# Finding the period of a nonlinear ODE

I am studying a periodic physical system with a nonlinear ODE $$x''=f(x)+g(x)x'^2$$

I think the periodicity comes from the $x'^2$ term because this provides two possible numbers to give a same right hand side value.

The following shows three numerical curves of this equation with $f(x)=x-x^3$ and $g(x)=2/x-x$. We can see that the curve is oscillating about the fixed point (by making $x''=0$ and $x'=0$, here the fixed point is $x^*=1$)

I can solve the period for the almost fixed-point solution $x(t)=x^*+\epsilon \cdot \cos{\omega t}$ and this perturbation gives me a frequency $\omega=\sqrt{-f'(x^*)}$. For the particular example I gave, $\omega=\sqrt{-f'(1)}=\sqrt{-(1-3\cdot 1^2)}=\sqrt{2}$ so $T=2\pi/\omega=\sqrt{2}\pi \approx 4.44$ and it matches the red curve pretty well.

My question is how can I analytically solve the period of the curves far away from the fixed point solution?

Thank you for you attention!

If you have a Hamilton function without separable terms like $$H(x,p)=\frac1{2m(x)}p^2+V(x)$$ the resulting dynamic is \begin{align} \dot x &= ~~~H_p=\frac{p}{m(x)}\\ \dot p &= -H_x=\frac{m'(x)}{2m(x)^2}p^2-V'(x) \end{align} Now eliminate $p,\dot p$ to get $$\ddot x=-\frac{m'(x)}{m(x)^2}\dot xp+\frac{\dot p}{m(x)} =-\frac{m'(x)}{m(x)}\dot x^2+\frac12\frac{m'(x)}{m(x)}\dot x^2-\frac{V'(x)}{m(x)} \\~\\ \ddot x+\frac{m'(x)}{2m(x)}\dot x^2+\frac{V'(x)}{m(x)}=0$$ Now for scalar $x$ the equations $(\ln|m(x)|)'=-2g(x)$ and $V'(x)=-m(x)f(x)$ are always integrable, which means that your ODE always has a first integral. As now all solutions have to remain on the level curves of this first integral, they have to be periodic as long as that level curve does not contain a stationary point. The period can be computed as $$\frac T2=\int_{x_1}^{x_2}\frac{dx}{\sqrt{2(V(x_1)-V(x))/m(x)}}$$ where $x_1<x_2$, $V(x_2)=V(x_1)$ are the extremal points in $x$ direction of a level curve.

In your example I get $m(x)=e^{x^2}/x^4$, $V'(x)=-e^{x^2}\frac{1-x^2}{x^3}=e^{x^2}(x^{-1}-x^{-3})$. As $$\frac d{dx}e^{x^2}x^{-2} = e^{x^2}(2x^{-1}-2x^{-3})$$ we get $$V(x)=\frac{e^{x^2}}{2x^2}$$ so that the lower turning point can be computed via the Lambert-W function from the higher one, $$-x_1^2e^{-x_1^2}=-x_2^2e^{-x_2^2}\implies x_1=\sqrt{-W_{0}(-x_2^2e^{-x_2^2})}$$

Numerical integration of the above integral for the period gives the graph from scipy.special import lambertw

E = 1.001+np.linspace(0,30,150+1); V0s = E*np.exp(1)

def integrand(V0): return lambda x: 1/(x*(V0*x**2*np.exp(-x**2)-1)**0.5)
def x1(V0): return (-lambertw(-1/V0    ).real)**0.5
def x2(V0): return (-lambertw(-1/V0, -1).real)**0.5
T = np.array([ 2*quad(integrand(V0), x1(V0), x2(V0)) for V0 in V0s])

plt.plot(E,T/(2**0.5*np.pi)); plt.grid();
plt.xlabel("$V_0=V(x_{1/2})$ in multiples of $e/2$");
plt.ylabel("$T$ in multiples of $\sqrt{2}\pi$"); plt.show()

• Excuse me, what do you mean by "separable terms"? – Giuseppe Negro May 31 '18 at 10:14
• In the common examples of mechanics, you have a kinetic energy term that solely depends on the velocity resp. impulse, and a potential energy term that solely depends on the position, $H(x,p)=T(p)+V(x)$. This then leads to the usual Newton formulas $m\ddot x=-\nabla V(x)$. As you can see, if the terms are not separable in that way, there will be more terms in the second order equation. – Dr. Lutz Lehmann May 31 '18 at 10:53