6
$\begingroup$

I am considering the following equation

$$\begin{pmatrix} -\frac{d}{dx} + \lambda \sin(2\pi x) & \lambda - \lambda \cos(2\pi x) \\ -\lambda-\lambda \cos(2\pi x) & -\frac{d}{dx} - \lambda \sin(2\pi x) \end{pmatrix}\varphi(x)=0$$

and I am wondering if there is an explicit characterization of $\lambda \neq 0$ for which there exist $1$-periodic solutions to this ODE.

This ODE, though matrix-valued has a nice symmetry in the sense that the off-diagonal entries have only even Taylor polynomials whereas the diagonal ones have odd ones.

However, it is of course difficult to say based on a Taylor series, if this will yield a periodic function.

Observation:

Close to zero, i.e. $x=0$, the solution should heuristically look like $\varphi(x)=(0,e^{-\lambda \pi x^2}),$ as it solves the equation "to leading-order"

$\endgroup$

2 Answers 2

8
$\begingroup$

Your equation in fact admits a simple explicit solution. As noted in the previous answer, the general solution is encoded in a matrix ODE, $$ M'(x) = \lambda A(x)M(x), \qquad M(0)=I\,. $$ Now, the simplification becomes apparent if change to a new basis, using the $x$-dependent rotation $$ R(x) = \begin{pmatrix} \cos(\pi x) & -\sin(\pi x) \\ \sin(\pi x) & \cos(\pi x)\end{pmatrix}. $$ Define $N(x)$ by $M(x)=R(x)N(x)$, and the ODE becomes $$ N' = \left(\lambda R^{-1}AR-R^{-1}R' \right)N = \begin{pmatrix} 0 & \pi \\ -\pi-2\lambda & 0\end{pmatrix} N. $$

Writing $\lambda = \frac{\pi}{2}(\mu^2-1)$, the solution is $$ N(x) = \begin{pmatrix} \cos (\pi\mu x) & \frac{1}{\mu} \sin (\pi\mu x) \\ -\mu \sin (\pi\mu x) & \cos (\pi\mu x) \end{pmatrix} \implies M(1) = \begin{pmatrix} -\cos (\pi\mu) & -\frac{1}{\mu} \sin(\pi\mu) \\ \mu \sin(\pi\mu) & -\cos(\pi\mu) \end{pmatrix}. $$ We get periodic solutions when $\mu=2n+1$, $n=1,2,3,\ldots$, so $\lambda=2\pi n(n+1)$.

$\endgroup$
1
  • $\begingroup$ Nicely done ! Perhaps I should have be more confident in explicit calculations. $\endgroup$ Jun 19, 2021 at 14:19
10
$\begingroup$

I doubt that this vector ODE be explicitly integrable. However you can do the following.

Write it as $\frac{d\phi}{dx}=\lambda A(x)\phi$, where $$A(x)=\begin{pmatrix} \sin2\pi x & 1- \cos2\pi x \\ -1- \cos2\pi x & - \sin2\pi x \end{pmatrix}.$$ Notice ${\rm Tr} A\equiv0$.

The general solution of the Cauchy problem is given in the form $\phi(x)=M(\lambda;x)\phi(0)$. Your question amounts to whether $\mu=1$ is an eigenvalue of $M(\lambda;1)$, or not.

First of all $\frac{d\det M}{dx}=\lambda ({\rm Tr}A(x))\det M\equiv0$, hence $\det M\equiv1$ (because $M(\lambda;0)=I_2$). Thus $1$ is an eigenvalue if and only if the trace of $M(\lambda;1)$ equals $2$.

Thus we are led to the study of $\lambda\mapsto{\rm Tr}M(\lambda;1)$. To this end, we notice that $$\frac{dM}{dx}=\lambda A(x)M,\qquad M(\lambda;0)=I_2.$$ The map $\lambda\to M$ is analytic. Expanding $M=I_2+\lambda M_1+\lambda^2M_2+\cdots$, we have $M_1'=A$ and $M_2'=AM_1$ with $M_j(\lambda;0)=0_2$. This can be integrated explicitly. Obviously ${\rm Tr}M_1\equiv0$. If my calculation is correct, we have ${\rm Tr}M_2(1)=-1$. Thus $|{\rm Tr}M(\lambda;1)|<2$ for small $\lambda\ne0$.

This is a useful information: when $\lambda\ne0$ is small, then the eigenvalues of $M(\lambda;1)$ are $e^{\pm i\theta}$, where $\theta(\lambda)$ varies analytically, with $\theta'(0)\ne0$. For a dense subset near zero, the frequency has the form $\theta(\lambda)=\frac{2\pi p}q$ where $\frac pq$ is rational (irreducible). Then every solution is $q$-periodic.

To answer properly and positively your question, there remains to check whether $\lambda\to{\rm Tr}M(\lambda;1)$ takes values above $2$. If so, there is a parameter $\lambda$ for which this trace equals $2$, and thus $\mu=1$ is a double eigenvalue. Then the ODE admits a $1$-periodic solution.

$\endgroup$

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.