{"id":"f87a7a99-d15c-49b4-a7e8-46481fd3e689","arxiv_id":"2411.17883","paper_version":3,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"Under the independence axiom, an expected utility representation exists if and only if there are n indifferent lotteries whose differences span a hyperplane in the n-dimensional simplex.","lead":"This short theory paper replaces the usual continuity assumption in expected utility theory with a geometric condition: the decision maker must be indifferent between enough lotteries to span a hyperplane. Under the independence axiom, this condition is equivalent to standard continuity and to the existence of an expected utility representation.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The theorem is plausible, but the written proof of (iii)=>(i) contains an algebraic error in the indifference-hyperplane construction; until corrected, the claimed equivalence is not fully proved.","rationale":"The paper's central claim — that under Independence, the geometric Indifferent Points axiom is equivalent to Strong Continuity and to expected utility — is likely correct. The Independence Axiom is the right structural assumption; the examples and surrounding independence results are coherent. However, the written proof is not yet complete. The (iii)=>(i) direction contains a concrete algebraic mistake: the displayed linear system does not describe the indifference hyperplane for the given utility function, and the proposed uniform lottery does not solve it. This is not a mere typo in an unimportant direction; it is the entire construction showing that expected utility implies IP. A repair is straightforward by writing the correct linear equation and choosing points near the uniform lottery, but as published the proof does not establish the implication. Lemma A2's proof also has a logical jump: the indifference q ∼ αq + (1-α)p does not by itself imply p∼q without an additional line-argument excluding strict preference in either direction. That gap is also fixable using the midpoint argument and Lemma A4. I therefore do not see a reason to reject the theorem, but I also cannot accept the proof as written. The reader's CONDITIONAL verdict is appropriate; my concern is not about the Independence Axiom but about these proof details, so my reading leaves the verdict unchanged.","tokens_in":5341,"tokens_out":23154,"duration_ms":204441,"concrete_test":"Independently re-derive (iii)=>(i) for the case n=2 (three prizes) with utilities u(x0)=0, u(x1)=1, u(x2)=2. Use the correct equation Σ pi·(ui-u0) = u - u0, i.e., p1 + 2p2 = 1, and exhibit two lotteries in the simplex satisfying it (e.g., (p1,p2) = (1/2,1/4) and (1/4,3/8)) with distinct vectors. Verify that their difference is nonzero, so IP holds. Then generalize: show the uniform lottery lies on the indifference hyperplane and that a basis of its kernel yields n affinely independent points in the simplex.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The most load-bearing defect is in the proof of Theorem 1, direction (iii)=>(i) (Appendix A, final paragraph). The author defines u as the average utility over prizes and then proposes the linear system M·(p1,...,pn)ᵀ = (u,1)ᵀ with first row (u(x1),...,u(xn)) and second row (1,...,1). This system ignores u(x0): for a lottery p, EU(p) = p0·u(x0) + Σ pi·u(xi) = u(x0) + Σ pi·(u(xi)-u(x0)), so the correct indifference equation is Σ pi·(u(xi)-u(x0)) = u - u(x0). The displayed second row Σ pi = 1 is also wrong, since the simplex allows Σ pi ≤ 1 with p0 = 1 - Σ pi; indeed the proposed uniform solution pi = 1/(n+1) gives Σ pi = n/(n+1) < 1 and does not satisfy the displayed system. Thus the constructed points are not shown to be indifferent, and the 'if' direction of the equivalence is unsupported. A secondary gap appears in Lemma A2: after deriving q ∼ αq + (1-α)p, the proof immediately concludes p∼q; this needs the line-strictness Lemma A4 (or a midpoint argument) to rule out p≻q and p≺q. Both issues are fixable — the correct indifference hyperplane exists and Lemma A2 can be proved by a midpoint IA argument — so the central theorem is likely correct, but the written proof is not complete.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"This paper proposes a geometric axiom, Indifferent Points (IP), as a substitute for topological continuity in the expected utility theorem for finite prize spaces. With X = {x_0, ..., x_n} and L identified with the n-dimensional simplex in R^n, IP requires n indifferent lotteries p_1 ∼ ... ∼ p_n whose affine hull is a hyperplane in R^n (for three prizes, a single binary indifference suffices). Theorem 1 claims that, under the Independence Axiom, IP is equivalent to Strong Continuity (SC) and to the existence of an expected utility representation. The proof is built on auxiliary lemmas (A1–A4) that establish convexity and line-degeneracy of indifference classes, then shows (i)⇒(ii) by identifying the strictly better set with an open halfspace, (ii)⇒(iii) via the standard Mas-Colell result, and (iii)⇒(i) by constructing indifferent points from the utility representation. Proposition 1 positions IP relative to strong continuity, mixture continuity, and (weak Wold-)solvability, and Example 1 shows that IP alone, without IA, implies none of the standard continuity axioms.","tokens_in":5616,"tokens_out":35716,"duration_ms":299129,"significance":"If the main theorem is correct, this is a clean structural contribution: in the finite-prize expected utility theorem, the topological content of continuity is replaced by a single, finitely checkable geometric condition on the indifference set — the existence of an indifference hyperplane. The axiom is behavioral in nature, the paper is self-contained in its main line (with the (ii)⇒(iii) direction appropriately deferred to the standard Mas-Colell representation theorem), and the paper introduces no free parameters or auxiliary entities. The mapping of IP against solvability, mixture continuity, and strong continuity (Proposition 1 and Example 1) is a useful addition to the axiomology of decision theory. However, the written proof has two defects in load-bearing places: the indifference-hyperplane construction in direction (iii)⇒(i) is algebraically erroneous in several respects, and Lemma A2's final inference is under-justified. In addition, the abstract overclaims logical independence of IP from Weak Continuity and the Archimedean axiom without providing the required reverse example. All of these issues appear repairable within the manuscript's scope.","major_comments":[{"comment":"The indifference-hyperplane construction is not correct as written, in three ways. First, the first row of the displayed system ignores u(x_0): under the identification of footnote 1, p(x_0) = 1 − Σ_i p_i, so EU(p) = u(x_0) + Σ_i (u(x_i) − u(x_0)) p_i, and indifference at utility level \\bar u = (1/(n+1))Σ_k u(x_k) is the single affine equation Σ_i (u(x_i) − u(x_0)) p_i = \\bar u − u(x_0), not Σ_i u(x_i) p_i = \\bar u. Second, the second row Σ p_i = 1 is inconsistent with the simplex geometry of footnote 1, where Σ_{i=1}^n p_i ≤ 1; the proposed particular solution \\bar p = (1/(n+1), ..., 1/(n+1)) has Σ \\bar p_i = n/(n+1) ≠ 1 and is therefore not a solution of the displayed system. Third, the nullity claim is wrong: for the 2 × n matrix M, rank(M) ≤ 2 gives dim ker(M) ≥ n − 2, not n − 1, so the proof cannot supply the n − 1 kernel vectors b_2, ..., b_n needed to obtain n points spanning a hyperplane. The intended statement is nevertheless true: using the corrected single affine equation (with a trivial separate sentence for the constant-utility case), the kernel of the 1 × n row has dimension n − 1, and perturbing the uniform lottery along n − 1 independent kernel vectors at small ε yields the required indifferent points. This paragraph must be rewritten.","section":"Appendix A, Theorem 1 (iii)⇒(i)"},{"comment":"The final inference 'applying the IA we find q = α* \\bar p + (1−α*) p ∼ α* q + (1−α*) p and thus p ∼ q ∼ p_1' is not justified. The IA step only yields q ∼ z for z := α* q + (1−α*) p on the segment between p and q; concluding p ∼ q requires ruling out p ≻ q and q ≻ p, which is precisely the content of Lemma A4 (case 0 < t < 1), or of a standard midpoint argument. Lemma A4 is proved later and its proof does not depend on Lemma A2, so the fix is local, but as written the claim that the whole affine hull of indifferent points lies in the indifference class is not established. This matters because the (i)⇒(ii) direction of Theorem 1 uses Lemma A2 to conclude that the parallel hyperplane through any p is contained in L_{∼p}.","section":"Appendix A, Lemma A2"},{"comment":"The abstract claims that IP is 'logically independent of Weak Continuity and the Archimedean axiom,' but the body establishes only one direction of this claimed independence: Example 1 exhibits a preference satisfying IP but none of the standard continuity axioms. No preference is constructed or cited that satisfies Weak Continuity or the Archimedean axiom while violating IP, so the asserted independence is not substantiated. This is not a purely stylistic point: nonconstant continuous EU preferences satisfy IP by the corrected (iii)⇒(i) construction, so the reverse direction needs a genuinely different example or a proof; otherwise the claim should be weakened to 'IP does not imply Weak Continuity or the Archimedean axiom.'","section":"Abstract"}],"minor_comments":[{"comment":"Terminology is inconsistent: the abstract and Section 1 use 'prices' where the majority and standard usage in the paper is 'prizes' (e.g., 'three prices' versus 'three prizes'); please standardize.","section":"Throughout"},{"comment":"The auxiliary point denoted p (the uniform mixture (1/m)Σ_k p_k) clashes with the arbitrary element p of A ∩ L; rename the former, e.g., \\bar p, and correct what appears to be 'q ≠ p' to 'q ≠ \\bar p'.","section":"Appendix A, Lemma A2"},{"comment":"Please verify the Mas-Colell et al. (1995) numbering: the continuity axiom is usually Definition 6.B.2 and the expected utility theorem is Proposition 6.B.3, so citing 'Definition 6.B.3' for mixture continuity looks misnumbered.","section":"Appendix A, Theorem 1 (ii)⇒(iii)"},{"comment":"Small typos: 'L ≻P' with a capital P in the (i)⇒(ii) proof should be L_{≻p}; 'Achimedean' in Example 1 should be 'Archimedean'; and 'Independent Points' after Theorem 1 should be 'Indifferent Points' for consistency with the axiom's name.","section":"Throughout"},{"comment":"In the uncountable-case construction, the assertion that aff(q_1, ..., q_{k+1}) does not contain p or r is left implicit; since the segment from s to r meets aff(q_1, ..., q_k, p, r) only at r when s lies outside it, one line of justification would improve readability.","section":"Appendix A, Proposition 1"},{"comment":"The example writes X = {x_1, ..., x_n}, dropping the x_0 of the main convention X = {x_0, ..., x_n}; please align the notation or state explicitly that the indexing is redefined.","section":"Example 1"}],"recommendation":"major_revision","confidential_remarks":"To the editor: the two proof defects identified in the report are real but localized, and in my judgment the central theorem is correct after the described fixes; this is therefore a major revision rather than a rejection. The abstract's logical-independence claim regarding Weak Continuity and the Archimedean axiom is the one claim that may not be fixable by a small example, so it deserves the author's explicit attention, but it does not affect Theorem 1. The reliance on the standard Mas-Colell representation result for (ii)⇒(iii) is appropriate, and there are no citation-pattern or novelty-disclosure concerns."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Gerrit,\n\nThe headline: this is a genuinely new axiomatization — replacing topological continuity with a geometric 'indifferent points' condition in the finite-outcome expected utility theorem. The main equivalence (IP ⇔ strong continuity ⇔ EU representation, under IA) is almost certainly true, and the logical independence results are clean. But the written proof has two fixable flaws, so it's not yet ready to be taken as published.\n\nWhat's good: The IP axiom is novel and the equivalence theorem is not in the cited literature. Example 1 correctly shows IP doesn't imply continuity without IA. The proof idea is elegant: under IA, indifference classes are convex; if an indifference class contains an affine hyperplane, then it is the entire separating hyperplane, giving open upper/lower contour sets. The paper also correctly connects to solvability and mixture continuity. No circularity, no fitted parameters, no self-citation problem.\n\nThe soft spots are both in Appendix A. Lemma A2 claims that the affine hull of indifferent points is indifferent, but the proof glides over the key step: after constructing q~p1, it asserts 'q~p~p1' without justification. You need Lemma A4 (line-strictness) or a midpoint argument to rule out strict preference along the line between p-bar and p. That's a genuine gap, though easily patched.\n\nMore serious: the (iii)=>(i) direction. The proposed linear system uses a 2×n matrix with second row (1,...,1) and RHS (u,1). But in the paper's parametrization, the simplex is {p∈R^n : p_i≥0, Σp_i≤1}, so the equation Σp_i=1 is not valid on the simplex. Also, the first row ignores u(x0); the correct indifference equation is Σ p_i (u(x_i)-u(x0)) = u - u(x0). And the claim that the kernel of a 2×n matrix has dimension at least n−1 is wrong; it's n−2. The uniform lottery p=(1/(n+1),...,1/(n+1)) doesn't satisfy Σp_i=1. All this is fixable: the correct construction uses a single linear functional (utility differences) whose kernel is n−1 dimensional; take a point of the right utility and add small kernel vectors. But as written, the proof doesn't establish the direction.\n\nOverall: the theorem is likely correct, the contribution is real, and the defects are repairable. This deserves a serious referee; I'd recommend sending it out with a request for a corrected Appendix A. If the author fixes those two points, it's publishable.\n\nFor a reading group, it's a nice short paper to discuss the IP axiom, though you'd want to give the appendix a skeptical eye.","headline":"A genuinely new geometric axiom for expected utility, probably true, but the written proof has two fixable errors in the appendix.","tokens_in":6162,"tokens_out":4654,"would_cite":true,"duration_ms":38908,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["91B16","91B08"],"pacs":[],"model":"deepseek-v4-flash","headline":"Under the Independence Axiom, a finite set of indifferent lotteries spanning a hyperplane is exactly equivalent to continuity and to an expected-utility representation.","keywords":["expected utility","Indifferent Points","Independence Axiom","continuity axioms","lexicographic preferences","solvability","mixture continuity","hyperplane spanning"],"falsifier":"For $n=2$, take two distinct lotteries $p$ and $q$ with $p \\sim q$; the theorem predicts the entire indifference class is the straight line through them and the better/worse sets are the two open half-planes. A complete, transitive preference satisfying the Independence Axiom where these two indifferent points have an indifference class that is not a line segment would falsify the theorem.","tokens_in":5081,"feed_emoji":"🎲","tokens_out":5715,"duration_ms":52237,"temperature":0.7,"pith_summary":"This paper shows that, for preferences over lotteries on a finite set of prizes, the topological continuity condition in the classic expected-utility theorem can be replaced by a finite geometric condition. The condition, called Indifferent Points, asks for $n$ pairwise indifferent lotteries whose difference vectors span a hyperplane in the $n$-dimensional simplex; for three prizes a single indifference between two lotteries is enough. The main theorem states that, under the Independence Axiom, Indifferent Points is equivalent to strong continuity and to the existence of an expected-utility representation. The paper also shows the new axiom is strictly weaker than solvability and independent of Weak Continuity and the Archimedean axiom, so it fills a genuinely different role.","feed_headline":"Indifferent points replace continuity in expected utility","feed_subtitle":"A single pair of equally liked lotteries suffices for three prizes, so continuity can be checked by finite data.","key_machinery":"The central object is the Indifferent Points axiom (IP): there exist $p_1 \\sim \\cdots \\sim p_n$ in the $n$-dimensional simplex such that the directional vectors $\\{p_k - p_1\\}_{k=2}^n$ are linearly independent, so their affine hull is a hyperplane. Four lemmas carry the argument: indifference sets are convex; the entire affine hull of indifferent points inside the simplex belongs to the same indifference class; translations of indifference sets remain indifference sets; and strict preference orders points on any line through two lotteries. Together these force the hyperplane of indifferent points to split the simplex into exactly the open strictly-better and strictly-worse sets, which is strong continuity. The converse uses the linear system $M\\cdot (p_1,\\dots,p_n)^\\top = (\\bar{u}, 1)^\\top$ and the fact that its kernel has dimension at least $n-1$ to construct a spanning family of indifferent lotteries from any expected-utility representation.","core_discovery":"The central claim is Theorem 1: for any complete and transitive preference on the lottery simplex satisfying the Independence Axiom, the following three statements are equivalent: (i) there exist $n$ indifferent lotteries whose pairwise difference vectors span a hyperplane; (ii) the strictly-better and strictly-worse sets of every lottery are open; and (iii) the preference is represented by expected utility for some utility function on prizes. The proof shows that, under the Independence Axiom, the affine hull of a set of indifferent points lies entirely inside the same indifference class, so a hyperplane of indifferent points separates the simplex into two open half-spaces that are exactly the strictly better and worse sets. The converse direction constructs the utility function by solving a linear system whose kernel has dimension at least $n-1$, yielding the required spanning indifferent lotteries. In the three-prize case, one indifference between two distinct lotteries already supplies the hyperplane.","pith_inferences":["A testable consequence not drawn in the paper: in experiments with three prizes, eliciting at least one indifference pair and checking independence could let researchers verify the continuity implied by expected utility without collecting data on infinitely many open sets.","For larger prize sets, the theorem suggests a combinatorial check: find $n-1$ linearly independent indifference directions; if they exist, the preference must be continuous. Designing such elicitations is a natural extension the paper does not pursue.","The finite-prize assumption is essential to the linear-algebra proof, so extending this result to infinite prize spaces would require a new argument; 'spanning a hyperplane' has no direct analogue there.","The paper's Example 1 indicates that when IA fails, IP can coexist with lexicographic-like structures, which suggests IP measures a different kind of regularity than standard continuity rather than a weakened version of it."],"forward_implications":["For any finite-prize decision problem, verifying continuity reduces to checking a finite list of indifferences: if $n$ equally ranked lotteries span a hyperplane, the whole preference is continuous.","In the popular three-prize case, a single elicited indifference between two lotteries, together with the Independence Axiom, is enough to guarantee an expected-utility representation.","Because IP is strictly weaker than solvability, every preference satisfying strong continuity, mixture continuity, or weak Wold-solvability automatically satisfies IP.","Without the Independence Axiom, IP does not imply continuity; the paper's lexicographic-style example satisfies IP but violates every standard continuity axiom, so IA is an indispensable part of the equivalence.","IP is logically independent of Weak Continuity and the Archimedean axiom, so it is not merely a disguised version of the usual continuity postulates."],"supporting_citations":[{"why":"Supplies the standard equivalence between strong continuity and expected-utility representation used for the (ii) implies (iii) direction.","marker":"Mas-Colell et al., 1995"},{"why":"Provides definitions of mixture continuity and solvability and the implication that mixture continuity implies solvability, used in Proposition 1.","marker":"Ozbek, 2024"},{"why":"Supplies the consolidated definitions of continuity postulates and solvability axioms that the paper compares Indifferent Points against.","marker":"Ghosh et al., 2023"}],"fun_headline_variants":["Hyperplane of indifferent lotteries replaces continuity","For three prizes, a single indifference pair suffices","Continuity equivalent to spanning indifferent points","Finite indifference data replace topological continuity","Two indifferent lotteries replace continuity for three prizes"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is the Independence Axiom: the theorem only goes through for preferences that preserve rankings when both options are mixed with a common third lottery, and without it Example 1 shows Indifferent Points can hold while every continuity axiom fails.","fun_headline_variants_meta":{"raw":{"variants":["Hyperplane of indifferent lotteries replaces continuity","For three prizes, a single indifference pair suffices","Continuity equivalent to spanning indifferent points","Finite indifference data replace topological continuity","Two indifferent lotteries replace continuity for three prizes"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.001241,"raw_usage":{"total_tokens":5197,"prompt_tokens":777,"completion_tokens":4420,"prompt_tokens_details":{"cached_tokens":768},"prompt_cache_hit_tokens":768,"prompt_cache_miss_tokens":9,"completion_tokens_details":{"reasoning_tokens":4350}},"tokens_in":9,"tokens_out":4420,"duration_ms":61401,"temperature":1.0,"reasoning_tokens":4350,"cache_read_input_tokens":768,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-12T11:45:13.360519+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"For $n=2$, take two distinct lotteries $p$ and $q$ with $p \\sim q$; the theorem predicts the entire indifference class is the straight line through them and the better/worse sets are the two open half-planes. A complete, transitive preference satisfying the Independence Axiom where these two indifferent points have an indifference class that is not a line segment would falsify the theorem.","supporting_citations":[{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the consolidated definitions of continuity postulates and solvability axioms that the paper compares Indifferent Points against."}],"review_version":1}