Especially when coupled to a spin.
Even more so if they are the relatively long lived and fast cavities you guys are constructing on the 4th floor.
Cavity (EM oscillator) coupled to a qubit (a two-level system)
\(\hat{H}_{0}=\omega_{q}\mid e\rangle\langle e\mid+\omega_{c}\hat{n}-\chi\mid e\rangle\langle e\mid \hat{n}\)
\(\hat{H}_{cavity}=\varepsilon\left(t\right)e^{i\omega_{c}t}\hat{a}^{\dagger}+h.c.\)
Displacement operator: \(\hat{D}(\alpha)=\exp\left(\alpha\hat{a}^{\dagger}-\alpha^{*}\hat{a}\right)\) where \(\alpha=i\int\varepsilon(t)dt\) acting only on the ground subspace \(\{\left|0\right\rangle \dots\left|n\right\rangle\dots \}\otimes\left|g\right\rangle\)
\(\hat{H}_{qubit}=\Omega\left(t\right)e^{i\omega_{q}t}\left|e\rangle\langle g\right|+h.c.\)
The control can be selective on the number of photons!
\(\Omega(t) = \Omega e^{-in\chi t}\) with \(\Omega\ll\chi\)
\(H = H_{0} + H_{cavity}(t) + H_{qubit}(t)\)
Use the selective control on the qubit to take closed paths on the Bloch sphere. Always end in the ground state.
\(\left|g,n\right\rangle \rightarrow e^{i\theta_{n}}\left|g,n\right\rangle\)
Selective on Number Arbitrary Phase
\(\hat{S}_{n}\left(\theta_{n}\right)=e^{i\theta_{n}\left|n\rangle\langle n\right|}\)
We can address multiple pairs of states in parallel.
\(\hat{S}\left(\vec{\theta}\right)\,=\prod_{n=0}^{\infty}\hat{S}_{n}\left(\theta_{n}\right)\,=\sum_{n=0}^{\infty}e^{i\theta_{n}}\mid n\rangle\langle n\mid\)
The usable Hilbert space is the \(\{\left|0\right\rangle \dots\left|n\right\rangle\dots \}\otimes\left|g\right\rangle\) subspace.
For most of the rest of the presentation we will restrict ourselves to the ground subspace.
We have these two basic operation acting on the ground subspace (processor instructions in a CPU analogy):
Displacement:
\(\hat{D}(\alpha)=\exp\left(\alpha\hat{a}^{\dagger}-\alpha^{*}\hat{a}\right)\)
SNAP Gate:
\(\hat{S}\left(\vec{\theta}\right)\,=\prod_{n=0}^{\infty}\hat{S}_{n}\left(\theta_{n}\right)\,=\sum_{n=0}^{\infty}e^{i\theta_{n}}\mid n\rangle\langle n\mid\)
Both of which act only on the ground subspace.
Can we use them to prepare any state in the ground subspace?
Consider \(\vec{\theta}_\epsilon=(\underset{n}{\underbrace{\epsilon,\dots,\epsilon}},0,\dots)\)
Sandwich the corresponding SNAP gate with a similar Displacement gate into a group commutator:
\(\hat{D}(\epsilon)\hat{S}(\vec{\theta}_\epsilon)\hat{D}(-\epsilon)\hat{S}(-\vec{\theta}_\epsilon)\)
\(\approx \exp\left(i\epsilon^{2}\sqrt{n+1}(\mid n\rangle\langle n+1\mid+h.c.)\right)\)
Nearest neighbours are coupled and by iteration we can couple all levels.
Given the state \(\left|n\right\rangle\) we want to create the state \(\left|\text{target}\right\rangle=\cos(\theta)\left|n\right\rangle + \sin(\theta)\left|n+1\right\rangle\)
Inspired by the above, consider the non-infinitesimal operation:
\(\hat{U}_{n}=\hat{D}(\alpha_{1})\hat{R}_{n}\hat{D}(\alpha_{2})\hat{R}_{n}\hat{D}(\alpha_{3})\)
where
\(\hat{R}_{n}=-\sum_{n'=0}^{n}\left|n'\rangle\langle n'\right|+\sum_{n'=n+1}^{\infty}\left|n'\rangle\langle n'\right|\)
\(\hat{U}_{n}=\hat{D}(\alpha_{1})\hat{R}_{n}\hat{D}(\alpha_{2})\hat{R}_{n}\hat{D}(\alpha_{3})\)
\(\hat{R}_{n}=-\sum_{n'=0}^{n}\left|n'\rangle\langle n'\right|+\sum_{n'=n+1}^{\infty}\left|n'\rangle\langle n'\right|\)
Optimize \(F=\left|\left\langle\text{target}\right|\hat{U}_{n}\left|n\right\rangle\right|\) wrt \(\alpha_1\), \(\alpha_2\), and \(\alpha_3\) with some good initial guesses.
Restrict to "non-negative" \(\mid \psi \rangle\)
\(\mid \psi \rangle = \left|\text{target}\right\rangle =\sum_{n=0}^{N}c_n\left|n\right\rangle\), \(\,\,\,c_n\ge 0\)
Construction by "unrolling":
Apply a final SNAP gate to impart any missing phases.
Fidelity better than \(0.999\).
To create an arbitrary N-dimensional state
We want to perform an arbitrary unitary operation on the oscillator state (even when the state is unknown), not just prepare a target state from another given state.
As in the case of state preparation for \(\left|n\right\rangle \rightarrow \cos(\theta)\left|n\right\rangle + \sin(\theta)\left|n+1\right\rangle\)
we want to use
\(\hat{U}=\hat{D}(\alpha_{1})\hat{R}_{n}\hat{D}(\alpha_{2})\hat{R}_{n}\hat{D}(\alpha_{3})\)
to implement
\(\hat{U}_{target}=\left[\begin{array}{c|c|c} Id_{n\times n} & 0 & 0\\ \hline 0 & \begin{array}{cc}\cos\theta & -\sin\theta\\ \sin\theta & \cos\theta \end{array} & 0\\ \hline 0 & 0 & Id \end{array}\right]\)
on the \(\{|n\rangle,|n+1\rangle\}\) subspace.
Optimize \(F=\left|\frac{1}{N_{cutoff}}Tr\left(\hat{U}^{\dagger}\hat{U}_{target}\right)\right|\) wrt \(\alpha_1\) and \(\alpha_2\) (\(\alpha_3 = -\alpha_1-\alpha_2\))
We want to construct \(\hat{U}_{target}\)
\(\hat{U}_{target}^{-1}=\left[\begin{array}{c|c} \hat{W}_{n\times n} & 0\\ \hline 0 & Id \end{array}\right]\)
We can apply a SNAP gate to make the last column positive
and then chain \(N-1\) 2-by-2 rotations:
\(\hat{V}_{n-1,n}\dots\hat{V}_{1,2}\hat{S}_{n}\hat{U}_{target}^{-1}=\left[\begin{array}{c|c} \begin{array}{c|c} \hat{W}_{n-1\times n-1} & 0\\ \hline 0 & 1 \end{array} & 0\\ \hline 0 & Id \end{array}\right]\)
Removing a column requires \(N-1\) 2-by-2 matrices. To ensure high fidelity, at the end of the column removal we perform a global optimization over all \(2N\) displacement parameters.
The cost function is the "leakage" outside of \(\hat{W}_{n-1\times n-1}\)
\(\left[\begin{array}{c|c} \begin{array}{c|c} \hat{W}_{n-1\times n-1} & 0\\ \hline 0 & 1 \end{array} & 0\\ \hline 0 & Id \end{array}\right]\)
We repeat the procedure for all columns. Then we do one final global optimization over all the displacement coefficients (~\(N^2\) of them) against the fidelity
\(F=\left|\frac{1}{N_{cutoff}}Tr\left(\hat{U}^{\dagger}\hat{U}_{target}\right)\right|\)
Target: permutation matrix
Result: \(\hat{U}_{target}\hat{U}_{constructed}^{-1}\)
Fidelity for random matrices:
Fidelity for Fourier and Permutation matrices:
To create an arbitrary N-by-N unitary matrix
Efficient scheme to control a multi-level system that can store multiple qubits of information.
Look at small \(\theta\):
For certain elegant definition of infidelity:
We can prove
\(\text{infid}\left(U_{N\times N}\right)\le M\left(\frac{N}{n_{snaps\ per\ SO(2)}}\right)^{2}\)
Given that \(\text{#gates} \propto n_{snaps\ per\ SO(2)} N^2\):
\(\text{#gates} \propto \frac{N^3}{\sqrt{\text{infid}}}\)
The preparation of Fock states by folding can be generalized to the preparation of any "sparse" state where the population is centered around a single Fock state.
We were working on only half the Hilbert space (the \(\mid g \rangle\) subspace).
If we can:
We can easily extend the protocol to work on the entire Hilbert space.
Ask Chao (MLS, Mar 30, 2015) and Reinier (MLS, next week).
Implementing a binary search for a Fock state using a SNAP-like gate and conditional measurement.
Reinier will be presenting it next week. They are designing even shorter pulses with some in-depth numerical optimizations.
Efficient, Fast and Extendable New Protocol for Universal Control
Thank you for your attention!
Questions?
(image credit to SMBC comics)
\(H = -\frac{1}{2}\Delta\hat{\sigma}_z + H_{green} + H_{blue}\)
\(H_{green} = \frac{1}{2}g(e^{i\beta}\hat{a}^\dagger\hat{\sigma}_- + h.c.)\)
\(H_{blue} = \frac{1}{2}\chi(\cos\phi\hat{\sigma}_x + \sin\phi\hat{\sigma}_y)\)
C. K. Law and J. H. Eberly, Phys. Rev. Lett. 76, 1055 (1996)
Their SNAP-like gate is very expensive (they do provide faster non-analytic version).
\(T_{total}\propto \frac{N^{18.5}}{\text{infid}^3}\)
Brian Mischuck and Klaus Mølmer, Phys. Rev. A 87, 022341 (2013)