Given the inhomogeneous ODE on the form: $u''(x)-Cu(x)=-f(x)$

where $0 < x < 1, C > 0$ and $f(x)$ is cont. on the interval for x.

With IC: $u(0) = u(1) = 0$

Solving that yields the general solution for the homogeneous part: $u_c = Ae^\sqrt{C} + Be^{-\sqrt{C}}$

Inserting initial conditions yields the equality $\sqrt{C} = -\sqrt{C}$ (which can't be since C is strictly positive, right?) since we find that $A = -B$

Also how would one go on finding a particular solution when the form of f(x) is unknown?


If discretizing this system would that be a valid method to prove this has a unique solution, or would that only be valid for the discrete system?

Analytically i guess we would use Variation of parameters as follows:

A solution would be of the form: $y = y_c + y_p$

An attempt at an solution using variation of parameters yield:

$y_p = y_1u_1+y_2u_2$

$y_1 = e^{(\sqrt{C})x}$

$y_2 = e^{-(\sqrt{C})x}$

$f = f(x)$

$u_1 = \int \dfrac{-y_2 f}{W(y_1,y_2)} dx = -\int \dfrac{e^{-(\sqrt{C})x}f}{1} dx = -\int {e^{-(\sqrt{C})x}f} dx $

$u_2 = \int \dfrac{y_1 f}{W(y_1,y_2)} dx = \int \dfrac{e^{(\sqrt{C})x}f}{1} dx = \int {e^{(\sqrt{C})x}f} dx $

Where (W = Wronskian determinant):

$W(y_1,y_2) = e^{(\sqrt{C}-\sqrt{C})x} = 1$

A solution would be of the form:

$y = y_c + y_p$

$y_c = 0$ so:

$y = y_p = y_1u_1+y_2u_2 = -e^{(\sqrt{C})x} \int {e^{-(\sqrt{C})x}f} dx + e^{-(\sqrt{C})x} \int {e^{(\sqrt{C})x}f} dx $

Inserting for IC yields (all the expontentials cancel out):

$y(0) = y_p = -\int {f(0)} dx + \int {f(0)} dx = 0$

$y(1) = y_p = - \int {f(1)} dx + \int {f(1)} dx = 0$

So $y = 0$ and does this tell us that the solution is unique? But is 0 a valid solution? This solution differs from the Green kernel method in the answer below, so this solution appears to be wrong.

Now this seems problematic, now am I using the wrong technique or using it wrongly?

  • $\begingroup$ In the inhomogeneous equation, the coefficients of the homogeneous part of the solution are again indeterminate. You can write it as $$y =-e^{(\sqrt{C})x} \Bigl(c_1+\int_0^x e^{-(\sqrt{C})s}f(s) ds\Bigr) + e^{-(\sqrt{C})x} \Bigl(c_2+\int_0^x {e^{(\sqrt{C})s}f(s)} ds\Bigr)$$ and $c_1,c_2$ are to be determined by the boundary conditions. $\endgroup$ – LutzL Nov 6 '15 at 16:03
  • $\begingroup$ The integrals cancel each other out, right, or how does one solve these integrals? And y is a function of x: $y = y(x)$ With $x = 0$ each integral equals 0 $\endgroup$ – jibo Nov 7 '15 at 15:49
  • $\begingroup$ So then we know at least that $c_1=c_2$ $\endgroup$ – jibo Nov 7 '15 at 15:55

It should of course be $$ u_c(x) = Ae^{\sqrt{C}x} + Be^{-\sqrt{C}x} $$ giving you the conditions $$ A+B=0\\ Ae^{\sqrt{C}} + Be^{-\sqrt{C}}=0 $$ which indeed tells you that there are no solution except the trivial solution $u_c\equiv 0$.

For the inhomogeneous equation you could either use variation of constants or use the Green kernel to express the same. For the latter, find the solutions with $u(0)=0=v(1)$ and $u'(0)=1=v'(1)$ which are $$ u(x)=\frac1{\sqrt C}\sinh(\sqrt Cx),\qquad v(x)=\frac1{\sqrt C}\sinh(\sqrt C(x-1)) $$ resulting in a Green function $$ G(x,y)=-\frac1C·\sinh(\sqrt C\min(x,y))·\sinh(\sqrt C(1-\max(x,y))) $$ for the particular solution $$ u_p(x)=\int_0^1 G(x,y)f(y)\,dy $$

  • $\begingroup$ Thanks! Not familiar with Green kernel, so trying to use the variation of constants technique (even though Green-method seems much shorter), which i guess is the same as variation of parameters. I'm unfamiliar with the last one, but i believe it's the method I'm supposed to use. My attempt at it however doesn't seem correct. $\endgroup$ – jibo Nov 6 '15 at 11:20

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.