Say I have a (non-unital) algebra $A$ which decomposes as a direct sum $A = V \oplus W$, where $V$ and $W$ are subalgebras. In an algebra, the multiplication is distributive over addition. Therefore, for two elements $v \in V$ and $w \in W$, we have that $$ (v+ w)(v+w) = v^2 + vw + wv + w^2. $$

On the other hand, since $A = V \oplus W$, we have that the multiplication is component-wise, $$(v+w)(v+w) = v^2 + w^2.$$

Is there a way to see these two are compatible? I.e. that, in this setup, $vw + wv = 0$?

EDIT: Ok, so here's another way to look at my problem. Suppose $A$ is an algebra, so in particular it's a vector space. Let $V$ and $W$ be subspaces such that $A = V \oplus W$ where $\oplus$ is a vector space direct sum. The question is then, if $V$ and $W$ are also subalgebras of $A$, does this then extend to a direct sum of algebras?

  • 1
    $\begingroup$ This can never happen (for $V, W \neq 0$) unless your notion of subalgebra does not include sharing the multiplicative unit. The closest thing that can happen is that $A$ is the direct product $V \times W$ of algebras, in which case $vw = wv = 0$ for all $v \in V, w \in W$. $\endgroup$ – Qiaochu Yuan Jul 27 '16 at 23:13
  • $\begingroup$ A isn't necessarily unital, so neither are the subalgebras. $\endgroup$ – user353840 Jul 27 '16 at 23:17
  • $\begingroup$ I'm not sure what you're asking for. You've just given a proof that if $V$ and $W$ are subalgebras of $A$ such that $V\oplus W\cong A$ as algebras by sending $(v,w)$ to $v+w$, then $vw+wv$ for all $v\in V$ and $w\in W$. What more do you want? $\endgroup$ – Eric Wofsey Jul 27 '16 at 23:19
  • $\begingroup$ In the direct product $R\times S$ of rings $R$ and $S$, we abuse notation by referring to elements $(r,0)\in R\times S$ as just $r$ and elements $(0,s)\in R\times S$ as just $s$. Thus, $rs$ is shorthand for $(r,0)(0,s)$, which by definition equals $(r\cdot0,0\cdot s)=(0,0)$, i.e. the zero element in $R\times S$. (In the case our rings are algebras, they are in particular vector spaces, which may inspire us to use the symbol $\oplus$ instead of $\times$.) $\endgroup$ – arctic tern Jul 27 '16 at 23:25
  • 1
    $\begingroup$ Show it? Um, if you're using $\oplus$ to denote direct sum of algebras, then it's part of what the notation means. It's part of the definition of $V\oplus W$ (interpreted as an internal algebra direct sum) that $v$ times $w$ equals $0$ for all $v\in V,w\in W$. If it weren't the case that all those products were $0$, then we wouldn't be able to claim that our algebra is an algebra direct sum of $V$ and $W$, and thus wouldn't have the right to write the notation $V\oplus W$ (again, interpreted as an internal algebra direct sum) in the first place. $\endgroup$ – arctic tern Jul 28 '16 at 0:11

Let's clear up some things from the comments. For categorical reasons, writing a direct product of rings $R\times S$ as a direct sum $R\oplus S$ might be misleading since $R\times S$ is not a coproduct, although in the case of algebras they are in particular vector spaces which we like writing $\oplus$ for, so in the context of algebras we might use $\oplus$ too. However we even use $\oplus$ for vector space decompositions of algebras (in order to signify grading), such as $\mathbb{C}[x]=\mathbb{C}\oplus\mathbb{C}x\oplus\mathbb{C}x^2\oplus\cdots$. In this answer I will decide to use $\times$ for direct products of rings in order to distinguish it from internal vector space direct sum. Also, let's distinguish internal from external perhaps.

If $R$ and $S$ are any two rings, the external direct product $R\times S$ is the the set of all ordered pairs $(r,s)$ with $r\in R$ and $s\in S$ with "componentwise" multiplication $(r_1,s_1)(r_2,s_2)=(r_1r_2,s_1s_2)$. If we're talking about unital rings $R$ and $S$, then $(1_R,1_S)$ is the unique identity for $R\times S$, and the elements $e_R=(1_R,0_S)$ and $e_S=(0_R,1_S)$ are not identity elements.

Instead they are idempotents (i.e. $e^2=e$), they are central (so $ex=xe$ for all $x\in R\times S$) and they are orthogonal (since $e_Re_S=0_{R\times S}$). I'll leave you to verify my parenthetical statements as an exercise.

There are (non-unital) ring embeddings $R\to R\times S$ and $S\to R\times S$, in which (by abuse of notation) we may write $r$ when we mean $(r,0_S)$ or write $s$ when we mean $(0_R,s)$. In this case, the product $rs$ is short for $(r,0_S)(0_R,s)=(r\cdot0_R,0_S\cdot s)=(0_R,0_S)=0_{R\times S}$. In particular, elements of $R$ and $S$ are identified with zero divisors in $R\times S$. With this abuse of notation, multiplication-by-$e_R$ is just the projection map $R\times S\to R$, and multiplication-by-$e_S$ is the projection map $R\times S\to S$.

Now let's switch gears. Suppose $A$ is any rings, and $R$ and $S$ are subrings for which every $a\in A$ may be uniquely written as $a=r+s$ for some $r\in R$, $s\in S$. If in addition we have $rs=0$ for all $r\in R$ and $s\in S$, then we write $A=R\times S$ and we call $A$ an internal direct product of $R$ and $S$. Note that the external direct product $R\times S$ is an internal direct product of $R\times\{0_S\}$ and $\{0_R\}\times S$ under these definitions.

They're both ways of talking about the same thing, the difference is we use the term "internal" when we're viewing $R$ and $S$ as subrings of an already-existing algebra $A$, and we use the term "external" when we're creating a new algebra out of already-existing rings $R$ and $S$.

If $A$ is unital and $A=R\times S$ is an internal direct product, then $1_A=e_R+e_S$ for some unique elements $e_R\in R$ and $e_S\in S$. As an exercise, check that $e_R$ and $e_S$ are orthogonal, central idempotents. Next, check that $R:=e_RA$ and $S:=e_SA$ are ideals, and in particular they are subrings, and that $A=R\times S$ is an internal direct product. Third, verify that if $e\in R$ is a central idempotent, then $e$ and $1-e$ are orthogonal central idempotents. Finally, prove that if $1_A\in A$ is a primitive idempotent then $A$ is not the internal direct product $A=R\times S$ of any two subrings $R\times S$. (Here let's not assume subrings share the same identity element.)

In conclusion, for unital rings, direct factors correspond to central, orthogonal idempotents. Sorry for the entire chapter, but I thought you might like this theory.

Anyway, if $A$ is an algebra, then it is possible for $A$ to be the internal vector space direct sum $V\oplus W$ of subalgebras $V$ and $W$ without being an internal direct product $V\times W$ of subrings $V$ and $W$. For instance, if we don't care about having identity elements, $A=\mathbb{R}\times\mathbb{R}[\varepsilon]/(\varepsilon^2)$ is a vector space direct sum of subalgebras $V=\mathbb{R}\times\mathbb{R}$ and $W=\{0\}\times\mathbb{R}\varepsilon$ but it is not a direct product of them as subrings, since $(0,1)$ is not orthogonal to $(0,\varepsilon)$. However it is a direct product of subrings $Y=\mathbb{R}$ (the first component) and $X=\mathbb{R}[\varepsilon]/(\varepsilon^2)$. Notice that $\dim V=\dim X=2$ and $\dim W=\dim Y=1$ but $X\not\cong V$ as algebras.

  • $\begingroup$ This is a great answer thank you. Do you by any chance have the book 'Representations of Finite and Compact Groups' by Barry Simon? Because the reason I posted this question is that I'm trying to understand a proof in there. But now that I understand direct sums more, I understand the proof less. $\endgroup$ – user353840 Jul 29 '16 at 9:16
  • $\begingroup$ @user353840 I have access to it. What's up? $\endgroup$ – arctic tern Jul 29 '16 at 9:50
  • $\begingroup$ It's theorem III.7.1 on page 56. In the book it is previously shown that the group algebra is isomorphic as an algebra to a direct sum of matrix algebras. (theorem III.1.5 page 38) Now, in the proof of theorem III.7.1 they construct the group algebra as a direct sum of vectorspaces $M_\alpha$. Then they state that $M_\alpha$ is isomorphic to a matrix algebra, to which I agree, but only as a vector space. But then they use the fact that it's a direct sum of algebras to show that $p^2 =p$ iff $p_\alpha^2 = p_\alpha$. So I'm really confused about the use of direct sums there. $\endgroup$ – user353840 Jul 29 '16 at 10:31
  • $\begingroup$ @user353840 The isomorphism $$\Bbb C[G]\cong \bigoplus_{{\rm Irr}(G)}{\rm End}(V)$$ (as $\Bbb C$-algebras and as $\Bbb C[G]$-bimodules in fact not just as vector spaces) is called the Artin-Wedderburn theorem or Peter-Weyl theorem. The slickest proof of this decomposition for finite groups that I know, which also explicitly solves for the idempotents/projections associated to irreps and derives the Schur orthogonality relations as corollary, I wrote another another account here. $\endgroup$ – arctic tern Jul 29 '16 at 16:43

If $V$ and $W$ are subalgebras and $A=V\oplus W$ as a vector space, it is not necessarily true that $A=V\oplus W$ as an algebra. For instance, consider a commutative algebra $A$ with basis $\{x,y,z\}$ with $xy=y^2=z$ and $x^2=xz=yz=z^2=0$. Take $V$ to be the span of $\{x\}$ and $W$ to be the span of $\{y,z\}$. Then $V$ and $W$ are subalgebras and $A=V\oplus W$ as a vector space. But for $v=x$ and $w=y$, $vw+wv=2z\neq 0$ (assuming our field does not have characteristic $2$).


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.