Skip to the Main Content

Note:These pages make extensive use of the latest XHTML and CSS Standards. They ought to look great in any standards-compliant modern browser. Unfortunately, they will probably look horrible in older browsers, like Netscape 4.x and IE 4.x. Moreover, many posts use MathML, which is, currently only supported in Mozilla. My best suggestion (and you will thank me when surfing an ever-increasing number of sites on the web which have been crafted to use the new standards) is to upgrade to the latest version of your browser. If that's not possible, consider moving to the Standards-compliant and open-source Mozilla browser.

June 14, 2023

The Wedderburn–Artin Theorem

Posted by John Baez

Now that I finally understand the proof of the Wedderburn–Artin Theorem, I want to do what any good category theorist would do: formulate it more generally and abstractly, so nobody can understand me and I look like a genius everybody can see how simple it actually is.

The Wedderburn–Artin Theorem says that any semisimple ring RR is a finite product of matrix algebras over division rings. In symbols:

R i=1 nM k i(D i) \displaystyle{ R \cong \prod_{i = 1}^n M_{k_i}(D_i) }

where D 1,,D nD_1, \dots , D_n are division rings.

But what’s a semisimple ring? We can define it to be one whose category of finitely generated modules is semisimple. Left or right modules? It turns out not to matter, though this is not obvious! So, pick one.

But what’s a semisimple category? It’s usually defined as an abelian category where every object is a finite direct sum of simple objects.

But what’s a simple object? It’s an object XX with no quotients other than 00 and XX itself. In an abelian category, this equivalent to saying XX has no subobjects other than 00 and XX. We use both of these facts to see how division rings show up in the Wedderburn–Artin theorem. Here’s how:

Schur’s Lemma. If XX and YY are simple objects in an abelian category, then any morphism f:XYf \colon X \to Y is either an isomorphism or 00. Thus, if XX is a simple object in an abelian category, Hom(X,X)Hom(X,X) is a division ring.

Proof. If XX and YY are simple the kernel of any morphism f:XYf \colon X \to Y must be 00 or XX, and its cokernel must be 00 or YY. If the kernel is XX or the cokernel is YY we must have f=0f = 0. The only alternative is that the kernel is 00 and the cokernel is 00. In an abelian category this implies that ff is an isomorphism. ▮

The proof of Wedderburn–Artin then goes as follows. Suppose RR is a ring whose category Mod RMod_R of finitely generated right modules is semisimple. Let R RMod RR_R \in Mod_R stand for RR viewed as a right module over itself. Then R RR_R is a finite direct sum of nonisomorphic simple objects X 1,,X kX_1, \dots, X_k, where each object X iX_i can show up any number of times in the direct sum, say n in_i times:

R R i=1 kX i n i \displaystyle{ R_R \cong \bigoplus_{i = 1}^k X_i^{n_i} }

Thus by Schur’s Lemma and the way direct sums work,

End(R R) i=1 kM n i(End(X i)) \displaystyle{ End(R_R) \cong \bigoplus_{i = 1}^k M_{n_i}(End(X_i)) }

so End(R R)End(R_R) is a finite product of matrix algebras over division rings!

Finally, we use the ring-theoretic analogue of Cayley’s Theorem: every endomorphism of RR as a right RR-module is given by left multiplication by some unique element of RR. This is easy to prove, and it yields

End(R R)R End(R_R) \cong R

so we get what we want: for any semisimple ring RR we have

R i=1 kM n i(D i) R \cong \bigoplus_{i = 1}^k M_{n_i}(D_i)

for some division rings D i=End(X i)D_i = End(X_i).

Analysis

Most of what we did was show that the endomorphism ring of any object in a semisimple category is a finite product of matrix algebras over division rings. Only at the last minute did we use the fact that a semisimple ring RR actually is such an endomorphism ring, namely End(R R)End(R_R). So arguably the Wedderburn–Artin theorem is mainly about endomorphism rings of objects in semisimple categories. Let’s think about it that way.

Furthermore, while semisimple categories are necessarily abelian, we didn’t use much about abelian categories. We used some facts about kernels and cokernels, but only for proving Schur’s Lemma. And while the hom-sets of an abelian category are required to be abelian groups, we didn’t use subtraction anywhere.

So, let’s strip down the assumptions to see the essence of Wedderburn–Artin. For this I should review a bit of category theory — since I don’t really want nobody to understand me.

Let’s start with a semiadditive category: that is, a category with a zero object and biproducts. A zero object is an object that is both initial and terminal; we write it as 00. Biproducts are what I’d been calling direct sums: the point is that they are both coproducts and products. More precisely, if two objects X,YX,Y in a category with a zero object have both a coproduct X+YX+Y and a product X×YX \times Y, we get a canonical morphism X+YX×YX + Y \to X \times Y. If this is an isomorphism we call either X+YX + Y and X×YX \times Y the biproduct of XX and YY, and write it as XYX \oplus Y.

In a semiadditive category, for any objects XX and YY the hom-set Hom(X,Y)Hom(X,Y) becomes a commutative monoid! In fact any semiadditive category is enriched over commutative monoids: for the proof go here or read this:

As a consequence, for any object XX in a semiadditive category, its set of endomorphisms End(X)End(X) is not just a monoid but a rig. This is like a ring, but possibly without additive inverses. While mathematicians talk much more about division rings much more, the concept of a division rig makes perfect sense: it’s a rig where every nonzero element has a multiplicative inverse. For example, the nonnegative rational numbers, or the nonnegative elements of any ordered field, are a division rig.

We can also deal with matrices. Given a rig RR there’s a new rig M n(R)M_n(R) consisting of n×nn \times n matrices with entries in RR, with matrix addition and multiplication defined in the usual way. Suppose XX is an object in a semiadditive category, and let X nX^n stands for the nn-fold biproduct XXX \oplus \cdots \oplus X. (This is both a coproduct and a product, so the notation X nX^n is not so bad.) Then you can check that

End(X n)M n(X) End(X^n) \cong M_n(X)

This uses almost all the properties of biproducts.

Now that we’ve got division rigs and matrix rigs, we’re ready to state and prove an easy generalization of the Wedderburn–Artin theorem!

Generalization

Suppose A\mathsf{A} is a semiadditive category. Say an object XAX \in \mathsf{A} is simple if every nonzero f:XXf \colon X \to X is an isomorphism. Say an object XAX \in \mathsf{A} is semisimple if it can be written as a biproduct of simple objects X 1,,X kX_1, \dots, X_k with hom(X i,X j){0}hom(X_i,X_j) \cong \{0\}. We allow each object X iX_i to show up multiple times in this biproduct, so

X i=1 kX i n i \displaystyle{ X \cong \bigoplus_{i = 1}^k X_i^{n_i} }

for some positive integers n in_i. These concepts of ‘simple’ and ‘semisimple’ reduce to the usual ones when A\mathsf{A} is abelian.

With these definitions, we get this result:

Abstract Wedderburn–Artin Theorem. If RR is a semisimple object in a semiadditive category A\mathsf{A}, then End(R)End(R) is a finite product of matrix rigs over division rigs.

Proof. Write

R i=1 kX i n i \displaystyle{ R \cong \bigoplus_{i = 1}^k X_i^{n_i} }

for simple objects X 1,,X nX_1, \dots, X_n with hom(X i,X j){0}hom(X_i,X_j) \cong \{0\}, and then use the properties of biproducts to conclude

End(R) i=1 nM n i(End(X i)) \displaystyle{ End(R) \cong \bigoplus_{i = 1}^n M_{n_i}(End(X_i}))

where each End(X i)End(X_i) is a division rig since X iX_i is simple. ▮

When A\mathsf{A} is an abelian category, the division rings and the multiplicities n in_i are determined (up to permutation) by the object XX. In this generalization I’m not claiming that, but I haven’t examined the issue yet.

I believe we can use this generalization to get a Wedderburn–Artin theorem for rigs. For this let A=Mod R\mathsf{A} = Mod_R be the category of right modules of some rig RR: that is, modules where it acts on commutative monoids. Take R RMod RR_R \in Mod_R to be RR as a right module over itself. Then I believe we have an isomorphism of rigs

REnd(R R) R \cong End(R_R)

If so, when R RR_R is a semisimple object in Mod RMod_R we can use this generalization of Wedderburn–Artin to write RR itself as a finite product of matrix rigs over division rigs.

I’m not claiming this generalization is a big deal. But taking apart a car engine and successfully putting it back together is the best way to get to know that engine — or so I’ve heard. And if you can put it back together with fewer pieces, and it still runs, that’s even better. So I felt I had to do this, to really understand Wedderburn–Artin.

Posted at June 14, 2023 4:34 PM UTC

TrackBack URL for this Entry:   https://golem.ph.utexas.edu/cgi-bin/MT-3.0/dxy-tb.fcgi/3472

4 Comments & 0 Trackbacks

Re: The Wedderburn–Artin Theorem

Yay for semiadditive categories.

Posted by: a on June 15, 2023 11:44 AM | Permalink | Reply to this

Re: The Wedderburn–Artin Theorem

This is nice!

I think there are some typos in your statement and proof of Schur’s lemma. The second sentence of the statement (“Thus,…”) follows from the “It suffices to show” in the first sentence of the proof, but not (at least, not obviously) from the first sentence of the statement. And in the proof, it looks to me as though the roles of 00 and YY as cokernels are switched.

Posted by: Mike Shulman on June 16, 2023 6:50 AM | Permalink | Reply to this

Re: The Wedderburn–Artin Theorem

You’re right on all counts. And the fixed version is even shorter!

Schur’s Lemma. If XX and YY are simple objects in an abelian category, then any morphism f:XYf \colon X \to Y is either an isomorphism or 00. Thus, if XX is a simple object in an abelian category, Hom(X,X)Hom(X,X) is a division ring.

Proof. If XX and YY are simple the kernel of any morphism f:XYf \colon X \to Y must be 00 or XX, and its cokernel must be 00 or YY. If the kernel is XX or the cokernel is YY we must have f=0f = 0. The only alternative is that the kernel is 00 and the cokernel is 00. In an abelian category this implies that ff is an isomorphism. ▮

Posted by: John Baez on June 16, 2023 8:26 AM | Permalink | Reply to this

Re: The Wedderburn–Artin Theorem

Hi, I think there is a typo at the end of your “abstract Wedderburn-Artin theorem”. The statement mentions a finite product but you have written a finite sum. For the semiadditive cateogry these are obviously the same, but this is not the case for the category of rings (and probably rngs). So I think there is an extra step missing where you state that End takes sums to products rather than sums to sums. (What you have written is true for abelian groups but not for rigs).

Another thing that might be interesting to note is that there is a definition of “point-free biproduct” on the nlab where a biproduct is a product and coproduct with a pair of split orthogonal idempotents coming from the compositions of the projection and inclusion maps. There are also ring theory proofs of Wedderburn-Artin that use orthogonal idempotents to do a block decomposition. I suspect these two facts might be related.

Posted by: Ali Caglayan on April 15, 2024 6:58 PM | Permalink | Reply to this

Post a New Comment