Modelling, simulation and feedback of open quantum systems
Pierre Rouchon
Centre Automatique et Systèmes Mines ParisTech
PSL Research University
Fall 2018
1 / 39
Outline of the lectures (Autumn 2018 at Mines ParisTech )
Lect. 1& 2 (Oct. 24: 11:10-13:00 and 14:00–15:50) Feedback for classical and for quantum systems; the first experimental realization of a
quantum-state feedback (LKB photon box); three quantum features (Schrödinger; collapse of the wave packet, tensor product); Quantum Non Demolition (QND) measurement of photons (ideal Markov model and Matlab simulations).
Lect. 3 (Nov. 7: 11:10–13:00) Models of the LKB photon box: entanglement between the probe-qubit and the photons; qubit-measurement back-action on the photons; measurement errors and imperfections; decoherence as unread fictitious measurements; discrete-time Markov chain; Kraus map; quantum trajectories and realistic Matlab simulations of the photon box
Lect. 4 (Nov. 14: 11:10–13:00) Feedback stabilization of photon-number state:
resonant interaction, the Lyapunov feedback scheme, closed-loop simulations / experimental data.
Lect. 5 (Nov. 21: 9:00–10:50) The LPA super-conduction qubit under continuous-time measurements (counting versus homo/hetero-dyne measurements):
continuous-time stochastic master equation (Poisson versus Wiener);
Lindblad master equation; QND measurement of a qubit and Matlab simulations.
Lect. 6 (Nov. 28: 11:10–13:00) Feedback stabilization of the excited state of a qubit;
quantum-state feedback based on QND measurement;closed-loop simulations.
2 / 39
Main references
I
S. Haroche and J.M. Raimond. Exploring the Quantum:
Atoms, Cavities and Photons. Oxford University Press, 2006.
I
H.M. Wiseman and G.J. Milburn. Quantum Measurement and Control. Cambridge University Press, 2009.
I
M.A. Nielsen and I.L. Chuang. Quantum Computation and Quantum Information. Cambridge University Press, 2000.
I
C.W. Gardiner and P. Zoller Quantum Noise. Springer, 2010.
3 / 39
Outline
Notion of Feedback
Discrete-time systems: The LKB photon-box
Continuous-time systems: qubit in circuit QED
Continuous diffusive-jump SME
4 / 39
Model of classical systems
control
perturbation
measure system
For theharmonic oscillatorof pulsationωwithmeasured positiony, controlled by the forceuand subject to an additional unknown force w.
x = (x1,x2)∈R2, y =x1
d
dtx1=x2, d
dtx2=−ω2x1+u+w
5 / 39
Feedback for classical systems
feedback observer/controller
perturbation
set point
control
measure system
Proportional Integral Derivative (PID) for dtd22y =−ω2y+u+w with the set pointv =yc
u=−Kp y −yc
−Kd
d
dt y−yc
−Kint
Z
y −yc with the positivegains(Kp,Kd,Kint)tuned as follows (0<Ω0∼ω, 0< ξ∼1, 0< 1:
Kp= Ω20, Kd =2ξ q
ω2+ Ω20, ,Kint=(ω2+ Ω20)3/2.
6 / 39
Quantum feedback: the back-action of the measurement.
A typical stabilizing feedback-loop for a classical system
system
controller
wTwo kinds of stabilizing feedbacks for quantum systems 1. Measurement-based feedback: controller is classical;
measurement back-action on the systemS is stochastic (collapse of the wave-packet); the measured outputy is a classical signal; the control inputuis a classical variable appearing in some controlled Schrödinger equation;u(t) depends on the past measurementsy(τ),τ≤t.
2. Coherent/autonomous feedback and reservoir engineering:the systemSis coupled tothe controller, another quantum system; the composite system,HS⊗Hcontroller, is an
open-quantum system relaxing to some target (separable) state.
7 / 39
The first experimental realization of a
quantum state feedbackThe photon box of the Laboratoire Kastler-Brossel (LKB):
group of S.Haroche (Nobel Prize 2012), J.M.Raimond and M. Brune.
u y
1
Stabilization of a quantum state with exactlyn=0,1,2,3, . . .photon(s).
Experiment:C. Sayrin et. al., Nature 477, 73-77, September 2011.
Theory:I. Dotsenko et al., Physical Review A, 80: 013805-013813, 2009.
R. Somaraju et al., Rev. Math. Phys., 25, 1350001, 2013.
H. Amini et. al., Automatica, 49 (9): 2683-2692, 2013.
1Courtesy of Igor Dotsenko.Sampling period 80µs. 8 / 39
Three quantum features emphasized by the LKB photon box
21. Schrödinger: wave funct.|ψi ∈ Hor density op. ρ∼ |ψihψ| d
dt|ψi=−~iH|ψi, d
dtρ=−~i[H, ρ], H=H0+uH1
2. Origin of dissipation: collapse of the wave packetinduced by the measurement of observableOwith spectral decomp.P
µλµPµ:
I measurement outcomeµwith proba.
Pµ=hψ|Pµ|ψi= Tr(ρPµ)depending on|ψi,ρjust before the measurement
I measurement back-action if outcomeµ=y:
|ψi 7→ |ψi+= Py|ψi
phψ|Py|ψi, ρ7→ρ+= PyρPy Tr(ρPy) 3. Tensor product for the description of composite systems(S,M):
I Hilbert spaceH=HS⊗ HM
I HamiltonianH=HS⊗IM+Hint+IS⊗HM
I observable on sub-systemM only:O=IS⊗OM.
2S. Haroche and J.M. Raimond.Exploring the Quantum: Atoms, Cavities and Photons.Oxford Graduate Texts, 2006.
9 / 39
Composite system built with an harmonic oscillator and a qubit.
I System
S corresponds to a quantized harmonic oscillator:
HS = ( ∞
X
n=0
ψn|
n
i(ψn)∞n=0∈
l
2(C) ),
where
|n
irepresents the Fock state associated to exactly n photons inside the cavity
I Meter
M is a qu-bit, a 2-level system (idem 1/2 spin system) :
HM =C2, each atom admits two energy levels and is described by a wave function c
g|g
i+c
e|e
iwith
|
c
g|2+|c
e|2=1; atoms leaving B are all in state
|g
iI State of the full system|Ψi ∈ HS⊗ HM
:
|Ψi=
+∞
X
n=0
Ψng|
n
i ⊗ |g
i+ Ψne|n
i ⊗ |e
i, Ψne,Ψng ∈C.Ortho-normal basis:
(|n
i ⊗ |g
i,|n
i ⊗ |e
i)n∈N.
10 / 39
S: quantum harmonic oscillator (spring system)
I Hilbert space:
HS =n P
n≥0ψn|ni, (ψn)n≥0∈l2(C)o
≡L2(R,C)
I Quantum state space:
D={ρ∈ L(HS), ρ† =ρ, Tr(ρ) =1, ρ≥0}.
I Operators and commutations:
a|ni=√
n|n-1i,a†|ni=√
n+1|n+1i; N=a†a,N|ni=n|ni;
[a,a†] =I,af(N) =f(N+I)a;
Dα=eαa†−α†a. a=X +iP= √1
2 x+∂x∂
, [X,P] =ıI/2.
I Hamiltonian:HS/~=ωca†a+uc(a+a†).
(associated classical dynamics:
dx
dt =ωcp, dpdt =−ωcx−√ 2uc).
I Classical pure state≡coherent state|αi α∈C: |αi=P
n≥0
e−|α|2/2√αn
n!
|ni;|αi ≡π1/41 eı
√2x=αe−(x−
√ 2<α)2 2
a|αi=α|αi,Dα|0i=|αi.
|0
|1
|2 ω
c| n
ω
cu
c... ...
11 / 39
M: 2-level system, i.e. a qubit (half-spin system)
I Hilbert space:
HM =C2=n
cg|gi+ce|ei, cg,ce∈C o
.
I Quantum state space:
D={ρ∈ L(HM), ρ† =ρ, Tr(ρ) =1, ρ≥0}.
I Operators and commutations:
σ-=|gihe|,σ+=σ-†=|eihg| σx=σ-+σ+=|gihe|+|eihg|; σy=iσ-−iσ+=i|gihe| −i|eihg|; σz =σ+σ-−σ-σ+=|eihe| − |gihg|; σx2=I,σxσy =iσz,[σx,σy] =2iσz, . . .
I Hamiltonian:HM/~=ωqσz/2+uqσx.
I Bloch sphere representation:
D=n
1
2 I+xσx+yσy+zσz (x,y,z)∈R3, x2+y2+z2≤1o
| g
| e ω
qu
q12 / 39
The Markov ideal model (1)
C
B
D R
1R
2B R
2I When atom comes outB,|ΨiB of the full system isseparable
|ΨiB=|ψi ⊗ |gi.
I Just before the measurement inD, the state is in general entangled(not separable):
|ΨiR2 =USM |ψi ⊗ |gi
= Mg|ψi
⊗ |gi+ Me|ψi
⊗ |ei whereUSM is a unitary transformation (Schrödinger propagator) defining the linear measurement operatorsMgandMeonHS. SinceUSMis unitary,M†gMg+M†eMe=I.
13 / 39
The Markov ideal model (2)
C
B
D R
1R
2H S M
The unitary propagatorUSM is derived from Jaynes-Cummings HamiltonianHSMin the interaction frame.
Two kinds of qubit/cavity Hamiltonians:
resonant,HSM/~=i Ω(vt)/2
a†⊗σ-−a⊗σ+ , dispersive,HSM/~= Ω2(vt)/(2δ)
N⊗σz, whereΩ(x) = Ω0e−x
2
w2,x =vt withv atom velocity,Ω0vacuum Rabi pulsation,wradial mode-width and whereδ=ωq−ωc is the detuning between qubit pulsationωqand cavity pulsationωc (|δ| Ω0).
14 / 39
Dispersive and resonant Jaynes-Cummings propagators
The solution of
ıdtdU =−~iHSM(t)U, withU0=Ireads
I
for
HSM(t)/~=i f
(t) a†⊗ |g
ihe
| −a⊗ |e
ihg
|(resonant)
Ut =
cos
θt2
√N
⊗ |
g
ihg
|+cos
θt2
√N+I
⊗ |
e
ihe
|−a
sin
θt
2
√N
√N ⊗ |
e
ihg
| +sin
θt
2
√N
√N a†⊗ |
g
ihe
|.I
for
HSM(t)/~=f
(t)N⊗(|e
ihe
| − |g
ihg
|)(dispersive)
U(t) =exp
(iθ(t)N)⊗ |g
ihg
|+exp
(−iθ(t
)N)⊗ |e
ihe
|.where
θ(t) =Rt0
f
(τ)d
τ.
15 / 39
Quantum trajectories
Just beforedetectorDthe quantum state isentangled:
|ΨiR
2= (Mg|ψi)⊗ |gi+ (Me|ψi)⊗ |ei
Just afteroutcomey, the state becomesseparable3:
|ΨiD =
My q
hψ|M†yMy|ψi|ψi
⊗ |yi.
Outcomeyobtained with probabilityPy =hψ|M†yMy|ψi..
Quantum trajectories(Markov chain, stochastic dynamics):
|ψk+1i=
Mg q
hψk|M†gMg|ψki|ψki, yk =gwith probabilityhψk|M†gMg|ψki;
Me
√
hψk|M†eMe|ψki|ψki, yk =ewith probabilityhψk|M†eMe|ψki;
with state|ψkiand measurement outcomeyk ∈ {g,e}at time-stepk:
3Measurement operatorO=IS⊗(|eihe| − |gihg|).
16 / 39
Quantum Non Demolition (QND) measurement of photons
4|ΨiR
2 =UR2UCUR1 |ψi ⊗ |gi
C
B
D R1
R2
B R2
UR1 =IS⊗|gi+|ei
√2
hg|+−|gi+|ei
√2
he| UC =e−i
φ0 2N
⊗ |gihg|+ei
φ0 2N
⊗ |eihe| UR2 =UR1
UR1 |ψi ⊗ |gi
=√1
2(|ψi ⊗ |gi+|ψi ⊗ |ei) UCUR1 |ψi ⊗ |gi
=√1
2
e−i
φ0 2N
|ψi
⊗ |gi+
ei
φ0 2N
|ψi
⊗ |ei
|ΨiR
2 =12
e−i
φ0 2N
|ψi
⊗(|gi+|ei) +
ei
φ0 2N
|ψi
⊗(−|gi+|ei)
=
−isin(φ20N)|ψi
⊗ |gi+
cos(φ20N)|ψi
⊗ |ei ThusMg=−isin(φ20N)andMe=cos(φ20N).
Quantum Monte-Carlo simulations with MATLAB:
IdealModelPhotonBoxWaveFunction.m
4M. Brune, . . . : Manipulation of photons in a cavity by dispersive atom-field coupling: quantum non-demolition measurements and generation of "Schrödinger cat"
states . Physical Review A, 45:5193-5214, 1992.
17 / 39
The Markov ideal model (3)
Just beforeD, the field/atom state isentangled:
Mg|ψi ⊗ |gi+Me|ψi ⊗ |ei
Denote byµ∈ {g,e}the measurement outcome in detectorD: with probabilityPµ=hψ|M†µMµ|ψiwe getµ. Just after the measurement outcomeµ=y,the state becomes separable:
|ΨiD=√1
Py
(My|ψi)⊗ |yi= q My
hψ|M†yMy|ψi|ψi
!
⊗ |yi. Markov process(density matrix formulationρ∼ |ψihψ|)
ρ+=
MgρM†g
Tr(MgρM†g), with probabilityPg= Tr
MgρM†g
;
MeρM†e
Tr(MeρM†e), with probabilityPe= Tr
MeρM†e .
Kraus map:
E
(ρ+/ρ) =K(ρ) =MgρM†g+MeρM†e.18 / 39
LKB photon-box: Markov process with detection efficiency
I With pure stateρ=|ψihψ|, we have ρ+=|ψ+ihψ+|= 1
Tr
MµρM†µMµρM†µ
when the atom collapses inµ=g,ewith proba. Tr
MµρM†µ .
I Detection efficiency:the probability to detect the atom is
η∈[0,1]. Three possible outcomes fory:y =gif detection ing, y =eif detection ineandy =0 if no detection.
The only possible update is based onρ: expectationρ+of|ψ+ihψ+| knowingρand the outcomey ∈ {g,e,0}.
ρ+=
MgρM†g
Tr(MgρMg) ify=g, probabilityηTr(MgρMg) MeρM†e
Tr(MeρMe) ify=e, probabilityηTr(MeρMe)
MgρM†g+MeρM†e ify=0, probability 1−η
ρ+does not remain pure: the quantum stateρ+becomes a mixed state;|ψ+ibecomes physically irrelevant.
19 / 39
LKB photon-box: Markov process with detection errors (1)
I With pure stateρ=|ψihψ|, we have ρ+=|ψ+ihψ+|= 1
Tr
MµρM†µMµρM†µ
when the atom collapses inµ=g,ewith proba. Tr
MµρM†µ .
I Detection error rates: P(y =e/µ=g) =ηg∈[0,1]the probability of erroneous assignation toewhen the atom collapses ing;P(y =g/µ=e) =ηe∈[0,1](given by the contrast of the Ramsey fringes).
Bayes law: expectationρ+of|ψ+ihψ+|knowingρand the imperfect detectiony.
ρ+=
(1−ηg)MgρM†g+ηeMeρM†e
Tr((1−ηg)MgρM†g+ηeMeρM†e)ify=g, prob. Tr
(1−ηg)MgρM†g+ηeMeρM†e
; ηgMgρM†g+(1−ηe)MeρM†e
Tr(ηgMgρM†g+(1−ηe)MeρM†e)ify=e, prob. Tr
ηgMgρM†g+ (1−ηe)MeρM†e .
ρ+does not remain pure: the quantum stateρ+becomes a mixed
state;|ψ+ibecomes physically irrelevant. 20 / 39
LKB photon-box: Markov process with detection errors (2)
We getρ+=
(1−ηg)MgρM†g+ηeMeρM†e
Tr((1−ηg)MgρM†g+ηeMeρM†e), with prob. Tr
(1−ηg)MgρM†g+ηeMeρM†e
; ηgMgρM†g+(1−ηe)MeρM†e
Tr(ηgMgρM†g+(1−ηe)MeρM†e) with prob. Tr
ηgMgρM†g+ (1−ηe)MeρM†e .
Key point:
Tr
(1−ηg)MgρM†g+ηeMeρM†e
and Tr
ηgMgρM†g+ (1−ηe)MeρM†e are the probabilities to detecty =g ande, knowingρ.
Generalizationby merging aKraus mapK(ρ) =P
µMµρM†µwhere P
µM†µMµ=I with aleft stochastic matrix(ηµ0,µ):
ρ+= P
µηy,µMµρM†µ Tr
P
µηy,µMµρM†µ when we detecty =µ0. The probability to detecty =µ0knowingρis Tr
P
µηy,µMµρM†µ .
21 / 39
Cavity decay (decoherence) seen as unread measurements The cavity mirrors play the role of a detector with two possible outcomes:
I zero photon annihilation
during
∆T: Kraus operator
M0=I−∆T2 L†−1L−1, probability
≈Tr
M0ρtM†0
with back action
ρt+∆T ≈ M0ρtM† 0
Tr
M0ρtM†0
.
I one photon annihilation
during
∆T: Kraus operator
M−1=√∆TL−1
, probability
≈Tr
M−1ρtM†−1
with back action
ρt+∆T ≈ M−1ρtM†
−1
Tr
M−1ρtM†−1
where
L−1= q 1
Tcava
is the
Lindbald operator associated to cavity damping(see bellow the continuous time models) with T
cavthe photon life time and
∆TT
cavthe sampling period (T
cav =100 ms and
∆T ≈
100
µsfor the LKB photon Box).
22 / 39
Cavity decoherence: cavity decay, thermal photon(s)
Three possible outcomes:I zero photon annihilationduring∆T: Kraus operator M0=I−∆T2 L†−1L−1−∆T2 L†1L1, probability≈ Tr
M0ρtM†0 with back actionρt+∆T ≈ TrM(M0ρ0ρtMtM†0†0).
I one photon annihilationduring∆T: Kraus operator M−1=√
∆TL−1, probability≈ Tr
M−1ρtM†−1
with back action ρt+∆T ≈ M−1ρtM
†
−1
Tr(M−1ρtM†−1)
I one photon creationduring∆T: Kraus operatorM1=√
∆TL1, probability≈ Tr
M1ρtM†1
with back actionρt+∆T ≈ TrM(M1ρ1tρMtM†1†1) where
L−1= q1+nth
Tcav a, L1=q
nth
Tcava†
are theLindbald operators associated to cavity decoherence:Tcav the photon life time,∆T Tcav the sampling period andnthis the average of thermal photon(s) (vanishes with the environment
temperature) (nth≈0.05 for the LKB photon box). 23 / 39
Experimental results
5Valeur moyenne du nombre de photons le long d’une longue séquence de mesure:
observation d’une trajectoire stochastique
Une trajectoire correspondant au résultat initial n=5
Sauts quantiques vers le vide dus à l’amortissement du champ Des mesures répétées
confirment n=5
Projection de l’état cohérent sur n=5
nNombre moyen de photons A partir de la probabilité Pi(n) inférée après chaque atome, on déduit le nombre moyen de photons:
Première observation des trajectoires stochastiques du champ, en très bon accord avec les prédictions théoriques (simulations
de Monte- Carlo. Voir cours précédents).
n = nPi(n)
n
! (6"10)
See the quantum Monte Carlo simulations of the Matlab script:
RealisticModelPhotonBox.m.
5From Serge Haroche, Collège de France, notes de cours 2007/2008.
24 / 39
Feedback: Kraus operators depend on the control input
u6u=0: dispersiveinteraction with
Mg(0) =cos
φ0N+φR
2
andMe(0) =sin
φ0N+φR
2
,
u=1: resonantinteraction with atom prepared in|ei
Mg(1) = sinθ
0 2
√ N
√
N a†andMe(1) =cos θ0
2
pN+I
u=−1: resonantinteraction with atom prepared in|gi
Mg(−1) =cosθ
0 2
√ N
andMe(−1) =−a sinθ
0 2
√ N
√ N (φ0, φR, θ0)are constant parameters.
6Zhou, X.; Dotsenko, I.; Peaudecerf, B.; Rybarczyk, T.; Sayrin, C.; S.
Gleyzes, J. R.; Brune, M.; Haroche, S. Field locked to Fock state by quantum feedback with single photon corrections. Physical Review Letter, 2012, 108, 243602.
25 / 39
The Lyapunov feedback law
I Computeuk as a function ofρk such thatρk converges towards the goal|¯nih¯n|.
I Lyapunov functionV(ρ) = Tr (N−¯n)2ρ
for example: when uk =0 we have
E
(V(ρk+1/ ρk,uk =0) =V(ρk) (martingale)I Lyapunov control: chooseuk in{0,1,−1}at each stepk in order to minimizeu7→
E
(V(ρk+1/ ρk,u).I In closed-loop,
E
(V(ρk+1)will be decreasing. It is reasonable to guess thatV(ρk)tends to 0, i.e., thatρk converges to|n¯ihn¯|.26 / 39
Closed-loop experimental results
Zhou et al. Field locked to Fock state by quantum feedback with single photon corrections.
Physical Review Letter, 2012, 108, 243602.
See the closed-loop quantum Monte Carlo simulations of the Matlab script:RealisticFeedbackPhotonBox.m.
27 / 39
Stochastic Master Equation (SME) and quantum filtering
Discrete-time modelsareMarkov processesρk+1= Kyk(ρk)
Tr(Kyk(ρk)), with proba.pyk(ρk) = Tr(Kyk(ρk)) where eachKy is a linear completely positive map admitting the expression
Ky(ρ) =X
µ
My,µρM†y,µ with X
y,µ
M†y,µMy,µ=I. K =P
yKy corresponds to aKraus maps(ensemble average, quantum channel)
E
(ρk+1|ρk) =K(ρk) =Xy
Ky(ρk).
Quantum filtering (Belavkin quantum filters)
data: initial quantum stateρ0, past measurement outcomes yl forl ∈ {0, . . . ,k−1};
goal: estimation ofρk via the recurrence (quantum filter) ρl+1= Kyl(ρl)
Tr(Kyl(ρl)), l =0, . . . ,k−1.
28 / 39
Continuous/discrete-time Stochastic Master Equation (SME)
Discrete-time modelsareMarkov processesρk+1= Kyk(ρk)
Tr(Kyk(ρk)), with proba.pyk(ρk) = Tr(Kyk(ρk)) associated toKraus maps(ensemble average, quantum channel)
E
(ρk+1|ρk) =K(ρk) =Xy
Ky(ρk)
Continuous-time modelsarestochastic differential systems dρt = −~i[H, ρt] +X
ν
LνρtL†ν−12(L†νLνρt+ρtL†νLν)
! dt
+X
ν
√ην
Lνρt+ρtL†ν− Tr
(Lν+L†ν)ρt ρt
dWν,t driven byWiener process7dWν,t =dyν,t−√ην Tr
(Lν+L†ν)ρt dt with measuresyν,t, detection efficienciesην ∈[0,1]and
Lindblad-Kossakowskimaster equations (ην ≡0):
d
dtρ=−~i[H, ρ] +X
ν
LνρtL†ν−12(L†νLνρt+ρtL†νLν)
7and/or Poisson processes, see next slides. 29 / 39
Ito stochastic calculus
Given a SDE
dXt =F(Xt,t)dt+X
ν
Gν(Xt,t)dWν,t,
we have the following chain rule summarized by the heuristic formulae:
dWν,t =O(√
dt), dWν,tdWν0,t =δν,ν0dt. Ito’s ruleDefiningft =f(Xt)aC2function ofX, we have
dft = ∂f
∂X Xt
F(Xt,t) +1 2
X
ν
∂2f
∂X2 Xt
(Gν(Xt,t),Gν(Xt,t))
! dt
+X
ν
∂f
∂X Xt
Gν(Xt,t)dWν,t.
Furthermore
E
d dtft
Xt
=
E
∂X∂fXt
F(Xt,t) +1 2
X
ν
∂2f
∂X2 Xt
(Gν(Xt,t),Gν(Xt,t))
! .
30 / 39
Continuous/discrete-time diffusive SME
With a single imperfect measuredyt=√ ηTr
(L+L†)ρt
dt+dWtand detection efficiencyη∈[0,1], the quantum stateρtis usually mixed and obeys to
dρt=
−i
~[H, ρt] +LρtL†−1 2
L†Lρt+ρtL†L dt +√
η
Lρt+ρtL†− Tr
(L+L†)ρt
ρt
dWt
driven by the Wiener processdWt(Gaussian law of mean 0 and variancedt).
WithIt¯o rules, it can be written as the following "discrete-time" Markov model
ρt+dt =
MdytρtM†dy
t + (1−η)LρtL†dt
Tr
MdytρtM†dy
t+ (1−η)LρtL†dt
withMdyt =I+
−i
~H−12 L†L
dt+√
ηdytL. The probability to detectdytis given by the following density
P
dyt∈[s,s+ds]
= Tr
MsρtM†s+ (1−η)LρtL†dt
√
2πdt e−s
2 2dt ds
close to a Gaussian law of variancedtand mean√ ηTr
(L+L†)ρt
dt.
31 / 39
A key physical example in circuit QED
8Superconducting qubit dispersively coupled to a cavity traversed by a microwave signal (input/output theory).
The back-action on the qubit state of a single measurement of both output fieldquadraturesIt
andQt is described by a simple SME for the qubit density operator.
dρt = −2i[uσx+vσy, ρt] +γ(σzρσz−ρt) dt +p
ηγ/2 σzρt+ρtσz−2 Tr(σzρt)ρt
dWtI+ıp
ηγ/2[σz, ρt]dWtQ withIt andQt given bydIt =p
ηγ/2 Tr(2σzρt)dt+dWtI and
dQt =dWtQ, whereγ≥0 is related to the measurementstrengthand η∈[0,1]is the detection efficiency.uandvare the two control inputs.
8M. Hatridge et al. Quantum Back-Action of an Individual Variable-Strength Measurement. Science, 2013, 339, 178-181.
32 / 39
Qubit, density matrix and Bloch sphere
With
|ψi=ψg|g
i+ψe|e
isatisfying
ı~dtd|ψi=H|ψi, the density operator corresponds to
ρ=|ψihψ|. Then
ρis non negative Hermitian operator such that Tr
(ρ) =1 and obeying to
d
dt
ρ=−ı~[H, ρ].
For mixed states,
ρ= I+xσx+yσ2 y+zσzwith
x
=Tr
(σxρ),y
=Tr
(σyρ)and z
=Tr
(σzρ).Then
(x,y
,z)
∈R3are the cartesian coordinates of vector M
~inside
Bloch sphere( Tr
ρ2=
x
2+y
2+z
2≤1):
d dt
M
~ = (u~ı+v~
)×M.
~is another formulation of
dtdρ=−2i[uσx+v
σy, ρ]. Hereu stands for the
rotation speedaround x -axis and v the rotation speed around y -axis.
33 / 39
QND measurement: asymptotic behavior in open-loop
Almost such convergence:Consider the SMEdρt = −2i[uσx+vσy, ρt] +γ(σzρσz−ρt) dt +p
ηγ/2 σzρt+ρtσz−2 Tr(σzρt)ρt
dWtI+ıp
ηγ/2[σz, ρt]dWtQ
withu=v =0 andη >0.
I For any initial stateρ0, the solutionρt converges almost surely ast → ∞to one of the states|gihg|or|eihe|.
I The probability of convergence to|gihg|(respectively|eihe|) is given bypg= Tr(|gihg|ρ0)(respectively Tr(|eihe|ρ0)).
Proof based on the martingalesVg(ρ) = Tr(|gihg|ρ) = (1−z)/2 and Ve(ρ) = Tr(|eihe|ρ) = (1+z)/2, and on the sub-martingale
V(ρ) = Tr2(σzρ) =z2:
E
dVg|ρt=
E
(dVe|ρt) =0,E
(dV|ρt) =2ηγ 1−z22dt ≥0.
Confirmed by the quantum Monte Carlo simulations:
IdealModelQubit.m
34 / 39
Adding decoherence due to finite life-time of
|eid
ρt = −2i[uσx+v
σy, ρt] +γ(σzρσz −ρt)dt
+pηγ/2 σzρt+ρtσz−
2 Tr
(σzρt)ρtdWtI+ıp
ηγ/2[σz, ρt]dWtQ +
LeρtL†e− 12
L†eLeρt +ρtL†eLe
dt
where
Le=p1/T1|gihe|
and T
1is the life time of the excited state
|e
i.
For u
=v
=0, convergence of all trajectories towards
|g
i, the ground state. Proof based on the super-martingales
V
e(ρ) =Tr
(|e
ihe
|ρ) = (1+z)/2:
E
(dVe|ρt) =−1 T
1V
edt
.Confirmed by the quantum Monte Carlo simulations:
RealisticModelQubit.m
35 / 39
Feedback stabilization of the excited state
dρt = −2i[uσx+vσy, ρt] +γ(σzρσz−ρt) dt +p
ηγ/2 σzρt+ρtσz−2 Tr(σzρt)ρt
dWtI+ıp
ηγ/2[σz, ρt]dWtQ
Withuandv arbitrary, we have forV(ρ) = Tr(σzρ) =z,
E
(dVt|ρt) =u Tr(σyρt)−v Tr(σxρt) =uy−vx.With the quantum state feedback
u= sign(y)
T (1−Vt), v =−sign(x)
T (1−Vt)
we get in closed loop
E
(dVt|ρt) = |x|+|y|T (1−Vt)and thusV tends to converge towards 1, i.e.,z tends to converge towards 1. Confirmed by the closed-loop simulations:IdealFeedbackQubit.mRobustness of such feedback illustrated by the more realistic simulations:
RealisticFeedbackQubit.m
36 / 39
Continuous/discrete-time jump SME
With Poisson processN(t),hdN(t)i=θ+ηTr VρtV† dt,and detection imperfections modeled byθ≥0 andη∈[0,1], the quantum stateρt is usually mixed and obeys to
dρt =
−i[H, ρt] +VρtV†−12(V†Vρt+ρtV†V) dt +
θρt+ηVρtV† θ+ηTr(VρtV†)−ρt
dN(t)−
θ+ηTr VρtV† dt
ForN(t+dt)−N(t) =1we haveρt+dt= θρt+ηVρtV† θ+ηTr(VρtV†). FordN(t) =0we have
ρt+dt= M0ρtM0†+ (1−η)VρtV†dt Tr
M0ρtM0†+ (1−η)VρtV†dt withM0=I+ −iH+12 η Tr VρtV†
I−V†V dt.
37 / 39
Continuous/discrete-time diffusive-jump SME
The quantum stateρtis usually mixed and obeys to dρt =
−i[H, ρt] +LρtL†−1
2(L†Lρt +ρtL†L) +VρtV†−1
2(V†Vρt+ρtV†V)
dt +√
η
Lρt+ρtL†− Tr
(L+L†)ρt
ρt
dWt
+
θρt+ηVρtV† θ+ηTr(VρtV†)−ρt
dN(t)−
θ+ηTr
VρtV† dt
ForN(t+dt)−N(t) =1we haveρt+dt= θρt+ηVρtV† θ+ηTr(VρtV†). FordN(t) =0we have
ρt+dt = MdytρtMdy†
t+ (1−η)LρtL†dt+ (1−η)VρtV†dt
Tr
MdytρtMdy†
t + (1−η)LρtL†dt+ (1−η)VρtV†dt
withMdyt =I+ −iH−12L†L+12 ηTr VρtV†
I−V†V dt+√
ηdytL.
38 / 39