I am trying to solve the following problem:

$$\dfrac{\partial{\phi}}{\partial{t}} = \dfrac{\partial^2{\phi}}{\partial{x}^2} - \cos(x), \ x > 0, t > 0$$

$$\phi(x, 0) = 0, \ x > 0$$

$$\phi(0, t) = e^{-t}, \ t > 0$$

Taking the Laplace transform (in $t$, of course) gives

$$\mathcal{L} \left\{ \dfrac{\partial{\phi}}{\partial{t}} \right\} = \mathcal{L} \left\{ \dfrac{\partial^2{\phi}}{\partial{x}^2} - \cos(x) \right\}$$

$$\therefore s \mathcal{L}\{\phi\} - \phi(0) = \dfrac{d^2 \mathcal{L} \{ \phi \}}{dx^2} - \dfrac{\cos(x)}{s}$$

And since $\phi(x, 0) = 0$, we have

$$s \mathcal{L} \{ \phi \} = \dfrac{d^2 \mathcal{L} \{ \phi \}}{dx^2} - \dfrac{\cos(x)}{s}$$

This is an ODE with constant coefficients.

From now on, let $\mathcal{L}\{ \phi \} = \bar{\phi}$ (for the sake of conciseness).

So our constant-coefficient ODE is

$$\dfrac{d^2 \bar{\phi}}{dx^2} - s \bar{\phi} = \dfrac{\cos(x)}{s}$$

This is a inhomogeneous ODE and, judging by the inhomogeneous term $\dfrac{\cos(x)}{s}$, can be solved using the method of undetermined coefficients.

The homogeneous version of the ODE is

$$\dfrac{d^2 \bar{\phi}}{dx^2} - s \bar{\phi} = 0$$

The characteristic polynomial is

$$m^2 - s = 0$$

$$\therefore m = \pm \sqrt{s}$$

Therefore, the complementary equation for this inhomogeneous ODE is

$$y_c(x) = Ae^{x\sqrt{s}} + Be^{-x\sqrt{s}}$$

And our particular solution for this inhomogeneous ODE is

$$y_p(x) = -\dfrac{1}{s^2 - s} \cos(x)$$

Therefore, our solution to the inhomogeneous ODE is

$$y(x) = Ae^{x\sqrt{s}} + Be^{-x\sqrt{s}} - \dfrac{1}{s^2 - s} \cos(x)$$

Therefore, we have that

$$\bar{\phi}(x, s) = Ae^{x\sqrt{s}} + Be^{-x\sqrt{s}} - \dfrac{1}{s^2 - s} \cos(x)$$

Here is where I am stumped.

HOWEVER, one way that I have seen other problems proceed is by assuming another boundary condition:

$$\lim\limits_{x \to \infty} \phi(x, t) = 0$$

This makes sense from a physical standpoint, but it was not specified in the problem statement, so I am unsure if it is valid for me to use it?

Also, I don't have full solutions for these problems, so I have no way of checking the intermediate steps.

The final solution should be

$$\phi(x, t) = \text{erfc}\left( \dfrac{x}{2\sqrt{t}} \right) - \cos(1 - e^{-t})$$

I would greatly appreciate it if people could please help me solve this problem.

EDIT: If we proceed with assuming the extra boundary condition (that is, if we proceed with assuming that $\lim\limits_{x \to \infty} \phi(x, t) = 0$), then we continue from above as follows:

Now, since $\lim\limits_{x \to \infty} \phi(x, t) = 0$, we have that $Ae^{-x \sqrt{s}} = \dfrac{A}{e^{x \sqrt{s}}} \to 0$ as $x \to \infty$, and $- \dfrac{1}{s^2 - s} \cos(x)$ does not exist as $x \to \infty$, since $\cos(x)$ will not converge.

So what do we do from here? It seems that the $- \dfrac{1}{s^2 - s} \cos(x)$ term is causing us problems?


$$\bar{\phi}(x, s) = \int_0^\infty \phi(x, t) e^{-st} \ dt,$$

we have that $\bar{\phi}(x, s) \to 0$ as $x \to \infty$, and hence $B = 0$, since $Ae^{-x \sqrt{s}} = \dfrac{A}{e^{x \sqrt{s}}} \to 0$ as $x \to \infty$.

Therefore, we now have

$$\bar{\phi}(x, s) = Ae^{-x \sqrt{s}}.$$

We now use the given boundary conditions.

Letting $x = 0$ in the Laplace transform of $\phi$ gives

$$\bar{\phi}(0, s) = \int_0^\infty \phi(0, t)e^{-st} \ dt = \int_0^\infty e^{-t}e^{-st} \ dt = \int_0^\infty e^{-t(1 + s)} \ dt$$

$$= \dfrac{-1}{1 + s} \int_0^{-\infty} e^u \ du = \dfrac{1}{1 + s},$$

since $\phi(0, t) = e^{-t}$ for all $t > 0$.

  • $\begingroup$ @Ian Do you mean something is inconsistent with my work or the problem itself? Also, what do you mean by $\phi(x,0)=\lim_{s \to 0} s \overline{\phi}(x,s)$? I do not understand where this comes from. $\endgroup$ – The Pointer Aug 28 '18 at 14:38
  • $\begingroup$ Sorry, disregard, I made a silly mistake. $\endgroup$ – Ian Aug 28 '18 at 14:40
  • $\begingroup$ @Ian Oh, ok. So does my work seem correct? I wonder if it's the problem itself that is inconsistent? $\endgroup$ – The Pointer Aug 28 '18 at 14:41
  • $\begingroup$ $y_p$ should be $y_p = \frac{1}{s(s+1)}\cos(x)$ because $\cos'' = -\cos$ $\endgroup$ – DisintegratingByParts Aug 29 '18 at 10:45
  • $\begingroup$ @DisintegratingByParts Are you sure? I had it as that, and then was told that it should be $- \frac{1}{s(1+s)}$. See Dylan's answer in math.stackexchange.com/questions/2896937/… $\endgroup$ – The Pointer Aug 29 '18 at 16:08

Let $\widetilde{\phi}(x,s) = \int_{0}^{\infty}e^{-ts}\phi(x,t)dt$ be the Laplace transform of $\phi(x,t)$. As you said we have the following ODE for $\widetilde{\phi}(x,s)$: $$ \dfrac{\partial^2{\widetilde{\phi}(x,s)}}{\partial{x}^2} -s\widetilde{\phi}(x,s)- \frac{\cos(x)}{s} = 0 $$ The general solution is simply: $$ \widetilde{\phi}(x,s)=Ae^{\sqrt{s}x}+Be^{-\sqrt{s}x} -\frac{\cos(x)}{s(s+1)} $$ where $A$ and $B$ are two arbitrary constant that need two condition to be determined. The first condition is a physical one: the solution must be limited in time and so $A=0$. The second condition is that: $$\phi(0, t) = e^{-t}, \ t > 0 .$$ Now the Laplace transform of $\phi(0,t)$ is simply $\widetilde{\phi}(0,s)=\frac{1}{1+s}$ and so we have $B = \frac{1}{s}$. Finally we have that the Laplace transform of our $\phi(x,s)$ is: $$ \widetilde{\phi}(x,s)=\frac{e^{-\sqrt{s}x}}{s} - \frac{\cos(x)}{s(s+1)} $$

Now we have to transforming back. I don't know if you know complex analysis. If not, simply check a Laplace transform's table and see that the inverse Laplace transform of $\frac{e^{-\sqrt{s}x}}{s}$ is $\text{erfc}(\frac{x}{2\sqrt{t}})$ and the inverse L.T. of $\frac{1}{s(s+1)}$ is $1 - e^{-t}$ and so the solution is: $$ \phi(x,t) = \text{erfc}\left( \dfrac{x}{2\sqrt{t}} \right) - \cos(x)(1 - e^{-t}). $$ If you know complex analysis you have to perform the following complex integral: $$ \phi(x,t) = \frac{1}{2{\pi}i}\int_Ldze^{zt}\widetilde{\phi}(x,z) $$ where L is a vertical line in the complex plane on the right of all singularity of $\widetilde{\phi}(x,z)$. Be careful due to the presence of the square root: is a multivalued function.


Let $\Phi(s,x)=\int_{0}^{\infty}\phi(t,x)e^{-st}dt$ be the Laplace transform in time of the desired solution. Then the transform of the heat equation is $$ \int_{0}^{\infty}e^{-st}\frac{\partial\phi}{\partial t}dt =\int_{0}^{\infty}e^{-st}\frac{\partial^2\phi}{\partial x^2}dt-\cos(x)\int_{0}^{\infty}e^{-st}dt. $$ Expanding this out gives $$ e^{-st}\phi(t,x)|_{t=0}^{\infty}-\int_{0}^{\infty}(-se^{-st})\phi(t,x)dt \\ =\frac{\partial^2}{\partial x^2}\int_{0}^{\infty}e^{-st}\phi(t,x)dt-\left.\cos(x)\frac{e^{-st}}{-s}\right|_{t=0}^{\infty}. $$

Therefore, $$ -\phi(0,x)+s\int_{0}^{\infty}e^{-st}\phi(t,x)dt=\frac{\partial^2}{\partial x^2}\int_{0}^{\infty}e^{-st}\phi(t,x)dt-\frac{\cos(x)}{s}. $$

Let $\Phi(s,x)=\int_{0}^{\infty}e^{-st}\phi(t,x)dt$. Then $\phi(0,x)=0$ gives

$$ s\Phi(s,x)=\Phi_{xx}(s,x)-\frac{\cos(x)}{s} \\ \Phi_{xx}-s\Phi=\frac{\cos(x)}{s}. $$ A particular solution is $C\cos(x)$. But $\cos''=-\cos$, which gives $C(-1-s)=1/s$ or $C=-1/(s(1+s))$. So the general solution is

$$ \Phi(s,x)=A\cos(\sqrt{s}x)+B\sin(\sqrt{s}x)-\frac{\cos(x)}{s(1+s)}. $$ Now you have to transform back in $s$.


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.