7

I have a question about the basic rule of differentiation of an operator integrated with some parameters. For example, suppose that we have hermitian operator $A(t)$ (so that, $ A(t) = A(t)^\dagger $ for $\forall t$), and $t$ is a parameter of $A$. I'd like to know how to calculate following expression: \begin{align} X = \frac{d}{dt} \int_{t_0}^{t} A(t')\,dt', \end{align}

with $t_0$ indicating the initial parameter value which is considered constant. If $A$ is just a real function with the parameter $t$, it would be simply $ X= A(t)$.

Is this also true for operators in QM?

Qmechanic
  • 201,751
  • Would [math.se] be a better home for this question? – Qmechanic Jul 04 '22 at 05:37
  • Yes. You can prove this by revisiting the FTC proof or taking components of both sides. – J.G. Jul 04 '22 at 05:57
  • Can you explain a little bit more about "taking components of both sides"? Do you mean something like the Taylor expansion of (R.H.S) or the basic definition of integral? – e6-resident Jul 04 '22 at 06:17
  • Related: https://physics.stackexchange.com/q/549698/247642, https://physics.stackexchange.com/q/568519/247642 – Roger V. Jul 04 '22 at 09:38
  • 1
    This is true for bounded operators. The integral would then be interpreted as a Bochner integral and the FTC applies. All these nice theorems really rely on the fact that the set of bounded operators form a normed linear space. But in QM most operators are unbounded and I am not even aware of a notion of integration or differentiation for such a set. Unbounded operators not only do not have a norm but they do not even form a linear space since each operator may have a different domain. – Leonid Jul 11 '22 at 13:16
  • @Leonid, Bochner integral is like to kill a fly with a gun here...There are more easy ways adapted to the formalism if quantum theory and essentially based on the weak (and strong) operatorial topology. – Valter Moretti Aug 05 '22 at 17:04

1 Answers1

6

Yes it is true under a suitable choice of the topology used to define the integral of an operator-valued function.

First of all we have to define the integral of a function that is operator-valued $I\ni t \mapsto A_t$, where all operators $A_t: D\to H$ have a common domain $D$ in the Hilbert space $H$. If the integral were a sum, the following property would be very obvious. $$\left\langle y,\sum_i A_{t_i} \Delta t_i ~ x\right\rangle = \sum_i \langle y, A_{t_i} ~ x\rangle \Delta t_i \quad \forall x\in D, y\in H\:.$$

Independently of the notion of integral we want to use, we expect that this property be valid also for integrals since they are the limit of sums. More precisely, the minimal requirement on the used topology in computing the limit is that it is compatible with the continuity of the scalar product, so when passing from sums to integrals, taking a suitable limit of Riemann sums, we have $$\left\langle y,\int_I A_t dt ~ x\right\rangle = \int_I \langle y, A_t ~ x\rangle dt \quad \forall x\in D, y\in H\:.$$

Our first goal is therefore to define a, possibly unique, operator $\int_I A_tdt$ that satisfies the identity above.

The easiest case is that of a map $$I \ni t \mapsto A_t$$ where $I$ is a real interval and the operators $A_t$ have a common (sub)domain $D$ and satisfy the following requirement. For every $x\in D$ the map $I\ni t \mapsto A_tx$ is continuous in the Hilbert space topology (in other words, $I\ni t \mapsto A_t$ is strongly continuous on its natural domain).

As a consequence, assuming $I$ compact, there is a constant $C_x< +\infty$ such that $||A_t x||\leq C_x$ for all $t\in I$.

The typical situation is $A_t = e^{itB}$ for some selfadjoint operator $B$, and in that case $D$ is the whole Hilbert space $H$ and $C_x= ||x||$ and we can also drop the compactness of $I=\mathbb R$. However also $A_t:= e^{itB}B^n$ is another, less trivial, case with $D:=Dom(B^n)$ and $C_x= ||B^nx||$.

In the said hypotheses, the map $$H \ni y \mapsto f(y)_x:= \int_I \langle y, A_t x\rangle dt $$ is well defined, antilinear and continuous, because $$|f(y)_x| \leq ||y|| C_x \int_I dt\:.$$ Riesz' lemma implies that there is a unique vector denoted by $$\int_I A_t dt ~ x\in H$$ such that $$f(y)_x = \left\langle y,\int_I A_t dt ~ x\right\rangle $$

At this point, using uniqueness of the said vector and $x$-linearity of the map $f(y)_x$, it is not difficult to see that the map $$D\ni x \mapsto\int_I A_t dt ~ x \tag{1}$$ is linear as well. In summary, there exists a unique operator indicated as in (1) such that $$\left\langle y,\int_I A_t dt ~ x\right\rangle = \int_I \langle y, A_t ~ x\rangle dt \quad \forall x\in D, y\in H\:.$$

Since we have established a good definition of integral of operator valued functions, we are in a position to answer the raised question. I recall that a sequence of operators $A_{n}$ all defined on the common domain $D\subset H$ is said to weakly converge to $A: D \to H$ if $$\langle y, A_n x\rangle \to \langle y, Ax\rangle \quad \mbox{as $n\to +\infty$}\quad \forall x\in D, y\in H$$

This seminormed topology is the one for instance used in the LSZ formalism in QFT and it plays a crucial role in physics. It is one of the seven relevant topologies (see comments below).

PROPOSITION. Suppose that the above hypotheses to define the operator $\int_I A_t dt :D \to H$ are valid with $I:= [t_0,t_1]$ and take $t\in (t_0,t_1)$. Then in weak sense it holds $$\frac{d}{dt} \int_{t_0}^t A_s ds = A_t\:.$$

PROOF. The thesis is equivalent to $$\lim_{h\to 0} \frac{1}{h}\left(\int_{t_0}^{t+h}\langle y, A_s x\rangle ds -\int_{t_0}^{t}\langle y, A_s x\rangle ds \right) = \langle y, A_t ~ x \rangle\quad \forall x\in D, y\in H$$ This identity is true because it is equivalent to $$\frac{d}{dt}\int_{t_0}^t \langle y,A_s x\rangle ds = \langle y,A_t x\rangle$$ which is trivially true because the map $$I \ni t \mapsto \langle y,A_t x\rangle$$ is continuous and thus one may apply the Fundamental Theorem of Calculus. QED

ADDENDUM. In case $I\ni t\to A_t$ is such that all $A_t:H\to H$ are bounded and there is a common bound of the norms $||A_t||< C < +\infty$ for all $t\in I$, then everything works with the only hypothesis that $I\ni t \mapsto A_t$ is weakly continuous. The proof is a trivial re-adaptation of that above.

  • This formalism would also apply to, say, a one parameter sequence of differential operators right? for instance, denote the derivative operator by $D$, then for each $f \in Dom(D)$ the map $t \to tD(f)$ is continuous since $D(f)$ has finite $L^2$ norm hence satisfies your conditions. Also, would you happen to have a reference for those seven relevant topologies you speak of? – Leonid Aug 06 '22 at 13:25
  • 1
    Yes, it applies. The seven topologies are the topologies relevant in operator algebra theory, you can find them for instance in Bratteli-Robinson’s two volumes or in Kadison-Ringrose’s books, or in Haag’s book from a more physical perspective. I use them in some of my books… – Valter Moretti Aug 06 '22 at 13:34
  • Have a look at the diagram here https://en.wikipedia.org/wiki/Operator_topologies the seven topologies are those in the diagram omitting the Arens-Mackey and the weak Banach ones. – Valter Moretti Aug 06 '22 at 13:37
  • Great, thank you. – Leonid Aug 06 '22 at 13:41
  • Dear Valter Moretti, if you have time, I'd highly appreciate if you could take a look on this question on math stack exchange. I am a bit stuck and would appreciate any hint/comment/point to references or answer. – Tobias Fünke Jan 19 '23 at 08:00
  • 1
    @Tobias Fünke Unfortunately I am being very busy. I have had a look. I think that without a control on the domain of the detivative $A'(x)$ you cannot say anything in addition to what appears in the answer. However if you assume that the derivative exists for every vector $\psi$ in the Hilbert space, you also have that the derivative operator is bounded as a consequence of the uniform boundedness principle: look at the forst corollary here. – Valter Moretti Jan 19 '23 at 13:17
  • @ValterMoretti I see, thank you! Much appreciated. – Tobias Fünke Jan 19 '23 at 13:37
  • One minor note: It think that my answer was wrong, actually, since I've assumed in my last computations that it would hold for all $\phi,\psi \in H$; but I think it should hold for all $\phi \in H$ but only for all $\psi \in \mathcal D(A^\prime(x))$ if this domain is "only" dense and not the whole Hilbert space. I've corrected this by using your strategy: Assuming that $\mathcal D(A^\prime(x))=H$ instead. – Tobias Fünke Jan 19 '23 at 14:10
  • Sorry I was (and I am) in a hurry, I forgot to say that the first corollary is the one in the Wikipedia page on the uniform boundedness principle. – Valter Moretti Jan 19 '23 at 15:15
  • @ValterMoretti No worries, your help is very much appreciated - I also noted that. Thanks. I've updated my answer already. – Tobias Fünke Jan 19 '23 at 15:40