I'm faced with the following nonlinear 2nd order system of ODEs: $$ \phi''(r)+\frac{4r^3-1}{r^4-r}\phi'(r)+\frac{r^2 h(r)^2+2r(r^3-1)}{(r^3-1)^2}\phi(r)=0, \\ h''(r)+\frac{2}{r}h'(r)-\frac{2r\phi(r)^2}{r^3-1}h(r)=0. $$ Here, $\phi$ and $h$ are real-valued functions of $r\in[1,\infty)$ and $h(1)=0$ and $\phi(1)$ is finite and nonzero.

I've been able to find numerical solutions of this system. I did so by assuming $h$ and $\phi$ can be written as a power series around $1$ and by imposing initial conditions at $r=1.00001$ and integrating to larger $r$.

However, I am wondering now if the solutions I found are unique, because the system is singular at the point $r=1$.

So my question is: for this specific system, is it possible that there exist singular solutions (by which I mean that they are not uniquely fixed by the initial value problem at $r=1.00001$)? Or are the solutions I can find uniquely fixed by the initial values because I impose them at $r=1.00001$ rather than at $r=1$? Note: I am only looking for solutions which in fact do behave analytically at $r=1$!

Any reference to existence theorems relevant for systems like the one above is also appreciated!

  • $\begingroup$ The method you have used (looking for coefficients assuming analytic behavior of your functions at your origin) has limited range. Have you tried a differential system solver as exist on many systems, like ode43 on Matlab (based on Runge Kutta) in order to gain a better understanding of it ? I understand that you are looking for theoretical guarantees, but, as your system is fairly complicated, it is unlikely that good conditions will be found... $\endgroup$
    – Jean Marie
    Feb 7 '16 at 22:36
  • $\begingroup$ @JeanMarie I am looking only for solutions which should behave analytically at my origin. Does that restrict the uniqueness some more? My numerical skills are limited unfortunately, I only used Mathematica, but I will look up the method you mentioned if this can help me. $\endgroup$ Feb 8 '16 at 8:48
  • $\begingroup$ Also, theoretical guarantees of existence of singular solutions would be nice, but any statement like "it is not immediately clear that singular solutions cannot exist" also helps. $\endgroup$ Feb 8 '16 at 8:49

I have made a Matlab implementation of your system (script below) to get approximate solutions.

a) If $h(1)=0$ (your hypothesis), function $h$ is found by the solver to be identically zero. This is one of the singular solution you are looking for... even it is rather trivial, isn't it ? Fonctions $\phi$ have are decreasing steadily, in a smooth manner.

I have thus attempted to solve the first equation by replacing $h(r)$ by $0$.

I have turned to Mathematica for solving it;

DSolve[p''[r]+(4r^3-1)p'[r]/(r(r^3 - 1))+2 r p[r]/(r^3-1) == 0, p[r], r]

gives two exact independent solutions :

$p_1[r]= \ $ Hypergeometric2F1[1/3, 2/3, 1, r^3] and

$p_2[r]= \ $ MeijerG[{{}, {1/3, 2/3}}, {{0, 0}, {}}, r^3]}}

(I never met any Meijer before !).

The general solution is thus a linear combination of these two.

BUT a major drawback is that this solution is defined only for $0<r<1$... Should there exist an analytical continuation through the complex plane ?

b) If one relaxes the constraint on $h(0)$ by letting $h(0)=0.001$ for example, the solutions for $h$ are increasing functions ; the curves of the $phi$ functions are highly oscillatory (see figures which are but a number of very different cases I have tried).

Matlab program:

function soldiffequ;

clear all;close all;

rspan = 1.00001:0.001:15; % domain for r values

ode = @(r,y) sc(r,y);

for k=2:5

y0 = [2^k;0.0001;0.;0.];% initial values for phi,h,phi',h', resp.

[r,Y] = ode45(ode, rspan, y0);%Runge Kutta solver

functions phi and h as functions of r

figure(1);hold on;plot(r,Y(:,1),’r’); % phi(r)

figure(2);hold on;plot(r,Y(:,2),’k’); % h(r)


function yp = sc(r,y)


Dphi=Phi;% a name for the der. of phi

Dh=H;%a name for the der. of h



yp = [Dphi;Dh;DPhi;DH];

Fig. 1a : case h(1)=0 : in red, functions phi for initial values 4,8,16 ,32. Functions h remain identically 0… :

Fig. 2a and 2b : Cases where h(1) is non zero (though very small : 0.0001) showing that functions phi may have a very oscillatory behaviour, according to their initial values (here 4,8,16,32),; Functions h are no longer zero : they are even rapidly increasing.

Fig. 1

Fig. 2a

Fig. 2b

  • $\begingroup$ Thanks a lot for this! However, I am not sure I completely understand this result. Are you assuming here that $h'(1)=0$ and $\phi'(1)=0$? The numerical solutions I have found depend on the parameters $h'(1)$ and $\phi(1)$, so it might not be so surprising that putting $h'(1)=0$ yields the trivial solution for $h$. Also, it should follow from the ODEs that $3\phi'(1)=-2\phi(1)$. $\endgroup$ Feb 8 '16 at 13:36
  • $\begingroup$ This last condition follows from multiplying the first DE by $r^3-1$ and then evaluating at $r=1$, where if $h$ is analytic $h(r)\approx h'(1)(r-1)$ so that the term proportional to $h(r)^2$ vanishes (i.e. $h(r)^2/(r^3-1)=h'(1)^2(r-1)^2/(r^3-1)\rightarrow 0$. $\endgroup$ Feb 8 '16 at 13:43
  • 1
    $\begingroup$ Yes indeed, I had taken the most natural initial conditions (all derivatives equal to zero).The fact that there $3\phi'(1)=-2\phi(1)$ changes things..I will take it into account. $\endgroup$
    – Jean Marie
    Feb 8 '16 at 13:54
  • $\begingroup$ I just incorporated condition $3\phi'(1)=-2\phi(1)$ in the program : no change ; the results are the same as in fig. 1 for $\phi$ and $h$ is still identically zero. Can some similar condition be found on $h'(1)$ ? $\endgroup$
    – Jean Marie
    Feb 8 '16 at 14:11
  • $\begingroup$ When I say "the same", one must understand "have the same shape"... $\endgroup$
    – Jean Marie
    Feb 8 '16 at 14:22

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.