The Jacobson radical: measuring the failure of semisimplicity

Three faces of one ideal

The Jacobson radical J(R) is the two-sided ideal that every introductory account defines three different ways — and the magic is that they agree. (1) The intersection of all maximal left ideals. (2) The annihilator of all simple left modules: J(R) = ∩ Ann(S). (3) The set of x such that 1 − rxs is invertible for all r, s ∈ R. Definition (2) makes it manifestly two-sided even though (1) only mentions left ideals — a small noncommutative miracle.

Radical zero means semisimple

Here is why the radical matters. For an Artinian ring R (descending chain condition on left ideals — true for any finite-dimensional algebra), J(R) is nilpotent: J(R)^m = 0 for some m. And R is semisimple if and only if J(R) = 0. So the radical is precisely the obstruction to semisimplicity. Quotient it out and R/J(R) is semisimple, hence falls under Artin-Wedderburn.

The strategy for any Artinian ring is therefore two-step: study the semisimple quotient R/J(R) via Wedderburn, then study the nilpotent piece J(R) as a sort of infinitesimal thickening. Many hard structure questions reduce to understanding how J(R) sits on top of a clean semisimple bottom.

A worked radical

Let R be upper-triangular 2x2 matrices over a field k:
     R = { [a, b; 0, d] : a, b, d in k }.

Maximal left ideals correspond to the two simple modules
   S1 = k via  [a,b;0,d] -> a     (top-left action)
   S2 = k via  [a,b;0,d] -> d     (bottom-right action)

Ann(S1) = { [0,b;0,d] },   Ann(S2) = { [a,b;0,0] }.
J(R) = Ann(S1) ∩ Ann(S2) = { [0, b; 0, 0] } = k*E_12.

Check nilpotent:  [0,b;0,0]^2 = [0,0;0,0].  So J(R)^2 = 0.  Good.

Quotient:  R / J(R)  ≅  k × k  (just the diagonal a, d).
That is semisimple — two copies of M_1(k) — exactly as promised.

Computing J(R) for upper-triangular 2×2 matrices: the strictly-upper part is the radical, and R/J is k × k.

This little ring is the perfect cautionary tale: it is finite-dimensional, perfectly concrete, and not semisimple. The single off-diagonal slot k·E_12 is the entire obstruction. Notice too that R has only the trivial idempotents needed to split the diagonal but cannot split off the radical — the extension 0 → S_1 → (the natural module) → S_2 → 0 does not split. That non-splitting is the radical made visible.