Beruflich Dokumente
Kultur Dokumente
to the system Hamiltonian of interest, we can emulate the forward and backward time propagation, where the
ancilla plays the role of a quantum clock. Within this scheme, the continuous evolution of the entire system
(the system of interest and the ancilla) is governed by a time-independent Hamiltonian. Our protocol is immune
to errors that could occur when the direction of time evolution is externally controlled by a classical switch.
tanglement entropy [41] and spectrum [42]. In this paper, the With the cavity-QED implementation, the ancilla qubit can
ancilla, in addition to playing the role of the quantum clock, also be realized with the global cavity photon mode as
has the added benefit of being the probe of the system. Specif- z =12a a, if the cavity photon state is restricted in the 0-
ically, we show how the OTO could be obtained by measuring and 1-photon subspace. Hence the total Hamiltonian of the
the same ancilla. coupled system can also be expressed as
The primary advantage of our protocol utilizing a quantum
clock for both control and readout of the many-body states is Htot = (1 2a a) H. (2)
its robustness against statistical errors, such as imperfect rota-
tion, in each shot of the experiments. In particular, our quan- From now on, we call both the cavity and the ancilla qubit
tum clock does not modify the many-body Hamiltonian in- as ancilla without further specification, since they play the
situ, which is in contrast to a previous proposal of measuring same role and one can use either them for the protocol.
the same correlator using a classical switch to continuously In Eq. (1) and (2), the ancilla only dresses the many-body
tune the prefactor of the Hamiltonian [14]. In addition, we system H, and does not exchange excitations (photons) with
are also able to construct a local Hamiltonian, which is more the many-body system. Crucially, if the H we consider is a
physical from condensed matter and quantum field theoretic local Hamiltonian, the ancilla does not mediate long-range in-
viewpoint, and may also exhibit richer behavior of quantum teraction between the particles/spins in the many-body system
scrambling. and preserves the locality of H.
We also provide simple examples of embedding such an The only thing that the ancilla does is to control the overall
ancilla in cavity-QED systems, for both a non-local all-to-all sign of the many-body Hamiltonian H quantum coherently. If
coupled spin models and a local XY-spin or extended Bose- the cavity contains no photon, namely the ancilla is in state
Hubbard model. In the non-local model, qubits (spins) are | 0a i [44], the overall sign is +; if the cavity contains one
interacting with each other mediated by a passive cavity bus, photon, namely the ancilla is in state | 1a i, the overall sign is
which is itself dispersively coupled to another ancilla cav- . If we consider the dynamics of the coupled system, we
ity in order to control the sign of the Hamiltonian. To re- can express the evolution operator as
alize the local model, local cavities/resonators are coupled
by intermediate qubits, which are themselves coupled to a Utot (t) = eiHtot t = eiHt | 0a ih 0a | + eiHt | 1a ih 1a |. (3)
global cavity. When integrating out the qubit degrees of free-
This means that the many-body system H evolves forward in
dom and with proper choosing of parameters, the effective
time if the cavity contains no photon, and backward in time
Hamiltonian has an overall sign controlled by the state of the
if the cavity contains one photon. Namely the cavity pho-
global cavity. Such models can be realized with recently de-
ton number a a or the ancilla qubit z acts a binary quantum
veloped experimental platforms such as circuit-QED network
clock that controls the arrow of time. More interestingly
[2124, 26, 27, 43] and qubit/atomic array in a 3D cavity [28].
since the clock is quantum, the system can be in a parallel
The outline of our paper is as follows. In Sec. II, we present
superposition of evolving both forward and backward in time,
our general protocol of measuring the OTO correlator with a
for example when we prepare the clock being in the super-
quantum clock. In Sec. III, we show how such a quantum
position state 12 (| 0a i + | 1a i).
clock could be embedded in a physical model. In Sec. IV, we
discuss the implementation of the protocol with circuit-QED Now we discuss a general protocol to measure the out-
systems. In Sec. V, we analyze the stability of our protocol of-time-order (OTO) correlator hO2 (t)O1 (0)O2 (t)O1 (0)i intro-
against imperfections. We present the generalization of ap- duced earlier, where O1 and O2 are certain operators, and
proach for extended Bose-Hubbard model and disordered spin O(t) = eiHt OeiHt is the Heisenberg evolved operator. The
chains in Sec. VI. We provide the conclusion and outlook in average hi could be with respect to a certain initial state
Sec. VII. In Appendix A, we list the complete sequence of | iS or an ensemble average over a thermal density matrix
P H
gates in the protocol. We show the details of the experimen- S = S e Z | iS S h |, where Z is the partition function. For
tal realization of the local model which we construct in the the sake of convenience, we will focus on average with re-
main text with a circuit-QED network or a qubit array in a 3D spect to a given pure state |iS . If one is interested in average
cavity in Appendix B. In Appendix C, we compare the numer- with respect to a thermal ensemble, one can still work with a
ical diagonalization of the original and second-order effective pure state that is obtained by time-evolving an initial finite-
Hamiltonian. Finally, in Appendix D, we provide a complete energy density pure state with respect to H [45]. Assuming
formula of the second-order dispersive Hamiltonian we men- that the system is generic (non-integrable), the pure state av-
tion in Sec. III without integrating out the qubits. erage is then expected to match the thermal ensemble average
at a temperature determined by the energy density of the state
[4547].
II. GENERAL SCHEME In the Schrodinger picture, the correlator corresponding to
a particular initial state can be written as
We consider a many-body system governed by Hamiltonian
S h |e O2 e O1 eiHt O2 eiHt O1 | iS .
iHt iHt
H and couple it globally to an ancilla qubit z , with the total
Hamiltonian being
To measure this correlator, we apply the following Ramsey
Htot = z H. (1) interferometry protocol as illustrated in Fig. 1:
3
arrow of time
cancel
Measure
cancel
...
...
...
...
FIG. 1: (a) Illustration of the Ramsey interferometry protocol. The interferometry starts from the left, with the initial state | iS | 0a i. The
Hadamard rotation splits the time evolution of the many-body state | iS into two branches, conditioned by the ancilla. The time evolution
conditioned by ancilla state | 0a i (| 1a i) is forward (backward) in the beginning. After applying the x operations, the ancilla states on the
two branches interchange, and so are the directions of time evolution. The red dashed lines show the canceled time evolution. Conditional
operations O1 and O2 on either branch are applied. A final measurement of the ancilla in the x- and y-basis gives the real and imaginary part
of the OTO correlator. We emphasize that the actual experimental time always goes from left to right. (b) The quantum circuit description of
the same protocol.
1. Start with the many-body system in the state | iS with 5. Apply a conditional-O2 on the lower (| 1a i) branch:
respect to which we wish to measure the OTO cor-
relator. Thus, the coupled system can expressed as CO2 ,1 = O2 | 1a ih 1a | + IS | 0a ih 0a |. (6)
| iS | 0a i.
6. In order to reverse the arrow of time in both branches,
2. Apply a Hadamard gate, i.e. a /2-rotation (pulse) we simply apply a x operator (-pulse around the x-
around the y-axis to the ancilla state: axis) to flip the ancilla. Then we let the coupled system
The coupled system is thus prepared in the superposed evolve for a period of 2t and reach the state
state 12 | iS [| 0a i + | 1a i]. From now on, the evolu-
1
tion of the many-body system split into two branches, [e2iHt O2 eiHt O1 | iS | 0a i + e2iHt eiHt | iS | 1a i].
conditioned by the ancilla state | 0a i and | 1a i respec- 2
tively.
7. Perform the previous steps (3-6) with reversal order
3. Apply a conditional operation (with conditioned operations on the other branch) as
shown in Fig. 1(a), the coupled system ends up with
CO1 ,1 = O1 | 1a ih 1a | + IS | 0a ih 0a |, (4) the final state
so that O1 is applied only to the lower branch of the in- 1
| f i = [| R i | 1a i + | L i | 0a i], (7)
terferometer conditioned by the ancilla state | 1a i. The 2
coupled system forms an entangled state
where we have abbreviated the wavefunctions in two
1 branches as
[O1 | iS | 1a i + | iS | 0a i].
2 | R i eiHt O2 eiHt O1 | iS , | L i O1 eiHt O2 eiHt | iS .
4. Let the system evolve with total Hamiltonian Htot for
time t according to Utot (t) represented in Eq. (3). The 8. Measure the expectation value of x operator under the
coupled system is now in an entangled state of evolv- final state | f i, which effectively takes an overlap be-
ing forward and backward in time conditioned by the tween the many-body states in the two branches of the
photon number, namely interferometer and leads to
1 h x i f h f | IS x | f i = Re[h L | R i]
[eiHt O1 | iS | 1a i + eiHt | iS | 0a i]. (5)
2 =Re[S h |eiHt O2 eiHt O1 eiHt O2 eiHt O1 | iS ]. (8)
4
The outcome is the real part of the OTO correlator. Sim- (a)
(ancilla)
ilarly, one can extract the imaginary part by measuring
y , since hy i f = Im[h L | R i]. (bus)
The first term at the second order is the so-called quantum- and will be eventually integrated out. Note that this is differ-
bus interaction, i.e., the flip-flop interaction mediated by the ent than the non-local Hamiltonian in the previous subsection
virtual photon in the coupling cavity bus [20, 51]. The second where ~ s were active degrees of freedom while b j were pas-
term represents the Lamb shift induced by the cavity bus. The sive. In addition, similar to the non-local case, the qubits are
prefactors of both terms depend on the detuning b,na , which coupled to a global cavity (described by photon operator a),
is controlled by the ancilla state | na i. In order to reverse the which will serve as the ancilla. We proceed as before, and split
sign of these prefactors, we chose the cross-Kerr nonlinearity the entire Hamiltonian H s into two parts, i.e. H s = H0 + V:
such that b,1 = b,0 = b , which leads to the condition:
= 2( b ) 2b .
X 1 X z
(12) H0 =b bj b j + + Hdisp ,
j
2 j j, j+1
When enforcing this condition, the effective Hamiltonian in X
the rotating frame with frequency can be written as Hdisp =a a zj, j+1 ,
j
X 1 g2j
X
V =gb [bj (j, j+1 + j1, j ) + H.c.].
X g g 0
(15)
(+j j0 + H.c.) +
j j
Heff =(1 2a a) zj
j< j0
b j
2 b j
4
g j In place of the cross-Kerr interaction in Eq. (9), H0 now con-
+ O 3 . (13) tain a term Hdisp which represents the dispersive interaction
b
between the global cavity (a) and the qubits () with interac-
Here, the effective Hamiltonian has exactly the form sug- tion strength , and is also sometimes called dispersive shift.
gested in Eq. (2), and the arrow of time is controlled by the For convenience, we define is the renormalized frequency of
ancilla photon number a a = 0 or 1 as desired. As shown the qubits, with the Lamb shift due to the global cavity already
above, the Hamiltonian controlled by the ancilla is an all-to- absorbed into the definition.
all coupled XY model in the presence of external field (corre- We note that the dispersive interaction Hdisp can arise, e.g.,
sponding to the Lamb shift term). One can also easily realize from a Jaynes-Cummings interaction in the dispersive regime
disorder in the coupling strengths. Additional ZZ-interaction [16, 20], where we get the dispersive shift = g2a /a . Here, ga
arises in the fourth-order perturbation [35, 53] [and the em- is the JC interaction strength and a is the detuning between
bedding of the ancilla is also realized once Eq. (12) is satis- bare qubit () and global cavity (a) frequencies. For weakly-
fied]: anharmonic superconducting qubits such as transmons, the
derivation of dispersive interaction can be found in Ref. [49].
X 2g2j g2j0 Similar to the non-local case, the photon number a a is con-
VZZ = (1 2a a)
zj zj0 . (14) served, and we again restrict to 0- and 1-photon sectors. In the
j< j0
3b following, we want to eliminate the qubit degrees of freedom
() perturbatively and find an effective Hamiltonian that local
The ZZ-interaction strength can be made stronger than this cavities (b) form an XY model of which the sign is determined
if one uses the transmon qubits [54], where the third-level of by the ancilla photon number.
transmon contribute significantly to the ZZ interaction [35]. We consider the dispersive regime where the local cavities
Finally, we note that the presence of the Lamb shift is cru- and qubits are far detuned in both ancilla sectors, compared to
cial for implementing the controlled operations mentioned in the JC interaction strength, namely
Sec. II, as will be explained in detail in Sec. IV.
b,na = + na b gb (na = 0, 1).
(qubit) g2b X
Heff = (1 2a a) [(bj b j+1 + H.c.) + 2bj b j ]
b j
(local cavity) 4
gb
(b) (c) + O 3 . (20)
b
A. Classical switch
arrow of time
1
0
switch
0.9
-0.05
0.8 -0.1
Relative error
-0.15
0.7
-0.2
Measure
0.6 -0.25
0.2
FIG. 5: Measurement protocol using a classical switch to control the
0.1
arrow of time. An ancilla qubit is initialized as the superposition of
| 0a i and | 1a i and hence split the evolution into two branches in order 0
0 2 4 6 8 10 12 14 16 18 20
to do the Ramsey interference. The ancilla enables conditional-O1
operation but does not control the sign of the Hamiltonian. Another
classical switch (such as the detuning) is used to change the sign of
the Hamiltonian and hence flip the arrow of time. FIG. 6: Effect of imperfect sign change via classical switch for a
P
spin model. The model considered here is H= i ~ i .~
i+1 + hi zi
where hi are chosen randomly from a uniform distribution in
Another potential issue with this approach is that for a lo- the interval [0.5, 0.5]. The main figure shows the correlator
cal Hamiltonian such as the XY-spin model described above, hO2 (t)O1 (t )O2 (t)O1 (0)i with O1 =z2 and O2 =zL1 where L=12 is
changing the sign of the total Hamiltonian requires one to the total number of sites. We take to be random Gaussian vari-
change the sign of each individual local term in the Hamil- able with variance and averaging in hO2 (t)O1 (t )O2 (t)O1 (0)i is
tonian separately, by controlling the detuning in-situ site by performed over this ensemble. The inset shows the relative error
site, and therefore, it is not obvious how to make the scheme (hO2 (t)O1 (t )O2 (t)O1 (0)i/hO2 (t)O1 (0)O2 (t)O1 (0)i) 1.
scalable without incurring errors that grow with the system
size. In contrast, a built-in global quantum clock avoids this
problem. of quantum clock divide both upper and lower branches into
three sectors, 23 = 8 paths are generated. The two paths | L i
and | R i are always staying in either of the two branches, i.e.
B. quantum clock upper upper upper, and lower lower lower re-
spectively, which are the only paths that survive in the ab-
sence of error, i.e. 1 , 2 = 0. Once the error is present,
Two primary error introduced to our protocol are the imper- the other six paths, which bounce between the upper and
fection of the pulses (single qubit rotations) acted on the quan- lower branches, will have non-zero amplitude. For exam-
tum clock (ancilla) and the imperfection in the couplings. The ple, the path upper lower upper corresponds to the
first type of error is generated in situ, while the second type is weighted state (i sin 21 )(i sin 22 )[U (t)]3 O2 U (t)O1 | iS ,
static. In the following, we analyze the effects of both types while the path upper lower lower corresponds to the
of errors. state (i sin 21 )(cos 22 )[U (t)]3 U(t)| iS . The errors modify
the final state in Eq. (7) to
1. Imperfection in pulses 3
1 1 2 X
| f i = cos cos |Ri + ci | Ei i | 1a i
2 2 2
Both the initial Hadamard gate (/2-pulse) and the x oper- i=1
ation (-pulse) which flips the ancilla and hence the arrow of
6
1 2 X
+ cos |Li + ci | Ei i | 0a i .
time can suffer from errors, since the rotation angles are con- cos (22)
2 2
tinuous variables and hence may not be exact. For a rotation i=4
along certain axis n, we can simply parameterize the rotation
error as Here, the state | E1,2,3 i (| E4,5,6 i) comes from the other un-
wanted paths end up in the upper (lower) branch. The am-
Rn ( + ) = ei(+)n~/2 , plitudes of them are c1 = c4 = i sin 21 cos 22 , c2 = c5 =
i sin 22 cos 21 and c3 = c6 = sin 21 sin 22 .
where is a small random fluctuation which differs in differ- Note that the errors in the -pulse do not change the value
ent shots of measurement. of the Hamiltonian H and H for forward and backward prop-
Assuming the initial Hadamard being perfect, we first con- agation. Nor do the errors change the quantum states | R i and
sider the imperfection of the two x flip operations on the | L i, of which the overlap h L | R i is the OTO correlator. Now
ancilla (1 , 2 = , n = x). Note that due to the two flips the question is to what extent that our protocol can extract this
9
overlap from the unwanted noise. When we measure the x Last but not least, we emphasize that with the current quan-
operator according to the protocol, it leads to tum information technology such as circuit QED, the fidelity
of a single-qubit gate can reach over 99.9% [58]. Therefore,
h x i f h f | I x | f i errors in rotating angles are under control and will not change
the order of magnitude of the signal, and we have shown from
1 2
= cos2 cos2 Re[h L | R i] + Noise. (23) above that the signal is stable against small imperfection in
2 2
the gates.
The first term is a slightly shrunk signal proportional to the
real part of the overlap between | L i and | R i. The second
noise term compare from the real or imaginary part of the
overlap involving the unwanted paths | Ei i. Since the mag-
2. Imperfection in the couplings
nitude of the real or imaginary part of any overlap is bounded
by 1, i.e.|Re(Im)h Ei | E j i| 1, |Re (Im)h Ei | R(L) i| 1, one
can derive a bound for the Noise, namely Before doing the experiments, one needs to tune the pa-
rameters such as the detunings a and b (e.g., by sweep-
|Noise| | sin 1 | + | sin 2 | + | sin 1 || sin 2 | ing the magnetic fluxes penetrating the superconducting loops
1 2 in the SQUID) to satisfy the conditions in Eq. (12), (18) or
+ sin2 ( )(1 + | sin 2 |) + sin2 ( )(1 + | sin 1 |) (19) which allows the reversing of sign exactly. In addi-
2 2
1 2 tion, there may be inhomogeneity in the qubit-cavity cou-
+ sin2 sin2 pling, namely the coupling strength may have spatial depen-
2 2
dence: g g j , ga ga, j . For the nonlocal model discussed
= | sin 1 | + | sin 2 | + O(12 + 22 + 1 2 ). (24)
in Sec. III A this is not a problem since the inhomogeneity
The above expression suggests that the noise bound is con- only introduces disorder in the effective coupling strength but
trolled by the errors on the rotation angles. The same prefactor not does not affect the condition Eq. (12) which allows ex-
and bound for noise hold for the y measurement, correspond- actly flipping the sign with the ancilla. However, for the local
ing to the imaginary part of the overlap. The signal-to-noise model discussed in Sec. III B, spatial dependent tunability of
ratio of the overlap has the expression the qubit frequency j , or equivalently the tunability of detun-
ing a, j and b, j is needed to satisfy the required conditions in
cos2 21 cos2 22 Eq. (18) or (19). Once the tuning is done with high precision,
SNR |h L | R i|, (25) the static imperfection is removed, and no such errors will be
| sin 1 | + | sin 2 | introduced in situ.
which is also controlled by the error angles and the magnitude The key is to have a calibration procedure that makes sure
of the overlap. Therefore, the overlap can be resolved once its that the static imperfection is removed or under control. This
magnitude is much larger than the noise background. can be achieved by a simplified version of the Ramsey inter-
In addition, the imperfection in the initial Hadamard (0 = ference protocol, without applying the operators O1 and O2 ,
/2, n = y) leads to the preparation of an unequal superposi- such that the cancellation between the forward and backward
tion of the two branches, evolution could be verified.
r r
1 sin 0 1 + sin 0
| iS ( | 0a i + | 1a i).
2 2
VI. EXTENSIONS OF THE LOCAL MODEL
The unequal weight of the wavefunctions in the two branches
of the interferometer [conditioned by | 0a i and | 1a i respec-
tively as shown in Eq. (22)] remains in the final output | i f . In Sec. III B, we have shown concretely how a 1D XY-spin
Therefore, the measurement outcome in the presence of both model can be embedded with a global quantum clock to con-
types errors becomes trol the sign of the Hamiltonian. Here we extend the model in
terms of the interaction and lattice type, spatial disorder and
1 2 dimensionality.
h x(y) i f = cos 0 cos2 cos2 Re(Im)[h L | R i]
2 2
+ Noise . (26)
3. Soft-core photons and Hubbard model
An extra prefactor cos 0 further shrinks the magnitude of
the overlap. On the other hand, the phase of the overlap,
i.e. Arg[h L | R i] = arctan{Im[h L | R i]/Re[h L | R i]} is less af- Above we focused on hard-core photons which lead to ef-
fected by the three error angles, since the same prefactors on fective spin-1/2 models. Now we consider soft-core photons
both the real and imaginary parts cancel with each other. The which allows one to build further interactions. Carrying out
SNR ratio remains the same expression as in Eq. (25) since the Schrieffer-Wolff transformation to the 4th-order yields the
the same prefactor cos 0 is introduced to the noise term. following correction to the Hamiltonian in Eq. (17):
10
localization-delocalization transition could be very pertinent are applied (a permutation switch), it reduces the computa-
since the key difference between a thermal phase and a many- tional complexity of certain problems from O(n2 ) to O(n). It
body localized phase is precisely that the former is chaotic would be worthwhile to explore the possibility of obtaining
while the latter is not. As discussed in Sec.VI 4, this is pos- such speedups in quantum algorithms using our cavity-QED
sible within our setup. Similarly, simulating SYK models setup.
[3, 12] in cavity QED or cold atoms and measuring OTO cor-
Acknowledgements:
relators is another promising direction.
Conceptually, the idea of measuring OTO correlators us- GZ and MH were supported by ONR-YIP, ARO-MURI,
ing the quantum clock is reminiscent of the idea of quantum- AFOSR-MURI, NSF-PFC at the JQI, and the Sloan Foun-
controlled ordering of gates discussed in Ref.[60]. The basic dation. TG acknowledges startup funds from UCSD and
result of Ref.[60] is that if in quantum computing, one al- fellowship from the Gordon and Betty Moore Foundation
lows a control switch that switches the order in which gates (Grant4304).
[1] M. Gutzwiller, Chaos in Classical and Quantum Mechanics, mentado, H. E. Tureci, and A. A. Houck, Physical review letters
Springer 1990. 107, 053602 (2011).
[2] S. H. Shenker and D. Stanford, Journal of High Energy Physics [24] D. L. Underwood, W. E. Shanks, J. Koch, and A. A. Houck,
2014, 1 (2014). Physical Review A 86, 023837 (2012).
[3] Hidden correlations in the hawking radiation and thermal [25] S. Schmidt and J. Koch, Annalen der Physik 525, 395 (2013).
noise, talk given at Fundamental Physics Prize Symposium, [26] J. Raftery, D. Sadri, S. Schmidt, H. E. Tureci, and A. A. Houck,
Nov. 10, 2014. Physical Review X 4, 031043 (2014).
[4] J. Maldacena, S. H. Shenker, and D. Stanford, arxiv: [27] A. Chiesa, P. Santini, D. Gerace, J. Raftery, A. A. Houck, and
1503.01409 (2015). S. Carretta, arXiv (2015), 1504.05667v1.
[5] Although the relation between such correlators and semiclassi- [28] S. Hacohen-Gourgy, V. V. Ramasesh, C. De Grandi, I. Siddiqi,
cal chaos was first observed a while ago [6], it is only recently and S. M. Girvin, Phys. Rev. Lett. 115 (2015).
that progress has been made in more general settings. [29] M. D. Lukin, Reviews of Modern Physics 75, 457 (2003).
[6] A. I. Larkin and Y. N. Ovchinnikov, JETP 28, 1200 (1969). [30] M. Saffman, T. G. Walker, and K. Mlmer, Rev. Mod. Phys. 82,
[7] If O1 , O2 were canonically conjugate, then semi-classically this 2313 (2010).
dO (t) 2
correlator is indeed given by dO 2
, which is the canonical [31] J. Ningyuan, A. Georgakopoulos, A. Ryou, N. Schine, A. Som-
2 (0)
way to define sensitivity of time evolution to the initial condi- mer, and J. Simon, arXiv (2015), 1511.01872v3.
tions. [32] A. Sommer, H. P. Buchler, and J. Simon, arXiv (2015),
[8] D. A. Roberts, D. Stanford, and L. Susskind, Journal of High 1506.00341v1.
Energy Physics 2015, 1 (2015). [33] K. Kim, M. S. Chang, S. Korenblit, R. Islam, E. E. Edwards,
[9] D. A. Roberts and D. Stanford, Phys. Rev. Lett. 115, 131603 J. K. Freericks, G. D. Lin, L. M. Duan, and C. Monroe, Nature
(2015). 465, 590 (2010).
[10] J. Polchinski and V. Rosenhaus, Journal of High Energy Physics [34] E. T. Jaynes and F. W. Cummings, Proceedings of the IEEE
2016, 1 (2016). 51,89 (1963).
[11] J. Maldacena and D. Stanford, arXiv: 1604.07818 (2016). [35] L. DiCarlo, J. M. Chow, J. M. Gambetta, L. S. Bishop, B. R.
[12] S. Sachdev and J. Ye, Phys. Rev. Lett. 70, 3339 (1993). Johnson, D. I. Schuster, J. Majer, A. Blais, L. Frunzio, S. M.
[13] S. Sachdev, Phys. Rev. Lett. 105, 151602 (2010). Girvin, R. J. Schoelkopf, Nature 460, 240 (2009).
[14] B. Swingle, G. Bentsen, M. Schleier-Smith, and P. Hayden, [36] D. Comparat and P. Pillet, Journal of the Optical Society of
arXiv:1602.06271v1 (2016). America B 27, A208 (2010).
[15] K. M. Birnbaum, A. Boca, R. Miller, A. D. Boozer, T. E. [37] C. S. Hofmann, G. Gunter, H. Schempp, N. L. M. Muller, A.
Northup, and H. J. Kimble, Nature (London) 436, 87 (2005). Faber, H. Busche, M. Robert-de Saint-Vincent, S. Whitlock,
[16] L. Jiang, G. K. Brennen, A. V. Gorshkov, K. Hammerer, M. and M. Weidemu ller, Frontiers of Physics 9, 571 (2013).
Hafezi, E. Demler, M. D. Lukin, and P. Zoller, Nature Physics [38] K. M. Maller, M. T. Lichtman, T. Xia, Y. Sun, M. J. Piotrowicz,
4, 482 (2008). A. W. Carr, L. Isenhower, and M. Saffman, Physical Review A
[17] J. S. Douglas, H. Habibian, C. L. Hung, A. V. Gorshkov, H. J. 92, 022336 (2015).
Kimble, and D. E. Chang, Nature Photonics 9, 1 (2015). [39] M. Muller, I. Lesanovsky, H. Weimer, H. P. Buchler, and P.
[18] A. G. lez Tudela, C. L. Hung, D. E. Chang, J. I. Cirac, and H. Zoller, Phys. Rev. Lett. 102, 170502 (2009).
J. Kimble, Nature Photonics 9, 1 (2015). [40] M. Knap, A. Kantian, T. Giamarchi, I. Bloch, M. D. Lukin, and
[19] R. J. Schoelkopf and S. M. Girvin, Nature 451, 664 (2008). E. Demler, Phys. Rev. Lett. 111, 147205 (2013).
[20] A. Blais, J. Gambetta, A. Wallraff, D. I. Schuster, S. M. Girvin, [41] D. A. Abanin and E. Demler, Phys. Rev. Lett. 109, 020504
M. H. Devoret, and R. J. Schoelkopf, Physical Review A 75, (2012).
032329 (2007). [42] H. Pichler, G. Zhu, A. Seif, P. Zoller, and M. Hafezi, arXiv
[21] A. A. Houck, H. E. Tureci, and J. Koch, Nature Physics 8, 292 (2016), 1605.08624v1.
(2012). [43] G. Zhu, S. Schmidt, and J. Koch, New Journal of Physics 15,
[22] J. Koch, A. A. Houck, K. Le Hur, and S. M. Girvin, Physical 115002 (2013).
Review A 82, 043811 (2010). [44] In our convention, the ancilla state | 0a i corresponds to the | i
[23] A. J. Hoffman, S. J. Srinivasan, S. Schmidt, L. Spietz, J. Au- spin state.
12
[45] M. Srednicki, Journal of Physics A: Mathematical and General 1. Initialize the coupled system as | iS | 0a i.
32, 1163 (1999).
[46] J. M. Deutsch, Phys. Rev. A 43, 2046 (1991). 2. Apply a Hadamard gate (/2-pulse around the y-axis)
[47] M. Srednicki, Physical Review E 50, 888 (1994). to the ancilla:
[48] J. Jin, D. Rossini, R. Fazio, M. Leib, and M. J. Hartmann, arXiv:
1302.2242v1 (2013). IS Ha .
[49] S. E. Nigg, H. Paik, B. Vlastakis, G. Kirchmair, S. Shankar, L.
Frunzio, M. H. Devoret, R. J. Schoelkopf, and S. M. Girvin,
Phys. Rev. Lett. 108, 240502 (2012).
3. Apply a conditional operation
[50] Alternatively, one can replace the ancilla cavity (a) with an an-
cilla qubit, such as the transmon qubit [54] which can be ap- CO1 ,1 = O1 | 1a ih 1a | + IS | 0a ih 0a |. (A1)
proximated as a weakly anharmonic oscillator and is directly
coupled to the passive cavity bus (b) with a Josephson junction 4. Conditional evolution for time t
[49].
[51] J. Majer, J. M. Chow, J. M. Gambetta, J. Koch, B. R. Johnson, J. eiHt | 0a ih 0a | + eiHt | 1a ih 1a |
A. Schreier, L. Frunzio, D. I. Schuster, A. A. Houck, A.Wallraff,
A. Blais, M. H. Devoret, S. M. Girvin and R. J. Schoelkopf,
Nature 449, 443 (2007). 5. Apply another conditional operation
[52] The third-order term is zero in this model. The complete for-
mula of the second-order effective Hamiltonian without pro- CO2 ,1 = O2 | 1a ih 1a | + IS | 0a ih 0a |. (A2)
jecting onto the zero qubit excitation subspace is shown in Ap-
pendix D. 6. (a) Apply a x operator (-pulse around the x-axis) to
[53] Besides the two-spin ZZ interaction, there also exists two flip the ancilla:
other types of effetctive interactions in fourth-order perturba-
tion, namely four-spin ring exchange interaction +i j +k l IS x .
and three-spin assisted hopping zi +j k .
[54] J. Koch, T. M. Yu, J. Gambetta, A. A. Houck, D. I. Schus-
ter, J. Majer, A. Blais, M. H. Devoret, S. M. Girvin, and R. (b) Conditional evolution for time 2t
J. Schoelkopf, Physical Review A 76, 042319 (2007).
[55] We note that in this situation, if the dispersive interaction e2iHt | 0a ih 0a | + e2iHt | 1a ih 1a |
Hdisp is realized by JC interaction perturbatively, the non-local
second-order flip-flop interactions between the qubits [Eq. (11)] 7. (a) Apply another x operator to the ancilla:
do not play any role, since all the qubits are in the down states.
Therefore, the dispersive Hamiltonian Hdisp is indeed a valid IS x .
description.
[56] J. R. Schrieffer and P. A. Wolff, Phys. Rev. 149, 491 (1966).
[57] S. Bravyi, D. P. DiVincenzo, and D. Loss, Annals of Physics
(b) Apply another conditional operation:
326, 2793 (2011), ISSN 0003-4916.
[58] R. Barends, J. Kelly, A. Megrant, A. Veitia, D. Sank, E. Jeffrey, CO2 ,0 = IS | 1a ih 1a | + O2 | 0a ih 0a |. (A3)
T. C. White, J. Mutus, A. G. Fowler, B. Campbell, Y. Chen,
Z. Chen, B. Chiaro, A. Dunsworth, C. Neill, P. OMalley, P. (c) Conditional evolution for time t
Roushan, A. Vainsencher, J. Wenner, A. N. Korotkov, A. N.
Cleland, and J. M. Martinis, Nature 508, 500 (2014). eiHt | 0a ih 0a | + eiHt | 1a ih 1a |.
[59] V. Oganesyan and D. A. Huse, Physical Review B 75, 155111
(2007). (d) Apply another conditional operation:
[60] M. Araujo, F. Costa, and C. Brukner, Phys. Rev. Lett. 113,
250402 (2014). CO2 ,0 = IS | 1a ih 1a | + O2 | 0a ih 0a |. (A4)
[61] One could also choose the /2 mode, which is the fundamental
mode of the resonator, leading to an opposite sign of the phase
variable on the two ends. However, eventually the sign can be 8. Measure the expectation value of x and y operator un-
gauged out in a 1D chain. der the final state | f i to measure the real and imagi-
[62] N. M. Sundaresan, Y. Liu, D. Sadri, L. J. Szocs, D. L. Under- nary part of the OTO correlator:
wood, M. Malekakhlagh, H. E. Tureci, and A. A. Houck, Phys-
ical Review X 5, 021035 (2015). h x(y) i f h f | IS x(y) | f i = Re(Im)[h L | R i]. (A5)
[63] M. Mariantoni, F. Deppe, A. Marx, R. Gross, F. K. Wilhelm,
and E. Solano, Physical Review B 78, 104508 (2008).
Appendix B: Circuit and cavity QED architecture realizing the
local models
Appendix A: The complete sequence of the measurement In this appendix, we discuss the details about the circuit-
protocol
QED architecture which realize our desired local model de-
scribed by Eq. (15) and illustrated in Fig. 3, and the corre-
We recapitulate the steps of our protocol for completeness: sponding experimental protocols.
13
a. 2D circuit-QED network ever, one can select one to play the major role by tuning the
qubit frequency close to the frequency of the selected mode.
We first discuss the realization with 2D on-chip circuit- When the global control qubit is detuned from local qubits,
QED network and illustrate it in Fig. 7(a). As an example, we the only interaction survives rotating-wave approximation is
show in Fig. 7(a) the realization of qubits with the Cooper-pair the ZZ coupling:
box/transmon, composed of two Josephson junctions and one 0 z X z
capacitor. The level structure and qubit frequency are tuned in Hdisp HZZ = j, j+1 . (B1)
situ by the external magnetic flux threading the junction loop. 2 j
In general, any type of superconducting qubits can be used in
the network, such as flux and fluxonium qubits. Such ZZ interaction is frequently used for a control-phase gate
The local -mode transmission-line resonators [61] are cou- on many platforms. For example in circuit-QED, such ZZ
pled capacitively to the qubits [21]. We represent the voltage interaction exists due to the contribution of the third-level of
on the ends of the resonator as V rj , and the electric charge the transmon qubits [35]. One can easily see that by doing the
on the upper superconducting island (non-grounded one) of replacement z = 1 2a a in the 0- and 1-photon subspace,
q
the qubit as V j, j+1 . The capacitive coupling between res- the above HZZ is formally identical to Hdisp [in Eq. (15)] up
onator and qubit on its right leads to the following interac- to a constant frequency shift, which can be absorbed into the
right q renormalized local qubit frequency .
tion T j = CV rj V j, j+1 , where C is the intermediate ca-
An alternative for the global transmission resonator can be a
pacitance. Canonical quantization allows us to represent the resonator array [21, 22, 25, 43], where we can use the common
phase variables with creation/annihilation of photon opera- mode (k = 0) as the ancilla. Besides the above approach us-
q
tors, i.e. V rj = Vrms
r
(b j +bj ) and V j, j+1 = eCg xj, j+1 , where V rms ing capacitive coupling and JC interaction to generate the dis-
is the root-mean-square Voltage of the resonator, Cq the qubit persive interaction perturbatively, one can also directly couple
capacitance, and e the unit charge. Therefore, with a rotating- each resonator in the array to the qubits with a Josephson junc-
wave approximation which drops the counter-rotating term, tion [48, 49]. In this way, the dispersive interaction strength
the interaction can be expressed as the Jaynes-Cummings is only proportional to the Josephson energy E J and does
right
form T j = gb (bj j, j+1 + H.c.), where the JC interaction not depend on the detuning in the form of g2a /a , and hence
strength is gb = 2eCq Vrms r
. The interaction between the res- can remain sizable even when the resonator and qubit is far
onator and the qubit on its left has an identical expression. detuned. With this method, the condition Eq. (18) for sign flip
Sum of all the pairwise interaction terms leads to the realiza- is even easier to be satisfied.
tion of the JC interaction V in Eq. (15). Since we eventually
need hard-core bosons to simulate spin models, we introduce
nonlinearity into the resonators by embedding qubits, which is b. 3D cavity-QED with superconducting qubit array
illustrated in the inset of Fig. 7(a). The presence of the qubit
inside the resonator leads to photon blockade [23]. Now we consider a 3D version of the experimental realiza-
The global transmission-line resonator (cavity) is coupled tion. Instead of considering a hybrid resonator-qubit network
to all the qubits between the local resonators. In order to as mentioned above, here we only consider a superconduct-
make sure the coupling is uniform, we put the qubits in the ing qubit array in a 3D cavity [c.f. Fig. 7(b)]. The word
peaks (positive or negative) of the resonator mode, imply- qubit here is not restricted to two-level systems, but actu-
ing the length of the resonator is at least N/2, where is ally refers to multi-level artificial atoms, which is an accurate
the microwave frequency and N is the total number of qubits. description for any superconducting qubits, such as transmons
This also means the control photon occupies the N th -harmonic [54]. Experimental realization of a Bose-Hubbard model with
mode. Such a super-long transmission-line resonator has been transmon array in a 3D cavity has been achieved recently in
explored experimentally in Ref. [62]. Due to the dressing of Ref. [28]. Still, the array consists of two different types of ar-
the qubit, the level structure of the global resonator also be- tificial atoms [illustrated with red and blue in Fig. 7(b)] with
comes anharmonic, therefore allowing one to manipulate the different level structures, achieved for example by choosing
photon state in the truncated 0- and 1-photon subspace. different size of the junction loop between the two supercon-
In addition, one could add another ancilla/control qubit ducting islands. The red qubits play the role of passive cou-
coupled to the global cavity. Instead of exploiting the non- plers that mediate interactions between the blue qubits, con-
linearity of the global resonator, one could also use an ancilla sistent with the schematic diagram in Fig. 3(a).
qubit to manipulate the photon state through the combination In order to only couple the red qubits but not the blue qubits
of control-phase gate induced by dispersive interaction and to the 3D cavity, we exploit the directional property of dipole
single qubit rotation [16]. coupling and so choose different orientations of the red and
An alternative to realize the dispersive-type coupling is to blue qubits. As illustrated in Fig. 7(b), the dipole of the blue
directly couple the ancilla qubit (represented by Pauli opera- qubits ~p1 , originating from the Cooper pair tunneling between
tor ) to all the local qubits, mediated by the virtual photon the two islands, is perpendicular to the cavity electric field
in the cavity. In this case, the cavity serves as a quantum bus ~ Therefore, the dipole interaction for the blue qubits ~p1 E~
E.
and hence has no photon occupation. One subtle point is that is zero. On the other hand, the dipole of the red qubits ~p2
multiple modes are mediating the dispersive interaction, how- is rotated so as not to be perpendicular to the electric field,
14
...
tion, we compare the numerical diagonalization of the original
S
model Eq. (15) and the full 2nd-order effective Hamiltonian
micrwave ... Eq. (17) or (20).
drive We start with the simplest dimer case as shown in Fig. 8(a),
S
original Hamiltonian
(b) (d)
effective Hamiltonian (2nd order)
original Hamiltonian
effective Hamiltonian (2nd order)
FIG. 8: Numerical comparison of the original and 2nd-order effective Hamiltonian for a dimer. Parameters: b =50MHz, a =800MHz, and
=50MHz (or equivalently ga =200MHz), on-site photon cut-off nmax b = 3. (a) The setup for numerical simulations contains two local cavities,
one qubit, and one global cavity. (b) Comparison of the spectrum between the exact (blue circle) and effective (yellow square) Hamiltonian
obtained from
P numerical exact diagonalization. The spectrum is separate into two ancilla sectors. The red circle show states in the 1-photon
manifold ( j hbj b j i 0, hz i 0). (c) The relative error between the exact and effective spectrum for gb /b = 0.1. (d) The average photon
and qubit excitation numbers for the low-lying states in both ancilla sectors, obtained from exact (blue circle) and effective (yellow square)
Hamiltonian. The red circles show the states in the 1-photon manifold. (e) The energy splitting in the 1-photon manifold for both ancilla
sectors obtained from exact diagonalization of the original Hamiltonian, and the prediction 2g2b /b from second-order perturbation theory.
there is an effective hopping amplitude t = (2a a 1)g2b /b is composed of three local cavities, three qubits in between,
between neighboring local cavities, which mediated by the in- and a global ring cavity. We choose the same parameters and
termediate qubit. The sign of the hopping amplitude changes focus still on the 1-photon manifold ( j hbj b j i 1) of the ex-
P
when the ancilla is flipped, while the magnitude |t| = g2b /b act numerical spectrum as shown in panel (b-e). We can see
should remain the same. Therefore, there should be a splitting from the zoom-in insets in panel (b) and (d) that the lowest of
= 2|t| = g2b /b between the symmetric and anti-symmetric the three states in na = 0 sector is singly degenerate, while in
single-particle states of the dimer, namely 21 (| 0b 1b i + | 1b 0b i) the na = 1 sector the lowest states are doubly degenerate. This
and 12 (| 0b 1b i | 1b 0b i). In panel (e), we plot the splitting in can be simply understood by the formula of the effective hop-
both ancilla sectors from the exact model, namely na =0 and ping amplitude t = (2a a 1)g2b /b from Eq. (20). For na = 0
na =1 as a function of gb /b , and compare them with the value ancilla sector, the effective hopping is t = g2b /b , which is
2g2b /b predicted by the perturbation theory. The match is negative according to the current parameter choice. In this sit-
very good for small gb /b when perturbation theory is valid. uation, the spectrum in the 1-photon manifold is {2|t|, |t|, |t|},
In addition, we note that even when the exact result deviates and the unique ground state in this manifold corresponds to
from the 2nd-order perturbation theory prediction, the split- the symmetric state 13 (| 1b 0b 0b i + | 0b 1b 0b i + | 0b 0b 1b i). The
ting for both ancilla sectors still match. This fact suggests that two degenerate states with higher energy can be chosen as
our prediction of the equal magnitude of the prefactors in both two counter-propagating states with opposite chirality, namely
ancilla sectors may go much beyond the second-order pertur- 1 (| 1b 0b 0b i + ei2/3 | 0b 1b 0b i + ei2/3 | 0b 0b 1b i). For na = 1
3
bation and may extend to all orders. In the main text, we al-
ancilla sector, the effective hopping is t = g2b /b , which is pos-
ready see this to be true for the fourth-order terms in Eq. (27)
itive and hence leads to frustration of the ring. In this situation,
and (28), with the prefactor (1 2a a)g4b /3b . Similarly, for kth
the spectrum in the 1-photon manifold is {|t|, |t|, 2|t|}, and
order perturbation, a prefactor of the form (1 2a a)gkb /k1 b the doubly-degenerate ground states correspond to the two op-
is expected. posite chiral states, while the symmetric state has higher en-
From the above verification of the dimer case, we see that ergy. Therefore, the signature of sign flipping is clearly shown
there is indeed a symmetry of the magnitude of the prefactors in the two insets. In addition, for both ancilla sectors, the split-
in both ancilla sectors. However, we are not able to check the tings (na =0 and na =1 ) between the lower and higher states is
sign flip induced by the ancilla from the spectrum, since the fixed to be 3|t| = 3g2b /b . We compare the splittings from the
spectrum of a dimer is invariant under the sign flip of the hop- exact diagonalization to the prediction 3g2b /b from perturba-
ping, which is equivalent to a gauge transformation. However, tion theory in panel (f) as a function of gb /b , and we can see
no gauge transformation can flip the hopping signs for a three- a very good match for small gb /b . Also, the symmetry of
site periodic ring, such as the setup shown in Fig. 9(a), which the magnitude of the splitting in both ancilla sectors is again
16
(f)
(exact)
(exact)
(perturbation)
(c) (e)
FIG. 9: Numerical results for a three-site ring. (a) The setup for numerical simulations contains three local cavities, three qubits, and one
global cavity, which form a periodic ring. (b-e) The average photon and qubit excitation numbers for the low-lying states in both ancilla
sectors, obtained from exact (blue circle) and effective (yellow square) Hamiltonian. The red circles show the states in the 1-photon manifold,
and the insets show the zoom-in spectrum in that manifold. (f) The splitting in the 1-photon manifold for both ancilla sectors obtained from
exact diagonalization of the original Hamiltonian, and the prediction 3g2b /b from second-order perturbation theory.
verified. We see from the first term that the qubits, like a local quan-
tum switch, mediate qubit-state-dependent hopping of pho-
tons on neighboring cavities, which has been previously ex-
Appendix D: Complete formula of the second-order effective plored in the context of superconducting circuits [63]. On the
Hamiltonian other hand, the second term shows the flip-flop interaction be-
tween neighboring qubits are only mediated by virtual pho-
In the main text, we have derived the effective Hamiltonian tons (meaning there is no presence of the photon operators),
of the local model constrained in the sector with zero qubit i.e. the so-called quantum bus interaction [20, 43, 51]. The
excitation, i.e. | i, which corresponds to a projection third term represents the Lamb shift of the qubits induced by
PS z =0 . Here, we release such an constraint, and show the full the neighboring local cavities, and the last term represents the
effective Hamiltonian in the dispersive regime up to second dispersive shifts (AC-Stark shifts), which shows the mutual
order: dressing of photons and qubits [20, 43].
X g2 X
Heff =H0 + b
[(bj b j+1 + H.c.)zj, j+1
n
b,na j
a