4
$\begingroup$

I am trying to solve the following nonlinear ODE for a function $P(x)$:

$$\left(1-x^2\right)P''(x)+k(k+1)P(x)=cP(x)^\frac{k-1}{k+1}.$$

Here, $k$ and $c$ are arbitrary parameters. By rescaling $P(x)$, one can without loss of generality set $c=1$, so really this equation is only parameterized by $k$.

I wonder if there exists a closed form solution for this equation, at least for integer $k$? Although it looks very simple, my efforts to solve it have so far been unsuccessful. Nonetheless, I have been able to find some very simple solutions: for instance,

  • when $k=1$, the equation can be integrated, resulting in

$$P(x)=\frac{1}{2}+a\left(1-x^2\right)+b\left[x+\left(1-x^2\right)\mathrm{arctanh}\,x\right],$$

  • when $k=-2$, by inspection I found

$$P(x)=\sqrt{2}\frac{4x}{3+x^2},$$

  • when $k=-3$, by inspection I found

$$P(x)=120\frac{1-x^2}{\left(5-x^2\right)^2}.$$

These solutions look simple enough that I'm hopeful that a general solution exists. Perhaps there even exists some standard transformation to linearize such equations?

$\endgroup$
1
  • $\begingroup$ Oh, I should add that the $c=0$ case is of course trivial: the equation becomes linear and the general solution is given by a superposition of two 2F1 hypergeometric functions. I am interested in the $c\neq0$ case. $\endgroup$ Jul 1, 2017 at 10:44

2 Answers 2

5
$\begingroup$

Using contact transformations, Lie showed that second order equations of this type can be transformed to the trivial equation $\frac{d^2 P^\prime}{dx^{\prime 2}} = 0$.

Edit: an algorithm for this is described in Kumei and Bluman When Nonlinear Differential Equations are Equivalent to Linear Differential Equations, SIAM J. Math. 42(5) (1982) 1157-1173.

Alternatively, a more general technique is described in Bluman and Dridi New Solutions for ordinary differential equations, J. Symb. Comp. 47 (2012) pp. 76-88. It uses a combination of invertible and non-invertible mappings. They treat several examples in detail.

Edit(2): The paper by N. Euler, Transformation Properties of $\ddot{x}+f_1(t)\dot{x}+f_2(t)x+f_3(t)x^n=0$ J. Nonlin. Math. Phys. 4(3-4) (1997) pp. 310-337 provides the conditions for an equation of this type to be integrable, as well as the conditions for it to have Lie symmetries. He also considers non-point transformations $X=F(x,t)$, $dT = G(x,t)dt$ to transform the equation to $\frac{d^2X}{dT^2}+k_1\frac{dX}{dT}+k_2X^p=0$ for constants $k_1$, $k_2$ and $p\in \mathscr{Q}$. I've tried substituting the corresponding values for your equation into his conditions, but the results underline your point about the lack of symmetries.

$\endgroup$
6
  • $\begingroup$ Thank you for your answer, Phil. Correct me if I'm wrong, but isn't finding the contact transformation in question just as hard as solving the equation directly? $\endgroup$ Jul 3, 2017 at 12:38
  • $\begingroup$ I am looking at the paper by Bluman and Dridi now. Once I understand how their Euler operator (18) acts, I'll apply their method to my equation. $\endgroup$ Jul 3, 2017 at 12:39
  • $\begingroup$ @AlexLupsasca It may be useful to look at Dridi's thesis as well - tel.archives-ouvertes.fr/tel-00264288. $\endgroup$ Jul 4, 2017 at 7:28
  • $\begingroup$ Hi Phil, I applied the method described in the paper by Bluman and Dridi. Unfortunately, it only produces the trivial solution to the linear equation that is obtained by setting c=0. This is because this restricts the equation to a subspace of solution space with enhanced symmetry--the situation is exactly analogous to the example they consider starting in the introduction. But thanks for mentioning the paper---their method is good to know! $\endgroup$ Jul 9, 2017 at 10:56
  • $\begingroup$ @AlexLupsasca Very disappointing. I've added a reference to my original answer on a linearisation procedure. $\endgroup$ Jul 10, 2017 at 11:51
3
$\begingroup$

For $k=0$, according to Maple the general solution is

$$ \text{arctanh}(x) \pm \int^{P(x)/\sqrt{x^2-1}} \dfrac{du}{\sqrt{u^2 - 2 c \ln(u) + b}} = a $$

Another family of solutions (for all $k$) is $$ P(x) = \left( {\frac {k \left( k+1 \right) }{c}} \right) ^{-(k+1)/2}{x}^{k+1 } $$

For $k = -3$, there are some other closed-form solutions, e.g.

$$ P(x) = \frac{8}{c(1-x^2)} $$ and $$ P(x) = \frac{24 (21 x^2-5)}{c (3x^2+5)^2}$$

$\endgroup$
5
  • $\begingroup$ Maxima happily confirmed the closed form solutions for $k=-3$, but didn't like the family of solutions for all $k$. $\endgroup$ Jul 2, 2017 at 9:48
  • $\begingroup$ There are some problems with branches when $x < 0$, but it should be OK for $x > 0$. $\endgroup$ Jul 2, 2017 at 21:41
  • $\begingroup$ Thank you Robert! Unfortunately, the general family of solutions you found is not useful for me, because I am really trying to solve another (much nastier) equation for a function $y(x)$. I simplified the equation by a transformation from $y(x)$ to $P(x)$ but the general solution you found goes to a constant under the inverse transformation :( $\endgroup$ Jul 3, 2017 at 12:42
  • $\begingroup$ That being said, I am curious how you found the other solutions? Perhaps the method can be generalized to other values of $k$? That would be almost as good as a general solution. I just tried to input the equation into Maple but I can't get it to spit out any of the answers you wrote, except for the $k=0$ one; is there a special argument that's required? Thank you once again! $\endgroup$ Jul 3, 2017 at 12:44
  • 2
    $\begingroup$ What I did was guess that the solution should be a rational function with numerator and denominator of particular degrees, substitute that into the differential equation, and solve for the coefficients. $\endgroup$ Jul 3, 2017 at 19:08

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.