I'm trying to find the error in my logic here.

Let's say we are given the Laplace operator in polar coordinates:

$$ \frac{\partial^2}{\partial r^2} + \frac{1}{r}\frac{\partial}{\partial r} + \frac{1}{r^2}\frac{\partial}{\partial \theta^2} \tag{1} $$

and we're interested in transforming back to Cartesian coordinates. We could first make use of the usual coordinate transformation, namely

$$ u = r \sin(\theta), ~~v = r \cos(\theta), \tag{2}$$

to write the partial derivatives in polar coordinates as follows,

$$ \frac{\partial}{\partial r} = \sin(\theta) \frac{\partial}{\partial u} + \cos(\theta)\frac{\partial}{\partial v}, \tag{3}$$

$$ \frac{\partial}{\partial \theta} = r \cos(\theta) \frac{\partial}{\partial u} - r \sin(\theta)\frac{\partial}{\partial v}. \tag{4}$$

I think we should then rewrite all $\theta$ and $r$ in terms of $u$ and $v$, making use of $r^2 = u^2 + v^2$, so that in particular

$$ \sin(\theta) = \frac{u}{\sqrt{u^2 + v^2}}, \tag{5}$$ $$ \cos(\theta) = \frac{v}{\sqrt{u^2 + v^2}} \tag{6}.$$

With this, (3) and (4) then become

$$ \frac{\partial}{\partial r} = \frac{u}{\sqrt{u^2 + v^2}} \frac{\partial}{\partial u} + \frac{v}{\sqrt{u^2 + v^2}} \frac{\partial}{\partial v}, \tag{3*}$$

$$ \frac{\partial}{\partial \theta} = v \frac{\partial}{\partial u} - u \frac{\partial}{\partial v}. \tag{4*}$$

Our next task is to then compute $\frac{\partial^2}{\partial r^2}$ and $\frac{\partial^2}{\partial \theta^2}$, which we can carefully do by multiplying the right-hand sides of (3*) and (4*), keeping in mind that the operators will act on the coefficients.

However, executing this and adding the terms together yields nothing that looks remotely near the known form, namely $$\frac{\partial^2}{\partial u^2} + \frac{\partial^2}{\partial v^2}.$$ So I'm wondering, what am I missing here?

In addition to this particular example, I'm interested in any general information you have about transforming partial derivatives from polar to Cartesian coordinates.

  • 4
    $\begingroup$ Wikipedia says the third term in your expression for the polar Laplace operator should have an additional factor of $\frac{1}{r^2}$. $\endgroup$ – Qiaochu Yuan Mar 15 '12 at 7:27
  • $\begingroup$ @Qiaochu: Indeed-- twas quite late when I posted this. I have corrected it now. $\endgroup$ – tentaclenorm Mar 15 '12 at 17:46

Here is my proof: $$ \frac{\partial}{\partial x}=\frac{\partial}{\partial r} \frac{\partial r}{\partial x} +\frac{\partial}{\partial \theta}\frac{\partial \theta}{\partial x}$$ so by applying the product rule

$$ \frac{\partial^2}{\partial x^2}=\frac{\partial}{\partial x}\left(\frac{\partial}{\partial r} \frac{\partial r}{\partial x}+\frac{\partial}{\partial \theta}\frac{\partial \theta}{\partial x}\right)= \frac{\partial^2}{\partial r^2}\left(\frac{\partial r}{\partial x}\right)^2+ \frac{\partial}{\partial r}\frac{\partial^2 r}{\partial x^2} + \frac{\partial^2}{\partial \theta^2}\left(\frac{\partial \theta}{\partial x}\right)^2+ \frac{\partial}{\partial \theta}\frac{\partial^2 \theta}{\partial x^2}$$ $$ +\frac{\partial^2 }{\partial r \partial \theta}\frac{\partial r}{\partial x} \frac{\partial \theta}{\partial x}$$

You get the same relation with $y$. Now you just have to calculate the derivatives of $r,\theta$ with respect to $x,y$. You have $$ r=\sqrt{x^2+y^2},\ \theta=\arctan \frac{y}{x}$$.

What follows is a simple calculus exercise on derivatives. You just need to prove that $$\left(\frac{\partial r}{\partial x}\right)^2+\left(\frac{\partial r}{\partial y}\right)^2=1, $$ etc.(the relations you need so that when you sum $\frac{\partial^2}{\partial x^2}$ and $\frac{\partial^2}{\partial y^2}$ you'll get the polar form of the laplacian).

Going on the lines you started, you shouldn't change $\sin \theta,\cos \theta$ in terms of $x,y$. Just calculate the next derivative with respect to $r,\theta$, using appropriately the formula of the partial derivative of composition of functions.

| cite | improve this answer | |
  • $\begingroup$ Beni, thanks for this. I feel like I'm missing something here, though. Is there a reason why the mixed partial derivatives do not appear in the second equation? [which I believe you meant to write as $\frac{\partial}{\partial x}\left(\frac{\partial}{\partial r} \frac{\partial r}{\partial x} + \frac{\partial}{\partial \theta}\frac{\partial \theta}{\partial x} \right) $] $\endgroup$ – tentaclenorm Mar 18 '12 at 6:12
  • $\begingroup$ @AmeliaYzaguirre: Yes, I forgot the mixed derivatives. Sorry about that. $\endgroup$ – Beni Bogosel Mar 18 '12 at 11:22
  • $\begingroup$ @BeniBogosel Isn't the mixed derivative supposed to have a factor 2? $\endgroup$ – Michiel Aug 20 '14 at 6:51

Using a bit of differential geometry. The key is the writing $\Delta f = \text{div grad}f$: if we manage to express div and grad in a coordinate-independent manner, we can write $\Delta$ easily in any coordinate system, be it cartesian or polar.

The metric and the inverse metric are $$ g_{ab}=\left(\begin{array}[cc] \ 1&0\\ 0&1 \end{array} \right)=g_{ab}^{-1} $$ in cartesian coordinates and $$ g'_{ab}=\left(\begin{array}[cc] \ 1&0\\ 0&r^2 \end{array} \right),\ \ \ g'^{-1}_{ab}=\left(\begin{array}[cc] \ 1&0\\ 0&r^{-2} \end{array} \right) $$ in polar coordinates. The volume form is $\omega=dx\wedge dy=r\, dr\wedge d\varphi$ and the differential of a function is $$ df = \frac{\partial f}{\partial x}dx + \frac{\partial f}{\partial y}dy,\ \ \ dF= \frac{\partial F}{\partial r}dr + \frac{\partial F}{\partial \varphi}d\varphi. $$ Now, div is given by $ d[\omega(E)]=\text{div}(E) \omega, $ for any vector field $E=E^x \partial_x + E^y\partial_y$ or $B=B^r \partial_r + B^\varphi\partial_\varphi$; so $$ d[\omega(E)] = d[E^x dy - E^y dx]=\left(\frac{\partial E^x}{\partial x}+\frac{\partial E^y}{\partial y}\right) dx\wedge dy\implies\text{div}(E)=\frac{\partial E^x}{\partial x}+\frac{\partial E^y}{\partial y}. $$ Or $$ d[\omega(B)] = d[B^r r\,d\varphi - B^\varphi r\,dr] =\left(\frac{1}{r}\frac{\partial (rB^r)}{\partial r}+\frac{\partial B^\varphi}{\partial \varphi}\right)r\, dr\wedge d\varphi\implies\text{div}(B)= \frac{1}{r}\frac{\partial (rB^r)}{\partial r}+\frac{\partial B^\varphi}{\partial \varphi}. $$ And grad$f=g^{-1}df$, so $$ \text{grad}f = \frac{\partial f}{\partial x}\partial_x + \frac{\partial f}{\partial y}\partial_y \ \text{ and } \ \text{grad}F = \frac{\partial F}{\partial r}\partial_r + \frac{1}{r^2}\frac{\partial F}{\partial \varphi}\partial_\varphi. $$ Finally, $\Delta=$div grad: $$ \Delta f = \frac{\partial^2 f}{\partial x^2}+ \frac{\partial^2 f}{\partial y^2} \ \text{ and } \ \Delta F = \frac{1}{r}\left( r \frac{\partial F}{\partial r}\right)+\frac{1}{r^2}\frac{\partial^2 F}{\partial \varphi^2}. $$

| cite | improve this answer | |
  • $\begingroup$ Could you explain how one obtains the metric in polar coordinates (I always get $dx^2+dy^2=2xdx+2ydy=dr^2$, what am I doing wrong)? Then there is an explicit expression of the Laplacian in local coordinates which only uses the metric tensor, right? Thanks in advance. $\endgroup$ – Luke Mathwalker Sep 12 '18 at 0:10
  • 1
    $\begingroup$ @LukeMathwalker Of course: $dx=dr \cos\varphi - r \sin\varphi\, d\varphi$, $dy=dr \sin\varphi+r\cos\varphi\, d\varphi$; this gives you the matrix $M$ with $M_{11}=\cos\varphi$, $M_{12}=-r\sin\varphi$, $M_{21}=\sin\varphi$ and $M_{22}=r\cos\varphi$. Then, $g'=M^T g M$. Concerning your calculation, you should not interpret $dx^2$ as $d(x\cdot x)$ but rather as $dx\otimes dx$. $\endgroup$ – Brightsun Sep 12 '18 at 17:16
  • 1
    $\begingroup$ @LukeMathwalker I think you are referring to $\sqrt{\mathrm{det}(g)} \Delta F = \partial_\mu (\sqrt{\mathrm{det}(g)} g^{\mu\nu} \partial_\nu F)$, where I assumed the Einstein summation convention on the indices $\mu$, $\nu$. One can show that this formula coincides with the one given above whenever the volume form is $dx^1\wedge \cdots \wedge dx^n$ in those coordinates $x^1$, $\ldots$, $x^n$ in which the metric tensor reads diag$(1,1,\ldots,1)$. The above approach has the advantage of also deriving the expressions for the divergence and the gradient. $\endgroup$ – Brightsun Sep 12 '18 at 17:21
  • $\begingroup$ Ah okay thank you! This helped me solving a question I asked here as well math.stackexchange.com/questions/2910656/…. Sadly I had to write the $w_2,w_3$ occuring there as $\cos(t),\sin(t)$ in order to transform it like you described. I would be glad if you could take a short look on that. (This is the last question, I promise :)). $\endgroup$ – Luke Mathwalker Sep 13 '18 at 22:43
  • 1
    $\begingroup$ @LukeMathwalker B. Schutz, Geometrical methods of mathematical physics. $\endgroup$ – Brightsun Sep 14 '18 at 19:28

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.