REVIEW 3 major objections 6 minor 4 references
Expected Utility Without Assuming Continuity
T0 review · 3 major / 6 minor · reviewed 2026-08-12 · deepseek-v4-flash
Pith's one-line read Under the Independence Axiom, a finite set of indifferent lotteries spanning a hyperplane is exactly equivalent to continuity and to an expected-utility representation.
desk verdict A genuinely new geometric axiom for expected utility, probably true, but the written proof has two fixable errors in the appendix. read the letter →
The pith
A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.
The reading
What carries the argument
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.
What would settle it
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.
Extended reading notes
Core claim
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.
Load-bearing premise
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.
Editorial extensions
If this is right
- 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.
Reading between the lines
- 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.
Editorial analysis
A structured set of objections, weighed in public.
Referee Report
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.
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 (3)
- [Appendix A, Theorem 1 (iii)⇒(i)] 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.
- [Appendix A, Lemma A2] 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}.
- [Abstract] 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.'
minor comments (6)
- [Throughout] 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.
- [Appendix A, Lemma A2] 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'.
- [Appendix A, Theorem 1 (ii)⇒(iii)] 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.
- [Throughout] 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.
- [Appendix A, Proposition 1] 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.
- [Example 1] 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.
Circularity Check
No significant circularity: derivation is self-contained and the new axiom is the object of study.
full rationale
The paper's central theorem is an equivalence result, and each direction is proved from the stated axioms rather than from the conclusion. Lemma A1, A2, A3, and A4 derive geometric properties of indifference sets directly from the Independence Axiom. Direction (i)=> (ii) uses these lemmas and a hyperplane-separation argument to show that strictly better and worse sets are open. Direction (ii)=> (iii) invokes the standard Mas-Colell representation theorem as independent external support. Direction (iii)=> (i) constructs an indifference hyperplane from the utility function delivered by the representation; it does not assume IP as an input. There are no fitted parameters, no renamed empirical patterns, and no self-citations of the author's prior work that carry load. The new axiom IP is the object of study, not a hidden premise, and the proof chain does not reduce any claimed implication to its own assumptions. Any possible algebraic gap in the (iii)=> (i) construction would be a correctness issue, not a circularity issue.
Assumptions & free parameters
assumptions (6)
- domain assumption The preference relation is complete and transitive.
- domain assumption Independence Axiom: p is preferred to q iff alpha p + (1-alpha) r is preferred to alpha q + (1-alpha) r for all p,q,r and alpha in (0,1].
- domain assumption The prize space X is finite with n+1 elements and L is the full n-dimensional probability simplex.
- standard math Standard affine geometry: affine hull, hyperplane separation, and finite-dimensional linear algebra.
- standard math Baire category theorem for the simplex in Proposition 1.
- domain assumption The equivalence of strong continuity and expected utility representation from Mas-Colell et al. (1995).
Cite this review
Pith. "Pith review of Expected Utility Without Assuming Continuity." pith.science (2026). https://pith.science/paper/7RNUKFWP
@misc{pith2026241117883,
author = {Pith},
title = {Pith review of: Expected Utility Without Assuming Continuity},
year = {2026},
howpublished = {\url{https://pith.science/paper/7RNUKFWP}},
note = {Machine review of arXiv:2411.17883}
}
read the original abstract
I provide an axiomatization of expected utility in which topological continuity is replaced by a geometric axiom. The axiom requires a finite set of indifferent lotteries that span a hyperplane. In the case of three prizes, two indifferent lotteries suffice. The axiom is weaker than Solvability, as well as logically independent of Weak Continuity and the Archimedean axiom.
Reference graph
Works this paper leans on
-
[1]
Ghosh, A., M. A. Khan, and M. Uyanık (2023). Continuity postulates and solvability axioms in economic theory and in mathematical psychology: a consolidation of the theory of individual choice.Theory and Decision 94(2), 189–210. Mas-Colell, A., M. D. Whinston, J. R. Green, et al. (1995).Microeconomic theory, Volume
work page 2023
-
[2]
Oxford university press New York. Ozbek, K. (2024). Expected utility, independence, and continuity.Theory and Decision 97(1), 1–22. A Proofs Before proving the main theorem, I provide self-contained proofs of auxiliary Lemmata, all of which assume a weak preferences fulfilling the Independence Axiom, IA. Essentially, they prove a generalization of the com...
work page 2024
-
[4]
1 | {z } =:M · p1 p2 ... pn = u 1 . Note that p:= ( 1 n+1 , 1 n+1 , . . . ,1 n+1 ) is a particular solution. Since rank(M)≤2, the kernel ker(M) of the linear map induced byMis a vector space of at least dimensionn−1 and the solution set of the system is given by p+ ker(M). Let b2, . . . , bn be linearly independent elements of ker(M). Since ...
work page 2024
-
[1995]
and note that we use an even stronger. continuity axiom. Proof of Theorem 1.(i)⇒(ii): Letp 1, . . . , pn ∈ L∼p1 such that{p k −p 1}n k=2 are linearly independent inR n. ThenA:= aff(p 1, . . . , pn) is a hyperplane of Rn, i.e., dimA=n−1. Hence, there exists a normal vector⃗ n∈R n such thatA={q∈R n| ⟨q−p1, ⃗ n⟩= 0}, where⟨·,·⟩is the scalar product onR n. Co...
work page 1995
Reviewed August 12, 2026 · model on record in the stance chip above.
Discussion (0). Continue with ORCID to comment.