I have the operator $$ A : -e^{-2ax} \frac{\partial}{\partial x} \left(e^{2ax}\frac{\partial}{\partial x}\right)\\ D_A = \left( v \in C^2[0,L] \quad | \quad v(0) = v(L) = 0 \right) $$ and I need to find the eigenfunction $p$. But first I have a question. How do I apply this operator to a function $u$? I see no $u$ in the formula. Is it a typo, should it be $\frac{\partial u }{\partial x}$ in the last partial derivative within the parenthesis? If this is the case, this is my attempt: $$Ap = \lambda p$$ $$-2ap' - p'' - \lambda p = 0$$ $$p'' + 2ap' + \lambda p = 0$$ $$p(x) = C_1 e^{-a + \sqrt{a^2 - \lambda}} + C_2e^{-a - \sqrt{a^2 - \lambda}}.$$ But this doesn't at all look like the correct answer $$e^{-ax}\sin{\frac{k\pi x}{L}}, k=1,2,3,...$$ What am I doing wrong and how do I find the right answer (without complicating things)?


How do we apply an operator such as

$A = -e^{-2ax} \dfrac{\partial}{\partial x} \left (e^{2ax} \dfrac{\partial}{\partial x} \right ) \tag 0$

to a function $u$? Typically we work from right to left, taking the argument $u$ to occupy the first or right-most "empty slot"; bearing this in mind:

We start with the following calculation:

$Au = -e^{-2ax} \dfrac{\partial}{\partial x} \left (e^{2ax} \dfrac{\partial}{\partial x} \right )u = -e^{-2ax} \dfrac{\partial}{\partial x} \left (e^{2ax} \dfrac{\partial u}{\partial x} \right )$ $=-e^{-2ax} \left ( 2ae^{2ax} \dfrac{\partial u}{\partial x} + e^{2ax} \dfrac{\partial^2u}{\partial x_2} \right ) = -2a\dfrac{\partial u}{\partial x} - \dfrac{\partial u^2}{\partial x^2}; \tag 1$

if $\lambda$ is an eigenvalue of this operator, then

$-e^{-2ax} \dfrac{\partial}{\partial x} \left (e^{2ax} \dfrac{\partial}{\partial x} \right )u = \lambda u, \tag 2$

then, via (1),

$-2a\dfrac{\partial u}{\partial x} - \dfrac{\partial u^2}{\partial x^2} = \lambda u, \tag 3$


$\dfrac{\partial u^2}{\partial x^2} + 2a\dfrac{\partial u}{\partial x} + \lambda u = 0; \tag 4$

since this is a linear, constant coefficient, second order ordinary differential equation, we may find solutions of the general form $e^{\rho x}$ where $\rho$ is a root of the quadratic equation

$\rho^2 + 2a \rho + \lambda = 0; \tag 5$

we have

$\rho_\pm = \dfrac{-2a \pm \sqrt{4a^2 - 4 \lambda}}{2} = -a \pm \sqrt{a^2 - \lambda}; \tag 6$

thus we examine functions of the form

$u(x) = C_+ e^{\rho_+ x} + C_- e^{\rho_- x} \tag 7$

which satisfy

$u(0) = u(L) = 0; \tag 8$

we note that (7) and (8) yield

$C_+ + C_- = u(0) = 0, \tag 9$

so that

$C_+ = -C_-; \tag{10}$

we can thus take

$C_+ = C = -C_-, \tag{11}$

and write

$u(x) = C(e^{\rho_+ x} - e^{\rho_- x}); \tag{12}$

we next note that for $\lambda < a^2$ in (6), we obtain two distinct real $\rho_\pm$, and if we apply the condition

$u(x) = C(e^{\rho_+ L} - e^{\rho_- L}) = u(L) = 0, \tag{13}$

we see that with $C \ne 0$ it leads to

$e^{\rho_+ L} = e^{\rho_- L}, \tag{14}$

which is impossible if $\rho_+ \ne \rho_-$; therefore (4), (5) have no solutions in $D_A$ with $\lambda < a^2$; since solutions to (4) are eigenfunctions of the operator $A$, we see it has no eigenvalues $\lambda < a^2$. If $\lambda = a^2$, then we see via (6) that (5) has a repeated root $\rho = \rho_\pm = -a$; it is well-known the solutions in this case are $e^{\rho x} = e^{-ax}$ and $xe^{-ax}$; then

$u(x) = C_1 e^{-ax} + C_2 x e^{-ax} = (C_1 + C_2x) e^{-ax}, \tag{15}$


$C_1 = u(0) = 0, \tag{16}$

and hence

$u(x) = C_2 x e^{-ax}, \tag{17}$


$C_2 L e^{-aL} = u(L) = 0 \Longrightarrow C_2 = 0 \Longrightarrow u(x) = 0, \tag{18}$

eigenfunctions cannot be zero; thus we rule out the case $\lambda = a^2$.

We are left only with the possibility that

$\lambda > a^2; \tag{19}$

then we have

$\rho_\pm = -a \pm i \sqrt{\lambda - a^2}; \tag{20}$

formulas (7), (9) and (12) still hold so we still write

$u(x) = C(e^{\rho_+ x} - e^{\rho_- x}); \tag{21}$

with $\rho_\pm$ as in (20) we find

$e^{\rho_\pm x} = e^{-ax} e^{\pm \sqrt{\lambda - a^2} x} = e^{-ax}(\cos \sqrt{\lambda - a^2} x \pm i \sin \sqrt{\lambda - a^2} x), \tag{22}$

so that

$e^{\rho_+ x} - e^{\rho_- x} = 2i e^{-ax} \sin \sqrt{\lambda - a^2} x, \tag{23}$


$u(x) = 2C i e^{-ax} \sin \sqrt{\lambda - a^2} x; \tag{24}$

the criterion

$u(L) = 0 \tag{25}$


$\sin \sqrt{\lambda - a^2} L = 0, \tag{26}$


$\sqrt{\lambda - a^2} L = k\pi, \; 0 \ne k \in \Bbb Z \tag{27}$

(we preclude $k = 0$ to enforce $u(x) \ne 0$); then

$\lambda = a^2 + \dfrac{k^2 \pi^2}{L^2}, \; 0 \ne k \in \Bbb Z, \tag{28}$

and the eigenfunctions may be taken to be

$e^{-ax} \sin \sqrt{\lambda - a^2} x = e^{-ax} \sin \dfrac{x k\pi}{L}, 0 \ne k \in \Bbb Z. \tag{29}$


Hint: Your functions are $$ p(x) = C_1 e^{(-a + \sqrt{a^2 - \lambda})x} + C_2e^{(-a - \sqrt{a^2 - \lambda})x} $$ Consider how they behave when $\lambda > a^2$. Then consider what $\lambda$ and the $C_i$ need to be to satisfy the boundary conditions.


How do I apply this operator to a function $u$?

Your $D_A$ looks like the function space of all 2 times differentiable functions on $[0,L]$. So this seems like a function of one variable.

And therefore these seem ordinary differential operators, no partial ones, in the definition of $A$.

Thus I believe $A$ applies like this $$ \begin{align} A u &= -e^{-2ax} \partial_x \left(e^{2ax} \partial_x u \right) \\ &= -e^{-2ax}\left(e^{2ax} 2a \partial_x u + e^{2ax} \partial_x^2 u\right) \\ &= - (\partial_x^2 u + 2a \partial_x u)\\ \end{align} $$ to $u \in D_A$.

  • $\begingroup$ Okay yes, that's what I thought and used as well. But using it, my answer does not seem coherent with the correct answer $\endgroup$ – Heuristics Jul 25 '18 at 18:25
  • $\begingroup$ Then I am at a loss what kind of function a feasible $u$ is. $\endgroup$ – mvw Jul 25 '18 at 18:27
  • $\begingroup$ Is my method of finding the eigenfunction correct (although somewhat incomplete)? $\endgroup$ – Heuristics Jul 25 '18 at 18:28
  • $\begingroup$ And you have to apply the boundary conditions from the $D_A$ definition. This might reduce your exponential solution to that damped sine function. $\endgroup$ – mvw Jul 25 '18 at 18:30
  • $\begingroup$ Have a look at Universal oscillator. $\endgroup$ – mvw Jul 25 '18 at 18:37

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.