1

Consider the first-order non-autonomous ODE: \begin{align} \dfrac{\mathrm{d}x}{\mathrm{d}t}=q(t)\cos^2 x-\sin ^2 x, \qquad \tag{$\star$} \end{align} where $q(t)$ is a continuous function with period $1$. Assume that this equation has a bounded solution.

(a) Show that any solution of $(\star)$ is bounded.

(b) For any $x\in \mathbb{R}$, denote $\phi(t,x)$ be the solution of $(\star)$ with initial value $\phi(0,x)=x$. Define the function $f:\mathbb{R}\to\mathbb{R}$ by $f(x)=\phi(1,x)$. Show that $f$ has exactly two fixed points $x_{1},x_{2}$ in $[0,\pi]$ with $f'(x_{i})\neq 1, i=1,2$.

For question (a), let $\phi_{0}$ be a bounded solution, then for any $n\in\mathbb{Z}$, $\phi_{0}+n\pi$ is also a bounded solution. By the uniqueness of solutions, for any two solutions of $(\star)$, their curves cannot intersect. Therefore, for any given solution of $(\star)$, its curve must be between $x=\phi_{0}(t)+n\pi$ and $x=\phi_{0}(t)+(n+1)\pi$ for some $n$, hence bounded.

However, I have no idea about (b). I would appreciate any answers and comments

Gonçalo
  • 15,869
czw
  • 103

1 Answers1

1

I have an idea of finding one fixed point of $f$.

\begin{align} \dfrac{\mathrm{d}x}{\mathrm{d}t}=q(t)\cos^2x-\sin^2x. \tag{$\star$} \end{align}

Fact 1: Due to the periodicity of $q$ and $\sin,\cos$, we can easily verify if $\phi(t)$ is a solution of $(\star)$, then so are $\phi(t+1)$ and $\phi(t)+\pi$.

For any $x\in\mathbb{R}$, let $\phi(t,x)$ be the solution of $(\star)$ with initial value $\phi(0,x)=x$.

Claim: $f(x)=\phi(1,x)$ satisfies:
(1) $f^{n}(x):=f(f(\cdots f(x)))=\phi(n,x)$;
(2) $f(x+\pi)=f(x)+\pi$;
(3) $f$ is strictly increasing.

Proof: For (1), $\phi(t+1,x)$ is the solution of $(\star)$ with initial value $\phi(1,x)=f(x)$, thus $\phi(t+1,x)=\phi(t,f(x))$. In particular, take $t=1$, we have $\phi(2,x)=f(f(x)):=f^2(x)$. Inductively, we can get $\phi(n,x)=f(f(\cdots f(x)))=f^{n}(x)$.

For (2), $\phi(t,x)+\pi$ is the solution of $(\star)$ with initial value $x+\pi$, thus $\phi(t,x)+\pi=\phi(t,x+\pi)$. In particular, take $t=1$, we have $f(x)+\pi=f(x+\pi)$.

To see (3), we compute the equation that $\partial_{x}\phi(t,x)$ satisfies. \begin{align} \dfrac{\partial }{\partial t}\left(\dfrac{\partial}{\partial x}\phi(t,x)\right)&=\dfrac{\partial }{\partial x}\left(\dfrac{\partial}{\partial t}\phi(t,x)\right)=\dfrac{\partial }{\partial x}\left(q(t)\cos^2\phi(t,x)-\sin^2\phi(t,x)\right)\\ &=-2(q(t)+1)\sin\phi(t,x)\cos\phi(t,x) \dfrac{\partial }{\partial x}\phi(t,x). \end{align} Therefore, \begin{align*} \dfrac{\partial }{\partial x}\phi(t,x)=\exp\left\{\int_{0}^{t}-2(q(s)+1)\sin\phi(s,x)\cos\phi(s,x)\mathrm{d}s\right\}>0. \end{align*} In particular, $f'(x)=\dfrac{\partial }{\partial x}\phi(1,x)>0$, hence $f$ is strictly increasing.

Now, by the property of rotation number, the limit \begin{align} \rho(f):=\lim_{n\to\infty}\dfrac{f^{n}(x)-x}{n} \end{align} exists, and it is independent on $x$. In particular, take $x=0$. By $(2)$, $f^{n}(0)=\phi(n,0)$. Since $\phi(t,0)$ is bounded, then we must have \begin{align} \rho(f)=\lim_{n\to\infty}\dfrac{f^{n}(0)}{n}=0. \end{align} In fact, this implies $f$ has a fixed point in $[0,\pi]$. If not, then $g(x):=f(x)-x$ has no zero in $[0,\pi]$. WLOG, assume $g>0$. Note that $g(x+\pi)=f(x+\pi)-x-\pi=f(x)-x=g(x)$, then $g$ has period $\pi$. Now, we can assume that $\min_{\mathbb{R}}g=\min_{[0,\pi]}g=\delta>0$. Then \begin{align} f(x)-x&=g(x)\ge \delta\\ f(f(x))-f(x)&=g(f(x))\ge \delta\\ &\vdots\\ f^{n}(x)-f^{n-1}(x)&=g(f^{n-1}(x))\ge\delta \end{align} Therefore, $f^{n}(x)-x\ge n\delta$, hence $0=\lim_{n\to\infty}\dfrac{f^{n}(x)-x}{n}\ge\delta>0$, a contradiction.

czw
  • 103
  • 1
    For 3) you could also simply use that different solutions can not touch or even cross. There is no need to invoke the rotation number, the ensuing argument directly shows that if $g$ has a positive minimum, then all solutions must grow indefinitely (linearly on average), which contradicts the established boundedness. Note that if a continuous periodic function like $g$ crosses the zero line downwards, then it has also to cross it upwards in the same period. – Lutz Lehmann Aug 11 '24 at 12:38