# Prove that if $A$ is diagonalizable then there is a matrix $B$ such that $B^{2012} = A$

Given: $$A \in M_{n\times n} (\mathbb C) \; , \; A \; \text{is diagonalizable}$$

We need to prove that:

$$\exists B \in M_{n\times n} (\mathbb C) \; : B^{2012} = A$$

What I said so far:

If $A$ is diagonalizable, then $\exists P$ and $D$ such that $A = P D P^{-1}$ where $D = \begin{pmatrix} \lambda_1 & 0 &\cdots& 0 \\ 0 & \lambda_2 & \cdots & 0 \\ \vdots & \vdots & \ddots & \vdots \\ 0 & 0 & \cdots & \lambda_n \end{pmatrix}$ , $P = [[v_1], [v_2], \dotsc ,[v_n]]$, $\det(P) \neq 0$ and also $\lambda_1 , \lambda_2 , \dotsc , \lambda_n$ are all distinct.

also,

$A^{2012} = P D^{2012} P^{-1}$

But where do I exactly go from there?

• It need not be true that the $\lambda_i$ are all distinct; for example $A$ could be the identity. The important thing is that the $\lambda_i$ all have $2012$-th roots, as they are elements of $\mathbb{C}$. – mdp May 2 '13 at 9:31
• Thanks for the heads up. Can you please explain why all $\lambda_i$ all have 2012th roots? They are a "number" afterall.. – TheNotMe May 2 '13 at 9:33
• Consider the polynomial $x^{2012}-\lambda_i$, and the fundamental thereom of algebra. The key property here is that $\mathbb{C}$ is algebraically closed. – mdp May 2 '13 at 9:34
• Any non-zero complex number has $2012$ distinct $2012$th roots. You only need one of them for each $\lambda_i$. – Arthur May 2 '13 at 9:35

The crucial step is that if you have a matrix of the form $ABA^{-1}$ and you take some power of it, this is equal to $(ABA^{-1})^n=AB^nA^{-1}$.
In the case you mention, you have $A=PDP^{-1}$. If you define $\tilde{D}$ to be the one obtained from $D$ by replacing every $\lambda_i$ by $\sqrt[2012]{\lambda_i}$ (where the choice of root doesn't really matter), this results in $(P\tilde{D}P^{-1})^{2012} = P\tilde{D}^{2012}P^{-1} = PDP^{-1}=A$ and so $B=P\tilde{D}P^{-1}$ would solve your question.
• In that case I will be able to prove that $B^n=A$ for any n? – user67773 May 2 '13 at 9:36
• Wow, thank you so much, very simple! And @Umakant I believe so, but only if the matrixes are both $\in \mathbb C$ – TheNotMe May 2 '13 at 9:38
• Not quite, since $B$ is constructed for $n=2012$ explicitly. Just taking an arbitrary power won't give you $A$, in general. However, this says you can construct such a matrix $B$ with $B^n=A$ for any $n$, since the matrices are over $\mathbb{C}$ indeed. – HSN May 2 '13 at 9:38
Consider $B = P D_1 P^{-1}$ such that $$D_1 = \left ( \begin{array} {cccc} \lambda_1^{\frac 1{2012}} & 0 & \cdots & 0\\ 0 & \lambda_2^{\frac 1{2012}} & \cdots & 0\\ \vdots & \ddots & \vdots & \vdots \\ 0 & \cdots & 0 & \lambda_n^{\frac 1{2012}} \end{array}\right)$$ then $$B^{2012} = PD_1^{2012}P^{-1} = PDP^{-1}$$ where $$D = \left ( \begin{array} {cccc} \lambda_1 & 0 & \cdots & 0\\ 0 & \lambda_2 & \cdots & 0\\ \vdots & \ddots & \vdots & \vdots \\ 0 & \cdots & 0 & \lambda_n \end{array}\right)$$