REVIEW 1 major objections 2 minor 14 references
Sometime a Paradox, Now Proof: Non-First-Order-izability of Yablo's Paradox
T0 review · 1 major / 2 minor · reviewed 2026-08-14 · deepseek-v4-flash
Pith's one-line read Yablo's paradox, formalized in second-order logic, cannot be captured by any first-order sentence or theory.
desk verdict Correct compactness proof for non-first-orderizability of a kernel sentence, but the identification with Yablo's paradox is too broad: over strict linear orders Y is first-order, so the headline claim needs qualification. 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 load-bearing object is the second-order sentence $Y$ together with its translation $Y^s$ into the language of a unary successor function. On any model of the successor axiom $\forall x,y\,(s(x)=s(y)\to x=y)$, the universe splits into disjoint copies of $\mathbb{N}$, $\mathbb{Z}$, and finite cycles; a subset $K$ is a kernel (every element is in $K$ iff its successor is not) exactly when no odd cycle is present. That equivalence turns the question of whether $Y$ is first-order expressible into the question of whether the theory $\{\neg\exists x\,s^{2n+1}(x)=x\}_{n\in\mathbb{N}}$ is finitely axiomatizable, and it is not: arbitrarily large odd cycles witness the failure of each finite subtheory.
What would settle it
A concrete way to test the claim: find a first-order sentence $\eta$ in the language $\{R\}$ such that $\eta\leftrightarrow Y$ holds in every structure, or a first-order theory whose models are exactly the directed graphs satisfying $Y$. Equivalently, exhibit a finite axiomatization of the theory of a successor function with no odd cycles; the paper proves none exists, so such an exhibit would overturn Theorem 3.3.
Extended reading notes
Core claim
The paper's central claim is that the second-order sentence $Y$, defined as $\neg\exists X\,\forall x\,(X(x)\leftrightarrow \forall y\,[xRy\to \neg X(y)])$, is not logically equivalent to any first-order sentence in the language of the binary relation $R$ (Theorem 3.3), and not equivalent to any first-order theory whatsoever (Theorem 3.4). The sentence says that no subset of the universe can contain exactly the elements whose $R$-successors all lie outside it—that is, that the directed graph $\langle D;R\rangle$ has no kernel. The proof shows that after replacing $xRy$ by $s(x)=y$, the negation of the translated sentence says that a model of the successor axioms has a kernel, which exists exactly when the model has no odd cycles; the theory of 'successor with no odd cycles' is not finitely axiomatizable. Consequently any first-order approximation to $Y$ would have to either miss some model or admit some unwanted model.
Load-bearing premise
The load-bearing premise is that replacing the natural 'later than' ordering in Yablo's paradox with an arbitrary binary relation $R$ preserves the paradox's content; if the paradox only arises for the genuine ordering of the natural numbers, the non-first-orderizability proof would not apply to that standard formulation.
Editorial extensions
If this is right
- No first-order sentence or theory over the language of directed graphs can have exactly the same models as the second-order Yablo sentence; any first-order surrogate either admits a directed graph with no kernel or excludes one with a kernel.
- The existence of a kernel in a directed graph, and the non-existence of such a kernel, are properties that are not first-order definable, since $\neg Y$ is exactly the assertion that a kernel exists.
- Formalizing Yablo's paradox as $Y$ does not by itself make a contradiction or a tautology; some condition on $R$ (such as $\forall x\exists y\,(xRy\wedge\forall z\,[yRz\to xRz])$) is required to force the paradox.
- The paradox yields a theorem of the same kind as Russell's or Liar's paradoxes: a logical impossibility result, here about the expressive boundary between first- and second-order logic.
Reading between the lines
- If the generalized relation $R$ is essential to the proof, the theorem does not directly settle whether first-order treatments that build in the actual natural-number ordering of Yablo's original list are first-order definable; the paper's formulation deliberately abstracts away the order.
- The same model-theoretic template—translate a paradox into a non-finitely-axiomatizable theory and invoke compactness—could apply to other infinite, non-self-referential paradoxes, isolating the order-type or graph property that generates the expressive gap.
- One testable extension is to ask whether the conjecture that $\neg Y$ is also not first-order-theory-equivalent holds; a counterexample would need a first-order theory whose models are precisely the directed graphs that possess a kernel.
- For finite directed graphs, kernel existence is a combinatorial property whose first-order undefinability can be checked on arbitrarily large odd and even cycles; this gives a concrete finite-model-theoretic version of the result.
Signed reviews
Editorial analysis
A structured set of objections, weighed in public.
Referee Report
Summary. The paper formalizes Yablo's paradox as the second-order sentence Y := ¬∃X∀x(X(x) ↔ ∀y(xRy → ¬X(y))) over an arbitrary binary relation R. It claims that Y is not equivalent to any first-order sentence (Theorem 3.3) and not equivalent to any first-order theory (Theorem 3.4). The proof translates the problem into the language of unary functions with an injectivity axiom, connects the existence of a kernel in the associated directed graph to the absence of odd cycles, and uses a compactness argument to show that the resulting theory is not finitely axiomatizable. The paper also presents sufficient conditions under which Y is valid (Theorem 2.3).
Significance. The compactness argument in Section 3 is sound and establishes a genuine model-theoretic result: the statement that a directed graph has no kernel is not first-order expressible and not first-order axiomatizable. This is a valuable contribution to the model theory of directed graphs. However, the result is proved for the generalized arbitrary-relation sentence Y, not for the usual order-theoretic Yablo paradox. Over strict linear orders, Y is first-order equivalent to 'there is no maximum,' so the advertised relevance to Yablo's paradox is overstated. The paper would be strengthened by presenting the result as being about kernel existence in arbitrary directed graphs and by acknowledging this limitation explicitly. The proof of Theorem 2.3(1) also contains errors, although that theorem is not used in the main proof.
major comments (1)
- [§2 (Definition 2.1) and §3 (Theorems 3.3–3.4)] The non-first-orderizability claim is proved for the sentence Y with an arbitrary binary relation R, but the title and abstract present this as a result about Yablo's paradox. Over the class of strict linear orders—the natural formalization of the 'later than' relation—Y is first-order equivalent to ∀x∃y(xRy). Indeed, in a strict linear order, a set X satisfies the kernel condition iff it is the singleton of the maximum element when a maximum exists, and no such X exists otherwise; hence Y holds exactly when there is no maximum. Therefore the theorem does not show that the order-theoretic Yablo paradox is non-first-orderizable. The paper should state this limitation explicitly and adjust its framing accordingly; as it stands, the central claim is broader than what the proof actually supports.
minor comments (2)
- [§2, Theorem 2.3(1)] The proof of Theorem 2.3(1) contains technical errors. In the induction step, from θ_{n+1}(a) = ∃y(aRy ∧ ∀z[yRz → θ_n(z)]) one only obtains θ_n(z) for all z with bRz, not aRz for all such z as the proof claims, and θ_n(z) is evaluated in the original graph, not in the induced subgraph on D_b. Thus the claim that ⟨D_b, R∩D_b²⟩ satisfies ∀xθ_n(x) is unsupported. In the base case, from b∉X the kernel condition yields ∃c(bRc ∧ c∈X), not c∉X as written. Since Section 3 does not depend on Theorem 2.3, these errors do not affect the main result, but the theorem's proof should be repaired or the theorem proved by a different argument.
- [Throughout] There are numerous typos and spacing errors, including 'Seconc-Order Logic' in the Section 2 heading, 'mathem atics' in the abstract, 'form' for 'from' in the introduction, and inconsistent spacing in 'nonfirstorderizability'. I recommend a careful proofreading pass before publication.
Circularity Check
No significant circularity: the non-first-orderizability proofs are self-contained model-theoretic arguments.
full rationale
The paper's central results, Theorems 3.3 and 3.4, are derived from Lemma 3.2 and standard compactness/axiomatizability facts cited from van Dalen's Logic and Structure. Lemma 3.2 is proved directly by analyzing kernels on successor structures and odd cycles, with no appeal to the target theorem. The proof that S' = S ∪ {¬∃x(s^{2n+1}(x)=x) : n∈N} is not finitely axiomatizable uses only the existence of arbitrarily large odd cycles, an independent compactness argument. No fitted parameter is renamed as a prediction, no uniqueness theorem from the author's prior work is invoked, and no ansatz is smuggled in via self-citation. The identification of Yablo's paradox with the second-order sentence Y in Definition 2.1 is a stipulative formalization, not a derivation that assumes its own conclusion; one may question whether the abstraction to an arbitrary binary relation preserves the intended paradox, but that is an interpretive limitation, not circularity. The paper's self-citations, [7] and [8], appear only as contextual references for translations into linear temporal logic and are not load-bearing for the main proof. The flawed-looking induction step in Theorem 2.3 is not used by Section 3, and in any case a proof gap is a correctness issue, not a circularity. Thus the honest finding is no significant circularity.
Assumptions & free parameters
assumptions (3)
- standard math Compactness theorem for first-order logic
- standard math Van Dalen's Lemma 4.2.10: a class that is both axiomatizable and co-axiomatizable is finitely axiomatizable
- domain assumption Models of the successor theory S are disjoint unions of copies of N, Z, and finite cycles
Cite this review
Pith. "Pith review of Sometime a Paradox, Now Proof: Non-First-Order-izability of Yablo's Paradox." pith.science (2026). https://pith.science/paper/NOCJ3KAY
@misc{pith2026190801496,
author = {Pith},
title = {Pith review of: Sometime a Paradox, Now Proof: Non-First-Order-izability of Yablo's Paradox},
year = {2026},
howpublished = {\url{https://pith.science/paper/NOCJ3KAY}},
note = {Machine review of arXiv:1908.01496}
}
read the original abstract
Paradoxes are interesting puzzles in philosophy and mathematics, and they could be even more fascinating, when turned into proofs and theorems. For example, Liar's paradox can be translated into a propositional tautology, and Barber's paradox turns into a first-order tautology. Russell's paradox, which collapsed Frege's foundational framework, is now a classical theorem in set theory, implying that no set of all sets can exist. Paradoxes can be used in proofs of some other theorems; Liar's paradox has been used in the classical proof of Tarski's theorem on the undefinability of truth in sufficiently rich languages. This paradox (and also Richard's paradox) appears implicitly in G\"{o}del's proof of his celebrated first incompleteness theorem. In this paper, we study Yablo's paradox from the viewpoint of first and second order logics. We prove that a formalization of Yablo's paradox (which is second-order in nature) is non-first-order-izable in the sense of George Boolos (1984).
Reference graph
Works this paper leans on
-
[1]
Berge, Claude ; Graphs, North Holland (3rd ed. 1991). isbn: 9780444876034
work page 1991
-
[2]
Boolos, George ; To Be is To Be a Value of a Variable (or to be some values of some variables ), The Journal of Philosophy 81:8 ( 1984) 430–449. doi: 10.2307/2026308 Reprinted in: Boolos, G.; Logic, Logic and Logic (isbn: 9780674537668) Harvard University Press (1998) pp. 54–72
- [3]
-
[4]
Cook, Roy ; The Y ablo Paradox: An Essay on Circularity , Oxford University Press (2014). isbn: 9780199669608
work page 2014
-
[5]
doi: 10.2143/LEA.235.0.3170108
Forster, Thomas & Gor ´e, Rajeev ; Yablo’s Paradox as a Theorem of Modal Logic , Logique et Analyse 59:235 ( 2016) 283–300. doi: 10.2143/LEA.235.0.3170108
-
[6]
Halbach, Volker & Zhang, Shuoying ; Yablo Without G¨ odel, Analysis 77:15 ( 2017) 53–59. doi: 10.1093/analys/anw062
-
[7]
Karimi, Ahmad & Salehi, Saeed ; Diagonal Arguments and Fixed Points , Bulletin of the Iranian Mathematical Society 43:5 ( 2017) 1073–1088. https://bit.ly/31pQVYF
work page 2017
-
[8]
Karimi, Ahmad & Salehi, Saeed ; Theoremizing Yablo’s Paradox, arXiv:1406.0134 (2014) 7 pp
work page Pith review arXiv 2014
Show all 14 references
-
[9]
doi: 10.1007/s11229-005-6201-6
Ketland, Jeffrey ; Yablo’s Paradox and ω -Inconsistency, Synthese 145:3 ( 2005) 295–302. doi: 10.1007/s11229-005-6201-6
2005 doi
-
[10]
Priest, Graham ; Unstable Solutions to the Liar Paradox , in: S. J. Bartlett & P. Suber (eds.), Self-Reference: Reflections on Reflexivity (isbn: 9789024734740), Martinus Nijhoff Pub- lishers ( 1987) pp. 145–175. doi: 10.1007/978-94-009-3551-8 9
1987 doi
-
[11]
van Dalen, Dirk ; Logic and Structure , Springer (5th ed. 2013). isbn: 9781447145578
2013
-
[12]
Gabbay & F
Visser, Albert; Semantics and the Liar Paradox , in: D. Gabbay & F. G¨ unthner (eds.), Hand- book of Philosophical Logic, V olume IV: T opics in the Philos ophy of Language (isbn: 9789401070218), Reidel ( 1989) pp. 617–706. doi: 10.1007/978-94-009-1171-0 10
1989 doi
-
[13]
doi: 10.1007/BF00249368
Yablo, Stephen; Truth and Reflection , Journal of Philosophical Logic 14:3 (1985) 297–349. doi: 10.1007/BF00249368
1985 doi
-
[14]
doi: 10.1093/analys/53.4.251
Yablo, Stephen ; Paradox Without Self-Reference , Analysis 53:4 ( 1993) 251–252. doi: 10.1093/analys/53.4.251
1993 doi
Reviewed August 14, 2026 · model on record in the stance chip above.
Discussion (0). Continue with ORCID to comment.