On the first experimental realization of a quantum state feedback
55th IEEE Conference on Decision and Control Las Vegas, USA, December 12-14, 2016
Pierre Rouchon
Centre Automatique et Systèmes, Mines ParisTech, PSL Research University Quantic Research Team, Inria
1 / 31
Technologies for quantum simulation and computation
1© OBrien
Superconduc�ng circuits
Photons
© S. Kuhr
Ultra-cold neutral/Rydberg
atoms
© Bla� & Wineland
Trapped ions
© Pe�a
Quantum dots
© IBM
Requirement:
Scalable modular architecture
Control software from the very beginning
1Courtesy of Walter Riess, IBM Research - Zurich.
2 / 31
Nobel Prize in Physics 2012
Serge Haroche David J. Wineland
" This year’s Nobel Prize in Physics honours the experimental inventions and discoveries that have allowed themeasurement and control of individual quantum systems. They belong to two separate but related technologies: ions in a harmonic trap and photons in a cavity"
From the Scientific Background on the Nobel Prize in Physics 2012 compiled by the Class for Physics of the Royal Swedish Academy of Sciences, October 9, 2012.
3 / 31
Outline
The photon-box experiment of the Haroche group Modelling based on three quantum rules Measuring photons without destroying them Stabilizing measurement-based feedback
Model structures (decoherence, measurement back-action) Discrete-time models (hidden Markov chain)
Diffusive models (Wiener process) Jump models (Poisson process) Feedback for quantum systems
Measurement-based feedback
Coherent (autonomous) feedback and dissipation engineering
4 / 31
The first experimental realization of a quantum-state feedback microwave photons
(10 GHz)
Experiment:C. Sayrin, I. Dotsenko, X. Zhou, B. Peaudecerf, T. Rybarczyk, S. Gleyzes, P. Rouchon, M. Mirrahimi, H. Amini,M. Brune, J.M. Raimond,S. Haroche:
Real-time quantum feedback prepares and stabilizes photon number states.
Nature,2011, 477, 73-77.
Theory:I. Dotsenko,M. Mirrahimi,M. Brune,S. Haroche, J.M. Raimond,P. Rouchon:
Quantum feedback by discrete quantum non-demolition measurements: towards on-demand generation of photon-number states.Physical Review A,2009, 80:
013805-013813.
M. Mirrahimi et al. CDC 2009, 1451-1456,2009.
H. Amini et al. IEEE Trans. Automatic Control, 57 (8): 1918–1930,2012.
R. Somaraju et al., Rev. Math. Phys., 25, 1350001,2013.
H. Amini et. al., Automatica, 49 (9): 2683-2692,2013.
5 / 31
Three quantum features emphasized by the photon box
2 1. Schrödinger: wave funct.|ψiof norm one in Hilbert spaceHd
dt|ψi=−i
~H|ψi, H=H0+uH1, uclassical control input Dirac notations:|ψi ≡
•
•
,|ψi†=hψ| ≡ • •
,H=H†≡ • •
• •
. 2. Origin of dissipation: collapse of the wave packetinduced by the
measurement of observableOwith spectral decomp.P
yλyPy:
I measured outputy with probabilityPy =hψ|Py|ψi;
I back-action:
|ψi 7→ |ψi+= Py|ψi phψ|Py|ψi 3. Tensor product for composite system(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.
6 / 31
Composite system
(S,M): harmonic oscillator⊗qubit.
I SystemS
corresponds to a quantized harmonic oscillator:
HS = (∞
X
n=0
xn|ni
(xn)∞n=0∈l2(C) )
,
where
|niis the photon-number state with
nphotons (hn
1|n2i=δn1,n2).
I MeterM
is a qubit, a 2-level system:
HM =
xg |gi+xe|ei
xg,xe∈C
,
where
|gi(resp.
|ei) is the ground (resp. excited) state(hg|gi
=he|ei=1 and
hg|ei=0)
I State of the composite system|Ψi∈ HS⊗ HM
:
|Ψi=
+∞
X
n=0
xng |ni ⊗ |gi+xne |ni ⊗ |ei, xne,xng ∈C.
Ortho-normal basis:
|ni ⊗ |gi,|ni ⊗ |ein∈N
.
7 / 31
The hidden Markov chain (1)
C
B
D R
1R
2B R
2I When atom comes outB, the quantum state|ΨiBof the composite system isseparable:|ΨiB=|ψi ⊗ |gi.
I Just before the measurement inD, the state is in general entangled(not separable):
|ΨiR
2=USM |ψi ⊗ |gi
= Mg|ψi
⊗ |gi+ Me|ψi
⊗ |ei whereUSM is a unitary transformation (Schrödinger propagator) defining measurement operatorsMgandMeonHS.
USM unitary impliesM†gMg+M†eMe≡I.
8 / 31
The hidden Markov chain (2)
Just beforethe atom detectorDthe quantum state isentangled:
|ΨiR
2 =Mg|ψi ⊗ |gi+Me|ψi ⊗ |ei
Just afteroutcome3y ∈ {g,e}, the state becomesseparable:
|ΨiD= q My
hψ|M†yMy|ψi|ψi
!
⊗ |yi with probabilityhψ|M†yMy|ψi.
Hidden Markov chain:|ψi ⊗ |gi −→ q My
hψ|M†yMy|ψi|ψi
!
⊗ |yi
|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 outputyk ∈ {g,e}at time-stepk:
3Measurement operatorO=IS⊗(|eihe| − |gihg|).
9 / 31
Why density operators ρ instead of wave functions |ψi
Assume known|ψ0ianddetector out of order:what about|ψ1i?
I Expectation value of|ψ1ihψ1|knowing|ψ0i:4 E
|ψ1ihψ1|
| {z }
ρ1
|ψ0i
=Mg|ψ0ihψ0|
| {z }
ρ0
M†g+Me|ψ0ihψ0|
| {z }
ρ0
M†e.
I SetK(ρ),MgρM†g+MeρM†efor any operatorρ.
I ρk expectation of|ψkihψk|knowingρ0=|ψ0ihψ0|:
ρ1=K(ρ0), ρ2=K(ρ1), . . . ,ρk =K(ρk−1).
Linear mapK: positivity and trace preserving (M†gMg+M†eMe=I) calledKraus mapor quantum channel.
Density operatorsρ: Hermitian non-negative operators of trace one.
4|ψihψ|:orthogonal projector on line spanned by unitary vector|ψi.
10 / 31
The Markov chain with
ρas hidden state
Detector efficiencyη∈[0,1]. Measured outputy ∈ {g,e,∅}:
ρk+1=
Kg(ρk)
Tr Kg(ρk),yk =gwith probability Tr Kg(ρk)
; Ke(ρk)
Tr(Ke(ρk)),yk =ewith probability Tr(Ke(ρk));
K∅(ρk)
Tr(K∅(ρk)),yk =∅with probability Tr(K∅(ρk));
with Kraus maps
Kg(ρ) =ηMgρM†g, Ke(ρ) =ηMeρM†e K∅(ρ) = (1−η)
MgρM†g+MeρM†e . We still have:
E
ρk+1ρk,K(ρk) =MgρkM†g+MeρkM†e=Xy
Ky(ρk).
Imperfections, decoherence modeled similarly
11 / 31
A controlled Markov process (input
u, hidden stateρ, outputy)
Inputu: classical amplitude of a coherent micro-wave pulse Stateρ: the density operator of the photon(s) trapped in the cavity Outputy: measurement of the probe atom
ρk+1=
DukKg(ρk)D†uk
Tr Kg(ρk) ,yk =gwith probability Tr Kg(ρk)
; DukKe(ρk)D†uk
Tr(Ke(ρk)) ,yk =ewith probability Tr(Ke(ρk));
DukK∅(ρk)D†uk
Tr(K∅(ρk)) ,yk =∅with probability Tr(K∅(ρk));
Controlled displacement unitary operator(u∈R):Du=eu(a†−a)with a=upper diag(√
1,√
2, . . .)the photon annihilation operator.
Measurement Kraus operators with linear dispersive interaction Mg =cos
φ0N+φR
2
andMe=sin
φ0N+φR
2
:M†gMg+M†eMe=I with N=a†a=diag(0,1,2, . . .)the photon number operator.
12 / 31
Open-loop dynamics (u
=0): experimental data
y
diag(ρ) ρ
13 / 31
u =
0: Quantum Non Demolition (QND) measurement of photons.
ρk+1=
Kg(ρk)
Tr(Kg(ρk)),yk =gwith probability Tr Kg(ρk)
;
Ke(ρk)
Tr(Ke(ρk)),yk =ewith probability Tr(Ke(ρk));
K∅(ρk)
Tr(K∅(ρk)),yk =∅with probability Tr(K∅(ρk));
Photon-number state|¯nih¯n|is a steady-state:Ky(|¯nih¯n|)∝|¯nih¯n|.
MartingalesWg(ρ) = Tr(g(N)ρ) for any functiong:
E
Wg(ρk+1)/ρk=Wg(ρk).Convergence:W(ρ) =1−P
nhn|ρ|ni2super-martingale,
E
W(ρk+1)/ρk=W(ρk)−Q(ρk)
whereQ(ρ)≥0 andQ(ρ) =0 iff∃¯nsuch thatρ=|¯nih¯n|.
Probability to converge to|nih¯ n|: Tr¯ (|nih¯ n|ρ¯ 0) =hn|ρ¯ 0|¯ni.
14 / 31
Feedback stabilization around 3-photon state: experimental data
y
u
diag(ρ) ρ
15 / 31
Structure of the stabilizing quantum-state feedback scheme
With a sampling time of 80µs, the controller is classicalI Goal: stabilization towards|¯nih¯n|.
I Stepk−1 to stepk
1. readyk−1the measurement outcome for probe atomk−1.
2. computeρkfromρk−1viaρk =Duk−1Kyk−1(ρk−1)D†uk−1 Tr Kyk−1(ρk−1) , 3. computeuk as a function ofρk(state feedback).
4. apply the micro-wave pulse of amplitudeuk. Observer/controllerstructure:
1. real-time state estimation: the quantumBelavkinfilter 2. u=f(ρ)based on astrict control Lyapunov functionV(ρ)
derived from open-loop martingales Tr(g(N)ρ):
E
V(ρk+1)ρk,uk =f(ρk)=V(ρk)−Q(ρk) whereQ(ρ)≥0 andQ(ρ) =0 iffρ=|¯nih¯n|.
16 / 31
Experimental closed-loop data
C. Sayrin et. al., Nature 477, 73-77, Sept. 2011.Decoherence due to finite photon life-time (70 ms) Detection efficiency 40%
Detection error rate 10%
Delay 4 sampling periods The quantum filter includes decoherence, detector imperfections and delays (Bayes law).
Truncation to 9 photons
Stabilization around 3-photon state
V u y
17 / 31
Outline
The photon-box experiment of the Haroche group Modelling based on three quantum rules Measuring photons without destroying them Stabilizing measurement-based feedback
Model structures (decoherence, measurement back-action) Discrete-time models (hidden Markov chain)
Diffusive models (Wiener process) Jump models (Poisson process)
Feedback for quantum systems Measurement-based feedback
Coherent (autonomous) feedback and dissipation engineering
18 / 31
Discrete-time Stochastic Master Equations (SME)
Trace preserving Kraus mapKudepending onu:
Ku(ρ) =X
ξ
Mu,ξρM†u,ξ with X
ξ
M†u,ξMu,ξ=I.
Take aleft stochastic matrix ηy,ξ
and setKu,y(ρ) =P
ξηy,ξMu,ξρM†u,ξ. Associated Markov chain reads:
ρk+1= Kuk,yk(ρk)
Tr(Kuk,yk(ρk)) withyk with probability Tr(Kuk,yk(ρk)). Classical inputu, hidden stateρ, measured outputy.
Ensemble average given byKusince
E
ρk+1ρk,uk=Kuk(ρk).Markov model useful for:
1. Monte-Carlo simulations of quantum trajectories(decoherence, measurement back-action).
2. quantum filteringto get the quantum stateρk fromρ0and(y0, . . . ,yk−1) (Belavkin quantum filterdeveloped for diffusive models).
3. feedback design and Monte-Carlo closed-loop simulations.
19 / 31
Stochastic master equation driven by Wiener process
Lindblad equation(HHermitian,Larbitrary, depending onu):
d
dtρ,L(ρ) = −i
~[H,ρ]
| {z } Schrödinger
+LρL†−1
2(L†Lρ+ρL†L)
| {z } ,DL(ρ)decoherence For anyt≥0,ρ07→etL(ρ0) =ρt is a trace preserving Kraus map.
Continuous-time outputt7→ytwithdyt=√
ηTr (L+L†)ρt
dt+dWt: dρt =
−i
~[H,ρt] +LρtL†−1
2(L†Lρt+ρtL†L)
dt +√
η
Lρt+ρtL†− Tr
(L+L†)ρt ρt
dWt
driven by Wiener processdWt (ηdetection efficiency).
Another formulation withIt¯o differentiation rule:
ρt+dt = Kdyt(ρ)
Tr(Kdyt(ρ))with proba. density Tr(Kdyt(ρ))fordyt
Kdy(ρ) =e−dy
2 2dt
MdyρM†dy+ (1−η)LρtL†dt Mdy =I+ −i
~H−12 L†L dt+√
ηdyL.
20 / 31
Stochastic master equation driven by Poisson process
Lindblad equation: dtdρ,L(ρ) =−i
~[H,ρ] +LρL†−12(L†Lρ+ρL†L) Countert7→y(t)∈N(detection imperfectionsϑ≥0,η∈[0,1]):
dy(t) =0or1:
E
dy(t)ρt=ϑ+ηTrLρtL† dtDynamics:
dρt =
−i
~[H,ρt] +LρtL†−1
2(L†Lρt+ρtL†L)
dt
+ ϑρt+ηLρtL† ϑ+ηTr LρtL†−ρt
!
dy(t)−
ϑ+ηTr
LρtL† dt
dy(t) =0:ρt+dt = K0(ρt)
Tr(K0(ρt))with probability Tr(K0(ρt)) dy(t) =1:ρt+dt = K1(ρt)
Tr(K1(ρt))with probability Tr(K1(ρt)) K0(ρ) =
M0ρM†0+ (1−η)LρL†dt
(1−ϑdt), K1(ρ) = ϑρ+ηLρL† dt M0=I+ −i
~H−12L†L dt
21 / 31
Outline
The photon-box experiment of the Haroche group Modelling based on three quantum rules Measuring photons without destroying them Stabilizing measurement-based feedback
Model structures (decoherence, measurement back-action) Discrete-time models (hidden Markov chain)
Diffusive models (Wiener process) Jump models (Poisson process)
Feedback for quantum systems Measurement-based feedback
Coherent (autonomous) feedback and dissipation engineering
22 / 31
Measurement-based feedback
system
controller
quantum world
classical world
y
u
decoherence
P-controller (Markovian feedbacka):
foru(t) =f(y(t)), average closed-loop dynamics ofρremains governed by a Lindblad master equation.
PID controller:no Lindblad master equation in closed-loop;
Nonlinear hidden-state stochastic systems:convergence analysis, Lyapunov exponents, dynamic output feedback, delays, robustness, . . .
aH.M. Wiseman:Quantum Trajectories and Feedback. PhD Thesis,
University of Queensland, 1994.
Short sampling times limit feedback complexity
23 / 31
First SISO measurement-based feedback for a superconducting qubit5
Analog Multiplier
d Feedback circuit
X*Y
Rabi reference 3.0 MHz 10 MHz
Phase error = Feedback signal X Y
a Signal generation
Rabi drive c Homodyne setup
I LO Readout
drive Q
LO Digitizer/
computer Q RF
RF I
b Dilution refrigerator
Transmon qubit
IN
(x2) Paramp HEMT
Q Q
OUT |0>
I Readout
cavity
|1>
|0>
|1> I
5R. Vijay, . . . , I. Siddiqi.Stabilizing Rabi oscillations in a superconducting qubit using quantum feedback.Nature 490, 77-80, October 2012.
24 / 31
First MIMO measurement-based feedback for a superconducting qubit6
FM Rabi
b)
~ ~
FM
~ ~
~
Rabi
JPC out in
Drift
a) ,
c) 3 inputs
2 outputs
6P. Campagne-Ibarcq, . . . , B. Huard:Using Spontaneous Emission of a Qubit as a Resource for Feedback Control.Phys. Rev. Lett. 117(6), 2016.
25 / 31
Measurement-based feedback stabilization of √1
2(|gi⊗|ei+|ei⊗|gi) 7
Alice Bob
7D. Ristè,. . . , L. DiCarlo:Deterministic entanglement of superconducting qubits by parity measurement and feedback. Nature 502, 350-354 (2013).
26 / 31
Coherent (autonomous) feedback (dissipation engineering)
Quantum analogue of Watt speed governor: adissipative mechanical system controls another mechanical system8system
controller quantum world classical world
decoherence
decoherence
u
y
u
cy
cOptical pumping (Kastler1950), coherent population trapping (Arimondo1996) Dissipation engineering, autonomous feedback: (Zoller, Cirac, Wolf, Verstraete, Devoret, Schoelkopf, Siddiqi, Lloyd, Viola, Ticozzi, Mirrahimi, Sarlette, ...)
(S,L,H) theoryandlinear quantum systems: quantum feedback networks based on stochastic Schrödinger equation, Heisenberg picture (Gardiner, Yurke, Mabuchi, Genoni, Serafini, Milburn, Wiseman, Doherty,Gough, James, Petersen, Nurdin, Yamamoto, Zhang, Dong, . . . )
Stability analysis: Kraus maps and Lindblad propagators are always contractions (non commutative diffusion and consensus).
8J.C. Maxwell:On governors.Proc. of the Royal Society, No.100, 1868.
27 / 31
Reservoir engineering to stabilize "Schrödinger cats"|αi+|-αi9 Cavity mode (system)
atom (controller) R1
R2
Box of atoms
Aim:
engineer atom-mode interaction, to stabilize |-α +|α
DC field:
(controls atom frequency) ENS experiment
Jaynes-Cumming Hamiltionian
H(t)/~=ωca†a⊗IM+ωq(t)IS⊗σz/2+iΩ(t) a†⊗σ-−a⊗σ+ /2 with the open-loop controlt7→ωq(t)combiningdispersiveωq6=ωc andresonantωq=ωcinteractions.
Key issues:convergenceofρk+1=K(ρk) =MgρkM†g+MeρkM†e.
9A. Sarlette et al:Stabilization of nonclassical states of the radiation field in a cavity by reservoir engineering.Phys. Rev. Lett. 107(1), 2011.
28 / 31
Autonomous feedback stabilization of
√12(|gi⊗|ei−|ei⊗|gi)10
Lindblad master equation:
d
dtρ=−i[H(t),ρ] +κDa(ρ) + 1
T1ADσA
−(ρ) + 1
2TφADσA z(ρ) + 1
T1BD
σB−(ρ) + 1
2TφBD
σBz(ρ) with
H(t)/~=χ
A
2σzA+χ2BσBz a†a +2ccos
χA+χB
2 t a+a† + Ω0
σAx+σBx +Ωn
e−in
χA+χB 2 t
(σ+A−σ+B) +h.c.
10S. Shankar, . . . , M.H. Devoret.Autonomously stabilized entanglement between two superconducting quantum bits.Nature, 504: 419-422, 2013.
29 / 31
Combining measurement-based feedback with dissipation engineering11
11Y. Liu, . . . , M.H. Devoret:Comparing and combining measurement-based and driven-dissipative entanglement stabilization.Phys. Rev. X 6, 2016.
30 / 31
Concluding remarks
I Three key quantum rules (highlighted by Haroche photon-box) 1. Schrödinger equationsdefining unitary transformations, 2. partial collapseof the wave packet: dissipation induced by
measurement of observables withdegeneratespectra, 3. tensor productforcomposite systems,
explain structure ofstochastic master equations, illustrate importance ofspin/spring models.
I Feedback control ofcoherenceandentanglementin composite systems: important issues for quantum computing, simulation, metrology and communication.
I Composite systems:curse of dimensionalityfor quantum networks with delays in coherent feedback loops.
Importance of collaborations with experimental quantum physicists for defining relevant control engineering questions
31 / 31