\begin{align*} \frac{d^{2}\phi }{dx^{2}}+\left ( \lambda -x^{2}\right ) \phi & =0\\ \phi ^{\prime }\left ( 0\right ) & =0\\ \phi \left ( 1\right ) & =0 \end{align*}
Putting the equation in the form \frac{d^{2}\phi }{dx^{2}}-x^{2}\phi =-\lambda \phi
Now the Rayleigh quotient is \lambda =\frac{-\left ( p\phi \phi ^{\prime }\right ) _{0}^{1}+\int _{0}^{1}p\left ( \phi ^{\prime }\right ) ^{2}-q\phi ^{2}dx}{\int _{0}^{1}\sigma \phi ^{2}dx}
Hence \lambda _{1}\leq 2.1429
\begin{align*} \frac{d^{2}\phi }{dx^{2}}+\left ( \lambda -x^{2}\right ) \phi & =0\\ \phi ^{\prime }\left ( 0\right ) & =0\\ \phi ^{\prime }\left ( 1\right ) & =0 \end{align*}
Putting the equation in the form \frac{d^{2}\phi }{dx^{2}}-x^{2}\phi =-\lambda \phi
Now the Rayleigh quotient is \lambda =\frac{-\left ( p\phi \phi ^{\prime }\right ) _{0}^{1}+\int _{0}^{1}p\left ( \phi ^{\prime }\right ) ^{2}-q\phi ^{2}dx}{\int _{0}^{1}\sigma \phi ^{2}dx}
Problem
Consider eigenvalue problem \frac{d}{dr}\left ( r\frac{d\phi }{dr}\right ) =-\lambda r\phi ,0<r<1 subject to B.C. \left \vert \phi \left ( 0\right ) \right \vert <\infty (you may also assume \frac{d\phi }{dr} bounded). And \frac{d\phi }{dr}\left ( 1\right ) =0. (a) prove that \lambda \geq 0. (b) Solve the boundary value problem. You may assume eigenfunctions are known. Derive coefficients using orthogonality.
Notice: Correction was made to problem per class email. Book said to show that \lambda >0 which is error changed to \lambda \geq 0.
From the problem we see that p=r,q=0,\sigma =r. The Rayleigh quotient is\begin{align} \lambda & =\frac{-\left ( p\phi \phi ^{\prime }\right ) _{0}^{1}+\int _{0}^{1}p\left ( \phi ^{\prime }\right ) ^{2}-q\phi ^{2}dr}{\int _{0}^{1}\sigma \phi ^{2}dr}\nonumber \\ & =\frac{-\left ( r\phi \phi ^{\prime }\right ) _{0}^{1}+\int _{0}^{1}r\left ( \phi ^{\prime }\right ) ^{2}dr}{\int _{0}^{1}r\phi ^{2}dr} \tag{1} \end{align}
The term -\left ( r\phi \phi ^{\prime }\right ) _{0}^{1} expands to -\left ( \left ( 1\right ) \phi \left ( 1\right ) \phi ^{\prime }\left ( 1\right ) -\left ( 0\right ) \phi \left ( 0\right ) \phi ^{\prime }\left ( 0\right ) \right )
case 1 If \phi ^{\prime }\neq 0 then we are done. The numerator is positive and we conclude that \lambda >0.
case 2 If \phi ^{\prime }=0 then \phi is constant and this means \lambda =0 is possible hence \lambda \geq 0. Now we need to show \phi being constant is also possible. Since \phi ^{\prime }\left ( 1\right ) =0, then for \phi ^{\prime }=0 to be true everywhere, it should also be \phi ^{\prime }\left ( 0\right ) =0 which means \phi \left ( 0\right ) is some constant. We are told that \left \vert \phi \left ( 0\right ) \right \vert <\infty . Hence means \phi \left ( 0\right ) is constant is possible value (since bounded). Hence \phi ^{\prime }=0 is possible.
Therefore \lambda \geq 0. QED.
The ODE is
\begin{align} r\phi ^{\prime \prime }+\phi ^{\prime }+\lambda r\phi & =0\qquad 0<r<1\tag{1}\\ \left \vert \phi \left ( 0\right ) \right \vert & <\infty \nonumber \\ \phi ^{\prime }\left ( 1\right ) & =0\nonumber \end{align}
In standard form the ODE is \phi ^{\prime \prime }+\frac{1}{r}\phi ^{\prime }+\lambda \phi =0. This shows that r=0 is a regular singular point. Therefore we try \phi \left ( r\right ) =\sum _{n=0}^{\infty }a_{n}r^{n+\alpha }
Substituting back into the ODE gives\begin{align*} r\sum _{n=0}^{\infty }\left ( n+\alpha \right ) \left ( n+\alpha -1\right ) a_{n}r^{n+\alpha -2}+\sum _{n=0}^{\infty }\left ( n+\alpha \right ) a_{n}r^{n+\alpha -1}+\lambda r\sum _{n=0}^{\infty }a_{n}r^{n+\alpha } & =0\\ \sum _{n=0}^{\infty }\left ( n+\alpha \right ) \left ( n+\alpha -1\right ) a_{n}r^{n+\alpha -1}+\sum _{n=0}^{\infty }\left ( n+\alpha \right ) a_{n}r^{n+\alpha -1}+\lambda \sum _{n=0}^{\infty }a_{n}r^{n+\alpha +1} & =0 \end{align*}
To make all powers of r the same, we subtract 2 from the power of last term, and add 2 to the index, resulting in \sum _{n=0}^{\infty }\left ( n+\alpha \right ) \left ( n+\alpha -1\right ) a_{n}r^{n+\alpha -1}+\sum _{n=0}^{\infty }\left ( n+\alpha \right ) a_{n}r^{n+\alpha -1}+\lambda \sum _{n=2}^{\infty }a_{n-2}r^{n+\alpha -1}=0
But a_{0}\neq 0 (we always enforce this condition in power series solution), which implies \fbox{$\alpha =0$}
For n\geq 2, now all terms join in, and we get a recursive relation\begin{align*} \left ( n\right ) \left ( n-1\right ) a_{n}r^{n-1}+\left ( n\right ) a_{n}r^{n-1}+\lambda a_{n-2}r^{n-1} & =0\\ \left ( n\right ) \left ( n-1\right ) a_{n}+\left ( n\right ) a_{n}+\lambda a_{n-2} & =0\\ a_{n} & =\frac{-\lambda a_{n-2}}{\left ( n\right ) \left ( n-1\right ) +n}\\ & =\frac{-\lambda }{n^{2}}a_{n-2} \end{align*}
For example, for n=2, we get a_{2}=\frac{-\lambda }{2^{2}}a_{0}
From tables, Bessel function of first kind of order zero, has series expansion given by \begin{align} J_{o}\left ( z\right ) & =\sum _{n=0}^{\infty }\frac{\left ( -1\right ) ^{n}}{\left ( n!\right ) ^{2}}\left ( \frac{z}{2}\right ) ^{2n}\nonumber \\ & =1-\left ( \frac{z}{2}\right ) ^{2}+\frac{1}{\left ( 2\right ) ^{2}}\left ( \frac{z}{2}\right ) ^{4}-\frac{1}{\left ( \left ( 2\right ) \left ( 3\right ) \right ) ^{2}}\left ( \frac{z}{2}\right ) ^{6}+\cdots \nonumber \\ & =1-\frac{z^{2}}{2^{2}}+\frac{1}{2^{2}4^{2}}z^{4}-\frac{1}{2^{2}3^{2}2^{6}}z^{6}+\cdots \nonumber \\ & =1-\frac{z^{2}}{2^{2}}+\frac{z^{4}}{2^{2}4^{2}}-\frac{z^{6}}{2^{2}4^{2}6^{2}}+\cdots \tag{3} \end{align}
By comparing (2),(3) we see a match between J_{o}\left ( z\right ) and \phi \left ( r\right ) , if we let z=\sqrt{\lambda }r we conclude that \fbox{$\phi _1\left ( r\right ) =a_0J_0\left ( \sqrt{\lambda }r\right ) $}
We can now normalized the above eigenfunction so that a_{0}=1 as mentioned in class. But it is not needed. The above is the first solution. We now need second solution. For repeated roots, the second solution will be \phi _{2}\left ( r\right ) =\phi _{1}\left ( r\right ) \ln \left ( r\right ) +r^{\alpha }\sum _{n=0}^{\infty }b_{n}r^{n}
The boundedness condition has eliminated the second solution altogether. Now we apply the second boundary conditions \phi ^{\prime }\left ( 1\right ) =0 to find allowed eigenvalues. Since \phi ^{\prime }\left ( r\right ) =-cJ_{1}\left ( \sqrt{\lambda }r\right )
or\begin{align*} \lambda _{1} & =14.682\\ \lambda _{2} & =49.219\\ \lambda _{3} & =103.5\\ & \vdots \end{align*}
Hence \begin{align} \phi _{n}\left ( r\right ) & =c_{n}J_{0}\left ( \sqrt{\lambda _{n}}r\right ) \tag{4}\\ \phi \left ( r\right ) & =\sum _{n=1}^{\infty }\phi _{n}\left ( r\right ) \nonumber \\ & =\sum _{n=1}^{\infty }c_{n}J_{0}\left ( \sqrt{\lambda _{n}}r\right ) \nonumber \end{align}
To find c_{n}, we use orthogonality. Per class discussion, we can now assume this problem was part of initial value problem, and that at t=0 we had initial condition of f\left ( r\right ) , therefore, we now write
f\left ( r\right ) =\sum _{n=1}^{\infty }c_{n}J_{0}\left ( \sqrt{\lambda _{n}}r\right )
Where \Omega is some constant. Therefore c_{n}=\frac{\int _{0}^{1}\sum _{n=1}^{\infty }f\left ( r\right ) J_{0}\left ( \sqrt{\lambda _{m}}r\right ) rdr}{\Omega }
From textbook, equation 5.9.8, we are given that for large \lambda \begin{align} \phi \left ( x\right ) & \approx \left ( \sigma p\right ) ^{\frac{-1}{4}}\exp \left ( \pm i\sqrt{\lambda }\int _{0}^{x}\sqrt{\frac{\sigma \left ( t\right ) }{p\left ( t\right ) }}dt\right ) \nonumber \\ & =\left ( \sigma p\right ) ^{\frac{-1}{4}}\left ( c_{1}\cos \left ( \sqrt{\lambda }\int _{0}^{x}\sqrt{\frac{\sigma \left ( t\right ) }{p\left ( t\right ) }}dt\right ) +c_{2}\sin \left ( \sqrt{\lambda }\int _{0}^{x}\sqrt{\frac{\sigma \left ( t\right ) }{p\left ( t\right ) }}dt\right ) \right ) \tag{1} \end{align}
Where c_{1},c_{2} are the two constants of integration since this is second order ODE. For \phi \left ( 0\right ) =0, the integral \int _{0}^{x}\sqrt{\frac{\sigma \left ( t\right ) }{p\left ( t\right ) }}dt=\int _{0}^{0}\sqrt{\frac{\sigma \left ( t\right ) }{p\left ( t\right ) }}dt=0 and the above becomes\begin{align*} 0 & =\phi \left ( 0\right ) \\ & =\left ( \sigma p\right ) ^{\frac{-1}{4}}\left ( c_{1}\cos \left ( 0\right ) +c_{2}\sin \left ( 0\right ) \right ) \\ & =c_{1}\left ( \sigma p\right ) ^{\frac{-1}{4}} \end{align*}
Hence c_{1}=0 and (1) reduces to \phi \left ( x\right ) =c_{2}\left ( \sigma p\right ) ^{\frac{-1}{4}}\sin \left ( \sqrt{\lambda }\int _{0}^{x}\sqrt{\frac{\sigma \left ( t\right ) }{p\left ( t\right ) }}dt\right )
Since \phi ^{\prime }\left ( L\right ) =0 then 0=c_{2}\left ( \sigma p\right ) ^{\frac{-1}{4}}\sqrt{\frac{\lambda \sigma }{p}}\cos \left ( \sqrt{\lambda }\int _{0}^{L}\sqrt{\frac{\sigma \left ( t\right ) }{p\left ( t\right ) }}dt\right )
\phi ^{\prime \prime }+\lambda \left ( 1+x\right ) \phi =0
Now, from textbook, equation 5.9.8, we are given that for large \lambda \begin{align} \phi \left ( x\right ) & \approx \left ( \sigma p\right ) ^{\frac{-1}{4}}\exp \left ( \pm i\sqrt{\lambda }\int _{0}^{x}\sqrt{\frac{\sigma \left ( t\right ) }{p\left ( t\right ) }}dt\right ) \nonumber \\ & =\left ( \sigma p\right ) ^{\frac{-1}{4}}\left ( c_{1}\cos \left ( \sqrt{\lambda }\int _{0}^{x}\sqrt{\frac{\sigma \left ( t\right ) }{p\left ( t\right ) }}dt\right ) +c_{2}\sin \left ( \sqrt{\lambda }\int _{0}^{x}\sqrt{\frac{\sigma \left ( t\right ) }{p\left ( t\right ) }}dt\right ) \right ) \tag{1} \end{align}
Where c_{1},c_{2} are the two constants of integration since this is second order ODE. For \phi \left ( 0\right ) =0, the integral \int _{0}^{x}\sqrt{\frac{\sigma \left ( t\right ) }{p\left ( t\right ) }}dt=\int _{0}^{0}\sqrt{\frac{\sigma \left ( t\right ) }{p\left ( t\right ) }}dt=0 and the above becomes\begin{align*} 0 & =\phi \left ( 0\right ) \\ & =\left ( \sigma p\right ) ^{\frac{-1}{4}}\left ( c_{1}\cos \left ( 0\right ) +c_{2}\sin \left ( 0\right ) \right ) \\ & =c_{1}\left ( \sigma p\right ) ^{\frac{-1}{4}} \end{align*}
Hence c_{1}=0 and (1) reduces to \phi \left ( x\right ) =c_{2}\left ( \sigma p\right ) ^{\frac{-1}{4}}\sin \left ( \sqrt{\lambda }\int _{0}^{x}\sqrt{\frac{\sigma \left ( t\right ) }{p\left ( t\right ) }}dt\right )
Hence for non-trivial solution we want, for large positive integer n\begin{align*} \sqrt{\lambda }\int _{0}^{1}\sqrt{\frac{\sigma \left ( t\right ) }{p\left ( t\right ) }}dt & =n\pi \\ \sqrt{\lambda } & =\frac{n\pi }{\int _{0}^{1}\sqrt{\frac{\sigma \left ( t\right ) }{p\left ( t\right ) }}dt}\\ & =\frac{n\pi }{\int _{0}^{1}\sqrt{1+t}dt} \end{align*}
But \int _{0}^{1}\sqrt{1+t}dt=1.21895, hence\begin{align*} \sqrt{\lambda } & =\frac{n\pi }{1.21895}=2.5773n\\ \lambda & =6.6424n^{2} \end{align*}
Therefore, solution for large \lambda is\begin{align*} \phi \left ( x\right ) & =c_{2}\left ( \sigma p\right ) ^{\frac{-1}{4}}\sin \left ( \sqrt{\lambda }\int _{0}^{x}\sqrt{\frac{\sigma \left ( t\right ) }{p\left ( t\right ) }}dt\right ) \\ & =c_{2}\left ( \sigma p\right ) ^{\frac{-1}{4}}\sin \left ( 2.5773n\int _{0}^{x}\sqrt{\frac{\sigma \left ( t\right ) }{p\left ( t\right ) }}dt\right ) \\ & =c_{2}\left ( 1+x\right ) ^{\frac{-1}{4}}\sin \left ( 2.5773n\int _{0}^{x}\sqrt{1+t}dt\right ) \\ & =c_{2}\left ( 1+x\right ) ^{\frac{-1}{4}}\sin \left ( 2.5773n\left ( -\frac{2}{3}+\frac{2}{3}\left ( 1+x\right ) ^{\frac{3}{2}}\right ) \right ) \end{align*}
To plot this, let us assume c_{2}=1 (we have no information given to find c_{2}). What value of n to use? Will use different values of n in increasing order. So the following is plot of \phi \left ( x\right ) =\left ( 1+x\right ) ^{\frac{-1}{4}}\sin \left ( 2.5773n\left ( -\frac{2}{3}+\frac{2}{3}\left ( 1+x\right ) ^{\frac{3}{2}}\right ) \right )