12

I'm having a little difficulty with this. Given some positive operator $A$, show that it is also hermitian.

(A positive operator is defined as $\langle Ax,x\rangle\ge 0$ for all $x \in V$ where $V$ is some vector space.)

Here's what I have so far. We can construct $A = B + iC$ where $B,C$ are hermitian operators $B = (A + A^*)/2$, $C = (-iA + iA^*)/2$ where $^*$ is the conjugate transpose.

I'm trying to show that $B$ and $C$ are diagonalizable by the same vectors, and that the eigenvalues of $C$ are $0$. I'm not sure how to do this though.

user1551
  • 139,064
randomafk
  • 517
  • 4
    But $A=\begin{bmatrix}1&0.1\0&1\end{bmatrix}$ is positive but not Hermitian according to your definition. –  Feb 05 '13 at 23:44
  • I had some difficulty finding the definition of positive operator. That's what I found here: http://www.quantiki.org/wiki/Positive_operator – randomafk Feb 05 '13 at 23:51
  • 1
    Well, who asked you "Given some positive operator $A$, show that it is also Hermitian"? They should give you their definition of a positive operator. –  Feb 06 '13 at 00:15
  • 3
    Seeing Christopher's answer, indeed $A=\begin{bmatrix}1&0.1\0&1\end{bmatrix}$ is not positive over a complex vector space. For $x=\begin{bmatrix}1\i\end{bmatrix}$ we have $\langle Ax,x\rangle = \left\langle\begin{bmatrix}1+0.1i\i\end{bmatrix},\begin{bmatrix}1\i\end{bmatrix}\right\rangle = 2\pm0.1i$ depending on your convention, and it is not true that (or rather, it doesn't make sense to assert that) $2\pm0.1i\ge0$. I should stop sticking my nose into questions about complex linear algebra. –  Feb 06 '13 at 03:47
  • I think you are right in being picky. I'm sure the exercise statement can be understood "from the context", but drooping just the statement here is simply incomplete. – user2820579 Apr 08 '21 at 09:15

4 Answers4

13

The following result is what you are trying to prove:

If $V$ is a finite-dimensional inner product space over $\mathbb{C}$, and if $A: V \rightarrow V$ satisfies $\langle Av, v \rangle \geq 0$ for all $v \in V$, then $A$ is Hermitian.

The result is not true if $V$ is taken to be a real inner product space. That was the key missing ingredient from your question. Here are some strong hints to obtain the proof:

  • Prove that, for all $v \in V$, $\langle (A - A^{\ast})v, v \rangle = 0$, by using the positivity assumption. Remember that over a complex space the inner product is conjugate-linear.
  • Notice that $A - A^{\ast}$ is a normal operator. Then, by applying the spectral theorem, show that $A - A^{\ast}$ must in fact be the zero operator.
Christopher A. Wong
  • 22,445
  • 3
  • 51
  • 82
  • I showed the first pretty easily by just noting that $<Av, v> = <v, Av> \ge 0$

    But I'm not too sure how to use the spectral thm to show the second, nor what to use these hints for! thanks a lot for the help

    – randomafk Feb 06 '13 at 04:23
  • What does the spectral theorem tell you about normal operators? Try to combine that result with the equation $\langle (A - A^*)v, v \rangle = 0$ for all $v \in V$. – Christopher A. Wong Feb 06 '13 at 04:50
  • 1
    Because the spectral thm tells us that $A-A^$ is diagonalizable, it must be similar to the zero operator since $<(A-A^)v,v)=0$ for all $v \in V$ (doesn't this fact alone imply that it's the zero operator, though? Why do we need to invoke the spectral theorum)

    Beyond that, I'm still not sure how to procede. How does that relate to showing A is hermitian?

    – randomafk Feb 06 '13 at 05:40
  • If $\langle Tv, v \rangle = 0$ for every $v$, you have to justify why this implies that $T$ is the zero operator. There are many ways to do it, but the way I'm suggesting, since $T = A - A^*$ is normal, is to use the fact that it is diagonalizable. If something is similar to the zero operator, what does that imply? – Christopher A. Wong Feb 06 '13 at 08:56
  • $A-A* = P 0 P^{-1}=0 \rightarrow A=A*$

    therefore A is hermitian?

    – randomafk Feb 06 '13 at 14:56
  • Is this not true for infinite-dimensional vector spaces? – CyCeez Mar 08 '22 at 02:27
  • @CyCeez If $B = A - A*$, then $B$ is normal and $(Bv,v) = 0$ for every $v$. So it remains to apply an appropriate infinite-dimensional spectral theorem, which requires some assumptions on $A$ and the vector space, e.g. bounded operator on a Hilbert space. – Christopher A. Wong Mar 08 '22 at 04:58
9

Let's go this way.

You already know how to show that any operator $A$ can be written as $A = B + iC$, where $B$ and $C$ are both Hermitian.

As $A$ is positive, for any $|v\rangle$ we should have $\langle v|A|v\rangle$ is a non-negative real number. As $B, C$ are Hermitian they have all real eigenvalues, and a spectral decomposition can be done. $B = \sum_j \lambda_j |j\rangle\langle j|$,$C = \sum_k \lambda_k |k\rangle\langle k|$, where $\lambda_j, \lambda_k \in \mathbb{R}$. Thus $\langle v|B + iC|v\rangle = \sum_j \lambda_j \langle v|j\rangle\langle j|v\rangle + \sum_k i\lambda_k\langle v|k\rangle\langle k|v\rangle$.

As $\langle v|j\rangle\langle j|v\rangle$ is always non-negative real, the second part must always be 0. Thus $A = B$ is Hermitian

Harper
  • 305
0

A simpler way is as follows:

  1. Prove that if any operator $B$ satisfies $\langle x,By\rangle = 0$ for every $x$ and $y$ in the inner product space in $\mathbb{C}$, then $B=0$.
  2. Expand $\langle x+y,B(x+y)\rangle$ and $\langle x+iy,B(x+iy)\rangle$.

These two facts will show that if $B$ satisfies $\langle x,Bx\rangle = 0$ for every $x$ and it is a complex inner product space, then $B=0$. In our case, we take $B=A-A^*$ and conclude the result. Note that all operators are from $V\to V$.

Gautam Shenoy
  • 10,318
0

Since $A=B+iC$ with $B$ and $C$ hermitian, we can decompose the inner product thanks to the linearity in its second argument

$$ \langle \psi | A \psi \rangle=\langle \psi | (B \psi+iC \psi) \rangle=\langle \psi | B \psi \rangle+\langle \psi | iC \psi \rangle $$

We can use the spectral decomposition: $B=Q^\dagger DQ$ and $C=Q'^\dagger D'Q'$ where $Q,Q'$ are orthogonal and $D,D'$ are diagonal matrices of real values.

$$ \langle \psi | B \psi \rangle+\langle \psi | iC \psi \rangle= \langle \psi | Q^\dagger DQ \psi \rangle+\langle \psi | iQ'^\dagger D'Q' \psi \rangle $$

Since the inner product is invariant with respect multiplication of orthogonal matrices

$$ \langle \psi | Q^\dagger DQ \psi \rangle+\langle \psi | iQ'^\dagger D'Q' \psi \rangle= \langle \psi | D \psi \rangle+\langle \psi | iD' \psi \rangle $$

If the $j$-th component of $|\psi\rangle$ is $(x_j+iy_j)$, the innero product $\langle \psi | D \psi \rangle$ can be written as $$ b=\sum_j (x_j+iy_j)(x_j-iy_j)d_j=\sum_j (x_j^2+iy_j^2)d_j $$ that is a real number.

Reasoning in a analogues way we have that $\langle \psi | iC \psi \rangle$ is a complex number, therefore it should be null.

We can conlcude that $A=B=Q^\dagger D Q$, i.e. $A$ is an hermitian operator.