Let $f\in L^{2}(0,1).$ Consider the following unique continuation problem: $$ \left\{ \begin{array}{ccc} af(x-r)+bf(x)=0, & \mathrm{if} & x\in (r,1) \\ & & \\ cf(x+1-r)+df(x)=0 & \mathrm{if} & x\in (0,r)% \end{array}% \right. \Longrightarrow f=0?.\text{ }r\in (0,1). $$

This is what I have done:

I have started by writing Fourier expansion of $f:$ $$ f(x)=\sum_{n\in %TCIMACRO{\U{2124} }% %BeginExpansion \mathbb{Z} %EndExpansion }c_{n}e^{2in\pi x}\text{ with }c_{n}=\int_{0}^{1}f(t)e^{-2in\pi t}dt. $$

The goal is to prove that $c_{n}=0,$ $\forall n\in %TCIMACRO{\U{2124} }% %BeginExpansion \mathbb{Z} %EndExpansion .$ \begin{eqnarray*} c_{n} &=&\int_{0}^{1}f(t)e^{-2in\pi t}dt=\int_{0}^{r}f(t)e^{-2in\pi t}dt+\int_{r}^{1}f(t)e^{-2in\pi t}dt \\ &&\overset{\text{by definition}}{=}-\frac{c}{d}\int_{0}^{r}f(t+1-r)e^{-2in% \pi t}dt-\frac{a}{b}\int_{r}^{1}f(t-r)e^{-2in\pi t}dt \\ &=&-\frac{c}{d}e^{-2in\pi r}\int_{1-r}^{1}f(t)e^{-2in\pi t}dt-\frac{a}{b}% e^{-2in\pi r}\int_{0}^{1-r}f(t)e^{-2in\pi t}dt. \end{eqnarray*}

So, $$ \left( 1+\frac{c}{d}e^{-2in\pi r}\right) \int_{1-r}^{1}f(t)e^{-2in\pi t}dt+\left( 1+\frac{a}{b}e^{-2in\pi r}\right) \int_{0}^{1-r}f(t)e^{-2in\pi t}dt=0. $$

For example if $$ 1+\frac{c}{d}e^{-2in\pi r}=1+\frac{a}{b}e^{-2in\pi r},\text{ }\forall n\in %TCIMACRO{\U{2124} }% %BeginExpansion \mathbb{Z} %EndExpansion , $$

the we get $$ c_{n}=\int_{0}^{1}f(t)e^{-2in\pi t}dt=0,\text{ }\forall n\in %TCIMACRO{\U{2124} }% %BeginExpansion \mathbb{Z} %EndExpansion , $$

but this condition is clearly not necessary.

Any ideas or other approaches to handle this problem?.

Thank you in advance.


1 Answer 1


A partial answer: some cases of uniqueness and some cases of non-uniqueness.

Since $f$ is only defined a.e., we may identify $0$ and $1$ and pose the problem in $\mathbb{R}/\mathbb{Z}$, which we may identify with $[0,1)$ as a measure space. If we denote $\tau:[0,1)\to[0,1)$ the translation $x\mapsto x-r\mod 1$, and define $\alpha:=-\frac c d\chi_{[0,r)}-\frac ab\chi_{[r,1)}$ the conditions can be rewritten as a fixed point equation: $f(x)=\alpha(x)f(\tau(x))$ (a.e.).

Situation 1. Assume $|c|<|d|$ and $|a|<|b|$ (or also, $|c|>|d|$ and $|a|>|b|$). Then $\|\alpha\|_\infty<1$ (respectively, $\|\frac1 \alpha\|_\infty<1$ ) and taking the $L_2$ norms $\|f\|_2\le\|\alpha\|_\infty\|f\circ\tau\|_2=\|\alpha\|_\infty\|f\|_2$ whence $f=0$ (the other case is analogous).

Situation 2. Now assume $r$ is rational, thus $r=\frac kn$ with $0<k<n$ and $(k,n)=1$. Then the $n$ intervals $I_j:=\tau^j([0,\frac1n))$, $0\le j<n$ are a partition of $[0,1)$, $k$ of which are included in $[0,r)$, the other $n-k$ being included in $[r,1)$.

If we define freely $f$in the interval $[0,\frac1n)$ the equation determines it uniquely on $[0,1)$, with a compatibility condition, namely $f(x)=\Big[\prod_{0\le j<n}\alpha(\tau^j(x))\Big]f(x).$ Note that for any $x\in[0,1)$ the $n$ iterates $\{\tau^j(x)\}_{0\le j<n}$ are a choice of representatives for the mentioned partition, so the compatibility condition writes $$\Big(-\frac cd\Big)^k\Big(-\frac ab\Big)^{n-k}=1,$$ and we have a closed infinite dimensional linear space of solutions, or no nonzero solutions, according whether this condition holds or does not.

  • $\begingroup$ Thank you Mr. Majer for this great answer. This helped me so much. $\endgroup$
    – Goga
    Aug 8, 2021 at 19:58
  • $\begingroup$ You're welcome. An interesting subcase of 2 is when both $-a/b$ and $-c/d$ are positive real numbers, because the condition for existence of non-zero solutions writes equivalently $(-c/d)^r(-a/b)^{1-r}=1$, that also make sense for irrational $r$, and suggests approximation arguments with rational $r_m\to r$ to prove existence of nontrivial solutions. $\endgroup$ Aug 8, 2021 at 22:08
  • $\begingroup$ Thank you a lot sir for the remark. $\endgroup$
    – Goga
    Aug 9, 2021 at 18:32
  • $\begingroup$ Actually i think there is more. I'll try to add something in a few days $\endgroup$ Aug 9, 2021 at 20:26
  • $\begingroup$ Thank you Mr. Majer in advance. $\endgroup$
    – Goga
    Aug 10, 2021 at 19:51

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.