2
$\begingroup$

Question: Is there a linear map $\mathcal F$ from the Hilbert space of $\ell^2$ functions on the discrete Heisenberg group to some Hilbert space of functions $ L^2(\bigcup \{\Omega_n\}) $, such that:

  1. $\mathcal{F}(f *g)|_{\Omega_n}=\mathcal{F}(f)|_{\Omega_n} {\cdot } \mathcal{F}(g)|_{\Omega_n}$ for some non-commutative product $\cdot $,

  2. $\langle f,g \rangle=\langle \mathcal{F}(f),\mathcal F(g) \rangle$, i.e. the Plancherel formula holds, and

  3. The spaces $\Omega_1$, $\Omega_2$, ... are as small as possible?

Motivation: This is the setting of a non-commutative circle method that could be used in number theory. Specifically, the problem I am investigating is whether all odd numbers $q$ can be written in the form $wx-xy+yz$ for $w$, $x$, $y$, $z$ primes.

I used the discrete Heisenberg group as an encoding, i.e. let $A$ be the set of matrices $$ \begin{bmatrix} 1 & a & 0 \\ 0 & 1 & b\\ 0 & 0 & 1 \end{bmatrix} $$ $B$ the set of matrices $$ \begin{bmatrix} 1 & a & 0 \\ 0 & 1 & -b\\ 0 & 0 & 1 \end{bmatrix} $$

where $a$ and $b$ are primes, and $C$ the set of all matrices with entry $C_{1,3}=q$.

Let $I_A$ be the indicator function of $A$ times some cutoff function to make $I_A$ an $\ell^2$ function, and similarly are $I_B$ and $I_C$. Then the problem is equivalent to computing the inner product $\langle I_A * I_B * I_A, I_C \rangle$.

Remark: If the discrete Heisenberg group were finite, we could take the $\Omega$s to be the irreducible representations of the group, and $\cdot$ the product of matrices in the irreducible representations. Then all three conditions would be satisfied. It's also the case when the discrete Heisenberg group is replaced by the integers, i.e. the original circle method, though the $L^2$ norm is defined differently.

$\endgroup$
3
  • 3
    $\begingroup$ I think the best you can do in this direction is take the Fourier transform with respect to the action of the centre of the Heisenberg group (which is isomorphic to ${\mathbf Z}$), and any partition of the Pontryagin dual of that centre (which is isomorphic to ${\mathbf R}/{\mathbf Z}$) into countably many measurable pieces induces a decomposition of the form you ask for (though it would be rather artificial to identify the resulting Hilbert spaces as L^2 spaces). I suspect that these are essentially the only such decompositions, though I don't know a particularly slick way to establish this. $\endgroup$
    – Terry Tao
    Commented Jan 4, 2022 at 5:32
  • 3
    $\begingroup$ As a substitute, would you be able to do anything for your intended applications if you worked with "step functions" defined on the real Heisenberg group? (There one has a very explicit Plancherel formula) $\endgroup$
    – Yemon Choi
    Commented Jan 4, 2022 at 6:07
  • 3
    $\begingroup$ Another thought: could you encode your problem into a "reduced" version of the integer Heisenber group, by which I mean: identify the centre with ${\bf Z}$, and then quotient out the whole group by a central copy of $n{\bf Z}$ for some large integer $n$? This would yield a group that is virtually abelian, where once again there is an explicit description of the Plancherel formula (worked out in detail in Folland's A Course in Harmonic Analysis, 2nd ed.) $\endgroup$
    – Yemon Choi
    Commented Jan 4, 2022 at 6:16

0

You must log in to answer this question.