1
$\begingroup$

A heat equation problem with Dirichlet boundary conditions on the domain $[x_1,x_2]$

$$\frac{\delta u}{\delta t} = k \frac{\delta^2 u}{\delta x^2}$$

$$u(x_1,t) = u(x_2,t) = 0$$

Would have eigenfunctions corresponding to sines. Whereas Neumann boundary conditions $$u'(x_1,t) = u'(x_2,t) = 0$$

Results in cosine eigenfunctions.

When evaluating steady-state solutions, with Neumann conditions we can solve for the first coefficients of the Fourier Cosine Series expansion. In the case of Dirichlet conditions, the sine series do not have a $A_0$ coefficient describing steady-state.

Intuitively the steady-state solution for Dirichlet conditions should always decay to zero, as we are allowing heat exchange on the borders and the solution for u has a time-dependent exponential decay. This is constrained with the Neumann conditions (no heat exchange at borders = isolated body), having a non-trivial steady-state.

Is this intuition correct? Is the steady-state solution for the heat equation with Dirichlet B.C always zero? Or is there something I am missing?

$\endgroup$
1
$\begingroup$

Your reasoning is correct. The steady state solutions can be obtained by setting $\partial u/\partial t=0$, leading to $$ u = c_1 x+c_2. $$ For the homogeneous Dirichlet B.C., the only solution is the trivial one (i.e., $u=0$. For the Neumann B.C., a uniform solution $u=c_2$ exists. Also, the steady state solution in this case is the mean temperature in the initial condition. Integrating the heat equation along $x_1<x<x_2$, $$ \frac{\partial}{\partial t} \int_{x_1}^{x_2} u dx =k \left.\frac{\partial u}{\partial x} \right|_{x_1}^{x_2}=0, $$ i.e., the total energy $E=\int_D u dx$ inside the domain is constant. For the steady state solution, $$ \int_{x_1}^{x_2} u(t\to \infty,x) dx=\int_{x_1}^{x_2} u(t=0,x) dx $$ $$ c_2=\frac{1}{x_2-x_1} \int_{x_1}^{x_2} u_0(x) dx, $$ in which $u_0(x)$ is the initial condition for $u$.

$\endgroup$
2
  • 1
    $\begingroup$ This is very helpful. How are you getting $u = c_1 x + c_2$ after setting $\delta u / \delta t = 0$? The solution I obtain for $u$, for general $\lambda > or < 0$ is $u(x,t) = \sum_n A_n sin(\lambda_n x) e^{-\lambda_n t}$. I could get your solution only by setting $\lambda = 0$ and getting eigenfunctions for that: $f(x) = c_1 x + c_2$ $\endgroup$
    – hirschme
    Dec 4 '18 at 16:08
  • 1
    $\begingroup$ If $\partial u/\partial t=0$, the 'remaining' equation is $\partial^2 u/\partial x^2=0$. Furthermore, since there is no dependence on $t$ anymore, we can change the partial derivative to total derivative, i.e., $$\frac{d^2 u}{dx^2}=0.$$ Integrating twice, we have $u=c_1 x+c_2$. See that this solution does not correspond to the solution for the original equation for any time $t$, but is the solution of the steady state equation. One can even say that, if $u$ does satisfy $$\frac{\partial u}{\partial t} = k \frac{\partial ^2 u}{\partial x^2},$$ then $$\lim_{t\to \infty} u(x,t) = c_1x+c_2.$$ $\endgroup$
    – rafa11111
    Dec 4 '18 at 16:19

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.