I have a few questions about differentiating the on-shell action.

Here is what I currently understand (or think I do!):

  1. Given that a system with Lagrangian $\mathcal{L}(\mathbf{q}, \dot{\mathbf{q}}, t)$ has the coordinate $\mathbf{q}_1$ at time $t_1$, and the coordinate $\mathbf{q}_2$ at time $t_2$, there exists a unique 'extremal path' $\gamma(t_1, \mathbf{q}_1, t_2, \mathbf{q}_2; t)$ which makes the action functional $$ \mathcal{S}[\mathbf{q}(t)] = \int_{t_1}^{t_2} \mathcal{L}(\mathbf{q}, \dot{\mathbf{q}}, t)\text{d}t $$ stationary. In other words, $\gamma$ satisfies the Euler-Lagrange equations, $$ \left.\left(\frac{\partial \mathcal{L}}{\partial q} -\frac{\text{d}}{\text{d}t}\frac{\partial \mathcal{L}}{\partial \dot{q}} \right)\right|_{q(t) = \gamma(t)} = \mathbf{0}, $$ and has $\gamma(t_1, \mathbf{q}_1, t_2, \mathbf{q}_2; t_1) = \mathbf{q}_1$ and $\gamma(t_1, \mathbf{q}_1, t_2, \mathbf{q}_2; t_2) = \mathbf{q}_2$.

  2. Moreover, the existence of this function allows the velocity, momentum etc. to be defined at the endpoints, e.g. the momentum at $(t_2, \mathbf{q}_2)$ is $$ \mathbf{p}_2 = \left.\frac{\partial \mathcal{L}}{\partial \dot{\gamma}(t)}\right|_{t=t_2}, $$ where $\dot{\gamma} \equiv \partial \gamma(t_2, \mathbf{q}_2; t) /\partial t$.

  3. Ignoring $t_1$ and $\mathbf{q}_2$ for simplicity, this allows the on-shell action (see here) to be defined as $$ s(t_2, \mathbf{q}_2) = \int_{t_1}^{t_2} \mathcal{L}(\gamma(t_2, \mathbf{q}_2; t), \dot{\gamma}(t_2, \mathbf{q}_2; t), t)\, \text{d} t. \tag{1} $$ Importantly, $s$ is a function of $t_2$, $\mathbf{q}_2$, and not a functional. It can therefore be differentiated as any other function.

  4. It is shown in Landau that

    $$ \frac{\partial s}{\partial t_2} = -\mathcal{H}_2, \quad \frac{\partial s}{\partial \mathbf{q}_2} = \mathbf{p}_2, \tag{2} $$

    but I don't follow the argument given.

I would like to derive the equations (2) by directly differentiating (1). I have read several answers which derive this in a different way (here, here and here), but I still have some questions. Firstly, here is my attempt at differentiating with respect to $\mathbf{q}_2$.

\begin{align} \frac{\partial s}{\partial \mathbf{q}_2} &= \frac{\partial}{\partial \mathbf{q}_2} \int_{t_1}^{t_2} \mathcal{L}(\gamma(t_2, \mathbf{q}_2; t), \dot{\gamma}(t_2, \mathbf{q}_2; t), t)\, \text{d} t \\ &= \int_{t_1}^{t_2} \frac{\partial}{\partial \mathbf{q}_2} \mathcal{L}(\gamma(t_2, \mathbf{q}_2; t), \dot{\gamma}(t_2, \mathbf{q}_2; t), t)\, \text{d} t\\ &= \int_{t_1}^{t_2} \frac{\partial \mathcal{L}}{\partial \gamma}\cdot\frac{\partial \gamma}{\partial \mathbf{q}_2} +\frac{\partial \mathcal{L}}{\partial \dot{\gamma}}\cdot\frac{\partial \dot{\gamma}}{\partial \mathbf{q}_2} \text{d}t. \end{align} Now, $$ \frac{\partial \dot{\gamma}}{\partial \mathbf{q}_2} =\frac{\partial}{\partial \mathbf{q}_2} \frac{\text{d}\gamma}{\text{d} t} = \frac{\text{d}}{\text{d} t} \frac{\partial \gamma}{\partial \mathbf{q}_2}, $$ so we can integrate by parts to yield \begin{align} \frac{\partial s}{\partial \mathbf{q}_2} &= \left[ \frac{\partial \mathcal{L}}{\partial \dot{\gamma}} \cdot \frac{\partial \gamma}{\partial \mathbf{q}_2}\right]_{t_1}^{t_2} + \int_{t_1}^{t_2} \underbrace{\left(\frac{\partial \mathcal{L}}{\partial \gamma} - \frac{\text{d}}{\text{d} t} \frac{\partial \mathcal{L}}{\partial \dot{\gamma}}\right)}_{\mathbf{0}}\cdot\frac{\partial \gamma}{\partial \mathbf{q}_2} \text{d} t\\ &=\mathbf{p}_2\cdot\frac{\partial \gamma}{\partial \mathbf{q}_2}(t_2). \end{align} For (2) to be true, we ought to have $\frac{\partial \gamma}{\partial \mathbf{q}_2}(t_2) = \mathbf{I}$. Is it valid interchange the order of evaluation and differentiation to write $$ \left.\frac{\partial \gamma(t_2, \mathbf{q}_2; t)}{\partial \mathbf{q}_2}\right|_{t=t_2} = \frac{\partial \gamma(t_2, \mathbf{q}_2; t_2)}{\partial \mathbf{q}_2} = \frac{\partial \mathbf{q}_2}{\partial \mathbf{q}_2} =\mathbf{I}?\tag{3} $$ If so, why? If not, then how else is it possible to arrive at equation (2) from here?

Secondly, here is my attempt at differentiating with respect to $t_2$. \begin{align} \frac{\partial s}{\partial t_2} &= \frac{\partial}{\partial t_2} \int_{t_1}^{t_2} \mathcal{L}(\gamma(t_2, \mathbf{q}_2; t), \dot{\gamma}(t_2, \mathbf{q}_2; t), t)\, \text{d} t \\ &= \mathcal{L}_2 + \int_{t_1}^{t_2} \frac{\partial}{\partial t_2} \mathcal{L}(\gamma(t_2, \mathbf{q}_2; t), \dot{\gamma}(t_2, \mathbf{q}_2; t), t)\, \text{d} t\\ &=\mathcal{L}_2 + \int_{t_1}^{t_2} \frac{\partial \mathcal{L}}{\partial \gamma}\cdot\frac{\partial \gamma}{\partial t_2} +\frac{\partial \mathcal{L}}{\partial \dot{\gamma}}\cdot\frac{\partial \dot{\gamma}}{\partial t_2} \text{d}t\\ &= \mathcal{L}_2 +\left[ \frac{\partial \mathcal{L}}{\partial \dot{\gamma}} \cdot \frac{\partial \gamma}{\partial t_2}\right]_{t_1}^{t_2} + \int_{t_1}^{t_2} \underbrace{\left(\frac{\partial \mathcal{L}}{\partial \gamma} - \frac{\text{d}}{\text{d} t} \frac{\partial \mathcal{L}}{\partial \dot{\gamma}}\right)}_{\mathbf{0}}\cdot\frac{\partial \gamma}{\partial t_2} \text{d} t\\ &=\mathcal{L}_2 + \mathbf{p}_2\cdot\frac{\partial \gamma}{\partial t_2}(t_2) \end{align} To get from the first to the second line I used Leibniz' rule for differentiating integrals. For equation (2) to be true, we ought to have $$ \frac{\partial \gamma}{\partial t_2}(t_2) = -\dot{\mathbf{q}}_2.\tag{4} $$ Is this correct? If so, how can it be shown?

I would be very grateful for any help anyone is able to give!


1 Answer 1



  • Eq. (3) follows from the boundary condition $$\gamma(t_2, \mathbf{q}_2; t\!=\!t_2)~=~\mathbf{q}_2. \tag{A}$$

  • Eq. (4) follows by differentiating eq. (A) wrt. $t_2$: $$ \left.\frac{\partial\gamma(t_2, \mathbf{q}_2; t)}{\partial t_2}\right|_{t=t_2} + \left. \frac{\partial\gamma(t_2, \mathbf{q}_2; t)}{\partial t}\right|_{t=t_2}~=~0.\tag{B}$$


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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