Professional Documents
Culture Documents
and numerical methods we observe the essential signatures where σiγ (γ = x, z) are the Pauli matrices acting on the ith spin,
of MBL: initial-state memory retention, Poissonian distributed Ji,j is the coupling strength between spins i and j, B is a uniform
energy level spacings, and evidence of long-time entanglement effective transverse field, Di is a site-dependent disordered potential,
growth. Our platform can be scaled to more spins, where a and h̄ = 1 (see Supplementary Information). After the chain evolves
detailed modelling of MBL becomes impossible. for some time, we collect the state-dependent fluorescence on
It is exceedingly rare in nature for systems to localize, or retain an intensified charge-coupled device camera for site-resolved
local information about their initial conditions at long times. imaging. This, in addition to our ability to perform high-fidelity
In an important counterexample, Anderson demonstrated that rotations, allows measurement of the single-site magnetization
localization can arise due to the presence of disorder, which can hσiγ i(γ = x, y, z) as well as arbitrary spin correlation functions along
destructively scatter propagating waves and prevent transport of any direction.
energy or particles3 . Although this interference effect can be applied Ji,j is a tunable, long-range coupling that falls off approximately
to generic quantum systems, most experimental work has been algebraically as Ji,j ∝ Jmax /|i − j|α (ref. 15), where Jmax is typically
restricted to the narrow parameter regime of low excitation energies 2π (0.5 kHz). Here, we tune α between 0.95 and 1.81, although
and no interparticle interactions4–6 . for most of the data α ≈ 1.13. We directly measure the complete
Whether such localization persists in the more general case of spin–spin coupling matrix (Fig. 1a), demonstrating the long-range
arbitrary excitation energy and non-zero interparticle interactions interactions required to exhibit MBL in this model. Moreover, it has
was theoretically explored by Anderson3 , and more recently by oth- been shown numerically that (1) exhibits MBL for the experimental
ers7–10 . This MBL phase is predicted to emerge for a broad set of in- parameters16 .
teraction ranges and disorder strengths, although the precise phase The site-specific programmable disorder term Di is sampled from
diagram is not well known11 because equilibrium statistical mechan- a uniform random distribution with Di ∈ [−W , W ], where W
ics breaks down in the MBL phase and numerical simulations are characterizes the strength of the disorder. The disorder is generated
limited to ∼20 particles8,9 . Very recent experiments searching for by site-dependent laser-induced Stark shifts (see Supplementary
MBL have measured constrained mass transport and the breakdown Information), which also allow for preparation of the system
of ergodicity in disordered atomic systems with interactions12,13 . into any desired product state. To ensure we observe the general
Here we report the direct observation of MBL in a long- behaviour of the disordered Hamiltonian, we average over 30
range transverse field Ising model with programmable, random distinct random instances of disorder (Fig. 1b,c), which leads to
disorder. This is a non-integrable model that cannot be mapped to a sampling error that is smaller than the features of interest (see
noninteracting particles (a necessary condition for MBL7 ) and we Supplementary Information).
can easily tune the disorder strength and interaction range over a An important signature of the MBL phase is manifested in the
parameter space that exhibits this phenomenon. Our experiment spectral statistics of adjacent energy levels of the Hamiltonian. In the
is effectively a closed quantum system over the timescales of thermalizing phase, the energy levels are given by the eigenvalues of
interest, because the system localizes approximately 60 times a random matrix, a matrix whose elements are given by a random
1
Joint Quantum Institute, University of Maryland, Department of Physics and National Institute of Standards and Technology, College Park,
Maryland 20742, USA. 2 Department of Physics, Indiana University, Bloomington, Indiana 47405, USA. 3 Institute for Quantum Optics and Quantum
Information of the Austrian Academy of Sciences, 6020 Innsbruck, Austria. 4 Institute for Theoretical Physics, University of Innsbruck, 6020 Innsbruck,
Austria. 5 Physik Department, Technische Universität München, 85747 Garching, Germany. 6 Physics Department, Princeton University, Princeton,
New Jersey 08544, USA. *e-mail: htimsjacob@gmail.com
a 0.5 b 8
0.1 −2
−4 Ion positions
1 2 10
3 4 8 9 −6
5 5 6 7
Spin 6
num 7 8 9
4 ber
ber 2 3 in num −8
10 1 Sp
c
30
d W = 8Jmax
25 0.3
Fraction of occurrences
Instance of
20
0.2
15
disorder
0.1
10
Figure 1 | An interacting spin model with random disorder. a, Directly measured elements of the spin–spin coupling matrix Jij , equation (1) (increasing
interaction strength from blue to red). The long-range interactions decay as Jmax /r1.13 . b, A specific instance of the random disordered field with a
schematic illustration of the long-range interactions. c, Random values of the disordered field for all 30 instances of disorder for several different disorder
strengths and for each ion (red indicates positive values and blue indicates negative values, with values between −0.5 and 0.5). d, Level statistics
calculated from the measured spin–spin coupling matrix in a and applied disorders in c are Poisson-distributed (black line is the expected level spacings for
a Poisson distribution), as predicted for a MBL system.
distribution, due to level repulsion. However, in the MBL phase, this When B P J , the Hamiltonian is effectively an XY model17,18
level repulsion is greatly suppressed because eigenstates typically and conserves i σiz , because Ising processes that flip spins along
differ by multiple spins flips. As a result, the level spacing between the large field are energetically forbidden. Thus, being in a spin
adjacent energy eigenvalues are Poisson-distributed8,9 . Using our configuration with half of the spins up and half of the spins down
directly measured spin–spin couplings and applied realizations for maximizes the accessible energy states. In addition, the Néel state is
the strongest experimental disorder W = 8Jmax and B = 4Jmax , we never an eigenstate, even for B J and W J , because the uniform
calculate the distribution of adjacent energy level splittings and find B field at each site still allows spin exchange in the z-basis.
them to be Poisson-distributed, as expected for a MBL state (Fig. 1d). If a system is thermal, the Eigenstate Thermalization Hypothesis
Before searching for evidence of localization in the system’s (ETH) provides a general framework where observables reach the
time evolution, we first find parameters that cause the measured value predicted by the microcanonical ensemble19–21 . This allows us
state to thermalize in the absence of disorder. We increase the to calculate the expected thermal value of the reduced density matrix
transverse field B and look for conditions that result in the single-site given the Hamiltonian and an initial state (see Supplementary
magnetization along two orthogonal directions approaching and Information). To further establish that the system is thermalizing,
remaining at their thermal equilibrium values (see Supplementary we measure the reduced density matrix for each spin, ρi = Tr{j6=i} ρ,
Information). without applied disorder and B = 4Jmax , as shown in Fig. 2a. In
Figure 2a shows the measured dynamics of hσiz i for B = 4Jmax our experiment, the spins are initially prepared in a product state
and Di = 0 with the spins initialized in the Néel ordered state, with high fidelity. However, at long times, the measured reduced
|↑↓↑↓↑↓↑↓↑↓iz along the z direction. This configuration has an density matrices show that each of the spins is very close to the
energy equivalent to an infinite temperature thermal state, because zero magnetization mixed state, implying the system has locally
the expectation value of the Hamiltonian is zero. At long times, each thermalized.
expectation value hσiz i approaches zero, losing memory of the initial We apply the random disordered potential, Di 6 = 0, with B = 4Jmax
ordering. As the transverse field B is increased, the system appears and observe the emergence of MBL as we increase the strength
to thermalize more quickly and the level statistics approach those of of disorder. Because the many-body eigenstates in the MBL phase
random matrices rather than Poissonians, as expected for a generic are not thermal, transport of energy and spins is suppressed,
thermodynamic system (see Supplementary Information). and ETH fails. Thus, observables will not relax to their thermal
Z magnetization, 〈σ iz 〉
Z magnetization, 〈σ iz 〉
0.78
0.53 0.47
0.03 0.5 0.5 0.22
0.04
0.02 〈↓| 0.01 0.05 0.08 〈↓|
| ↑〉 |↑〉 〈↓| |↑〉 〈↓| |↑ 〉
| ↓〉 〈↑| 0.0 |↓〉 〈↑| |↓〉 〈↑| 0.0 |↓〉 〈↑|
Increasing disorder
b c d e
1.0
W = 1Jmax W = 2Jmax W = 4Jmax W = 6Jmax
Z magnetization, 〈σ iz 〉
0.5
0.0
−0.5
−1.0
0 2 4 6 8 10 0 2 4 6 8 10 0 2 4 6 8 10 0 2 4 6 8 10
Jmaxt Jmaxt Jmaxt Jmaxt
Figure 2 | Emergence of a MBL state. a, Time-evolved single-site magnetizations hσiz i (different colours represent different ions) for the Hamiltonian in
equation (1) for B = 4Jmax with no applied disorder (Di = 0). The initial-state reduced density matrices for ions 1 and 10 show the spins start in a product
state along the z direction. The time-averaged reduced density matrices for Jmax t > 5 (colours from blue to red indicate increasing values of the elements of
the density matrix) agree with the values predicted by the ETH, implying the system has thermalized locally. b–e, As the disorder strength increases, the
spins retain more information about their initial state, indicating a transition towards MBL. f, Dynamics of hσiz i for the strongest applied disorder,
W = 8Jmax . The initial-state and steady-state time-averaged reduced density matrices for ions 1 and 10 now show that information is preserved about the
initial spin configuration at the end of the evolution. Statistical error bars (1 s.d.) are smaller than the data points.
1 1 X b 0.50
c 0.35
D(t) = − hψ0 | σiz (t)σiz (0) |ψ0 i (2) W = 8Jmax
Steady-state 〈 (t)〉
Steady-state 〈 (t)〉
2 2N i
0.33
0.31
which gives the number of spin flips away from the initial state, 0.40
ψ0 , normalized by the length of the chain, N . At long times, 0.29
the HD approaches 0.5 for a thermalizing state and remains at 0 0.27
for a fully localized state. In Fig. 3a, we measure that the long- 0.30 0.25
0.1 1 10 1.0 1.2 1.4 1.6 1.8
time HD is 0.5 in the absence of disorder, and becomes smaller
Disorder strength, W (Jmax) Power-law exponent, α
as the disorder strength is increased and the system more strongly
localizes. This time evolution agrees with exact diagonalization
numerical calculations for (1) (see Supplementary Information). Figure 3 | The Hamming distance (HD). a, The HD exhibits time dynamics
Figure 3b shows that for finite but weak disorder, the time- that reach their steady-state values after Jmax t ≈ 5. For increasing disorder,
averaged HD for Jmax t > 5 is essentially unchanged, indicating weak the system becomes more strongly localized, and the steady-state HD
or no localization. However, once the random field is sufficiently decreases. (Different colours represent different disorder strengths.) b, The
strong we observe a crossover from a thermalizing to a localized steady-state HD with respect to the strength of the random potential
state. Once in this regime, the system becomes more localized with indicates the state is not or only weakly localized for small disorder, but
increasing disorder strength. after the random field is sufficiently strong it becomes more localized with
There is great theoretical interest in mapping the MBL increased disorder. c, The steady-state HD with respect to the power-law
phase diagram with respect to interaction range and disorder exponent indicates that the system becomes less localized in the presence
strength11,22,23 . We have taken the first steps towards this goal of longer-range interactions (smaller α). Error bars, 1 s.d.
a b Data availability. The data that support the plots within this paper
1.5 No disorder W = 6Jmax
1.1 and other findings of this study are available from the corresponding
1.4 author upon reasonable request.
QFI
QFI
1 Many-body
Methods for:localization
Many-bodyin a quantuminsimulator
localization with
a quantum simu-
2 programmable random disorder
lator with programmable random disorder
4 C. Monroe∗
6 optical dipole forces to ions confined in a 3-layer linear Paul trap with a 4.8 MHz radial frequency.
7 Two off-resonant laser beams with a wavevector difference ∆k along a principal axis of transverse
8 motion globally address the ions and drive stimulated Raman transitions. The two beams contain a
9 pair of beatnote frequencies symmetrically detuned from the resonant transition at ν0 = 12.642819
10 GHz by a frequency µ, comparable to the transverse motional mode frequencies. In the Lamb-
11 Dicke regime, this results in the Ising-type Hamiltonian in Eq. (1) 1–3 with
N
2 bi,m bj,m
Ji,j = Ω ωR 2 , (2)
m=1
µ2 − ωm
12 where Ω is the global Rabi frequency, ωR = h̄∆k 2 /(2M ) is the recoil frequency, bi,m is the normal-
13 mode matrix 4 , and ωm are the transverse mode frequencies. The coupling profile may be approx-
14 imated as a power-law decay Ji,j ≈ Jmax /|i − j|α , where in principle α can be tuned between 0
15 and 3 by varying the laser detuning µ or the trap frequencies ωm 2, 5 . In this work, α was tuned ap-
∗
Joint Quantum Institute, University of Maryland Department of Physics and National Institute of Standards and
Technology, College Park, MD 20742
†
Department of Physics, Indiana University, Bloomington, IN, 47405
‡
Institute for Quantum Optics and Quantum Information of the Austrian Academy of Sciences, 6020 Innsbruck,
Austria
§
Institute for Theoretical Physics, University of Innsbruck, 6020 Innsbruck, Austria
¶
Physik Department, Technische Universitt Mnchen, 85747 Garching, Germany
Physics Department, Princeton University, Princeton, NJ 08544, USA
1
NATURE PHYSICS | www.nature.com/naturephysics 1
© 2016 Macmillan Publishers Limited. All rights reserved.
SUPPLEMENTARY INFORMATION DOI: 10.1038/NPHYS3783
16 proximately between 0.95 and 1.81 by changing µ. By asymmetrically adjusting the laser beatnote
17 detuning µ about the carrier by a value of B we apply a global Stark shift that can be thought of as
19 We generate the effective disorder by applying a site-dependent Stark shift using a single
20 355nm laser beam that is focused down tightly to a 1/e2 waist of ∼ 1.8µm. The ion separation is
21 ∼ 2.5µm, thus the crosstalk between ions is negligible with a measured ratio of nearest-neighbor
22 Rabi frequencies of ∼ 20 : 1. We use an acousto optic modulator (AOM) with a full width at half
23 maximum bandwidth of ≈100 MHz to apply the Stark shift to each ion. The AOM is not imaged
24 onto the ions, so that driving the AOM with different frequencies allows the position of the beam to
25 be scanned over the length of a 10 ion chain, ∼ 20µm. The Stark shift is proportional to I 2 . Thus,
26 to achieve larger applied Stark shifts, we raster through the AOM drive frequencies corresponding
27 to addressing each ion with a total cycle time of ∼ 5µs instead of applying them simultaneously.
28 Since we cannot control the sign of the site-specific Stark shift, to center the disorder strength
29 around the global transverse field, we adjust the asymmetric detuning by W Jmax /2.
30 Sampling Error We determine that averaging over 30 different random realizations of disorder is
31 sufficient to have a sampling error smaller than the effect we observe by looking at the change in
32 the time-averaged HD with respect to a change in the disorder strength. Figure 3b makes explicit
33 that this error is much smaller than than the change in the time-averaged HD with respect to a
35 Measuring the spin-spin coupling matrix In order to observe the dynamics between just two
36 of the ions in the chain, we shelve the other spins out of the interaction space. This is done
37 by performing a π rotation between |↓z , 2 S1/2 |F = 0, mF = 0, and one of the Zeeman states,
2
38 S1/2 |F = 1, mF = −1, while shifting the two ions of interest out of resonance by applying a
39 large Stark shift with the individual addressing beam. We then apply our Hamiltonian which now
40 acts only on the two ions left in the interaction space and determine the elements of the spin-spin
41 coupling matrix by fitting the measured interaction Rabi flopping frequency between each pair of
42 spins.
43 Arbitrary product state preparation State initialization starts with optically pumping the spins
44 with high-fidelity to |↓↓↓ · · · z . Then we perform a global π/2 rotation to bring the ions to
45 |↓↓↓ · · · x . At this point we apply a Stark shift with the individual addressing beam to the spins
46 that are to be flipped and allow the chain to evolve until these ions are π out of phase with rest
47 of the ions. This, along with our ability to perform high fidelity global rotations, allows for the
48 preparation of any arbitrary product state along any direction of the Bloch sphere. Individual spin
49 flips can be achieved with a fidelity of ∼ 0.97, while arbitrary state preparation can be done with a
50 fidelity of ∼ (0.97)N , where N is the number of spins flipped with the individual addressing beam.
51 Determining a Set of Thermalizing Parameters Extended Data Figure 1 shows the time evolu-
52 tion of σix for different values of B for the spins initialized in the randomly chosen product state
53 |↓↓↓↑↓↓↓↑↓↑x . Without a transverse field, the spins are in an eigenstate of the Ising interaction
54 and undergo no evolution. Once a transverse field is added the individual spins begin to lose mem-
55 ory of their initial conditions and as its strength is increased, the ions thermalize faster and more
56 robustly.
57 To confirm the system is thermalizing, we measure the time evolution of the single site mag-
58 netization, σiz , along an orthogonal direction for different strengths of the transverse magnetic
59 field starting with the spins initialized in the Néel ordered state. As seen in Extended Data Fig. 1
60 the spins have lost information about their initial conditions in the z direction for all values of B.
61 We calculate the spectral statistics of adjacent energy levels for the Hamiltonian and find they
62 are not Poisson distributed for B = 4Jmax and Di = 0 indicating that with no applied disorder, the
63 system is not in a localized phase. Furthermore, one can determine if a system is in a thermal or
64 localized regime by finding the correlation between adjacent energy splittings by calculating the
66 where δn = En+1 − En ≥ 0. For a localized phase, where one expects a Poisson energy spectrum,
67 the probability distribution of this order parameter is given by Pp (r) = 2/(1 + r)2 and thus r ≈
68 0.39. For energy level spacings following a random-matrix as predicted for a thermalizing regime,
69 we calculate r ≈ 0.53 for a chain of 10 spins. Extended Data Figure 2 shows that r saturates
70 to the expected value for a random matrix distribution, indicating that the Hamiltonian is thermal
72 Calculating the density matrix expected by the Eigenstate Thermalization Hypothesis Given
73 a Hamiltonian and an initial state |ψ0 , the corresponding energy is ψ0 | H |ψ0 . For a thermalizing
74 system satisfying ETH this energy should be equal to the classical energy:
Tr[He−βH ]
E= (4)
Tr[e−βH ]
75 for the appropriate β = 1/(kB T ). When partitioning the entire system into subsystems A and B,
76 with the size of A much smaller than B (perhaps even a single spin), then, the density matrix on
TrB [e−βH ]
ρA = (5)
Tr[e−βH ]
78 Since we start in the Néel ordered state, the initial energy given the Hamiltonian in Eq. (1)
79 is equal to zero, ψ0 | H |ψ0 = 0. Equating this to the right hand side of Eq. (4) and solving for
80 β gives β = 0, or T = ∞. Using this β in Eq. (5) gives a value for any reduced thermal density
81 matrix of:
1/2 0
0 1/2
82 in agreement with the measured reduced density matrices in Fig. 2a.
83 Comparison to Numerics To demonstrate the MBL we observe is a general feature of our Hamil-
84 tonian we perform numerical simulations using exact diagonalization. Extended Data Figure 3
85 compares the experimentally measured time evolution of the normalized HD with numerics and
86 shows excellent agreement between them. We see similar agreement between experimental data
87 and numerics for the time evolution of the single-spin magnetizations (not shown). The aspects of
89 tics of (1)7 .
90 Decoherence and Dephasing To measure our system’s coupling to the environment we fit an
91 exponential decay to the dynamics in the upper left panel of Extended Data Fig. 1 as we expect
92 no time evolution of σix because the initial state is an eigenstate of the Hamiltonian and thus any
93 dynamics give an estimate of the decoherence rate. We find this estimate to be JM ax t = 64.6 which
95 Extended Data Fig. 4 shows a numerical simulation of the extended dynamics for the model
96 Hamiltonian with (green curves) and without (blue curves) crosstalk error between ions from the
97 individual addressing and laser intensity noise on the Ising couplings. It is clear that the localization
98 persists well beyond the experimental timescales even when accounting for experimental noise.
99 We model the crosstalk noise on the disordered field by adding 5% of the Stark shift applied
100 to adjacent ions to the size of the intended Stark shift which is consistent with the spillover we
101 measure between ions. To incorporate noise on the spin-spin couplings, we scale the strength of
102 the Ising couplings by a value we pull randomly from a Gaussian distribution centered around
103 µ = 1 with σ = 0.05 for each instance of disorder because the laser intensity noise is slower
104 than the duration of an experiment. The size of this simulated noise is consistent with the directly
106 Quantum Fisher Information The quantum Fisher information (QFI) has recently been shown
8, 9
107 to witness genuinely multipartite entanglement . From a quantum metrology perspective, the
108 QFI quantifies the sensitivity of a given input state to a unitary transformation eiϑÔ generated by
N
110 For a local operator Ô = i=1 Ôi (where the difference between largest and smallest eigen-
fQ ≡ FQ /N > 1 . (7)
112 To characterize the growth of entanglement out of the initial Néel state, the natural choice
N i z
113 of the generator Ô is the staggered magnetization, Ô = i=1 (−1) σi /2. Remarkably, this QFI is
114 proportional to the variance of the HD D(t) given by Eq. (2) of the main text,
FQ = 4N 2 (∆D̂)2 = [(−1)i+j σiz σjz ] − [ (−1)i σiz ]2 , (8)
i,j i
N i z
115 when associating D(t) = D̂(t), with D̂ = 1/(2N )[1 − i=1 (−1) σi ].
116 The QFI as defined in Eq. (6) assumes a pure state, i.e., that time evolution is purely unitary.
117 For mixed states, the QFI cannot be expressed as a simple expectation value of the operator Ô10 .
118 In general, decoherence reduces the purity of the system’s state over experimental time scales. To
119 show that the measured increase of FQ as defined in Eq. (8) is indeed due to coherent dynamics, we
120 compare to numerical calculations for a unitary time evolution using the experimental parameters.
121 Extended Data Fig. 5 shows the experimental data is always below the theoretical prediction for
122 a unitary time evolution. The loss of purity or other experimental imperfections such as detection
123 error, therefore, do not generate a false positive indicator of entanglement in our system.
124 Extended Data Figure 6 further establishes this point, showing an increase in the QFI we
125 measure and strong agreement between experimental data and theory when postselecting for mea-
126 sured states with 5-spin excitations. We postselect because when B J the Hamiltonian is
127 effectively an XY model and conserves i σiz , because Ising processes that flip spins along the
128 large field are energetically forbidden. However, because of camera detection error we find there is
129 significant leakage out of states with 5-spin excitations (≈ 70% expected numerically, ≈ 35% de-
130 tected) into states with 4 and 6-spin excitations (≈ 20% detected) which should not be populated
131 as the transverse magnetization is conserved modulo two spin flips in the transverse field Ising
132 model. Thus, we post-select for states with 5-spin excitations. Extended Data Figures 5 and 6
133 show a clear difference between the interacting case and a theory of free-fermions (see below) for
134 the experimental data and numerical simulations, thus, establishing that the growth in QFI in the
136 To study how the localization changes with system size, we performed a numerical finite-size
137 scaling. In order to obtain a well-behaved scaling, we use the Kac prescription11 , i.e., we adjust
138 the couplings as Jij = JN −1 |i − j|−α , where N = (N − 1)−1 i<j |i − j|−α . Note that using
139 this prescription the fundamental energy scale J differs by about a factor of 2 from Jmax , the value
141 For α > 2, the disordered long-range Ising Hamiltonian shows MBL behavior at large
142 disorder12 . In Extended Data Fig. 7, we plot the dynamics of the QFI for α = 3, where it grows
143 consistent with the characteristic long-time growth of entanglement for an MBL state. In par-
144 ticular, within a time window 2α < tJ < 3α where only next-nearest-neighbor interactions are
145 relevant, the system essentially behaves as a nearest-neighbor Ising model with a weak next-to-
146 nearest-neighbor coupling. For such a system, a logarithmic growth of entanglement is expected,
147 as we indeed find in that regime, see inset in Extended Data Fig. 7.
148 Moreover, in Extended Data Fig. 7, we compare our numerical results to the appropriate
149 long-range free-fermionic theory (see below), which shows a quick system-size independent sat-
150 uration of the QFI without further growth. Therefore, we conclude that the observed increase of
151 the QFI is not possible in a quantum system without many-body interactions, thus giving a clear
153 The situation is more complex at α = 1.13. For B = 0, it has been predicted that within
154 the range 1 < α < 2 delocalized behavior could be expected in the thermodynamic limit12 . As
155 seen in Extended Data Fig. 8, for the considered system sizes up to N = 14 the model displays all
156 essential signatures of MBL, as found for α = 3. However, the important question of whether this
157 localization persists in the thermodynamic limit can only be addressed with system sizes larger
158 than those accessible using exact diagonalization. Here, scaling our quantum simulator to larger
159 system sizes could thus resolve a difficult open question, namely of the existence of ergodicity in
160 the range 1 < α < 2. However, we would like to emphasize that the essential features of MBL
161 are nevertheless captured by the 10-spin experimental system. In particular, we still find a time
162 window consistent with a logarithmic growth of entanglement, see inset in Extended Data Fig. 8.
163 In order to make a stronger connection between growth in the QFI and growth of entangle-
164 ment we calculate the entropy of entanglement between two halves of the chain:
165 where ρA = T rB [ρ] and B is the other half of the spin chain. The entanglement entropy quantifies
167 In Extended Data Figs. 7 and 8 it is clear that there is long-time growth of the entanglement
168 entropy that is consistent with the expected growth for a MBL state13, 14 and is absent in the free-
169 fermion numerics. The difference between the numerics for the model Hamiltonian and the non-
170 interacting theory for the QFI and the entanglement entropy in Extended Data Figs. 9 and 10
171 distinguishes between the two cases for the experimental system size and timescale. These figures
172 also establish a qualitative connection between growth in the QFI and growth in entanglement.
173 To show that the QFI growth is truly due to interactions, we also compare numerics with the
174 experimentally applied Hamiltonian to a close approximation of H, Eq. (1) of the main text, with a
175 non-interacting theory. Using the Jordan-Wigner transformation, σj− → e−iθj cj , with the phase of
†
176 the string operator θj = π j<i cj cj , the Hamiltonian Eq. (1) can be mapped to a fermionic theory
H= Jij (c†i ei(θj −θi ) cj + c†i ei(θj +θi ) c†j + h.c.) − (B + Di )c†i ci . (10)
i<j i
178 If Jij contained only nearest-neighbor interactions, this Hamiltonian would become equivalent
10
179 to a free-fermionic theory. For general Jij , however, the string operators generate interactions
180 between the fermions. Over short times, and especially in a localized regime, the phases θj are
181 dominated by their initial values, i.e., it is a good approximation to replace (for the initial Neel
j
182 state) θj → π j<i ((−1) + 1)/2 in the Hamiltonian. This replacement amounts to approximating
183 H by a non-interacting fermionic theory with long-range hopping and pairing. The QFI for that
184 case is included in Extended Data Figs. 7 and 8. As one can see, the QFI quickly saturates to
185 values below fQ = 1. The experimentally and numerically observed further growth of the QFI
186 is thus truly due to interactions, and cannot be captured within a free-fermionic theory, even with
189
188 1. Mølmer, K. & Sørensen, A. Multiparticle Entanglement of Hot Trapped Ions. Phys. Rev. Lett.
191 2. Porras, D. & Cirac, J. I. Effective Quantum Spin Systems with Trapped Ions. Phys. Rev. Lett.
193 3. Kim, K. et al. Entanglement and Tunable Spin-Spin Couplings between Trapped Ions Using
194 Multiple Transverse Modes. Phys. Rev. Lett. 103, 120502 (2009).
195 4. James, D. F. V. Quantum dynamics of cold trapped ions with application to quantum compu-
197 5. Islam, R. et al. Emergence and Frustration of Magnetism and Variable-Range Interactions in
11
199 6. Oganesyan, V. & Huse, D. A. Localization of interacting fermions at high temperature. Phys.
201 7. Wu, Y-L. & Das Sarma, S. Understanding analog quantum simulation dynamics in coupled
203 8. Hyllus, P. et al. Fisher information and multiparticle entanglement. Phys. Rev. A 85, 022321
204 (2012).
205 9. Tóth, G. Multipartite entanglement and high precision metrology. Phys. Rev. A 85, 022322
206 (2012).
207 10. Braunstein, S. L. & Caves, C. M. Statistical distance and the geometry of quantum states.
209 11. Kac, M. Can One Hear the Shape of a Drum? Am. Math. Monthly 73, 1-23 (1966).
210 12. Burin, A. L. Energy delocalization in strongly disordered systems induced by the long-range
212 13. Bardarson, J. H., Pollmann, F. & Moore, J. E. Unbounded Growth of Entanglement in Models
214 14. Pino, M. Entanglement growth in many-body localized systems with long-range interactions.
12
0.5
0.0
-0.5
-1.0
1.0
Z Magnetization, 〈σiz〉
0.5
0.0
-0.5
-1.0
0 2 4 6 8 10 0 2 4 6 8 10 0 2 4 6 8 10 0 2 4 6 8 10
216
Jmaxt Jmaxt Jmaxt Jmaxt
217 Extended Data Figure 1: Measured thermalization in the transverse field Ising model. The
218 upper panels show the time dynamics of σix (different colors represent different ions) for 10
219 spins prepared in the random product state |↓↓↓↑↓↓↓↑↓↑x , for different transverse magnetic field
220 strengths. For B = 0 the spins are in a eigenstate and do not thermalize. However, as the strength
221 of B is increased the system begins to thermalize more robustly and quickly. The lower panel plots
222 the time evolution of σiz with 10 spins prepared in the Néel ordered in the z direction for different
223 transverse magnetic field strengths. We conclude that the system is in the thermalizing regime for
224 B = 4Jmax since we observe thermalizing behavior along two orthogonal directions. Error bars
13
0.54
0.52
0.50
r
0.48
0.46
1 2 5 10 20 50
226 BJmax
227 Extended Data Figure 2: Thermalizing level statistics. The calculated value of r with respect
228 to B saturates close to the predicted value for a random-matrix distribution (dashed black line)
229 implying that the Hamiltonian is in the thermal phase for sufficiently large B.
14
0.6
0.4
D(t)
0.2
W/J max =8
W/J max =6
W/J max =4
W/J max =2
0
0 2 4 6 8 10
230 Jmax t
231 Extended Data Figure 3: Comparison of the experimental data (crosses) with exact numerical
232 simulations (blue lines) for normalized Hamming distance. There is excellent agreement be-
233 tween the numerical simulations using the experimental parameters and the measured data. This
234 demonstrates that the observed effects are general features of the Hamiltonian.
15
0.6
0.4
D(t)
236 Extended Data Figure 4: Numerical simulations of the extended time evolution of the Ham-
237 ming distance for the model Hamiltonian (blue curves) and with experimental noise (green
238 curves). The localization we observe persists much longer than the experimental timescale in
239 the model Hamiltonian (blue curves) even when accounting for laser intensity noise and crosstalk
240 between the ions from the individual addressing beam (green curves).
16
1.4
W/J max = 6 W/J max =8
1.2
fQ (t)
0.8
0.1 1 10 0.1 1 10
241 Jmax t Jmax t
242 Extended Data Figure 5: Comparison of experimental data (green dots) with exact numerical
243 simulations for the experimentally applied Hamiltonian (thick blue lines) and free-fermion
244 Theory (orange) for QFI. The solid straight lines represent logarithmic fits to the numerical (light
245 blue) and experimental data (light green). Deviations from the ideal coherent dynamics due to
246 decoherence and other imperfections in the experimental setup, such as detection error, lead to a
247 reduction of the QFI. Importantly, this suggests that experimental imperfections do not generate a
248 false positive for entanglement. Moreover, there is long-time growth in the QFI from the measured
249 data and applied Hamiltonian numerics that is absent in the free-fermion theory.
17
1.4
W/J max = 6 W/J max = 8
1.2
fQ (t)
0.8
0.1 1 10 0.1 1 10
250 Jmax t Jmax t
251 Extended Data Figure 6: Comparison of postselected experimental data (green dots) with ex-
252 act numerical simulations for the experimentally applied Hamiltonian (thick blue lines) and
253 free-fermion Theory (orange) for QFI. The solid straight lines represent logarithmic fits to the
254 numerical results for the experimentally applied Hamiltonian (light blue) and postselected data
255 for results with 5 spin flips (light green). The increase in the postselected QFI and the agreement
256 between the postselected data and numerical simulations supports the claim that experimental im-
257 perfections decrease the value of the QFI for the full experimental data.
18
1.5 2
1
fQ (t)
S(t)
1
1.3
0.5
fQ (t)
1.1
0.9
1 10 100
0 0
0.1 1 10 100 1000 0.1 1 10 100 1000
258 Jt Jt
259 Extended Data Figure 7: QFI and entanglement entropy from exact diagonalization (α = 3
260 and W/J = 8). Left panel: When subject to disorder, the QFI of the staggered magnetization
261 shows a characteristic growth of entanglement (blue lines; from dark to light: N = 8, 10, 12, 14
262 averaged over 106 , 3 × 105 , . . . , 2500 disorder realizations). This growth is absent in a theory of
263 free-fermions with long-range hopping and pairing (green dashed lines with N = 14 (dark green)
264 averaged over 10000 realizations). Left panel inset: In a time window dominated by next-nearest
265 neighbor interactions, 2α < tJ < 3α , one observes a characteristic logarithmic entanglement
266 growth, expected for a MBL system with short-range interactions. Right panel: The entanglement
267 entropy between two halves of the chain shows long-time logarithmic growth for the interacting
268 case and saturates for the free-fermion theory consistent with the expectation for a MBL state and
269 single-particle localized state, respectively, and a qualitative agreement between growth in QFI and
19
1.5 3
1 2
fQ (t)
S(t)
1.3
0.5 1
fQ (t)
1.1
0.9
1 10 100
0 0
0.1 1 10 100 1000 0.1 1 10 100 1000
271 Jt Jt
272 Extended Data Figure 8: QFI and entanglement entropy from exact diagonalization (α =
273 1.13 and W/J = 8). Same color coding as in Extended Data Fig. 7. Importantly, for the experi-
274 mentally relevant system size of N = 10, we again find a time window consistent with a logarithmic
275 growth of entanglement in the growth of QFI (see left inset) and half-chain entanglement entropy.
20
0.25 1
L =8 L =8
L =10 L =10
0.2 L =12 L =12 0.8
L =14 L =14
0.15 0.6
δ fQ (t)
δ S(t)
0.1 0.4
0.05 0.2
0 0
0.1 1 10 100 1000 0.1 1 10 100 1000
276 Jt Jt
277 Extended Data Figure 9: Difference between numerics for the interacting model Hamiltonian
278 and free-fermion theory for QFI and entanglement entropy (α = 3 and W/J = 8). There is
279 a clear departure between the numerically calculated QFI and entanglement entropy for the model
21
0.5 2
L =8 L =8
L =10 L = 10
L =12 L = 12
L =14 L = 14
δ fQ (t)
δ S(t)
0.25 1
0 0
0.1 1 10 100 1000 0.1 1 10 100 1000
281 Jt Jt
282 Extended Data Figure 10: Difference between numerics for the interacting model Hamiltonian
283 and free-fermion theory for QFI and entanglement entropy (α = 1.13 and W/J = 8).
284 There is a clear departure between the numerically calculated QFI and entanglement entropy for
285 the model Hamiltonian and the free-fermion theory for N=10 on the experimental timescale.
22