictions of the demand map to open or convex subsets, so injectivity can be verified on a region of interest using the same test.","The law-of-demand condition and the connected-substitutes condition are not nested, so the line-segment test covers demand systems with complementarity that connected-substitutes methods cannot handle."],"supporting_citations":[{"why":"Supplies the classical global-univalence theorem for quasi-definite Jacobians that the differentiable results generalize.","marker":"Gale and Nikaido [1965]"},{"why":"Provides the textbook result that inverse images of monotone operators are convex, used in Lemma 1 for the full-space case.","marker":"Rockafellar and Wets [2009]"},{"why":"Extends convexity of preimages of monotone operators to restricted domains, cited for Lemma 1 when the domain is not all of $\\mathbb{R}^K$.","marker":"Kassay, Pintea, and Szenkovits [2009]"},{"why":"Establishes injectivity under connected substitutes; the paper compares its law-of-demand condition with this competing shape restriction.","marker":"Berry, Gandhi, and Haile [2013]"},{"why":"Supplies conditions under which aggregate demand obeys the law of demand, one of the main economic justifications for Assumption 1.","marker":"Hildenbrand [1983]"},{"why":"Shows additive random utility models satisfy the law of demand, the application used in Example 4.","marker":"McFadden [1981]"},{"why":"Provides the equivalence between the law of demand and weakly quasi-definite Jacobians used in Lemma 2.","marker":"Parthasarathy [2006]"},{"why":"Underlies Lemma 3 connecting differentiability of the aggregator to injectivity of quasilinear demand.","marker":"Rockafellar [1970]"}],"fun_headline_variants":["Global injectivity? Just check for flat line segments","No flat segments, no collisions: demand injectivity","Demand injectivity localizes to line segments","Law of demand turns injectivity into line check"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that the demand mapping obeys the law of demand, meaning $(Q(u)-Q(\\tilde u))\\cdot(u-\\tilde u)\\ge 0$ for every pair in the domain; this monotonicity is what forces inverse images to be convex, and without it local injectivity need not imply global injectivity, as the example $Q(u)=u^2$ on $\\mathbb{R}$ shows.","fun_headline_variants_meta":{"raw":{"variants":["Global injectivity? Just check for flat line segments","No flat segments, no collisions: demand injectivity","Demand injectivity localizes to line segments","Law of demand turns injectivity into line check"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.00039,"raw_usage":{"total_tokens":1987,"prompt_tokens":810,"completion_tokens":1177,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":426,"completion_tokens_details":{"reasoning_tokens":1116}},"tokens_in":426,"tokens_out":1177,"duration_ms":9552,"temperature":1.0,"reasoning_tokens":1116,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T13:07:52.075177+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Take a continuous mapping $Q:U\\subset\\mathbb{R}^2\\to\\mathbb{R}^2$ satisfying $(Q(u)-Q(\\tilde u))\\cdot(u-\\tilde u)\\ge 0$ on an open convex $U$, and search for a point $y$ whose preimage $Q^{-1}(y)$ contains two distinct points but not the segment joining them. Finding even one such example would falsify Lemma 1 and the equivalence between global injectivity and the line-segment condition.","supporting_citations":[{"cited_title":"The jacobian matrix and global univalence of mappings","cited_arxiv_id":null,"evidence_quote":"Supplies the classical global-univalence theorem for quasi-definite Jacobians that the differentiable results generalize."},{"cited_title":"On convexity of preimages of monotone operators","cited_arxiv_id":null,"evidence_quote":"Extends convexity of preimages of monotone operators to restricted domains, cited for Lemma 1 when the domain is not all of $\\mathbb{R}^K$."},{"cited_title":"Connected substitutes and invertibility of demand","cited_arxiv_id":null,"evidence_quote":"Establishes injectivity under connected substitutes; the paper compares its law-of-demand condition with this competing shape restriction."},{"cited_title":"law of demand","cited_arxiv_id":null,"evidence_quote":"Supplies conditions under which aggregate demand obeys the law of demand, one of the main economic justifications for Assumption 1."},{"cited_title":"Econometric models of probabilistic choice","cited_arxiv_id":null,"evidence_quote":"Shows additive random utility models satisfy the law of demand, the application used in Example 4."},{"cited_title":"On global univalence theorems, volume 977 of Lecture Notes in Mathematics","cited_arxiv_id":null,"evidence_quote":"Provides the equivalence between the law of demand and weakly quasi-definite Jacobians used in Lemma 2."},{"cited_title":"Convex Analysis","cited_arxiv_id":null,"evidence_quote":"Underlies Lemma 3 connecting differentiability of the aggregator to injectivity of quasilinear demand."}],"review_version":1}