I have this conceptual question: In Landau's book of classical mechanics, about the principle of least action, it's written:

$$\left. \delta S =\frac{\partial L}{\partial v} \delta q \right\rvert_{t_1}^{t_2} + \ \int\limits_{t_1}^{t_2} \ dt \left(\frac{\partial L}{\partial q}\ - \frac d {dt}\frac {\partial L}{\partial v}\right) \delta q \ =0 ,$$

where $q=q(t)$ is the position function, $v=v(t)$ velocity function, $S$ the action, and L the Lagrangian of the system.

There is the condition $\delta q(t_1)=\delta q(t_2)=0$. So the first term is zero, and then it says " there remains an integral which must vanish for all values of $\delta q$. This can be so only if the integrand is zero identically.

Well I can't understand why

$$\left(\frac{\partial L}{\partial q}\ - \frac d {dt}\frac {\partial L}{\partial v}\right) =0 \;.$$

  • $\begingroup$ Because you are minimizing the action, that whole expression must be zero in order for it to be a minima. Since t1 and t2 are arbitrary, the integrand has to be zero. $\endgroup$ – Señor O Apr 15 '17 at 0:56
  • $\begingroup$ But if the result is x^2 it also vanishes everywhere, isn't it? $\endgroup$ – santimirandarp Apr 15 '17 at 0:59
  • 1
    $\begingroup$ The mathematical statement you seem to be asking about is known as the "fundamental lemma of variational calculus". $\endgroup$ – Adomas Baliuka Apr 15 '17 at 1:02
  • $\begingroup$ What do you mean "if the result is x^2"? What result is x^2? $\endgroup$ – Señor O Apr 15 '17 at 1:06
  • $\begingroup$ Oh, i was saying something silly. I am sorry. But i still wonder if there is a better way to justify that the integrand is zero. $\endgroup$ – santimirandarp Apr 15 '17 at 1:07

Consider the following integral $$\int_{t_1}^{t_2}F(t)\delta q(t) dt=0,$$ for every possible $\delta q$. Then choose a function $\delta q$ which has a large value but is different from zero only in an infinitesimal neighborhood of a point $t_0\in[t_1,t_2]$. Then, $$0=\int_{t_1}^{t_2}F(t)\delta q(t) dt=\int_{t_0-\epsilon}^{t_0+\epsilon}F(t)\delta q(t) dt=F(t_0)\int_{t_0-\epsilon}^{t_0+\epsilon}\delta q(t) dt.$$ In the last equal sign we took $F$ out of the integral because the function is approximately constant in the infinitesimal interval $2\epsilon$. Now, the last integral above is different from zero, so $F(t_0)$ has to vanish. Repeat this argument for all $t\in[t_1,t_2]$ and you obtain that $F(t)$ is identically zero. This is basically the fundamental lemma of variational calculus.

  • $\begingroup$ Okey, i think it is fairly good.. $\endgroup$ – santimirandarp Apr 15 '17 at 1:38

The point is that if $\delta q(t_1) = \delta q(t_2) = 0$ then

$$ 0 = \int_{t_1}^{t_2}{\rm d}t \left(\frac{{\rm d}}{{\rm d}t}\frac{\partial L}{\partial v} - \frac{\partial L}{\partial q}\right)\delta q $$

and this must be true regardless of the selection of $t_1$, $t_2$ or $\delta q$. Imagine that $q$ live in a $n$-dimensional space, all the variations are independent of each-other, this quadrature can be interpreted as a linear combination of such variations. Since they are independent, and add up to zero, this means that the coefficients must vanish, that is

$$ \frac{{\rm d}}{{\rm d}t}\frac{\partial L}{\partial v} - \frac{\partial L}{\partial q} = 0 $$

  • $\begingroup$ Sorry but i am a poor chemist, is there a simpler way to undestand it? $\endgroup$ – santimirandarp Apr 15 '17 at 1:02

When a particle travels from point 1 to point 2, the fundamental law is $F = ma$. One way to find the path a particle follows is to start with the velocity and acceleration of the particle and point 1, and use $F = ma$ to predict the new position after a small time step. Then find the velocity and acceleration at the new place, and follow it another small step. Repeat. Newton used this approach.

Lagrange had a different approach. Start with the initial and final positions and velocities. Consider all paths that connect them. The particle only follows one of these paths. Find a condition that is true for the actual path, and false for all the others. The condition is that $F = ma$ for all points of the path. This turns out to be equivalent to the condition that the integrand is $0$ on the path.

You can find a good derivation of this and a discussion of the benefits of this approach at David Tong's website.

Landau uses a different approach. He chooses the principal of least action as the fundamental law instead of $F=ma$. This is fine. You can derive one from the other.


Maybe this approach will help:

Since the integral $ \ \int\limits_{t_1}^{t_2} \ dt \left(\frac{\partial L}{\partial q}\ - \frac d {dt}\frac {\partial L}{\partial v}\right) \delta q \ $ is zero for any $\delta q$ , it must also be zero for $$\delta q = \epsilon \left(\frac{\partial L}{\partial q}\ - \frac d {dt}\frac {\partial L}{\partial v}\right), $$ with $\epsilon$ a "small enough" real positive number. In this specific case, the action principle requires that, $$ \epsilon \ \int\limits_{t_1}^{t_2} \ dt \left(\frac{\partial L}{\partial q}\ - \frac d {dt}\frac {\partial L}{\partial v}\right)^2 =0. $$ The only way a non-negative integrant to yield an integral equal to zero, is for the integrant to be strictly zero, hence $$ \frac{\partial L}{\partial q}\ - \frac d {dt}\frac {\partial L}{\partial v}=0. $$


protected by Qmechanic Apr 15 '17 at 4:43

Thank you for your interest in this question. Because it has attracted low-quality or spam answers that had to be removed, posting an answer now requires 10 reputation on this site (the association bonus does not count).

Would you like to answer one of these unanswered questions instead?

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