Let $f=f(q(t),\dot q(t),t)$, where $q(t)=\{q_1(t),...,q_N(t) \}=\{q_{a}\}_{a=1}^N$ and $\dot q:=\frac{dq}{dt}$. I want to show that if the following equations (Euler-Lagrange) are satisfied identically (independent of $q_a(t)$)

$$\frac{d}{dt}\frac{\partial f}{\partial \dot q_{a}}-\frac{\partial f}{\partial q_{a}}=0 \qquad(1)$$

then $f$ must have the following form

$$f(q(t),\dot q(t),t)=\frac{d \Lambda (q(t),t)}{dt} \qquad(2)$$

I read a book where they rewrite (1) as

$$\sum_{b=1}^N \left[ \frac{\partial^2f}{\partial \dot q_b \partial \dot q_a} \ddot q_b+\frac{\partial^2f}{\partial q_b \partial \dot q_a}\dot q_b \right]+\frac{\partial^2f}{\partial t \partial \dot q_a}=\frac{\partial f}{\partial q_a}\qquad(3)$$

Then the book says that, since we want (1) to be satisfied independent of $q_a$ then the coefficients of $\ddot q_a$ in (3) must be zero. So they find that $f$ must be a linear function of the $q_a$'s:

$$f(q,\dot q,t)=\sum_{b=1}^N A_b(q,t)\dot q_b+B(q,t)\qquad(4)$$

My question is: are they forgetting that the coefficients of the $\dot q_a$ must be zero as well?

The book inserts (4) into (1) and obtain a set of equations of the form

$$\sum_{b=1}^NC_{ab}\dot q_b+\frac{\partial A_a}{\partial t}=\frac{\partial B}{\partial q_a}\qquad(5)$$

Then they say again that since we want (1) to be satisfied independent of $q_a$ then the coefficients of $\dot q_a$ in (5) must be zero. And they find the functional form of $f$.

So my question can be restate as: Why do they have to do this step-by-step argument instead of making zero the coefficients of $\ddot q_a$ and $\dot q_a$ in (3) from the beginning?


Here's my guess: none of the partials can generate a $\ddot{q}_b$ therefore, they must vanish, thus all of the $\frac{\partial^2f}{\partial \dot{q}_b \partial \dot{q}_a}$'s must vanish. However, both $\frac{\partial^2 f}{\partial t \partial \dot{q}_a}$ and $\frac{\partial f}{\partial q_a}$ can create $\dot{q}_b$'s (and $\dot{q}_a$) therefore it's not clear whether or not there is a way to write the function such that $\frac{\partial^2 f}{\partial q_b \partial \dot{q}_a} \neq 0$ (i.e. you cannot assume these are zero). In fact, I'm pretty sure it's not the case that that partial has to be $0$. So if you go through it, it's ugly, but here's what it looks like (keep in mind that $\Lambda$ has no dependence on $\dot{q}$).

\begin{align} f(q, \dot{q}, t) =& \frac{d\Lambda(q, t)}{dt} = \frac{\partial \Lambda}{\partial t} + \sum_1^N \frac{\partial \Lambda}{\partial q_b}\dot{q}_b & \qquad(A)\\ \frac{\partial f}{\partial \dot{q}_a} =& \frac{\partial \Lambda}{\partial q_a} &\qquad(B)\\ \frac{\partial^2f}{\partial q_b \partial \dot{q}_a} =& \frac{\partial^2\Lambda}{\partial q_b \partial q_a} \neq 0 &\qquad(C) \end{align}

Notice in equation (B), that most of the terms vanish because $\frac{\partial\Lambda}{\partial t}$ and all of the $\frac{\partial \Lambda}{\partial q_b}$'s have no dependence on any of the $\dot{q}$'s therefore $\frac{\partial^2 \Lambda}{\partial \dot{q}_i \partial t} = \frac{\partial^2 \Lambda}{\partial \dot{q}_i\partial q_j} = 0$. However, $\frac{\partial \Lambda}{\partial q_a}$ can be dependent on other $q_b$'s!

Just to go through the rest of the derivation, once you assume that $\frac{\partial^2 f}{\partial \dot{q}_b\partial \dot{q}_a} = 0$, this means that $\frac{\partial f}{\partial \dot{q}_a}$ must be independent of all of the $\dot{q}$'s which basically means they cannot be multiplied by each other, divided, put into functions, etc.--there can only be linear terms or no terms (since the linear terms will go away after the first partial). Which gives:

\begin{align} f(q, \dot{q}, t) =& B(q, t) + \sum_1^N A_b(q, t)\dot{q}_b & \qquad(D) \\ \frac{\partial f}{\partial q_a} =& \frac{\partial B}{\partial q_a} + \sum_1^N \frac{\partial A_b}{\partial q_a}\dot{q}_b & \qquad(E) \\ \frac{\partial f}{\partial \dot{q}_a} =& A_a(q, t) & \qquad(F)\\ \frac{d}{dt}\frac{\partial f}{\partial \dot{q}_a} =& \frac{\partial A_a}{\partial t} + \sum_1^N \frac{\partial A_a}{\partial q_b} \dot{q}_b & \qquad(G) \end{align}

Now the only way to generate the $\dot{q}$'s is with those summed partials, so they must be zero:

\begin{align} 0=&\sum_1^N\left(\frac{\partial A_a}{\partial q_b} - \frac{\partial A_b}{\partial q_a}\right)\dot{q}_b & \qquad(H) \\ \frac{\partial A_a}{\partial q_b} =& \frac{\partial A_b}{\partial q_a} & \qquad(I) \\ \frac{\partial B}{\partial q_a} =& \frac{\partial A_a}{\partial t} & \qquad(J) \end{align}

Other than through inspection, I can't come up with a great reason as to how they found the functional form. So here's something you should note (hopefully you remember this about partials of "well-behaved" functions):

$$ \frac{\partial^2 f}{\partial x \partial y} = \frac{\partial^2 f}{\partial y \partial x} $$

So when you see two partials of something set equal, one way to enforce equality is to make the thing you are taking the partial of a partial derivative itself:

$$ \frac{\partial f_1}{\partial x} = \frac{\partial f_2}{\partial y} \rightarrow f_1 = \frac{\partial F}{\partial y}, f_2 = \frac{\partial F}{\partial x} $$

Using that this would give:

$$ \frac{\partial A_a}{\partial q_b} = \frac{\partial A_b}{\partial q_a} \rightarrow A_a = \frac{\partial \Lambda}{\partial q_a}, A_b = \frac{\partial \Lambda}{\partial q_b} \\ \frac{\partial B}{\partial q_a} = \frac{\partial A_a}{\partial t} \rightarrow B = \frac{\partial \Gamma}{\partial t}, A_a = \frac{\partial \Gamma}{\partial q_a} $$

But we already said that $A_a = \frac{\partial \Lambda}{\partial q_a}$, so clearly $\Gamma = \Lambda$ and we finally arrive at:

\begin{align} f(q, \dot{q}, t) =& \frac{\partial \Lambda}{\partial t} + \sum_1^N \frac{\partial \Lambda}{\partial q_b}\dot{q}_b = \frac{d \Lambda(q, t)}{dt} & \qquad(K) \end{align}

  • $\begingroup$ OK. I just have a question. What do you mean with "generate"? "none of the partials can generate a $\ddot q_b$" $\endgroup$ – Ana S. H. Mar 28 '14 at 9:16
  • $\begingroup$ Wait! Are you saying that due to the RHS of my (3) equation is a function of $(q,\dot q, t)$ then the coefficients of $\ddot q_b$ in the LHS of (3) must be zero? (i.e. we have to have a function of $(q,\dot q, t)$ in the LHS as well). And the same argument for my (5) equation? $\endgroup$ – Ana S. H. Mar 28 '14 at 9:19
  • 1
    $\begingroup$ Yes, the RHS of eq. (3) is a function of $(q,\dot{q}, t)$, but also $\frac{\partial^2 f}{\partial t \partial \dot{q}_a}$ and $\frac{\partial^2 f}{\partial q_b \partial \dot{q}_a} \dot{q}_b$. So the only place where $\ddot{q}$ appears is in that single term (which we say the coefficients of must be zero). Then, yes the argument is similar for eq. (5), except now the RHS and $\frac{\partial A_a}{\partial t}$ are functions of only $(q, t)$ and now only the $\sum C_{ab} \dot{q}_b$ contributes $\dot{q}$ terms and thus those coefficients (which are a difference of two partials) must be zero. $\endgroup$ – Jared Mar 28 '14 at 12:13

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.