12

Suppose $M$ is a Riemannian manifold. Consider flow $\frac{\partial}{\partial t}g_{ij}=-2(R_{ij}+\nabla_i \nabla_j f)$, where $f$ is a time-dependent function. I would like to prove that flows of this form are equivalent, up to diffeomorphism, to the Ricci flow $\frac{\partial}{\partial t}g_{ij}=-2R_{ij}$, that is:

By defining a 1-parameter family of diffeomorphism $\Psi(t):M\to M$ by

$$\frac{d}{dt}\Psi(t)=\nabla_{g(t)}f(t),$$ $$\Psi(0)=id_M$$

I want to show that $\bar{g}(t):=\Psi(t)^*g(t)$ satisfy $$\frac{\partial}{\partial t}\bar{g}_{ij}=-2 \bar{R}_{ij}.$$

My problem is that I don't know how to calculate $\frac{\partial}{\partial t}\Psi(t)^*g(t)$. I know that $\frac{\partial}{\partial t}\Psi(t)^* \alpha=\mathcal{L}_{\nabla f} \alpha$, where $\mathcal{L}$ is Lie derivative and $\alpha$ is a time-independent object, but I faced problem when $\alpha$ is a time-dependent object.

Can someone point me in the right direction? Thanks in advance for your time.

1 Answers1

8

You can get the correct answer by applying the formal Leibniz/product rule to the expression:

\begin{eqnarray*} \frac{\partial}{\partial t}\left(\Psi\left(t\right)^{*}g\left(t\right))\right) & = & \frac{\partial\Psi\left(t\right)^{*}}{\partial t}g\left(t\right)+\Psi\left(t\right)^{*}\frac{\partial g\left(t\right)}{\partial t}\\ & = & \Psi^{*}\left(\mathcal{L}_{\nabla f}g\right)-2\Psi^{*}\left(\textrm{Rc}+\nabla\nabla f\right)\\ & = & \Psi^{*}\left(-2\textrm{Rc}\right). \end{eqnarray*}

This calculation is written out in slightly more detail in Chapter 9.4 of http://maths-people.anu.edu.au/~andrews/book.pdf.

For a more rigorous proof, start by defining $f(u,v) = \Psi(u)^* g(v),$ so that we are trying to compute $\frac{d}{dt}f(t,t).$ Since everything here is smooth, we can apply the multivariable chain rule to rewrite this as

$$ \begin{eqnarray*} \frac{d}{dt}f(t,t) & = & \left(\frac{\partial f}{\partial u} + \frac{\partial f}{\partial v}\right)_{u=t,\;v=t} \\ & = & \left( \frac{\partial}{\partial u}\left(\Psi(u)^{*}g(v)\right) + \frac{\partial}{\partial v}\left(\Psi(u)^{*}g(v)\right) \right)_{u=t,\;v=t} \end{eqnarray*}. $$

The first term here is $\Psi(u)^*\left(\mathcal L_{\nabla f}g(v)\right)$ by the definition of the Lie derivative, and the second term is $\Psi(u)^* \frac{\partial}{\partial t}g(v)$ because $\Psi(u)^*$ is pointwise linear. Then just make the substitutions $u = t, v = t$ to obtain the result I claimed earlier.

  • 1
    It makes sense. I thought that $\Psi (t)^* g(t)$ has the form of $f(g(t))$ and I must use chain rule for calculating its derivative!! thanks a lot. – Sepideh Bakhoda Sep 13 '13 at 11:06
  • @AnthonyCarapetis: What is the definition of the derivative? Isn't it $$\frac{\partial}{\partial t}\left(\Psi\left(t\right)^{}g\left(t\right)\right)=\lim_{h\to0}\frac{1}{h}\biggl(\Psi\left(t+h\right)^{}g\left(t+h\right)-\Psi\left(t\right)^{*}g\left(t\right)\biggr)?$$ – C.F.G Feb 24 '20 at 11:48
  • Haha, Thanks. Seemly, your answer is suitable for my question. – Enhao Lan Nov 18 '21 at 04:30
  • @C.F.G Maybe, my answer https://math.stackexchange.com/questions/4308471/how-to-show-partial-t-hat-g-sigmat-psi-t-g-sigmat-psi-t-pa/4310381#4310381 is worth to discuss. In fact, I want your judge :-) – Enhao Lan Nov 19 '21 at 09:10
  • @SepidehBakhoda $\Psi(t)^*$ is linear transformation, any linear transformation can be treated as time a matrix, therefore, it meet Leibniz/product rule. – Enhao Lan Nov 19 '21 at 09:12