Skip to content

The 3-5 Switch and the Conclusion

Summary

The 3-5 switch is the elegant final building block in Wiles' proof. The Langlands–Tunnell theorem provides residual modularity for \(p = 3\). When the 3-representation is reducible, Wiles switches to \(p = 5\) and constructs an auxiliary curve that provides the entry point. In this way, every semistable elliptic curve is recognised as modular – and Fermat's Last Theorem is proved.

Prerequisites


1. The Problem with \(p = 3\)

Recall: The proof strategy

Wiles' proof of the modularity of semistable curves has two stages:

  1. Residual modularity: Show that \(\bar{\rho}_{E,p}\) is modular.
  2. Lifting: Prove \(R = T\) to pass from residual to full modularity.

Stage 2 is done (Taylor–Wiles trick). Stage 1 remains: where does residual modularity come from?

Langlands–Tunnell

For \(p = 3\) there is a powerful result:

Theorem (Langlands–Tunnell)

Let \(\bar{\rho}: G_{\mathbb{Q}} \to \text{GL}_2(\mathbb{F}_3)\) be a continuous, odd representation. Then \(\bar{\rho}\) is modular – there exists a modular form \(f\) of weight 1 with \(\bar{\rho} \cong \bar{\rho}_f\).

The proof uses a special property of \(\text{GL}_2(\mathbb{F}_3)\): this group is solvable (it has order 48 and is isomorphic to an extension of \(S_3\) by \(\mathbb{Z}/2\mathbb{Z}\)). For solvable groups, the tools of the Langlands programme are available – in particular base change for \(\text{GL}_2\).

Langlands proved the modular forms correspondence for solvable Galois representations in 1980, and Tunnell refined the result in 1981.

From weight 1 to weight 2

Langlands–Tunnell provides a modular form of weight 1, but we need weight 2 (for the connection with elliptic curves). Wiles solves this via a lifting argument:

From the weight-1 form \(f\), one constructs a weight-2 form \(g\) with \(\bar{\rho}_g \cong \bar{\rho}_{E,3}\). This uses Hida's theory of ordinary \(p\)-adic modular forms.


2. Why \(p = 3\) Is Not Enough

The irreducibility condition

Wiles' proof of \(R = T\) (and the Taylor–Wiles patching) requires that \(\bar{\rho}_{E,p}\) be irreducible. For the Frey curve, this is automatically satisfied for \(p \geq 5\) (by Mazur). But for a general semistable curve \(E\), \(\bar{\rho}_{E,3}\) can be reducible.

When is \(\bar{\rho}_{E,3}\) reducible?

The representation \(\bar{\rho}_{E,3}\) is reducible if and only if \(E\) has a rational 3-isogeny kernel – a \(G_{\mathbb{Q}}\)-stable subspace of \(E[3]\) of dimension 1 over \(\mathbb{F}_3\). Geometrically, this means: there is an isogeny \(E \to E'\) of degree 3 with rational kernel.

This does occur – there are infinitely many such curves. For these curves, the proof with \(p = 3\) fails.

The situation

\(\bar{\rho}_{E,3}\) Modular? Irreducible? \(R = T\) possible?
irreducible Yes (Langlands–Tunnell) Yes ✓ Proof works
reducible Yes (Langlands–Tunnell) No ✗ Irreducibility missing

3. The Switch to \(p = 5\)

The idea

When \(\bar{\rho}_{E,3}\) is reducible, consider \(\bar{\rho}_{E,5}\) instead. The 5-representation has a good chance of being irreducible – because a curve having both a 3-isogeny kernel and a 5-isogeny kernel would be very special.

Irreducibility of \(\bar{\rho}_{E,5}\)

Wiles shows: if \(E\) is semistable and \(\bar{\rho}_{E,3}\) is reducible, then \(\bar{\rho}_{E,5}\) is irreducible. This follows from Mazur's classification of possible torsion structures of rational elliptic curves:

Theorem (Mazur, 1978)

Let \(E/\mathbb{Q}\) be an elliptic curve. If \(E\) has a rational isogeny of degree \(\ell\) (for a prime \(\ell\)), then \(\ell \leq 19\) or \(\ell \in \{37, 43, 67, 163\}\).

For a semistable curve, the possibilities can be further restricted. In particular, there is no semistable curve with a simultaneous rational 3- and 5-isogeny kernel (this would imply a rational 15-isogeny, which Mazur excludes).

The new problem

Now we have \(\bar{\rho}_{E,5}\) irreducible – but where does residual modularity come from? Langlands–Tunnell works only for \(p = 3\), not for \(p = 5\) (because \(\text{GL}_2(\mathbb{F}_5)\) is not solvable).

Here comes the ingenious trick.


4. Constructing an Auxiliary Curve

The strategy

Wiles seeks a second elliptic curve \(E'\) with the following properties:

  1. \(E'[5] \cong E[5]\) as a Galois module (the 5-torsion agrees)
  2. \(\bar{\rho}_{E',3}\) is irreducible (so that \(p = 3\) works for \(E'\))
  3. \(E'\) is semistable

Why such a curve exists

The set of elliptic curves \(E'\) with \(E'[5] \cong E[5]\) is parametrised by a modular curve \(X(5)\) – a curve of genus 0 (hence rational!). There are therefore infinitely many such curves \(E'\).

Among these infinitely many candidates, Wiles needs to find one that: - is semistable (at all primes), and - has \(\bar{\rho}_{E',3}\) irreducible.

Since the condition "\(\bar{\rho}_{E',3}\) reducible" excludes a proper subset and the parametrisation runs over a rational curve, such a curve \(E'\) exists.

The construction

Concretely: over the function field \(\mathbb{Q}(t)\) there is a "universal" curve with the correct 5-torsion. By specialising at suitable rational values \(t = t_0\), one obtains the desired auxiliary curve \(E'\).


5. Closing the Chain

Step 1: \(E'\) is modular (via \(p = 3\))

Since \(\bar{\rho}_{E',3}\) is irreducible, we can apply the entire proof apparatus for \(p = 3\):

\[ \bar{\rho}_{E',3} \text{ irreducible} \xrightarrow{\text{Langlands–Tunnell}} \bar{\rho}_{E',3} \text{ modular} \xrightarrow{R = T} \rho_{E',3} \text{ modular} \implies E' \text{ modular.} \]

Step 2: \(E[5]\) is modular

Since \(E'\) is modular, in particular: \(\bar{\rho}_{E',5}\) is modular (the 5-residual representation comes from a modular form). Since \(E'[5] \cong E[5]\), it follows:

\[ \bar{\rho}_{E,5} \cong \bar{\rho}_{E',5} \quad \text{is modular.} \]

Step 3: \(E\) is modular (via \(p = 5\))

Now we have the entry point for \(p = 5\): \(\bar{\rho}_{E,5}\) is modular and irreducible. The Taylor–Wiles trick gives \(R = T\) for \(p = 5\):

\[ \bar{\rho}_{E,5} \text{ modular + irreducible} \xrightarrow{R = T} \rho_{E,5} \text{ modular} \implies E \text{ modular.} \]

The complete chain

\[ \boxed{E'[5] \cong E[5]} \quad + \quad \boxed{E' \text{ modular (via } p=3\text{)}} \implies \boxed{\bar{\rho}_{E,5} \text{ modular}} \xrightarrow{R=T} \boxed{E \text{ modular}} \]

6. FLT Is Proved

The case distinction

For every semistable elliptic curve \(E/\mathbb{Q}\):

Case 1: \(\bar{\rho}_{E,3}\) irreducible. Langlands–Tunnell + Taylor–Wiles (\(p = 3\)) → \(E\) modular. ✓

Case 2: \(\bar{\rho}_{E,3}\) reducible. Then \(\bar{\rho}_{E,5}\) is irreducible. Construct auxiliary curve \(E'\) with \(E'[5] \cong E[5]\) and \(\bar{\rho}_{E',3}\) irreducible. Then: \(E'\) modular (Case 1) → \(\bar{\rho}_{E,5}\) modular → Taylor–Wiles (\(p = 5\)) → \(E\) modular. ✓

In both cases, \(E\) is modular. Since the Frey curve is semistable, Fermat's Last Theorem follows.

The complete proof diagram

\[ \text{FLT solution} \xrightarrow{\text{Frey}} E_{\text{Frey}} \xrightarrow{\text{Wiles (3 or 3-5)}} \text{modular} \xrightarrow{\text{Ribet}} \text{contradiction} \]

Theorem (Wiles, Taylor–Wiles, 1995)

Every semistable elliptic curve over \(\mathbb{Q}\) is modular.

Corollary (Fermat's Last Theorem)

The equation \(x^n + y^n = z^n\) has no solution in positive integers for \(n \geq 3\).


7. Retrospective: The Complete Proof Structure

From Fermat to Wiles

The complete proof connects ideas from over four centuries of mathematics:

Year Mathematician Contribution
1637 Fermat The conjecture
1640 Fermat Proof for \(n = 4\) (infinite descent)
1770 Euler Proof for \(n = 3\)
1955 Taniyama, Shimura TSC: elliptic curves ↔ modular forms
1980 Langlands, Tunnell Modularity of solvable representations
1985 Frey FLT solution → "impossible" elliptic curve
1986 Ribet Level-lowering: TSC ⟹ FLT
1989 Mazur Universal deformation rings
1995 Wiles \(R = T\) for semistable curves
1995 Taylor–Wiles Patching argument (closing the gap)

The logical structure

FLT solution (hypothetical)
  ↓  [Frey 1985]
Frey curve E (semistable, extreme discriminant)
  ↓  [Wiles 1995: 3-5 switch + R=T + Taylor–Wiles]
E is modular
  ↓  [Ribet 1986: level-lowering]
ρ̄(E,p) comes from newform of level 2
  ↓  [S₂(Γ₀(2)) = 0]
Contradiction → FLT solution does not exist   □

What makes the proof so remarkable

  1. Indirectness: FLT is not proved directly but by contradiction – via the detour of elliptic curves, modular forms, and Galois representations.
  2. Unification: The proof unifies number theory, algebraic geometry, complex analysis, and representation theory.
  3. Depth: The methods (deformation theory, patching) are not specific to FLT but tools of general significance.
  4. Humanity: The story of Wiles' seven-year secret project, the gap, and its closure is unique in the history of mathematics.

Outlook

The next and final article looks beyond Wiles' proof:

Article Topic
08 – What Came After BCDT, Serre's conjecture, the Langlands programme

Sources

  • Andrew Wiles: Modular elliptic curves and Fermat's Last Theorem, Annals of Mathematics 141 (1995), §5
  • Robert Langlands: Base change for GL(2), Annals of Mathematics Studies 96 (1980)
  • Jerrold Tunnell: Artin's conjecture for representations of octahedral type, Bulletin of the AMS 5 (1981)
  • Barry Mazur: Rational isogenies of prime degree, Inventiones Mathematicae 44 (1978)
  • Gary Cornell, Joseph Silverman, Glenn Stevens (eds.): Modular Forms and Fermat's Last Theorem, Springer (1997), Chapters XV–XVI