# Matrix diagonalisable in R, but not in C.

I know is quite easy to find a matrix $A\in\mathbb{R}^{2,2}$ that is diagonalisable if the base field is $\mathbb{C}$, but not diagonalisable if the base field is $\mathbb{R}$. The easiest example can be: $$\begin{pmatrix} 0&-1 \\ 1&0 \end{pmatrix}$$ because then we have the eigenvalues equation in the form of $\lambda^2+1=0$.

But what if we would like to find a matrix $B\in\mathbb{C}^{2,2}$ that is diagonalisable if the base field is $\mathbb{R}$, but not diagonalisable if the base field is $\mathbb{C}$? Is this even possible? I came across this question in a math question bank and I have huge concerns about it.

I would like to ask you one more question. What if we would like to find a matrix $B\in\mathbb{Q}^{2,2}$ that is diagonalisable if the base field is $\mathbb{R}$, but not diagonalisable if the base field is $\mathbb{Q}$? Will this matrix do? $$\begin{pmatrix} \pi&0 \\ 0&\pi \end{pmatrix}$$ Thank you very much.

That is impossible: if a matrix is diagonisable in a field K, it is also diagonisable in any field $L$ that contains $K$. The change of basis matrix that will lead to the diagonal form will not change. It is just a matter of extending the scalars from $K$ to $L$.

• Thank you very much. How about the second part? – marco11 May 10 '15 at 23:47
• The general answer I gave is all so about that case. Isn't it clear? – Bernard May 10 '15 at 23:52
• It is clear, but is my example to the second case valid? (I meant the last part) – marco11 May 10 '15 at 23:55
• Well, it's not a matrix in $M_2(\mathbf Q)$, so it's meaningless to try to diagonalise it over $\mathbf Q$. The field which you'll compute must contain the matrix coefficients. – Bernard May 10 '15 at 23:58
• Ah, you are right. What would you say about matrix $\begin{pmatrix} 0&2\\1&0\end{pmatrix}$? Then the eigenvalue equation is $\lambda^2-2=0$ and so it is not diagonalisable in $\mathbb{Q}$ but diagonalisable in $\mathbb{R}$. Am I right? – marco11 May 11 '15 at 0:01

Your example with $\pi$ does not work as the matrix is not even defined over the rationals.
Take for example \begin{pmatrix} 0 & 2 \\ 1 & 0 \end{pmatrix} The characteristic polynomial is $X^2 - 2$. The matrix is diagonalizable over the reals (the polynomial decomposes into linear factors, and the multiplicity of each eigenvector is $1$) but it cannot be diagonalizable over the rationals as the eigenvectors are not rational, and for a matrix to be diagonalizable over a given field you need that the characteristic polynomial factors completely over that field.