Paradoxes of Infinity as Reductio ad Absurdum

Russell’s antinomy, Gödel’s incompleteness, and Turing’s undecidability are usually read as limitations to be endured within an infinitary foundation. We propose a different diagnosis: each classical paradox combines constitutive baseline principles—consistency and effective determinacy—with one completed-infinity premise: absolute totalities (AT), an internalized registry (IR: the object theory interprets Robinson’s $Q$), or unrestricted Choice (AC$\infty$). Turing’s undecidability separates as an external non-existence theorem.
The rows state the paper’s claims in their current form. The source column names the body statement that carries each row and links its witnesses as tokens — py, lean and leanML, each followed by the row’s label: the python check that decides the row, its declaration in the core (no axioms) and its declaration on Mathlib.
#DescriptionStatus Sources
A. Inputs
A1p05001The classical theorems, accepted unchanged: Gödel I and Rosser, Tarski’s undefinability, Turing’s halting theorem, Cantor’s theorem and uncountability, Russell, Burali–Forti, Skolem, Vitali, Hausdorff, Banach–Tarski; the operational paradoxes (Hilbert’s hotel, Zeno). No classical derivation is disputed.I
[G"odel 1931; Rosser 1936; Tarski 1956; Turing 1937; Cantor 1891; Russell 1903; Burali-Forti 1897; Skolem 1922; Vitali 1905; Hausdorff 1914; Banach & Tarski 1924]
A2p05002Metamathematics: Robinson’s $Q$, essentially undecidable, every model infinite; $\mathrm{Con}(Q)$ by cut elimination in $I\Delta_0+\mathrm{superexp}$, not in $I\Delta_0+\mathrm{exp}$; Tait’s finitism; Dedekind’s categoricity of the second-order naturals; true second-order arithmetic not arithmetical; Tarski’s decidable real field; Solovay’s model and Shelah’s inaccessible.I
[Tarski et al. 1953; H'ajek & Pudl'ak 1993; Wilkie & Paris 1987; Tait 1981; Shapiro 1991; Tarski 1951; Solovay 1970; Shelah 1984]
A3p05003Combinatorics and set theory: Fraenkel–Mostowski permutation models (symmetry as an obstruction to choice); de Bruijn–Erdős compactness of colouring; aperiodic tile sets and the undecidability of the domino problem (Berger, Robinson); Zermelo’s choice.I
[Jech 1973; de Bruijn & ErdH os 1951; Berger 1966; Robinson 1971; Zermelo 1904]
A4p05004The corpus: the totality finite and complete (00:A2), counting closes by return (00:A5), a bounded observer a proper part and every reading a projection (00:A4, A6), Gödel vacuous over the finite $\Omega$ (00:Z1), the $\Omega$-hard demarcation (00:Z3); finite mathematics as ontologically primary (Lev, Zeilberger).I
B. Gödel’s bundle and the frames
B1p05005The standing hypotheses: (Eff) effective axiomatisation, (Arith) interpretability of $Q$, (Cns) consistency. Two doors for infinity: the meta-level syntax, potential and retained; and (IR), the object theory interprets $Q$ — an injective non-cyclic successor with sequence coding, the completed registry inside the theory.D
Section 3; § 3.1
B2p05006The bundle $\{$Eff, Cns, Cmp, IR$\}$ is inconsistent, with no soundness assumption: under Eff and IR the Rosser sentence is available, and Cmp decides it against plain consistency.T
Prop. 1; Thm. 3.3
B3p05007Witnessed independence: every three-element subset is satisfiable — $Q$ itself ($\neg$Cmp, consistent by finitary means, the superexponential cost exact); $\mathrm{Th}(W_N)$ ($\neg$IR: consistent, complete, decidable, singly axiomatised, interpreting no $Q$ since every model of $Q$ is infinite); the inconsistent extension ($\neg$Cns); true arithmetic ($\neg$Eff). Four legs, all load-bearing.T
python sourceLean, Mathlib
Thm. 3.4
master: 00:Z1
B4p05008The arithmetic frame $W_N$: domain $[0,N)$, constant $0$, order $<$, the graphs $S$, $A$, $M$ of successor, sum and product restricted to the domain (overflow fails the atom); frame truth is truth in $W_N$; a $\Delta_0$ sentence is transcribed through the graphs with guarded bounded quantifiers.D
Def. 1; § 3.2
B5p05009Per-frame completeness: $\mathrm{Th}(W_N)$ is consistent, complete and decidable, axiomatised by the categorical sentence $\sigma_N$; truth in any finite structure is decided by evaluation, and the theory of any structure is complete.T
python sourceLean, Mathlib
Prop. 2
master: 00:Z1
B6p05010Bounded stability, schema form: a $\Delta_0$ sentence $\varphi$ with bound $t(\varphi)$ has $W_N\models\varphi^{*}\iff\N\models\varphi$ for every $N>t(\varphi)$, each instance $\mathrm{St}(\varphi,M)$ a finite conjunction of finite checks; proved for an explicit $\Delta_0$ language with $t(\varphi)$ computed; Goldbach below $20$ has $t=400$ and value true (below $10$, $t=100$, decided by the kernel).T
python sourceLean, the core
Thm. 3.6; Ex. 3.7
B7p05011The unbounded quantifier is the internalised registry: a $\Pi^{0}_{1}$ sentence receives a verdict from each frame about its own domain and none about all frames at once; frame-relative completeness covers every executed verification (a bounded $\Delta_0$ trace) and no $\Pi^{0}_{1}$ closure.Ω
§ 3.2; D2; 00:Z3
master: 00:Z3
C. Where incompleteness goes
C1p05012The bounded agent of capacity $K$: records of at most $K$ symbols over $s\ge2$ letters, fewer than $s^{K+1}$ of them; certification is a retained record encoding a completed proof-check, one sentence per record, no schema compression.D
Def. 2
C2p05013No internal mirror: fewer than $s^{K+1}$ records exist, so an agent with $s^{K+1}<N$ holds no injective representation of the domain of $W_N$ — a proper part cannot mirror the whole (pigeonhole).T
python sourceLean, the coreLean, Mathlib
Prop. 3
master: 00:A4
C3p05014Vanishing certified fraction: at least $c^{L}$ normal-form sentences of length $\le L$, half of each cohort frame truths by completeness, fewer than $s^{K+1}$ certified, so the certified fraction is at most $s^{K+1}/\lfloor c^{L}/2\rfloor\to0$.T
python sourceLean, Mathlib
Prop. 4
C4p05015The migration: without IR every sentence about the totality is decided and incompleteness reappears as a resource bound on agents — the Gödel sentence’s role passes to truths whose shortest certificate exceeds the capacity; truth for $W_N$ is definable one level up (Tarski relativised to a resource hierarchy).D
§ 3.3; 00:Z1
master: 00:Z1
C5p05016Horizon separation: halting for machines of bounded space $K$ is decided from above within $|Q|\,K\,s^{K}$ steps (a configuration repeats) and never from within (diagonalisation); Turing’s theorem is an external non-existence theorem, classified with Cantor’s, not an instance of the bundle.T
python sourceLean, the core
§ 3.3
D. The reductio principle and the forced selection
D1p05017The premises and baselines: AT (an absolute totality), IR (the internalised registry), AC$\infty$ (unrestricted choice over uncountable families); Cns, Cmp, Meas (the exact measure axioms of each geometry), and Det — a totality decides every well-posed question about itself, elements, subsets and relations, with its full second-order theory effectively presented; declared as an axiom.D
§ 4.1
D2p05018Antinomy normal form: Cns $+$ AT $+$ comprehension is inconsistent (Russell, Burali–Forti); the witness of Cns $+\neg$AT is bounded comprehension — every set $u$ has a subset not among its members and no universal set exists, decided on $V_4$.T
python source
Prop. 5
D3p05019Diagonal normal form: $\{$Eff, Cns, Cmp, IR$\}$ inconsistent (B2), Tarski a variant with the truth predicate for Cmp; the diagonal is a theorem when external — Cantor’s theorem, $n<2^{n}$ on every frame, uncountability, halting — and a paradox engine only over an internalised registry with completeness demanded.T
python sourceLean, the coreLean, Mathlib
Prop. 6; Rem. 4.1
D4p05020Choice normal forms: Vitali (countably additive translation-invariant measure on all subsets $+$ AC$\infty$ on $\R/\Q$, a countable amenable group) and Banach–Tarski (finitely additive isometry-invariant measure $+$ AC$\infty$ $+$ a free non-amenable rotation subgroup) are inconsistent bundles with distinct mechanisms; the finite witness is counting measure with definable choice.T
python source
Prop. 7
D5p05021The reductio principle: for each normal form a baseline $B$ and one infinitary premise $I$ with $B+I$ contradictory and $B+\neg I$ satisfiable by an explicit witness — bounded ranks, $\mathrm{Th}(W_N)$, counting measure (Solovay’s model as classical corroboration); the rejection of $I$ is coherent and witnessed.T
Thm. 4.2; Table 1; B3, D2, D4
D6p05022The selection thesis: every normal form conjoins completed infinity with a totalising operation (comprehension, completeness, total selection, a total measure); each is harmless alone, contradiction tracks the conjunction; classical mathematics rejected the totalising conjunct at each crisis, this paper the shared one, the selection settled by Det rather than convention.D
§ 4.4; § 2.5
D7p05023Forced exit: Det supplies Eff and Cmp, so $\{$Cns, Det, IR$\}$ is inconsistent and, Cns and Det being constitutive, the reductio is forced against IR — no determinate totality internalises an unbounded registry; the four-clause bundle condenses to this triad. Conditional on Det.T
Prop. 8; B2
D8p05024Finiteness of determinate totalities: in ZF with every infinite set Dedekind-infinite, a structure satisfies Det iff it is finite — finite: the full higher-order theory decided by exhaustive evaluation; infinite: a copy of $\N$ inside gives an effective reduction of true second-order arithmetic, which is not arithmetical. The real field’s first-order tameness is silence about its subsets.T
python sourceLean, Mathlib
Prop. 9
D9p05025Corollaries: no absolute totality is determinate (a universal set is infinite by the iterated singletons); Meas is decided content — a Det totality decides every measure question, trivially on a finite one (counting measure). The totalising conjuncts of two normal forms follow from D8.T
python source
Cor. 1; Cor. 2
D10p05026The three gates: syntactic infinity (meta-level, potential, retained), semantic infinity (a completed object of the theory, the target in all three forms), diagonalisation (a method, dangerous only downstream of the second gate); the paper argues with an unbounded supply of finite objects only.D
§ 4.5; Table 2
E. Choice recovered
E1p05027Global choice on a finite universe: with a canonical order, $\operatorname{ch}(A)=\min A$ is a uniform pointwise choice rule; over a fixed finite base AC is a definable theorem; finite products are nonempty without it.T
python sourceLean, the coreLean, Mathlib
Prop. 10
E2p05028Periodic choice: an equality-periodic family $A_{i+N}=A_i$ has the choice $f(i)=\operatorname{ch}(A_i)$ of the same period; group-periodic families as equivariant surjections $\pi:A\to I$ with equivariant sections.T
python sourceLean, the coreLean, Mathlib
Lemma 5.2; Def. 5
E3p05029Equivariant periodic choice: a $G$-equivariant section exists iff each stabiliser action $G_i\curvearrowright A_i$ has a fixed point (one orbit at a time); the Fraenkel–Mostowski mechanism in positive form.T
python sourceLean, Mathlib
Thm. 5.3
E4p05030The definable basis: every subspace of a finite vector space has the greedy $\min$-basis, one definable function on all subspaces; the equivariant obstruction — $\Z/2$ acting by $v\mapsto-v$ on the line over $\F_5$ fixes no basis ($-v\neq v$), over $\F_2$ it does: arithmetic, not formal.T
python sourceLean, the coreLean, Mathlib
Thm. 5.4; Ex. 5.5
E5p05031Periodic Kőnig: in a finite digraph with out-degree $\ge1$ the greedy walk repeats a vertex within $|D|+1$ steps and is eventually periodic with period $\le|D|$ — the pigeonhole on an iteration over finitely many states.T
python sourceLean, the core
Lemma 5.6
E6p05032Periodic de Bruijn–Erdős, dimension one: a $\sigma$-periodic graph of bounded range whose finite subgraphs are $c$-colourable admits a definable $\sigma^{m}$-periodic $c$-colouring (the transfer digraph $D_c$, its sub-digraph of vertices beginning walks of every length, E5).T
python source
Thm. 5.7
E7p05033The dimension boundary: in $\Z^{2}$ aperiodic tile sets exist — every finite patch admissible, no periodic configuration — and the domino problem is undecidable; the passage from finite admissibility to infinite realisability has no finite certificate, its demarcation none either. The finite reading asserts patch admissibility only.Ω
§ 5.4; A3
E8p05034The choice paradoxes vanish: on a finite set counting measure is complete and invariant, no decomposition doubles the set ($X\sqcup X$ does not inject into $X$), and every ultrafilter is principal; periodic variants inherit this on one period.T
python sourceLean, Mathlib
§ 5.5
F. The reading
F1p05035Ontology and conservativeness: actual infinity is an idealisation of the finite; the bounded fragment is kept exactly, the unbounded fragment in schema form; analysis as approximation over finite grids; choice in the small. Positioning: finitism, intuitionism, ultrafinitism, structuralism; the potential-infinity rule retained at the meta-level as a commitment, not a neutrality.D
§ 6; § 6; § 6.1; 00:A2
master: 00:A2
F2p05036The Reductio Rule and its two tribunals: the economic ledger (the classical exit keeps $\Pi_1$, $\Pi_2$ number theory at a superexponential consistency grant; the finite exit keeps the bounded fragment and declines one classically provable existence claim, E7) and the constitutive Det, on which the verdict rests; objections O1–O5 answered.D
§ 6.2; § 6.3
F3p05037The catalogue: the pathologies of the completed infinite in four classes — antinomies, consistent-but-unregistrable, underdetermination, physical divergences — each removed by finiteness; the one exclusion retained, the schema over finite instances and the successor act.D
Appendix A
T. Tasks
T1p05041The standing of the iteration rule — whether the potential-infinity rule of the meta-level can be rejected, and at what cost — is the subject of companion work; the present results are stated to survive either outcome.O
§ 6.1

Status

I
Import. A standard result, or a corpus result, used without reproof.
D
Definition. A naming or declared convention, carrying no empirical content.
T
Theorem. Derived, and machine-verified in exact arithmetic by the named witness.
Ω
$\Omega$-hard. Decided by the totality; not closeable by a bounded observer. A demarcation, not an open item.

Sources

python source
Gold: the python check that decides the predicate, at its line in the script (the source, rendered on this site); on hover the deciding family and the families that corroborate it, and the verdict of the last run (red when it fails). The script page's bar runs the predicate's cell of the package's notebook on Google Colab (one cell per predicate; the whole notebook from the header badge).
Lean, the coreLean, Mathlib
The predicate as one Lean declaration, named by its accession key (p05007) — the conjunction of the theorems that prove it — at its line in the rendered module, one mark per library that proves it, the core's first. Green: the core (lean/FrcCore) — the statement on the shell ring, by computation and induction, no Mathlib, no axioms: the finitist witness. Violet: Mathlib (lean/FrcLedger) — the classical statement for every shell on Mathlib's hierarchy (its axioms, propext, Classical.choice, Quot.sound, on hover). Red: a proof the gate rejects (an unfinished proof, native_decide) — not a witness. A predicate proved in both libraries shows both marks; one mark, one proof. The module page's bar opens the module as one executable file in Lean Live (the core on plain Lean, no Mathlib; the file elaborates in the browser and prints every predicate's axioms at its end); the header's two Lean badges count the marks of each colour and open the two executables.
section
The statement in the paper's body that carries the predicate: a proposition, a remark or a section.
rests on
The predicates it rests on, by label: realisations, definitions and imports of this ledger, predicates of other ledgers (32:B3) and of the master (00:D15).
What the reader must grant. Four imports: the classical theorems of A1, the metamathematics of A2, the combinatorics and set theory of A3, the corpus of A4. Ten declarations (B1, B4, C1, C4, D1, D6, D10, F1–F3): the hypotheses and the two doors, the frame, the bounded agent, the migration, the premises and baselines, the selection thesis, the three gates, the reading. Twenty-one theorem rows (B2, B3, B5, B6, C2, C3, C5, D2–D5, D7–D9, E1–E6, E8): B6, C2, C5, D3, E1, E2, E4, E5 are proved in the core with no axioms, B3, B5, C2, C3, D3, D8, E1–E4, E8 on Mathlib’s hierarchy, and every one is decided by the package on the instances named; B2, D5 and D7 are the classical argument applied, D7 as an implication. Two $\Omega$-hard rows: B7, the $\Pi^{0}_{1}$ closure that no frame decides, and E7, the two-dimensional boundary that no finite certificate demarcates. One open row, T1, a task. What is conditional is exactly one thing, treating Det as constitutive (D7, D8 read as a verdict); everything else is unconditional. No realisation row: the paper identifies no physical term. The title claim — the paradoxes as reductio against completed infinity — is D5 with its witnesses (B3, D2, D4), forced under Det by D7 and D8, and scoped by E7. \endappendices