On "uniqueness" of Fourier expansion

1.1k Views Asked by At

It is well known that Taylor series is unique, i.e., if $$ \sum_{n_1, \ldots, n_d \ge 0} a(n_1, \ldots, n_d)x_1^{n_1} \cdots x_d^{n_d} = 0 $$ holds around $(x_1, \cdots, x_d)=0$ and the left hand side converges absolutely, then $ a(n_1, \ldots, n_d) = 0 $ for all $ n_1, \ldots, n_d \ge 0$.

I am interested in the case of Fourier series.

$\mathbf {Definition}$

Let $d \in \mathbb Z_{\gt 0}$, $n = (n_1, \ldots, n_d)$ be a variable in $\mathbb Z^d$ and $a(n) \in \mathbb C$.

Let $X$ be an order of summation of $\sum_{n \in \mathbb Z}$.

Let $$ \mathscr F _X (a, x) = \sum_{n \in \mathbb Z^d} a(n) \mathrm {exp} (2 \pi i(n \cdot x)) $$ be a convergent Fourier series summed up in the order $X$, where $x \in \mathbb R^d$ and $n \cdot x$ is the standard inner product.

$\mathbf {Question}$

Suppose that $\mathscr F_X(a,x) = 0$. Then, can we conclude that $a(n) = 0$ for all $n \in \mathbb Z^d$ ?

Of course, if X is the standard (cubic) summation (i.e., $\lim_{N \to \infty} \sum_{-N \le n_1, \ldots n_d \le +N}$) and $\mathscr F_X(a, x) $ converges uniformly in $x$ on compact sets, then $a(n) = 0$ because $$a(n) = \int _{x \in {[0,1]}^d} \mathscr F_X(a,x) \mathrm{exp}(-2 \pi i n \cdot x) dx$$ are the Fourier coefficients of $\mathscr F_X(a,x) = 0$. On the other hand, if $\mathscr F_X(a, x)$ does not converge uniformly, we may have a chance that $a(n) \ne 0$ for some $n \in \mathbb Z^d$. But I do not have any example.

If anyone knows a hint, an example or a good reference, please tell me.

1

There are 1 best solutions below

1
On

Interesting you should raise this question; trying to give a definitive answer to this question is what led Georg Cantor to develop set theory (this is laid out in a good biography of Cantor by J. W. Dauben.

Cantor's final result:Let the zero function be represented by a Fourier series within a given interval. Let there be a set of points in the interval (possibly empty) called the exception set; we assume the series converges in the interval except on this set, where we make no assumption (it may converge, it may not). The exception set may have accumulation points; call the set of accumulation points the first derived set. This set may itself have accumulation points; call them the second derived set. And third, fourth, and so on.

THEOREM: Suppose the N'th derived set, for some finite N, is empty. Then the representation of the zero function on that interval is unique; i.e, all Fourier coefficients must be zero.

THEOREM: Suppose the N'th derived set is never empty for any finite N. Then the representation of the zero function on that interval need not be unique.

I don't know exactly how this extends to multi-dimensional space, but I suspect the relevant theorems are essentially the same.