The abc Conjecture and IUT Theory

⚡ Key Takeaways

One-line statement: The abc conjecture asserts that for coprime positive integers with $a + b = c$ , the product $c$ cannot be much larger than the radical (squarefree kernel) of $abc$ ; Inter-universal Teichmüller (IUT) theory is Shinichi Mochizuki's vast framework claiming to prove it. Key result: For every $\varepsilon > 0$ , only finitely many coprime triples satisfy $c > \operatorname{rad}(abc)^{1+\varepsilon}$ . Where it applies: Diophantine geometry, number theory, anabelian geometry, the structure of elliptic curves and their associated theta functions. Who/when: Conjectured by Joseph Oesterlé and David Masser (1985); claimed proof via IUT by Shinichi Mochizuki (2012, published 2021), whose validity remains contested.

💡 The Core Insight

Addition and multiplication of integers are "independent" in a precise sense: if you know the prime factorizations of $a$ and $b$ , you learn almost nothing about the prime factorization of $a+b$ . The abc conjecture turns this intuition into a quantitative law — the radical $\operatorname{rad}(abc)$ , which strips out exponents, controls $c$ up to an arbitrarily small power. Equivalently, large prime powers ("highly multiplicative" numbers) are additively rigid: they cannot conspire to form an additive relation $a+b=c$ unless their radical is comparably large.

DefinitionRadical of an integer

For a nonzero integer $n$ , the radical (or squarefree kernel) is the product of the distinct primes dividing $n$ :

1

\quad \operatorname{rad}(n) = \prod_{p \mid n} p.

Derivation hint: Take the prime factorization $n = \prod_p p^{e_p}$ and set every nonzero exponent $e_p$ to $1$ ; the radical discards multiplicity entirely. Here $p$ ranges over the distinct prime divisors of $n$ . For example $\operatorname{rad}(72) = \operatorname{rad}(2^3 \cdot 3^2) = 2 \cdot 3 = 6$ .

Definitionabc triple and quality

An abc triple is an ordered triple $(a,b,c)$ of positive integers with

2

\quad a + b = c, \qquad \gcd(a,b) = 1.

Derivation hint: Coprimality of $a$ and $b$ forces $\gcd(a,c)=\gcd(b,c)=1$ as well, since any common prime would divide the third by $a+b=c$ . The quality (or merit) of the triple is

3

\quad q(a,b,c) = \frac{\log c}{\log \operatorname{rad}(abc)}.

Derivation hint: Take logarithms of the heuristic relation $c \approx \operatorname{rad}(abc)^{q}$ ; quality measures how much "more multiplicative depth" $c$ has than its radical.

TheoremThe abc Conjecture (Oesterlé–Masser, 1985)

For every real $\varepsilon > 0$ there exists a constant $K_\varepsilon > 0$ such that for every abc triple,

4

\quad c \;<\; K_\varepsilon \cdot \operatorname{rad}(abc)^{\,1+\varepsilon}.

Derivation hint: This is the assertion that $q(a,b,c) \to 1$ in the limit — only finitely many triples exceed any fixed quality above $1$ , after absorbing the slack into $K_\varepsilon$ . Equivalently, for each $\varepsilon > 0$ the set of abc triples with $q(a,b,c) > 1+\varepsilon$ is finite. The strong (effective) form would specify $K_\varepsilon$ explicitly; the weak form merely asserts existence.

TheoremEquivalent radical-difference form

The conjecture is equivalent to: for all $\varepsilon>0$ ,

5

\quad \max(|a|,|b|,|c|) \ll_\varepsilon \operatorname{rad}(abc)^{1+\varepsilon}

for every coprime solution of $a+b+c=0$ in nonzero integers. Derivation hint: Set $c' = -c$ in $a+b=c$ to obtain the symmetric homogeneous form $a+b+c=0$ ; the radical and max are symmetric under sign changes since $\operatorname{rad}(-n)=\operatorname{rad}(n)$ .

ProofProbabilistic heuristic for finiteness

We sketch why triples of high quality should be rare. Let $R = \operatorname{rad}(abc)$ . 1. Density of squarefree-radical-bounded numbers. The number of integers up to $x$ that are "products of primes from a fixed radical $R$ " grows only polylogarithmically; integers with all prime factors $\le R^{1+\varepsilon}$ and radical $\le R$ are sparse. 2. Counting solutions. Model $a$ , $b$ as random integers with the constraint that $\operatorname{rad}(abc) \le R$ . Heuristically the probability that a random pair near size $N$ sums to a third number while keeping the combined radical at most $N^{1/(1+\varepsilon)}$ decays. 3. Borel–Cantelli style estimate. Summing the expected number of triples with $q > 1+\varepsilon$ over dyadic ranges of $c$ gives a convergent series:

6

\quad \sum_{c} \Pr\big[q(a,b,c) > 1+\varepsilon\big] \;<\; \infty.

Derivation hint: The probability that a $c$ of given size has radical smaller than $c^{1/(1+\varepsilon)}$ falls off faster than $1/c$ once $\varepsilon>0$ , making the sum converge. 4. Conclusion. Convergence implies only finitely many "successes," matching the conjecture's finiteness assertion. $\blacksquare$ (heuristic only)

∎

ProofConditional derivation: FLT asymptotically from abc

Suppose abc holds. Consider Fermat's equation $x^n + y^n = z^n$ with coprime $x,y,z$ and $n \ge 4$ . 1. Set $a = x^n$ , $b = y^n$ , $c = z^n$ . Then $a+b=c$ and the triple is coprime. 2. The radical satisfies

7

\quad \operatorname{rad}(abc) = \operatorname{rad}(x^n y^n z^n) = \operatorname{rad}(xyz) \le xyz \le z^3.

Derivation hint: Radical removes exponents, so $\operatorname{rad}(x^n)=\operatorname{rad}(x)\le x$ ; bound each of $x,y$ by $z$ . 3. Apply [4] with $\varepsilon = 1$ (say): $z^n = c < K_1 \operatorname{rad}(abc)^{2} \le K_1 z^{6}$ . 4. Hence $z^{n-6} < K_1$ , which bounds $z$ for all $n > 6$ . So only finitely many solutions exist for each large $n$ , and none for $n$ sufficiently large relative to $K_1$ . $\blacksquare$

∎

mathematician · Shinichi Mochizuki | 1969–

Born in Tokyo and raised partly in the United States, Mochizuki entered Princeton at 16, completing his PhD under Gerd Faltings. He became a leading figure in anabelian geometry, proving the Grothendieck Conjecture for hyperbolic curves over number fields and developing " $p$ -adic Teichmüller theory." Over roughly two decades at the Research Institute for Mathematical Sciences (RIMS) in Kyoto, he constructed Inter-universal Teichmüller theory, releasing the four foundational papers in August 2012. The work claims a proof of the abc conjecture (in the Szpiro-inequality form) but is written in an idiosyncratic self-contained idiom. After an extended and contentious refereeing process, the papers appeared in Publications of RIMS (2021), with Mochizuki himself the journal's editor-in-chief — a fact that intensified controversy.

🧠 Intuition

Think of a number's "additive identity" (its size) versus its "multiplicative identity" (which primes, with what exponents, build it). A perfect power like $2^{30}$ is enormous but multiplicatively thin — its radical is just $2$ . The abc conjecture says you cannot have $a+b=c$ where all three are simultaneously this thin. If $a=2^{m}$ , $b=3^{n}$ , then $c=a+b$ is forced to be multiplicatively "fat": it must drag in many new primes, keeping $\operatorname{rad}(abc)$ large. Additive combination generates new prime factors. The famous record triple

2 + 3^{10}\cdot 109 = 23^5

shows how hard nature works to find even slightly fat radicals — and it remains the highest-quality triple known, with $q \approx 1.6299$ . The conjecture says such gems are finite for every quality threshold.

Figure

The abc inequality:

c

vs.

\operatorname{rad}(abc)

on a log-log plot

ExampleComputing quality of a record triple

Given: The triple $a = 2$ , $b = 3^{10}\cdot 109 = 6{,}436{,}341$ , $c = 23^5 = 6{,}436{,}343$ . Verify $a+b=c$ and compute $q$ .

Steps: 1. Check: $2 + 6{,}436{,}341 = 6{,}436{,}343 = 23^5$ . ✓ Coprime since $a=2$ , $b$ odd, $c$ odd. 2. Radical: $\operatorname{rad}(abc) = \operatorname{rad}(2 \cdot 3^{10} \cdot 109 \cdot 23^5) = 2 \cdot 3 \cdot 109 \cdot 23 = 15{,}042.$ 3. Logarithms: $\log c = \log(6{,}436{,}343) \approx 15.677$ ; $\log \operatorname{rad} = \log(15{,}042) \approx 9.619$ . 4. Quality:

8

\quad q = \frac{\log c}{\log \operatorname{rad}(abc)} = \frac{15.677}{9.619} \approx 1.6299.

Derivation hint: Substitute the natural logs from steps 2–3 directly into [3]; the base cancels so any logarithm base works.

Answer: $\boxed{q \approx 1.6299}$ — the largest known quality (Eric Reyssat's triple).

ExampleUsing abc to bound a Fermat-type equation

Given: Assume the explicit form $c < \operatorname{rad}(abc)^{2}$ (i.e. $K_\varepsilon=1$ , $\varepsilon=1$ , hypothetically). Bound solutions of $x^5 + y^5 = z^5$ in coprime positives.

Steps: 1. Set $a=x^5,\, b=y^5,\, c=z^5$ . 2. $\operatorname{rad}(abc) = \operatorname{rad}(xyz) \le xyz \le z^3$ (since $x,y \le z$ ). 3. Apply assumed bound: $z^5 = c < (\operatorname{rad}(abc))^2 \le (z^3)^2 = z^6$ . 4. This gives $z^5 < z^6$ , always true — so $n=5$ is not ruled out by $\varepsilon=1$ . Retry with the asymptotic exponent: solving $z^{n} < z^{3(1+\varepsilon)}$ forces $n < 3(1+\varepsilon)$ .

9

\quad n \le 3(1+\varepsilon) \implies \text{no solutions once } n > 3(1+\varepsilon).

Derivation hint: Cancel $z$ exponents from $z^n < z^{3(1+\varepsilon)}$ ; large $n$ contradicts the inequality, eliminating solutions for big $n$ .

Answer: $\boxed{\text{For } \varepsilon=1, \text{ no coprime solutions when } n \ge 7.}$ (The threshold sharpens as $\varepsilon\to 0$ with an effective $K_\varepsilon$ .)

⚠️ Common mistake

abc is not "proven" by community consensus. Although Mochizuki's papers are published, leading experts including Peter Scholze and Jakob Stix identified what they describe as a fatal gap in Corollary 3.12 of the third IUT paper (their 2018 report). Mochizuki disputes this, claiming a fundamental misunderstanding of his framework. As of now, the proof is not accepted by the broad mathematical community. Treat "abc is a theorem" as unverified.

⚠️ Common mistake

The constant $K_\varepsilon$ is not effective in the weak form. The conjecture as stated only asserts existence of $K_\varepsilon$ . Many spectacular Diophantine consequences (e.g. effective Mordell) require an effective version where $K_\varepsilon$ is computable, which is strictly stronger and not implied by mere finiteness.

CounterexampleWhy the exponent cannot be

1

(no

\varepsilon

)

The inequality $c < K \cdot \operatorname{rad}(abc)$ (i.e. $\varepsilon = 0$ ) is false for any constant $K$ . Consider $a = 1$ , $b = 2^{2^n} - 1$ , $c = 2^{2^n}$ . One shows $2^n \mid \operatorname{rad}(abc) \cdot (\text{something})$ is too weak: in fact using $a=1,\ b = 9^n - 1,\ c=9^n$ one can produce triples with $\dfrac{c}{\operatorname{rad}(abc)} \to \infty$ .

10

\quad \limsup_{(a,b,c)} \frac{c}{\operatorname{rad}(abc)} = +\infty.

Derivation hint: Take $c=2^{n}$ and exploit that $2^n -$ a structured power has a small radical infinitely often; the ratio is unbounded, killing $\varepsilon=0$ . The $\varepsilon>0$ slack is essential.

connection

The abc conjecture sits at a crossroads of arithmetic geometry. It is essentially equivalent to Szpiro's conjecture on the conductor and discriminant of Elliptic Curves: for a minimal elliptic curve, $|\Delta| \ll_\varepsilon N^{6+\varepsilon}$ , where $\Delta$ is the discriminant and $N$ the conductor — and IUT actually proves this Szpiro-type inequality directly. It implies the asymptotic Fermat's Last Theorem, generalizes the Mason–Stothers Theorem (the polynomial abc, which is proven elementarily), and tightly relates to the Mordell Conjecture (Faltings' theorem) through height bounds. IUT theory itself is built on Anabelian Geometry and the étale Fundamental Group, encoding arithmetic into the symmetries of Hodge–Arakelov Theory and theta functions.

TheoremMason–Stothers: the polynomial analogue (proven)

For nonzero coprime polynomials $a(t),b(t),c(t)$ over a field of characteristic $0$ with $a+b=c$ and not all constant,

11

\quad \max(\deg a, \deg b, \deg c) \;\le\; \big(\text{number of distinct roots of } abc\big) - 1.

Derivation hint: Differentiate $a+b=c$ to form a Wronskian $W = a b' - a'b$ ; its order of vanishing at each root is controlled by multiplicities, and $\deg W$ bounds the degrees — the function-field "radical" is the count of distinct roots. This is the exact analogue of abc with $\operatorname{rad}$ replaced by the squarefree part, and it is fully proven, lending strong credence to the number-theoretic conjecture.

🔬 Where this appears

⏳ History

- 1983 — Richard Mason (and independently A. Stothers, 1981) prove the polynomial abc theorem (Mason–Stothers), the function-field model.

- 1985 — Joseph Oesterlé, studying Szpiro's conjecture on elliptic curves, and David Masser, formulate the abc conjecture in its modern quantitative form.

- 1988 — Lucien Szpiro's conjecture is recognized as essentially equivalent to abc, tying it to the geometry of elliptic curves.

- 1991 — Noam Elkies shows abc implies an effective form of Mordell's conjecture (Faltings' theorem).

- 2006 — The ABC@Home distributed-computing project begins systematically cataloguing high-quality triples.

- August 2012 — Shinichi Mochizuki posts the four IUT papers claiming a proof via inter-universal Teichmüller theory.

- 2015 — Major workshops (Oxford, Kyoto/RIMS) attempt to disseminate IUT; experts report the core remains opaque.

- 2018 — Peter Scholze and Jakob Stix circulate a manuscript asserting a gap in Corollary 3.12; Mochizuki publishes detailed rebuttals.

- 2020–2021 — PRIMS accepts and publishes the IUT papers; the broader community remains unconvinced, and the abc conjecture is widely regarded as open.

open-problem

The central open question is simply whether the abc conjecture is true and, if Mochizuki's framework is correct, whether it can be communicated and verified in conventional terms. 1. The Scholze–Stix objection targets the identification of certain copies of arithmetic objects ("theaters") under indeterminacies; whether the claimed inequality survives a fully rigorous accounting of these indeterminacies is disputed. 2. Effectivity. Even granting abc, producing an explicit $K_\varepsilon$ remains a deep and largely open quantitative problem; IUT, if valid, yields an effective Szpiro inequality, but the constants are enormous. 3. The quality spectrum. Whether quality $q$ can equal or exceed specific thresholds (e.g. $q > 1.7$ ) for some triple is unknown; the supremum of $q$ is conjectured to be exactly... unknown, with $q \to 1^+$ being the abc bound but no a priori upper cap below the record. 4. Independent proofs. No proof of abc independent of IUT exists, despite partial results (e.g. abc holds with the exponent weakened, via Stewart–Yu's bounds $c < \exp(\operatorname{rad}(abc)^{1/3+\varepsilon})$ — far from polynomial).

Formal Statement

Why It Holds (Heuristic and Conditional Derivations)

Building Intuition

Worked Examples

Common Misconceptions

Connections

Real-World Applications

Historical Development

Subtleties & Limits

See Also