5

I "know" that $\mathbb{C} \otimes_\mathbb{R} \mathbb{C} \cong \mathbb{C} \oplus \mathbb{C}$ as rings, but I don't really know it, what I mean with this is that I don't know any explicit isomorphism $f: \mathbb{C} \otimes_\mathbb{R} \mathbb{C} \rightarrow \mathbb{C} \oplus \mathbb{C}$. I suspect that such an isomorphism should be easy to find, but I am really not finding anyone. Could anyone please help me?

Tedar
  • 559
  • Is it bilinear? – Tedar Sep 19 '13 at 22:17
  • 1
    At this point you should make precise what kind of isomorphism you want ($\mathbb C$-vector space? $\mathbb R$-vector space? Rings? $\mathbb C$-algebra? $\mathbb R$-algebra?). You have many options. The first question to ask yourself is "what do you actually "know"?" i.e. what type of isomorphism between the two do you have? – Patrick Da Silva Sep 19 '13 at 22:26
  • Isomorphism of rings, sorry – Tedar Sep 19 '13 at 22:26
  • 1
    @Tedar : Don't be sorry, I was not criticizing, I was giving you a tip to help you answer yourself. – Patrick Da Silva Sep 19 '13 at 22:27
  • For instance, the isomorphism of $\mathbb R$-vector spaces is clear because both have dimension $4$ over $\mathbb R$ by the properties of tensor product and direct sum. It is not clear though that an isomorphism of $\mathbb R$-vector spaces will preserve multiplication in both rings. – Patrick Da Silva Sep 19 '13 at 22:28
  • @YACP The incompatibility between linearity and bilinearity means that map is not well-defined. For example $$(1+i)\otimes(1+i)=1\otimes1+1\otimes i+i\otimes 1+i\otimes i$$ but $$(1+i)\otimes(1+i)\mapsto(1+i,1+i)$$ while $$1\otimes1+1\otimes i+i\otimes1+i\otimes i\mapsto(1,1)+(1,i)+(i,1)+(1,1)=(2+2i,2+2i)$$ yet $(1+i,1+i)\ne(2+2i,2+2i)$. – anon Sep 19 '13 at 23:27
  • Related: https://math.stackexchange.com/questions/263192 – Watson Nov 26 '18 at 09:28

3 Answers3

6

You don't mean the direct sum of rings (or algebras), you mean the direct product of rings. See also here.

In general, if $K/k$ is a field extension and $f \in k[x]$ is a polynomial which splits over $K$ into $n$ distinct linear factors $x-\alpha_i$, then there is an isomorphism of $K$-algebras

$$k[x]/(f) \otimes_k K \cong K[x]/(f) \cong \prod_i K[x]/(x-\alpha_i) \cong \prod_i K = K^n.$$ It is given by mapping $x \otimes 1$ to $(\alpha_1,\dotsc,\alpha_n)$, the rest is given by the information that it is an $K$-algebra isomorphism. Explicitly, we have $p \otimes \lambda \mapsto (\lambda p(\alpha_1),\dotsc,\lambda p(\alpha_n))$ for $\lambda \in K$ and $p \in k[x]$.

The inverse can be found just by looking at the proof of the Chinese Remainder Theorem which we have used above: Since $\alpha_1,\dotsc,\alpha_n$ are pairwise distinct, there are $p_j \in K[x]$ such that $p_j(\alpha_i)=\delta_{ij}$ (e.g. Lagrange polynomial). Then $\overline{p_j} \in K[x]/(f)$ gets mapped to the unit vector $e_j \in K^n$. Hence, the inverse map $K^n \to K[x]/(f)$ is given by $(\lambda_1,\dotsc,\lambda_n) \mapsto \sum_j \lambda_j p_j$.

Example: If $f = x^2+1 \in k[x]$ is irreducible and $\mathrm{char}(k) \neq 2$, we get $k(i) \otimes_k k(i) \cong k(i) \times k(i)$ as $k(i)$-algebras (where $k(i)$ acts on the right tensor factor) given by $i \otimes 1 \mapsto (i,-i)$ (and hence $1 \otimes i \mapsto (i,i)$ and $i \otimes i \mapsto (-1,1)$). We compute $p_1=\dfrac{x+i}{2i}$ and $p_2 = \overline{p_1} = \dfrac{i-x}{2i}$ in $k(i)[x]$. The images in $k(i)[x]/(x^2+1) \cong k(i) \otimes_k k(i)$ are $p_1 = 1 \otimes \frac{i}{2i} + i \otimes \frac{1}{2i} = \frac{1}{2} (1 \otimes 1 - i \otimes i)$ and $p_2 = 1 \otimes \frac{i}{2i} -i \otimes \frac{1}{2i} = \frac{1}{2} (1 \otimes 1 + i \otimes i)$.

These are the two orthogonal idempotents mentioned in the other answers; as you can see you don't have to guess them etc., you can compute them following a general algorithm. It is useful that many theorems have constructive proofs, such as here the Chinese Remainder Theorem.

Let me just indicate what happens in characteristic $2$. Here we have $k(i) \otimes_k k(i) \cong k(i)[x]/(x+i)^2 \cong k[a,b]/(a^2,b^2)$ with $b=x+i$ and $a=i+1$.

3

This isomorphism would be of $\Bbb R$-algebras. Clearly both ${\Bbb C}\otimes_{\Bbb R}\Bbb C$ and $\Bbb C\oplus\Bbb C$ are four-dimensional as $\Bbb R$-vector spaces. If ${\Bbb C}\otimes_{\Bbb R}{\Bbb C}\cong{\Bbb C}\oplus{\Bbb C}$ then there must be central orthogonal idempotents in the algebra ${\Bbb C}\otimes_{\Bbb R}{\Bbb C}$ corresponding to $(1,0)$ and $(0,1)$. I boldface COI because these elements are critical to decomposing algebras in very general situations. Of course everything is commutative so you don't need to worry about central in this situation.

Obviously $1\otimes1$ is idempotent. What about $i\otimes i$? Well $(i\otimes i)^2=-1\otimes-1=1\otimes 1$, so this is not an idempotent. However it is a nontrivial square root of one. This tells us that

$$1\otimes1=(i\otimes i)^2\iff (1\otimes1+i\otimes i)(1\otimes 1-i\otimes i)=0.$$

That is, $\alpha=1\otimes1+i\otimes i$ and $\beta=1\otimes1-i\otimes i$ are orthogonal, i.e. $\alpha\beta=0$. Are they idempotent? Not quite; check $\alpha^2=2\alpha\Leftrightarrow (\alpha/2)^2=\alpha/2$ and $\beta^2=2\beta\Leftrightarrow(\beta/2)^2=\beta/2$ so $\alpha/2$ and $\beta/2$ are central orthogonal idempotents in $\Bbb C\otimes_{\Bbb R}\Bbb C$.

In general if $R=Re\oplus R(1-e)$ where $e\in R$ is a central idempotent, then an isomorphism is given by $r\mapsto (re,r-re)$ (this works even if $R$ is non-unital!).

This gives us the algebra isomorphism $A=A\frac{\alpha}{2}\oplus A\frac{\beta}{2}$. Now we must find algebra isomorphisms between the ideals $(\alpha/2)$ and $(\beta/2)$ of $A=\Bbb C\otimes_{\Bbb R}\Bbb C$ and $\Bbb C$. The idempotents $\alpha/2,\beta/2$ correspond to $1\in\Bbb C$, so we only need to find square roots of the idempotents' negatives which correspond to the same in $\Bbb C$, i.e. $i\in\Bbb C$. Since $(\lambda\frac{\alpha}{2})^2=\lambda^2\frac{\alpha}{2}$ and $(\mu\frac{\beta}{2})^2=\mu^2\frac{\beta}{2}$, it suffices to find square roots $\lambda,\mu$ of $-1\otimes1$, which can easily be identified as $\lambda,\mu=1\otimes i,i\otimes1$ (it doesn't matter which of the four configurations we choose). For fun choose $\lambda,\mu=i\otimes1$. Compute

$$\lambda\frac{\alpha}{2}=i\otimes1\frac{1\otimes1+i\otimes i}{2}=\frac{i\otimes1-1\otimes i}{2} $$

$$\mu\frac{\beta}{2}=i\otimes1\frac{1\otimes1-i\otimes i}{2}=\frac{i\otimes1+1\otimes i}{2} $$

Therefore, after some (possibly mental) linear algebra,

$$\begin{array}{ll} a(1\otimes1)+b(i\otimes1)+c(1\otimes i)+d(i\otimes i) & = \color{Blue}{\frac{a+d}{2}}\left[\frac{1\otimes1+i\otimes i}{2}\right]+\color{Green}{\frac{a-d}{2}}\left[\frac{1\otimes1-i\otimes i}{2}\right] \\ & \,+\, \color{Magenta}{\frac{b-c}{2}}\left[\frac{i\otimes1-1\otimes i}{2}\right]+\color{Red}{\frac{b+c}{2}}\left[\frac{i\otimes1+1\otimes i}{2}\right]\end{array}$$

gets mapped to

$$\left(\color{Blue}{\frac{a+d}{2}}+\color{Magenta}{\frac{b-c}{2}}i,\color{Green}{\frac{a-d}{2}}+\color{Red}{\frac{b+c}{2}}i\right).$$

This completes our $\Bbb R$-algebra isomorphism $\Bbb C\otimes_{\Bbb R}\Bbb C\cong \Bbb C\oplus\Bbb C$.

Note the four configurations of $\lambda,\mu\in\{1\otimes i,i\otimes1\}$ correspond to the four distinct elements $(i,0),(-i,0),(0,i),(0,-i)$ (not in any particular order) in $\Bbb C\oplus\Bbb C$, analogous to how the conjugation map $i\leftrightarrow -i$ yields a $\Bbb R$-algebra automorphism of $\Bbb C$.

anon
  • 155,259
  • "The strongest isomorphism would be of $\mathbb{R}$-algebras." - No, see the comment by KCd as well as my answer. – Martin Brandenburg Sep 19 '13 at 23:56
  • I meant between the choice of vector spaces or R-algebras in my head. I will edit. – anon Sep 20 '13 at 00:04
  • The factors of $\tfrac{1}{2}$ above are incorrect: as written the identity $1\otimes 1$ is mapped to $(\tfrac{1}{2},\tfrac{1}{2})$. Removing these factors yields the correct reverse direction isomorphism asked for in the comments of another answer. – Adam Marsh Oct 11 '18 at 19:28
1

One example of an isomorphism $\varphi: \Bbb C \oplus \Bbb C \longrightarrow \Bbb C \otimes_{\Bbb R} \Bbb C$ is given on generators by $$\varphi(1, 0) = \tfrac{1}{2}(1 \otimes 1 + i \otimes i),$$ $$\varphi(0, 1) = \tfrac{1}{2}(1 \otimes 1 - i \otimes i).$$

Henry T. Horton
  • 18,984
  • 6
  • 65
  • 80
  • Sorry to ask for more but - I was mainly thinking about an isomorphism in the other direction. I guess one could try to invert this, but is there an easy way for the other direction? – Tedar Sep 19 '13 at 22:21
  • 3
    Since ${\mathbf C} \cong {\mathbf R}[x]/(x^2+1)$ as ${\mathbf R}$-algebras, we have a ${\mathbf C}$-algebra isomorphism ${\mathbf C} \otimes_{\mathbf R} {\mathbf C} \cong {\mathbf C} \otimes_{\mathbf R} {\mathbf R}[x]/(x^2+1) \cong {\mathbf C}[x]/(x^2+1)$, which by the Chinese remainder theorem is isomorphic to ${\mathbf C}[x]/(x-i) \times {\mathbf C}[x]/(x+i)$, and that's isomorphic to ${\mathbf C} \times {\mathbf C}$ (all isomorphisms being as ${\mathbf C}$-algebras). If you know the isomorphism in the Chinese Remainder Theorem you can get an isomorphism in the direction you want. – KCd Sep 19 '13 at 22:59