A real-time synchronization feedback for single-atom frequency standards
1Pierre Rouchon
Mines ParisTech, Centre Automatique et Syst `emes [email protected]
in collaboration with Mazyar Mirrahimi(INRIA)
Quantum/Classical Control in Quantum Information:
Theory and experiments.
September 13-20, 2008 Otranto, Italy
1Partially supported by ”ANR Blanc” project CQUID
Outline
Chip-scale Atomic clock The NIST MicroClock
The principle: Coherent Population Trapping The system and its synchronization scheme Master equations and quantum trajectories
The slow/fast master equation The slow master equation Quantum trajectories Convergence analysis Concluding remarks
The NIST MicroClock
2I Quartz crystal clocks: 1 second over few days.
I NIST chip-scale atomic clock: 1 second over 300 years
I High-Perf. atomic clocks: 1 second over 100 million years.
2NIST: National Institute of Standards and Technology, web-site:
http://tf.nist.gov/timefreq/index.html.
The principle: Coherent Population Trapping
33From the web-site:http://tf.nist.gov/timefreq/index.html.
The synchronization via extremum seeking
s=dtd, constant>0 parameters(h,k,a,ω).
Here u = ωdiode and y = f(ωdiode) where f admits a sharp maximum at the unknown value u¯=ωatom.
Extremum seeking via feedback: u(t) =v(t) +asin(ωt)where v(t)≈ωatomis adjusted via adynamic time-varying output feedback(withω,a,h,√
kωatom):
d
dtv(t) =−ksin(ωt)(y(t)−ξ(t)) with d
dtξ=h(y(t)−ξ(t)) Our contribution4: a real-time synchronization schemewhen the atomic cloud is replaced by a single atom.
4Mirrahimi-R, 2008, arxiv:0806.1392v1
The system and its synchronization scheme
Input: Ω˜1,Ω˜2∈C and u = dtd∆.
Output: photo-detector click times corresponding to stochas- tic jumps from|eito|g1ior|g2i.
Synchronization goal: stabilize the unknown detuning∆to 0.
Two time-scales:
|Ω˜1|,|Ω˜2|,|∆e|,|∆| Γ1,Γ2
Modulationof Rabi complex amplitudesΩ˜1andΩ˜2: Ω˜1(t) = Ω1−ıεΩ2cos(ωt), Ω˜2(t) =ıεΩ1cos(ωt) + Ω2, withΩ1,Ω2>0 constant,ωΓ1,Γ2and 0<ε1.
Detuning update ∆N+1= ∆N−K2Ω1Ω2
Ω21+Ω22cos(ωtN) at each detected jump-timetN. The gainK >0 fixes the standard deviationσK: 4σK2 =εKΩΓ21+Ω22
1+Γ2.
Closed-loop quantum trajectory (
matlab M-file:QCCQI08PR.m)
0 1 2 3 4 5
x 104 0
500 1000
ω t / 2 π
Click number
0 1 2 3 4 5
x 104 0
20 40
ω t / 2 π
Normalized de−tuning Δ/σK
Λ-system parameters:Γ1= Γ2=10,∆e=2.0
Modulation parameters:Ω1= Ω2=1.0,ω=2.8,ε=0.14 Feedback gainK=0.0023 leading to a standard deviation σK =0.0057
Robustness
0 1 2 3 4 5
x 104 0
500 1000
ω t / 2 π
Click number
0 1 2 3 4 5
x 104 0
20 40
ω t / 2 π
Normalized de−tuning Δ/σK
Detector efficiencyof 50%,wrong jump detectionof 50%, feedback-loop delayofτ withω τ=π/4.
The slow/fast master equation
Master equation of theΛ-system d
dtρ=−ı
h¯[ ˜H,ρ] +1 2
2
∑
j=1
(2QjρQj†−Qj†Qjρ−ρQ†jQj),
with jump operatorsQj=p Γj
gj
he|and Hamiltonian H˜
h¯ =∆
2(|g2i hg2| − |g1i hg1|)−
∆e+∆ 2
|ei he|
+Ωe1|g1i he|+Ωe∗1|ei hg1|+Ωe2|g2i he|+Ωe∗2|ei hg2|.
Since|Ω˜1|,|Ω˜2|,|∆e|,|∆| Γ1,Γ2we havetwo time-scales: a fast exponential decay for ”|ei”and aslow evolution for
”(|g1i,|g2i)”.
The slow master equation
Geometric reduction via center manifold techniques5leads to a reduced master equationthat is still of Lindblad type with a slow HamiltonianH andslow jump operatorsLj:
d
dtρ=−ı
¯
h[H,ρ] +1 2
2
∑
j=1
(2LjρL†j −L†jLjρ−ρL†jLj),
withH=∆2σz= ∆(|g2ihg22|−|g1ihg1|) andLj=p γ˜j
gj b
Ωe
and whereγ˜j=4|eΩ(Γ1|2+|eΩ2|2
1+Γ2)2 Γj and bΩ˜
is thebright state:
b
Ωe
= Ωe1 q
|eΩ1|2+|eΩ2|2
|g1i+ Ωe2 q
|eΩ1|2+|eΩ2|2
|g2i
For∆ =0,ρ converges towards thedark state
dΩ˜ dΩ˜ :
d
Ωe
=− Ωe∗2 q
|eΩ1|2+|eΩ2|2
|g1i+ Ωe∗1 q
|eΩ1|2+|eΩ2|2
|g2i.
5Mirrahimi-R 2008, arXiv:0801.1602v1, accepted in IEEE-AC.
Quantum trajectories for the slow approximation
In the absence of the quantum jumps,ρ evolves on theBloch sphereaccording to (˜γ=4|Ωe1Γ|2+|Ωe2|2
1+Γ2 ) 1
γ˜ d
dtρ=−ı∆
2˜γ[σz,ρ]− b
Ωe b
Ωe
ρ+ρ
bΩe b
Ωe
2 +
bΩe
ρ
b
Ωe
ρ.
At each time stepdt,ρ may jump towards the state|g1i hg1|or
|g2i hg2|with ajump probabilitygiven by:
dt γ˜ bΩe
ρ
bΩe
SinceΩ˜1(t) = Ω1−ıεΩ2cos(ωt)andΩ˜2(t) =ıεΩ1cos(ωt) + Ω2, γ˜
bΩe b
Ωe
=γ(|bi+ıεcos(ωt)|di) (hb| −ıεcos(ωt)hd|) withγ=4|Ω1Γ|2+|Ω2|2
1+Γ2 ,|bi=Ω1√|g1i+Ω2|g2i
Ω21+Ω22 and|di= −Ω2√|g1i+Ω1|g2i
Ω21+Ω22
Quantum trajectories in Bloch-sphere coordinates
Withβ=2 arg(Ω1+ıΩ2)and
ρ= 1+X(|bihd|+|dihb|)+Y(ı|bihd|−ı|dihb|)+Z(|dihd|−|bihb|)
2 :
d
dtX=−∆cosβY−γ
εcos(ωt)Y+1−ε2cos2(ωt)
2 Z
X d
dtY= ∆cosβX−∆sinβZ+γ εcos(ωt)
−γ
εcos(ωt)Y+1−ε2cos2(ωt)
2 Z
Y d
dtZ= ∆sinβY+γ
1−ε2cos2(ωt) 2
−γ
εcos(ωt)Y+1−ε2cos2(ωt)
2 Z
Z Thejump probability per unit of timeis
Pjump= γ
2(1−Z−2εcos(ωt)Y+ε2cos2(ωt)(1+Z)).
Just after a jump(X,Y,Z)is reset to±(sinβ,0,cosβ).
Convergence of the no-jump dynamics
d
dtX=−∆cosβY−γ
εcos(ωt)Y+1−ε2cos2(ωt)
2 Z
X d
dtY= ∆cosβX−∆sinβZ+γ εcos(ωt)−γ
εcos(ωt)Y+1−ε2cos2(ωt)
2 Z
Y d
dtZ= ∆sinβY+γ
1−ε2cos2(ωt) 2
−γ
εcos(ωt)Y+1−ε2cos2(ωt)
2 Z
Z
For|∆|<2γ and 0<ε1, the above time-periodic nonlinear system admits aquasi-global asymptotically stable periodic orbit(proof: Poincar ´e-Bendixon and perturbation). It reads
(X,Y,Z) =
0 , −sinβ∆
γ +γ2cos(ωt) +γ ωsin(ωt)
ω2+γ2 ε , 1
up to second order terms inε and ∆γ. Whenωγ,Pjump≈γ
εcos(ωt) +∆sin2γβ 2
if the last jump occurs more that few−logε/γ second(s) ago.6.
6ReplaceZ by 1−X2+Y2 2 in previous formula givingPjump.
Detuning update as a discrete-time stochastic process
Our analysis neglects the transient just after a jump.
When a jump occurs attN, we have
∆N+1=∆N−Ksinβcos(ωtN) and its probability was proportional to
εcos(ωtN) +∆N2γsinβ 2
. The phaseϖ=ωtN can be seen as a stochastic variable in [0,2π]with the following probability densityP∆N(ϖ)on[0,2π]:
P∆N(ϖ) =
εcos(ϖ) +∆N2γsinβ2
2π
ε2 2 +∆
2 Nsin2β
4γ2
The de-tuning update is thus adiscrete-time stochastic process
∆N+1=∆N−Ksinβcosϖ where the probability ofϖ∈[0,2π]depends on∆N.
Convergence proof
We assume here|∆| ε γ (rememberγωΓ1+ Γ2):
∆N+1=∆N−Ksinβcosϖ
withϖ of probability densityP∆N(ϖ)≈ 2π1 +∆Nπ ε γsinβcosϖ . Simple computations yield to7
E(∆N+1|∆N) = 1−Ksin2β ε γ
!
∆N
For 0<K ≤ ε γ
sin2β,E(∆N)tends to zero.
Similarly, we have
E(∆2N+1|∆N) = 1−2Ksin2β ε γ
!
∆2N+K2sin2β 2 For 0<K ≤ ε γ
2 sin2β,E(∆2N)converges toσK2 =ε γK4 .
7E(∆N+1|∆N)stands for the conditional expectation-value of∆N+1 knowing∆N.
Summary: scales and feedback-gain design
Rabi frequency modulations:
Ω˜1(t) = Ω1−ıεΩ2cos(ωt) Ω˜2(t) =ıεΩ1cos(ωt) + Ω2 withΩ1,Ω2Γ = Γ1+ Γ2, 0<ε1 and
Ω21+Ω22
Γ1+Γ2 =γωΓ Detuning update
∆N+1 = ∆N −Ksinβcos(ωtN) withK >0,β =2 arg(Ω1+ıΩ2).
Adiscrete-time stochastic processwhere the gainK >0 drives
I the convergence speedwith a contraction of
1−Ksin2β
ε γ
forE(∆N)at each iteration
I the precisionvia the asymptotic root-mean-square σK =
√
ε γK 2 .
Concluding remarks
I For anonlinear convergence proofwith∆<γ/2,ε small enough and well tuned gainK, see Mirrahimi-R 2008, arxiv:0806.1392v1. Sensitivity analysis to wrong jump detection and noise remains to be done.
I Such simple feedback can be also developed for other single quantum systems such as the 3-level system illustrating theDehmelt’s electron shelving scheme8
I Such feedback scheme could be a preliminary guide for inventing the ”quantum regulator”, a quantum analogue of theclassical PID regulator.
8C. Cohen-Tannoudji, J. Dalibard: Single atom Laser spectroscopy:
looking for dark periods in fluorescent light. Europhys. Lett. 1 (9), pp:441-448, 1986.