I have a question about a rule from the calculus of variations.

Assume we consider the space of differentiable functions on $C^1(\mathbb{R})$ (or for the sake of simplicity the smooth functions $C^{\infty}(\mathbb{R})$.

My question is why the "variational derivative" commutes with a total derivative, namely why for $q \in C^1(\mathbb{R})$ holds

$$\frac{\delta}{\delta q(\tilde{t})} \frac{d}{dt} = \frac{d}{dt} \frac{\delta}{\delta q(\tilde{t})} ~?$$


OP is essentially asking the following.

Why the total time derivative $$ \frac{d}{dt}~=~\frac{\partial}{\partial t} + \sum_{m=0}^{\infty}\sum_{i=1}^n q^{i(m+1)}(t)\frac{\partial}{\partial q^{i(m)}(t)} \tag{1} $$ and the functional/variational derivative $$ \frac{\delta}{\delta q^j(t^{\prime})} \tag{2}$$ commute?

That's a good question. The intuitive reason is that the differentiations refer to different variables. But actual calculations make it less obvious (cf. e.g. eq. (3) below). Assume that the derivatives (1) & (2) act on some space ${\cal F}$.

  1. For instance, say that ${\cal F}$ is the space of functions of the form $f(q^{i(m)}(t),t)$. Then we may write the functional derivative (2) as $$ \frac{\delta}{\delta q^j(t^{\prime})} ~=~ \sum_{\ell=0}^{\infty}\delta^{(\ell)}(t\!-\!t^{\prime})\frac{\partial}{\partial q^{j(\ell)}(t)},\tag{2'}$$ because an infinitesimal variation is of the form $$\begin{align}\int\! dt^{\prime} \sum_{j=1}^n \frac{\delta f(q^{i(m)}(t),t)}{\delta q^j(t^{\prime})}\delta q^j(t^{\prime}) &~=~\delta f(q^{i(m)}(t),t) ~=~\sum_{j=1}^n \sum_{\ell=0}^{\infty}\frac{\partial f(q^{i(m)}(t),t)}{\partial q^{j(\ell)}(t)}\delta q^{j(\ell)}(t)\cr &~=~\int\! dt^{\prime} \sum_{j=1}^n \sum_{\ell=0}^{\infty}\delta^{(\ell)}(t\!-\!t^{\prime})\frac{\partial f(q^{i(m)}(t),t)}{\partial q^{j(\ell)}(t)} \delta q^j(t^{\prime}). \end{align} \tag{2"} $$ When we calculate the commutator $$\begin{align}\left[\frac{\delta}{\delta q^j(t^{\prime})}, \frac{d}{dt}\right] &\stackrel{(1)+(2')}{=}~ \sum_{m=0}^{\infty}\sum_{i=1}^n \left[\frac{\delta}{\delta q^j(t^{\prime})},q^{i(m+1)}(t)\right]\frac{\partial}{\partial q^{i(m)}(t)} -\sum_{\ell=0}^{\infty}\left[\frac{d}{dt}, \delta^{(\ell)}(t\!-\!t^{\prime})\right]\frac{\partial}{\partial q^{j(\ell)}(t)} \cr &~~~=~~~ \sum_{m=0}^{\infty} \delta^{(m+1)}(t\!-\!t^{\prime}) \frac{\partial}{\partial q^{i(m)}(t)} - \sum_{\ell=0}^{\infty} \delta^{(\ell+1)}(t\!-\!t^{\prime})\frac{\partial}{\partial q^{j(\ell)}(t)}~=~0,\end{align} \tag{3} $$ we get zero!

  2. If we extend ${\cal F}$ with functions of finite many other times, or with functionals (say, with internal time-integrations), or both, similar calculations show that the derivatives (1) & (2) commute.

| cite | improve this answer | |
  • $\begingroup$ Hi. Thank you for your enlightening explanations. One point is unclear: how do you obtain the expression $(2')$ for $\frac{\delta}{\delta q^j(t^{\prime})} $? $\endgroup$ – KarlPeter Apr 6 '19 at 20:40
  • $\begingroup$ the only applyable tool to calculate $\frac{\delta}{\delta q^j(t^{\prime})}$ I found at your linked wiki page en.wikipedia.org/wiki/Functional_derivative. Here it is stated that for every functional $F\colon M \rightarrow \mathbb{R} $ by definition holds (*)$\begin{align} \int \frac{\delta F}{\delta\rho}(x) \phi(x) \; dx &= \lim_{\varepsilon\to 0}\frac{F[\rho+\varepsilon \phi]-F[\rho]}{\varepsilon} \\&= \left [ \frac{d}{d\epsilon}F[\rho+\epsilon \phi]\right ]_{\epsilon=0},\end{align}$ $\endgroup$ – KarlPeter Apr 7 '19 at 2:26
  • $\begingroup$ I'm not sure which "simple cases" you have in mind but I think that we can take for example the "evaluation functional" $ F_{t_0}: M \to R, f \mapsto f(x_0)$ at $t_0$. Then the right hand side of (*) is $\phi(x_0)$ and on the left side we are looking for a $\frac{\delta F}{\delta\rho}(t) $ such that the integral equals $\phi(t_0)$. We see that $\frac{\delta F}{\delta\rho}(t):= \delta(t -t_0) $ solves the problem. But $\frac{\delta F}{\delta\rho(t')}(t) \neq \delta(t -t_0) \frac{\partial}{\partial \rho(t)} F$. $\endgroup$ – KarlPeter Apr 7 '19 at 2:26
  • $\begingroup$ So this contradicts (2') or do I have overseen something? Futhermore - assume that my calculations are just wrong - how to verify that (2') holds in general? $\endgroup$ – KarlPeter Apr 7 '19 at 2:26
  • $\begingroup$ For starters, eq. (2') applies to functions, not functionals. $\endgroup$ – Qmechanic Apr 7 '19 at 10:57

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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