Global existence for nonlinear heat equation and harmonic maps between Riemannian manifolds
Table of Contents
The PDF version of this page can be downloaded by replacing html
in the its address by
pdf
.
For example /html/sheaf-cohomology.html
should become /pdf/sheaf-cohomology.pdf
.
Let \( M \) be a compact Riemannian manifold. We want to solve the following nonlinear heat equation where \( F: M \longrightarrow M'\subset B\subset V = \mathbb{R}^N \): \[ \frac{d F_t}{d t} = -\Delta F_t + \Gamma(F_t) (\nabla F_t)^2 \] We have proved that the solution exists in short-time and is smooth whenever it exists. We will now establish long-time existence using continuity method: we will show that if the solution exists on \( [\alpha,\omega_n] \) where \( \omega_n \) is an increasing sequence to \( \omega \), then the solution exists on \( [\alpha,\omega]\). We then apply short-time existence to gain a small open interval where solution still exists. We then conclude that the solution exists globally on \( [\alpha,+\infty) \) since this interval is connected.
The crucial step to prove that the solution can be extended on \( [\alpha,\omega] \) is to uniformly bound all of its derivatives in time of evolution \( [\alpha,\omega) \). These estimates will also be useful to justify the convergence of \( F_t \) in \( C^\infty(M) \) to a smooth function \( F_\infty \) which will eventually be a harmonic map from \( M \) to \( M' \).
Recall that we proved in Corollary cor:bound-2-2, under the hypothesis of negative curvature, the boundedness of \( \|F_t\|_{W^{2,2}(M)} \) by a constant \( C \) depending only on curvatures of \( M, M' \) and the initial total energies. Since \( \frac{dF_t}{dt } \) relates to spatial derivatives of \( F \) by the nonlinear heat equation, it is easy to see that \( \|F_t\|_{W^{2,2}(M\times[\tau,\tau +\delta])} \) is bounded by a constant independent of \( \tau \). We will denote \( W^{k,p}(M\times [\beta,\gamma]) \) by \( W^{k,p}([\beta,\gamma]) \).
Suppose \( \Riem(M') \leq 0 \). There exists a constant \( C \) depending only on \( \delta \), the metrics and initial total energies such that \[ \|F\|_{W^{2,2}(\tau,\tau+\delta)}\leq C\quad\text{for all } \alpha \leq \tau <\omega-\delta. \]
Since \[ \|F\|^2_{W^{2,2}([\tau,\tau+\delta])} \leq \int_\tau^{\tau+\delta} \|F_t\|^2_{W^{2,2}(M)} dt + 2\int_\tau^{\tau+\delta} \left\| \Delta F_t \right\|^2_{L^2}dt + 2\int_\tau^{\tau+\delta} \left\| \Gamma(F_t)(\nabla F_t)^2 \right\|^2_{L^2}dt \] The first term and the second term are bounded by \( C^2\delta \), the third one, since \( \Gamma(F_t) \) is bounded, by \( C^2\delta \) where \( C \) is a constant only depending on the metrics and initial total energies.
The estimates of higher derivatives of \( F \) will be established in the same strategy as the bootstrap: first in \( W^{2,p} \) for all \( p \) then in \( W^{k,p} \) for all \( k,p \), then in \( C^\infty \).
1 Estimate of higher derivatives.
Suppose \( \Riem(M') \leq 0 \). For all \( p \in (1,+\infty) \), there exists a constant \( C>0 \) depending only on \( \delta \), \( p \), the metrics and initial energies such that for all \( \alpha +\delta \leq \tau \leq \omega-\delta \): \[ \|F\|_{W^{2,p}([\tau,\tau+\delta])}\leq C \]
Applying Gårding Inequality to the parabolic equation \( AF = \Gamma(F) (\nabla F)^2 \) where \( A:= \frac{\partial}{\partial t} + \Delta \) is the heat operator, one has \[ \|F\|_{W^{2,p}([\tau,\tau+\delta])} \leq C\left( \|\Gamma(F)(\nabla F)^2\|_{L^p([\tau -\frac{\delta}{3},\tau +\delta ])} + \|F\|_{W^{2,2}([\tau-\frac{\delta}{3},\tau +\delta])} \right) \] The second term of RHS is already bounded by applying Theorem thm:bound-2-2 to \( \frac{4\delta}{3} \). For the first term: \[ \left\| \Gamma(F) (\nabla F)^2\right\|_{L^p([\tau-\frac{\delta}{3},\tau+\delta])} \leq C(M') \| |\nabla F|^2 \|_{L^p([\tau-\frac{\delta}{3}, \tau+\delta])} = C(M') \| e(F) \|_{L^p([\tau - \frac{\delta}{3},\tau+\delta])}. \]
Recall that by Theorem thm:den-pot, the potential density satisfies \( \frac{de}{dt} +\Delta e -Ce \leq 0 \) for certain constant \( C \) depending only on the metric of \( M \). By Maximum principle (Theorem thm:max-princ-d), one has \( e \leq \psi_\tau \) where \( \psi_\tau \) is the solution of
\begin{cases} \frac{d}{dt}\psi_\tau + \Delta\psi_\tau -C\psi_\tau = 0 \\ \restr{\psi_\tau}{\tau - \frac{\delta}{2}} = \restr{e}{\tau - \frac{\delta}{2}} \end{cases}We apply Gårding Inequality again for \( \psi_\tau \) and obtain
\begin{equation} \label{eq:lem:bound-2-p:2} \| e(F) \|_{L^p([\tau - \frac{\delta}{3},\tau+\delta])} \leq \|\psi_\tau\|_{L^p([\tau- \frac{\delta}{3},\tau+\delta])} \leq C \|\psi_\tau\|_{L^1([\tau- \frac{\delta}{2},\tau+\delta])}. \end{equation}Now apply \( L^1 \)-Comparison Theorem thm:1-comparison-d to \( \psi_\tau \), one has
\begin{equation} \label{eq:lem:bound-2-p:3} \|\psi_\tau\|_{L^1([\tau- \frac{\delta}{2},\tau+\delta])} \leq \int_0^{3\delta/2} \|\restr{\psi_\tau}{\tau-\frac{\delta}{2}}\|_{L^1}e^{Bt}dt \leq \int_0^{3\delta/2}e^{Bt}dt. \|\restr{e}{\tau-\frac{\delta}{2}}\|_{L^1}\leq C. \end{equation}The lemma follows from \eqref{eq:lem:bound-2-p:2} and \eqref{eq:lem:bound-2-p:3}.
We can now estimate higher order derivatives.
Suppose \( \Riem(M') \leq 0 \). For all \( p\in (1,+\infty) \) and \( k<+\infty \), there exists \( C \) depending only on \( k\), \(p\), the metrics and initial energies such that \[ \|F\|_{W^{k,p}([\tau, \tau+\delta])} \leq C \] for all \( \alpha +\delta \leq\tau\leq\omega-\delta \).
Applying Gårding Inequality to the equation \( \frac{dF}{dt} + \Delta F_t = \Gamma(F)(\nabla F)^2 \) then Regularity estimate for the quadratic term (Theorem thm:reg-quad), one has for \( \epsilon \ll \delta \):
\begin{align*} \|F\|_{W^{k,p}([\tau,\tau+\delta])} &\leq C_\epsilon \left( \|F\|_{W^{2,p}([\tau-\epsilon,\tau+\delta])} + \|\Gamma(F)(\nabla F)^2\|_{W^{k-2,p}([\tau-\epsilon,\tau+\delta])}\right) \\ &\leq C_\epsilon\left(1 + C\left(1+\|F\|_{W^{s,q}([\tau-\epsilon,\tau+\delta])}\right)^{q/p}\right) \end{align*}as long as \( k-1 < s \) and \( \frac{1}{p} > \frac{k}{s}.\frac{1}{q} \). Therefore if \( \|F\|_{W^{s,q}([\tau,\tau+\delta])}\leq C(\delta,s,q) \) for all \( \beta\leq \tau\leq\omega-\delta \) and \( q\in (1,+\infty) \), we just proved that \[ \|F\|_{W^{k,p}([\tau,\tau+\delta])} \leq C({\epsilon}, k,p) \] for all
\begin{cases} \beta + \epsilon \leq \tau \leq \omega-\delta \\ k < s+1, p \in(1,+\infty) \end{cases}since \( \|F\|_{W^{s,q}([\tau-\epsilon,\tau+\delta])} \leq 2C(\delta,s,q) \).
One can then conclude by induction on \( k \), with step \( \frac{1}{2} \), starting with \( k=2 \) and \( \epsilon=\frac{\delta}{2} \) divided by 2 after each induction step.
2 Global existence for nonlinear heat equation.
Suppose \( \Riem(M') \leq 0 \). The solution of nonlinear heat equation
\begin{equation} \label{eq:thm:global-heat} \frac{dF}{dt} = -\Delta F +\Gamma(F) (\nabla F)^2 \end{equation}with smooth initial condition exists globally for all time \( t >\alpha \).
Let \( F_n \) be a sequence of solution of \eqref{eq:thm:global-heat} on \( [\alpha,\omega_n] \) with \( \omega_n \) increasing to \( \omega \) then they coincide by uniqueness of the solution. As discussed in the beginning of this part, it is sufficient to prove that the solution extends to \( [\alpha,\omega] \). Let \( F \) be the solution on \( [\alpha,\omega) \) such that \( \restr{F}{[\alpha,\omega_n]} = F_n \), then by Theorem thm:bound-k-p, for all \( \tau \in [\alpha,\omega-\delta) \): \[ \| D^u_t D^v_x F \|_{L^\infty(M\times[\tau,\tau+\delta])} \leq C_{\rm Sobolev} \| D^u_t D^v_x F \|_{W^{k,p}(M\times[\tau,\tau+\delta])} \leq C_{\rm Sobolev} . C(k,p,\delta) \] where, if we choose \( k \) sufficiently large, \( C_{\rm Sobolev} \) is the constant of Sobolev imbedding \( W^{k,p}(M\times[0,\delta]) \hookrightarrow C(M\times[0,\delta]) \) and \( C(k,p,\delta) \) is the constant provided by Theorem thm:bound-k-p.
So all partial derivatives of \( F \) are uniformly bounded on \( [\alpha,\omega) \). This proves that \( F \) extends to a solution on \( [\alpha,\omega] \). In fact \(\restr{F}{\tau}:=\restr{F}{M\times\{\tau\}} \) converges in \( C^\infty(M) \) as \( \tau\to \omega \), since \[ \|D^\alpha \restr{F}{\tau} - D^\alpha\restr{F}{\tau'}\|_{L^\infty} \leq \max_{\|\beta\| = \|\alpha\|+1}\|D^{\beta}F\|_{L^\infty}|\tau-\tau'|. \]
We have just proved the first part of the following theorem.
Let \( M, M' \) be compact Riemannian manifolds with \( \Riem(M') \leq 0 \). Then for every smooth map \( f_0:\ M \longrightarrow M'\subset B\subset \mathbb{R}^N \), the nonlinear heat equation
\begin{equation*} \begin{cases} \frac{df_t}{dt} = \tau(f_t), & \text{for all $t\geq 0$} \\ \restr{f}{t=0} = f_0, \end{cases} \end{equation*}admits a globally defined smooth solution \( f_t \). Moreover, all derivatives \( D^\alpha f_t \) remain bounded as \( t\to +\infty \).
- For a suitable sequence \( \{t_n\} \) increasing to \( +\infty \) the sequence \( \{f_{t_n}\} \) converges in \( C^\infty(M) \) to a function \( f_\infty \) with \( \tau(f_\infty)=0 \). Therefore any map \( f_0:\ M \longrightarrow M' \) is homotopic to a harmonic map.
For any sequence \(\{t_n\}\), one can extract from \( \{f_{t_n}\} \), since their derivatives are uniformly bounded, a subsequence \( \{f_{t_{n_i}}\} \) converging in \( C^k(M, \mathbb{R}^N) \). By a diagonal argument, one can extract from any sequence \( \{f_{t_n}\} \) a subsequence converging in \( C^\infty(M, \mathbb{R}^N) \) to \( f_\infty \). Abusively denote this subsequence by \( \{f_{t_n}\} \), by Theorem thm:den-kin \[ \lim_{n\to\infty} K(f_{t_n}) = \lim_{n\to\infty} \int_M |\tau(f_{t_n})|^2 = 0 \] Therefore \( \tau(f_{t_n}) \to 0 \) in \( L^2(M)^{\oplus N} \). But also \( \tau(f_{t_n}) \to \tau(f_\infty) \) in \( C^\infty(M, \mathbb{R}^N) \), one has \( \tau(f_\infty)=0 \).