Fermat's Last Theorem — Wiles' Strategy

⚡ Key Takeaways

Fermat's Last Theorem (FLT): for integers $n > 2$ , the equation $x^n + y^n = z^n$ has no solution in positive integers $x, y, z$ . Wiles' strategy (1994, with Taylor): prove the modularity of a large enough class of elliptic curves so that a hypothetical FLT solution — packaged into the Frey curve — would yield a semistable elliptic curve whose Galois representation cannot exist, by Ribet's level-lowering theorem. Applies to number theory, arithmetic geometry, the Langlands program. Conjectured by Fermat (1637); proved by Andrew Wiles and Richard Taylor (1994–95).

💡 The Core Insight

A counterexample $a^p + b^p = c^p$ ( $p$ an odd prime) would produce the elliptic curve $y^2 = x(x-a^p)(x+b^p)$ whose mod- $p$ Galois representation is too unramified to come from any modular form (Ribet), yet would have to come from one if the curve is modular (Wiles). Proving every semistable elliptic curve is modular therefore makes the counterexample logically impossible. The engine is a numerical criterion — the Wiles–Lenstra–Taylor " $R=T$ " isomorphism — showing a universal deformation ring of Galois representations equals a Hecke algebra.

DefinitionFermat's Last Theorem

For every integer $n \ge 3$ , there are no positive integers $x, y, z$ satisfying

1

\quad x^n + y^n = z^n.

Derivation hint: This is the statement to be disproved-impossible; the symbols $x,y,z$ are positive integers, $n$ an integer exponent. It suffices to prove the cases $n = 4$ (Fermat, by infinite descent) and $n = p$ an odd prime, since any $n \ge 3$ is divisible by $4$ or by an odd prime, and a solution for $n$ yields one for any divisor of $n$ .

DefinitionElliptic curve over

\mathbb{Q}

A smooth projective genus-1 curve with a rational point, given in Weierstrass form

2

\quad E: \; y^2 = x^3 + a_2 x^2 + a_4 x + a_6, \qquad a_i \in \mathbb{Z},

with discriminant $\Delta_E \ne 0$ and conductor $N_E = \prod_\ell \ell^{f_\ell}$ . $E$ is semistable if $E$ has multiplicative or good reduction at every prime, equivalently $N_E$ is squarefree. Derivation hint: The conductor measures ramification of the Tate module; squarefree conductor $\Leftrightarrow$ reduction never additive.

DefinitionThe Frey–Hellegouarch curve

Given a hypothetical solution $a^p + b^p = c^p$ with $p \ge 5$ prime, $\gcd(a,b,c)=1$ , and $a \equiv 3 \pmod 4$ , $b$ even, define

3

\quad E_{a,b,c}: \; y^2 = x\,(x - a^p)\,(x + b^p).

Derivation hint: The three 2-torsion points $0, a^p, -b^p$ encode the Fermat solution; multiplicativity of the equation forces a remarkable discriminant. Its minimal discriminant and conductor are

4

\quad \Delta = 2^{-8}\,(abc)^{2p}, \qquad N = \prod_{\ell \mid abc} \ell .

Derivation hint: Compute $\Delta = 16\,(a^p b^p c^p)^2$ from the roots $0,a^p,-b^p$ ; the global minimal model strips $2^8$ ; squarefree conductor follows because each prime divides $abc$ to a multiplicity divisible by $p$ , giving multiplicative reduction.

TheoremModularity (Wiles–Taylor, semistable case)

Every semistable elliptic curve $E/\mathbb{Q}$ is modular: there exists a weight-2 newform $f$ of level $N_E$ such that for all primes $\ell \nmid N_E$ ,

5

\quad a_\ell(E) = a_\ell(f), \qquad a_\ell(E) := \ell + 1 - \#E(\mathbb{F}_\ell).

Derivation hint: Equate the $L$ -function of $E$ (Euler product over $a_\ell$ ) with the Mellin transform of a cusp form; equivalently the Galois representation on $T_\ell E$ matches the one attached to $f$ .

TheoremRibet's level-lowering (1986)

Let $f$ be a newform of level $N$ giving rise to an irreducible mod- $p$ representation $\bar\rho_f$ . If $\bar\rho_f$ is unramified at a prime $q \mid N$ with $q^2 \nmid N$ and arises from the $p$ -torsion in a semistable context, then $\bar\rho_f$ also arises from a newform of level $N/q$ .

TheoremFLT, deduced

The conjunction of the two preceding theorems implies Fermat's Last Theorem for all primes $p \ge 5$ (and hence all $n \ge 3$ , combined with $n=3,4$ ).

ProofFLT from Modularity + Ribet

Step 1 (Setup). Suppose $a^p + b^p = c^p$ with $p \ge 5$ prime, nontrivial and coprime; normalize $a \equiv 3 \pmod 4$ , $b$ even. Form the Frey curve $E = E_{a,b,c}$ of equation [3].

Step 2 (Semistability). By [4], $N_E = \prod_{\ell\mid abc}\ell$ is squarefree, so $E$ is semistable.

Step 3 (Galois representation). Let $\bar\rho_{E,p}: G_\mathbb{Q} \to \mathrm{GL}_2(\mathbb{F}_p)$ be the action on the $p$ -torsion $E[p]$ . Mazur's theorem (irreducibility of $E[p]$ for semistable $E$ , $p\ge 5$ ) gives that $\bar\rho_{E,p}$ is irreducible.

Step 4 (Ramification is tiny). Because $p \mid v_\ell(\Delta)$ for every prime $\ell \mid abc$ (each appears to power $2p$ in $\Delta$ ), the representation $\bar\rho_{E,p}$ is unramified outside $\{2,p\}$ and finite flat at $p$ . Its Serre conductor is $N=2$ .

6

\quad \bar\rho_{E,p} \text{ is unramified at all odd } \ell, \text{ Serre level } N(\bar\rho)=2.

Derivation hint: At $\ell\mid abc$ the curve has multiplicative reduction with $p \mid \mathrm{ord}_\ell(\Delta)$ ; the Tate parameter then shows the inertia acts through $p$ -th powers, i.e. trivially mod $p$ .

Step 5 (Modularity). By the Modularity Theorem [5], $E$ is modular: $\bar\rho_{E,p}$ arises from a weight-2 newform of level $N_E$ .

Step 6 (Level lowering). Apply Ribet's theorem repeatedly to strip every odd prime $q \mid N_E$ (legal by Step 4's unramifiedness). One descends to a newform of level $2$ .

Step 7 (Contradiction). The space of weight-2 cusp forms of level $2$ is zero-dimensional: $S_2(\Gamma_0(2)) = 0$ . There is no newform for $\bar\rho_{E,p}$ to come from.

7

\quad \dim_\mathbb{C} S_2(\Gamma_0(2)) = g\big(X_0(2)\big) = 0.

Derivation hint: The modular curve $X_0(2)$ has genus $0$ ; cusp forms of weight 2 correspond to holomorphic differentials, of which there are $g$ . Thus no Frey curve exists, so no Fermat solution exists. $\blacksquare$

∎

DefinitionGalois deformation ring

R

Fix a residual representation $\bar\rho:G_\mathbb{Q}\to\mathrm{GL}_2(\mathbb{F}_p)$ that is absolutely irreducible, modular, and satisfies a chosen set $\mathcal{D}$ of local conditions (ramification type, determinant, flatness). The functor sending a complete local Noetherian $\mathbb{Z}_p$ -algebra $A$ to the set of deformations of $\bar\rho$ to $A$ of type $\mathcal{D}$ is representable by a universal deformation ring $R_\mathcal{D}$ with universal lift

8

\quad \rho^{\mathrm{univ}}: G_\mathbb{Q} \to \mathrm{GL}_2(R_\mathcal{D}), \qquad \rho^{\mathrm{univ}} \bmod \mathfrak{m}_{R} = \bar\rho.

Derivation hint: Mazur's deformation theory: Schlessinger's criteria are satisfied because $\mathrm{End}_{G}(\bar\rho)=\mathbb{F}_p$ (absolute irreducibility kills nontrivial endomorphisms).

DefinitionHecke algebra

T

Let $\mathbb{T}_\mathcal{D}$ be the $\mathbb{Z}_p$ -subalgebra of $\mathrm{End}$ of the relevant space of cusp forms generated by Hecke operators $T_\ell$ ( $\ell\nmid N$ ) and $U_\ell$ ( $\ell\mid N$ ), completed and localized at the maximal ideal $\mathfrak{m}$ attached to $\bar\rho$ . Eichler–Shimura attaches to $\mathbb{T}_\mathcal{D}$ a Galois representation, hence a surjection of $\mathbb{Z}_p$ -algebras

9

\quad \varphi: R_\mathcal{D} \twoheadrightarrow \mathbb{T}_\mathcal{D}.

Derivation hint: The modular Galois representation valued in $\mathbb{T}$ is itself a deformation of type $\mathcal{D}$ , so the universal property yields the surjection $\varphi$ .

ProofThe

R = T

theorem (numerical criterion)

Goal. Show $\varphi$ in [9] is an isomorphism; then every type- $\mathcal{D}$ deformation, in particular the one coming from a semistable elliptic curve, is modular.

Step 1 (Reduce to a numerical inequality). Wiles' commutative-algebra criterion: $\varphi:R\to T$ is an isomorphism of complete intersections iff

10

\quad \#\,\big(\Phi_R\big) \;=\; \#\,\big(\mathcal{O}/\eta_T\big) \;<\;\infty,

where $\Phi_R = \mathfrak{p}/\mathfrak{p}^2$ is the cotangent space of $R$ at the chosen prime and $\eta_T$ is the congruence ideal of $T$ . Derivation hint: For a surjection of finite flat $\mathcal{O}$ -algebras with $T$ a complete intersection, $\#\Phi_R \ge \#(\mathcal{O}/\eta_T)$ always; equality forces $R\cong T$ a complete intersection (Wiles–Lenstra).

Step 2 (Bound the Selmer side). The cotangent space $\Phi_R$ is identified with a Selmer group $H^1_\mathcal{D}(G_\mathbb{Q}, \mathrm{ad}^0\bar\rho)$ of the adjoint representation:

11

\quad \Phi_R \;\cong\; H^1_{\mathcal{D}}\big(\mathbb{Q}, \mathrm{ad}^0\,\rho^{\mathrm{univ}}\big).

Derivation hint: First-order deformations are cocycles valued in $\mathrm{ad}\,\bar\rho$ ; trace-zero part $\mathrm{ad}^0$ fixes the determinant condition; local conditions $\mathcal{D}$ cut out the Selmer subgroup.

Step 3 (Bound the Hecke side). The congruence module $\mathcal{O}/\eta_T$ is computed by the special value of an $L$ -function / a ratio of Petersson norms (the Flach–Hida formula), giving an exact size.

Step 4 (Patching — Taylor–Wiles). When the minimal case is hard to verify directly, introduce auxiliary sets $Q_n$ of Taylor–Wiles primes (primes $q\equiv 1\bmod p^n$ with $\bar\rho(\mathrm{Frob}_q)$ having distinct eigenvalues). One builds a tower of deformation/Hecke rings $R_{Q_n}, T_{Q_n}$ , then patches them into power-series rings to deduce that $R$ is a complete intersection and free over $\mathcal{O}$ , forcing equality in [10].

12

\quad R_\infty \cong \mathcal{O}[[x_1,\dots,x_r]] \;\twoheadrightarrow\; T_\infty, \qquad \dim R_\infty = \dim T_\infty .

Derivation hint: Each Taylor–Wiles prime adds one Selmer/tangent dimension and one Hecke generator; a Galois-cohomology count (global Euler characteristic) keeps the two sides in lockstep; the patched ring is regular, so $R=T$ .

Step 5 (Propagate). The minimal case gives $R=T$ for the minimal deformation problem; an induction on the level / Ihara's lemma (Wiles' "raising the level" and the numerical isomorphism criterion) propagates modularity to all semistable types, completing the Modularity Theorem [5]. $\blacksquare$

∎

⚠️ The 1993 gap

Wiles' first announced proof (June 1993) contained a gap in the bound of Step 3/Step 4: the Euler-system/Flach–Kolyvagin method failed to give the required upper bound on the Selmer group in full generality. The fix (September 1994), found with Richard Taylor, replaced the flawed argument by the Taylor–Wiles patching of [12], avoiding the Euler system entirely. The published result appeared as two 1995 Annals of Mathematics papers (Wiles; Taylor–Wiles).

mathematician · Andrew Wiles | b. 1953

Wiles encountered FLT as a ten-year-old in a Cambridge library and resolved to prove it. After foundational work on Iwasawa theory and the Birch–Swinnerton-Dyer conjecture (with Coates) and the proof of the main conjecture of Iwasawa theory (with Mazur), he worked in near-total secrecy from 1986 (after Ribet's theorem made the strategy viable) until 1993. His decisive innovations were the systematic use of deformation theory of Galois representations (building on Mazur), the $R=T$ numerical criterion (with Lenstra's later simplification), and — with Taylor — the patching argument. Awarded the Abel Prize (2016); the Fields Medal age limit precluded a 1998 medal, so he received a special IMU silver plaque.

🧠 Intuition

Think of three parallel worlds for a 2-dimensional Galois representation $\bar\rho$ :

Galois world ( $R$ ): all lifts of $\bar\rho$ satisfying prescribed local symmetry rules — a moduli space of arithmetic symmetries.
Automorphic world ( $T$ ): only those symmetries that come from modular forms — eigenvalues of Hecke operators.
Geometry world ( $E$ ): the symmetry attached to an actual elliptic curve.

Wiles proves the Galois world equals the automorphic world ( $R=T$ ): there is nothing exotic — every allowed Galois symmetry is secretly modular. The Frey curve lives in the geometry world, hence (via $R=T$ ) in the automorphic world; but Ribet shows it would have to descend to level 2, where no modular form lives. The "size-matching" in [10] is the engine: count Galois deformations (Selmer group) and count modular congruences ( $L$ -value); their equality is a tangent-space dimension count that cannot leave room for a non-modular curve.

Figure

Logical architecture of Wiles' proof

```svg

```

ExampleComputing the Frey curve's invariants for a toy "solution"

Given. Pretend (illegally) $a=3,\ b=4,\ c$ with $a^p+b^p=c^p$ , $p=5$ . We just exercise the formulas of [3]–[4] with $a^5=243$ , $b^5=1024$ .

Steps. 1. Roots of the cubic: $0,\ a^5=243,\ -b^5=-1024$ . 2. Curve: $y^2 = x(x-243)(x+1024)$ . 3. Naive discriminant $\Delta_0 = 16\,(a^5 b^5)^2 (a^5+b^5)^2 = 16\,(243\cdot1024)^2(1267)^2$ , i.e. $16(abc)^{2p}$ form with $c^5=1267$ . 4. Strip the factor $2^8$ from the minimal model: $\Delta = 2^{-8}(abc)^{2p}$ . 5. Conductor (radical of $abc$ ): $abc = 3\cdot 4 \cdot c$ ; squarefree part collects the distinct primes dividing $abc$ .

Boxed answer.

\boxed{\;E: y^2=x(x-243)(x+1024),\quad \Delta=2^{-8}(abc)^{10},\quad N=\mathrm{rad}(abc)\;}

(No real Fermat solution exists; this only illustrates [3]–[4]. Units: all quantities dimensionless integers.)

ExampleWhy level 2 is empty — genus of

X_0(2)

Given. Need $\dim S_2(\Gamma_0(2)) = g(X_0(2))$ .

Steps (genus formula for $X_0(N)$ ). For $X_0(N)$ ,

g = 1 + \frac{\mu}{12} - \frac{\nu_2}{4} - \frac{\nu_3}{3} - \frac{\nu_\infty}{2},

with index $\mu=[\mathrm{SL}_2(\mathbb{Z}):\Gamma_0(N)] = N\prod_{p\mid N}(1+\tfrac1p)$ . 1. $N=2$ : $\mu = 2(1+\tfrac12)=3$ . 2. Elliptic points of order 2: $\nu_2 = 1$ (since $\big(\tfrac{-1}{2}\big)$ rule gives one). 3. Elliptic points of order 3: $\nu_3 = 0$ . 4. Cusps: $\nu_\infty = 2$ (the cusps $0$ and $\infty$ ). 5. Plug in: $g = 1 + \tfrac{3}{12} - \tfrac14 - 0 - \tfrac22 = 1 + 0.25 - 0.25 - 1 = 0$ .

Boxed answer.

\boxed{\,g(X_0(2)) = 0 \;\Rightarrow\; \dim S_2(\Gamma_0(2)) = 0\,}

Hence no weight-2 newform of level 2 exists — the contradiction in [7].

⚠️ What Wiles did and did not prove

Wiles did not prove the full Taniyama–Shimura–Weil conjecture; he proved it for semistable elliptic curves — exactly the class the Frey curve falls into. The full modularity theorem (all elliptic curves over $\mathbb{Q}$ ) was completed in 2001 by Breuil, Conrad, Diamond, and Taylor, extending the Taylor–Wiles method to additive-reduction primes. Also, Wiles did not "find Fermat's marginal proof": the genuine proof uses 20th-century arithmetic geometry inconceivable in the 17th century.

CounterexampleWhere the strategy genuinely fails without semistability

The reduction depends on the Frey curve being semistable so that $N_E$ is squarefree and Ribet's level lowering can strip all odd primes down to level 2. If one tried the analogue for $p=3$ , the residual representation $\bar\rho_{E,3}$ may be reducible (Mazur's irreducibility requires $p\ge 5$ in this setting), so the deformation theory's hypothesis $\mathrm{End}_G(\bar\rho)=\mathbb{F}_p$ collapses and $R$ need not be representable. This is why FLT is reduced to primes $p\ge 5$ and the cases $n=3,4$ are handled separately by classical descent (Euler, Fermat).

connection

Wiles' strategy sits at the crossroads of several theories. The bridge from Fermat to elliptic curves is the Frey Curve and the Modularity Theorem; the descent step is Ribet's Level-Lowering Theorem, itself depending on Serre's Modularity Conjecture (the epsilon conjecture). The proof engine is Galois Deformation Theory (Mazur) compared against Hecke Algebras via Eichler–Shimura Theory; the size matching uses Selmer Groups and special $L$ -values in the style of Iwasawa Theory. The whole edifice is a celebrated instance of the Langlands Program, relating $\mathrm{GL}_2$ automorphic forms to 2-dimensional Galois representations. The Taylor–Wiles patching has since become the standard tool for Modularity Lifting Theorems.

🔬 Where this appears

⏳ History

- 1637 — Pierre de Fermat writes the marginal note in his copy of Diophantus, claiming a "truly marvelous proof" of [1].

- 1770 — Euler proves the case

n=3

(with a gap later filled), using descent in

\mathbb{Z}[\sqrt{-3}]

.

- 1825 — Dirichlet and Legendre settle

n=5

; Lamé later treats

n=7

.

- 1847 — Kummer proves FLT for all regular primes via ideal numbers, founding algebraic number theory.

- 1955–58 — Taniyama and Shimura formulate the modularity conjecture; Weil adds precision (the level = conductor).

- 1972/1975 — Hellegouarch and (later, sharply) Frey associate an elliptic curve to a hypothetical Fermat solution.

- 1985–86 — Serre formulates the epsilon conjecture; Ribet proves level lowering, making "modularity

\Rightarrow

FLT" rigorous.

- 1986–93 — Wiles works in secret, develops

R=T

via deformation theory; announces the proof at Cambridge, June 1993.

- 1994–95 — Gap fixed with Taylor (Taylor–Wiles patching); the two Annals papers appear; FLT is finally a theorem.

- 2001 — Breuil–Conrad–Diamond–Taylor prove the full Modularity Theorem for all elliptic curves over

\mathbb{Q}

.

open-problem

Effective / explicit forms. While FLT is proved, the modular method does not by itself bound solution sizes for generalized Fermat equations uniformly; many remain open (e.g. the full Fermat–Catalan and Beal conjectures). Modularity over general number fields is still incomplete: modularity of all elliptic curves over an arbitrary number field is not known (great progress over totally real and some imaginary quadratic fields). The structure of $R$ in non-minimal / non-semistable cases — local deformation rings can be singular (Kisin's resolution via local models is delicate). Finally, a conceptual direct proof of FLT avoiding modularity, or one of low enough complexity to fit "a margin," is widely believed not to exist, but this is a meta-mathematical intuition, not a theorem.

Fermat's Last Theorem — Wiles' Strategy

Formal Statement

Why It Holds — The Reduction

The Heart: Wiles' Proof of Modularity

Building Intuition

Worked Examples

Common Misconceptions

Connections

Real-World Applications

Historical Development

Subtleties & Limits

See Also