3
$\begingroup$

Let $y''+fy=0$ be a second-order linear ode on $y$, where $f(x)>0$, and $I=\left[ a,b \right)$ be an interval. Suppose we want to estimate the number of zeros of a (not identically zero) solution of the ode on the interval $I$. Note that this question is well defined, because by the Sturm separation theorem, choosing a different solution can change the number of zers by at most 1.
It follows from the Sturm–Picone comparison theorem that if $0<m \leq f(x) \leq M$ on the interval ($m,M$ are numbers), and $N$ is the number of zeros of a solution on $I$, then $$\left \lfloor \frac{(b-a)\sqrt{m}}{\pi} \right \rfloor \leq N \leq \left \lceil \frac{(b-a)\sqrt{M}}{\pi} \right \rceil$$ Now, we can break $I$ into subintervals, and hopefully we get better estimates. A hopeful thought lead to hoping that the number of zeros on $I$ can be approximated by the integral $S=\int_{a}^{b} \frac{\sqrt{f(x)}}{\pi} dx$.

My question is, how good is this approximation? What can be said about the error $\left| N-S \right|$?

$\endgroup$
  • $\begingroup$ It looks like one always has $N\ge S-1$ but we can easily have $S\approx 0$ for any $N$ without extra assumptions. The second observation is trivial (sawtooth function with smoothed corners). I can try to check the first if it is of any interest for you. $\endgroup$ – fedja Jun 17 '14 at 21:11
  • $\begingroup$ Your formula is too sensitive to the precise shape of local singularities ($f\approx \delta$, as suggested by fedja). What matters here is $\int_I f$ over small intervals. Also, a very general precise formula would allow you to locate eigenvalues of $-y''+fy$ with high accuracy, so seems unlikely. $\endgroup$ – Christian Remling Jun 18 '14 at 5:13
  • $\begingroup$ $N\ll S$ is possible too: we can take $f(x)=cx^{-2}$ on $(1,\infty)$, where we must choose $c>0$ small enough so that the equation becomes non-oscillatory. See en.wikipedia.org/wiki/…. So any solution of $y''+fy$ has at most $N$ zeros on $(1,b)$, with $N$ independent of $b$, but of course we can make $S(b)$ arbitrarily large. $\endgroup$ – Christian Remling Jun 18 '14 at 5:33
  • 1
    $\begingroup$ In physics the approximation of the number of zeroes in $I$ by $S$ is known as WKB (or sometimes JWKB) approximation. Google "WKB approximation". The Wikipedia article is a good starting point. $\endgroup$ – Johannes Trost Jun 18 '14 at 11:23
0
$\begingroup$

Here is a specific example with a large $S$ constructed by starting with $y$. Consider $$y=\sqrt[3]{sin(x)}$$ for the interval $[\delta,\pi-\delta]$ with $0 \lt \delta$ quite small. enter image description here

Actually, an interval like $[\delta,1]$ is adequate.

The point is that $y''+fy=0$ sets $$f=\frac{2+\sin^2(x)}{9\sin^2{x}} \gt\frac{2}{9x^2}$$ so $$\int_{\delta}^1\sqrt{f}dx \gt \frac{\sqrt{2}}{3}(-\ln{\delta})$$ which can be made arbitrarily large by choosing $\delta$ sufficiently small.

I haven't worked out any details, but I'm sure that this could be made into an oscillating example (with arbitrarily large $|S-N|$ ) by some adjustment on the intervals $[k\pi-\delta,k\pi+\delta].$

$\endgroup$
  • $\begingroup$ The WKB approximation of the number of zeroes in $I$ with (the so called action integral) $S$ is good, if $|df/dx|\ll f^2$ in the interval $I$, which is not the case for your example. See my comment on the question. $\endgroup$ – Johannes Trost Jun 18 '14 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.