0

I want to solve a seemingly simple eigenvalue problem. I have a fixed set of boundary conditions given and want to change the complex parameter omega in to minimize exponentially falling solutions for some large r value. However, for the regime of omega, in which I expect to get exponentially falling solutions, I only obtain exponentially growing solutions. Therefore, I can't even attempt to minimize the result at some finite but large r.

rstart = 2 + 0.001;
f[r_] := 1 - 1/r
eq = f[r] D[f[r] D[R[r], r],r] - (w^2 + f[r] ((l (l + 1))/r^2 + m^2))R[r] == 0 /. {l -> 1,m -> 4/10, w -> wr + I wi}
sol = ParametricNDSolve[{eq, R[rstart] == 1, R'[rstart] == 1},R, {r, rstart, 100}, {wr, wr}]
test = Table[Log[Abs[R[wr, wi][r]]] /. sol, {wr, 0.88, 1, 0.01}, {wi, 10^-4,10^-1, 10^-2}];
Plot[test, {r, rstart, 100}]

I have also looked at other people's posts, though, there the minimization was done with respect to the initial conditions, which are fixed in my case. I have also tried the same with an increased MaxStepSize, but also that was unsuccessful.

xzczd
  • 65,995
  • 9
  • 163
  • 468
Nils
  • 127
  • 3
  • Probably you have an implicit BC of the solution being zero at $ r = \infty $, right? Just force it to be zero at the end of your range. – b3m2a1 Mar 18 '19 at 17:16
  • But how would I force it to be zero, if I have the given initial conditions and want to determine w, by setting the solution to zero at infinity? – Nils Mar 18 '19 at 17:24
  • 1
    "However, for the regime of omega, in which I expect to get exponentially falling solutions, I only obtain exponentially growing solutions. " Then are you sure the underlying model itself is correct? – xzczd Mar 22 '19 at 04:25
  • 2
    @xzczd: it's possible that the OP's DE does have a growing and a decaying solution, but the condition of the problem is bad enough that integrating forward yields a linear combination of the growing and decaying solution, which is of course itself growing. – J. M.'s missing motivation Mar 22 '19 at 12:09
  • 3
    The ODE has the asymptotic limit R''[r] - (w^2 + m^2} R[r] == 0, which will have exponential or oscillatory solutions, depending on the value of (w^2 + m^2}. – bbgodfrey Mar 22 '19 at 12:17
  • @J.M.isslightlypensive Do you mean something similar to this? – xzczd Mar 22 '19 at 12:38
  • @xzczd, like that, yes. Similar things happen when trying to numerically integrate the modified Bessel DE, where any inexact computation will have $K$ swamping out any of $I$'s contributions in the long run. – J. M.'s missing motivation Mar 22 '19 at 12:41
  • So the numerical solution is leaking into the exponentially growing branch, which then starts to dominate at some point? But then I should at least obtain a exponentially decaying solution in the beginning (given the initial data and parameters would give an exponentially decaying solution). – Nils Mar 22 '19 at 15:40

0 Answers0