The problem is to estimate the lowest eigenvalue of the following boundary value problem:

$$ u_{xx} + \lambda u = 0, \ \ \ \ x \in [0,1], \ \ \ \ u(0)=0, \ \ u_x(1)+u(1) = 0. $$

We are supposed to use the Rayleigh-Ritz method, with a trial function of the form $u=x^2 + \alpha x$.


So far I have found the trial function to: $u(x)=x^2-1.5x$, by applying boundary conditions to the trial function.

My problem arises because when I use the Rayleigh Ritz method, $$F(u(x))=\frac{\int_0^1 p(x)u_x^2-q(x)u^2dx}{\int_0^1w(x)u^2dx}=\frac{\int_0^1u_x^2dx}{\int_0^1u^2dx} =\frac{\int_0^1(2x-1.5)^2dx}{\int_0^1(x^2-1.5x)^2dx}=2.92$$ I get an eigenvalue which is smaller than the exact value, $\lambda_{exact}=4.12$, which can't be right? Does it have something to do with a boundary term $[p(x)u_xu]_0^1$ somewhere not being equal to zero?


Yes, you're absolutely right. Since $[p(x)u_x u]_0^1 \neq 0$, you need to estimate the lowest eigenvalue using this formula: $$F[u(x)] = \frac{\int_0^1 \left[- u\frac{d}{dx} \left(p \frac{du}{dx}\right) - qu^2\right]dx}{\int_0^1 w u^2 dx} = \frac{-\int_0^1 u u_{xx} dx}{\int_0^1 u^2 dx}$$ [You can verify that this expression does indeed give an upper bound on the lowest eigenvalue by writing $u$ as a linear combination of (orthogonal) eigenfunctions - presumably this is done in your textbook.]

At first glance, you might believe that the expression I wrote down is equivalent to what you wrote down, by a familiar integration-by-parts trick. But this is not true! If you try to integrate the numerator by parts, you will pick up the non-zero term $-[p(x)u_x u]_0^1$.

Anyway, I quickly calculated this on Mathematica and got an answer of $4.17$, which is very close to the real answer!


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.