This is mostly a re-hash of this thread, but it did not receive an adequate answer. In the derivation that I am reading, there is one step that is not justified. Perhaps obvious, but it is not clear to me. Here is the build-up:

Suppose we have the following (one-dimensional) integral that we want to minimize by finding a suitable function $Y(x)$: $\int_{x_1}^{x_2} f(Y(x),Y'(x),x) dx$. Assume we already know the function that minimizes this to be $y(x)$. Define $Y(x,\alpha) = y(x)+\alpha \eta(x)$ and require $\eta(x)$ to be a differentiable function with $\eta(x_1) = \eta(x_2) = 0$ so that this is a suitable path between the points of interests, and assume $\alpha$ is real. From this, define

$$S(\alpha) = \int_{x_1}^{x_2} f(Y(x,\alpha),Y'(x,\alpha),x) dx$$

It is clear that the minimum path is simply $Y(x,0)$, which implies that $\frac{\partial S}{\partial \alpha}\Big|_{\alpha = 0} = 0$. Now, here is where my confusion is:

With no dependence on $\alpha$ for the bounds, we know that

$$\frac{\partial S}{\partial \alpha}\Big|_{\alpha = 0} = \Big[\int_{x_1}^{x_2} \frac{\partial f}{\partial \alpha} dx \Big]_{\alpha = 0} = \Big[\int_{x_1}^{x_2} \Big(\eta(x)\frac{\partial f}{\partial Y}+\eta'(x)\frac{\partial f}{\partial Y'}\Big) dx\Big]_{\alpha = 0} = 0$$

and my author, without justification, immediately makes the jump to

$$\Big[\int_{x_1}^{x_2} \Big(\eta(x)\frac{\partial f}{\partial Y}+\eta'(x)\frac{\partial f}{\partial Y'}\Big) dx\Big]_{\alpha = 0} = \int_{x_1}^{x_2} \Big(\eta(x)\frac{\partial f}{\partial y}+\eta'(x)\frac{\partial f}{\partial y'}\Big) dx$$

Perhaps in clearer form, the following step is taken:

$$ \frac{\partial f(y+\alpha \eta,y'+\alpha \eta',x)}{\partial \alpha} = \eta \frac{\partial f}{\partial y} + \eta'\frac{\partial f}{\partial y'}$$

Can someone provide justification for this step? My intuition tells me that

$$\frac{\partial f}{\partial Y(x,\alpha)}\Big|_{\alpha = 0} = \frac{\partial f}{\partial Y(x,0)} = \frac{\partial f}{\partial y(x)} $$

but some essential part of that just feels wrong. It reminds me of the silly mistake that a careless calculus student can make, saying that

$$\frac{\partial f(x,y)}{\partial x}\Big|_{(x,y) = (x_o,y_o)} = \frac{\partial f(x_o,y_o)}{\partial x}$$

Which is clearly incorrect by noting that $f(x_o,y_o)$ is simply some constant. In the above case I think the issue is much subtler.

  • $\begingroup$ Your problem is the part, where the $\partial Y$ and $\partial Y'$ are replaced by $\partial y$ and $\partial y'$? The line above can be taken as known? $\endgroup$ – Thomas Sep 3 '14 at 6:24
  • 1
    $\begingroup$ That is the chain rule. $\endgroup$ – James S. Cook Sep 3 '14 at 6:32
  • $\begingroup$ Indeed, take the line above as known, as you can work out by bringing in the partial derivative and applying the chain rule. $\endgroup$ – dsm Sep 3 '14 at 6:33
  • $\begingroup$ the intuitive step is wrong because $f$ is a function of both $Y$ and $Y'$ which are technically independent variables here. Moreover, both depend on $\alpha$ so ignoring the $Y'$ does matter. $\endgroup$ – James S. Cook Sep 3 '14 at 6:47

$$ \frac{\partial f(y+\alpha \eta,y'+\alpha \eta',x)}{\partial \alpha} = \eta \frac{\partial f}{\partial y} + \eta'\frac{\partial f}{\partial y'}$$ Why? Well, if $f$ is a function with arguments $f(Y,Y',x)$ and $Y,Y',x$ are thought of as functions of $\alpha$ then the chain-rule is: $$ \frac{\partial}{\partial \alpha}f(Y,Y',x) = \frac{\partial f}{\partial Y}\frac{\partial Y}{\partial \alpha}+ \frac{\partial f}{\partial Y'}\frac{\partial Y'}{\partial \alpha}+ \frac{\partial f}{\partial x}\frac{\partial x}{\partial \alpha}.$$ But, for you variation about the optimal path $y$, $$ \frac{\partial Y}{\partial \alpha} = \eta \qquad \& \qquad \frac{\partial Y'}{\partial \alpha} = \eta'$$ whereas $\frac{\partial x}{\partial \alpha}=0$. Finally, setting $\alpha =0$ changes the evaluation of those partials from arbitrary $Y$ to $y$ and writing the partial derivative w.r.t. $y$ merely expresses that evaluation at the optimal path.

| cite | improve this answer | |
  • $\begingroup$ Thanks for the response. Yes, I understand how the expression with $Y$ is generated with the chain rule. It's the switch from $\partial Y$ to $\partial y$ in the denominator that I wanted justification for (your very last comment). Sure, it may seem obvious that along the optimal path that $\partial Y$ = $\partial y$ since we are holding $\alpha$ constant at zero. For me, though, this lacks rigor, and kind of masks the somewhat clear calculus that I imagine goes into it's justification. $\endgroup$ – dsm Sep 3 '14 at 14:28
  • $\begingroup$ This is a standard notation, we write $df/du$ to denote $f'(u)$. To be explicit, we could write $(df/dx)(u)$, but $df/du$ compactly expresses the derivative evaluated at $u$. It is the same with $Y$ verses $y$. $\endgroup$ – James S. Cook Sep 3 '14 at 19:54

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.