MathLabs

Worked solution: Wiles's modularity proof via the Taylor–Wiles method (1994)

Step 8 of 11: The 33–55 switch: bypassing reducible mod-33 representations
In plain words

What if an elliptic curve EE has a reducible mod-33 representation ρ‾E,3\overline{\rho}_{E,3}, so Langlands–Tunnell cannot get the induction started at p=3p = 3? Both ρ‾E,3\overline{\rho}_{E,3} and ρ‾E,5\overline{\rho}_{E,5} cannot be badly behaved at the same time for a semistable curve, so when p=3p = 3 fails, the mod-55 representation ρ‾E,5\overline{\rho}_{E,5} is irreducible.

In May 1993, Wiles found a clever relay trick (the 33–55 switch): construct a second semistable elliptic curve E′/QE'/\mathbb{Q} that shares with EE the exact same 55-torsion shadow ρ‾E′,5≅ρ‾E,5\overline{\rho}_{E',5} \cong \overline{\rho}_{E,5}, while its 33-torsion shadow ρ‾E′,3\overline{\rho}_{E',3} is irreducible. Then E′E' is modular via p=3p = 3, which makes ρ‾E′,5≅ρ‾E,5\overline{\rho}_{E',5} \cong \overline{\rho}_{E,5} modular, and now Wiles can run his lifting machine at p=5p = 5 to prove EE itself is modular!

ρ‾E,3 reducible   ⟹  ∃ E′/Q with ρ‾E′,5≅ρ‾E,5 and ρ‾E′,3 irreducible (hence modular)\overline{\rho}_{E,3} \text{ reducible } \implies \exists\, E'/\mathbb{Q} \text{ with } \overline{\rho}_{E',5} \cong \overline{\rho}_{E,5} \text{ and } \overline{\rho}_{E',3} \text{ irreducible (hence modular)}
Detailed analysis

As Wiles explains in the introduction and Chapter 5 of his paper (Wiles 1995, p. 444, p. 448, p. 452–453, and Chapter 5), a semistable elliptic curve E/QE/\mathbb{Q} automatically satisfies the local hypotheses of his lifting theorem at p=3p = 3, except possibly when ρ‾E,3\overline{\rho}_{E,3} is reducible (or reducible over Q(−3)\mathbb{Q}(\sqrt{-3})). If both ρ‾E,3\overline{\rho}_{E,3} and ρ‾E,5\overline{\rho}_{E,5} were reducible, EE would yield a rational point on the modular curve X0(15)X_0(15) (or a closely related curve), and the known classification of rational points on X0(15)X_0(15) shows that every such semistable curve is already modular.

When ρ‾E,3\overline{\rho}_{E,3} is reducible and ρ‾E,5\overline{\rho}_{E,5} is irreducible, Wiles considers the twisted modular curve X(ρ‾E,5,ρ‾E,3)X(\overline{\rho}_{E,5}, \overline{\rho}_{E,3}) parameterizing elliptic curves E′E' whose 55-torsion agrees with E[5]E[5] and whose 33-torsion has a prescribed irreducible image. Because this moduli curve has genus 00 with rational points, Hilbert's irreducibility theorem produces a semistable elliptic curve E′/QE'/\mathbb{Q} such that ρ‾E′,5≅ρ‾E,5\overline{\rho}_{E',5} \cong \overline{\rho}_{E,5} while ρ‾E′,3\overline{\rho}_{E',3} is irreducible (Wiles 1995, p. 448 and Chapter 5).

Applying the p=3p = 3 modularity lifting theorem (Theorem 0.3) to E′E' proves that E′E' is modular. Since E′E' is modular, its mod-55 representation ρ‾E′,5\overline{\rho}_{E',5} comes from a modular form—and because ρ‾E,5≅ρ‾E′,5\overline{\rho}_{E,5} \cong \overline{\rho}_{E',5}, the mod-55 representation of EE is now known to be modular! Wiles can then apply his modularity lifting theorem a second time, now with p=5p = 5, to conclude that EE is modular (Wiles 1995, Theorem 0.4, p. 448).

Terms in this step
Irreducible vs. reducible representation
A two-dimensional Galois representation is reducible if there is a one-dimensional line in Fp2\mathbb{F}_p^2 fixed by every Galois symmetry (so all matrices become upper-triangular in a chosen basis), and irreducible if no such invariant line exists.
The 33–55 switch
Wiles's May 1993 technique (Chapter 5 of Wiles 1995) that replaces a semistable curve EE whose mod-33 representation is reducible by a companion semistable curve E′E' sharing the same mod-55 representation ρ‾E′,5≅ρ‾E,5\overline{\rho}_{E',5} \cong \overline{\rho}_{E,5} but having an irreducible mod-33 representation.
Knowledge used in this step