So the Heisenberg nilmanifold is an addition rule on triples $(a,b,c) + (x,y,z) \equiv (a+x, b+y + m\; xc ,c+z)$. This rule is associative and $$ n(a,b,c) = \left(na, nb + m\frac{n(n-1)}{2}ac, nc\right) $$ However, we are going to quotient by those triples whose coordinates are integers. $$H = (\mathbb{R},\mathbb{R},\mathbb{R}) \mod (\mathbb{Z},\mathbb{Z},\mathbb{Z}) $$ In this relation the quotient should be (can someone help me shrink these fractions please?) $$n(a,b,c) \equiv \left( \{ n a\},\left\{ nb + m \frac{n(n-1)}{2}ac - m\;\; na\lfloor nc\rfloor \right\} , \{nc \} \right)$$ I use additive notation since if $n = 0$, this is just the 3-torus $\mathbb{T}^3$ and it really looks like addition.
In this way it is a circle bundle over the torus (Euler class = n in this case, for $\mathbb{T}^3$, Euler class is $0$.). It can also be thought of the mapping cylinder of the linear map, $ \left( \begin{array}{cc} 1 & 1 \\ 0 & 1 \end{array} \right) $
These come in handy when reading the papers by Tao, Green and Ziegler on Gowers's norms, where they look for patterns in the primes by correlating with various sequences mod 1. $\leftarrow$ If you can help me understand that paper would be nice
I would like to know how functions $f:H \to \mathbb{R}$, i.e. in $f \in L^2(H)$ decomposes into a "Fourier-like series" (maybe the term is spectral decomposition).
If this were just the 3-torus ($n=0$), the modes would be $e^{i(mx+ny+pz)}$, but these Heisenberg groups are not quite abelian.
If I recall... the starting point if to observe the middle copy of the circle, $(0, \mathbb{T}, 0)$ is in the center of the group and then you can write induced representations...
- $(a,b,c) \mapsto e^{i(ma + nc)}$ with integers $m, n \in \mathbb{Z}$ is a two-dimensional family...
- Wikipedia suggests you can act on functions, $(a,b,c)f(x) = e^{i(cx + mb)}f(x+m a)$.
If this were over $H(\mathbb{Z}_p)$ (with $\mathbb{Z}_p =\{ e^{2\pi i k /p}: k = 0 \dots, p-1 \}$ ) instead of $H(\mathbb{T})$ we'd have $p^2$ 1-dimensional representations and $p-1$ of $p$-dimensional representations. Then $p^2\cdot 1 + (p-1)\cdot p^2 = p^3 = |H(\mathbb{Z}_p)|$, so we have all of them. Then let $p \to \infty$ (doesn't have to be prime) and $\mathbb{Z}_p \to S^1$.
I would expect the representations of $H$ to form some kind of 3-dimensional lattice (if $n=0$ representations of $\mathbb{T}^3$ are indexed by $\mathbb{Z}^3$. and then I still don't know of they make up all of $L^2(H)$
Finally, I'm worried I'm not getting the 2-step nilcharacters mentioned in Tao Green Ziegler's papers such as $e^{n^2 x}$ or $e^{n \sqrt{2} \lfloor n \sqrt{3} \rfloor}$ .