We define $\phi:=(\phi^1,\phi^2):\Omega\subset\mathbb{R}^2\to\phi(\Omega)$ with $\Omega$ such that $\phi$ is a diffeomorphism by

$$x^1:=\phi^1(r,\theta)=r\cos\theta\qquad\text{and}\qquad x^2:=\phi^2(r,\theta)=r\sin\theta$$
for $r,\theta\in\mathbb{R}$.

Compute the pushforward $(\phi^{-1})_*V$ of the vector field $V(x^1,x^2) = \frac{\partial }{\partial x^1}$

  • 1
    $\begingroup$ What have you tried? Do you know the formula for computing the pushforward of vector fields in coordinates? $\endgroup$ – Tyler Holden Sep 25 '13 at 19:53
  • $\begingroup$ I don't know the formula... $\endgroup$ – Thomas Produit Sep 25 '13 at 20:22

The OP clearly does not know where to start, so I will try to give a full answer.

Forget all of the fancy notation. Your function $\phi$ changes from polar coordinates to Cartesian coordinates. If you give it an $r$ and a $\theta$ then $\phi$ will give you an $x$ and a $y$. We have:

$$\phi : (r,\theta) \longmapsto (r\cos\theta,r\sin\theta).$$

The Jacobian matrix of $\phi$ is the matrix of partial derivatives: $$J_{\phi} = \left[\begin{array}{cc} \partial\phi_1/\partial r & \partial\phi_1/\partial \theta \\ \partial\phi_2/\partial r & \partial\phi_2/\partial \theta\end{array}\right] = \left[\begin{array}{cc} \cos\theta & -r\sin\theta \\ \sin\theta & r\cos\theta \end{array}\right]$$

Notice that $\det(J_{\phi}) = r\cos^2\theta + r\sin^2\theta \equiv r$, and so $\phi$ is a diffeomorphism if and only if $r \neq 0$.

Given a vector field in polar form, say $v=a(r,\theta)\partial_r + b(r,\theta)\partial_{\theta}$, we can find $(\phi_*)(v)$:

$$(\phi_*)(v) \sim \left[\begin{array}{cc} \cos\theta & -r\sin\theta \\ \sin\theta & r\cos\theta \end{array}\right]\left[\begin{array}{c} a \\ b \end{array}\right]=\left[\begin{array}{c} a\cos\theta-rb\sin\theta \\ a\sin\theta + br\cos\theta \end{array}\right]$$ Hence $(\phi_*)(v) = (a\cos\theta-rb\sin\theta)\partial_x+(a\sin\theta+br\cos\theta)\partial_y$.

Similarily, given the vector field $\partial_x$, we might want to know which vector field in polar form $v$, gets sent to $\partial_x$. For this you want $(\phi_*)(v)=\partial_x$. In other words, $v=(\phi^{-1}_*)(\partial_x)$. Let $v=a\partial_r+b\partial_\theta$, then $(\phi_*)(a\partial_r+b\partial_{\theta}) = \partial_x$ becomes the matrix equation

$$\left[\begin{array}{cc} \cos\theta & -r\sin\theta \\ \sin\theta & r\cos\theta \end{array}\right]\left[\begin{array}{c} a \\ b \end{array}\right]=\left[\begin{array}{c} 1 \\ 0 \end{array}\right]$$

The matrix of $\phi$ is $J_{\phi}$ and the matrix of $\phi^{-1}$ is $(J_{\phi})^{-1}$. If we multiply on the left by the inverse of $J_{\phi}$ we get $a = \cos\theta$ and $b=-\frac{1}{r}\sin\theta$. Hence:

$$(\phi_*)(\cos\theta\partial_r-\tfrac{1}{r}\sin\theta\partial_{\theta}) = \partial_x$$

Equivalently, this may be written as

$$(\phi^{-1}_*)(\partial_x) = \cos\theta\partial_r-\tfrac{1}{r}\sin\theta\partial_{\theta}$$

However, we need to write $r$ and $\theta$ in terms of $x$ and $y$. Notice that since $x=r\cos\theta$ and $y=r\sin\theta$ we have $r=\sqrt{x^2+y^2}$ and so:

$$(\phi^{-1}_*)(\partial_x) = \frac{x}{\sqrt{x^2+y^2}}\partial_r-\frac{y}{\sqrt{x^2+y^2}}\partial_{\theta}$$

The vector $\partial_x$ based at $(x,y)$ gets sent the what is written above.



If you have fixed coordinates then the Jacobian matrix is the matrix of the push-forward.

Calculate the Jacobian and then multiply it by the appropriate vector.

  • $\begingroup$ What do you mean by vector ? Vector fields apply on function, aren't they ? $\endgroup$ – Thomas Produit Sep 25 '13 at 20:04
  • 2
    $\begingroup$ If you have two manifolds, say $M$ and $N$, and a smooth function between them, say $\operatorname{f} : M \to N$, then the push-forward - also called the differential - is a function between tangent bundles: $\operatorname{f}_* : TM \to TN$. At a given point, say $x \in M$, the push-forward is a linear map between tangent spaces $(\operatorname{f}_*)_x : T_xM \to T_{\operatorname{f}(x)}N$. At each point, it takes a vector in $T_xM$ to a vector in $T_{\operatorname{f}(x)}N$. In general, it takes a vector field on $M$ to a vector field on $N$. $\endgroup$ – Fly by Night Sep 26 '13 at 15:58
  • 2
    $\begingroup$ +1 for the answer, but in the comment a bit of care must be taken. Vectors can always be pushed forward to new vectors, but vector fields may not push forward to vector fields. Specifically, if $f$ fails to be injective with $f(x) = f(y)$, There is no reason $d_x f\, V(x) = d_y f \,V(y)$ should hold, so the image isn't neccesarily a vector field. Think of immersing $S^1$ into $\mathbb{R}^2$ as a figure $8$ for a concrete counter example. $\endgroup$ – Jason DeVito Sep 26 '13 at 18:00
  • 2
    $\begingroup$ @JasonDeVito In general, I agree. But this question deals with a diffeomorphism while your example is an immersion. For a map to be a diffeomorphism, the differential must be bijective at each point. Although, I guess your point is that I should have mentioned that, and in that case: I agree. Thanks for pointing that out. $\endgroup$ – Fly by Night Sep 26 '13 at 18:26
  • $\begingroup$ @FlybyNight: Agreed about the diffeomorphism vs immersion distinction. I recall in grad school I once talked about push-forwarding a v.f. and the professor harped on me. I guess it stuck ;-). Sorry to bother! $\endgroup$ – Jason DeVito Sep 26 '13 at 18:35

To elaborate upon Fly by Night's response, if $F: M \to N$ is a smooth map and $p\in M$ and $f: M \to \mathbb R$, then in a coordinate chart $(x^i)$ about $p$ and $(y^i)$ about $F(p)$ we have $$\left(F_*\left. \frac{\partial}{\partial x^i}\right|_p\right)f = \left( \frac{\partial F^j}{\partial x^i}(p) \frac{\partial}{\partial y^j}\right) f$$ So the Jacobian of your transformation describes how to compute pushforwards in coordinates.

  • $\begingroup$ Does it mean that I have to multiply the Jacobian of $\phi$ to $\begin{pmatrix} \frac{\partial}{\partial r}\\ \frac{\partial}{\partial \theta} \end{pmatrix}$ ? $\endgroup$ – Thomas Produit Sep 25 '13 at 20:20
  • 2
    $\begingroup$ @ThomasProduit No, the symbols $\partial/\partial r$ and $\partial/\partial \theta$ represent basis vector fields. Think back to linear algebra. They play the role of $e_1$ and $e_2$. If your vector is $1e_1+2e_2$ then your represent that as $(1,2)^{\top}$. $\endgroup$ – Fly by Night Sep 26 '13 at 16:03
  • $\begingroup$ I am curious as to why this was downvoted (especially 5 months later). It is a correct statement which answers the question, so...? $\endgroup$ – Tyler Holden Feb 4 '14 at 21:37
  • 1
    $\begingroup$ @TylerHolden: Don't hold your breath, there is not much rime and reason to it; it seems some either enjoy downvoting or believe that hurting you , rather than explaining why you're wrong ( unless the question is of extremely-low quality, which this is not) , is an effective way of learning. Go figure. $\endgroup$ – gary Jan 18 '18 at 17:25

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.