3
$\begingroup$

This is a similar question to this but with a different ODE.

Let $f$ be a continuous function in $L^2([1,\infty)$ satisfying $\sup_{r\geq 1} r|f(r)| <\infty$. Let $\ell$ be a positive integer, $R>1$, and $a$ be a real number.

Let $u(r)$ be a function on $[R,\infty)$ solving the following IVP: $$(r^2-1)u''(r)+2ru'(r) - \ell(\ell+1) u(r) = f(r), \quad r\in [R,\infty)$$ $$u(R)= a, \quad \lim_{r\to \infty} u(r) = 0$$

We should be able to estimate the solution $u$ by $f$ and $a$. I want to understand precisely how $\ell$ will appear in this estimate.

I am speculating the following: there exists a constant $C$ independent of $\ell$, $a$, $u$ and $f$ satisfying $$\ell \sup_{r\geq R}r|u(r)| + \sup_{r\geq R}r^2|u'(r)| \leq C(\ell^{-1}\sup_{r\geq R} r|f(r)| + \ell^{-1}\lVert f\rVert_{L^2} + \ell|a|)$$

Can anyone verify if this is true? How would I prove this estimate (or the correct one in case the above is incorrect)?

The solutions to the homogeneous problem are the Legendre polynomials of the first and second kind $P_{\ell}$, $Q_{\ell}$. We can use the method variation of parameters to get

$$u(r) = AQ_{\ell}(r) + P_{\ell}(r) \int_{r}^{\infty} Q_{\ell}(t) f(t)W(t)^{-1}(t^2-1)^{-1}dt + Q_{\ell}(r) \int_{R}^r P_{\ell}(t) f(t) W(t)^{-1}(t^2-1)^{-1} dt$$

where $W = P_{\ell}Q_{\ell}'-P_{\ell}'Q_{\ell}$ and $A$ is some constant that will depend on $a$. I have tried to use this formula to get the estimate but couldn't.

Any help is appreciated.

EDIT: Using the Frobenius method we get,

$$P_{\ell}(r) = \sum_{n=0}^{\ell} a_n r^{\ell-n}, \quad Q_{\ell}(r) = \sum_{n=0}^{\infty}b_n r^{-\ell-1-n}$$ where $a_n$, $b_n$ are recursively defined by $a_0 = b_0 = 1$ (by choice), $a_1 = b_1 = 0$, and for $n\geq 2$ $$a_n = \frac{(\ell-n+1)(\ell-n+2)}{n^2-n(2\ell+1)} a_{n-2}, \quad b_n = \frac{(\ell+n-1)(\ell+n)}{n(n+2\ell+1)}b_{n-2}$$

Then the Wronskian satisfies: $$W(t)(t^2-1) = 2\ell+1$$ in which I used Lagrange's identity to deduce that $W(t)(t^2-1)$ is a constant and took the limit as $t$ goes to $\infty$ to find the constant.

We would then like to get the following estimates:

$$\left| rP_{\ell}(r) \int_{r}^{\infty} Q_{\ell}(t) f(t)dt \right|\leq \frac{C}{\ell} \sup_{s\geq R}s|f(s)|$$ $$\left| rQ_{\ell}(r) \int_{R}^r P_{\ell}(t) f(t)dt \right|\leq \frac{C}{\ell} \sup_{s\geq R}s|f(s)|$$ for some $C$ independent of $\ell$. This will then imply $ \sup_{r\geq 1}r|u(r)| \leq C\ell^{-2}\sup_{r\geq R} r|f(r)|$, which is part of what we're trying to show.(Note that the second factor of $\ell$ in the right side comes from $W$).

It will be sufficient if we have estimates on $Q_{\ell}$ and $P_{\ell}$ looking like:

$$(1) \qquad \sup_{r\geq R}r^{\ell+1}|Q_{\ell}(r)| \leq C, \quad \sup_{r\geq R}r^{-\ell} |P_{\ell}(r)| \leq C$$

Where $C$ is independent of $\ell$ (it can depend on $R$). If so, then the estimate I'm trying to get follows using the same calculation @IgorKhavkine performed.

However, I haven't been able to get those estimates in equation (1).

I think I could use the recursive relation: $$(\ell+1)P_{\ell+1}(r) = (2\ell+1)r P_{\ell}(r) - \ell P_{\ell-1}$$ to get the desired estimate on $P_{\ell}$.

I have tried to make use of the identity: $$Q_{\ell}(r) = B\int_{0}^{\infty} (r+(r^2-1)^{1/2}\cosh \theta)^{-\ell-1} d\theta$$ where $B^{-1} = \int_{0}^{\infty} (1+\cosh \theta)^{-\ell-1} d\theta$ ( added so that it agrees with the expression of $Q_{\ell}$ written earlier). But I couldn't make use of this to get the desired estimate on $Q_{\ell}$. I would need to bound the expression $$\frac{\int_{0}^{\infty} (1+(1-R^{-2})^{1/2}\cosh \theta)^{-\ell-1} d\theta}{\int_{0}^{\infty} (1+\cosh \theta)^{-\ell-1} d\theta}$$ by a constant independent of $\ell$. But it seems too difficult.

$\endgroup$
7
  • 1
    $\begingroup$ Your variation of constants formula is wrong. It should be $$ u(r) = P_\ell(r) \int_r^\infty Q_\ell(t) f(t) W(t)^{-1} (t^2 - 1)^{-1} ~dt + Q_\ell(r) \int_R^r P_\ell(t) f(t) W(t)^{-1} (t^2 - 1)^{-1} ~dt $$ With the extra decay inserted I think exactly the same calculation that Igor performed for you in your previous question should work. (I suspect you got your formula from the situation where the equation is normalized so that $u''$ has coefficient $1$.) $\endgroup$ Commented May 2 at 7:19
  • 1
    $\begingroup$ In addition, by Lagrange's identity, $(r^2-1) W(r)$ is simply a constant. What you are still missing are some asymptotic estimates on $P_\ell(r), Q_\ell(r)$ for large $r$, which you can get from the Frobenius method. Otherwise, it's just as Willie said $\endgroup$ Commented May 2 at 7:59
  • $\begingroup$ I have tried using Frobenius method and got all the coefficients, but the expression was very. But I didn't know that $(r^2-1)W(r)$ is a constant! This simplifies the expression significantly since there is no sum in the denominator! I will try this today. Thank you Willie and Igor. $\endgroup$
    – Laithy
    Commented May 2 at 16:41
  • $\begingroup$ @IgorKhavkine I have tried what you recommended. The required estimates we need on $P_{\ell}$ and $Q_{\ell}$ seem to be difficult unless I'm missing something obvious. I edited the post to show my progress. $\endgroup$
    – Laithy
    Commented May 3 at 16:50
  • 1
    $\begingroup$ The Frobenius method gives you asymptotic estimates as $r\to\infty$ for fixed $\ell$, so you already have them. It looks like the sticky point for you is the dependence on $\ell$, meaning that you should look for bounds like $P_\ell(r) \le C r^\ell$ and $Q_\ell(r) \le C r^{-\ell-1}$ that are uniform in $\ell$ on $r\in [1,\infty)$. While there are methods to get such bounds directly from the differential equation (e.g., WKB uniform approximation) it will probably be easier to find them in the literature. Have a look for instance at MO304693. $\endgroup$ Commented May 4 at 7:28

1 Answer 1

0
$\begingroup$

The needed bounds on the Legendre functions $P_{\ell}$, $Q_{\ell}$ are proven here: https://arxiv.org/abs/2411.02801 See appendix A.3

$\endgroup$

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .