M.E. Irizarry-Gelpí

Physics impostor. Mathematics interloper. Husband. Father.

Hamiltonian for Quantum Fourier Transform 3


Given the unitary Fourier matrix \(F\) and its inverse \(F^{\dagger}\) you can construct the two Hartley matrices \(P\) and \(Q\) via:

\begin{align*} P &\equiv \frac{1}{2}(1 + i)F + \frac{1}{2}(1 - i)F^{\dagger} & Q &\equiv \frac{1}{2}(1 - i)F + \frac{1}{2}(1 + i)F^{\dagger} \end{align*}

Note that both \(P\) and \(Q\) are hermitian:

\begin{align*} P^{\dagger} &= P & Q^{\dagger} &= Q \end{align*}

Both of the Hartley matrices square to the identity:

\begin{align*} P^{2} &= I & Q^{2} &= I \end{align*}

The product of the Hartley matrices gives \(S\):

\begin{equation*} P Q = S \end{equation*}

Note that \(S^{2} = I\) also. Thus, the four matrices \(I\), \(P\), \(Q\), and \(S\) are a representation of the group \(\mathbb{Z}_{2} \times \mathbb{Z}_{2}\). On the other hand, the four matrices \(F\), \(F^{2} = S\), \(F^{3} = F^{\dagger}\), and \(F^{4} = I\) are a representation of the group \(\mathbb{Z}_{4}\).

The Fourier transform matrix and its inverse can be re-written in terms of the Hartley matrices:

\begin{align*} F &= \frac{1}{2}(1 - i) P + \frac{1}{2} (1+i) Q \\ F^{\dagger} &= \frac{1}{2}(1 + i) P + \frac{1}{2} (1-i) Q \end{align*}

In terms of the Hartley matrices, you have involutions. These allow you to use familiar results:

\begin{align*} \exp{(i a P)} &= I\cos{(a)} + i P \sin{(a)} \\ \exp{(i b Q)} &= I\cos{(b)} + i Q \sin{(b)} \\ \exp{(i c S)} &= I\cos{(c)} + i S \sin{(c)} \end{align*}

Then

\begin{equation*} \exp{(i a P)} \exp{(i b Q)} \exp{(i c S)} = I C_{I}(a, b, c) - P C_{P}(a, b, c) - Q C_{Q}(a, b, c) - S C_{S}(a, b, c) \end{equation*}

with

\begin{align*} C_{I}(a, b, c) &= \cos{(a)} \cos{(b)} \cos{(c)} - i \sin{(a)} \sin{(b)} \sin{(c)} \\ C_{P}(a, b, c) &= \cos{(a)} \sin{(b)} \cos{(c)} - i \sin{(a)} \cos{(b)} \sin{(c)} \\ C_{Q}(a, b, c) &= \sin{(a)} \cos{(b)} \cos{(c)} - i \cos{(a)} \sin{(b)} \sin{(c)} \\ C_{S}(a, b, c) &= \sin{(a)} \sin{(b)} \cos{(c)} - i \cos{(a)} \cos{(b)} \sin{(c)} \end{align*}

You need

\begin{align*} C_{I}(a, b, c) &= 0 & C_{P}(a, b, c) &\neq 0 & C_{Q}(a, b, c) &\neq 0 & C_{S}(a, b, c) &= 0 \end{align*}

Use

\begin{align*} a &= x + y & b &= x - y \end{align*}

Then

\begin{align*} 2 C_{I}(x, y, c) &= \left[ \cos{(2y)} + \cos{(2x)} \right] \cos{(c)} - i \left[ \cos{(2y)} - \cos{(2x)} \right] \sin{(c)} \\ 2 C_{P}(x, y, c) &= \left[ \sin{(2x)} - \sin{(2y)} \right] \cos{(c)} - i \left[ \sin{(2x)} + \sin{(2y)} \right] \sin{(c)} \\ 2 C_{Q}(x, y, c) &= \left[ \sin{(2x)} + \sin{(2y)} \right] \cos{(c)} - i \left[ \sin{(2x)} - \sin{(2y)} \right] \sin{(c)} \\ 2 C_{S}(x, y, c) &= \left[ \cos{(2y)} - \cos{(2x)} \right] \cos{(c)} - i \left[ \cos{(2y)} + \cos{(2x)} \right] \sin{(c)} \end{align*}

The following solutions guarantee \(C_{I} = C_{S} = 0\):

\begin{align*} x_{k} &= \frac{(4 k + 1)\pi}{4} & x_{l} &= \frac{(4 l - 1)\pi}{4} & y_{m} &= \frac{(4 m + 1)\pi}{4} & y_{n} &= \frac{(4 n - 1)\pi}{4} \end{align*}

Note that

\begin{align*} \sin{(2x_{k})} &= 1 & \sin{(2x_{l})} &= -1 & \sin{(2y_{m})} &= 1 & \sin{(2y_{n})} &= -1 \end{align*}

The \((x_{k}, y_{m})\) solution gives

\begin{align*} C_{P} &= -i \sin{(c)} & C_{Q} &= \cos{(c)} \end{align*}

Note that

\begin{align*} a_{km} &= \frac{\pi}{2} + (k + m)\pi & b_{km} &= (k - m) \pi \end{align*}

The \((x_{k}, y_{n})\) solution gives

\begin{align*} C_{P} &= \cos{(c)} & C_{Q} &= -i\sin{(c)} \end{align*}

Note that

\begin{align*} a_{kn} &= (k + n) \pi & b_{kn} &= \frac{\pi}{2} + (k - n) \pi \end{align*}

The \((x_{l}, y_{m})\) solution gives

\begin{align*} C_{P} &= -\cos{(c)} & C_{Q} &= i\sin{(c)} \end{align*}

Note that

\begin{align*} a_{lm} &= (l + m) \pi & b_{lm} &= -\frac{\pi}{2} + (l - m) \pi \end{align*}

The \((x_{l}, y_{n})\) solution gives

\begin{align*} C_{P} &= i \sin{(c)} & C_{Q} &= -\cos{(c)} \end{align*}

Note that

\begin{align*} a_{ln} &= -\frac{\pi}{2} + (l + n) \pi & b_{ln} &= (l - n) \pi \end{align*}

With \(c = \pm \pi / 4\) and some more work, you can recover \(F\) as an exponential of an anti-hermitian matrix.