Irrationality Complex: Measures & Structures

• Irrationality Complex is a framework that quantifies non-rationality using measures from Diophantine approximation, continued fractions, and geometric invariants.
• It integrates arithmetic, geometric, and algorithmic layers to assess approximation quality, certificate constructions, and computability in both numbers and algebraic varieties.
• The approach bridges number theory and algebraic geometry by linking irrationality measures with birational invariants, monodromy obstructions, and choice behavior metrics.

Searching arXiv for recent and core papers on irrationality measures, degree of irrationality, and related uses of “irrationality complexity.” In the literature surveyed here, “irrationality complex” functions as an umbrella label for the structured ways in which irrationality is quantified, encoded, or obstructed. In Diophantine approximation it is organized by the irrationality measure $\mu(\alpha)$, continued fractions, partial quotients, trigonometric estimates, and associated convergence phenomena; in birational geometry it is expressed by the degree of irrationality $\operatorname{irr}(X)$ and related fibrational invariants; and in monodromy, proof theory, and choice theory it refers to broader networks of quantitative constraints separating rational from non-rational behavior (Carella, 2019, Levinson et al., 2023, Smith, 2019, Carpentiere et al., 2023).

1. Diophantine approximation as the primary arithmetic layer

For an irrational real number $\alpha$, the irrationality measure or irrationality exponent $\mu(\alpha)$ describes how well $\alpha$ can be approximated by rational numbers. One standard definition is

$$\mu(\alpha)=\inf\Bigl\{\mu\ge 1:\exists\,c(\alpha,\mu)>0\text{ such that }\Bigl|\alpha-\frac pq\Bigr|>\frac{c(\alpha,\mu)}{q^\mu}\text{ for all but finitely many }\frac pq\in\mathbb Q\Bigr\}.$$

Equivalently, $\mu(\alpha)$ is the supremum of the exponents $z$ for which

$0<\left|x-\frac pq\right|<\frac1{q^z}$

has infinitely many integer solutions $(p,q)$ with $\operatorname{irr}(X)$0. Dirichlet’s approximation theorem implies $\operatorname{irr}(X)$1 for every irrational $\operatorname{irr}(X)$2. Liouville numbers are exactly those with $\operatorname{irr}(X)$3, while irrational algebraic numbers and Lebesgue-almost every real number satisfy $\operatorname{irr}(X)$4 (Carella, 2019, Becher et al., 2014).

This makes $\operatorname{irr}(X)$5 a numerical measure of what one summary explicitly calls “Diophantine complexity”: $\operatorname{irr}(X)$6 corresponds to the generic and minimal irrational case, whereas large $\operatorname{irr}(X)$7 signals exceptionally close rational approximations (Carella, 2019). Within this framework, the 2019 paper “Irrationality Measure of Pi” claims that for every $\operatorname{irr}(X)$8,

$\operatorname{irr}(X)$9

has only finitely many rational solutions, and therefore $\alpha$0 (Carella, 2019). That claim is situated against the historical bounds $\alpha$1 due to Mahler and $\alpha$2 due to Salikhov. The same paper also records that one of its internal arguments uses the claimed boundedness of the partial quotients of $\alpha$3, while the broader literature regards boundedness of the partial quotients of $\alpha$4 as open, so that line of reasoning is highly nontrivial and not accepted (Carella, 2019).

The arithmetic scope of the irrationality complex also includes collective irrationality statements when individual cases remain unresolved. A 2021 result proves that at least two of $\alpha$5 are irrational, and that at least one of $\alpha$6 is irrational (Lai et al., 2021). In a different but related direction, a 2013 transcendence result shows that for transcendental $\alpha$7, at least one of $\alpha$8 and $\alpha$9 is transcendental; in particular, at least one of $\mu(\alpha)$0 and $\mu(\alpha)$1 is transcendental, and at least two of $\mu(\alpha)$2, $\mu(\alpha)$3, and $\mu(\alpha)$4 are transcendental (Lima, 2013).

2. Continued fractions, geometric profiles, and computability strata

A central structural encoding of irrationality is the continued fraction expansion $\mu(\alpha)$5 with convergents $\mu(\alpha)$6. Large partial quotients correspond to unusually good rational approximations, bounded partial quotients force $\mu(\alpha)$7, and periodic continued fractions characterize quadratic irrationals (Carella, 2019, Morales-Almazan, 2015). The 2015 paper “A geometrical approach to measure irrationality” recasts this data geometrically: for $\mu(\alpha)$8, let $\mu(\alpha)$9 be the largest circular sector of radius $\alpha$0, centered at the origin, symmetric with respect to the line $\alpha$1, and containing no integer lattice point in its interior. Its area is

$\alpha$2

where $\alpha$3 is the aperture. The function $\alpha$4 is piecewise controlled by the convergents $\alpha$5, and its local extrema are bounded in terms of the ratios $\alpha$6 and the continued fraction coefficients (Morales-Almazan, 2015).

For quadratic irrationals, continued fractions are eventually periodic, the ratios $\alpha$7 are asymptotically periodic, and $\alpha$8 remains trapped between positive finite bounds (Morales-Almazan, 2015). By contrast, the same paper proves that $\alpha$9 is a subsequential limit of the local minima and the local maxima are unbounded if and only if the partial quotients $$\mu(\alpha)=\inf\Bigl\{\mu\ge 1:\exists\,c(\alpha,\mu)>0\text{ such that }\Bigl|\alpha-\frac pq\Bigr|>\frac{c(\alpha,\mu)}{q^\mu}\text{ for all but finitely many }\frac pq\in\mathbb Q\Bigr\}.$$0 are unbounded. This suggests that $$\mu(\alpha)=\inf\Bigl\{\mu\ge 1:\exists\,c(\alpha,\mu)>0\text{ such that }\Bigl|\alpha-\frac pq\Bigr|>\frac{c(\alpha,\mu)}{q^\mu}\text{ for all but finitely many }\frac pq\in\mathbb Q\Bigr\}.$$1 is a geometric re-expression of the same approximation complexity measured arithmetically by $$\mu(\alpha)=\inf\Bigl\{\mu\ge 1:\exists\,c(\alpha,\mu)>0\text{ such that }\Bigl|\alpha-\frac pq\Bigr|>\frac{c(\alpha,\mu)}{q^\mu}\text{ for all but finitely many }\frac pq\in\mathbb Q\Bigr\}.$$2 (Morales-Almazan, 2015).

Computability theory supplies a further layer. The 2014 paper “The Irrationality Exponents of Computable Numbers” proves that a real number $$\mu(\alpha)=\inf\Bigl\{\mu\ge 1:\exists\,c(\alpha,\mu)>0\text{ such that }\Bigl|\alpha-\frac pq\Bigr|>\frac{c(\alpha,\mu)}{q^\mu}\text{ for all but finitely many }\frac pq\in\mathbb Q\Bigr\}.$$3 is the irrationality exponent of some computable real number if and only if $$\mu(\alpha)=\inf\Bigl\{\mu\ge 1:\exists\,c(\alpha,\mu)>0\text{ such that }\Bigl|\alpha-\frac pq\Bigr|>\frac{c(\alpha,\mu)}{q^\mu}\text{ for all but finitely many }\frac pq\in\mathbb Q\Bigr\}.$$4 is the upper limit of a computable sequence of rational numbers, equivalently if $$\mu(\alpha)=\inf\Bigl\{\mu\ge 1:\exists\,c(\alpha,\mu)>0\text{ such that }\Bigl|\alpha-\frac pq\Bigr|>\frac{c(\alpha,\mu)}{q^\mu}\text{ for all but finitely many }\frac pq\in\mathbb Q\Bigr\}.$$5 is right-computably enumerable in $$\mu(\alpha)=\inf\Bigl\{\mu\ge 1:\exists\,c(\alpha,\mu)>0\text{ such that }\Bigl|\alpha-\frac pq\Bigr|>\frac{c(\alpha,\mu)}{q^\mu}\text{ for all but finitely many }\frac pq\in\mathbb Q\Bigr\}.$$6 (Becher et al., 2014). Consequently there exist computable real numbers whose irrationality exponent is not computable. The same paper recalls Jarník’s dimension statement

$$\mu(\alpha)=\inf\Bigl\{\mu\ge 1:\exists\,c(\alpha,\mu)>0\text{ such that }\Bigl|\alpha-\frac pq\Bigr|>\frac{c(\alpha,\mu)}{q^\mu}\text{ for all but finitely many }\frac pq\in\mathbb Q\Bigr\}.$$7

and constructs Cantor-like sets whose natural measure concentrates on numbers with prescribed irrationality exponent $$\mu(\alpha)=\inf\Bigl\{\mu\ge 1:\exists\,c(\alpha,\mu)>0\text{ such that }\Bigl|\alpha-\frac pq\Bigr|>\frac{c(\alpha,\mu)}{q^\mu}\text{ for all but finitely many }\frac pq\in\mathbb Q\Bigr\}.$$8 (Becher et al., 2014). In this sense, the irrationality complex of a computable real includes both Diophantine data and arithmetical-hierarchy data.

3. Certificates, determinant methods, phase integrals, and automated discovery

A recurrent theme is the search for finite irrationality certificates. The classical criterion says that if integers $$\mu(\alpha)=\inf\Bigl\{\mu\ge 1:\exists\,c(\alpha,\mu)>0\text{ such that }\Bigl|\alpha-\frac pq\Bigr|>\frac{c(\alpha,\mu)}{q^\mu}\text{ for all but finitely many }\frac pq\in\mathbb Q\Bigr\}.$$9 satisfy $\mu(\alpha)$0 and $\mu(\alpha)$1, then $\mu(\alpha)$2 is irrational. Zudilin’s determinantal refinement replaces single linear forms by Hankel determinants of moment sequences $\mu(\alpha)$3, and under an integral representation $\mu(\alpha)$4 with divisibility and growth control, irrationality follows from the weaker inequality

$\mu(\alpha)$5

instead of $\mu(\alpha)$6 (Zudilin, 2015). This yields, among other consequences, a new proof of the irrationality of $\mu(\alpha)$7, as well as determinantal re-proofs of the irrationality of $\mu(\alpha)$8 and $\mu(\alpha)$9 (Zudilin, 2015).

A different reformulation comes from oscillatory integrals. The 2013 paper “Geometric Phase Integrals and Irrationality Tests” shows that the existence of isolated real solutions of analytic systems, and in particular the rationality of $z$0, can be encoded by convergence of the phase of a complex integral

$z$1

as $z$2, where $z$3 is a nonnegative analytic “geometric Lagrangian” vanishing exactly on the target solution set (Napoletani et al., 2013). For the Euler–Mascheroni constant, this produces an exact reformulation of the statement “$z$4 is rational” as existence of a certain phase limit, but not an irrationality proof (Napoletani et al., 2013).

Experimental and symbolic-computation approaches push the certificate paradigm in a different direction. The 2019 paper “Automatic Discovery of Irrationality Proofs and Irrationality Measures” uses the Almkvist–Zeilberger algorithm and creative telescoping to generate recurrences for integral families $z$5, convert them into linear forms in constants such as $z$6, dilogarithms, or logarithmic triples, and extract irrationality measures from dominant and subdominant characteristic roots. It defines an empirical exponent

$z$7

for rational approximants $z$8, and if $z$9 stabilizes to $0<\left|x-\frac pq\right|<\frac1{q^z}$0, this suggests $0<\left|x-\frac pq\right|<\frac1{q^z}$1 (Zeilberger et al., 2019). The same paper presents Maple packages for Alladi–Robinson-type, Beukers-type, and Salikhov-type constructions (Zeilberger et al., 2019).

The 2026 paper “Tail Criteria, No-Go Audits, and Apéry-Type Certificate Obstructions for the Irrationality of $0<\left|x-\frac pq\right|<\frac1{q^z}$2” turns certificate search itself into an object of study. It proves that $0<\left|x-\frac pq\right|<\frac1{q^z}$3 is equivalent to three eventual factorial-arithmetic phenomena: an eventual ceiling recurrence for $0<\left|x-\frac pq\right|<\frac1{q^z}$4, an eventual factorial-Cantor digit condition $0<\left|x-\frac pq\right|<\frac1{q^z}$5, and an eventual divisibility condition $0<\left|x-\frac pq\right|<\frac1{q^z}$6 for a natural sequence $0<\left|x-\frac pq\right|<\frac1{q^z}$7 (Yu, 15 Jun 2026). It then formulates an Apéry-type certificate framework based on integer linear forms $0<\left|x-\frac pq\right|<\frac1{q^z}$8 and audits several low-complexity mechanisms, including mixed Padé approximation, crossed separate approximations to $0<\left|x-\frac pq\right|<\frac1{q^z}$9 and $(p,q)$0, simple $(p,q)$1-fractions, holonomic ansatzes, Rodrigues-type families, and an integer kernel-lattice search. In the final kernel-lattice audit, 145 raw candidates reduce to 133 primitive records; the best signals are dominated by continued-fraction shadows, while non-CF candidates do not form a degree-continuing family (Yu, 15 Jun 2026).

4. Birational irrationality as a geometric complexity invariant

In algebraic geometry, irrationality complex is expressed by the degree of irrationality. For an irreducible complex projective variety $(p,q)$2 of dimension $(p,q)$3,

$(p,q)$4

This is a birational invariant, $(p,q)$5 if and only if $(p,q)$6 is rational, and for curves it coincides with gonality (Levinson et al., 2023, 1603.05543). Related invariants include the stable degree of irrationality $(p,q)$7, the unirational degree of irrationality $(p,q)$8, the covering gonality, and the connecting gonality, with

$(p,q)$9

for smooth projective varieties (1603.05543).

The 2023 paper “Minimal degree fibrations in curves and the asymptotic degree of irrationality of divisors” introduces the minimal fibering degree $\operatorname{irr}(X)$00, the minimal $\operatorname{irr}(X)$01-degree of curves that appear as general fibers of a dominant rational map $\operatorname{irr}(X)$02 (Levinson et al., 2023). Its Theorem A states that for $\operatorname{irr}(X)$03 smooth projective of dimension $\operatorname{irr}(X)$04, $\operatorname{irr}(X)$05 ample, $\operatorname{irr}(X)$06 effective, and $\operatorname{irr}(X)$07 smooth with $\operatorname{irr}(X)$08, any map $\operatorname{irr}(X)$09 computing $\operatorname{irr}(X)$10 factors through a minimal $\operatorname{irr}(X)$11-degree fibration $\operatorname{irr}(X)$12, and

$\operatorname{irr}(X)$13

When $\operatorname{irr}(X)$14, one has $\operatorname{irr}(X)$15; if every minimal degree fibration is regular, then $\operatorname{irr}(X)$16 (Levinson et al., 2023).

This yields explicit asymptotics for complete intersections. If $\operatorname{irr}(X)$17 is a general complete intersection of sufficiently large and sufficiently unbalanced degrees $\operatorname{irr}(X)$18, then for every $\operatorname{irr}(X)$19,

$\operatorname{irr}(X)$20

(Levinson et al., 2023). Here irrationality becomes a quantitative birational complexity measure rather than a yes-or-no rationality test.

For smooth surfaces $\operatorname{irr}(X)$21 of degree $\operatorname{irr}(X)$22, the 2016 paper “On irrationality of surfaces in $\operatorname{irr}(X)$23” gives a nearly complete hierarchy. It proves

$\operatorname{irr}(X)$24

and

$\operatorname{irr}(X)$25

while $\operatorname{irr}(X)$26 (1603.05543). For a very general $\operatorname{irr}(X)$27, one has $\operatorname{irr}(X)$28, computed only by projections from points of $\operatorname{irr}(X)$29 (1603.05543).

5. Asymptotics in families, ruled targets, and rationally connected examples

The 2019 paper “Fano hypersurfaces with arbitrarily large degrees of irrationality” shows that irrationality complexity can be large even for rationally connected varieties (Chen et al., 2019). For a very general Fano hypersurface $\operatorname{irr}(X)$30 of dimension $\operatorname{irr}(X)$31 and fixed Fano index $\operatorname{irr}(X)$32, there exists $\operatorname{irr}(X)$33 such that for all $\operatorname{irr}(X)$34,

$\operatorname{irr}(X)$35

More precisely, the same lower bound holds for $\operatorname{irr}(X)$36, the minimal degree of a dominant rational map from $\operatorname{irr}(X)$37 to a ruled variety (Chen et al., 2019). The paper notes that these are the first examples of rationally connected varieties with degree of irrationality greater than $\operatorname{irr}(X)$38 (Chen et al., 2019).

The method combines degeneration to characteristic $\operatorname{irr}(X)$39, Kollár’s positivity construction, and a specialization theorem for maps to ruled varieties. In a flat projective family, if the generic fiber admits a dominant generically finite rational map of degree at most $\operatorname{irr}(X)$40 to a ruled variety, then every irreducible component of the special fiber does as well (Chen et al., 2019). This makes $\operatorname{irr}(X)$41 especially stable under specialization, and in certain families of surfaces and strict Calabi–Yau threefolds it implies corresponding specialization control for $\operatorname{irr}(X)$42 itself (Chen et al., 2019).

One consequence is that every complex abelian surface $\operatorname{irr}(X)$43 satisfies

$\operatorname{irr}(X)$44

(Chen et al., 2019). The broader implication is that irrationality complexity in algebraic geometry behaves simultaneously as a quantitative invariant and as a specialization-sensitive structure on families.

6. Hodge-theoretic and monodromy obstructions

For cubic threefolds, irrationality can be encoded in monodromy rather than in approximation or degree estimates. The 2019 paper “Irrationality and monodromy for cubic threefolds” studies the universal family $\operatorname{irr}(X)$45 of smooth cubic threefolds and its cohomological monodromy

$\operatorname{irr}(X)$46

equivalently the monodromy of the intermediate Jacobian map $\operatorname{irr}(X)$47 (Smith, 2019). Its main theorem states that $\operatorname{irr}(X)$48 does not factor through the mapping class group of any closed oriented surface of total genus $\operatorname{irr}(X)$49, and in particular does not factor through $\operatorname{irr}(X)$50 (Smith, 2019).

The proof uses Lönne’s presentation of $\operatorname{irr}(X)$51 as a quotient of an Artin group, Picard–Lefschetz transvections, rigidity results for homomorphisms $\operatorname{irr}(X)$52, and an explicit obstruction to realizing the required Artin graph by curves on a genus-$\operatorname{irr}(X)$53 surface (Smith, 2019). This gives what the paper calls a geometric group theory perspective on the well-known irrationality of cubic threefolds (Smith, 2019).

This suggests a further layer of the irrationality complex: irrationality may be witnessed not only by approximation exponents or by birational degrees, but also by the impossibility of realizing a variation of Hodge structure as curve-like monodromy. In this setting, the obstruction is not a numerical distance from rationality but a non-factorization theorem in symplectic and mapping-class-group terms.

7. Extension of the term beyond arithmetic geometry

The phrase also appears in a metric theory of choice behavior. The 2023 paper “A rational measure of irrationality” defines deterministic choice behavior as a quasi-choice correspondence $\operatorname{irr}(X)$54, chooses rationalizable behaviors as the benchmark of rationality, equips the space of behaviors with a metric, and defines the degree of irrationality by

$\operatorname{irr}(X)$55

Here irrationality becomes, in the paper’s own words, a graded, geometric notion: how far a behavior lies from the rational core, and in what way (Carpentiere et al., 2023).

The same paper introduces a refined metric $\operatorname{irr}(X)$56 via local rationalizations $\operatorname{irr}(X)$57, designed to reflect Chernoff’s Axiom $\operatorname{irr}(X)$58 and Sen’s Axiom $\operatorname{irr}(X)$59, and then extends the framework to stochastic choice by taking the random utility model as the benchmark of rationality and using Block–Marschak polynomials to measure deviations from it (Carpentiere et al., 2023). This is not a number-theoretic use of irrationality, but it preserves the same structural idea: irrationality is measured by a profile of constraints, distances, and local obstructions rather than by a single binary label.

Across these settings, the common content of the irrationality complex is therefore not a single invariant but a family of interlocking profiles. For real numbers it is the web formed by $\operatorname{irr}(X)$60, continued fractions, geometric sectors, computability classes, and certificate constructions; for varieties it is the network of $\operatorname{irr}(X)$61, $\operatorname{irr}(X)$62, stable and unirational refinements, specialization behavior, and monodromy obstructions; and in adjacent quantitative theories it is the metric distance from rational benchmarks. The term names the architecture of irrationality rather than one isolated test for it.