0

Given $1>k>0$, I want to show that $\forall t>0$, \begin{equation}%\label{eqn: 1st time derivative} \lim_{|x|\to\infty}\frac{u_x(x,t)}{(u(x,t))^k}=0, \end{equation} where \begin{equation} u(x,t)=\frac{1}{\sqrt{2\,\pi\,t}}\int_{-\infty}^{\infty} g(y)\,e^{-\frac{(y-x)^2}{2\,t}}\,dy \end{equation} is a solution of the heat equation $u_t=\frac{1}{2}u_{xx}$. Here, we assume $g>0$ and $g$ is smooth. My first attempt is to evaluate \begin{equation}%\label{eqn: 1st time derivative} \int_{-\infty}^{\infty}\frac{u_x(x,t)}{(u(x,t))^k}\,dx = \int_{u=0}^{u=0}u(x,t)^{-k}\,du(x,t) =0, \end{equation} using the fact $\lim_{|x|\to\infty}u(x,t)=0$. However, this does not give \begin{equation}%\label{eqn: 1st time derivative} \int_{-\infty}^{\infty}\left|\frac{u_x(x,t)}{(u(x,t))^k}\right|\,dx=0, \end{equation} and we cannot conclude $\lim_{|x|\to\infty}\frac{u_x(x,t)}{(u(x,t))^k}=0$. On the other hand, the initial condition $g(y)=e^{-y^2}$ gives \begin{equation} u(x,t)=\frac{e^{-\frac{x^2}{2 t+1}}}{\sqrt{2 t+1}}, \end{equation} and it is easy to see that $\lim_{|x|\to\infty}\frac{u_x(x,t)}{(u(x,t))^k}=0$, which seems to be true for more general $g(y)$. Any reference, suggestion, idea, or comment is welcome. Thank you!

LCH
  • 815
  • Is that representation the inverse Fourier transform? And if so what are your boundary conditions because it can be easier to work problems like this out without using an explicit definition of $u$ – Henry Lee May 03 '21 at 17:01
  • Just something "manual" to try that came to mind: use the Leibniz rule to take the derivative and then integrate by parts. You get asked to differentiate $g$, which is not allowed a priori, but you can fix that by using the semigroup property to rewrite $u(x,t)=H_{t/2} * (H_{t/2} * g)$ where $H_t$ is the heat kernel, and now $H_{t/2} * g$ is smooth. So now assuming WLOG that $g$ is smooth you can integrate the expression for $u_x$ by parts...does the result you get help at all? – Ian May 03 '21 at 17:02
  • (Of course, you might be trying to use this statement to show the smoothing property, and in this case this direction of attack is circular.) – Ian May 03 '21 at 17:05
  • yes, this helps! I'm now doing "use the Leibniz rule to take the derivative and then integrate by parts." Is the desired result reasonable? – LCH May 03 '21 at 17:15
  • Is this correct? $u_x(x,t)=\frac{1}{\sqrt{2,\pi,t}}\int_{-\infty}^{\infty} e^{-\frac{z^2}{2,t}}g_z(z+x),dz$. Yes, you are right, the limit diverges for $k\ge1$. – LCH May 03 '21 at 17:28
  • How to deal with $(u(x,t))^k$ and compare it with $u_x(x,t)$? – LCH May 03 '21 at 17:38
  • I've performed some calculations below. What's the next step to find the limit? – LCH May 03 '21 at 17:56

1 Answers1

1

It seems we need to assume $|g(y)|\le M_1\,e^{ay^2}$ and $|g_y(y)|\le M_2\,e^{by^2}$ for some $a,b,M_1,M2>0$, or assume $|g_y(y)|\le M_3\,|g(y)|$ for some $M_3>0$?

\begin{align}%\nonumber \notag u(x,t)=&\frac{1}{\sqrt{2\,\pi\,t}}\int_{-\infty}^{\infty} g(y)\,e^{-\frac{(y-x)^2}{2\,t}}\,dy && \text{} \\[1ex] \notag u_x(x,t)=&\frac{1}{\sqrt{2\,\pi\,t}}\int_{-\infty}^{\infty} g(y)\,\frac{(y-x)}{t}\,e^{-\frac{(y-x)^2}{2\,t}}\,dy && \text{Leibniz integral rule} \\[1ex] \notag =&\frac{1}{\sqrt{2\,\pi}t^{\frac{3}{2}}}\int_{-\infty}^{\infty} g(y)\,(y-x)\,e^{-\frac{(y-x)^2}{2\,t}}\,dy && \text{simplify} \\[1ex] \notag =&\frac{1}{\sqrt{2\,\pi}t^{\frac{3}{2}}}\int_{-\infty}^{\infty} g(z+x)\,z\,e^{-\frac{z^2}{2\,t}}\,dz && \text{let $z:=y-x$ $\iff$ $y=z+x$} \\[1ex] \notag =&\frac{1}{\sqrt{2\,\pi}t^{\frac{3}{2}}}\int_{-\infty}^{\infty} g(z+x) (-t)\,d\left(e^{-\frac{z^2}{2\,t}}\right) && \text{rewrite} \\[1ex] \notag =&\frac{-1}{\sqrt{2\,\pi\,t}}\int_{-\infty}^{\infty} g(z+x)\,d\left(e^{-\frac{z^2}{2\,t}}\right) && \text{simplify} \\[1ex] \notag =&\frac{-1}{\sqrt{2\,\pi\,t}} \left( g(z+x)\,e^{-\frac{z^2}{2\,t}}\bigg|_{-\infty}^{\infty} - \int_{-\infty}^{\infty} e^{-\frac{z^2}{2\,t}}g_z(z+x)\,dz \right) && \text{IBP} \\[1ex] \notag =&\frac{1}{\sqrt{2\,\pi\,t}} \int_{-\infty}^{\infty} e^{-\frac{z^2}{2\,t}}g_z(z+x)\,dz && \text{assume $g$ is bounded} \\[1ex] \notag =&\frac{1}{\sqrt{2\,\pi\,t}} \int_{-\infty}^{\infty} e^{-\frac{(y-x)^2}{2\,t}}g_y(y)\,dy && \text{$z:=y-x$ $\iff$ $y=z+x$} \end{align}

\begin{align}%\nonumber \notag u(x,t)=&\frac{1}{\sqrt{2\,\pi\,t}}\int_{-\infty}^{\infty} g(y)\,e^{-\frac{(y-x)^2}{2\,t}}\,dy && \text{} \\[1ex] \notag (u(x,t))^k=&\left(\frac{1}{\sqrt{2\,\pi\,t}}\right)^k\left(\int_{-\infty}^{\infty} g(y)\,e^{-\frac{(y-x)^2}{2\,t}}\,dy\right)^k && \text{$k$th power} \end{align}

\begin{align}%\nonumber \notag \frac{u_x(x,t)}{(u(x,t))^k}=& \left(\frac{1}{\sqrt{2\,\pi\,t}}\right)^{1-k} \frac{\int_{-\infty}^{\infty} e^{-\frac{(y-x)^2}{2\,t}}g_y(y)\,dy} {\left(\int_{-\infty}^{\infty} g(y)\,e^{-\frac{(y-x)^2}{2\,t}}\,dy\right)^k} && \text{} \\[1ex] \notag =&\left(\frac{1}{\sqrt{2\,\pi\,t}}\right)^{1-k} \frac{\int_{-\infty}^{\infty} e^{-\frac{(y-x)^2}{2\,t}}g_y(y)\,dy} {\int_{-\infty}^{\infty} g(y)\,e^{-\frac{(y-x)^2}{2\,t}}\,dy} \left(\int_{-\infty}^{\infty} g(y)\,e^{-\frac{(y-x)^2}{2\,t}}\,dy\right)^{1-k} && \text{} \end{align}

LCH
  • 815
  • You can get this boundedness using the same $u=H_{t/2} * (H_{t/2} * g)$ trick I mentioned. – Ian May 03 '21 at 23:37
  • Could you make your trick more explicit? – LCH May 03 '21 at 23:41
  • Look up "semigroup property of the heat equation". It is a little hard to explain quickly, there is some notation to be developed. Basically the idea is that you can determine $u(t+s,x)$ by solving the heat equation up to time $t$ and then "restart" and evolve what you currently have by an additional time increment of $s$. – Ian May 03 '21 at 23:46
  • Do you mean (7.1.7) and (7.1.8) in https://link.springer.com/content/pdf/10.1007%2F978-0-387-49319-0_8.pdf – LCH May 04 '21 at 04:02