Let $(M_t,g_t)$ be a Riemannian manifold evolve by volume preserving mean curvature flow. So , for second fundamental form, we have $$ \partial_t h_{ij}=\Delta h_{ij}-2H h_{im}h^m_j+hh_{im}h^m_j + |A|^2 h_{ij} $$ $H=g^{ij}h_{ij}$ is mean curvature, $|A|^2=g^{ij}g^{kl}h_{ik}h_{jl}$ is inner product of second fundamental form. If the initial manifold is uniformly convex : the eigenvalues of its second fundamental form are strictly positive everywhere. Then, how to show $M_t$ still be uniformly convex for all $t\ge 0$ where the solution exists ?

enter image description here

  • 1
    $\begingroup$ Just apply Hamilton's tensor maximum principle to that evolution equation - at a null eigenvector of $A$ it reduces to $\partial_t A - \Delta A = |A|^2A\ge0.$ $\endgroup$ – Anthony Carapetis Mar 18 '17 at 2:46
  • $\begingroup$ @AnthonyCarapetis How to get the $|A|^2A$? Choice a normal coordinate, such that $g_{ij}=\delta_{ij}$ and $h_{ij}$ is diagonal. Then the above evolution equation is $\partial_t h_{ii}=\Delta h_{ii}-2Hh_{ii}^2+hh_{ii}^2+|A|^2h_{ii}$, then $\partial_t A =\Delta A +(h-H)|A|^2$. $\endgroup$ – lanse7pty Mar 18 '17 at 7:45
  • $\begingroup$ By $A$ I mean the full second fundamental form tensor. What I wrote is not quite true - I really mean that $(\partial_t h_{ij}-\Delta h_{ij}) v^i v^j \ge 0$ whenever $h_{ij} v^j = 0$, which is the "null eigenvector condition" required for the tensor maximum principle. $\endgroup$ – Anthony Carapetis Mar 18 '17 at 8:50

Use Theorem 9.1 of Hamilton's THREE-MANIFOLDS WITH POSITIVE RICCI CURVATURE, which tells you that if a time-dependent symmetric tensor field $h$ satisfies $$\partial_t h_{ij} = \Delta h_{ij} + N_{ij}$$

with the reaction term $N$ satisfying the null-eigenvector condition $$h_{ij} v^i = 0 \implies N_{ij}v^iv^j \ge 0,$$

then positive-definiteness $h \ge 0$ is preserved in time. You can derive this from the scalar maximum principle by studying the scalar function $v \mapsto h(v,v)$ on the unit tangent bundle.

In this case we have $N_{ij} = -2H h_{im}h^m_j+hh_{im}h^m_j + |A|^2 h_{ij}$, so assuming $h_{ij}v^i = 0$ we see $N_{ij} v^i = 0$. Thus the inequality is preserved.

  • $\begingroup$ Thanks, what is the relation of uniformly convex and second fundamental form ? I just Google a uniformly convex space. en.wikipedia.org/wiki/Uniformly_convex_space $\endgroup$ – lanse7pty Mar 18 '17 at 11:31
  • 1
    $\begingroup$ In this context I assume uniformly convex means $h \ge \epsilon g$ for some $\epsilon > 0$; i.e. the eigenvalues of the second fundamental form are bounded below by a uniform constant. Of course, since the manifold is compact, this uniformity doesn't really mean anything - it's equivalent to ordinary convexity $h \ge 0$. This maximum principle argument doesn't rule out the convexity constant $\epsilon$ decreasing in time - you need to study something like $h - \epsilon g$ (or I think Huisken uses $h - \epsilon H g$?) for that, which will be a little more subtle. $\endgroup$ – Anthony Carapetis Mar 18 '17 at 11:38
  • $\begingroup$ Thanks, seemly , Huisken is to prove $h\ge 0$ , then to prove $h \ge \epsilon Hg$, maybe , there is a typo on Huisken's paper. Besides, do you know any about this qustion math.stackexchange.com/questions/2184477/… . I fail to calculate it . $\endgroup$ – lanse7pty Mar 19 '17 at 6:46
  • 1
    $\begingroup$ @lanse7pty: that's a different notion for totally different spaces - we are talking about classical convexity of a domain in $\mathbb R^n$, not local convexity of a Banach space. Strictly speaking, uniform convexity is $h > 0$ and I've only proven convexity $h \ge 0$; but using a stronger maximum principle should preserve $h > 0$ using the same condition. $\endgroup$ – Anthony Carapetis Mar 19 '17 at 7:16
  • 1
    $\begingroup$ @lanse7pty: See e.g. Prop 12.47 of Chow & Knopf volume 2 (Ricci flow techniques and applications). $\endgroup$ – Anthony Carapetis Sep 27 '17 at 4:07

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.