22

If $ \partial u/\partial v=a $, then $ \partial v/\partial u=1/a$?

pioneer
  • 388

1 Answers1

34

Functions of a Single Variable

For functions of one variable, if $y=f(x)$ is strictly monotone and differentiable on an interval, and $f'(x)\ne 0$ in that interval, then the inverse function $x=f^{-1}(y)$ is also strictly monotone and differentiable in the corresponding interval and

$$\bbox[5px,border:2px solid #C0A000]{\frac{dx}{dy}=\frac{1}{\frac{dy}{dx}}}\tag 1$$


EXAMPLE:

Suppose $y=\sin(x)$ for $x\in (-\pi/2,\pi,2)$. Note that the sine function is monotone and differentiable on $(-\pi/2,\pi/2)$ with $\frac{dy}{dx}=\cos(x)$ and $\cos(x)\ne 0$.

The inverse function, call it $x=\arcsin(y)$ for $y\in (-1,1)$, is therefore monotone and its derivative is

$$\frac{dx}{dy}=\frac{1}{\cos(x)}=\frac{1}{\sqrt{1-y^2}}$$

Therefore, we have $\frac{d\,\arcsin(y)}{dy}=\frac{1}{\sqrt{1-y^2}}$.


Functions of a Two Variables

The relationship in $(1)$ does not apply, in general, to functions of more than one variable. As an example, examine the transformation of Cartesian coordinates $(x,y)$ to polar coordinates $(\rho,\phi)$ as given by

$$\begin{align} \rho &=\sqrt{x^2+y^2}\\\\ \phi &=\operatorname{arctan2}(y,x) \end{align}$$

and

$$\begin{align} x&=\rho \cos(\phi)\\\\ y&=\rho \sin(\phi) \end{align}$$

We examine the relationship between $\frac{\partial \rho }{\partial x}$ and $\frac{\partial x}{\partial \rho}$ to see if $(1)$ holds. Note that

$$\begin{align} \frac{\partial \rho }{\partial x}&=\frac{x}{\rho}\\\\ & =\cos(\phi)\\\\ &=\frac{\partial x}{\partial \rho} \end{align}$$

Therefore, $\frac{\partial \rho }{\partial x}\ne \frac{1}{\frac{\partial x}{\partial \rho}}$ and $(1)$ does not hold (unless $y=0$).

Instead of the relationship $(1)$ holding, we have instead

$$\begin{equation} \begin{pmatrix} \frac{\partial x}{\partial \rho} & \frac{\partial x}{\partial \phi} \\ \frac{\partial y}{\partial \rho} & \frac{\partial y}{\partial \phi} \end{pmatrix} \begin{pmatrix} \frac{\partial \rho}{\partial x} & \frac{\partial \rho}{\partial y} \\ \frac{\partial \phi}{\partial x} & \frac{\partial \phi}{\partial y} \end{pmatrix}=\begin{pmatrix} 1 & 0 \\ 0 & 1 \end{pmatrix} \end{equation}$$

whereupon matrix inversion becomes

$$\begin{equation} \begin{pmatrix} \frac{\partial x}{\partial \rho} & \frac{\partial x}{\partial \phi} \\ \frac{\partial y}{\partial \rho} & \frac{\partial y}{\partial \phi} \end{pmatrix} =\begin{pmatrix} \frac{\partial \rho}{\partial x} & \frac{\partial \rho}{\partial y} \\ \frac{\partial \phi}{\partial x} & \frac{\partial \phi}{\partial y} \end{pmatrix}^{-1} \tag 2\end{equation}$$

Note that $(2)$ is the analog of $(1)$ and applies whenever a transformation and its inverse exists and is prescribed by differentiable functions. Moreover, it can be generalized to any number of variables.

Ruslan
  • 6,775
Mark Viola
  • 179,405
  • I think you are incorrect as to whether (1) holds with functions of more than one variable. It is true that (1) only holds for total derivatives, which is what you proved. However, (1) holds just fine for the multivariate case.

    In your example $x = \rho \cos(\theta)$. So then, by differentiation, $dx = -\rho \sin(\theta),d\theta + \cos(\theta),d\rho$. Therefore, $$ \frac{dx}{d\rho} = -\rho \sin(\theta) \frac{d\theta}{d\rho} + \cos(\theta) $$

    – johnnyb Jan 09 '18 at 14:54
  • @johnnyb What are you talking about?? Equation $(1)$ holds for single variables. But for partial derivatives, it is not in general true that $\frac{\partial u}{\partial v}=\frac1{\frac{\partial v}{\partial u}}$ – Mark Viola Jan 09 '18 at 16:04
  • @johnnyb In your comment, you have assumed that $\theta$ and $\rho$ are dependent variables. That is not generally the case. – Mark Viola Jan 09 '18 at 16:10
  • As to your first comment, I was talking about the fact that (1) holds for multiple variables, as long as you are dealing in total differentials. I agree that it does not hold for partials, as the partial notation is somewhat nonsensical anyway (i.e., there is not a distinct "partial u". There are multiple "partial u"s. That's why you can't invert the ratio). – johnnyb Jan 10 '18 at 20:01
  • As to your second comment, I don't think I've made any such assumption. However, if I did, I would be greatly appreciative if you can show me an example of where the formula as I gave it breaks down because of variable dependence, where it doesn't also break down in the original equation. That would help my thinking a lot. – johnnyb Jan 10 '18 at 20:02
  • @johnnyb You wrote "as the partial notation is somewhat nonsensical anyway." Partial derivatives are well defined, so your statement is simply unfounded. I showed that (i) under the stated assumptions $y'(x)=1/x'(y)$, (ii) it is not in general true that when $(x,y)\mapsto (u,v)$ that $\frac{\partial u}{\partial x}$ is the reciprocal of $\frac{\partial x}{\partial u}$, and (iii) that the relationship in equation $(1)$ has as a generalization $(2)$. – Mark Viola Jan 10 '18 at 21:04
  • @johnnyb In your comment, you wrote "$\frac{dx}{d\rho}=-\rho\sin(\theta)\frac{d\theta}{d\rho}+\cos(\theta)$," whereby you are tacitly suggesting that the independent variables $\rho$ and $\theta$ are related. If so, we are back to a single variable problem where $x$ depends on $\rho$. – Mark Viola Jan 10 '18 at 21:07
  • I agree with your points (i), (ii), and (iii) above. I also agree that the concept of partial derivatives are well-defined; it is the notation that is non-sensical. – johnnyb Jan 11 '18 at 22:09
  • That's why so many books opt for $f_x$ instead of $\frac{\partial f}{\partial x}$. The reason why you can't treat $\frac{\partial f}{\partial x}$ as a fraction isn't because there is something fundamentally incompatible with treating differentials as fractions, but because the fraction has information about the numerator in the denominator. That's why the notation is non-sensical. If you actually separated out the information it would be workable as a fraction. I'm currently working on a paper on this. – johnnyb Jan 11 '18 at 22:09
  • "you are tacitly suggesting that the independent variables ρ and θ are related"

    I don't think so, though I could be mistaken. I'm saying that the ratio of dx and $d\rho$ is dependent on $\rho$ and $\theta$. The equation does not depend on $\rho$ and $\theta$ being dependent on each other.

    – johnnyb Jan 11 '18 at 22:11
  • As this thread is getting long, we should probably move this to chat. https://chat.stackexchange.com/rooms/71558/discussion-of-differentials-as-ratios – johnnyb Jan 11 '18 at 22:50
  • You have written $\frac{d\rho}{d\theta}$. If these are independent, then that derivative is $0$ and its appearance is superfluous. As far as notation, it's only notation. Neither a total derivative $\frac{df(x)}{dx}$ nor a partial derivative $\frac{\partial f(x,y)}{\partial x}$ is a fraction. So, I'm not sure as to the point you're trying to make regarding fractions. – Mark Viola Jan 11 '18 at 23:38
  • @MarkViola It's an interesting answer, +1 :) . What is the result that you mentioned in equation 2 called. Can you provide me with some reference? – Rishabh Jain Sep 27 '19 at 13:31
  • 1
    @RishabhJain, the equation right before (2) follows from the chain rule. For, if $(\rho, \phi) = P(x, y)$ is the change of variables to polar coordinates, and $(x, y) = C(\rho, \phi)$ the change of variables to Cartesian, then $C\circ P = id$ (on a properly selected domain). Taking the differential of both sides of this identity yields the equation right before (2). – Shlomi A Jan 02 '20 at 12:29