Mathematics Stack Exchange is a question and answer site for people studying math at any level and professionals in related fields. It's 100% free, no registration required.

Sign up
Here's how it works:
  1. Anybody can ask a question
  2. Anybody can answer
  3. The best answers are voted up and rise to the top

Let $V$ be a finite dimensional inner-product space, and suppose that $T_1$, $T_2$ are normal operators on $V$ that commute. How to show that $T_1+T_2$ and $T_1T_2$ are then normal?

It is clear if $T_1$ commutes with $T_2^*$ and $T_2$ with $T_1^*$, but is it true? If yes how to prove it?

Edit: It is an exercise in Schaum's Outlines of Linear Algebra.

share|cite|improve this question
up vote 4 down vote accepted

My preferred answer was given by Robert Israel, as it is a great opportunity to meet the Fuglede-Putnam-Rosenblum Theorem.

Here is an interpolation argument in the finite-dimensional case where you stand. I will first prove a lemma for matrices. Of course, the same holds for operators on a finite-dimensional complex inner-product space.

Lemma: a matrix $A\in M_n(\mathbb{C})$ is normal if and only if there exists a polynomial $P\in \mathbb{C}[X]$ such that $A^*=P(A)$.

Proof: the implication $\Leftarrow$ is obvious. So assume $A$ is normal. Equivalently, there exists a unitary matrix $U$ and a diagonal matrix $D=\mbox{Diag}\{\lambda_1,\ldots,\lambda_n\}$ such that $A=UDU^*$. Then $A^*=UD^*U^*$ where $D^*=\mbox{Diag}\{\overline{\lambda_1},\ldots,\overline{\lambda_n}\}$. By Lagrange interpolation, we can find a polynomial $P\in\mathbb{C}[X]$ such that $P(\lambda_j)=\overline{\lambda_j}$ for $j=1,\ldots,n$. Hence $D^*=P(D)$ and it follows that $$P(A)=P(UDU^*)=UP(D)U^*=UD^*U^*=A^*.$$ QED.

Now assume that $T$ is normal and commutes with $S$. Then $S$ commutes with every power of $T$, hence with every polynomial in $T$. By the above, $S$ commutes with $T^*$.

The facts you want to prove follow easily.

Note: this remains true in $\mathbb{R}$. But note that we need to be a little bit careful as real normal matrices are not diagonalizable in $M_n(\mathbb{R})$ in general.

share|cite|improve this answer
Thank you very much, your answer is enlightening. – Spenser Apr 21 '13 at 22:33
@Spenser You're welcome. I always liked this lemma. – 1015 Apr 21 '13 at 22:53
I tried to answer my question. I would be very grateful if you could look at it. Many thanks. – Spenser Apr 21 '13 at 23:25
(+1) by the way. – Spenser Apr 22 '13 at 0:07

See the Fuglede-Putnam-Rosenblum theorem (e.g. in Rudin, "Functional Analysis", thm 12.16).

share|cite|improve this answer
Hmm... It is an exercise in Schaum's outline of linear algebra. There should be a simpler proof. I know nothing about functional analysis. Note that I said finite dimensional. – Spenser Apr 21 '13 at 21:18
Like I said, that's my preferred answer, +1. – 1015 Apr 21 '13 at 22:17

Here is an attempt to answer my own question. Please feel free to comment and point out mistakes if you see some.

(Edit: I am learning this stuff right now, so I would really appreciate if an expert could confirm or refute my arguments.)

Lemma: Let $S,T$ be normal operator on a finite dimensional inner-product space, and suppose that $ST=TS$. Then, there is an orthonormal basis for which both $S$ and $T$ are diagonal.

Proof: Since $S$ is normal, $S$ is diagonalizable, so its minimal polynomial factors into distinct linear terms and thus, by the Primary Decomposition Theorem, $$V=E_{\lambda_1}\oplus \cdots\oplus E_{\lambda_k}$$ where $\lambda_i$ are the distinct eigenvalues of $S$ and $E_{\lambda_i}=\ker(S-\lambda_i I)$. Now, if $v\in E_{\lambda_i}$, then $$(S-\lambda_i I)T(v)=ST(v)-\lambda_i T(v)=TS(v)-\lambda_i T(v)=\lambda_i T(v)-\lambda_i T(v)=0$$ so $E_{\lambda_k}$ is $T$-invariant, and thus $T_i:=T|_{E_{\lambda_i}}$ is diagonalizable. Hence, $$E_{\lambda_i}=E_{\mu^i_1}\oplus\cdots\oplus E_{\mu^i_{n_i}}$$ where $\mu^i_1,\ldots,\mu^i_{n_i}$ are the distinct eigenvalues of $T_i$, and $E_{\mu^i_j}=\ker(T_i-\mu^i_j I)$.

Let $B^i_j$ be an orthonormal basis for $E_{\mu^i_j}$. Since $T$ is normal and, for fixed $i$, $\mu^i_1,\ldots,\mu^i_{n_i}$ are distinct, we have $E_{\mu^i_1}\bot\cdots\bot E_{\mu^i_{n_i}}$, so $B^i_1\cup\cdots\cup B^i_{n_i}$ is orthonormal. Moreover, since $S$ is normal and $\lambda_1,\ldots,\lambda_k$ are distinct, $B:=\cup B^i_j$ is still orthonormal. Now, $B$ is an orthonormal basis of eigenvectors for both $S$ and $T$.


Now the result is clear. We have $$[ST^*]_B=[S]_B[T]^*_B=[T]^*_B[S]_B=[T^*S]_B$$ so $ST^*=T^*S$, and similarly, $S^*T=TS^*$.





share|cite|improve this answer
That's almost perfect, +1. You need a little extra justification. As you said, since $S$ and $T$ commute, each $E_{\lambda_i}$ is invariant under $T$. But you need it to be invariant under $T^*$ too if you want to claim that $T_i$ is diagonalizable in an orthonormal basis. But $E_{\lambda_i}=\mbox{Ker}(S^*-\overline{\lambda_i})$ is also an eigenspace of $S^*$. And since $T^*$ commutes with $S^*$, it is indeed invariant under $T^*$. So $T_i^*$ is the restriction of $T^*$ and it follows that $T_i$ is normal. The rest should be ok. – 1015 Apr 22 '13 at 2:16
@julien Thank you so much julien. I learned a lot with this exercice. I understand the problem, I will edit it with your recommendations soon. – Spenser Apr 22 '13 at 2:27

If the operators are normal, then they can be diagonalised, even by a unitary matrix. If they commute, they are simultaneously diagonalisable. I am almost certain that they can be diagonalised by the same unitary matrix: if they have all different eigenvalues then the unitary matrix is unique, so this is clear, and if not then you can use a continuity argument. Since similarity with use of unitary matrix behaves well with the adjoint, you may assume both operators are diagonal. Thus, you are done!

share|cite|improve this answer


Denote normal operators: $$\mathcal{N}(\mathcal{H}):=\{N\in\mathcal{B}(\mathcal{H}):N^*N=NN^*\}$$

Denote bounded calculus: $$\mathcal{N}(N):=\{\eta(N)\in\mathcal{B}(\mathcal{H}):\|\eta\|_\infty<\infty\}$$

For commuting ones: $$MN=NM:\quad\mathcal{N}(M)\cup\mathcal{N}(N)\subseteq\mathcal{N}(M\wedge N)$$

Especially one has: $$MN\in\mathcal{N}(M\wedge N):\quad(MN)^*(MN)=(MN)(MN)^*$$

(No boundedness needed.)


For more details see: Superalgebra

share|cite|improve this answer

Your Answer


By posting your answer, you agree to the privacy policy and terms of service.

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