Roth's Theorem (Diophantine Approximation)¶

For each fixed real algebraic irrational and positive exponent margin, only finitely many rationals approximate it more closely than denominator power 2 plus that margin.

Version: v1 · 2026-10-17 · History
Domain-specific #: 14281
Domain group: Formal Sciences
Origin domain: Mathematics
Subdomain: Diophantine Approximation → Mathematics
Aliases: Roth's theorem, Thue–Siegel–Roth theorem

Core Idea¶

Roth's theorem gives a sharp restriction on rational approximations to a fixed real algebraic irrational number. For every such target α and every fixed ε > 0, only finitely many distinct reduced fractions p/q, with q > 0, satisfy

\[ 0 < |\alpha-p/q| < q^{-(2+\varepsilon)}. \]

The theorem is a class-wide finiteness guarantee. It permits exceptional fractions, perhaps with very large denominators; it does not list them or give an effective largest denominator. The exponent margin ε must stay positive and fixed while the fractions vary. Roth's original statement says that infinitely many approximations at exponent κ force κ ≤ 2; the displayed finite-exception form is its contrapositive for κ = 2 + ε.^[1]^[2]

This identity has two useful directions within number theory. Known algebraicity blocks an infinite sequence of overly close rational approximants; conversely, constructing such a sequence for an already irrational number rules out algebraicity. Both directions use the same fixed target, denominator-power threshold and quantifier order. They are consequences of the theorem, not extra roles in its statement.^[1]^[3]

Structural Signature¶

Sig role-phrases: fixed arithmetic target → distinct rational approximants → real error → fixed exponent margin → denominator threshold → finite-exception conclusion.

R1 · Fixed real algebraic irrational target. Hold α fixed, with algebraic degree at least two. Replacing it with an arbitrary transcendental or complex number changes the asserted class.^[1]
R2 · Reduced rational approximants. Let p/q range over distinct rationals, normalize q > 0 and gcd(p, q) = 1, and use q as the size parameter. Lowest terms avoids counting multiple spellings of one rational as different exceptions.^[1]
R3 · Absolute approximation error. Compare p/q with α using the ordinary absolute value on the real line. A different norm or algebraic approximants requires a different statement.^[1]
R4 · Fixed positive exponent margin. Choose one ε > 0 before varying p/q. An ε that shrinks with q is not covered by this fixed-parameter conclusion.^[1]
R5 · Denominator-power threshold. Test the strict inequality against q raised to minus (2 + ε). Exponent 2 itself is the sharp boundary, not a permitted positive margin.^[1]
R6 · Finite-exception guarantee. Only finitely many of the distinct fractions pass the R5 test for each fixed pair (α, ε). This is qualitative finiteness, not an enumerated exception list.^[1]^[2]

What It Is Not¶

Roth's theorem is not a ban on every extraordinarily good rational approximation. Finitely many such approximants are consistent with it. It is also not a finite-exception statement at ε = 0: exponent 2 is sharp because every irrational has infinitely many rational approximants at that scale.^[1]

It is not a converse saying that a real number with approximation exponent 2 must be algebraic. Nor does a collection of good fractions prove transcendence unless it contains infinitely many distinct reduced fractions satisfying one fixed positive exponent margin. A claimed threshold with ε = ε(q) tending to zero changes the quantifiers.^[1]^[3]

The entry does not assert a result for p-adic distance, arbitrary algebraic approximants or a general number field. The original paper's later paragraph on approximation by algebraic numbers of bounded degree has a separate corrigendum; the replacement formula has not been verified here and is excluded from this entry.^[4]

Scope of Application¶

Use this theorem in real Diophantine approximation when an algebraic irrational target and a denominator-dependent rational error bound are explicit. Its two source-grounded settings have different objects and inference directions:

Finiteness of a fixed cubic Thue equation. A solution of x³ − 2y³ = 1 with y ≠ 0 produces an unusually close rational approximation to the known algebraic irrational cube root of 2. Infinite equation solutions would contradict Roth at a fixed margin. This yields finiteness, not a computed solution list.^[1]
Transcendence of a sparse binary series. Partial sums of ξ = Σ(n≥0) 2⁽⁻³n) produce infinitely many distinct reduced rationals with a fixed exponent gap. If ξ were algebraic, Roth would forbid them; the contradiction establishes transcendence. Here algebraicity is a temporary assumption, not a fact about ξ.^[3]

These are theorem-native mathematical constructions, not independent empirical replications. The equation case applies the theorem forward to a known algebraic target; the series case applies it by contradiction to a target whose algebraicity is in question.

Clarity¶

Keep three claims apart. The theorem says finitely many fractions meet a fixed stronger-than-exponent-2 test. A separate calculation may show that one particular fraction meets it; that does not challenge the theorem. A still stronger calculation may construct infinitely many distinct fractions, but it rules out algebraicity only after irrationality and one fixed positive ε are established.^[1]^[3]

The distinction from an irrationality measure is a distinction of subject. A measure or exponent describes rational approximability of an individual number. Roth supplies a proved class-level conclusion for every real algebraic irrational under a stated convention. It evaluates a power-law boundary in that setting; the theorem is not itself the number-specific measure.^[1]^[3]

Manages Complexity¶

Rational approximants have unbounded denominators, so no finite inspection can establish the theorem's general conclusion. The useful compression is fix α and ε → normalize distinct p/q → compare error with q^(−2−ε) → infer only finitely many successes. The order matters: changing the target or ε while q grows does not instantiate the same claim.^[1]

The bound is qualitative. Bennett's discussion of effective irrationality measures distinguishes Roth's noncomputable constant in that account from special-class effective estimates. Nothing in the bare finite-exception statement supplies a computable maximum denominator, a numerical count of exceptions or a complete list of solutions to an equation.^[2]

Abstract Reasoning¶

Three moves recur. Direct application: fix a known algebraic irrational and show that an infinite family of proposed objects would induce infinitely many too-good rational approximations. Contrapositive use: construct infinitely many fractions at one fixed exponent gap, prove the target irrational, and reject algebraicity. Boundary check: distinguish ε > 0 from the sharp ε = 0 case before drawing a conclusion.^[1]^[3]

The cubic equation and sparse series below are complete worked carriers for the first two moves. Their explicit reductions are author derivations grounded in Roth's theorem and Waldschmidt's stated example; they are not presented as separately printed proofs or as a reproduction of Roth's twenty-page argument.

Knowledge Transfer¶

Transfer is literal across number-theoretic problems only when the same roles can be filled: a real algebraic irrational target, distinct reduced rationals, a q-power error estimate with fixed ε > 0, and a finite-versus-infinite conclusion. A cubic integer equation supplies approximants through a polynomial factorization; a sparse digit series supplies them through truncation tails. The carriers differ, but the theorem's arithmetic test stays the same.^[1]^[3]

Calling an unrelated process “hard to approximate” is analogy. Without rational denominators, the real algebraic target class and the exponent-2 boundary, this named theorem does not travel to that process. A broader idea about bounded approximation may warrant separate abstraction, but these sources do not establish one.

Examples¶

Canonical — cubic equation finiteness¶

Let α = ∛2 and consider integer solutions of x³ − 2y³ = 1. The polynomial T³ − 2 is irreducible over the rationals, so α is a real algebraic irrational. The case y = 0 gives only x = 1. For y ≠ 0 set t = x/y; then t³ = 2 + 1/y³, so 1 ≤ t < 2. Factorization gives

\[ 0 < |t-\alpha| = \frac{1}{|y|^3(t^2+\alpha t+\alpha^2)} < \frac{1}{3|y|^3} < |y|^{-5/2}. \]

For q = |y| choose p = sgn(y)·x; then p/q = t. Any common divisor of x and y would have its cube divide 1, so this fraction is reduced. If infinitely many integer solutions existed, their |y| would be unbounded and would supply infinitely many distinct fractions satisfying Roth's inequality with the fixed margin ε = ½. Thus only finitely many solutions exist. The exact cubic reduction is an author specialization of the original paper's homogeneous-form corollary; no solution list or effective size bound is claimed.^[1]

Mapped back: R1 is ∛2; R2 is p/q with q = |y|; R3 is the factored absolute error; R4 is ε = ½; R5 is q^(−5/2); R6 forbids infinitely many such fractions. The y = 0 case is finite independently, and the reduction is valid only because every proposed equation solution fills these roles.

Applied — sparse-series transcendence¶

Define ξ = Σ(n≥0) 2⁽⁻³n). Its binary expansion has 1s at positions 1, 3, 9, 27, ... with unbounded gaps, so it is not eventually periodic and is irrational. The Nth partial sum is p_N/q_N with q_N = 2⁽³N). Its last numerator term is 1 and all earlier terms are even, so p_N is odd and the fraction is reduced. The partial sums are distinct and q_N grows without bound.^[3]

The positive tail obeys 0 < ξ − p_N/q_N < 2q_N^(−3). For q_N > 4 this is below q_N^(−5/2), again a fixed ε = ½ gap. Suppose, for contradiction, that ξ were algebraic. Since ξ is already known irrational, Roth would permit only finitely many of these fractions, while the construction yields infinitely many. Therefore ξ is transcendental. Waldschmidt identifies the sparse 3^n series as a Roth consequence; the explicit numerator and tail calculations here spell out the construction.^[3]

Mapped back: R1 is the conditional assumption that ξ is a real algebraic irrational; R2 is the sequence of reduced partial sums; R3 is their positive tail; R4 is the fixed ε = ½; R5 is q_N^(−5/2); R6 yields the contradiction. The conclusion rejects R1 rather than claiming a transcendental number is an actual algebraic instance.

Structural Tensions¶

No source-grounded two-sided design trade-off is constitutive of this theorem. The sharp exponent-2 boundary, finitely permitted exceptions and lack of an effective exception list are mathematical scope and proof-output limits, not choices between competing objectives. Keeping this section empty of invented “tensions” preserves the exact quantified identity.^[1]^[2]

Structural–Framed Character¶

The theorem is structural within arithmetic but strongly domain-framed. Its evaluative weight is a formal truth claim, not a preference that one approximation is “better” for a user. Its statement does not depend on a human workflow, even though proof, notation and applications are mathematical practices. Its 1955 attribution and publication are historical facts, not institutions that make the quantified relation true.^[1]

Its vocabulary travels only while the arithmetic roles travel: real algebraic irrational, reduced p/q, ordinary absolute error and exponent 2 + ε. The two worked uses recognize the same theorem in unlike number-theoretic carriers; they do not import it into a domain lacking those roles. Its character: a universal guarantee over a specific arithmetic class, structural in its quantifier and threshold relation, and domain-specific in the objects that make that relation true.

Structural Core vs. Domain Accent¶

The bare phrase “a threshold admits only finitely many exceptions” sketches a portable relation. It is not enough to identify Roth's theorem: its real algebraic target, rational denominator, absolute error, fixed margin and exponent-2 boundary are constitutive, not detachable accents. Remove them and neither the theorem nor its two applications follows.^[1]

The live Prime Approximation concerns a deliberately chosen tractable surrogate with an error a use case can tolerate. Roth instead constrains all rational approximants to each target in its arithmetic class. A future Prime about class-wide finite exceptions would need unlike, independently grounded non-number-theoretic cases with the same necessary roles. The present sources establish no such broader identity; the generic skeleton remains a future-Prime question, not a parent asserted for this named theorem.

This entry is a kind of Formal theorem.

Approximation is related through rational error, but its chosen-surrogate and use-tolerance identity is different from Roth's all-approximants finiteness statement. Irrationality Measure is a related domain-specific quantity or bounding function assigned to a number; Roth proves a class-wide power-law restriction for algebraic irrationals. Diophantine Equation is an application class: only equations whose solutions yield the required approximation gap support this reduction. These distinctions are identity tests, not claims that every neighboring topic is a parent.^[1]^[2]

Relationships to Other Abstractions¶

Current abstraction Roth's Theorem (Diophantine Approximation) Domain-specific

Parents (1) — more general patterns this builds on

Roth's Theorem (Diophantine Approximation) is a kind of Formal theorem Domain-specific

Roth's Theorem is a proved mathematical statement with a specific algebraic rational-approximation finiteness claim.

Hierarchy path (1) — routes to 1 parentless root

Roth's Theorem (Diophantine Approximation) → Formal theorem

Neighborhood in Abstraction Space¶

Roth's Theorem (Diophantine Approximation) sits in a sparse region of the domain-specific corpus (77^th percentile for distinctiveness): few abstractions share its structure, so a faithful description tends to retrieve it precisely.

Family — Number-Theoretic Properties & Tests (22 abstractions)

Nearest neighbors

Arakelian's Theorem — 0.83
Trial Division — 0.83
Average Order of an Arithmetic Function — 0.83
Least Common Multiple — 0.82
Arithmetic–Geometric Mean — 0.82

Computed from structural-signature embeddings · 2026-10-17

Not to Be Confused With¶

The 1953 Roth theorem on three-term arithmetic progressions. It is a different theorem by the same mathematician, not a statement about rational approximation.^[5]
Liouville-type bounds. Earlier algebraic approximation bounds have different exponents and do not by themselves give Roth's sharp 2 + ε finite-exception conclusion.^[1]
Ridout's restricted-denominator theorem. A sparse series with exponent-growth ratio 2 does not get the same fixed positive margin from the argument above; a stronger restricted-denominator result is separate.^[3]
A general irrationality measure or exponent. That is a number-specific description. Roth is a theorem about every member of a specified algebraic class.^[1]
An effective calculation of exceptions. The bare theorem does not compute all admissible fractions, an exact count or a maximum denominator.^[2]

References¶

[1] K. F. Roth, “Rational Approximations to Algebraic Numbers,” Mathematika 2, no. 1 (1955): 1–20, DOI 10.1112/S0025579300000644, original published article PDF, especially printed pp. 1–2 for inequality, theorem, sharpness and homogeneous-form corollary. The full article was retained, but its general proof was not independently reproduced. registry ↩a ↩b ↩c ↩d ↩e ↩f ↩g ↩h ↩i ↩j ↩k ↩l ↩m ↩n ↩o ↩p ↩q ↩r ↩s ↩t ↩u ↩v ↩w

[2] Michael A. Bennett, “Effective Measures of Irrationality for Certain Algebraic Numbers,” Journal of the Australian Mathematical Society (Series A) 62 (1997): 329–344, DOI 10.1017/S144678870000104X, original article PDF, especially introduction pp. 329–330 on the noncomputable Roth constant in that account and special effective bounds. No numerical bound from damaged OCR is used here. registry ↩a ↩b ↩c ↩d ↩e ↩f

[3] Michel Waldschmidt, “Words and Transcendence,” in Analytic Number Theory (2009), 449–470, author-hosted typeset chapter, §4.1, printed pp. 463–464, Theorem 4.2 and the explicit Σg⁽⁻³n) example. This is an author exposition of the theorem and example, not a second independent proof of Roth. registry ↩a ↩b ↩c ↩d ↩e ↩f ↩g ↩h ↩i ↩j

[4] K. F. Roth, “Corrigendum,” Mathematika 2, no. 2 (1955): 168, DOI 10.1112/S0025579300000826, publisher metadata and partial extract. The full one-page correction was not inspected; its exact replacement statement is excluded. registry ↩

[5] Ben Green, “Arithmetic Progressions at the Journal of the LMS,” Journal of the London Mathematical Society 113, no. 3 (2026): e70483, DOI 10.1112/jlms.70483, open-access publisher article, §2 immediately after Theorem 2.1. The full publisher HTML explicitly distinguishes Roth’s 1953 three-term-progression theorem from the Diophantine-approximation theorem. Green's retrospective was inspected; Roth's source paper for the progression result was not. registry ↩