Observation of a Discrete Time Crystal

Scientists have confirmed a brand new form of matter: time crystals
Perpetual motion without energy.
J. Zhang, P. W. Hess, A. Kyprianidis, P. Becker, A. Lee, J. Smith, G. Pagano,1 I.-D. Potirniche, 2 A. C. Potter, 2, 3 A. Vishwanath, 2, 4 N. Y. Yao, 2 and C. Monroe 1

1 Joint Quantum Institute, University of Maryland Department of Physics and National Institute of Standards and Technology, College Park, MD 20742
2 Department of Physics, University of California Berkeley, Berkeley, CA 94720, USA
3 Department of Physics, University of Texas at Austin, Austin, TX 78712, USA
4 Department of Physics, Harvard University, Cambridge, MA 02138, USA
(Dated: September 29, 2016)
About Newswire Expert

Newswire Expert is your daily connection to the wire. Covering U.S. and International business, entertainment, technology and financial news releases.
Time spin Ion Time Crystal ag ne tiz Disorder
Scientists have confirmed a brand new form of matter: time crystals
Perpetual motion without energy.
Observation of a Discrete Time Crystal
About Newswire Expert

Newswire Expert is your daily connection to the wire. Covering U.S. and International business, entertainment, technology and financial news releases.
Time spin Ion Time Crystal ag ne tiz Disorder
<p>Observation of a Discrete Time Crystal
J. Zhang, P. W. Hess, A. Kyprianidis, P. Becker, A. Lee, J. Smith, G. Pagano,1
I.-D. Potirniche,2 A. C. Potter,2, 3 A. Vishwanath,2, 4 N. Y. Yao,2 and C. Monroe1
1Joint Quantum Institute, University of Maryland Department of Physics and
National Institute of Standards and Technology, College Park, MD 20742
2Department of Physics, University of California Berkeley, Berkeley, CA 94720, USA
3Department of Physics, University of Texas at Austin, Austin, TX 78712, USA
4Department of Physics, Harvard University, Cambridge, MA 02138, USA
(Dated: September 29, 2016)
Spontaneous symmetry breaking is a funda-
mental concept in many areas of physics, rang-
ing from cosmology and particle physics to con-
densed matter1. A prime example is the breaking
of spatial translation symmetry, which underlies
the formation of crystals and the phase transi-
tion from liquid to solid. Analogous to crystals
in space, the breaking of translation symmetry in
time and the emergence of a “time crystal” was
recently proposed2,3, but later shown to be for-
bidden in thermal equilibrium4–6. However, non-
equilibrium Floquet systems subject to a periodic
drive can exhibit persistent time-correlations at
an emergent sub-harmonic frequency7–10. This
new phase of matter has been dubbed a “discrete
time crystal” (DTC)10,11. Here, we present the
first experimental observation of a discrete time
crystal, in an interacting spin chain of trapped
atomic ions. We apply a periodic Hamiltonian to
the system under many-body localization (MBL)
conditions, and observe a sub-harmonic temporal
response that is robust to external perturbations.
Such a time crystal opens the door for studying
systems with long-range spatial-temporal corre-
lations and novel phases of matter that emerge
under intrinsically non-equilibrium conditions7.
For any symmetry in a Hamiltonian system, its spon-
taneous breaking in the ground state leads to a phase
transition12. The broken symmetry itself can assume
many different forms. For example, the breaking of spin-
rotational symmetry leads to a phase transition from
paramagnetism to ferromagnetism when the temperature
is brought below the Curie point. The breaking of spatial
symmetry leads to the formation of crystals, where the
continuous translation symmetry of space is replaced by
a discrete one.
We now pose an analogous question: can the trans-
lation symmetry of time be broken? The proposal of
such a “time crystal”2 for time-independent Hamiltoni-
ans has led to much discussion, with the conclusion that
such structures cannot exist in the ground state or any
thermal equilibrium state of a quantum mechanical sys-
tem4–6. A simple intuitive explanation is that quantum
equilibrium states have time-independent observables by
Tim
e 
(100 Floquet periods)
H1
H2
H3
H1 = g(1 - ε) σi
y
H2 = Jij σi
x σj
x 
H3 = Di σi
x 
H1
H2
H3
H1
H2
H3
Global rotation
Interactions
Disorder
FIG. 1: Floquet evolution of a spin chain. Three Hamil-
tonians are applied sequentially in time: a global spin flip of
nearly pi (H1), long-range Ising interactions (H2), and strong
disorder (H3). The system evolves for 100 Floquet periods of
this sequence.
construction; thus, time translation symmetry can only
be spontaneously broken in non-equilibrium systems7–10.
In particular, the dynamics of periodically-driven Flo-
quet systems possesses a discrete time translation sym-
metry governed by the drive period. This symmetry can
be further broken into “super-lattice” structures where
physical observables exhibit a period larger than that of
the drive. Such a response is analogous to commensurate
charge density waves that break the discrete translation
symmetry of their underlying lattice1. The robust sub-
harmonic synchronization of the many-body Floquet sys-
tem is the essence of the discrete time crystal phase7–10.
In a DTC, the underlying Floquet drive should gener-
ally be accompanied by strong disorder, leading to many-
body localization13 and thereby preventing the quantum
system from absorbing the drive energy and heating to
infinite temperatures14–17.
Here, we report the direct observation of discrete time
translation symmetry breaking and DTC formation in a
spin chain of trapped atomic ions, under the influence
of a periodic Floquet-MBL Hamiltonian. We experimen-
tally implement a quantum many-body Hamiltonian with
long-range Ising interactions and disordered local effec-
tive fields, using optical control techniques19,20. Follow-
ar
X
iv
:1
60
9.
08
68
4v
1 
 [q
ua
nt-
ph
]  
27
 Se
p 2
01
6
2
0
20 40 60 80 100
- 1.0
- 0.5
0.0
0.5
1.0
0
20 40 60 80 100
- 1.0
- 0.5
0.0
0.5
1.0
0
20 40 60 80 100
- 1.0
- 0.5
0.0
0.5
1.0
0
20 40 60 80 100
- 1.0
- 0.5
0.0
0.5
1.0
0
20 40 60 80 100
- 1.0
- 0.5
0.0
0.5
1.0
0
20 40 60 80 100
- 1.0
- 0.5
0.0
0.5
1.0
0
20 40 60 80 100
- 1.0
- 0.5
0.0
0.5
1.0
0
20 40 60 80 100
- 1.0
- 0.5
0.0
0.5
1.0
0
20 40 60 80 100
- 1.0
- 0.5
0.0
0.5
1.0
0
20 40 60 80 100
- 1.0
- 0.5
0.0
0.5
1.0
(a)
W t3 = π
ε = 0.03, 2πJ0 t2 /(W t3)=0.07
(b)
(c)
(d)
(e)
ε = 0.11, 2πJ0 t2 /(W t3)=0.07
Ion # 1
Ion # 2
Ion # 3
Ion # 4
Ion # 5
Ion # 6
Ion # 7
Ion # 8
Ion # 9
Ion # 10
0
20
40
60
80
100
- 1.0
- 0.5
0.0
0.5
1.0
Time (T)
S
in
gl
e
Io
n
M
ag
ne
tiz
at
io
ns
0
20
40
60
80
100
- 1.0
- 0.5
0.0
0.5
1.0
S
in
gl
e
Io
n
M
ag
ne
tiz
at
io
ns
0
20
40
60
80
100
- 1.0
- 0.5
0.0
0.5
1.0
S
in
gl
e
Io
n
M
ag
ne
tiz
at
io
ns
0
20
40
60
80
100
- 1.0
- 0.5
0.0
0.5
1.0
S
in
gl
e
Io
n
M
ag
ne
tiz
at
io
ns
0.4
0.5
0.6
0
0.02
0.04
0.06
Frequency
0.4
0.5
0.6
0
0.02
0.04
0.06
Frequency (1/T)
0.4
0.5
0.6
0
0.02
0.04
0.06
Frequency
0.4
0.5
0.6
0
0.02
0.04
0.06
Interactions On
(1/T)
Frequency (1/T)
(1/T)
Time (T)
Time (T)
Time (T)
Ion # 3
Ion # 8
(f)
0
20 40 60 80 100
- 1.0
- 0.5
0.0
0.5
1.0
0
20 40 60 80 100
- 1.0
- 0.5
0.0
0.5
1.0
ε = 0.03, 
W t3 = 0
ε = 0.03, 
FF
T 
sp
ec
tru
m
FF
T 
sp
ec
tru
m
FF
T 
sp
ec
tru
m
FF
T 
sp
ec
tru
m
Interactions Off
Time (T)
Time (T)
Time (T)
Time (T)
Time (T)
Time (T)
M
ag
ne
tiz
at
io
n
M
ag
ne
tiz
at
io
n
M
ag
ne
tiz
at
io
n
M
ag
ne
tiz
at
io
n
FIG. 2: Spontaneous breaking of discrete time translation symmetry. Top panel: Time-evolved magnetizations of
each spin 〈σxi (t)〉 and their Fourier spectra, showing sub-harmonic response of the system to the Floquet Hamiltonian. (a)
When only the H1 spin flip is applied, the spins oscillate with a sub-harmonic response that beats due to the perturbation
ε = 0.03 from perfect pi-pulses, with a clear splitting in the Fourier spectrum. (b) With both the H1 spin flip and the disorder
H3, the spins precess with various Larmor rates in the presence of different individual fields. (c) Finally, adding the spin-spin
interaction term H2 (shown with the largest interaction phase J0t2 = 0.036 rad), the spins lock to the sub-harmonic frequency
of the drive period. Here the Fourier spectrum merges into a single peak even in the face of perturbation ε on the spin flip
H1. (d) When the perturbation is too strong (ε = 0.11), we cross the boundary from the discrete time crystal into a symmetry
unbroken phase10. Bottom panel: Individually resolved time traces. (e) Spin magnetization for all 10 spins corresponding to
the case of (b). (f) Spin 3 and 8 corresponding to the case of (c). Each point is the average of 150 experimental repetitions.
Error bars are computed from quantum projection noise and detection infidelities.
ing the evolution through many Floquet periods, we mea-
sure the temporal correlations of the spin magnetization
dynamics.
A DTC requires the ability to control the interplay be-
tween three key ingredients: strong drive, interactions,
and disorder. These are reflected in the applied Floquet
Hamiltonian H, consisting of the following three succes-
sive pieces with overall period T = t1 + t2 + t3 (see Fig.
1) (~ = 1):
H =

H1 = g(1− ε)
∑
i σ
y
i ,
time t1
H2 =
∑
i Jijσ
x
i σ
x
j ,
time t2
H3 =
∑
iDiσ
x
i
time t3.
(1)
Here, σγi (γ = x, y, z) is the Pauli matrix acting on the
ith spin, g is the Rabi frequency with small perturbation
ε, Jij is the coupling strength between spins i and j, and
Di is a site-dependent disordered potential sampled from
a uniform random distribution with Di ∈ [0,W ].
To implement the Floquet Hamiltonian, each of the
effective spin-1/2 particles in the chain is encoded in
the 2S1/2 |F = 0,mF = 0〉 and |F = 1,mF = 0〉 hyperfine
‘clock’ states of a 171Yb+ ion, denoted |↓〉z and |↑〉z and
separated by 12.642831 GHz21. We store a chain of 10
ions in a linear rf Paul trap, and apply single spin rota-
tions using optically-driven Raman transitions between
the two spin states22. Spin-spin interactions are gener-
ated by spin-dependent optical dipole forces, which give
rise to a tunable long-range Ising coupling23,24 that falls
off approximately algebraically as Jij ∝ J0/|i− j|α. Pro-
grammable disorder among the spins is generated by the
ac Stark shift from a tightly focused laser beam that ad-
dresses each spin individually20. The Stark shift is an
effective site-dependent σzi field, so we surround this op-
eration with pi2 -pulses to transform the field into the x
direction of the Bloch sphere (see Methods). Finally, we
measure the magnetization of each spin by collecting the
spin-dependent fluorescence on a camera for site-resolved
imaging. This allows access to the single-site magnetiza-
3
tion, σγi , along any direction with a detection fidelity
> 98% per spin.
The unitary time evolution under a single Floquet pe-
riod is
U(T ) = e−iH3t3e−iH2t2e−iH1t1 .
(2)
The first evolution operator e−iH1t1 nominally flips all
the spins around the y-axis of the Bloch sphere by an
angle 2gt1 = pi, but also includes a controlled perturba-
tion in the angle, εpi, where ε < 0.15. This critical ro-
tation step is susceptible to noise in the Rabi frequency
(1% rms) from laser intensity instability, and also op-
tical inhomogeneities (< 5%) across the chain due to
the shape of the Raman laser beams. In order to ac-
curately control H1, we use the BB1 dynamical decou-
pling echo sequence25 (see Methods) to suppress these
effects, resulting in control of the rotation angle to a pre-
cision < 0.5%. The second evolution operator e−iH2t2
applies the spin-spin Ising interaction, where the maxi-
mum nearest-neighbor coupling J0 ranges from 2pi(0.04
kHz) to 2pi(0.25 kHz) and decays with distance at a power
law exponent α = 1.51. The duration of the interaction
term is set so that J0t2 < 0.04 rad of phase accumulation.
The third evolution operator e−iH3t3 provides disorder to
localize the system, and is programmed so that the vari-
ance of the disorder is set by Wt3 = pi. In this regime,
MBL in the thermodynamic limit is expected to persist
even in the presence of long-range interactions26–28.
To observe the DTC, we initialize the spins to the state
|ψ0〉 = |↓〉x = 1√2 (|↓〉z + |↑〉z) through optical pumping
followed by a global pi2 rotation. After many periods of
the above Floquet unitary Eq.(2), we measure the mag-
netization of each spin along x, which gives the time-
correlation function
〈σxi (t)〉 = 〈ψ0|σxi (t)σxi (0) |ψ0〉 .
(3)
Figure 2 depicts the measured spin magnetization dy-
namics, both in the time and the frequency domain, up
to N = 100 Floquet periods. A single Floquet period
T is set to a value between 74-75 µs, depending on the
parameters in the Hamiltonian.
The global pi-pulse e−iH1t1 rotates the spins roughly
half way around the Bloch sphere, so that we expect a
response of the system at twice the drive period 2T , or
half of the Floquet frequency. The frequency of this sub-
harmonic response in the magnetization is sensitive to the
precise value of the global rotation in H1 and is therefore
expected to track the perturbation ε. This results in co-
herent beats and a splitting in the Fourier spectrum by
2ε (Fig. 2(a)). When we add disorder e−iH3t3 to the Flo-
quet period, the single spins precess at different Larmor
rates (Fig. 2(e)) and dephase with respect to each other
(Fig. 2(b)). Only upon adding Ising interactions e−iH2t2 ,
and hence many-body correlations, the spin synchroniza-
tion is restored (Fig. 2(c,f)).
0
0.05
0.1
0.00
0.02
0.04
0.06
0.08
0
0.05
0.1
0.00
0.02
0.04
0.06
0.08
0
0.05
0.1
0.00
0.02
0.04
0.06
0.08
0
0.05
0.1
0.00
0.02
0.04
0.06
0.08
0.048
0.012
0.072
2πJ0 t2 /(W t3)
2πJ0 t2 /(W t3)
2πJ0 t2 /(W t3)
2πJ0 t2 /(W t3)
ε
ε
ε
ε
0.024
(a)
P
ea
k
he
ig
ht
va
ria
nc
e
(a
rb
.)
(b)
2πJ0 t2 /(W t3)
0.05
0.1
0
0.02
0.04
0.06
P
er
tu
rb
at
io
n 
ε
Thermal
Discrete Time Crystal
Symmetry Unbroken
MBL
FIG. 3: Variance of the subharmonic peak amplitude
as a signature of the DTC transition. (a) Variances
of the central peak amplitude, computed over the 10 sites
and averaged over 10 instances of disorder, for four different
strengths of the long-range interation term J0. The cross-
over from symmetry unbroken state to a DTC is observed
as a peak in the measured variance of the sub-harmonic sys-
tem response. Dashed lines: numerical results, scaled verti-
cally to fit the experimental data (see supplementary infor-
mation). Experimental error bars are standard error of the
mean. (b) Cross-over determined by a fit to the variance peak
location (dots). Dashed lines: numerically determined phase
boundaries with experimental long-range coupling parame-
ters10. Grey shaded region indicates 90% confidence level of
the DTC to symmetry unbroken phase boundary.
The key result is that with all of these elements, the
temporal response is locked to twice the Floquet period,
even in the face of perturbations to the drive in H1.
This can be seen clearly as the split Fourier peaks from
Fig. 2(b) merge into a single peak in Fig. 2(c). This repre-
sents the “rigidity” of the discrete time crystal10, which
persists under moderate perturbation strengths. How-
ever, for large ε, the DTC phase disappears as evinced
by the decay of the sub-harmonic temporal correlations
and the suppression of the central peak heights, as shown
in Fig. 2(d). In the thermodynamic limit, these pertur-
bations induce a phase transition from a DTC to a sym-
metry unbroken MBL phase7–10, which is rounded into a
cross-over in finite size systems.
The phase boundary is defined by the competition be-
tween the drive perturbation ε and strength of the inter-
actions J0. We probe this boundary by measuring the
variance of the sub-harmonic spectral peak height, com-
puted over the 10 sites and averaged over 10 instances
of disorder. Figure 3(a) shows the variances as a func-
tion of the perturbation ε, for four different interaction
4
strengths. As we increase ε, the variance growth distinc-
tively captures the onset of the transition, with increased
fluctuations signaling the crossing of the phase bound-
ary. When the perturbations are too large, the crystal
“melts”. Figure 3(b) shows the fitted centers of the vari-
ance curve, on top of numerically computed phase bound-
aries with experimental parameters. The measurements
are in agreement with the expected DTC to time crystal
“melting” boundary, which displays approximately linear
dependence on the perturbation strength in the limit of
small interactions10.
0.00
0.02
0.04
0.06
0.08
0.10
0.12
0.00
0.02
0.04
0.06
0.08
P
ea
k
he
ig
ht
(a
rb
.u
ni
ts
)
Perturbation ε
2πJ0 t2 /(W t3)
0.072
0.048
0.024
0.012
FIG. 4: Subharmonic peak amplitude as a function of
the drive perturbation. Central subharmonic peak ampli-
tude in the Fourier spectrum as a function of the perturba-
tion ε, averaged over the 10 sites and 10 disorder instances,
for four different interaction strengths. Solid lines are guides
to the eye. The height decreases across the phase boundary
and eventually diminishes as the single peak is split into two.
Error bars: 1 s.d.
Figure 4 illustrates the amplitude of the subharmonic
peak as a function of ε, for the four different applied in-
teraction strengths. The amplitude drops off with larger
perturbations, and this slope is steeper as we decrease
the interaction strength. The sub-harmonic peak ampli-
tude observable in Fig. 4 can serve as an auxilary probe
of the rigidity. It is expected to be an order parameter
for the DTC phase and thus, to scale similarly to the
mutual information between spins8,10. This connection
also provides insight into the Floquet many-body quan-
tum dynamics, and in particular to the correlations or
entanglement that underly the DTC phase. Indeed, the
eigenstates of the entire Floquet unitary are expected to
resemble GHZ or spin-“Schro¨dinger Cat” states8,18. The
initial product state in the experiment can be written as
a superposition of two cat states: |↓↓ · · · ↓〉x = 1√2 (|φ+〉+
|φ−〉), where |φ±〉 = 1√2 (|↓↓ · · · ↓〉x ± |↑↑ · · · ↑〉x). These
two states evolve at different rates corresponding to their
respective quasi-energies, giving rise to the sub-harmonic
periodic oscillations of physical observables. Such oscil-
lations are expected to persist at increasingly long times
as the system size increases7,8,10.
In summary, we present the first experimental obser-
vation of discrete time translation symmetry breaking
into a DTC. We measure persistent oscillations and syn-
chronizations of interacting spins in a chain and show
that the discrete time crystal is rigid, or robust to
perturbations in the drive. Our Floquet-MBL system
with long-range interactions provides an ideal testbed for
out-of-equilibrium quantum dynamics and the study of
novel phases of matter that exist only in a Floquet set-
ting7–10,30,31. Such phases can also exhibit topological
order31–35 and can be used for various quantum infor-
mation tasks, such as implementing a robust quantum
memory36.
ACKNOWLEDGEMENTS
We acknowledge useful discussions with Mike Zaletel
and Dan Stamper-Kurn. This work is supported by
the ARO Atomic and Molecular Physics Program, the
AFOSR MURI on Quantum Measurement and Verifica-
tion, the IARPA LogiQ program, the IC Postdoctoral
Research Fellowship Program, the NSF Physics Frontier
Center at JQI, and the Miller Institute for Basic Re-
search in Science. A. V. was supported by the AFOSR
MURI grant FA9550- 14-1-0035 and Simons Investigator
Program.
[1] Chaikin, P. & Lubensky, T. Principles of Condensed
Matter Physics, vol. 1 (Cambridge University Press,
1995).
[2] Wilczek, F. Quantum time crystals. Phys. Rev. Lett.
109, 160401 (2012).
[3] Wilczek, F.
Superfluidity and space-time translation
symmetry breaking. Phys. Rev. Lett. 111, 250402
(2013).
[4] Bruno, P. Comment on “quantum time crystals”. Phys.
Rev. Lett. 110, 118901 (2013).
[5] Bruno, P. Impossibility of spontaneously rotating time
crystals: A no-go theorem. Phys. Rev. Lett. 111, 070402
(2013).
[6] Watanabe, H. & Oshikawa, M. Absence of quantum time
crystals. Phys. Rev. Lett. 114, 251603 (2015).
[7] Khemani, V., Lazarides, A., Moessner, R. & Sondhi, S. L.
Phase structure of driven quantum systems. Phys. Rev.
Lett. 116, 250401 (2016).
[8] Else, D. V., Bauer, B. & Nayak, C. Floquet time crystals.
Phys. Rev. Lett. 117, 090402 (2016).
[9] von Keyserlingk, C. W., Khemani, V. & Sondhi, S. L.
Absolute stability and spatiotemporal long-range order
in floquet systems. Phys. Rev. B 94, 085112 (2016).
[10] Yao, N. Y., Potter, A. C., Potirniche, I.-D. & Vish-
wanath, A. Discrete time crystals: rigidity, criticality,
and realizations. arXiv:1608.02589 (2016).
[11] This phase is also referred to as a pi-spin glass7 or a Flo-
quet time crystal8.
5
[12] Sachdev, S. Quantum Phase Transitions (Cambridge
Univ. Press, Cambridge, 1999).
[13] Nandkishore, R. & Huse, D. A. Many-body localiza-
tion and thermalization in quantum statistical mechan-
ics. Annual Review of Condensed Matter Physics 6, 15–
38 (2015).
[14] D’Alessio, L. & Rigol, M. Long-time behavior of isolated
periodically driven interacting lattice systems. Phys. Rev.
X 4, 041048 (2014).
[15] Ponte, P., Chandran, A., Papic, Z. & Abanin, D. A. Pe-
riodically driven ergodic and many-body localized quan-
tum systems. Annals of Physics 353, 196 – 204 (2015).
[16] Eckardt, A. Atomic quantum gases in periodically driven
optical lattices. arXiv:1606.08041v1 (2016).
[17] We note that under certain conditions, time-crystal dy-
namics can persist for rather long times even in the ab-
sence of localization before ultimately being destroyed by
thermalization18.
[18] Else, D. V., Bauer, B. & Nayak, C. Pre-thermal time
crystals and floquet topological phases without disorder.
arXiv:1607.05277 (2016).
[19] Smith, J. et al. Many-body localization in a quantum
simulator with programmable random disorder. Nat.
Phys. advance online publication, – (2016).
[20] Lee, A. C. et al. Engineering large stark shifts for con-
trol of individual clock state qubits. arXiv:1604.08840
(2016).
[21] Olmschenk, S. et al. Manipulation and detection of a
trapped Yb+ hyperfine qubit. Phys. Rev. A 76, 052314
(2007).
[22] Hayes, D. et al. Entanglement of atomic qubits using an
optical frequency comb. Phys. Rev. Lett. 104, 140501
(2010).
[23] Porras, D. & Cirac, J. I. Effective quantum spin systems
with trapped ions. Phys. Rev. Lett. 92, 207901 (2004).
[24] Kim, K. et al. Entanglement and tunable spin-spin cou-
plings between trapped ions using multiple transverse
modes. Phys. Rev. Lett. 103, 120502 (2009).
[25] Brown, K. R., Harrow, A. W. & Chuang, I. L. Arbitrarily
accurate composite pulse sequences. Phys. Rev. A 70,
052318 (2004).
[26] Burin, A. L.
Localization in a random xy model
with long-range interactions: Intermediate case between
single-particle and many-body problems. Phys. Rev. B
92, 104428 (2015).
[27] Yao, N. Y. et al. Many-body localization in dipolar sys-
tems. Phys. Rev. Lett. 113, 243002 (2014).
[28] There has been recent work29 questioning the stability
of MBL with anything longer-range than exponential-in-
distance interaction. However, the proposed mechanism
is not relevant on experimentally accessible length- and
time- scales.
[29] De Roeck, W. & Huveneers, F. Stability and instability
towards delocalization in mbl systems. arXiv:1608.01815
(2016).
[30] Bordia, P. & Luschen, H. & Schneider, U. & Kanp, M.
& Bloch, I. Periodically Driving a Many-Body Localized
Quantum System. arXiv:1607.07868 (2016).
[31] von Keyserlingk, C. W. & Sondhi, S. L. Phase structure
of one-dimensional interacting floquet systems. i. abelian
symmetry-protected topological phases. Phys. Rev. B
93, 245145 (2016).
[32] Else, D. V. & Nayak, C. Classification of topolog-
ical phases in periodically driven interacting systems.
arXiv:1602.04804 (2016).
[33] von Keyserlingk, C. W. & Sondhi, S. L. Phase struc-
ture of one-dimensional interacting floquet systems. ii.
symmetry-broken phases. Phys. Rev. B 93, 245146
(2016).
[34] Potter, A. C. & Morimoto, T. & Vishwanath, A. Topo-
logical classification of interacting 1D Floquet phases.
arXiv:1602.05194 (2016).
[35] Roy, R. & Harper, F.
Abelian Floquet symmetry-
protected topological phases in one dimension. Phys.
Rev. B 94, 125105 (2016).
[36] Bahri, Y. & Vosk, R. & Altman, E. & Vishwanath, A.
Localization and topology protected quantum coherence
at the edge of hot matter. Nature Communications 6,
7341 (2016).
METHODS SUMMARY
Dynamical decoupling sequence.
We use a pair of Raman laser beams globally illuminat-
ing the entire ion chain to drive qubit rotations22. The
ion chain has 25 µm length, and we shape the beams
to have 200 µm full width half maximum along the ion
chain, resulting in ∼ 5% intensity inhomogeneity. When
a fixed duration is set for H1 in Eq. 1 of the main text,
the time dependent magnetization for different ions ac-
quire different evolution frequencies, resulting in the net
magnetization of the system dephasing after about 10
pi-pulses. In addition, the Raman laser has rms inten-
sity noise of about 1%, which restricts the spin-rotation
coherence to only about 30 pi-pulses (80% contrast).
To mitigate these imperfections, we employ a BB1 dy-
namical decoupling pulse sequence for the drive unitary
U1 (written for each spin i):
U1(ε) = e
−iH1t1 = e−i
pi
2 σ
θ
i e−ipiσ
3θ
i e−i
pi
2 σ
θ
i e−i
pi
2 (1−ε)σyi ,
where
in addition to the perturbed pi
rotation
e−i
pi
2 (1−ε)σyi , three additional rotations are applied: a pi-
pulse along an angle θ = arccos(−pi
4pi ), a 2pi-pulse along
3θ, and another pi-pulse along θ, where the axes of these
additional rotations are in the x-y plane of the Bloch
sphere with the specified angle referenced to the x-axis.
In this way, any deviation in the original rotation from
the desired value of pi(1− ε) is reduced to second order.
See Ref.25 for detailed discussions.
Generating the effective Ising Hamiltonian
We generate spin-spin interactions by applying spin-
dependent optical dipole forces to ions confined in a 3-
layer linear Paul trap with a 4.8 MHz transverse center-
of-mass motional frequency.
Two off-resonant laser
beams with a wavevector difference ∆~k along a princi-
pal axis of transverse motion globally address the ions
6
and drive stimulated Raman transitions. The two beams
contain a pair of beatnote frequencies symmetrically de-
tuned from the spin transition frequency by an amount µ,
comparable to the transverse motional mode frequencies.
In the Lamb-Dicke regime, this results in the Ising-type
Hamiltonian in Eq. (1)23,24 with
Ji,j = Ω
2ωR
N∑
m=1
bi,mbj,m
µ2 − ω2m
,
(4)
where Ω is the global Rabi frequency, ωR = ~∆k2/(2M)
is the recoil frequency, bi,m is the normal-mode matrix,
and ωm are the transverse mode frequencies. The cou-
pling profile may be approximated as a power-law decay
Ji,j ≈ J0/|i − j|α, where in principle α can be tuned
between 0 and 3 by varying the laser detuning µ or the
trap frequencies ωm. In this work, α is fixed at 1.51 by
setting the axial trapping frequency to be 0.44 MHz, and
Raman beatnode detuning to be 155 kHz.
Apply disorder in the axial direction.
We apply the strong random disordered field with a
fourth-order ac Stark shift20, which is naturally an effec-
tive σzi operator. To transform this into a σ
x
i operator,
we apply additional pi/2 rotations. Hence the third term
in the Floquet evolution U3 (written for each spin i) is
also a composite sequence:
U3 = e
−iH3t3 = ei
pi
4 σ
y
i e−iDiσ
z
i t3e−i
pi
4 σ
y
i = e−iDiσ
x
i t3 .
METHODS
Experimental time sequence.
The Floquet time evolution is realized using the timing
sequence illustrated in Fig. 5.
The chain of 10 trapped ions is initialized in the
ground motional state of their center of mass motion us-
ing doppler cooling and sideband cooling (not shown).
Optical pumping prepares the ions in the |↓〉z state. We
then globally rotate each spin vector onto |↓〉x by per-
forming a pi2 -pulse around the y-axis (“Initialization” in
Fig. 5).
The H1 Hamiltonian (perturbed pi-pulses) lasts 14-15
µs depending on ε and it consists of a four pulse BB1
sequence as described in the methods summary. Our
carrier Rabi frequency is set such that we perform a pi-
pulse in less than 3 µs. Including the three compensating
pulses, the BB1 sequence requires 5 pi-pulse times to im-
plement U1(ε).
The H2 Ising Hamiltonian is applied for 25 µs (“Spin-
spin interactions” in Fig. 5). The pulse time was suffi-
ciently long that with the pulse shaping described below
the effects of the finite pulse time spectral broadening
were largely reduced. To compensate for the residual
off-resonant carrier drive, we apply a small amplitude
transverse field (“Compensation” in Fig. 5) for 2 µs.
For the H3 disorder Hamiltonian, we apply
∑N
i=1Diσ
z
i
(“Strong random disorder” in Fig. 5) generated by Stark
shifts as described in the Methods summary. This is
sandwiched in-between rotations around the y-axis to
convert this into disorder in σxi . After up-to 100 applica-
tions of the Floquet evolution, we rotate the state back
to the z-axis (“Prepare for measurement” in Fig. 5) and
detect the spin state |↑〉z or |↓〉z using spin-dependent
fluorescence.
Pulse shaping for suppressing off-resonant
excitations.
The optical control fields for generating H1, H2 and H3
are amplitude modulated using acousto-optic modulators
(AOMs) to generate the evolution operators. If the RF
drive to these modulators is applied as a square pulse will
be significantly broadened in the Fourier domain due to
fast rise and fall times at the edges (100 ns). As the pulse
duration decreases, the width of the spectral broadening
will expand (see Fig 5 inset). The components of the
evolution operator must be sufficiently short in order to
evolve 100 Floquet periods within a decoherence time of
< 8 ms.
This spectral broadening is problematic when generat-
ing the interaction Hamiltonian H2 due to off-resonant
driving. The spin-spin interactions are applied using
beatnote frequencies detuned 4.8 MHz from the carrier
transition and 155 kHz from the sidebands. A broad
pulse in the frequency domain can drive either the qubit
hyperfine transition at the carrier frequency or phonons
via sideband transitions.
Similar issues occur while we apply the disordered field
in H3. The fourth-order ac Stark shift is generated from
a frequency comb that has a closest beatnote which is 23
MHz away from hyperfine and Zeeman transitions. To
apply large average Stark shifts (33 kHz max.) across
the 10 ions, we raster the laser beam once within a single
cycle (30 µs), for a single-ion pulse duration of 3µs. The
fast rastering also produces off-resonant carrier driving
that resemble σyi fields.
We mitigate both these effects by shaping the pulses
with a 25% “Tukey” window , a cosine tapered function
for the rise and fall. This largely removes off-resonant
terms in all Hamiltonian chapters, while minimizing the
reduction in pulse area (80%). We carefully character-
ize any residual effects in H2 with a single ion where no
interaction dynamics are present, and apply a small com-
pensation field in σyi to cancel residual effects. We deduce
an upper limit of 0.3% (relative to H1) on the residuals
from the envelopes of the dynamically decoupled pi pulse
7
sequence.
Emergence of time crystal stabilized by the
interactions and the disorder.
In the main text we have shown different scenarios
which occur when we modify sub-sections of the total
Floquet evolution operator. Here we first expand the
results shown in Fig.2 of the main text by highlighting
the effect of the interactions. Figure 6 shows individual
magnetization evolutions with the same ε and instance
of disorder {Di} but with increasing interaction J0, from
left to right. This shows how, all else being equal, as
we turn up the interaction strength the synchronization
across the ions gradually builds up signaling the forma-
tion of a discrete time crystal.
In a complementary fashion, Fig. 7 addresses the role
of disorder in stabilizing the time crystal formation. In
particular we compare the magnetization dynamics, both
in the time and frequency domain, including or removing
the disorder chapter from the total Floquet evolution.
We show on the side the corresponding exact numerics
calculated by applying U(T )N on the initial state |ψ0〉
with the measured experimental parameters. Figure 7(a)
shows that, with no disorder, the ions coherent dynamics
tracks the perturbation ε, which results in coherent beats
in agreement with our exact results. On the other hand,
Figure 7(b) shows that, all else being equal, adding the
disorder chapter locks the subharmonic response of all the
ions. Although the numerics is qualitatively in agreement
with the experimental data, nevertheless we observe a
decay in the magnetization that cannot be explained by
our numerical unitary simulations.
This damping can be due to two possible sources: one
is the residual off-resonant drive of the disorder chap-
ter which is not totally eliminated by the pulse shap-
ing (see pulse-shaping section above). This small resid-
ual effect can behave like residual σxi and σ
y
i terms, or
coupling between the clock spin states and the Zeeman
states |F = 1,mF = ±1〉, resulting in a decay in the co-
herent oscillations. This effect varies across the different
disorder instances and the different interaction strengths
and leads to an overall decrease of the Fourier subhar-
monic peak height and of its variance with respect to
what is expected from the theory (see Fig.7(b)). To take
this effect into account in the theory, we perform a least
squares fit of the amplitude of the four theory curves to
the experimental data. With this procedure we obtain
the scaling factors (0.56, 0.53, 0.51, 0.78) for the inter-
action strengths J0t2/(Wt3) = (0.72, 0.48, 0.24, 0.12) re-
spectively, which are consistent with the decays discussed
above.
Data analysis and fitting procedure
The peak height variance data in Fig. 3(a) are fit to
a lineshape in order to extract the cross-over transition
boundary εp in Fig. 3(b). We use the following phe-
nomenologial lineshape, a Lorentzian of the log10()
F (ε) = A
1
1 +
(
log10(ε/εp)
γ
)2 +B
(5)
A statistical error is extracted from a weighted non-
linear fit to the data, which yields a fractional standard
error bar of a few percent. The error in the peak height
is limited by systematic error in the finite number of in-
stances we realized in the experiment. For each value of
J0 and ε we average over the same 10 instance of disorder.
These 10 instances are averaged and fit to F (ε).
In order to estimate the error due to the finite disorder
instances, we perform random sampling from a numerical
dataset of 100 instances of disorder. Sampling 10 of these
and fitting the peak height variance data to F (ε) yields
a Gaussian distribution of extracted peak centers p over
10000 repetitions (Fig. 8). We take the one standard
deviation (67%) confidence interval in the sample as the
systematic uncertainty in the fit, which yield a fractional
error of ∼ 15%. This systematic uncertainty dominates
over the statistical, so we apply error bars in Fig. 3(b)
equal to this computed finite sampling error.
8
U1
U2
Spin-spin
interactions
Programmable
random disorder
Compensation
Rotate z
 into x
π 2π π
π(1-ε) rotation
Rotate x
 back to z
Raster across 
10 sites
Pulse shaping
U3
14~15 μs
27 μs
33 μs
BB1 dynamical 
decoupling
Time
U (T)
Single
period
Initialization
Prepare for 
measurement
Spin state
detection (1 ms)
optical
pumping
(15 μs)
101 Floquet periods
Fourier spectrum
Unshaped
Shaped
(Tukey window)
FIG. 5: Experimental pulse sequence. See text for detailed explanations.
0
20
40
60
80
100
- 1.0
- 0.5
0.0
0.5
1.0
S
in
gl
e
Io
n
M
ag
ne
tiz
at
io
ns
0
20
40
60
80
100
- 1.0
- 0.5
0.0
0.5
1.0
S
in
gl
e
Io
n
M
ag
ne
tiz
at
io
ns
0
20
40
60
80
100
- 1.0
- 0.5
0.0
0.5
1.0
S
in
gl
e
Io
n
M
ag
ne
tiz
at
io
ns
0.4
0.5
0.6
0
0.02
0.04
0.06
S
pe
ct
ra
lD
en
si
ty
(a
rb
.)
0.4
0.5
0.6
0
0.02
0.04
0.06
S
pe
ct
ra
lD
en
si
ty
(a
rb
.)
0.4
0.5
0.6
0
0.02
0.04
0.06
S
pe
ct
ra
lD
en
si
ty
(a
rb
.)
Frequency (1/T)
Frequency (1/T)
Time (T)
Time (T)
2πJ0 t2 /(W t3)=0.024
2πJ0 t2 /(W t3)=0.07
ε = 0.035, 2πJ0 t2 /(W t3)=0.012
(a)
(b)
(c)
Frequency (1/T)
Time (T)
ε = 0.035, 
ε = 0.035, 
FIG. 6: Build up of a discrete time-crystal. From (a) to (b) and (c), we fix the disorder instance and the perturbation,
while gradually increase the interactions. The temporal oscillations are synchronized with increasing interactions, and the
Fourier sub-harmonic peak is enhanced.
9
0
20
40
60
80
100
- 1.0
- 0.5
0.0
0.5
1.0
Time (T)
S
in
gl
e
Io
n
M
ag
ne
tiz
at
io
ns
0
20
40
60
80
100
- 1.0
- 0.5
0.0
0.5
1.0
Time (T)
S
in
gl
e
Io
n
M
ag
ne
tiz
at
io
ns
0
20
40
60
80
100
- 1.0
- 0.5
0.0
0.5
1.0
Time (T)
S
in
gl
e
Io
n
M
ag
ne
tiz
at
io
ns
0
20
40
60
80
100
- 1.0
- 0.5
0.0
0.5
1.0
Time (T)
S
in
gl
e
Io
n
M
ag
ne
tiz
at
io
ns
0.4
0.5
0.6
0
0.05
0.1
Frequency (1/T)
S
pe
ct
ra
lD
en
si
ty
(a
rb
.)
0.4
0.5
0.6
0
0.05
0.1
Frequency (1/T)
S
pe
ct
ra
lD
en
si
ty
(a
rb
.)
0.4
0.5
0.6
0
0.05
0.1
Frequency (1/T)
S
pe
ct
ra
lD
en
si
ty
(a
rb
.)
0.4
0.5
0.6
0
0.05
0.1
Frequency (1/T)
S
pe
ct
ra
lD
en
si
ty
(a
rb
.)
Without Disorder
With Disorder
Experiment
Exact numerics
Experiment
Exact numerics
ε = 0.035,  2J0 t2 = 0.036, W t3=0
ε = 0.035,  2J0 t2 = 0.036, W t3=0
ε = 0.035,  2J0 t2 = 0.036, W t3=π
ε = 0.035,  2J0 t2 = 0.036, W t3=π
Ion 1
Ion 2
Ion 3
Ion 4
Ion 5
Ion 6
Ion 7
Ion 8
Ion 9
Ion 10
Ion 1
Ion 2
Ion 3
Ion 4
Ion 5
Ion 6
Ion 7
Ion 8
Ion 9
Ion 10
FIG. 7: Comparing ions dynamics with and without disorder. (a) Without disorder, the interaction suppresses the
beating. Left: experimental data; right: exact numerics calculated under the Floquet time-evolution. (b) With disorder, the
time-crystal is more stable. Left: experimnetal result for one single instance; right: numerical simulations.
0.00
0.05
0.10
0.15
0.20
0.25
0.30
0.00
0.01
0.02
0.03
0.04
Perturbation ε
P
ea
k
he
ig
ht
va
ria
nc
e
(a
rb
.)
0.030 0.035 0.040 0.045 0.050 0.055 0.060
0
10
20
30
40
50
60
Fit Center (εp)
P
D
F
FIG. 8: A random sampling from numerical evolutions. Averages of 10 disorder instances from numerical evolution
under H for 2piJ0t2/(Wt3) = 0.072. Left: An example random numerical dataset (points) and fit to Methods Eq. 5 (dashed
line). Right: The normalized probability distribution (PDF) of peak fit centers εp is shown in yellow, and a normal distribution
is overlaid in red. The normal distribution is fit using only the mean and standard deviation of the sample, showing excellent
agreement with Gaussian statistics. For this value of J0 the mean ε¯p = 0.046 and the standard deviation σεp = 0.006.
</p>