1

To show this extra term, here is an example problem. Let $x = u$ and $y = \sqrt{r^2 - u^2}$. Find $dx\,dy$ in terms of $u$ and $r$.

Using the Jacobian matrix J, we have that $dx\,dy = |J|\,du\,dr$, where $J = \begin{bmatrix}x_u & x_r \\ y_u & y_r \end{bmatrix}$. Plugging in values for this gives the correct answer $dx\,dy=\frac{r}{\sqrt{r^2-u^2}}\,du\,dr$.

Calculating the total differentials for $x$ and $y$ gives: $$dx = du$$ $$dy = y_r\,dr + y_u\,du=\frac{r}{\sqrt{r^2-u^2}}\,dr -\frac{u}{\sqrt{r^2-u^2}}\,du$$ Plugging in these values to $dx\,dy$: $$du\,\left[\frac{r}{\sqrt{r^2-u^2}}\,dr -\frac{u}{\sqrt{r^2-u^2}}\,du\right]$$ Where did the $-\frac{u}{\sqrt{r^2-u^2}}\,du^2$ term go using the Jacobian method? I'm guessing that we can somehow consider $du^2$ to be so small to ignore, but what if $dr$ were comparable to $du$, so that $dr\,du \approx du^2$? Why does the $du^2$ term go to zero, but not the $dr\,du$ term, and how does the Jacobian "know" to automatically ignore it?

  • 1
  • 3
    You seem to have discovered one of the reasons differentials are defined as elements of the exterior algebra, in which products like $du\land du$ are always zero. It has nothing to do with the "size" of $du$ but rather with the two factors being parallel and therefore spanning a parallelogram of zero area. – Andreas Blass Jun 19 '17 at 00:56
  • @AndrewD.Hwang The meaning of the Jacobian for approximating small patches of area (or volume) is clear to me, as is the total differential being used as an "instantaneous plane" approximation. What I can't reconcile is why the two linear approximations (albeit, of different things) differ by the $du^2$ term. I'm not sure why $du^2$ is "allowed" to go to zero, but $dr,du$ is not. I suppose the explanation of these things would involve analysis, which I have not yet undertaken. Thank you for the reply. – user407691 Jun 20 '17 at 23:37
  • 1
    @user407691: The geometric point is, an infinitesimal parallelogram with sides $du$ and $du$ has area zero. In other words, a parallelogram with sides $du$ and $a, dr + b, du$ has the same area as a parallelogram with sides $du$ and $a, dr$. :) – Andrew D. Hwang Jun 20 '17 at 23:56
  • 1
    @AndrewD.Hwang Ah, that makes sense. Not sure why I failed to see that. Thank you. :) – user407691 Jun 21 '17 at 00:26
  • 1
    @user407691: The notation for infinitesimals does tend to suggest they're scalar quantities rather than (co)vectors. :) – Andrew D. Hwang Jun 21 '17 at 00:52

0 Answers0