@2023-Quantum and Classical Discrete Time Crystals

Colloquium: Quantum and Classical Discrete Time Crystals
Michael P. Zaletel
Department of Physics,
University of California, Berkeley,
California 94720 USA
Materials Science Division,
Lawrence Berkeley National Laboratory, Berkeley,
Mikhail Lukin
Haarvard University, Cambridge,
arXiv:2305.08904v1 [quant-ph] 15 May 2023
MA 02138 USA
Christopher Monroe
Duke Quantum Center,
Department of Electrical and Computer Engineering,
Duke University,
Durham NC 27701 USA
Chetan Nayak
Microsoft Quantum, Station Q,
University of California, Santa Barbara,
CA 93106 USA
University of California, Santa Barbara,
CA 93106 USA
Frank Wilczek
Center for Theoretical Physics, MIT,
Cambridge MA 02139 USA;
T. D. Lee Institute and Wilczek Quantum Center,
SJTU, Shanghai;
Arizona State University,
Tempe AZ USA;
Stockholm University
Stockholm, Sweden
Norman Y. Yao
University of California, Berkeley,
(Dated: May 17, 2023)
The spontaneous breaking of time translation symmetry has led to the discovery of a
new phase of matter – the discrete time crystal. Discrete time crystals exhibit rigid
subharmonic oscillations, which result from a combination of many-body interactions,
collective synchronization, and ergodicity breaking. This Colloquium reviews recent
theoretical and experimental advances in the study of quantum and classical discrete
time crystals. We focus on the breaking of ergodicity as the key to discrete time crystals
and the delaying of ergodicity as the source of numerous phenomena that share many
of the properties of discrete time crystals, including the AC Josephson effect, coupled
map lattices, and Faraday waves. Theoretically, there exists a diverse array of strate-
gies to stabilize time crystalline order in both closed and open systems, ranging from
localization and prethermalization to dissipation and error correction. Experimentally,
many-body quantum simulators provide a natural platform for investigating signatures
of time crystalline order; recent work utilizing trapped ions, solid-state spin systems,
and superconducting qubits will be reviewed. Finally, this Colloquium concludes by
describing outstanding challenges in the field and a vision for new directions on both
the experimental and theoretical fronts.
2
CONTENTS crystalline and magnetic ordering to superfluidity, su-

perconductivity, and the generation of particle masses.
I. Introduction: Spontaneous Breaking of Time Translation Recently there has been a burst of activity surround-
Symmetry 2
ing the spontaneous breaking of symmetries that involve
II. What defines a time-crystalline phase of matter? 4 time translation (Else et al., 2016; Khemani et al., 2016;
A. Illustrative examples 5 Sacha, 2015; Shapere and Wilczek, 2012b; Wilczek, 2012;
Yao et al., 2017). These explorations have exposed new
III. Floquet Hamiltonian systems 7
A. Faraday waves 7 phenomena and new opportunities, but also important
B. Floquet Hamiltonians and emergent symmetries 7 subtleties. In this Colloquium, we will discuss both the
1. The rotating frame: constructing the Floquet resulting sharpening of theoretical concepts and the dis-
Hamiltonian 8 covery of previously unsuspected new phases of matter.
2. Time translation symmetry breaking as internal
symmetry breaking of the Floquet Hamiltonian 9
We will also discuss open questions and potential appli-
3. The spectral perspective 9 cations.
4. Application to many-body parametric resonance 10 Landau-Ginzburg theory is the starting point for many
C. Ergodicity, destroyer of time-crystals 11 theoretical treatments of spontaneous symmetry break-
IV. Closed, periodically-driven quantum systems 11
ing. In a simple but representative example of this frame-
A. Introduction to quantum Floquet phases 11 work, one considers the theory of a complex scalar field
B. Many-body localized discrete time crystal 13 φ(x, t) whose equations are invariant under a phase trans-
C. Experimental signatures of disordered time crystals 13 formation φ → eiλ φ. Physically, φ might represent the
1. Trapped ion spin chains 13
field associated with the creation and annihilation of
2. Spins in condensed matter 14
3. Superconducting transmon qubits 14 bosonic particles (e.g., 4 He atoms), s-wave spin singlet
D. Prethermal discrete time crystal 15 Cooper pairs, or the distribution of a planar spin den-
1. Floquet prethermalization 15 sity. Within the Landau-Ginzburg paradigm, a key role is
2. Prethermal discrete time crystal in a 1D trapped played by the energy (or more generally the free energy)
ion chain 16
E. Periodically driven Bose-Einstein Condensates 17
of a field configuration. The simplest energy functional
consistent with symmetries is given by:
V. Open, periodically-driven systems and stochastic
dynamics 19
A. “Activated” time-crystals 20
B. Experimental realizations of activated time-crystals: V (φ) = b4 |∇φ|4 + b2 |∇φ|2 + a2 |φ|2 + a4 |φ|4 = H = −L,
pendula, AC-driven charge density waves, and (1)
fractional Shapiro steps 21 where {a2 , a4 , b2 , b4 } are real-valued parameters while
C. Ergodicity in open systems 23 H, L denote the Hamiltonian and Lagrangian densities,
D. Probabilistic cellular automata 24
E. An absolutely stable open time-crystal: the π-Toom
respectively. Taking a4 , b4 > 0 insures stability against
model 24 rapid spatial variations and large fields.
F. Open Hamiltonian and quantum dynamics 26 If a2 < 0, states with hφi 6= 0 are energetically fa-
vorable, spontaneously breaking the phase rotation sym-
VI. Outlook and future directions 26
A. New venues for time crystals 26 metry. If b2 < 0, states with h∇φi = 6 0 are favorable,
1. Quantum many-body scars 26 spontaneously breaking the spatial translation symme-
2. Stark time crystals 26 try and indicating the emergence of spatial patterns. By
B. Prethermalization beyond Floquet quantum systems 27 reducing the symmetry, the system gains a non-zero en-
1. Classical prethermal discrete time crystals 27
2. Higher-M discrete time crystals 27
ergy per unit volume. In the limit of infinite volume, the
3. Prethermal time quasi-crystals 27 symmetry of all physically realizable states is less than
C. Applications: from metrology to quantum information the symmetry of the original equations. This is the mech-
benchmarking 28 anism of spontaneous symmetry breaking in equilibrium.
D. Concluding Remarks 29
Physicists are accustomed, in many contexts, to treat-
References 29 ing time and space on an equal footing. Thus, one could
imagine generalizing Eq. (1) to include time derivatives.
In a Lagrangian framework, it is natural to consider
I. INTRODUCTION: SPONTANEOUS BREAKING OF adding the leading order kinetic terms,
TIME TRANSLATION SYMMETRY
Lkin. = c4 |∂t φ|4 + c2 |∂t φ|2 , (2)
Spontaneous symmetry breaking is a remarkable col-
lective phenomenon: an assembly of constituents, each which corresponds to the energy density Hkin. =
interacting with only its nearby neighbors, manages to 3c4 |∂t φ|4 + c2 |∂t φ|2 . For c2 < 0, c4 > 0, it is energetically
align its behavior across large spatial and temporal sepa- favorable to have h∂t φi = 6 0, implying the spontaneous
rations. The concept has wide-ranging applications, from breaking of time translation symmetry.
3
This line of reasoning suggests that spontaneous time- behavior evolves with the same discrete time-translation
translation symmetry-breaking (Sτ B) is straightforward symmetry as the underlying equations of motion. In con-
to achieve. But subtleties abound (Shapere and Wilczek, trast, a system is said to exhibit a subharmonic response
2012b; Wilczek, 2012). Quantum mechanics requires if there are properties which oscillate at frequency ωD /m
Hamiltonians, and for non-singular Hamiltonians, Hamil- for some integer m > 1. Such a subharmonic response
ton’s equations, ∂t φ = ∂p H, ∂t p = −∂φ H, imply that spontaneously breaks the discrete time-translation sym-
a system is stationary at any energy minima. Thus a metry down to the smaller subgroup t → t + 2πn/ωD
ground state cannot exhibit periodic oscillations. The with n ∈ mZ. If this Sτ B is stable to perturbations, the
behavior of Eq. (2) is able to evade this this conclusion system is a “discrete time crystal” (Else et al., 2020b;
because ∂t φ is not a single-valued function of the canon- Guo and Liang, 2020; Khemani et al., 2019; Sacha, 2020;
ical momentum [i.e., p(∂t φ) = δL δ∂t φ does not have a
kin.
Sacha and Zakrzewski, 2017). Owing to the periodic
unique inverse ∂t φ(p)], and consequently, neither is the drive, it is not obvious how to stabilize such behavior.
Hamiltonian.(Shapere and Wilczek, 2012b) This is not Within the Landau paradigm, spontaneous symmetry-
necessarily an insurmountable problem. Indeed for such breaking phases retain their order in the face of fluctua-
singular Hamiltonians it is often possible to construct tions (both thermal and quantum) by virtue of the energy
consistent (in particular, unitary) quantum theories that penalty associated with misaligned regions. But this is
realize the corresponding classical dynamics in the limit not possible for a discrete time crystal, since energy is
of large quantum numbers (Chi and He, 2014; Choudhury not conserved in a driven system. The oscillations in dis-
and Guha, 2019; Shapere and Wilczek, 2012a,b; Zhao tant parts of the system must remain in lockstep even
et al., 2013). Moreover, singular Hamiltonians that sup- though the energy penalty for failing to do so can easily
port time-dependent minima can arise as limiting cases be over-ridden by the energy supplied by the drive.
of non-singular Hamiltonians as appropriate parameters The aim of this Colloquium is to give a unifying per-
are taken large. A limiting theory can be valuable, be- spective on various non-equilibrium mechanisms for dis-
cause it is often more tractable than the full theory, while crete Sτ B in closed (Fig. 1) and open (Fig. 2) systems.
being accurate within a large range of parameter space By centering our discussion in the language of dynami-
(Alekseev et al., 2020; Dai et al., 2020, 2019; Shapere cal systems, we hope to highlight the connections — and
and Wilczek, 2019). However, away from this limit sev- contrasts — between the rich literature on subharmonic
eral authors (Bruno, 2013; Nozières, 2013; Watanabe and responses in dynamical systems and more recent devel-
Oshikawa, 2015) have argued that persistent oscillations opments in the context of closed quantum systems. Until
cannot arise in a quantum system that is in equilibrium recently, many-body systems exhibiting Sτ B, such as pe-
with respect to a local Hamiltonian. riod doubling in coupled map lattices, were only found
These subtleties suggest that the application of the in open systems that relied fundamentally on dissipation.
Landau paradigm to Sτ B may not be as easy as Eq. (2) They could be viewed as a type of engine: the energy
would lead us to believe. An additional necessary ingredi- supplied by the frequency-ωD drive is converted to fre-
ent for Sτ B is a robust mechanism for ergodicity breaking quency ωD /m motion while releasing heat to a cold bath.
which enables the dynamics of the system to “remember” With the discovery of many-body localized (MBL) dis-
the initial condition (e.g. the complex phase of φ) out to crete time crystals, a qualitatively different form of Sτ B
infinite times. Spontaneous symmetry breaking is a spe- was discovered: one that does not generate any entropy
cial case of ergodicity breaking. Focusing on this aspect at all.
of spontaneous symmetry breaking leads to the following In an MBL time crystal, the time scale of Sτ B oscil-
formulations of Sτ B: it is a form of ergodicity breaking lations diverges exponentially as the system size alone is
in which a generic ensemble of initial conditions exhibits increased. However, for practical purposes it is also of
persistent oscillations with a temporal phase shift that interest to know if there are other physical parameters
remembers the initial condition (cf. (3)). whose limiting behavior can drive such an exponential
In recent years, a non-equilibrium route to Sτ B has increase, for example, the drive frequency ωD → ∞; the
been discovered in the context of Hamiltonians peri- temperature T → 0; or the particle density n → ∞.
odically driven at frequency ωD . Such “Floquet” sys- This will lead us to the notion of prethermal, activated,
tems (Floquet, 1883) feature a reduced, discrete time- and driven-BEC time crystals, respectively. A key tool
translation symmetry H(t + 2π/ωD ) = H(t), where which is common to both of these “exponentially good”
T0 = 2π/ωD is the driving period. When a simple har- time-crystals and to their MBL counterpart is the emer-
monic oscillator is driven at frequency ωD , it responds at gence of an effective time-independent Floquet Hamilto-
frequency ωD , regardless of its natural frequency. This is nian, Heff , which governs the dynamics out to exponen-
even true of many non-linear systems that are far more tially long time scales. Crucially, Heff can exhibit emer-
complex: when driven at frequency ωD , their observable gent symmetries that are protected by the underlying
properties respond at integer harmonics of ωD , regardless time-translation symmetry of the drive. In a pleasing re-
of their detailed structure. In this case, the observable turn to form, Sτ B can then be understood in terms of the
N. Y. Yao, A. C. Potter, I.-D. Potirniche, and A. Vishwanath
1
Department of Physics, University of California Berkeley, Berkeley, California 94720, USA
2
Department of Physics, University of Texas at Austin, Austin, Texas 78712, USA
3
RES EARCH | REPORT
Department of Physics, Harvard University, Cambridge, Massachusetts 02138, USA
(Received 5 November 2016; published 18 January 2017; publisher error corrected 19 January 2017)
4
Despite being forbidden in equilibrium, spontaneous breaking of time translation symmetry can occur in
periodically driven, Floquet systems with discrete time-translation symmetry. The period of the resulting
discrete time crystal is quantized to an integer multiple of the drive period, arising from a combination of
Closed / Deterministic
collective synchronization andSystems
many body localization. Here, we consider a simple model for a one-
dimensional discrete time crystal which explicitly reveals the rigidity of the emergent oscillations as the drive
is varied. We numerically map out its phase diagram and compute the properties of the dynamical phase
transition where the time crystal melts into a trivial Floquet insulator. Moreover, we demonstrate that
the model can be realized with current experimental technologies and propose a blueprint based upon a one
Classical Systems
dimensional chain of trapped ions.Quantum Systems
Using experimental parameters (featuring long-range interactions), we
identify the phase boundaries of the ion-time-crystal and propose a measurable signature of the symmetry
breaking phase transition.
DOI: 10.1103/PhysRevLett.118.030401
Classical Hamiltonian Classical/Quantum Hamiltonian Quantum Hamiltonian
Few-body Systems Spontaneous Floquet

symmetry Prethermalization
breaking—where a quantum state Many-body
In this Letter, we present Localization
three main results. First, by
breaks an underlying symmetry of its parent Hamiltonian— exploring the interplay between entanglement, many body
KAM Theorem à stability of subharmonic Largeadriving
represents unifying frequency
concept à in energy
modern absorption
physics [1,2]. localization Disorder
and TTSB, à breaking
we produce of ergodicity
a phase diagram and for a
response / time translation symmetry breaking from the
Its ubiquity spans drivefrom is exponentially
condensed matter suppressed
and atomic prevention of energy
(DTC) absorption fromlike driveother
Downloaded from http://science.sciencemag.org/ on July 23, 2021

discrete time crystal [24]. The DTC,
physics to high energy particle physics; indeed, examples
of the phenomenon abound in nature: superconductors,
Bose-Einstein condensates, (anti)ferromagnets, any crystal,
and Higgs mass generation for fundamental particles. This
diversity seems to suggest that almost any symmetry can be
broken.
ii
z
Spurred by this notion, and the analogy to spatial crystals,
h <latexit sha1_base64="I6TcQQtbEyJaTMeXtlWptdopZJ0=">AAACA3icbVDLSsNAFJ3UV62vqDvdDBbBVUlE0GXRjcsK9gFNDJPpJB06MwkzE6GGght/xY0LRdz6E+78G6dpFtp64MLhnHu5954wZVRpx/m2KkvLK6tr1fXaxubW9o69u9dRSSYxaeOEJbIXIkUYFaStqWakl0qCeMhINxxdTf3uPZGKJuJWj1PicxQLGlGMtJEC+8BjSMSMQE/RmKO7h4BCTxZSYNedhlMALhK3JHVQohXYX94gwRknQmOGlOq7Tqr9HElNMSOTmpcpkiI8QjHpGyoQJ8rPix8m8NgoAxgl0pTQsFB/T+SIKzXmoenkSA/VvDcV//P6mY4u/JyKNNNE4NmiKGNQJ3AaCBxQSbBmY0MQltTcCvEQSYS1ia1mQnDnX14kndOG6zTcm7N687KMowoOwRE4AS44B01wDVqgDTB4BM/gFbxZT9aL9W59zForVjmzD/7A+vwB1CqXoA==</latexit>
Wilczek proposed the intriguing concept of a “time crystal”—

a state which spontaneously breaks continuous time trans-
lation symmetry [3–5]. Subsequent work developed more
precise definitions trof such timelog(t) translation symmetry t⇤ breaking <latexit sha1_base64="+97l0TN4e+JZXiE1O8ESV7A4VCM=">AAAB6nicbVBNS8NAEJ3Ur1q/qh69LBZBPJRECnosevFY0X5AG8pmu2mXbjZhdyKU0J/gxYMiXv1F3vw3btsctPXBwOO9GWbmBYkUBl332ymsrW9sbhW3Szu7e/sH5cOjlolTzXiTxTLWnYAaLoXiTRQoeSfRnEaB5O1gfDvz209cGxGrR5wk3I/oUIlQMIpWesD+Rb9ccavuHGSVeDmpQI5Gv/zVG8QsjbhCJqkxXc9N0M+oRsEkn5Z6qeEJZWM65F1LFY248bP5qVNyZpUBCWNtSyGZq78nMhoZM4kC2xlRHJllbyb+53VTDK/9TKgkRa7YYlGYSoIxmf1NBkJzhnJiCWVa2FsJG1FNGdp0SjYEb/nlVdK6rHpu1buvVeo3eRxFOIFTOAcPrqAOd9CAJjAYwjO8wpsjnRfn3flYtBacfOYY/sD5/AH67Y2V</latexit>
(TTSB) [6–8] and ultimately, led to a proof of the “absence of

<latexit sha1_base64="+K3xE7puhevIG3xCcGLs8iSw3AE=">AAAB6nicbVBNS8NAEJ3Ur1q/oh69LBbBU0lE0GPRi8eK9gPaUDbbTbt0swm7E6GE/gQvHhTx6i/y5r9x2+agrQ8GHu/NMDMvTKUw6HnfTmltfWNzq7xd2dnd2z9wD49aJsk0402WyER3Qmq4FIo3UaDknVRzGoeSt8Px7cxvP3FtRKIecZLyIKZDJSLBKFrpAfu671a9mjcHWSV+QapQoNF3v3qDhGUxV8gkNabreykGOdUomOTTSi8zPKVsTIe8a6miMTdBPj91Ss6sMiBRom0pJHP190ROY2MmcWg7Y4ojs+zNxP+8bobRdZALlWbIFVssijJJMCGzv8lAaM5QTiyhTAt7K2EjqilDm07FhuAvv7xKWhc136v595fV+k0RRxlO4BTOwYcrqMMdNKAJDIbwDK/w5kjnxXl3PhatJaeYOYY/cD5/AGgcjd0=</latexit>
<latexit sha1_base64="cMgJeyN3YI3vEhpX04ao6WOVjiE=">AAAB7nicbVDLSgNBEOz1GeMr6tHLYBDiJeyKoMegF48RzAOSJcxOJsmQ2ZllplcISz7CiwdFvPo93vwbJ8keNLGgoajqprsrSqSw6Pvf3tr6xubWdmGnuLu3f3BYOjpuWp0axhtMS23aEbVcCsUbKFDydmI4jSPJW9H4bua3nrixQqtHnCQ8jOlQiYFgFJ3U6ko9rOBFr1T2q/4cZJUEOSlDjnqv9NXta5bGXCGT1NpO4CcYZtSgYJJPi93U8oSyMR3yjqOKxtyG2fzcKTl3Sp8MtHGlkMzV3xMZja2dxJHrjCmO7LI3E//zOikObsJMqCRFrthi0SCVBDWZ/U76wnCGcuIIZUa4WwkbUUMZuoSKLoRg+eVV0rysBn41eLgq127zOApwCmdQgQCuoQb3UIcGMBjDM7zCm5d4L96797FoXfPymRP4A+/zB7QQjyM=</latexit>
E.g. parametric resonance of a single non-linear (equilibrium)During quantum

Fig. 1. Experimental
the timeprotocol.
setup and
intermediatecrystals” (A) [9]. However,
Schematic
prethermal this proof
of the 25-ion
regime, chain body system to a featureless infinite temperature ensemble. (C) (Top) Phase
(35). Single-site
leaves the door addressing
open (top), global Raman
toorder
TTSB anbeams
inpersist (middle), and!/J
intrinsically state-
out-of- diagram of Heff. Owing to the antiferromagnetic nature of the Ising interactions
oscillator, Arnold tongues, etc. time-crystalline can for
dependent fluorescence (bottom) enable the preparation, evolution, and
⌧ ⇠ e Time crystalline order survives to ⌧ ! 1
Jij > 0, the ferromagnetic phase occurs at the top of the many-body spectrum.
<latexit sha1_base64="APUb2kdwUZGhMWBq0JXJtR9bRBo=">AAACAHicbVA9SwNBEN2LXzF+nVpY2CwGwSrciaBl0MYygvmA3BH2NnvJkr3dY3dOOUIa/4qNhSK2/gw7/42b5ApNfDDweG+GmXlRKrgBz/t2Siura+sb5c3K1vbO7p67f9AyKtOUNakSSnciYpjgkjWBg2CdVDOSRIK1o9HN1G8/MG24kveQpyxMyEDymFMCVuq5RwGQDAeaD4ZAtFaPOOAyhrznVr2aNwNeJn5BqqhAo+d+BX1Fs4RJoIIY0/W9FMIx0cCpYJNKkBmWEjoiA9a1VJKEmXA8e2CCT63Sx7HStiTgmfp7YkwSY/Iksp0JgaFZ9Kbif143g/gqHHOZZsAknS+KM4FB4WkauM81oyBySwjV3N6K6ZBoQsFmVrEh+IsvL5PWec33av7dRbV+XcRRRsfoBJ0hH12iOrpFDdREFE3QM3pFb86T8+K8Ox/z1pJTzByiP3A+fwD275ai</latexit>
equilibrium setting, and recent work [10,11] has demonstrated

<latexit sha1_base64="kZRsvNxq76nxTukRrSivxegsX6s=">AAACAXicbVDLSgNBEJz1GeNr1YvgZTAInuKuCHoMehFPEcwDsmuYnfQmQ2Zml5lZISzx4q948aCIV//Cm3/j5HHQxIKGoqqb7q4o5Uwbz/t2FhaXlldWC2vF9Y3NrW13Z7euk0xRqNGEJ6oZEQ2cSagZZjg0UwVERBwaUf9q5DceQGmWyDszSCEUpCtZzCgxVmq7+4EhGQ40Exju8yAR0CX4BN8M227JK3tj4HniT0kJTVFtu19BJ6GZAGkoJ1q3fC81YU6UYZTDsBhkGlJC+6QLLUslEaDDfPzBEB9ZpYPjRNmSBo/V3xM5EVoPRGQ7BTE9PeuNxP+8VmbiizBnMs0MSDpZFGccmwSP4sAdpoAaPrCEUMXsrZj2iCLU2NCKNgR/9uV5Uj8t+17Zvz0rVS6ncRTQATpEx8hH56iCrlEV1RBFj+gZvaI358l5cd6dj0nrgjOd2UN/4Hz+ABkPlf8=</latexit>
detection of the quantum dynamics. (B) For intermediate times (tpre < t < t*), (Bottom) Schematic of the stroboscopic magnetization dynamics in the trivial
that quantum systems subject to periodic driving can
the system approaches an equilibrium state of the prethermal Hamiltonian Heff. indeed (red) and PDTC (blue) phase (full and dashed curves represent even and odd
exhibit discrete
In the trivial FloquetTTSB [10–13];
phase, the suchafter
magnetization systems
tpre decaysdevelop
to zero. Inper-
the driving periods, respectively). When the energy density of the initial state is
FIG. 1 Schematic depicting strategies for sistent stabilizing
macroscopic
PDTC phase, time
the magnetization crystalline
oscillations
changes atsignan order
integer
every period, whichin
multiple periodically-driven,
leadsoftothe
a above the critical value ec, closed
the system issystems evolving
in the PDTC phase, and its lifetimevia
robust subharmonic
driving period, response. At times
manifesting in ta≫subharmonic
t&, Floquet heatingresponse
brings the many-
for follows the frequency-dependent heating time t*.
deterministic dynamics. In few-body classical systems, such as a parametrically driven non-linear oscillator, stable subharmonic
responses are ubiquitous. This stability can physical beobservables.
understood as a time-dependent extension
An important
exhibits constraint onbehavior
robust period-doubling symmetry up breaking in many-
simultaneously across FIG.
the full (a)of
1. chain Phasethe Kolmogorov-Arnold-Moser
(35),diagram
whereofsvithe discrete
is the time crystal
v-th component ofas
thea spin-1/2
function
(KAM) theorem (Givental et al., 2009; Kolmogorov, body
untilFloquet 1954;
systems
the frequency-controlled Möser,
is the need fortime
heating 1962),
disorder which
and
enabling proves
localiza-
the direct that quasiperiodic
of interaction
measurement ofstrength
the JPauli pulse orbits
z andoperator imperfectionsofion,
for the i-th dynamical
ϵ. (b)
andDepicts
we adopt the
scale, t(Sec.
* (Fig. 3B). In comparison, for the lat- Floquet dynamics location of the subharmonic Fourierℏpeak as a Jfunction oflong-
ϵ. In
systems remain robust to small perturbations tion [10–17]. III.B.4).
In There is no
the translation-invariant quantum
setting, Floquet ofanalog
both the magnetiza-
to this classical the convention ¼ 1. Here,
few-body ij >strategy.
0 is the
ter, all signatures of period doubling disappear tion and the energy density. the noninteracting caserange (J z ¼ 0), thewith
coupling peak tracksnearest-neighbor
average ϵ, while in the
For many-body systems, both classical and eigenstates
quantum
by the are short-range
dynamics
frequency-independent correlated
time scale can andexhibit
tpre resemble
The Floquetinfinite
Floquet prethermalization
interacting
drive alternates betweencasetwo (J z ¼interaction
0.15), the in the
peak
strength =limit
J0remains
2p · 0.33 of whereas
rigidly
kHz, large
locked
temperature
(Fig. 3A). By states whichthe
investigating cannot
lifetimeexhibit
of the symmetry breaking dynamics
types of Hamiltonian at ω=2. (Fig.
The 1B):
pink(i)region By =indicates
2p · 0.5 kHz theand Bz = 2p of
FWHM · 0.2the
kHz are global
base ofthe
the
driving frequencies (Sec. IV.D). During an intermediate prethermal
[15,18,19]. Under certain conditions, however, prethermal
time-crystalline order as a function of the
window
a global
of
p-pulse
time
around
(i.e.
the ŷ
before
axis and (ii) an
Floquet
effective
heating
magnetic fields.
occurs),
The Floquet unitary
ω=2 peak. Data are obtained at L ¼ 14 with 10 disorder averages. 2
system can exhibit discrete time crystalline order. The lifetime we of
time-crystal-like dynamics can persist for long times [20,21]
energy density of the initial state, identify this order
evolution for scales
time T exponentially
under a disorder-free, Uwith= U U the
implementsfrequency
(c)–(d) Representative realizations of the subharmonic Fourier
F 2 1 the dynamics of the
over a
the phase boundary for the PDTC. long-range, mixed-field Ising model. This is period of the drive and has frequency w = 2p/T.
drive, τ ∼ eωD /J . In strongly disordered, even in the absence
quantum of localization before
many-body
Our system consists of a one-dimensional systems, ultimatelya being
phenomenon response
described by the two evolution operators
corresponding
known as to in
many-body
ϵ (b). All Fourier transforms
localization
Within the prethermal window in time tpre <
are
destroyed
chain of 25by 171 thermalization [17,22].
Yb+ ions. Each ion encodes " computed # using 10 < tn <<t*150 , theFloquet periods.
can occur. This prevents Floquet heating (see Sec. III.C) and also provides a mechanism an ef-
pXN for the many-body system to avoid stroboscopic dynamics of the sys-
fective spin-1/2 degree of freedom in its hyper- Uinfinite syi tem (every other period) are well approximated
becoming ergodic. By breaking ergodicity, time crystalline order can persist to 1 ¼ exp "i times in the thermodynamic limit. Since
fine levels jF ¼ 0; mF ¼ 0i and jF ¼ 1; mF ¼ 2 by an effective prethermal Hamiltonian, which
0031-9007=17=118(3)=030401(6)
0i (Fig. 1A). Long-range Ising interactions 030401-1i © 2017
to lowest orderAmerican
in 1/w Physical
is given (30) Society
bystrategy.
many-body localization relies upon the discreteness of quantum mechanical "levels, there is no classical
!# analog to this
are generated through a pair of Raman laser
X X X N N N
U2 ¼ exp "iT Jij sxi sxj þ By syi þ Bz szi
beams (33, 34). Arbitrary effective magnetic N
X N
X
syi
i<j i¼1 i¼1
fields can be applied either locally or globally Heff ¼ Jij sxi sxj þ By ð2Þ
ð1Þ i<j i¼1
and single-site readout can be performed
breaking of these emergent symmetries and as an appli- nents” (Von Neumann, 1952). The answer to this ques-
Kyprianidis et al., Science 372, 1192–1196 (2021) 11 June 2021 2 of 5
cation of the Landau paradigm not to H(t), but rather tion is inextricably linked to the physical possibility of the
to Heff . most radical form of ergodicity breaking of all: classical
In the setting of Floquet dynamics, the environment and quantum error correction. From this perspective, if
couples to the system only via a coherent drive. But purely dissipative time crystals are a form of engine, and
in any physical experiment, the environment itself is a MBL time crystals a type of idealized perpetual motion,
many-body system, and thus, coupling to it inevitably then Sτ B in open systems is an embryonic example of an
comes along with dissipation and noise. Indeed, all ex- error-corrected computer program: repeatedly applying
periments on discrete time crystals to date show their the same “NOT” operation to all registers, the computer
fingerprints. The possibility of perfect Sτ B in the pres- settles into the period-2 output ...0101010101....
ence of a drive, dissipation, and noise remains an open
question.
II. WHAT DEFINES A TIME-CRYSTALLINE PHASE OF
The outlines of a possible answer to this question are MATTER?
provided by deep results from theoretical computer sci-
ence and non-equilibrium statistical physics. As empha- Discrete Sτ B leads to regular and long-lived oscilla-
sized, Sτ B is fundamentally a form of ergodicity break- tions with a period which is a multiple of the drive’s.
ing, and there is a history of rigorous results from these Oscillations in general, however, are ubiquitous through-
communities which show that ergodicity breaking can be out nature, so in this section we will formalize when
generically stable to noise (Gács, 2001; Toom, 1980). In such behavior signals the emergence of a genuine phase
that context, the motivation was to understand whether of matter. Our starting point is any dynamical system
“reliable systems can emerge from unreliable compo- (Birkhoff, 1927; Katok and Hasselblatt, 1997; Strogatz,
5
2018) whose state x evolves under a discrete-time update property of a dynamical phase of matter if it is robust to
rule Φ : x → Φ(x). The state of the system at time-step any small locality-preserving perturbation of either the
t is thus x(t) = Φ(t) (x(0)), where the superscript denotes initial condition x or the dynamics Φ. Under the pertur-
iteration. The time-independence of Φ implies there is bation Φ(x) → −x(1 − ), for example, oscillations are
a discrete time-translation symmetry. While the nota- damped at times beyond τ ∼ −1 , and Sτ B is destroyed.
tion highlights a discrete time-step, Φ may arise from When discussing stability, we need to be clear which
viewing a continuous-time system “stroboscopically.” In class of perturbations we demand stability against. Cor-
the context of classical and quantum dynamics defined respondingly this defines different possible classes of
by time-periodic Hamiltonian H(t + T ) = H(t), x is the Sτ B phases — Hamiltonian, unitary, Langevin, quan-
classical or quantum state, and Φ corresponds to integra- tum Lindladian, etc. — which would exhibit Sτ B robust
tion of Hamilton’s equations or the Schrodinger equation to arbitrary small perturbations within that dynamical
over one period of the drive. class. Depending on how broadly or narrowly we define
The dynamics exhibit m-fold time-translation symme- stability, we will have more or fewer examples, but weaker
try breaking if there exists a local observable O which or stronger implications.
exhibits periodic oscillations out to infinite times for a
measurable volume of initial conditions x:
A. Illustrative examples
τ τ
1X 1X
lim O(Φ(mn) (x)) 6= lim O(Φ(mn+p) (x)) The requirements surrounding Eq. (3) contain several
τ →∞ τ τ →∞ τ
n=1 n=1
subtleties, so it will prove helpful to walk through several
(3)
illustrative examples of “is X a time-crystal” with these
where 0 < p < m corresponds to the m-phases of the or- in mind. We first focus on examples which do exhibit
bit, with equality restored for p = m. A time-crystal thus oscillations (and are interesting in their own right), but
“remembers” which of m initial conditions it is in. This which do not satisfy the strict requirements of Eq.(3).
implies that Sτ B is a particular form of ergodicity break- Consider first the dynamics of an undriven oscillator
p2 ω02 2
ing (Cornfeld et al., 2012; Sinai, 1959; Walters, 2000): the H = 2m + 2m q + 4 q 4 . Obviously any initial condi-
time-averaged behavior of the m-fold iterated map Φ(m) tion x = (q, p) will oscillate forever - do such oscil-
depends on the initial condition. This behavior has been lations constitute a discrete time-crystal? Since H is
called “asymptotic periodicity” in the literature on clas- time-independent, for the purposes of testing Eq. (3) we
sical many-body dynamical systems (Lasota et al., 1984; may choose to stroboscopically observe the dynamics at
Lasota and Mackey, 2013; Losson and Mackey, 1996). whichever period T we suspect harbors the oscillations
While Eq.(3) implies infinitely long-lived oscillations, we (say T = 2π/ω0 ). However, stability requires us to ac-
may also consider a relaxed condition in which the time count for the non-linearity , which causes the oscillation
scale τ remains finite but diverges exponentially with a frequency to depend on the amplitude of the initial con-
control parameter such as the drive frequency or temper- dition. As a result, there is no fixed period T for which
ature. This will be the case for “prethermal” (Sec. IV.D) Eq. (3) is satisfied for a finite volume of initial conditions:
and “activated” (Sec. V.A) time-crystals, respectively. a generic ensemble of states will instead dephase, with no
We emphasize that Sτ B does not require that each Sτ B after some characteristic time τ .
state x itself undergoes perfectly periodic motion, only The time scale τ for such dephasing may be large, in
that there is an observable O which oscillates on average. which case the dynamics may appear to exhibit Sτ B in
For example, in the context of a single variable x, the ob- practice even while they do not satisfy Eq. (3) in princi-
servable O(x) = sign(x) might exhibit regular oscillations ple. Consider for example the AC-Josephson effect,
even while the motion of x itself is quite chaotic. This
n2
accounts for the fact that we rarely have direct access to H = −EJ cos φ + 2eV n + (4)
the microstate of the system, only coarse-grained mea- 2C
surements. This limitation also motivates the require- Here φ is the superconducting phase difference across a
ment that oscillations be observable for a finite volume Josephson junction, which is conjugate to the Cooper
of initial conditions, over which we rarely have exact con- pair number n; EJ is the Josephson energy for the tunnel-
trol. Taken together, an equivalent formulation of Eq.(3) ing of a pair across the junction, C is the junction capaci-
takes a statistical point of view by considering the evolu- tance, and V is the voltage difference across the junction.
tion of distributions over microstates ρ(x); we return to For a large junction the capacitance approaches C = ∞,
this formulation in our discussion of stochastic systems and the equations of motion then give φ(t) = φ0 + 2eV t,
in Sec. V. implying the supercurrent Is = ṅ = EJ sin(φ) oscillates
Even a single degree of freedom can trivially exhibit indefinitely in response to a DC voltage. However, for
Sτ B: take for example the map Φ(x) = −x. The prob- the generic case in which C is finite, a shift of variables
lem gains its richness when we demand stability: Sτ B is a from n → n − 2eV C brings H to the form of a pendulum
6
Open / Stochastic Systems

Classical Langevin / Quantum Linbladian
Temperature, T = 0 Temperature, T > 0
Dissipative Dynamical Systems Thermally Activated Toom / Gacs Models

Damping / friction à perturbations to either Low temperature à dissipation dominates “Error correction” à breaking of ergodicity
the state or the dynamics decay rapidly VOLUME 52, NUMBER 25 PHYSICAL REVIEW LETTERS 18 JUNE 1984
NbSe~
(aj V, (=SOmV
T=42K 1/2
1/2 1/1
25-
C?:
bQ 2/5 2/5 2/5
2/3
J,
15-
C3
C:
50=
CD
CD
20-
I5-
I I I I
-20 -10 0 10 20
sample voltage V (mV)
E.g. coupled logistic maps as a realization

FIG. 1. Differential resistance vs dc sample voltage (a) with and (b) without applied rf voltage at 25 MHz. The peaks
in (a) correspond to steps in the direct I- V curve. A few peaks are identified by p/q (see text).
E.g. AC-driven charge density wave materials, E.g. cellular automaton models (Gacs model in 1D,
of a dissipative many-body time crystal sured with and without externally applied rf voltage
parametrically
=
at co,„t/2n
driven coupled non-linear pendula
25 MHz. For the rather low rf voltage
of
„,
used here, ET is not reduced from the value in the
„,
quency locking between the internal CDW (or
Josephson) frequency rv;„, and all harmonics pt0,
the applied rf field. The harmonics of m, are Toom model in 2D) as many-body time crystals
absence of rf, but there appears a whole family of induced by the inherent nonlinearity of the system.
Time crystalline order survives to ⌧ !1

Time =crystalline order persists for ⌧ ⇠ e

peaks at impressed dc current such that
p~ext q int~ (3)
Direct spectral analysis of the current oscillations'
in the absence of applied rf reveals rich harmonic
current, with intensity of the harmonics qco;„, de-
<latexit sha1_base64="CZOkQDUvpqz1YAsAQcmSXbdpUxg=">AAACBHicbVDJSgNBEO1xjXEb9ZhLYxA8xRkR9BiiB48RskFmDD2dStKkZ6G7RghDDl78FS8eFPHqR3jzb+wsB018UPB4r4qqekEihUbH+bZWVtfWNzZzW/ntnd29ffvgsKHjVHGo81jGqhUwDVJEUEeBElqJAhYGEprB8HriNx9AaRFHNRwl4IesH4me4AyN1LELHrKUelqEFO4z7wYksjM67FRobdyxi07JmYIuE3dOimSOasf+8roxT0OIkEumddt1EvQzplBwCeO8l2pIGB+yPrQNjVgI2s+mT4zpiVG6tBcrUxHSqfp7ImOh1qMwMJ0hw4Fe9Cbif147xd6Vn4koSREiPlvUSyXFmE4SoV2hgKMcGcK4EuZWygdMMY4mt7wJwV18eZk0zkuuU3LvLorlyjyOHCmQY3JKXHJJyuSWVEmdcPJInskrebOerBfr3fqYta5Y85kj8gfW5w8YapcU</latexit>
/kB T Time crystalline order survives to ⌧ !1
where p and q are integers. The ratio cu;„$2rrIcDw caying slowly with q. Hence it is natural to interpret
was found to be 30 MHz cm2/A in agreement with the peaks as regions in which any harmonic of the
earlier studies. ' A few of the peaks are identified internal frequency locks to any harmonic of the
by p/q in the figure. The identifications were con- external field. ' To the extent that mode locking
firmed for several p/q by plotting Icnw vs co,„, and within such regions is complete, the CDW is unable
checking that the slopes were indeed p/q times that to respond to changes in the applied dc voltage, and
FIG. 2 Schematic depicting strategies for stabilizing time crystalline order in periodically-driven, open systems evolving via for the fundamental. The features with small p and
q are more conspicuous, being both taller and wid-
er. Figure 1 is a typical trace. By varying V,f and
the resistance rises to that of the normal electrons
alone. In the present experiment, the steps have
lower resistance than the normal electrons, indicat-
~e„,, either more or fewer steps can be observed. ing that the locking is not complete. The tendency
stochastic dynamics. Dissipative non-linear dynamical systems have long been known to exhibit stable, many-body time While the dependence on these parameters is not
strong, substantially increasing or decreasing either
to lock is weaker for larger p and q.
It is largely agreed that complete mode locking
' V,f or co,„, significantly reduces the number of steps occurs for Eq. (2) only because of the inertial
term. ~'
translation symmetry breaking. Indeed, coupled map " lattices
~ can satisfy all of the stated requirements for being a discrete time which can be observed.
The steps with q =1, p
previously
1 have been studied
and are well known to arise from fre-
In NbSe3 the inertial term is negligible
up to several hundred megahertz.
based on the frequency-dependent
' The argument,
conductivity
crystal. The key ingredient for the stability of Sτ B in this setting is a generalized version of dissipation—effectively, one can 2278
think of the microscopic dynamics in such systems as being coupled to a zero temperature bath (alternatively, one can say that
the microscopic dynamics are not information- or measure-preserving). This ensures that a finite volume of initial conditions
contracts towards, for example, a period-doubled fixed point (Sec. II.A). The stability of time crystalline order in many-body
systems coupled to a finite temperature bath are significantly more subtle. For example, in classical Langevin dynamics or
quantum Linbladian dynamics, at any T > 0, dissipation always comes with noise. In such systems (e.g. finite temperature,
parametrically-driven, coupled non-linear pendula), at low temperatures, time crystalline order can survive for an “activated”
time-scale ∼ e∆/kB T (Sec. V.A). Somewhat remarkably, by considering more generic stochastic dynamics (i.e. probabilistic
cellular automata), it is possible to realize finite-temperature, time crystalline order with an infinite lifetime (Sec. V.E).
whose behavior is similar to the non-linear oscillator dis- the internal motion of its many-body constituents. At
cussed
p above. The time scale for the resulting dephasing long times the system then relaxes to a state (e.g., van-
τ = C/EJ may be large - though it is worth pointing ishing particle number in the superfluid or spin aligned
out that this dependence is not an exponential. In ex- with the field in the magnet) wherein macroscopic mo-
periments, Josephson junctions are generally resistively tion ceases. To navigate this dichotomy, one must shield
shunted, and related interesting phenomena arise in this the system from explicit symmetry breaking apart from
open context as will be discussed in Sec. V. brief, intermittent measurement events, as in the exper-
Another interesting limit of Eq.(4) is EJ = 0, in which iment of Urbina et al., 1982. In this case, Sτ B is stable
case H has an internal U (1) phase symmetry which guar- only to perturbations which preserve the internal symme-
antees that φ̇ = 2eV + n/C remains constant. In this try. In contrast, we will find that a stronger form of Sτ B
case T = 2π/φ̇ This is an example of Sτ B which “piggy- is possible which depends only on the time-translation
backs” on the spontaneous symmetry breaking of an un- symmetry itself.
related internal symmetry. Equivalent examples are su- A many-body example which does satisfy all the re-
perfluids at non-zero chemical potential or an XY mag- quirements surrounding Eq. (3) is period doubling in de-
net in a perpendicular field. Such systems have a U (1) terministic dissipative systems. An iterated map x →
order parameter that precesses in time for generic initial f (x), such as the logistic map,(Strogatz, 2018) provides
conditions. The precession is unobservable (for example an idealized model of the unit-time evolution of such a
Is = 0) unless the U (1) symmetry, which makes all an- system. Taking for example f (x) = −x(1 + a − x2 ),
gles equivalent, is broken. But in the absence of U (1) there is a basin of attraction of initial states √
that set-
symmetry, there is no barrier to energy dissipating from tle into a limiting oscillation between x = ± a when
the macroscopic precession of the order parameter into a > 0. This simple one-body example can be promoted
7
to a many-body model (a “coupled map lattice”) by height q,

considering an array of variables xi which evolve under
Φ : xi → f (xi ) + v(xi−1 , xi , xi+1 ) with some generic lo- q̈k = −[ωk2 + δk cos(ωD t)]qk (5)
cal interaction v. The behavior of coupled map lattices is
very rich, since they include cellular automata (Gutowitz, where qk is the amplitude of the surface-wave at wave-
1991) as a special case (Bunimovich and Sinai, 1988; vector k, ωk is the natural frequency of the surface
Kaneko, 1984, 1992; Kaneko and Konishi, 1987; Kapral, wave, and δk is proportional to the driving amplitude.
1985). Coupled map lattices can support collective sub- The Mathieu equation features an exponentially-growing
harmonic responses (Gielis and MacKay, 2000; Losson solution (which precisely looks like a subharmonic re-
and Mackey, 1996; Losson et al., 1995) which are stable sponse) of the approximate form ∼ etΓ cos(tωD /2 + θ)
to smooth perturbations of f or v. when a mode satisfies ωD ∼ 2ωk (McLachlan, 1947).
As we will discuss, the non-linearity then regulates the
blow-up and stabilizes Sτ B. Since the Euler equations are
Hamiltonian (Olver, 1982), it might seem that Faraday-
III. FLOQUET HAMILTONIAN SYSTEMS wave instabilities then provide a Hamiltonian example of
Sτ B.
As noted above, if we consider systems with dissipa- However, while this analysis implies a linear subhar-
tion but no noise, as for example a coupled map lattice, it monic instability, it does not resolve whether the mo-
is relatively straightforward to obtain stable, many-body tion at long times is an example of Sτ B in the strong
Sτ B. This is the classical zero-temperature limit of the sense. While the non-linearities which are generically
general open system problem that we discuss at length present have the favorable property of regulating the ex-
in Section V. These systems illustrate that the existence ponential blow-up, they also couple different k-modes.
and stability of time crystals depends strongly on the A generic initial condition will contain some energy in
class of dynamical system considered (open, closed; clas- these high-k modes, which will then act as an effectively
sical, quantum). The presence of dissipation leads to noisy force on the motion of the k mode in which one is
a fixed-point orbit about which the linearized dynamics hoping to observe stable Sτ B. The key question can then
have eigenvalues |λ| < 1, so that a finite volume of initial be summarized as follows: Does Sτ B survive when treat-
conditions contracts towards the fixed point. In contrast, ing the Faraday wave problem as a genuine non-linear
for the canonical transformation which arises from inte- many-body system, and if not, what governs the auto-
grating Hamilton’s equations over one Floquet period, correlation time of the long-lived subharmonic response
the eigenvalues always come in conjugate pairs λ1 λ2 = 1 manifestly seen in experiments?
in order to ensure that the phase-space volume dq ∧ dp
is preserved in accord with Liouville’s theorem. In the
section, we discuss why stable Sτ B, i.e. an infinitely- B. Floquet Hamiltonians and emergent symmetries
long-lived time crystal, is not expected to occur in such
a situation. In order to answer this question, it will prove help-
ful to walk through a specific example and to develop
some intuition along the way. To this end, we will work
through the example of a shaken non-linear pendulum
A. Faraday waves (i.e. a parametrically driven non-linear oscillator); for a
single pendulum (and even in the few-body case), we
To illustrate why Hamiltonian structure makes the will see that one naturally gets stable Sτ B as a conse-
possibility of Sτ B considerably more challenging, it is quence of the Kolmogorov-Arnold-Moser (KAM) theo-
helpful to consider the concrete example of Faraday-wave rem. (Givental et al., 2009; Kolmogorov, 1954; Möser,
instabilities in shaken surface waves. Faraday observed 1962) In a truly many-body system, however, Sτ B be-
that when a container of water with a liquid-air inter- comes unstable. We will argue that there is an intuitive,
face is shaken vertically at frequency ωD , surface-waves general, mechanism underlying this obstruction: ergod-
n
develop which oscillate at frequencies ω = m ωD , which icity. Along the way, we will introduce and develop a
are rational subharmonics of the drive (Faraday, 1831; particularly important construct — namely, the Floquet
Rayleigh, 1883a). The sub-harmonics can be understood Hamiltonian — which will provide a unifying framework
as standing-wave modes that are forced via parametric for understanding many versions of Sτ B in closed Hamil-
resonance (Rayleigh, 1883b). A precise mathematical tonian systems (both quantum and classical).
understanding of the instability is obtained by lineariz- Consider the Hamiltonian for a shaken non-linear pen-
ing the incompressible Euler equations for surface waves dulum:
(Benjamin and Ursell, 1954) in the presence of periodic
acceleration. This reduces the problem to a set of “Math- p2 ω2
H(q, p, t) = + 0 (1 + δ cos(ωD t))q 2 + q 4 . (6)
ieu equations” for each Fourier component of the surface 2 2 4
8
58 R.S. Zounes, R.H. Rand / International Journal of Non-Linear Mechanics 37 (2002) 43}73
(a) (b) “Floquet Hamiltonian”: HF (Buchleitner et al., 2002;

Else et al., 2017, 2020a; Holthaus, 1995). This second
transformation takes the form of a so-called Magnus ex-
P pansion and the composite transformation is given by
TIME
TIME
K(t) = Kmagnus (t) ◦ K0 (t), again with K(mT ) = K(0).
MARKO RADUNOVIC ©123RF.COM

In summary, the strategy is to find a canonical transfor-
LY DRIVEN SYSTEM. In this schematic
ct how two of the spins (black and green)
mation K(t) : (q, p) → (Q, P ) such that the transformed
oscillating external source (dashed blue
ient behavior, the spins fall into lockstep
Q
FIGURE 2. CRYPTO-EQUILIBRIUM IN A DISCRETE TIME CRYSTAL.
Hamiltonian HF (Q, P ) is time-independent. This idea
he drive frequency. Such a subharmonic
In a 2T-periodic time crystal, the spin system’s overall orientation
a discrete time crystal: Its periodicity
nslation symmetry of the drive.
flips during each driving period, so it takesFig.
two 4.
periods
Levelforcurves
the spins
to return to something resembling their initial state. But to someone
that H(q, p, t) might actually be equivalent to an equi-
of the resonant Kamiltonian, K , as seen in the !}> plane. The resonance band is bounded by action values,
!
>+1.4 and >+2.8.
of a discrete time crystal and to draw

viewing the system at fixed intervals (that is, stroboscopically) from
FIG. 3 (a) Schematic depiction of a time-dependent rotat-
a frame of reference that flips with the spins, the system appears to
be in equilibrium, and it exhibits many features of an equilibrium
librium system (i.e. governed by HF ) in an appropriate
host of superficially similar-looking
ilibrium phenomena,10 some of which
ore (see the article by Ray Goldstein
system; no entropy is generated and oscillations need no other
ing frame, which transforms the original time-dependent H(t)
inputs to sustain them. We therefore say the system is in crypto- time-dependent rotating frame is referred to as “crypto-
equilibrium.
into the time-independent Floquet Hamiltonian, HF . Time- equilibrium” in (Else et al., 2020b; Yao and Nayak, 2018).
time crystal—and what isn’t oscillator, if our initial state is a superposition of two eigen-
translation symmetry breaking (here with period m = 2)
symmetry breaking manifests itself in states then the system will naturally exhibit oscillations at a fre-
We emphasize that a-priori, the Magnus expansion
quency given by the difference between the eigenstate ener-
can be understood as spontaneous symmetry breaking of an
#er a possible initial transient period,
ime oscillations with a period longer
gies. But in most systems, a generic initial state will not lead to
late-time oscillations with ω ≠ Ω. So any good definition of may not converge, in which case the pair of objects
TTSB in Floquet systems must generalize the ground-state or
internal Zm -symmetry of the Floquet Hamiltonian. Fig-
No entropy is generated by the late- equilibrium notion of spontaneous symmetry breaking in such
a way that the oscillatory behavior does not depend on the
ure adapted from (Yao and Nayak, 2018). (b) Contours
er: The oscillations remain in phase choice of initial state.
{K(t), HF } need not exist — this will be the precise ob-
ances and times. That requirement—and thus a precise definition of TTSB in
of equal “quasi-energy” HF (Q, P ) for a non-linear paramet-
s more precise is a bit subtle. When periodically driven systems—can be stated in a remarkably struction to stable Sτ B in the many-body Faraday-wave
a$er, one typically starts with a pre- compact form: A discrete time crystal is a state of ma$er in
tate, but not in discrete time crystals: which the Floquet eigenstates are necessarily “cat states,” that
ric oscillator near a 2:1 resonance. The original coordi-
ms do not have a ground state. Again, is, entangled superpositions of macroscopically distinct states.
case. Note that in the quantum setting (which we will
crystals helps. Consider a Floquet One immediate corollary is that because Fig. 5.
anyNumerically generated PoincareH map induced by solutions of the non-linear Mathieu equation: ""0.25, #"0.05, $"2.0, and
initial physical
nates of the oscillator q, p are related to the coordinates Q, P
period T, so that for any time t the
s H(t + T) = H(t). The discrete spatial-
%"1. The
state we can prepare must be a superposition solution
of such Floquet
cat states, all preparable initial states will exhibit oscillations.
! return to in just a few paragraphs), one can, in princi-
is strobed at times t "n2!/$, n"1, 2, 3,2 .
through a time-dependent canonical transformation given ap-

f a one-dimensional spatial crystal
aving a quasi-momentum that is only
Let us further unpack that definition. Since the Floquet
eigenstates of a discrete time crystal aren’t preparable, we can ple, always define a Floquet Hamiltonian by taking the
Using the resonance Kamiltonian, K , we obtain analytic expressions for the location of equilibrium points
ere a is the period of the crystal la$ice. readily distinguish the discrete time crystal from, for example,
!
proximately by and
time-translation symmetry of a peri- Eq. (7).of theThe
the width two
resonance band. Weminima oftheHdynamics
do not examine F , at
a simple harmonic oscillator. As mentioned above, whether the which
in the logarithm
stochastic layers arising from of the Floquet unitary; thus, in this case, one
eads to eigenstates having a quasi- harmonic oscillator displays time-periodic behavior depends
the intersection of the invariant manifolds of saddle points, since the integrable nature of the unperturbed
(Q, P ) are time-independent,
ed modulo Ω ≡ 2π/T, and there is no
minimum value of the quasi-energy.
system is preserved under themap back
perturbation to the two exactly-
strongly on the choice of initial state one prepares, and most
method.
importantly, nothing prohibits us from preparing a harmonic should replace the “need not exist” from the previous
TSB in periodically driven, nonequi-
period doubled orbits,
be cast in terms of ground-state or q (t) = −q (t + T0 ). As predicted by
oscillator in its ground state. By stark contrast, in a discrete
6.1. Location of∗equilibrium points ∗
time crystal every physically preparable initial state will ex-
ourselves to ground-state properties,

hibit oscillations at late times.
Eq.(11), time-translation symmetry manifests as a Z2 symme-
The key features of discrete time crystals explicitly distin-
sentence with the fact that HF will be highly non-local.
Consider the resonance band associated with periodic orbits in 2m : 1 resonance with the parametric
frequency ω ≠ Ω—a requirement for
rly easily. Even for a simple harmonic
guish them from a multitude of other systems (see the table on
try HF (Q + π, P ) = HF (Q, P ). Figure adapted from Fig. 4
page 46) that exhibit oscillations with unexpected periods.10
forcing. The band emerges from a resonant torus * an invariant closed curve of the unperturbed
As a preview, we note that the distinction between a
PoincareH
TEMBER 2018
of (Zounes and Rand, 2002). true discrete time crystal (Sec. IV.B) and the prethermal
discrete time crystal (Sec. IV.D) can precisely be under-
stood as whether the Magnus expansion converges. In
To connect to the Faraday wave discussion above, the the former case, it does and one has a well defined Flo-
reader may think of this pendulum as a specific k-mode, quet Hamiltonian, HF , while in the latter case, it almost
where we are now explicitly accounting for the non- does, leading to:
linearity but neglecting the inter-mode coupling. The
linearized equations of motion give rise to the Mathieu H(Q, P, t) = Heff (Q, P ) + V (t). (8)
equation [c.f. Eq. (5)] and the goal, with respect to time
crystalline order, would be to prove that the non-linearity Here, Heff is an effective Hamiltonian that captures the
stabilizes Sτ B. system’s stroboscopic dynamics for exponentially long
time-scales and V (t) is the (exponentially small) resid-
ual time-dependence which cannot be “rotated” away by
1. The rotating frame: constructing the Floquet Hamiltonian the Magnus expansion.
The concept of a “rotating frame” transformation in
Near a period doubled solution of Eq. 6, one expects on the quantum case is completely analogous. In particular,
physical grounds that trajectories should approximately within the quantum setting, the dynamics are captured
take the form by a unitary time evolution operator:
p Z t1
q(t) − ip(t) = 2P (t)ei(ωD t/2+Q(t)) (7)
U (t1 , t0 ) = T exp −i H(t)dt , (9)
t0
where Q(t), P (t) are slowly varying in comparison to ωD .
The transformation that takes (q, p) → (Q, P ) is a time- where T represents time-ordering. In this case, K(t)
dependent canonical transformation, which we denote by is a unitary transformation chosen such that the time-
K0 (t); this transformation can be thought of as taking evolution operator U (t1 , t0 ) decomposes as
the system to a particular “rotating frame” (Fig. 3a).
Note that the period of the rotation is sub-harmonic, U (t1 , t0 ) = K−1 (t1 )e−i(t1 −t0 )HF K(t0 ) (10)
K0 (mT ) = K0 (0) (in the period-doubled case, m = 2).
In the rotating frame, the Hamiltonian, H(Q, P, t) is with K(mT ) = K(0) and HF suitably local. K(t) is re-
still time-dependent, but the natural frequencies of the ferred to as the “micromotion,” since it contains the com-
system (i.e. those which govern the slowly-varying mo- ponents of the dynamics at harmonics of ωD /m, while
tion of Q(t), P (t)) are now off-resonant with ωD . Thus, the “slow” part of the motion resides in HF (Buchleitner
one can attempt to follow up with a second transforma- et al., 2002; Bukov et al., 2015; Eckardt and Anisimovas,
tion in order to generate a completely time-independent 2015; Rahav et al., 2003; Shirley, 1965; Zel’Dovich, 1973).
9
When restricting to the stroboscopic dynamics, the “Flo- frame using K−1 (t), one finds that the average behav-
quet unitary” which implements a single discrete time ior will oscillate with period m. More rigorously, sym-
step is given by UF = U (T, 0) = K−1 (T )e−iT HF K(0). To metry breaking implies that there are m different ini-
ensure that the connection to the notation in Eq. 3 is tial conditions yj = (Qj , Pj ) which are related by the
clear, we note that Floquet unitary, UF , then furnishes symmetry, yj+1 = Xyj . The time-averaged behaviors of
the dynamical map Φ which acts on the quantum state these different initial conditions are distinct (in the ro-
x = |ψi. tating frame),
Pτ e.g., there are local observables O with
Ō(yj ) ≡ τ1 n=0 O(yj (nT )) 6= Ō(yj+1 ). Transforming
back to the lab frame, xj (t) = K−1 (t)yj (t) and substi-
2. Time translation symmetry breaking as internal symmetry tuting in the definitions, one immediately finds that the
breaking of the Floquet Hamiltonian definition of time-crystalline order [c.f. Eq. (3)] is satis-
fied.
If the micromotion is periodic, K(T ) = K(0), there One thus arrives at a correspondence between m-fold
can be no Sτ B, since the stroboscopic dynamics be- Sτ B and the more familiar notion of Zm -spontaneous
come canonically equivalent to the un-driven dynamics symmetry breaking with respect to HF (in the rotat-
of HF , and Sτ B is forbidden in equilibrium (Bruno, ing frame). This illustrates a general principle of all
2013; Watanabe and Oshikawa, 2015). But in a time- closed Hamiltonian many-body time-crystals discovered
crystal, the frame rotates subharmonically (Fig. 3a), thus far (Fig. 1): In a Hamiltonian time-crystal, the dis-
i.e. K(mT ) = K(0), and the transformation X = crete time-translation symmetry of H(t) manifests as a
K(0)K−1 (T ) is non-trivial. The transformation X then spontaneously broken internal Zm -symmetry of the Flo-
plays a crucial role since it permutes through the m cy- quet Hamiltonian HF (Else et al., 2017, 2020a; von Key-
cles of the sub-harmonic response. In the specific case of serlingk et al., 2016). In a sense this bring us back full
the shaken non-linear pendulum [c.f. Eq. (7)], for exam- circle to our starting inspiration, the Landau paradigm
ple, X : (Q, P ) → (Q + π, P ). We will expand upon this (Sec. I), with the crucial replacement H → HF .
discussion in the next subsection (Sec. III.B.4).
Under rather general conditions, the Magnus ex-
3. The spectral perspective
pansion can be constructed to ensure that X =
K(nT )K−1 (nT +T ) is independent of n (Else et al., 2017,
While the above formalism applies equally well in the
2020a; Machado et al., 2020), which then ensures that
classical and quantum settings, in the quantum case an
X m = I. Substituting Eq. (10) into the time-translation
equivalent definition can be given in terms of the spectral
symmetry U (T, 0) = U (2T, T ), one finds that
properties of the Floquet operator UF alone (Buchleit-
ner et al., 2002; Holthaus, 1995; Sacha, 2015): a discrete
X −1 e−iT HF X = e−iT HF . (11)
time crystal is a phase of matter in which the eigenstates
In other words, X is an internal Zm symmetry of the Flo- of UF are necessarily “cat states,” i.e. superpositions of
quet Hamiltonian (Buchleitner et al., 2002; Else et al., macroscopically-distinct states (Else et al., 2016). This is
2017; Holthaus, 1995; von Keyserlingk et al., 2016). a direct generalization of the statement that when a con-
While the emergence of an internal symmetry may seem ventional symmetry is spontaneously broken in an equi-
mysterious, it is really just a re-expression of the origi- librium system, the low-energy eigenstates that trans-
nal discrete time-translation symmetry in a frame which form simply under the symmetry (e.g. eigenstates of the
is rotated by X at each step (Fig. 3a). In (Else et al., center or Cartan sub-algebra (Fulton and Harris, 2013)
2020a), this intertwining is referred to as “twisted time- of the symmetry group) are necessarily cat states.
translation”. However, the emergence of such a symme- To unpack this definition we investigate the impli-
try was pointed out well before the recent interest in Sτ B; cations of Sec. III.B.2 for the “Floquet eigenstates”
see for example the extensive literature on non-spreading UF |εF i = e−iT εF |εF i. Here the “quasienergies” εF are
wavepackets in periodically driven systems. (Buchleitner defined modulo the driving frequency εF ≡ εF + ωD . As
et al., 2002; Holthaus, 1995; Holthaus and Flatté, 1994). discussed, UF exhibits Sτ B when there is a decomposi-
tion UF = Xe−iT HF for which HF spontaneously breaks
The existence of a Floquet Hamiltonian HF and an
the symmetry [X, HF ] = 0. In the thermodynamic limit,
emergent symmetry X immediately leads to the possi-
spontaneous symmetry breaking implies that the eigen-
bility of Sτ B. In particular, suppose that the dynam-
states of HF come in degenerate pairs permuted by the
ics of HF spontaneously break the X symmetry. Intu-
symmetry, X |↑i = |↓i (we focus on X 2 = 1 for simplic-
itively, this means that the state space breaks up into
ity) 1 . However, in a finite system, the eigenstates of HF
m “superselection” sectors which are permuted by the
action of X but which are not connected by evolution
under HF . In the rotating frame, a trajectory thus
1 In conventional symmetry breaking, only eigenstates with energy
gets stuck within a sector, and mapping back to the lab
10
must simultaneously diagonalize X, so they come in pairs a particle bouncing off an oscillating mirror(Holthaus,
|±i = √12 (|↑i ± |↓i with eigenvalues ± = ± ∆ which 1995; Holthaus and Flatté, 1994). In this 0D case, the
are split by an amount ∆ ∼ e−L/ξ which is exponentially Sτ B arises because KAM stability allows the quasi-
small in the system size L if the system is to exhbit true energy to “split up” the {Q, P }-space into disconnected
Sτ B. The eigenstates of HF are thus “cat states,” since basins (Fig. 3b), rather than because of a collective
the constituent |↑ / ↓i are macroscopically distinct. The phenomena. Furthermore in 0D there is long-range
quasienergies of UF are then obtained by including their order in time, but not in space, and as such the reader
eigenvalue under X: + − may prefer not consider this case a time crystal. This
F = + ∆, and F = − ∆ + ωD /2.
The spectrum of UF thus consists of pairs of cat states analysis also illustrates why the quantum version of
with eigenvalues that differ by ωD /2 up to exponential a parametrically-driven non-linear oscillator will not
accuracy in L. feature infinitely long-lived Sτ B: while a quantum
To relate this spectral property back to Eq. (3), note Floquet Hamiltonian with a Z2 symmetry may exist, for
that an initial state |xi will generically have amplitude any finite barrier height there will be quantum tunneling
on both the |+i and |−i states of each pair. Due to between the two minima at a rate ∼ 1/τ . This leads to a
the eigenvalue pairing, the system will thus coherently unique Z2 -symmetric steady state, and the subharmonic
oscillate at frequency ωD /2, up to a dephasing time τ ∼ response will have a correlation time of τ . (Buchleitner
∆−1 which diverges exponentially in L. et al., 2002; Holthaus, 1995; Holthaus and Flatté, 1994;
Sacha, 2015) analyze several examples of such quantum
subharmonic responses and provide estimates of the
4. Application to many-body parametric resonance tunneling rate, including a particle bouncing off an
oscillating mirror, a topic we will return to in Sec. IV.E.
Let us now see how the formalism from The fly in the ointment comes when we attempt to
Secs. III.B.1, III.B.2 plays out for a single para- construct HF in the many-body setting. In particular,
metrically driven non-linear pendulum [c.f. Eq. 6]. let us consider an array of coupled pendula by adding a
Chirikov, 1979 and Zounes and Rand, 2002 showed nearest-neighbor interaction to Eq. 6,
that HF (Q, P ) indeed exists within a finite volume of X p2 ω2

phase space (calling it the “resonance Kamiltonian”), H(q, p, t) = i
+ 0 (1 + δ cos(ωD t))qi2 + qi4
and that courtesy of the KAM theorem, its existence i
2 2 4
X
is stable to small perturbations. The contours of equal −g (qi − qj )2 . (12)
“quasi-energy” HF (Q, P ) are shown in Fig. 3b. As hi,ji
aforementioned, we see that HF exhibits a Z2 -symmetry,
X : Q → Q + π, which exchanges two local minima Physically, the reader can think of this as treating
enclosed by a separatrix. The detailed construction of the pendulum as a macroscopic object composed of
K(t) is quite involved, but the approximate intuition is atoms (i.e. the Frenkel-Kontorova model (Kontorova and
as follows: One starts with the canonical transformation Frenkel, 1938)). It differs from the Faraday problem in
in Eq. (7), and removes the residual time dependence the details of the dispersion and non-linearity, but is oth-
order-by-order through a sequence of transformations erwise conceptually similar.
whose convergence is guaranteed by the KAM theorem. If the initial condition is uniform, qi , pi = q, p, the
The two minima of HF , Q, P = (± π2 , P∗ ), are fixed problem exactly reduces to Eq. (6) which we now know
points of HF ; mapping back to the lab frame, one exhibits stable period doubling. But stability for a
obtains two precisely period doubled orbits related by finite-volume of initial conditions requires one to con-
time translation, K−1 (t) : (± π2 , P∗ ) → q∗ (t ± T /2), p∗ (t). sider initial conditions of the form, qi = q∗ + δqi , in
Small deviations from the minima circulate around which the δqi are small but independent. Using the lan-
quasi-energy contours, corresponding to slow oscillations guage we introduced near Eq. 5, there is now weight
about the period doubled motion. Similar behavior in the higher-kP modes, qk , which are all-to-all coupled
4
is seen in other periodically driven problems, such as through the q
i i non-linearity, so the center-of-mass
the kicked rotor model, whose stroboscopic motion Φ mode (i.e. k = 0) cannot be examined in isolation. While
reduces to Chirikov’s standard map,(Chirikov, 1979) or one can attempt to employ the same time-dependent
canonical transformation K(t) site-by-site, for generic ini-
tial conditions the coupling g(qi −qj )2 will not be exactly
time-independent after this transformation; thus, since
densities below the symmetry-breaking temperature Tc have this the transformed Hamiltonian H(Qi , Pi , t) remains time-
structure. However, if many-body localization is able to exist in
Floquet Hamiltonians (Ponte et al., 2015; Sels and Polkovnikov,
dependent we cannot appeal to the conservation of HF
2021; Šuntajs et al., 2020), it is believed that symmetry breaking to ensure stability. One can attempt to remedy this by
must manifest at all energy densities (De Roeck et al., 2016; adjusting the canonical transformation via a perturba-
Moudgalya et al., 2020; Sahay et al., 2021). tive expansion in g, but there is a rather general reason
11
to expect this expansion will fail to converge in the ther- The existence of such a long time scale is certainly one
modynamic limit: the ubiquity of ergodicity. ingredient in Faraday’s observation of period-doubling in
a system which is Hamiltonian. An additional role is
played by viscosity, which converts the high-k oscillations
C. Ergodicity, destroyer of time-crystals of the surface into heat which is eventually dissipated
into the environment. When viscosity alone is added to
As we have discussed, Hamiltonian systems are distin- the equations of motion, the system is no longer measure
guished from more general dynamical systems like cou- preserving and becomes analogous to the coupled map
pled map lattices by the existence of a phase-space vol- lattices where true Sτ B is possible. However, this might
ume element, dµ = dq ∧ dp, which is invariant under the seem to be a contradiction, because viscosity is after all
dynamics, i.e., the dynamics are “measure preserving.” a phenomenological treatment of microscopic degrees of
An invertible measure preserving system is said to be er- freedom which are themselves Hamiltonian! The key is
godic if the only finite-volume subset A which is invariant that at any finite temperature (energy density), a mi-
under the discrete-time update rule Φ (i.e. Φ(A) = A) is croscopic bath invariably generates noise in addition to
the entire phase space. An S equivalent statement is that viscosity, as required by the fluctuation-dissipation the-
the orbit of any volume m Φ(m) (A) eventually fills all orem. Understanding the fate of Sτ B in the presence of
of phase space (Cornfeld et al., 2012). Full ergodicity dissipation and noise is a particularly rich direction; we
is incompatible with the existence of a time-independent will return to the possibility of such “open” time-crystals
HF (Q, P ) because the conservation of the quasi-energy in Sec. V.
HF partitions phase space into Φ-invariant quasi-energy To summarize, in dissipative dynamical systems like
contours (though the dynamics may be ergodic within the coupled map lattices, stable, many-body Sτ B is pos-
each contour). A time-dependent many-body system sible, and time-crystals in this context have a long his-
with a bounded state space is generically expected to tory. But their realization in measure-preserving systems
be ergodic in the thermodynamic limit. 2 which describe the universe at the microscopic level is far
Or, in the unbounded case: the generic fate of a driven, more subtle. While stable Sτ B is possible in few-body
many-body Hamiltonian system is to heat up. In an classical systems such as a parametric resonator (i.e. the
ergodic system, Birkhoff’s Ergodic Theorem (Birkhoff, shaken pendulum example in Sec. III.B.4), their existence
1931; Cornfeld et al., 2012) equates a temporal average in many-body systems requires, at minimum, a generic
of the sort defined in Eq. (3) with averages over the entire mechanism for breaking ergodicity. In closed classical
phase-space. Thus Sτ B necessarily requires Φ(m) to be systems, this is not thought to be possible except when
non-ergodic: a time-crystal “remembers” its initial con- the system is fine tuned. This is where the magic of quan-
dition for an infinitely long time because it is encoded in tum mechanics comes in, through the remarkable physics
the phase of the period-m oscillations. Sτ B is thereby of many-body localization (MBL). MBL allows for sta-
deeply connected with another profound phenomena: er- ble ergodicity breaking in the thermodynamic limit, and
godicity breaking. where there is ergodicity breaking, there will be time-
When the interactions are weak, or the driving is crystals.
strong, the time-scale required to explore all of phase
space can be very large, so in practice, Sτ B may persist
out to very long times τ . This behavior is sometimes
IV. CLOSED, PERIODICALLY-DRIVEN QUANTUM
referred to as slow Floquet heating. While energy is not SYSTEMS
conserved, in systems described by an effective Floquet
Hamiltonian [Eq. (8)], the quasienergy hHeff i will change A. Introduction to quantum Floquet phases
slowly due to the residual drive V (t), eventually allowing
the system to explore its full phase space. We will return As noted above, ergodicity-breaking is necessary for
to such anomalously long time scales when we discuss the stability of a time crystal. In fact, with the insights
“prethermal” time-crystals in Sec. IV.D. of the previous section in hand, one can utilize almost any
generic, robust form of ergodicity-breaking to stabilize a
time crystal. However, this wasn’t immediately appar-
2
ent when many-body localization (Abanin et al., 2019;
The ideas emanating from Boltzmann’s “ergodic hypothesis”
postulate that time-independent many-body systems are gener- Nandkishore and Huse, 2015) was originally discovered.
ically ergodic within each conserved energy shell.(Von Plato, Many-body localization occurs in 1D and, possibly, 2D
1991) This hypothesis relates to the Floquet setting by con- systems with strong quenched disorder and short-ranged
verting the time-dependent Hamiltonian H(t) into a time-
independent H = H(t) + πt , where (t, πt ) are an additional
interactions (Choi et al., 2016; Smith et al., 2016). It rep-
canonical pair living on the cylinder t ∼ t + T . If H is ergodic resents a failure of the eigenstate thermalization hypoth-
on an energy shell, then the stroboscopic dynamics of H(t) are esis (Deutsch, 1991; Srednicki, 1994) that occurs when
fully ergodic. disorder prevents equilibration by localizing the degrees
0 2 0 2 0 2 0 2
0.08 0.08 0.08 0.08
Peak height variance (a.u.)

0.10
0.08 0.08 Numerical
Peak height
0.06 0.06 0.06 0.06 simulations
0.06
LETTER RESEARCH 0.04 12
0.04 0.04 0.04 0.04 0.06
Peak height (a.u.)

0.02
As shown in Fig. 2c, for θ = 1.034π, this lifetime increases with the a 0.00
0.02 (a)interaction time
0.02(τ171 0.02 0.02 (c)0.9
Crystalline fraction, f
+ 0.00 0.04 0.08 0.12
Yb
1) and eventually approaches
measured spin depolarization time T
the independently
ρ
µs . This demonstrates that,
1 ≈ 60 0.04
1 = 275 ns
1 = 92 ns
0.6
for sufficiently long interaction times, the observed periodic order is
0.00 0.00 0.00 28. We associate this0.00
0 limited0.1
0.05 by only coupling
0 to the environment
0.05 0.1 0 0.05 0.1 with0 0.050.3 0.1
DTC order9–12. Within the DTC phase, 2 µmthe lifetime is essentially
(b)independent of θ, indicating exceptional robustness (Fig. 2d). (d) 0 0.02
We examined whether the observed periodic order could arise from
b an accidental
0.06 XY sequence14 or from inhomogeneous dephasing
b 800
resulting from the effective

Symmetry single-particle disorder in the dressed state
unbroken
f
basis. To avoid the former, τ1 is always chosen as an integer multiple of 0.00 0.8
2π /Ωx. For the latter, although it has been shown that disorder 600
(ns)
alone 0.04is insufficient for stabilizing a DTC phase in the absence of 0.00 0.02 0.04 0.06 0.08 0.10
Perturbation,
Discrete time
1
interactions9–12, we verified this experimentally; implementing a rotary
Interaction time,
echo sequence that reduces such dephasing, we find no change in the crystalline order Perturbation,
400 0.4
lifetime of the DTC order and an enhancement in the subharmonic
response
0.02 at late times (see Methods and Extended Data Fig. 1). In
Figure 4 | Subharmonic peak height as a function of the drive
principle, fast Markovian dephasing could also lead to apparent perturbation. Main panel, the central subharmonic peak height in th
periodic order at extremely small values of θ − π by eliminating 200
Discrete
coherences along both ŷ and ẑ , leaving only x̂ polarization dynamics.
time crystal Fourier spectrum as a function0of the perturbation ε, averaged over th
In such0 a case, the decay rate of periodic order should increase 10 sites and 10 disorder instances, for four different interaction streng
quadratically with θ − π. However, 0.025 this explanation 0.05 is inconsistent with
0.075 0 (see key at1 top). Solid lines1.2 are guides to the eye. The height decreases
the observed robustness of the lifetime of DTC order for a range of 0.8 0.9 1.1
Effective interaction
θ − π values (Fig. 2d) and the independently 2J0 t2 measured time dephasing rate across theangle,
Rotation phase (π) boundary and eventually diminishes as the single pea
(see Methods). c is split into two. Error d bars, ±1 s.d. Inset, numerical simulations given
Figure 3 | Variance of the subharmonic
To experimentally determine the DTC peak amplitude
phase boundary, we focus asona signature z
FIG. 4 (a)ofSignatures of time crystalline order were observed in a one
120
dimensional experimental chain parameters.
of Ytterbium ions. In this system,
the DTC transition. a, Variances of the central peak height, computed
the long-time behaviour of the polarization time traces (50 < n ≤ 100) 60 0.6
Fourier transform
and compute the ‘crystalline fraction’, in which is defined as the “clock”0 states of a 171 Yb +
intensity, |S( )|2

over
the effective the degree
spin 10 sites and
of averaged
freedom is over encoded 10 instances two of disorder,
hyperfine for four 0.5 ion. (b) By varying the interaction
)
/T
ratio of the ν = 1/2 peak intensity to the total spectral power, 0.95
(1
1 1.05 0.4
different which
strength (x-axis), strengths
f = Sis (ν =of 1/the2) 2 /long-range
effectively ∑ ν S (ν ) parameterized interaction term
2 (see Methods). Figure by anJ0. interaction
3a shows The
f as acrossover time-scale, procedure).
and the The strength measurementsof the perturbation
y
are in agreementon the with the expe
function ofal.,
θ for two different interactionabletimes. For weak interactions
from a symmetry
π-pulse (y-axis), (Zhang etunbroken 2017) state were to a DTC
(τ1 = 92 ns), f has a maximum at θ = π and rapidly decreases as θ
is
toobserved
observe as a
the peak in
system’s 120 behavior DTC to
change time fromcrystal ‘melting’
ax discrete boundary,
time crystalwhich
phasedisplays
to approxim
0.6
the measured
a time-translation symmetryvariance
deviates of the subharmonic
unbroken
by approximately 0.02π phase.
. However, system
Figure
for stronger response.
interactionsDashed
adapted from Fig. 60 3(b) of linear(Zhang dependence
0.5
et al., 2017).on the perturbation
(c) (Choi et al., strength
2017a) in the limit of s
0
)
lines, numerical results,
ns), wescaledobserve avertically to fit the ofexperimental as a data
/T
(τ1 = 275 arobust DTC phase, which manifests 0.4(∼10
utilized a diamond sample containing high density Nitrogen-Vacancy color interactions
1 centers 15. ppm) (Choi et al., 2020). In this
(1
0.95 1.05
large crystalline fraction over a wide range 0.86π < θ < 1.13π. We
system, the(see Methods
spin consists for of
associate detailed
atwo manalysis
s -sublevels
phenomenological procedures
phase and boundary NV and possible
centers
with f = 10% and sources ofwith one (π)another
interact Figurevia magnetic the
4 illustrates dipole-dipole
amplitude interactions.
of the subharmonic peak
decoherence).
The interplay between Experimental
observe thethat the boundary
long-ranged error bars,with
enlarges
dipolar s.e.m.; a.u., arbitrary
τ1,interaction
eventually saturatingand atunits.
Figure 3 | Phase diagram and transition. a, b, Crystalline fraction f (a)
the
and itsthree
associateddimensional
phasefunction
diagram (b) asof nature of
θ andthe sample isapplied
conjectured to strengths. I
b, Crossover τ1 ≈ 400 ns (Fig. 3b). The phase boundary can also be visualized as the
determined by a fit(Ho to the variance peak location
ε, for
a function ofthe four different
τ1 obtained interaction
lead to critically slowvanishing
thermalization and et al., 2017; Kucsko et (filled
from a2018),
al., Fourier transform
during at latewhich
times (50 < n ≤ 100). Thetime
discrete red diamonds
crystalline order can be
circles). Dashed
of the ν = 1/2 peak
line, numerically
incommensurate peaks (Fig. 3c).
the simultaneous
determined
emergence
phase boundary
of two mark the phenomenological
with presence of spin–spin
phase boundary, identified as a interactions,
10% the peak height falls off slowly
observed. (d) Experimentally measured phase boundary (red dashed line
crystalline added
fraction; as error
horizontal a guide)
bars denote ofthethe critical
statistical error (s.d.)discrete time crystal
experimental long-range
The rigidity of the coupling
ν = 1/2 peak parameters
can be qualitatively10
. Grey shaded
understood by region
from a super-Gaussian fit. increasing
In a, vertical error This
ε.bars of data slope is steeper as we turn down the intera
points (circles)
as a function of the effective
constructing interaction
effective eigenstatesstrength of 2T Floquet and the π-pulse
cycles, including spin– are imperfection.
limited by the noisestrength,
Figure adapted from Fig. 3(b) of (Choi et al.,
2017a). indicates 90%spin confidence
interaction. We level of the the
approximate DTC unitaryto symmetry
time evolutionunbroken over a indicate phase floor (see Methods)in agreement
and horizontal error with
the pulse uncertainty of 1%. Grey lines denote the fit to extract the
bars the trend of numerical simulations (
boundary. Interaction
single period asstrengths UT = Rθy e−iHeffare τ1 and normalized to be unitless,
solve for a self-consistent evolution referencing
phase boundary (see Methods). inset).InThis is characteristic
b, the colours of the higher susceptibility to perturba
of the data points (circles)
end, we consider represent the extracted crystalline fraction at the associated parameter set.
using product
to the fixed disorder states as a variational
accumulated phase ansatz.
π (ref. To this
the situation in which a typical spin returns to its initial state after 2T,
10). The dashed line corresponds This to asubharmonic
disorder-averaged theoreticalpeak height
prediction observable is expected to scale
10
ψ(0)〉 ∝ ψ(2T )〉 = e−iθS e iφiS e−iθS e−iφiS ψ(0)〉, and self-consistently asymmetric distributionsimilar way as the mutual information , and can serve as an o
y x y x for the phase boundary. Asymmetry in the boundary arises from an
of rotation angles (see Methods). c, Evolution of
of freedom of a system. Locally-injected energy can- 2013) considered
determine the interaction-induced rotation angle φi ≡ ∑ j Jij /r ij3 S xj τ1 the Fourier spectra as a parameter.
periodically-driven
function of θ for twoThis
versions of a Ma-
differentconnection
interaction times: also provides insight into the Flo
not spreadtrack the perturbation
throughout ≈Ji τ1 ψthe (0) S x ψ (0)ε., This
system where(Abaninresults
ψ(0)〉 is in the coherent
et al.,spin2019;
initial beats
state and andjorana
τ1a=splitting
385 ns (top) and τ1 = 92in
chain
iH3t 3 ofmany-body
single spin−trajectory
ns (bottom).
which d, Blochthe
quantum
sphere indicating a
time-dependent
under thedynamics,
Hamilto-
in particular the correlations or e
in the Fourier
Abanin and Papić, Floquet Ji = ∑
2017; spectrum
J / r 3
(see
j ij Nandkishore by 2ε
Methods). (Fig.
We expect
and 2a). φ When
to
Huse,
i change we
sign
2015). add
after disorder
each
nian e toggles
dipolar Hamiltonian
the 2T-periodic evolution
(red)between
and global rotation topological
(blue).
long-range
superconducting and
glement underlying the DTC phase. Indeed, the eigenstates of the e
ij
MBL wastooriginally
the Floquet period, the single spins precess ofcanathighly-
cycle, because the average polarization 〈ψ (0) S xdifferent Larmor rates
ψ (0) 〉 should
beformulated
flipped. Intuitively, the asself-consistent
a property solution be visualized as trivialin DTC orderinsulating Floquet
at ν = 1/3. Here, westates.
unitary
utilize all are
three These
expected
spin states of free
to
the fermion
resemble GHZ mod-
(Greenberger–Ho
(Fig. 2e) and dephase
a closed path on the with
Blochrespect
sphere (Fig. to 3d),each
where each other of the(Fig.
four arcs2b). Only on centre. We begin with all of the spins polarized in
nitrogen–vacancy 8
disordered time-independent
adding Isingsuchcorresponds to oneHamiltonians
interactions portion
e − iH of2tthe
2 2T-periodic Ĥevolution.
in which
, and hence many-body correlations, When θ = π, els the m s
avoid
= 0 〉 state and Zeilinger)
Floquet
evolve under theor
heating
barespin-‘Schrödinger
dipolar by virtue
Hamiltonian for a Cat’
ofstates
their . The initial product st
inte-
a solution always exists. More surprisingly, even for θ ≠ π a closed duration τ1 (see Methods). Next, we apply two resonant microwave
local observables fail
the spin synchronizationto relax to
path can still be found
their
is restored thermal
for sufficiently (Fig. 2c, strong
average
f). grability, which the experiment does notcansurvive
interactions, pulses, each of duration τ2, first on the transition ms = 0〉 → ms = −1〉
be written as a superposition
generic perturba- of two cat s
with respectThe to ekey−β Ĥ
result and,
Ji τ1 > 2 θi − πinstead,
is ;that in such cases,the
withthe long-time
alldeviation
of these in θ awaybehav-from π is com- tions
elements, theand then on
temporal
pensated by the dipolar interactions (Fig. 3d). We obtain a theoretical sequence of operations defines a single Floquet
quet
involving ↓↓…↓〉
the transition s = 0〉 x→ =ms 1= +1〉φ. In
minteractions.
2 (
For such
combination,
cycle with period
)
+〉 + φ−〉 , where φ±〉 = 2 ↓↓…↓〉x ± ↑↑…
this drives, the Flo-
1
(
ior dependsresponse
on the isphase
locked
details
boundary toof twice
by the the
numerically initial Floquet
averaging state, period, a clear
the self-consistent even
solution in the
T = τ1face+ 2τoperator
ofbefore,
2. As hThese takes
we measuretwothe on
states a particularly
i evolve
polarization at which
hP(nT), different is simple
rates form, to their re
icorresponding
over both disordered spin positions and local fields. The resultant defined as the population Pdifference between the ms P = 0〉 and B B
example of perturbations
ergodicity phase toboundary
the drive
breaking. in H1. This
It
is in reasonable
was canwith
subsequently
agreement betheseen clearly U
re-
experimental as
msF= the−1= 〉split exp
states (Fig. tive
ic When
4a). quasi-energies,
j iγ
each
A
j ofγthe
B
expgiving
j applied iJ rise
microwaves iγ
jcor-
toj the
γj+1subharmonic
(Prosen,periodic oscilla
alized thatFourier
MBL peaks from
canobservations
also occur Fig.
for short2b merge
into moderate
1D into a single
interaction
periodically-driven times τ1, but peak in Fig.
overesti-
1998; 2c. This
responds to an ideal πof
Thakurathi physical
pulse, this sequence observables. Such oscillations are expected to pers
realizes a cyclic transition
al., 2013).
etis explicitly For c near π/2, there
mates the boundary at large τ1 (dashed 10 line, Fig. 3b; see Methods). with Z 3 symmetry (Fig. 4b), which broken by any change
represents
systems with strongthe ‘rigidity’
disorder
Finally, Fig. 4 for of the DTC
appropriate
demonstrates that the , which driving
discrete persists pa-underare
time-translation inmoderate
the pulse duration.increasingly
The Fourier spectra long times
of P(nT) as the
for various pulse system size increases7,8,10.
“Floquet Majorana fermions” at quasienergy π. In
perturbation strengths. However, for large
symmetry can be
rameters (Abanin et al., 2016; Lazarides et al., 2015; further broken down to Z 3 ε, the DTC phase disappears,
(refs 10–12, 29), resulting durations and for two In summary,
different values of τ 1 are we present the experimental observation of dis
shown in Fig. 4c. With
as evinced other words, thereH 2are locally-indistinguishable states of

Ponte et al., 2015).byIn thesuch
decayFloquet-MBL
of the subharmonic temporal
systems,
© 2017 en-
Macmillan Publishers
correlations
Limited, parteven/odd
and time 9 M A R C translational
of Springer Nature. Allfermion
0 1 7 | V O L 5 4 3 | symmetry
N A T U R E | 2 2 3 breaking into a DTC. We mea
rights reserved. parity, associated with edge modes. In
the suppression
ergy absorbed during one of the central
part peakdrive
of the heights, as shown
cycle mustin Fig. 2d.undriven In the persistent oscillations and synchronizations of interacting spi
thermodynamic limit, these perturbations induce a phasean transition asystem, chain and Majorana
show thatedge modes time
the discrete would be de-
crystal is rigid, or robu
be returned before its completion (Bordia et al., 2017; 7–10
from a DTC to a symmetry unbroken MBL phase , which is rounded perturbations in the drive. Our Floquet-MBLperi-
generate (i.e. they would be zero modes), but the system with long-r
D’Alessio into
anda crossover
Polkovnikov, 2013;
in finite Ponte et al., 2015).
size systems. odic driving interactions splits theirprovides Floquetaneigenvaluesideal testbed by for out-of-equilibrium
π. quan
Consequently, the energy spreading
The phase boundary is defined andbythermalization
the competition between the dynamics and the study of novel phases of matter that exist only
This property of the spectrum does not imply subhar-
necessary drive
for drive-induced
perturbation ε and Floquet heating
strength cannot oc-J0. We probe this Floquet setting7–10,25–28. Such phases can also exhibit topological
of the interactions monic response of the system for generic initial conditions
boundary by measuring the variance of the subharmonic spectral peak and can be used for various quantum information tasks, such as im
cur (Sec. III.C).
(and, moreover, it is not stable to interactions). However,
height, computed over the 10 sites and averaged over 10 instances of menting a robust quantum memory26,29,30.
In the meantime, interest in Floquet systems had been
disorder. Figure 3a shows the variances as a function of theaperturba- Jordan-Wigner transformation of the driven Majorana
re-energized Online Content Methods, along that
with any additional subhar-
Extended Data display item
by seminal work demonstrating that pe-
tion ε, for four different interaction strengths. As we increase ε, yields chain the Source a driven Ising model exhibits
Data, are available in the online version of the paper; references uni
riodic driving can induce topological properties in an monic oscillations
variance growth distinctively captures the onset of the transition, with these sections appear only in the online paper. of the Ising order parameter; moreover,
otherwise increased
nominally fluctuations signalling
trivial system the crossing
(Inoue and Tanaka, of the phaseby boundary.
adding strong disorder leading to MBL, the system
Received 26 September 2016; accepted 10 January 2017.
When
2010; Jiang et the
al.,perturbations are too large, the
2011; von Keyserlingk andcrystalSondhi, ‘melts’. The could highest
be stabilized against interactions (Khemani et al.,
variances correspond
2016a,b; Lindner et al., 2011;to thePotirniche
crossover points.
et al., Figure 2017; 3b shows the fitted
2016). Initially it was not believed that the subharmonic
1. Chaikin, P. & Lubensky, T. Principles of Condensed Matter Physics Vol. 1
Potter etmaxima of the variance
al., 2016; Roy and curve, on top of2016,
Harper, numerically 2017;computed response phaseof the (Cambridge resultingUniv. “π Press,
spin 1995).glass” could survive in
Thakurathiboundaries
et al., with
2013;experimental
Wang et parameters
al., 2013).(seeInMethods par- forthe theabsence
fitting of 2. the Wilczek,
IsingF. Quantum
symmetry, time crystals.
implying Phys.that, in 109,
Rev. Lett. anal-160401 (2012).
ticular, (Jiang et al., 2011) and (Thakurathi et al., ogy to the discussions of the XY magnet in Sec. II.A, a
9 M A RC H 2 0 1 7 | VO L 5 4 3 | NAT U R E
© 2017 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.
13
time crystal stable against generic perturbations did not Floquet dynamics will be captured by this example. In
exist (Khemani et al., 2016). particular, consider Floquet time evolution (with a pe-
This sentiment was overturned in (Else et al., 2016), riod, T0 = t1 + t2 ) governed by alternating between two
which showed that discrete time crystals are a stable time-independent Hamiltonians:
phase of matter in Floquet-MBL systems with only time (
translation symmetry. In particular, a discrete time crys- H1 , for time t1
H(t) = (13)
tal in a closed system must satisfy Eq. 3 in the absence H2 , for time t2 .
of a bath and in a manner that is stable against arbitrary
weak perturbations respecting time translation symme- Let us take H1 , H2 as given by:
try. Other symmetries may or may not be present, but X X
Eq. 3 should hold even when those symmetries are vio- H1 = − Jij σiz σjz − (hzi σiz + hyi σiy + hxi σix )
hi,ji i
lated. These developments led rapidly to the first phase X
diagram for a time crystal and to proposals describing H2 = g σix . (14)
how to create and measure discrete time crystals in exper- i
iments (Yao et al., 2017), culminating in the first obser-
vations of time-crystalline behavior (Fig. 4) (Choi et al., with ~σ being Pauli spin operators and Jij , hxi , hyi , hzi
2017a; Zhang et al., 2017). sufficiently disordered to ensure that the spin chain is
many-body localized. The simplest way to see the (triv-
Thus far, the above discussions have focused on many-
ial) emergence of a period-doubled (i.e. ν = 1/2 sub-
body localization as a possible strategy for stabilizing
harmonic) response is to consider the decoupled limit
discrete time crystals. One of the caveats of using MBL
(Jij = 0) with only a longitudinal field, hz , along ẑ. For
is that the approach depends on the robustness of many-
any individual spin initially along the ẑ-axis, the spin will
body localization itself. While MBL has purportedly
Larmor precess around the x̂-field during the second por-
been proven to exist in certain undriven, one-dimensional
tion, H2 , of the Floquet evolution. When timed appro-
spin chain models with random local interactions (Imbrie,
priately, t2 = π/g, this evolution implements a so-called
2016), its stability in d > 1 and in long-range interacting
π-pulse, which flips the spin, and causes a period-doubled
systems remains an open question (De Roeck and Hu-
response. However, much like our discussion of the map
veneers, 2017; Khemani et al., 2021; Luitz et al., 2017;
Φ(x) = −x in Section II, this period doubling is not
Thiery et al., 2018; Yao et al., 2014, 2021); to date, there
rigid to perturbations and not indicative of a many-body
is no mathematical proof for the stability of many-body
phase. In particular, any imperfections in the timing of
localized Sτ B in a Floquet system in any dimension. An
the π-pulse will immediately lead to the breakdown of the
alternate strategy for stabilizing DTC order, termed Flo-
period-doubled response (Yao et al., 2017). However, by
quet prethermalization, can occur in any dimension and
turning on the Ising interactions between the spins, in
in the presence of long-range interactions, so long as the
conjunction with the disorder which leads to MBL, the
driving frequency, ωD , is sufficiently high. To this end,
system becomes robust to small imperfections and the
in the remainder of this section focused on closed sys-
subharmonic response is rigid to arbitrary, weak pertur-
tems (Fig. 1), we will focus in subsection IV.B on MBL
bations of both the initial state and the Hamiltonian (so
discrete time crystals in 1D and in subsection IV.D on
long as these perturbations respect the period of the Flo-
prethermal discrete time crystals in arbitrary dimension
quet evolution). We note that this last parenthetical dis-
and with power-law interactions.
tinguishes time crystalline order in closed systems from
time crystalline order in open systems, where perturba-
tions that explicitly break the discrete time translation
B. Many-body localized discrete time crystal
symmetry of the dynamics can still be allowed (see Sec-
tion V.E).
Our focus in this subsection will be on highlighting
the key features of time crystalline order in Floquet MBL
systems, by describing recent observations from three dis- C. Experimental signatures of disordered time crystals
tinct experimental platforms: trapped atomic ions (Yao
et al., 2017; Zhang et al., 2017), spins in solid state 1. Trapped ion spin chains
materials (Abobeih et al., 2019; Randall et al., 2021),
and superconducting qubits (Arute et al., 2019; Frey and Zhang et al., 2017 observed and characterized time
Rachel, 2022; Mi et al., 2021). crystalline behavior in a spin chain (Fig. 4a) composed
Before describing the experiments, we begin by intro- of trapped atomic ions. In the experiment, each ion
ducing a standard model for a period-doubled discrete acts as an effective spin, and interactions are controlled
time crystal in a spin-1/2 chain, and discuss several lim- through external, optical spin-dependent forces (Monroe
its of this model. While the experiments will not im- et al., 2021). The system is periodically driven by suc-
plement this specific Floquet unitary, the spirit of their cessively applying three time-independent Hamiltonians,
14
corresponding to a global drive, interactions, and disor- ing a variety of product initial states, (Randall et al.,
der (Smith et al., 2016). The disorder is programmed by 2021) demonstrates that dipolar interactions between the
individually addressing each spin with a tunable laser nuclear spins lead to robust period doubling for nearly
beam. The ion chain was composed of a relatively ∼ 103 Floquet cycles (note that each Floquet cycle lasts
small number of effective spins (between L = 10 − 14) 10 ms), independent of the initial state [Fig. 5(c)]. More-
and the time crystalline response exhibited some sen- over, they show that local thermalization occurs within
sitivity to the initial conditions (Khemani et al., 2019; ∼ 10 Floquet cycles, demonstrating that the observed
Zhang et al., 2017), suggesting that long-range interac- time crystalline order is not a result of slow thermaliza-
tions could be leading to a time crystalline response stabi- tion.
lized by prethermalization (Sec. IV.D) rather than local-
ization. The experiment was able to observe a cross-over
between the DTC regime and the symmetry unbroken 3. Superconducting transmon qubits
phase (Fig. 4b). To do so, Zhang et al., 2017 measured
the Fourier spectrum of each individual spin and stud- Signatures of time crystalline order have also re-
ied the variance of the amplitude of the ν = 1/2 subhar- cently been observed in three experiments utilizing su-
monic response as a function of the π-pulse imperfection, perconducting transmon qubits (Frey and Rachel, 2022;
ε. By increasing the strength of the interactions between Mi et al., 2021; Xu et al., 2021). Working with
the ion spins, they demonstrated that the location of the Google’s ‘Sycamore’ processor, (Mi et al., 2021) employ a
variance peak shifted toward larger values of ε, consis- quantum-circuit-based approach to demonstrating MBL
tent with the expectation that many-body interactions time crystalline order (Ippoliti et al., 2020). In par-
are essential for stabilizing time crystalline order. ticular, in order to work with a one dimensional sys-
tem, (Mi et al., 2021) isolate a nearest-neighbor cou-
pled chain of L = 20 qubits from the two-dimensional
2. Spins in condensed matter
array (Fig. 5d); an analogous strategy is taken by (Frey
and Rachel, 2022), who isolate an L = 57 qubit chain
Signatures of time crystalline order in solid-state sys- on IBM’s quantum processors: ibmq manhattan and
tems have been observed in both disordered Nitrogen- ibmq brooklyn. Both (Mi et al., 2021) and (Frey and
Vacancy (NV) ensembles (Choi et al., 2017a; Ho et al., Rachel, 2022) utilize one and two-qubit gates in order
2017) and NMR systems (Luitz et al., 2020; Pal et al., to implement a digital Floquet sequence. For the exper-
2018; Rovny et al., 2018a,b). In the context of NV cen- iment performed on the Sycamore processor, the specific
ters (Fig. 4c,d), despite the presence of strong disorder, Floquet sequence was given by (Fig. 5e),
it is known that dipolar interactions in three-dimensions UF = e− 2
i
P
i hi σiz − 4i
e
P
i Ji σiz σi+1
z
e − 2i πg
P
i σix
, (15)
cannot lead to localization; however, the resulting time-
crystalline order can exhibit an anomalously long life- consisting of random longitudinal fields, disordered
time owing to critically slow thermalization (Ho et al., nearest-neighbor Ising interactions, and approximate π-
2017). One of the puzzles arising from NMR experiments pulses. Deviations from the ideal π-pulse limit (i.e. g = 1
on phosphorus nuclear spins in ammonium dihydrogen in Eq. 15) are used to control the transition between
phosphate was the observation of period doubling de- the discrete time crystal phase (i.e. g = 0.97 in Fig. 5f)
spite the lack of disorder and a high temperature initial and the trivial thermal phase (i.e. g = 0.6 in Fig. 5f).
state; these observations were ultimately understood as To demonstrate that the observed subharmonic response
consequences of an approximate long-lived U (1) conser- is not affected by the choice of initial states, both (Mi
vation law (Luitz et al., 2020), which leads to effective et al., 2021) and (Frey and Rachel, 2022) probe the Flo-
prethermal time crystalline order (see Sec. IV.D). quet dynamics starting from random initial bit-strings.
Recent work in the solid-state has focused on pushing In addition, (Mi et al., 2021) also experimentally imple-
toward a realization of time crystalline order in a regime ment a finite-size-scaling analysis by varying the length of
compatible with many-body localization (Randall et al., their one dimensional chain between contiguous subsets
2021). As depicted in Fig. 5(a), Randall et al., 2021 uti- of 8, 12, and 16 transmon qubits; this reveals a tran-
lize a platform consisting of a precisely characterized ar- sition between the MBL discrete time crystal and the
ray of 27 nuclear spins surrounding a single NV center in thermal phase at a critical value of the π-pulse imperfec-
diamond (Abobeih et al., 2019). Owing to differences in tions, 0.83 & gc . 0.88. Finally, in order to characterize
the hyperfine interaction strengths, each nuclear spin is the intrinsic gate errors and decoherence of their sys-
individually addressable (for both initialization and read tem (with T1 , T2 ∼ 100µs (Arute et al., 2019; Kjaergaard
out [Fig. 5(b)]). In order to work with an effective one- et al., 2020)), (Mi et al., 2021) implement a benchmark-
dimensional geometry (thus avoiding issues regarding the ing “echo” Floquet sequence, which reverses the digital
stability of MBL in d > 1), (Randall et al., 2021) select a time evolution after a set number of Floquet cycles. Di-
specific 9-spin subset of the 27 nuclear spins. By prepar- viding by this benchmarking sequence leads to signatures
15
PRE-THERMAL PHASES OF MATTER PROTECTED BY … PHYS. REV. X 7, 011026
PRE-THERMAL PHASES(2017)
OF MATTER PROTECTED BY
(a) (b) (c)
symmetry generated by X. This Ising symmetry corre- (since ρ~ commutes with D). symmetry Therefore, generatedat odd by times, X. Thisthe Ising symmetry corre-
sponds to an approximate emergent symmetry UXU † of U f order parameter sponds to an approximate emergent symmetry UXU † of U f
(the term emergent is used for the reason stated above, and (the term emergent is used for the reason stated above, and
quasienergy
approximate because it is an exact symmetry of U ~ f , not Uf , approximate
z because it is an ð20Þ exact symmetry of U ~ f , not Uf ,
M2nþ1 ¼ hσ i iX ρ~ X ¼ −M 2n :
and therefore is approximately conserved for times t ≪ t! ). and therefore is approximately conserved for times t ≪ t! ).
Suppose that D spontaneously breaks the symmetry X Suppose that D spontaneously breaks the symmetry X
below some finite critical temperature τc . For example, Thus, the state of the system at odd times is different from
below some finite critical temperature τc . For example,
workingPin two dimensions or higher, we could have the state at even times, and time translation by T is
workingPin two dimensions or higher, we 2could have
D ¼ −J hi;ji σ zi σ zj plus additional smaller terms of strength spontaneously broken to time translation by 2T.
D ¼ −J hi;ji σ zi σ zj plus additional smaller terms 2 of strength
The above analysis took placeFloquet in thecycle, frame N rotated
(d) a which break integrability.
(e) a RandomWe are
Bit-strings interested in
Random the andregime
h c
( ) (f) Disorder which
and break
Initial-State integrability.
Averaged We
Autocorrelators are interested in the regime
es that can be executed simultaneously
i i
by U. However, we canandalso consider the expectation
array. We calibrated and benchmarked where Figurethe heating
1: atime
Programmable t! ≫
Random Bit-strings
t0pre−thermal
spin-based X , where
g tpre−thermal
quantum
Random and h i i is ( ) c 1.0
Z (h )simulator. (A) We program an
Disorder
effective
where 1D
Initial-State
the Averaged
heating Autocorrelators
time t ≫ t pre−thermal , where tpre−thermal is
Q : 0 1
values of operators in theof original frame, for example,
!
12)
nent and system level using a powerful 1
1
A ( 𝑈 ) A ( 𝑈 )
the chain
thermalization
of 9 spinsQtime :in0 an of10interacting
D.01 cluster of 27 Z13 C nuclear spins
1.0 Figure
(orange) 4: AObservation
close to aof†the
singlethe DTCNV response for generic initial states.(A) (A)Individual
Individualspin spinex-ex-
Figurez 4: Observation the DTC
z thermalization response for
time generic
of D. initial Astates.
g X (h )
ZZ (
11
marking . Finally, we used component-
12)
1
1
(A𝑈 )
0 (𝑈ECHO)z
A (𝑈 )
0 (𝑈ECHO)
dict the performance of the whole sys- Q: 1 X
g
Z (h )
hψðtÞjσ jψðtÞi
0.5 pectation ¼
values h ψðtÞjU
~ h) ijasi as σ Uj
aifunctionψðtÞi.
~ of The after
Nafter rotation
initializing U the is spins in the Néel state |"#"#"#"#"i
23)
2
ZZ (
tum information behaves as expected Now, consider
center. Connections theQ: 1 time
2
evolution
0indicate
1 nuclear-nuclear
g XjψðtÞi, starting Z from
couplings
(h ) a pectation
i
|Jjk0.5| >and1.5 values
Hz, and
A 0 ( 𝑈 h
ECHO
z
a
jblue (red) function
Now, linesof N
consider initializing
the time the spins
evolution A 0 ( 𝑈 in the
) Néel
jψðtÞi,correlation
ECHO state |"#"#"#"#"i
starting from a
ZZ (
closeand to applying
the identity in ✓the regime where 5the heating timetwo-point
ZZ ( 23)
applying for and ms.(B) (B)Averaged
Averaged two-pointcorrelation function ,
2
2
UFUFfor ✓==0.95⇡ 0.95⇡and ⌧ ⌧==5 ms. function
(0) ( )
1 X
g Z (h )
given short-range
represent negative correlated
Q: 0
Q : 0(positive)
0
statenearest-neighbor
jψð0Þi. We alsoZcouplings define within 0.0 the chain z (24). Magneticgiven short-range
field: z correlated state tojψð0Þi. We also define
34)
3
3
corresponding to the data in (A). The †DTC response persists similar high as for the po-
(0) ( )
X (h )
is large, so σ has large overlap with U σ U and therefore
g
0 1 N
ZZ (
corresponding to the data in (A). The DTC response persists to similar high as for
timespo-
the
ZZ ( 34)
3
the B rotated states jψðtÞi

3
~Q : 1 ¼0 UjψðtÞi. At stroboscopic Z (htimes 𝑈
i ECHO i jψðtÞi N
task
z ⇠ 403 G. (B) Experimental sequence:
X
The spins are will
initialized
0.0
by applying the
the rotated
PulsePol states UjψðtÞi. Atshowing
stroboscopic
g )
0 ~ of the ¼ data
45)
larized state (Fig.3B). 3B). (C) Fourier transform inin(B), (B),showing thetheperiod-doubled
period-doubled
4
4
macy, we compare our quantum proces-
Q: 1 0 0
g X
−iDT n Z (h ) display
⇥ #state
larized fractional
𝑈ECHO (Fig. frequency
(C) Fourier oscillations.
⇤transform [Specifically,
of the data
ZZ (
t ¼ sequence
nT, we find (29),that Q : 1jψðnTÞi jψð0Þi. Since −iDT n
ZZ ( 45)
4
ical computers in the task of sampling
followed ~ 1 by
4
0 ¼ ðXe
rotations X
g
ofÞ the~ form Z (h
R(#,
)
') = -0.5
exp response.
response. ifrom 𝑈 (D) (Average
(sin(')
2𝑈 (D)
x
𝑈 † )t + cos(')
Average t correlation
¼ nT, )we
correlation
y
for.for find that
even(upper
even (upper jψðnTÞi
~ curve)and
curve) ¼odd
and ðXe
odd(lower(lower jψð0Þi.
Þ curve)
~ NNfor
curve) Since
for9 9ran-
ran-
it -0.5
follows the construction of 2U that U¼ OðλTÞ,
5
quantum circuit 11,13,14
. Random circuits
−iDT Þ2 ¼ e−2iDT
5
(𝑈 †initial 1þ
evolution ,under
Q : we 1 see that Xat even multiplesUZF(h )= of[Uint (⌧and −iDT e−2iDTtime
1 0 g Z (h ) )t N
marking because they do not possess
r limited guarantees of computational
ðXeAfter 5
the Floquet X
g
sequence
5
) -1.0 Udomly
· λT x (✓)
domly ≪ 1· chosen
gU
chosen
is =int
0.60,
the (⌧ )]
initial
Qubit
regime
states,
,11states,
theðXe spins
where plusplusthe
Þ the
are
the ¼ polarized
polarized
heating , state
weisand
state seethethe
and
large.] that
gNéel0.97,at
= Néel state
Qubit even
state multiples
(indicated
(indicated
11 in in(E))
(E))of with
with
1
cuits to entangle a set of quantum bits
on of single-qubit and two-qubit logi- Figure 1: Programmable sequentially
the period, t ¼read Q : 0
2nT, out the
0 through
20 0
time evolution
the X NV electronic of jψðtÞi ~Z (hspin is
20
using ✓= =0.95⇡
electron-nuclear
-1.0 ✓ 0
g = 0.60,and
0.95⇡ 10andand
Qubit⌧11= 5 ms. Each data point is the average
⌧ =20 the
nuclear-nuclear
5 ms. 0 Each period, data
20 point t ¼ is 2nT,
40the averagethe60 time overeven/odd
g = 0.97,
over 80even/odd
evolution
Qubit 11 integers
of jψðtÞi
integers
100 ~ininthetherange
isrange
spin-based quantum Q : simulator. We recall that t theThree
20 conditions for fractional tfrequency
g )
0 1 (A) We program an effective 1D
20
0 N to N of
0 the states 20 are measured up 60to N = 80 theothers
otherstotoNN The
20
antum circuit’s output produces a set
chain of 9 spins in an
Qubitdescribed
interacting by
cluster the
of
Adjustable coupler
27 time-independent
two-qubit gates (see text). Colored 𝑈boxes
13
C nuclear spins Hamiltonian
(orange)
𝑈F ( t Cycles)
close to a D.
single Thus,
with ‘I’ denote oscillations
NV N to
re-initialization
10 + 10.
Floquet
N + 10. Three
into
cycle, the
N of described
the states
given state. areby the
measured
40
time-independent
Floquet up to
cycle,N =N 800,800,the
Hamiltonian100
D. = =Thus,
300.The
300.
0101, 1011100, …}. Owing to quantum
b F ( t Cycles)
d dashed in t the
black prethermalized
line isa fit
a fitofof| ||, |,averagedregime
averaged aret(gthe
over that (a) states
three D measured to N = 800, using a
stribution of the bitstrings resembles center. Connections we
indicate expect that,
nuclear-nuclear
(C) Coupling matrix after the
couplings time t
Thermal
|J | > 1.5 Hz, , the
and system
blue MBL-DTC
(red) appears
lines dashed black
Thermal (g line
= is
0.60) wethe expect that, after
over
MBL-DTC theXtime
the =three
0.97) tpre−thermal
states measured , the system appears a
to N = 800, using
b 1.0 for the the 9-spin chain.MBL-DTC (D) Average coupling magnitude
Thermal (g = as abreakfunction fof site
b jk pre−thermal
oduced by light interference in laser
gs are much more likely to occur than represent negative (positive) nearest-neighbor couplings within
Thermal
g = 0.60,chain (24). Magnetic field:g = 0.97,
dmust 1.0 spontaneously
phenomenological 0.60) function Ising
(N ) =Ae symmetry
MBL-DTC
AeN/N N/N(g 1/e=, 0.97)
givingupAto A= =a0.76(1)
0.76(1)and andNN1/e==472(17). 472(17).(E) (E)
to be in a thermal state of D at temperature τ. Thus, phenomenological function to fbe(N )=
in ↵a thermal , giving
1/e
state of D at N temperature τ.{",Thus,
his probability distribution becomes
FIG. Bz ⇠5 403 (a)G. (B) Schematic
Experimental distanceimage
sequence:across The the
depictingspinschain.
1.0
the = Orange
gQubit
0.60,
are initialized 9 by
11 spin line:
applying least-squares
subset the PulsePolof gQubit = 0.97,fit
Qubit 11 topre-characterized
⇤the1127 finite
a power-law
1.0
0.5 function
critical temperature
Calculated energydensity density
nuclear τEc ,Efor
J0 /|j and
fork|all
spins , possible
(b) the energy
surrounding statesofof density
theform
theformN single L
L j |mjNV
1/e
i, mj2 2{", (black
#}(black
ψð2nTÞih ψð2nTÞj ≈ ρ , where ρ ⇥is
11 a# thermal density matrix Calculated energy all possible states
≈ the |m j i, m #}
A/A0
e number of qubits (width) and number
j ~ bygiving ~ ~
0.5 Hz and
Qubit~ (E) Measured expectation
with
0.5
0.0
values
respect tohinitial z
after j ψð2nTÞih
~
initializing ψð2nTÞj
~ ρ
~ , where ρ~ is a thermal density
j matrix
jD of toUjψð0Þi NVinmust correspond to the a corresponding
sequence(b) (29),Eachfollowednuclear rotationsspinJof0 the
= can form
6.7(1) R(#,
be ') = exp↵ = 2.5(1).
i 2 (sin(') x
+ cos(') y
)and . lines). The istates states measured in (D) are indicated by j
colors.
A/A0
center. individually initialized read-out via
lines). Thecoupling
initial the
measured center.
(D) are The Floquet
indicated by sequence
τ,the
andcorresponding colors.
0.0
processor is working properly using a
After evolution underfor the D at some
Floquet sequence temperature
UF 0.5 = [Uint (⌧ τ,) and
· Ux (✓) the· Uapproximate
int (⌧ )] , the spins
N equality
are -0.5
for D1,at we some temperature the approximate equality
nchmarking 11,12,14
, which compares how
consists of U , which the state
includes|"""""""""i. a disordered The data is corrected
long-ranged for measurement
Ising interaction-0.5 errors
temperature
-1.0 (25).
and τ < a τ .
disordered
c In Fig. longitudinal show the expected
field followed by an
sequentially readint meansthethat
out through the expectation
NV electronic spin values of local
0.0using electron-nuclear andobservables
nuclear-nuclear are meansτ that the expectation values of local A/ A
A/A0 observables are
( )( )
experimentally with its corresponding 0

imulation on a classical computer. For
the two-qubit gates (see text). Colored boxes with ‘I’ denote re-initialization behavior
-1.020 at low temperatures and contrast it with the 1.0
asured bitstrings {x } and computeapproximate
i π-pulse, approximately Rx (θ). theThe 0.0
same. disorder Note
Instancethat
1 forinto
is naturally τ< the given
thestate.
τc, inherited Ising from2015the random positioning approximately of thethe nuclear
same.
15
spinsthat
Note within for 1.0 τ the
<
0.5 τ c , the Ising
ng fidelity (see also Supplementary (C) Coupling matrix for the 9-spin chain. (D) Average coupling magnitude as a function
11,13,14
of site expected behavior in a system which is not a time
15 crystal in 3
Qubit
Instance
Instance12
n of the simulated probabilities ofdiamond lattice. symmetry
(c)Orange
In the of D MBL is spontaneously
discrete 23broken
time crystal and ρ~Jphase,mustk|either ↵ the system
15 exhibits robust subharmonic oscillations lasting ∼0.5 10
prethermal regime. symmetry of D is spontaneously broken 0.0 and ~ must either
the -0.5 0.0ρ
distance across the chain. line: least-squares-0.5 fit to a power-law function , the1010
Qubit
10 mm
Instance
Instance 0 /|j
z z
n
"P(x )# − 1 Floquet
giving J 0 cycles.
= 6.7(1) Hz select
The
and ↵ =a nonzero
different
2.5(1). (E) value
colors
Measured for the
are
expectationorder
Instance
Instance34
associatedparameter
values h z
i afterM
with ¼ hσ
different
initializing
2n i
i ρ~ initial 5 states, which select
are a nonzero
characterized value by for the
their order parameter
quasienergy -0.5
-0.5-1.0 M 2n ¼ hσ i iρ
i i (1)
-1.0
Instance
Instance45 j
12 5 ~
in thethestateright Thehavedata islong-range
corrected for measurement errors (25).latterbehavior
Fig. 1 | The Sycamore processor. a, Layout of processor, showing a rectangular
|"""""""""i. or correlations. The 20 case 40 is 60 impos- or0state,
have long-range 40 correlations. 80The latter
panel. Crucially, the time Instance 5 -1.0 case is impos-
s, P(x ) is the probability of bitstring x
-1.0
array of 54 qubits (grey), each connected to its four nearest neighbours with
0 10crystalline
20 0 is80independent
100 0 of 10the initial 20 20as expected 60for an MBL 100time
0 B. Example: Periodically 20driven40Isingt spins
i i couplers (blue). The inoperable qubit is outlined. b, Photograph of the
0 10 t 20 0 20 40 t 60 80 100 10 t 20 0 60 80 100
m circuit, and the average is over the
, F is correlated with how oftencrystal.
XEB we (d) Schematic
Sycamore chip. sible given our initial
depicting thestate, as long-range
53t transmon qubit correlations
t Sycamore chip. (Mi t et al., 2021) sible isolate given our oneinitial
t dimensionalstate, as long-range chains of correlations
ings. When there are no errors in the
on of probabilities is exponential L (see = 8, 12,
connectivity 16, to20
was chosen
cannot
qubits be
be forward-compatible from generated
this
with error correc-
in finite time. Then,
two-dimensional grid. at odd (e) A times digital Let us now
Floquet consider consisting
sequence a cannot
concretebemodel ofgenerated
random that realizes finitethetime. Then,
inlongitudinal fields, at odd times
FIG.
this12 2. Observing a many-body localized discrete time-crystal. a, Schematic of the experimental circuit composed
t ¼ ð2n þ 1ÞT,Ising we have behavior described above. We
trandom consider anwe Isinghave ferromag-
26
nd sampling from this distribution will tion using the surface code . A key systems engineering advance of
r hand, sampling from the uniform disordered nearest-neighbor
device is achieving high-fidelity single- and two-qubit operations,FIG.
ofnot
t 2.interactions,
identicalObserving a many-body
cycles of and
the approximate
unitary Û localizedπ-pulses
. The local discrete
polarization time-crystal.
are applied
of each to
qubit, ¼
a, Schematic
h ð2n
Ẑ(t)i, þ is 1ÞT,
of the
“bit-string”
measured experimental
at the initial
end.circuit
In composed
states.
the (f)
following
F
net, with awith
oflongitudinal
each qubit, hfield Ẑ(t)i,applied to break initialthe Ising
n
2 and produce F = 0. Values of F just in isolation but also while performing a realistic computation with
XEB XEB of t identical cycles of the unitary ÛF . The local polarization is measured at the end. In the following
In simultaneous
e probability that no error has occurred the thermal gate operations on phase
many qubits. We(left panel),
panels, we
discuss the highlights
panels, we
the disorder
investigate
investigate
averaged
a number
−iDT
a number ofiDT
autocorrelation
of disorder
disorder
instances
instances
each
each with
function a di↵erent
a di↵erent
quickly random
random
decays to
bit-string
bit-string
zero.
initial state.
Instate.the b,MBL Experimental
b, Experimental
time iDT
i
crystal phase, the same correlation ψ(ð2n

obabilities P(x ) must be obtained from below; see also the Supplementary Information.
j ~ þ 1ÞT)ih ψ(ð2n
~
values of þ h 1ÞT)j
Ẑ(t)i ≈
measuredðXe at ρ
Þ~ ðe
qubit 11.XÞ Data are symmetry
shown for explicitly,
five representative
lasting ∼ circuit and drivenj ψ(ð2n
~ 2 atþ
circuit high
1ÞT)ih
instancesfrequency
ψ(ð2n
~ deep by
þ
in a very
1ÞT)j
the ≈ ðXe
thermal (g−iDT= Þ~ ρðe XÞ
0.60)
values of hfunction exhibits subharmonic oscillations 10 Floquet cycles. in the(Mi et al.,
(g =2021)
um circuit, and thus computing F is XEB In a superconducting circuit, conduction electrons condense into a
Ẑ(t)i measured at qubit 11. Data are shown for five representative instances deep thermal 0.60)
tum supremacy. However, with certain macroscopic quantum state, such that currents and voltages behave
2,30
btain quantitative fidelity estimates of quantum mechanically . Our processor uses transmon qubits , which and MBL-DTC
6
(g = 0.97) U phases., which c, Autocorrelator strongA transverse
= hẐ(0) Ẑ(t)i field. Thus,
at time we obtained
qubit crystal’s
11, take from averaging the results of 36
ning wide and deep quantum circuits. also implement a benchmarking
can be thought of as nonlinear superconducting resonators at 5–7and “echo”
GHz. MBL-DTC (g = 0.97) phases. sequence, ð18Þ demonstrates that
echo c, Autocorrelator A = hẐ(0)Ẑ(t)i at qubit 11, obtained from averaging the lifetime is consistent
† t the t results of 36 with ð18Þ
nough FXEB for a circuit with sufficient circuit instances. For the same circuit instances, the average autocorrelator at the output of ÛECHO = (†ÛtF ) tÛF is also measured
The qubit is encoded as the two lowest quantum eigenstates of the
being limited by experimentalcircuit
assical computing cost is prohibitively imperfections
instances. Forand
the samedecoherence. Figures
circuit instances,
resonant circuit. Each transmon has two controls: a microwave drive a-c are
the average adapted at
autocorrelator from (Randall
the output et al.,
of ÛECHO 2021),
= (Û F ) ÛF iswhile figures
also measured
ause our logic gates are imperfect and and its square root,¼A X0,ρ~,is
Xisshown ð19ÞA for comparison. d, Top HðtÞ
shown alongside
to excite the qubit, and a magnetic flux control to tune the frequency. ¼H
panels: TheðtÞ þ V;
ratio A/A0 obtained ¼X
ð21Þpanel
from
c.ρ~ X
c. Bottom ð19Þ
o create are sensitive to errors. A single
of the algorithm will completely shuffle
and
d-f are adapted from (Arute et its square
al.,
panels: 2019)
A/A
root,
andA0(Mi et al., 2021).A for comparison. d, Top panels: The0ratio A/A0 obtained from panel
alongside
Each qubit is connected to a linear resonator used to read out the qubit
0 as a function of t and qubit location.
state5. As shown in Fig. 1, each qubit is also connected to its neighbouring
Bottom
n close to zero fidelity11 (see also Sup- qubits using a new adjustable coupler31,32. Our coupler design allows us
panels: A/A0 as a function of t and qubit location.
ore, in order to claim quantum suprem- to quickly tune the qubit–qubit coupling from completely off to 40 MHz.
ssor that executes the program with One qubit did not function properly, so the device uses 53 qubits and (a) Time crystal (b) Non-time crystal (a) Time crystal
86 couplers.
are varied
varied in in order
orderto todemonstrate
The processor is fabricated using aluminium for metallization and
are demonstratestability stabilityofofthe thephase
phase explore explorethe theDTC DTCphase phaseand anditsits transition
transition intointo a thermal
a thermal
of time crystalline order that exhibit minimal
Josephson junctions, and indium for bump-bonds between two silicon
in anan extended
extended parameter effective
region andacrossacrossdisorder
disorderreal- real- phase. phase.AtAtg g==1,1,the themodelmodelimplements
implements a ⇡aorder
⇡ pulse which
rocessor wafers. The chip is wire-bonded to a superconducting circuit board
in parameter region and Non time crystal Time crystal pulse which
sor named ‘Sycamore’ which consists
4 transmon qubits, where each qubit decay.
is
and cooled to below 20 mK in a dilution refrigerator to reduce ambient
izations; The The limitations
limitationsofof(ii)
thermal energy to well below the qubit energy. The processor is con-
izations; (ii)finite
finitesize sizeand
and(iii)
(iii)finite
finite exactly exactlyflips flipsallallqubits
qubits(in(inthe thez basis)
z basis) and and returns
returns themthem
neighbours, in a rectangular lattice. The nected through filters and attenuators to room-temperature electronics,
coherence time
coherence time are are addressed,
addressed, respectively,
respectively,by varying totothe
byvarying theinitial
initialstate
stateover overtwo twoperiods.
periods.A A key key signature
signature of of
CTOBER 2019 the number of qubits in the
the number of qubits in the system and by separating system and by separating the DTC is the presence of robust period doubling, i.e.i.e.
the DTC is the presence of robust period doubling,
e↵ects of
e↵ects of extrinsic
extrinsicdecoherence
decoherencefrom fromintrinsic
intrinsicthermaliza-
thermaliza- extending extendingover overa afinitefiniteextentextentininparameter parameter space,
space, eveneven
D. Prethermal discrete timetion; crystal (iv) TheTheexistence
existenceofofeigenstate eigenstateorder orderacross
acrossthe theen- en- asasg gisistuned tunedaway
log(t)
awayfrom from1. 1. Strong StrongIsing Isinginteractions,
interactions, log(t)
tion; (iv)
<latexit sha1_base64="GjqyJsz/JswmPpgDM/ifxWQRHGg=">AAAB73icbVBNS8NAEJ3Ur1q/qh69LBahXkoigh6LXjxWsB/QhrLZbtqlm03cnQgl9E948aCIV/+ON/+N2zYHbX0w8Hhvhpl5QSKFQdf9dgpr6xubW8Xt0s7u3v5B+fCoZeJUM95ksYx1J6CGS6F4EwVK3kk0p1EgeTsY38789hPXRsTqAScJ9yM6VCIUjKKVOj0ZD0kVz/vliltz5yCrxMtJBXI0+uWv3iBmacQVMkmN6Xpugn5GNQom+bTUSw1PKBvTIe9aqmjEjZ/N752SM6sMSBhrWwrJXP09kdHImEkU2M6I4sgsezPxP6+bYnjtZ0IlKXLFFovCVBKMyex5MhCaM5QTSyjTwt5K2IhqytBGVLIheMsvr5LWRc1za979ZaV+k8dRhBM4hSp4cAV1uIMGNIGBhGd4hTfn0Xlx3p2PRWvByWeO4Q+czx8LjY9N</latexit>
<latexit sha1_base64="GjqyJsz/JswmPpgDM/ifxWQRHGg=">AAAB73icbVBNS8NAEJ3Ur1q/qh69LBahXkoigh6LXjxWsB/QhrLZbtqlm03cnQgl9E948aCIV/+ON/+N2zYHbX0w8Hhvhpl5QSKFQdf9dgpr6xubW8Xt0s7u3v5B+fCoZeJUM95ksYx1J6CGS6F4EwVK3kk0p1EgeTsY38789hPXRsTqAScJ9yM6VCIUjKKVOj0ZD0kVz/vliltz5yCrxMtJBXI0+uWv3iBmacQVMkmN6Xpugn5GNQom+bTUSw1PKBvTIe9aqmjEjZ/N752SM6sMSBhrWwrJXP09kdHImEkU2M6I4sgsezPxP6+bYnjtZ0IlKXLFFovCVBKMyex5MhCaM5QTSyjTwt5K2IhqytBGVLIheMsvr5LWRc1za979ZaV+k8dRhBM4hSp4cAV1uIMGNIGBhGd4hTfn0Xlx3p2PRWvByWeO4Q+czx8LjY9N</latexit>
tire spectrum is established. The

tire spectrum is established. The flexibility of our quan- flexibility of our quan- which produce
which produce long-range long-range spatial order, are
spatial order, are essential for essential for
7,10
1. Floquet prethermalization tum tum processor,
processor, combinedcombinedwith withthe thescalable
scalableexperimental
experimental this thisrobustness
robustness 7,10 . This is in contrast to a system of
. This is in contrast to a system of
protocolsdevised
protocols devisedin inthe thefollowing,
following,allows allowsusustotofulfill
fulfillthese
these decoupled decoupledqubits qubits( ( == 0)0)which which rotate
rotate byby a continuously
a continuously
criteria and
criteria and observe
observetime-crystalline
time-crystallineeigenstate eigenstateorder.
order. varyingangle
varying angle⇡g⇡gevery everyperiod periodinstead insteadof ofbeing beinglockedlocked
While many-body localization Short
The provides
The experiment
experiment a conducted
Prethermal isisconducted
quantum
Long ononap-
ananopen-ended,
open-ended, Short
linear Prethermal
linear atatperiod
perioddoubling.
doubling.
Long Prior
Prior Shorttheoretical
theoretical workwork
Prethermal 32 32 has Long
has shown shown
Time
chain of ofwith Time Time (Q Time Time Time
proach to realizing a time crystal chain LL = = 20 20 superconducting
ansuperconducting
infinite lifetime transmonqubits
transmon qubits (Q1 1 that thatmodel
model(1) (1)is isexpected
expected toto bebe in in anan MBL MBL DTC DTC phase
phase
through Q
through LQ2020)) that
that are are isolated
isolated fromaatwo-dimensional
from two-dimensional ininthe therange
rangeg g>>gcg, cand , andtransition
transition to a thermal
todependence
a †thermal phase
phase z at
at
in the thermodynamicFIG. 1. Thelimit, τ
expected ∼ e
time →
dependence∞, inof this
hσ z
qubitsi viai section,
at stroboscopic times, starting from a low-temperature FIG. state
1. with
The respect
expected to
timeUDU (for of hσ i i at stroboscopic time
grid. We
grid. We drive
drive the the qubits viaaatime-periodic
time-periodic FIG. (Floquet)
(Floquet)
6 Schematic a acritical
critical value
depicting ⇡c ⇡
valuegcgprethermal 0.84.
0.84. time crystalline order
example, for a state
we focus on an alternate, disorder-free with all spins polarized in the z direction.). Panel (a) shows the prethermal time-crystal
example, phase,
for a and
state panel
with all(b) shows
spins polarized in the z direction.). Panel
circuit with ttstrategy,
circuit ÛÛtFt with identicalcycles
identical
dubbed
cycles (Fig.2a)
(Fig. 2a)ofofÛÛ :: Achieving MBL g⇠
the non-time-crystal prethermal
F phase. FF
identified via the the
stroboscopic
Achieving MBL inin
non-time-crystalthismodel
behavior
this model forfor
prethermal
of the
g⇠ 1 requires
phase.
1magnetization
requires disor-
disor-
Floquet prethermalization (Fig. 1). iHere, iP
P the lifetime
iP
P of i i as PPa X̂ function ofdertime.
der ininthe thetwo-qubit
two-qubit
(a) In12,32 the interaction,
interaction,
Q non-time-crystal , which is
i , iwhich isprethermal
eveneven under
under
ÛÛFF == ee|scale ihhiẐẐi i e 4i 4 i i i Ẑ i ẐẐ
i Ẑi+1 2 ⇡g i X̂ Q
{z i i+1}} |ee| 2 {z (1) Ising
22 ⇡g i i i
the time crystalline order does not {zi iwith }} |e| the {z system (1) Isingsymmetry
symmetry
{z } 12,32 , X̂ ia, condition
a condition that was not met
| {z }
phase, the magnetization during , X̂
i ,
ieven and odd that was not met
26,27 Floquet cycles
i
size, but can be exponentially long
longitudinalfields
in a particular
longitudinal Isinginteraction
fields Ising interaction xxrotation
control 011026-5
rotationbyby⇡g⇡g bybysomesomepast pastDTC DTCexperiments
experiments 26,27
. .Ising-odd
Ising-odd terms,
terms,
becomes identical after
i.e.hih, iare
i.e.
a short
, areapproximately time
approximatelydynamically (corresponding
dynamically decoupled
decoupled
to local
byby thethe 01102
parameter, namely, the ratio where
where of X̂the
X̂ii and driving
and are frequency
ẐẐii are Pauli operators.
Pauli operators. to Each thermalization).
Each i i(h (h
i )i )isis During
x xpulses
pulses the
overtwo
over intermediate
twoperiods,
periods, thereby
thereby prethermal
lowering
lowering their
their regime,
e↵ective
e↵ective
local energy scales, J, within sampled
sampled randomly from
the randomly
system: from
τ ∼[ [e1.5⇡,ω1.5⇡,
. 0.5⇡]([([ ⇡,this
D /J 0.5⇡] ⇡,⇡])
⇡])forforev- ev- disorder
magnetization disorder may strength
strengthremain and andhindering
hindering
small, but localization
non-zero,
localization (in(inthethe
until ab-ab-
it
ery realization
ery realization of ofthethecircuit.
circuit. Overall,
Overall,ÛÛFFimplements
implements
decays toanan zero sence
sence
at ofofindependent
the independent
Floquet disorder
disorder
heating in inthe thei ).i ).Utilizing
time-scale Utilizing
(long newly
newly
time).
In static systems, the concept interacting
of
interacting Ising prethermalization
Isingmodel modelthat thatisisperiodically is
periodically“kicked”“kicked” bya aprethermal
developedCPHASE CPHASEgates gates(see (see SM forfor details) with con-
(b) In by the developed discrete time SM
crystal details)
phase, with
the con-
mag-
a powerful framework for understanding transversepulse
transverse pulsethat the
that thermaliza-
rotates
rotates allqubits
all qubitsby by⇡g⇡gabout
aboutthe thexx tinuously tinuouslytunable tunableconditional
conditionalphases phasesallows allows usus to to
engi-engi-
axis. In In this
thiswork,
work,gscales—the
gisistuned
tunedwithin within therange netization
range[0.5,
[0.5,1.0] oscillates
1.0]toto neer neerstrong throughout
strongdisorder
disorderinin i tothe prethermal regime, lead-
tion of systems with disparate axis. energy dy-
the i to fulfill
fulfill this thiskey key requirement.
requirement.
ing to stroboscopic curves that exhibit plateau behavior. The
namics (and thus thermalization) of “fast moving” de- prethermal time crystal ultimately melts at exponentially late
grees of freedom are governed by an effective Hamilto- times, controlled by the frequency of the driving field. Figure
nian where the slow degrees of freedom remain basically adapted from (Else et al., 2017).
frozen (Berges et al., 2004; Gring et al., 2012; Kagan and
16
Klinger, 1974). In driven Floquet systems, the presence Second, Heff must be able to host a symmetry-broken
of two distinct energy scales is particularly natural, since phase with respect to the emergent symmetry. Note that
the frequency of the drive and the interaction energy- owing to Landau-Peierls-type arguments, this naturally
scales within the Hamiltonian are independent. Indeed, places constraints on the interaction range and dimen-
it has recently been established that Floquet prether- sionality for realizing a prethermal discrete time crystal
malization is a generic feature of driven systems in the (PDTC) (Kyprianidis et al., 2021; Machado et al., 2020;
high-frequency regime (Abanin et al., 2017a,b, 2015; Else Pizzi et al., 2021a). Finally, the initial state of the many-
et al., 2017; Kuwahara et al., 2016; Mori et al., 2016; Wei- body system must have a sufficiently low energy density
dinger and Knap, 2017). From the perspective of heating, (measured with respect to Heff ), such that it equilibrates
as discussed in Sec. IV.A, one of the central consequences to the spontaneously symmetry broken phase during the
of Floquet prethermalization is that the frequency of the prethermal regime (Fig. 6).
drive exponentially controls the heating time scale, t∗
(Abanin et al., 2015; Machado et al., 2019; Mori et al.,
2016). The physical intuition for this exponential scal- 2. Prethermal discrete time crystal in a 1D trapped ion chain
ing is as follow: at large frequencies, ωD J (where J
is the local energy scale of the many-body system), the To highlight the dynamical signatures of a prethermal
system must undergo ∼ ωD /J local rearrangements in discrete time crystal, as well as the distinctions from an
order to absorb a single unit of energy from the drive. MBL time crystal (Sec. IV.C.2 and Sec. IV.C.3), we turn
This intuition holds for both quantum and classical sys- to a recent experiment performed on a trapped ion quan-
tems, although in the remainder of this subsection, we tum simulator (Kyprianidis et al., 2021). As discussed in
will focus on the quantum setting; we will return to the Sec. IV.C.1, some of the first experimental observations
classical case in the outlook (Sec. VI.B.1). of time crystalline behavior (Fig. 4a,b) were originally
Crucially, (Kuwahara et al., 2016) and (Abanin et al., observed in small-scale trapped ion experiments (Zhang
2017a) proved that up until the timescale, t∗ ∼ eωD /J , et al., 2017), and one of the central advances in recent
the system does not absorb energy from the drive and work is the ability to experimentally distinguish between
the stroboscopic Floquet dynamics are (up to exponen- local thermalization (e.g. “short time” in Fig. 6) and late-
tially small corrections) captured by an effective static time dynamics in the prethermal regime.
Hamiltonian, Heff . Unlike MBL, these features of Flo- The experiment consists of a one dimensional chain of
quet prethermalization are largely independent of disor- N = 25 Ytterbium ions and the Floquet evolution alter-
der, dimension and the range of interactions (Fan et al., nates between two types of dynamics. First, a global π-
2020; Machado et al., 2020, 2019; Peng et al., 2021; Pizzi pulse is applied and then second, the system evolves un-
et al., 2021a; Rubio-Abadal et al., 2020). der a disorder-less, long-range, mixed-field Ising model.
While the ability to exponentially delay the onset of At leading order in the Floquet-Magnus expansion, the
Floquet heating is crucial for realizing and stabilizing a stroboscopic dynamics of the ions are captured by the
discrete time crystal, one additional key insight is still effective Hamiltonian,
needed, namely, the observation that Heff can exhibit X X y
Heff = Jij σix σjx + By σi , (16)
an emergent symmetry (which need not be present in
i,j i
the original Floquet evolution) protected by the discrete
time-translation symmetry of the drive (Else et al., 2017; which exhibits an emergent Ising symmetry. To begin,
Machado et al., 2020). The presence of such an emergent (Kyprianidis et al., 2021) demonstrate that independent
symmetry in Heff allows one to sharply define phases of of the initial state of the ion chain, the system exhibits
matter (i.e. via symmetry-breaking) in the prethermal slow, frequency-dependent heating to infinite tempera-
regime (Fig. 6). A mathematical description of prether- ture. As aforementioned, in order for the system to ex-
mal discrete time crystals in the quantum setting closely hibit time crystalline order, the initial state of the ion
follows the discussion of sections III.B.1 and III.B.2. chain must reach a pseudo-equilibrium state (within the
This lays the foundation for the connection to time prethermal regime) in which it breaks the emergent Ising
crystalline order—if a many-body system prethermalizes symmetry (Fig. 7). For a one dimensional system at
to a state that spontaneously breaks the emergent sym- finite temperature, this is only possible for sufficiently
metry of Heff , it will also exhibit time-crystalline order, long-ranged interactions (Fisher et al., 1972; Kosterlitz,
corresponding to a subharmonic oscillation between the 1976; Thouless, 1969). The trapped ion system gener-
different symmetry sectors (Else et al., 2017; Machado ates such long-range interactions by using a pair of Ra-
et al., 2020). man laser beams to couple the internal spin states to
To recap, within the framework of Floquet prether- motional modes of the ion chain (Sørensen and Mølmer,
malization, realizing a time crystal requires a few ingre- 1999).
dients: first, the Floquet drive must induce an emergent Initializing the system with a Néel state (top, Fig. 7a),
symmetry in the prethermal effective Hamiltonian Heff . (Kyprianidis et al., 2021) observe that the magnetiza-
RES EARCH | R E P O R T 17
PN
(a) tion, M (t) = 1/N i=1 hσix (t)ihσix (0)i quickly decays to
zero, in agreement with the expectation that the sys-
tem equilibrates to a symmetry- unbroken paramagnetic
state. On the other hand, starting from a polarized initial
state, M (t) exhibits period doubling (bottom, Fig. 7a),
whose lifetime is directly controlled by the frequency of
REPORT RES EARCH | REPORT
the drive. Moreover, in this latter case, the lifetime of the
time-crystalline order matches with t∗ , consistent with
the intuition that Floquet heating ultimately melts the
PDTC at late times.
An intriguing open question is whether this melting
can be delayed or fully arrested by coupling the system
to a cold bath (Else et al., 2017). More broadly, the
stability and rigidity of many-body time crystalline order
in dissipative, open quantum systems [Fig. 8(c)] remains
an active area of exploration (Booker et al., 2020; Buča
and Jaksch, 2019; Buča et al., 2019; Chinzei and Ikeda,
(b)
2020; Dogra et al., 2019; Gambetta et al., 2019b; Gong
Fig. 3. Characterizing the PDTC phase. (A and B) (Top) Magnetization et al., 2018;(t*)Iemini
and magnetization decay Keßler
et al., 2018; (tPDTC) times for four
et al., different
2020, 2019; initial states at
dynamics, M(t), for the Néel state (A) and the polarized state (B). ForKeßler
the Néel et varying energy
al., 2021; densities (34). At low
Kongkhambut energy
et al., densities,
2022; tPDTC (orange) are
Lazarides
state, M(t) quickly decays to zero at time tpre (dashed vertical line), independent
et al., 2020;substantially
Lledó shorter
et al.,than t* (magenta)
2019; Lledó and
andindependent
Szymańska,of frequency,
of the drive frequency. For the polarized state, the subharmonic response (2T- highlighting the trivial Floquet phase. At high energies, tPDTC is similar to t*,
2020; Seibold et al., 2020; Tucker et al., 2018; Zhu et al.,
periodicity) persists well beyond tpre, and its lifetime is extended upon increasing highlighting the long-lived, frequency-controlled nature of the PDTC behavior. Th
2019).is Welocation
the drive frequency. The lifetime of the prethermal time-crystalline order t
return to it in Section V.
of the observed crossover in energy density is in agreement with an
PDTC
obtained by fitting the magnetization dynamics to an exponential decay (34). independent quantum Monte Carlo calculation (red and blue shaded regions)
Statistical error bars are of similar size as the point markers. (Bottom) (34). Error bars for the decay time correspond to fit errors, whereas error bars
^x $magnetization dynamics across the entire ion chain at w/J0 = 38. (C) E. Heating for
Periodically thedriven
energy density correspondCondensates
Bose-Einstein to statistical errors.

Fig. 3. Characterizing the PDTC phase. (A and B) (Top) Magnetization (t*) and magnetization decay (tPDTC) times for four different initial states at

ing the PDTC phase. (A and B) (Top) Magnetization (t*) and magnetization decay (tPDTC) times for four different initial states at
dynamics, M(t), for the Néel state (A) and the polarized state (B). For the Néel varying energy densities (34). At low energy densities, tPDTC (orange) are
the Néel state (A) and the polarized state (B). For thetoNéel varyingtpre energy densitiesline),
(34).independent
At low energy substantially
densities, tPDTC (orange)
than t*are
itstate, M(t)
is present quickly
in Hdecays
eff (Eq. 2).zeroWhen
at time such (dashed
emer- verticalthe Floquet Long-lived
dynamics shorter
of thesubharmonic
energy density (magenta)
responses andhave
In this independent
work, beenwe of frequency,
explored
report the experimenta
decays to zero at time tpre (dashed vertical
of the symmetry line),
drive frequency. independent substantially shorter than t * (magenta) and independent of frequency,
gent is For the polarized
present, state, the
the exact subharmonic
Flo- response
(Fig. 2A). This (2T-
contrast
extensively highlighting
highlights
in thean
driven trivial FloquetEinstein
essential
Bose phase. At high
observation energies,
of robust
condensates tPDTC is similar time-crystalline
prethermal
(BEC) to t*,
ncy. For the polarized state, the subharmonic
periodicity) persistsresponse
well (2T- t , highlighting
beyond and its theistrivial
lifetime Floquet
extended upon phase. At
increasing highhighlighting
energies, tthePDTC is similarfrequency-controlled
long-lived, to t*, nature of the PDTC behavior. The
quet dynamics are approximately generated pre point: Although t can bedisorder
*
without extended (Autti
by increasing behavior
et al., 2018, that Giergiel
2021; persists beyond
et al.,any early-time
s well beyond tpre, and its lifetime is extended
theevolving
drive frequency.upon increasing highlighting the long-lived, frequency-controlled natureofofthetheobserved
PDTC behavior. The
by underThe Hefflifetime of the
for time T,prethermal
followed time-crystalline
by the driving tPDTC is location
orderfrequency,
2020; no order
Hannaford, survives crossovertransient
beyond
2022;Monte Kongkhambut
in energy dynamics.
density is in Ouragreement
results
2022;shaded
Liao
with an
highlight the
. The lifetime of the prethermal time-crystalline
obtained by of
fitting order tPDTC is
the magnetization location oftothe anobserved crossover in energy density is inquantum
agreement an calculationet
withCarlo al.,
the action G. This latter part dynamics
suggests that exponential
tpre when decaythe(34).
system is in independent
the trivial Floquet phase. potential of(red and blue
periodic driving, regions)
in general, and
the magnetization dynamics to an exponential
Statistical error barsdecay
are (34).
of similar independent
size as the quantum
point markers.Monte
(Bottom) Carloet al., 2019;
calculation(34). Sacha,
(redError
and bluefor
bars 2015,
shaded 2020; Smits
regions)
the decay et toal., 2020,whereas
2021,
the time-crystalline order is naturally captured The Floquet dynamics starting from the time correspond
prethermalization, fit errors,
in particular, error bars
as a framework
rs are of similar size as the pointx
^ markers. (Bottom)
$magnetization dynamics across the(34).
entireError
ion bars for
chain at the0 decay
w/J = 38. time
(C) correspond
2018;
Heating Wang
for to
the fitenergy
eterrors,
al., whereas
2021b).
density errorAbars
correspond paradigmatic
to statistical errors. example is
bydecay
the system’s magnetization
for four differentdynamics;
initial statesthe polarized state are markedly distinct (Fig. 3B). for realizing and studying out-of-equilibrium
(t*) and magnetization (tPDTC) times at

ynamics across the entire7ionPrethermal
FIG. chain at w/J0 discrete
= 38. (C) Heating
time crystal for thein energy density correspond provided
a disorder-free to statistical byerrors.a BEC bouncing off an oscillating mirror
he Néel action
varying energy densities (34). of G !changes
At low
171 energy
+" the sign
densities, tPDTC of the order
(orange) are First, M(t) exhibits period doubling, with M > phenomena. Even in the presence of noise, we
pendent substantiallychain of Lparameter
shorter =is 25
than
it Yb
* (magenta)
tpresent insxiH ions.
and (Eq.(a)
every
independent
2). For
period.
Whenof initial
As
frequency,
such states whose
a result,
emer- the0 foren-
Floquet (Flatté
even dynamics
periods of and
and M 0Holthaus,
the< energy
for odd 1996; InSacha,
periods.
density find
this that2015).
work, the Let the
weprethermal
report us experimental
try
dynamics remain
eff
Hseeff(2T-
(Eq. 2). When ergy
highlighting thedensity
such trivial
emer-
gent
there (measured
Floquet phase.
the
symmetry
are two At
Floquetwith
high respect
energies,
is dynamics
present,
possibilities for to
of isH
tPDTC
the
the eff )energy
similar
the
exact lies
to tin
Flo-
prethermal *, the trivial
density
(Fig. In contrast
2A). This
Second, the this
decay
to work, we report
highlights
of this
understand the
an essential
period-doubling
this experimental
scenario observation
by stable, ofwhich
consideringrobustsuggests
aprethermal
particle time-crystalline
thatun-an external bath
creasing
is present, phase
highlighting
the exact of Flo-
the
queteffective
the long-lived,
dynamics, dynamics Hamiltonian
frequency-controlled
(Fig.depending
2A). This
are nature
contrast
approximately
on the (top
of the PDTCpanel),
highlightsbehavior.
generated
system’s energy the dynam-
The point:
an essential Although
behavior tder
observation
* can of
is directly be robust
extended
controlled prethermal
by the
by time-crystalline
increasing
frequency behavior that persists
at sufficiently lowoffbeyond any early-time
temperature can stabilize the
the influence of gravity which bounces an oscil-
tPDTC
re is locationics
approximately of the
of observed
thedensity
generated by crossover
magnetization
evolving
point: inAlthough
energy
(Fig.under
1B). is
Heff
If density
largely
thefor is in
time
t* prethermal
can be agreement
independent
T,extended
followed
state with
bybyan
of the
increasing
cor- drive
theofdriving frequency,
behavior
the drive. nothe
that
Third, order survives
persists
lifetime ofbeyond
beyond theany transient
early-time
time- dynamics.
prethermal Our results
dynamics for highlightlong
infinitely the time
lating mirror [Fig. 8(a)]. In the reference frame of the
Heff for independent
r(34). time T, quantumthe
frequency
followed Monte
and
by
responds
Carloofcalculation
quickly
action
the driving
to the This (red and
G.equilibrates
latter
frequency, bluesuggests
to
part
symmetry-respecting
shadedDespite
no zero.
order regions)
that
survives the
tpre
para- beyond rapidthe
when system
orderisdynamics.
transient
crystalline in the trivial
mirrors the Floquet
Our phase.
results
dynamics of thepotential
highlight the of
(29). Thisperiodic
standsdriving, in general,
in contrast and
to localization
(34). Error bars of
for the
thedecay time correspond to fit errors, whereas errorexhibits
bars Theslow mirror (instarting
whichfrom the the mirrorprethermalization,
is held constant at z = as a0),
This latter part decaysuggests that the
magnet, magnetization,
time-crystalline
tpre
thewhen theorder
magnetization the
system isenergy
naturally
is
isin theand
zero density
captured
trivial Floquet phase.
remains energyFloquet
densitydynamics
potential of periodic
shown in Fig.driving, in general,
2B, demonstrat- and
based approaches in particular,
for stabilizing framework
Floquet phases
Heating for the Floquet
energy density correspond
heating. Forto statistical
initial stateserrors. whose energy density lies the particle’s gravitational potential energy is given by
ne order is naturally captured by the system’s
unchanged The acrossmagnetization
Floquet a period.dynamics dynamics;
Conversely, starting ifthe frompolarized
the state are markedly
the thatprethermalization,
ing Floquet heating in distinct (Fig.
particular,
ultimately 3B). thefor realizing
as a framework
melts in whichand studying of
the presence out-of-equilibrium
an external bath tend
magnetization dynamics; in the symmetry-broken
action polarized
the
prethermal of Gstate !changes
"
phase
state are of
the
corresponds sign H
markedly of
toeffthethe (bottomorder(Fig.
distinct
symmetry- panel),
First,
3B).
PDTC the
M(t) for
at Utimes.
exhibits (z)
realizing
late =
period m(g
anddoubling,
studying+a withcos(ωM >D t))z, with
phenomena.
out-of-equilibrium to the
Evenposition
destabilize in
thethedynamics. of the
presence of
Ournoise,
work we point
ynamics
nges theof sign the energy
of thedensity
stroboscopic In this
dynamics
parameter sxi work, thewemagnetization
ofevery report the
period. aexperimental
Asdoubling, exhibit
result, subhar-
order
breaking First, M(t)
ferromagnet, exhibits theperiod
magnetization is 0M
with for>By
even periods
phenomena.
considering and
particletwo Madditional
Even <at0infor
z the
>odd 0.periods.
presence
initialClassically, find to
of noise,
states, that
we the prethermal
aa number
single of future dynamics
particle in this remain
directions: (i) exploring
contrast highlights an essential observation
there are two ofcurves
robust prethermal
possibilities for the time-crystalline
prethermal Second, the decay of this period-doubling
every period.monic
h t* can be extended
oscillations
As a result, nonzero
by increasing
0 and
for
behavior
(solid
even periodsevery
alternates
that persists
correspond
and
beyond
M <
period. 0 for
any
to
Theodd
early-time
even
re- periods
periods.
we explorefindthe that the
potential
stability prethermal
ofexhibits
the PDTCdynamics
aphase
2:1 as astable,
remain
subharmonic which suggests
generalizations response of that an external
which
Floquet is bath
prethermalization
ssibilities for the while dynamics,
dashed
prethermal curves
Second, depending
correspond
the decay on theto system’s
odd energy
periods), behavior
whose is
life- directly controlled by the frequency at sufficiently low temperature can stabilize the
quency, no order survives sultingtransient
beyond
time extends density
with
2T-periodic,
(Fig. dynamics.
1B).
increasing Ifdirectly
the Ourofresults
subharmonic
prethermal
frequency.
this highlight
period-doubling
dynamics
state
The cor- the offunction
lifetime thethis
of
stable,
drive.
of energywhich
Third, time
thelow
suggests
density.
conceptually
lifetime
that 3C
Figure
equivalent
oflifetime
an depicts
the time-
external
to that bath
toof
prethermal
a quasi-periodic
the driven non-linear
dynamics
drive (36), (ii) stabilizing
for infinitely long38), times
nding on the system’s energy
is the behavior
hallmark is
of a time controlled
crystal. by the frequency both theat sufficiently
heating
pendulum temperature
and the can of stabilize
the the
Floquet topological phases (37, and (iii
stem is in the trivialprethermal
Floquet phase. timepotential
responds to the
ofcrystalline
of symmetry-respecting
periodic
order driving,
isthe
consistentin general,
para-
with and crystalline
thebeing order
cut-prethermal mirrors thediscussed
dynamics ofintheSec. III.B
(29). This(Flatté
stands in and Holthaus,
contrast to localization-
t Ifdynamics
the prethermal
starting state
from cor- the
theWe prethermalization,
investigate drive. Third,
these intwo lifetime
regimes
particular, as aby of
mea-
framework time-
time-crystalline dynamics
order.
1996).in Fig. Near
In 2B, fortheinfinitely
thedemonstrat- bottom
quantum case, long
of times
leveraging nonequilibrium many-body dyna
off by the slow magnet, the magnetization
Floquet heating is zero and (b)
time-scale. remains energy density shown
Experimentally based this approachesleadsforto two “non-
stabilizing Floquet phases,
symmetry-respecting
are markedly distinct (Fig.para- suringfor
3B). crystalline
the order
autocorrelation
realizing and mirrors the dynamics
of theout-of-equilibrium
studying magnetization of the spectrum,
(29). This stands
where no in contrast to localization-
symmetry-breaking mics for enhanced metrology (39).
unchanged acrossof a period. Conversely, if the time ingcrystal
that Floquetspreading
heating ultimatelywavepackets” melts the localized in which the to presence
the double-minimaof an external bath tends
netization zeromeasured
ibits periodisdoubling, andwithremains
M>
phasephenomena.
diagram
energy density a prethermal
Evenshown in Fig.discrete
in the presence 2B, demonstrat-
of noise, we phase exists, based the approaches
decay offor thestabilizing
magnetization Floquet is phases,
iods and M < 0 as
for odda functionprethermal
periods. of the
find
state
energy
that
corresponds
the density
prethermal
to
of the
the symmetry-
dynamics initial state.
remain
PDTC at late times.
States of the effective Floquet to destabilize
Hamiltonian,
REFERENCES the dynamics.
corresponding
AND NOTES Our work to points
s a period. Conversely, if the ing that Floquet 1X N ! heating "! xultimately " with melts frequency-independent
the in which the presence of an external
and substantially bath tends
faster
decay near the
of thistoperiod-doubling breaking
edge of the ferromagnet,
spectrum the
(i.e.
x magnetization
which order is By
respect considering
to two
the time additional
twothe initial
period-doubled states, to
orbits a number
1. of future directions:
(Bialynicki-Birula et al.,Third
L. D. Landau, E. M. Lifshitz, Mechanics,(i) exploring
Edition: Course
corresponds the symmetry- stable, M ðt Þwhich
PDTC ¼ latesuggests
at times. si ðtthat
Þ san i ð0Þ
externalð3Þ bath than the toheating
destabilize dynamics.
scale. By contrast, Our work near points
ectly controlled byH eff )
the nonzero
exhibit
frequency PDTC
at andorderalternates
sufficiently N low while every
states
temperature period. in
can The
the re-
trivial
stabilize the we explore
phase of the stability of the PDTC phase as a generalizations
of TheoreticalofPhysics,
Floquet Volumeprethermalization
1 (Butterworth-Heinemann,
agnet, the magnetizationsulting is By considering
2T-periodic,
i¼1 two additional
subharmonic dynamics
initial states, the topof
function
1994;
toofenergy
a the
number Buchleitner
of future
spectrum,
density. Figurewhere and
directions:
3Ca depicts
symmetry-Delande,
(i) exploring 1995; Buchleitner
ed. 3, 1976).
to a quasi-periodic
et al.,
drive (36), (ii) stabilizing
Third,
ernates theevery
lifetime Hofeff the
period.
do time-
The
notre- prethermal
exhibit
we
prethermal
explore dynamics
the
time
stabilityfor infinitely
crystalline
of the longorder.
PDTC times
phase
Figures
as a generalizations of Floquet prethermalization
is the(Kyprianidis
hallmark of a ettime breaking ferromagnetic
2002; Flattéphase and exists, the
Holthaus, two 2. P. Mansfield,
1996). J. Phys. C Solidin
topological phases (37,con-
However, State Phys. 4, 1444–1452
der mirrors the dynamicsadaptedoffrom the (29). This stands al.,crystal.
in 2021).to localization-both the heating time and the lifetime of the Floquet(1971).
contrast 38), and (iii)
dic, subharmonic dynamics Starting function ofthese
with a low–energy energy density.
density Figure Néel 3C depicts
state to aare
time scales quasi-periodic
consistent with drive
one (36),
another (ii) stabilizing
of and
y shown in Fig. 2B, demonstrat-We based investigate approaches fortwo regimes
stabilizing by mea-
Floquet phases,time-crystallinetrast order.toNear the classical
the bottom case, quantum
leveraging tunneling between
3. L. M. K. Vandersypen, I. L. Chuang, Rev. Mod.dyna-
nonequilibrium many-body Phys. 76,
of
et aheating
time crystal.
ultimately melts the (Fig.
suring3A),both
the
in we thetheheating
observe
autocorrelation
which that M(t)
presence time
ofofthe and
anquickly thedecays
magnetization
external lifetime
bath to of
tends thethe
thus Floquet
Floquet
spectrum, where
the topological
heating nolimits
two phases
the PDTC
symmetry-breaking
minima (37,lifetime.
results 38),inand a (iii)
mics for1037–1069
finite enhanced
lifetime, metrology
(2005). τ , for (39).the
eimes.
these two regimes by mea- tpre
zero attotime-crystalline
, in agreement
destabilize order.
with the
the dynamics. Near
Our the points
expectation
work bottom of exists,
Our
phase leveraging
results aredecay
the nonequilibrium
consistent
of the with a phase
magnetization many-body
bound-
is dyna-
4. H. Zhou et al., Phys. Rev. X 10, 031003 (2020).
subharmonic oscillations. From theS. spectral
Kitamura, Annu.perspec-
0 Þ ≈ 2,REFERENCES AND NOTES
rrelation of the magnetization 5. T. Oka, Rev. Condens. Matter Phys. 10,
ng two additional that the
initial states, tothe spectrum,
asystem
number 1 of
X where
equilibrates
Nfuture
!
tonothe
directions:
" !
symmetry-breaking
symmetry-
" (i) exploring mics for
ary occurring
frequency-independent enhanced
at energy anddensitymetrology
substantially (39).
NJ
hHeff i=ðfaster
stability of the PDTC phase as unbroken,
a phase tÞ ¼exists, the
M ðparamagnetic
generalizations x
decay
phase.
of sFloquet
ð t Þ of
s x the
This
0 magnetization
behavior
prethermalization
ð Þ ð3Þ inisagreement tivewith [Sec. III.B.3],
independent this
numerical manifests
1. L. D. as
387–408
Landau, two
E.
(2019).
M. Floquet
Lifshitz, Mechanics, eigen-
Third Edition: Course
than the heating time scale. By contrast, near 6. A. C. Potter, T. Volume
Morimoto, A. Vishwanath, Phys. Rev. X 6,
is N i¼1 idrive
frequency-independent
frequency-independent, in
i
and
direct substantially
contrast to faster states,
REFERENCES |ψ
AND i
NOTES and |ψ i, whose of relative
Theoretical Physics,
Floquet 1 (Butterworth-Heinemann,
eigenval-
ergyN density.
1 X ! x "! x " Figure 3C depicts to a quasi-periodic (36), (ii) stabilizing thecalculations
top of the fromspectrum,quantum whereMonte
1 Carlo (34). ed. 3, 1976).
a symmetry- 2 041001 (2016).
ng timesand 1. L. D. Landau, E. M. Lifshitz, Mechanics, Third Edition: Course
i ðt Þtheslifetime
i ð0Þ ð3Þ Floquet
of the than the topological
heating time phasesscale. (37, By 38),contrast,
and (iii)breaking
near ferromagnetic phase exists, the two 2. P. Mansfield, J. Phys. C Solid State Phys. 4, 1444–1452
N
ne i¼1order. Near the bottom Starting of leveraging nonequilibrium many-body dyna-time scales are
of Theoretical Physics, Volume 1 (Butterworth-Heinemann,
(1971).
thewith topa of the spectrum,
low–energy density where
Néel a symmetry-
state ed.consistent
3, 1976). with one another and
where no symmetry-breaking Kyprianidis
mics Science 372,
et al.,enhanced 1192–1196 (2021) 11 June 2021 3. L. M. K. Vandersypen, I. L. Chuang, Rev. Mod. Phys. 76, 4 of 5
(Fig. 3A), wefor
breaking observe thatmetrology
ferromagnetic M(t) quickly (39).
phase exists,
decays to thethus twoFloquet 2. P.heating
Mansfield,limits
J. Phys.the
C Solid
PDTC Statelifetime.
Phys. 4, 1444–1452
1037–1069 (2005).
he decay of the magnetization
ow–energy density Néel state is tpre, in
time scales are (1971).
zero atREFERENCES ANDconsistent
agreement with thewith one anotherOur
expectation andresults are consistent with a phase bound-
NOTES
4. H. Zhou et al., Phys. Rev. X 10, 031003 (2020).
pendent and substantially faster 3. L. M. K. Vandersypen, I. L. Chuang, Rev. Mod. Phys.5.76,T. Oka, S. Kitamura, Annu. Rev. Condens. Matter Phys. 10,
rve that M(t) quickly decaysthat to the 1.thussystemFloquet
L. D. Landau,equilibrates
E. M.heating to theThird
limits
Lifshitz, Mechanics, symmetry-
theEdition:
PDTC Courselifetime.
ary occurring1037–1069 density hHeff i=ðNJ0 Þ ≈ 2,
at energy(2005).
ng time scale. By contrast, near 387–408 (2019).
reement with the expectation unbroken, Our paramagnetic
resultsPhysics,
of Theoretical phase.
are consistent
Volume Thiswith behavior
a phase bound-
1 (Butterworth-Heinemann, in agreement withet al.,
4. H. Zhou independent
Phys. Rev. X 10, numerical
031003 (2020). 6. A. C. Potter, T. Morimoto, A. Vishwanath, Phys. Rev. X 6,
spectrum, where a symmetry- ed. 3, 1976).
18
Rep. Prog. Phys. 81 (2018) 016401 Review
ues are perturbed away from π → π + 1/τ . Intuitively, (a) (b)

these two Floquet eigenstates correspond to the “bond-
ing” and “anti-bonding” configurations of the effective
double-well potential. When τ → ∞, the configurations
|ψ± i = √12 (|ψ1 i ± |ψ2 i) are exchanged under each period,
and a system prepared with a majority amplitude in one a
or the other will thus exhibit Sτ B. But with tunneling,
<latexit sha1_base64="b7/vCs5ze5KtVd66W3yyALYBfbk=">AAAB6HicbVBNS8NAEJ3Ur1q/qh69LBbBU0lEqMeiF48t2A9oQ9lsJ+3azSbsboQS+gu8eFDEqz/Jm//GbZuDtj4YeLw3w8y8IBFcG9f9dgobm1vbO8Xd0t7+weFR+fikreNUMWyxWMSqG1CNgktsGW4EdhOFNAoEdoLJ3dzvPKHSPJYPZpqgH9GR5CFn1FipSQflilt1FyDrxMtJBXI0BuWv/jBmaYTSMEG17nluYvyMKsOZwFmpn2pMKJvQEfYslTRC7WeLQ2fkwipDEsbKljRkof6eyGik9TQKbGdEzVivenPxP6+XmvDGz7hMUoOSLReFqSAmJvOvyZArZEZMLaFMcXsrYWOqKDM2m5INwVt9eZ20r6qeW/Wa15X6bR5HEc7gHC7BgxrU4R4a0AIGCM/wCm/Oo/PivDsfy9aCk8+cwh84nz/DXYzl</latexit>
<latexit
these configurations slowly precess into each other over

time3.τ Left
the tunnelingFigure panel: schematic
, restoring the symmetry plot of a system of N atoms
(Buchleit- Figure 4. Left panel: schematic plot of a system of N atoms bouncing
bouncing on an oscillating mirror and prepared in a many-body
ner et al., 2002;
Floquet Flatté
state (24). andEach Holthaus,
of two atomic 1996;
clouds Holthaus,
moves with a period on(c)an oscillating mirror as in figure 3 but after the measurement of
Figure 3. Left panel: schematic plot of a system of N atoms
1995; Holthaus and
2Tbouncing
but they Flatté,
exchange 1994;
their Sacha,
positions 2015).
after time T so that the entire theFigure
position 4. of a single
Left panel:atom—atomic
schematic plotcloud visible of
of a system in N theatoms
plot moves
bouncing
on an oscillating mirror and prepared in a many-body
with
on aanperiod 2T. Right
oscillating mirror panel
as inshowsfigurethe results
3 but afterofthe themeasurement
measurements of
However, τFloquet
can state
Floquet bestate isradically
periodic
(24). Each with atwo
ofextended period atomic T. Right by panel
clouds moves
replacing showswithtime a period
of the
positions
position of of
100 i.e. at t = 0 one
atoms,atom—atomic
a single measures
cloud visiblepositions
in the plot of moves
evolution
2T but theyof the corresponding
exchange their single particle
positions after probability
time T so that density
the entire
a single particle
(25) with
for
Floquet = an
N state 10is 4 N -particle Bose-Einstein con-
g0 N = −0.5
, periodic with, ω = 1.1 and
a period T. Right λ = panel0.06 . Reprinted
shows time
100withatoms, let the
a period 2T.remaining
Right panel atomsshows evolve and after
the results of T/2 one again
the measurements
measures
of positions positions
of 100 ofatoms,
100 atoms i.e. atand t =so0on. oneThe histograms
measures positions presented
of
densate with figure
an with
evolution permission
attractiveof the corresponding from Sachasingle
contact (2015b),
interaction particle Copyright Hint
probability (2015) by
=density in 100
the right panel indicate that time periodic evolution of theone system,
g0 4 the(25)
American
for N =Physical
10 4
, g Society.
N = −0.5 , ω = 1.1 and λ = 0.06 . Reprinted atoms, let the remaining atoms evolve and after T/2 again
− 4 |Ψ(z)| (Sacha, 2015). 0 In appropriate units, the after the spontaneous time translation symmetry
measures positions of 100 atoms and so on. The histograms presented breaking, can be
figure with permission from Sacha (2015b), Copyright (2015) by
observed
in the right in a panel
singleindicate
experimental that time realization.
periodicThe initial of
evolution number of
the system,
Hamiltonian forFigure
thethe field
American
3. operator
LeftPhysical
panel: schematic Society. Ψ̂ isplotgiven of a system by of N atoms atoms 104 and the othertranslation
Nthe=spontaneous parameterssymmetry as in figure 3. Reprinted
after
Figure 4. Left panel:time schematic plot of a systembreaking, of N atoms canbouncing
be
Z ∞ interacting
bouncing on an oscillating mirror
via δ-contact potential one has to consider the fol-FIG.and prepared
in a many-body
state1(24). a atomic clouds moves with a period figure observed
on8 an with
(a) permission
in a single
Schematic
oscillating mirror
fromdepicting
Sacha (2015b),
experimental
as in figure realization.
3 but
Copyright
Nafter The
atoms the initial
(2015)
measurementnumber
bouncing by the ofoff an
of
Ĥ = Ψ̂Floquet
†
dz lowing
2T (z) many-body
−
but they exchange ∂ 2 Each
+Floquet
(1 of two
+
their positions
Hamiltonian cos(ω after D t))|z| Ψ̂(z) American
atoms
the
oscillating Physical
Nmirror
position = 10 of
4 Society.
aand the atom—atomic
single
(Sacha, other2015).parameters cloud
(b) as inNumerical
figurein3.the
visible Reprinted
plot moves
simulations
onetime T so that the entire
z
interacting 2via δ-contact g
potential has to consider the fol- figure awith permission
0
i ! 2Tstate
Floquet !is ∞ periodic "with a period g0 T. Right panel # shows time of the with
matchesscenario
American
period 2T.
thePhysical (a) from
in Right
non-interacting
panelSacha
shows showsthat
result,
(2015b),
the the
results
|ψ ! ≈
Copyright
time (2015) by the
of theevolution
measurements
|N, 0!. However,of the
g0 H lowing
evolution
4= many-body
dt
of dz
the corresponding Floquet
ψ̂ H(t) Hamiltonian †
singleψ̂particle
+ ψ̂ − i∂
probability
ψ̂, densityatomic of positions of 100 Society.
atoms, i.e. at t = 0 one0measures positions of
F
− |Ψ̂(z)|(25) .for0!N = 10 t (17) when density exhibits Sτ B. At each time shown, the posi-
2T 0! , ∞
4
g0 N = −0.5 " ,ω= 2 1.1 and λ = 0.06 # . Reprintedtions matches 100 atoms, let the remaining atoms evolve and after T/2 one again
4 $ % g 0 † Copyright (2015) by of 10 2
the
atoms non-interacting
(out of 10 4
) result,
are
measures positions of 100 atoms and so on. The histograms presented |ψ
measured 0 ! ≈ |N, 0!
and . However,
a histogram
figure
H≈ J permission
with from 1 +(2015b),
Sacha N|U − 2U(c) > J,
ĉ†dt ĉ†2dz ψ̂ +H(t) ψ̂ ψ̂) − i∂t ψ̂,
F = |A (23)
Two approximationsthe − can
American ĉbe
0 1 Physical1− 0 used
ĉ2Society.to 4analyze (U − 22U the resulting
12 of thosewhen
in the right panelisindicate
positions depicted. that time12periodic evolution of thecondensate
Bose-Einstein system,
2 after
in aittransverselythe spontaneous time translation symmetry breaking, can be
" J$ % 1 # is energetically favorable
pumped to collect
high-finesse
N|U − 2U all bosons
optical in
cavity a single
can real-
dynamics (Sacha, 2015): (i)
† taking † N →† ĉ∞ while holding 12 | > J, The initial number (23)
× −(ĉ†1 )2ĉĉ122ĉ+
≈ 1− (ĉĉ†22)ĉ22ĉ21 + 2ĉ(U 1 1− ĉ†2 ĉ2U
2 12 +) const., (22) ize observedwave-packet
localized
signatures in aofsingleSτ Bexperimental
inand an open realization.
consequently system. the ground
In state
particular, of the
of
"2 reduce to a 4periodically
4
g0 N fixed, the dynamics # driven itatoms N = 10 and the
is energetically other parameters
favorable to collect as inallfigure
bosons 3. Reprinted
in which
aoscillation
single
interacting † 2 δ-contact the
atom-cavity Hamiltonian system (22) can is a Schrödinger
exhibit a cat-like
period-doubled state
where
Gross-Pitaevskii U×and(ĉvia
equation U
1 )12 ĉ22are
(GPE). + (ĉ † 2potential
equal
2 ) The ĉ21to+integrals2ĉ
GPE† one
1 ĉ1 ĉ2is
† has to consider the fol-
ĉover
2 a+ const.,
time
classical and space (22)
figure with permission from Sacha (2015b), Copyright (2015) by the
is localized
between cleardistinct
if one wave-packet
writes
density such and
wave a consequently
many-body
patterns Floquet the ground
(Keßler et state
eigenstate of
al., 2021).
lowing many-body Floquet Hamiltonian American Physical Society.
of products
Hamiltonian field theory of the over probability
a complex densitiesfield ofψ(t,
two z), evolving and local-Figuresin the Hamiltonian
another Fock
adapted from (Keßler (22)
basis is
|ñ a
, ñSchrödinger
! where ñ cat-like
and ñ state
= N −which
ñ
!U 2Tand !U12∞multiplied 1 2 et al., 2021; 1 Sacha, 2 2015; 1Sacha
where are equal " tobyintegrals . In over #time (22) and space
can therefore ized
beof
H
wave-packets
efficiently
products
= dtof simulated;
the dz
probability
ψ̂ H(t) (ii) ggone
densities
+ ψ̂ ψ̂ of
equation
00 † can project
− i∂
two evolving
ψ̂,
we areis matches
clear
and Zakrzewski,
local-
ifthe
occupations onenon-interacting
writes
2017). of the
√
such first a result,
many-body |ψ0 ! ≈
localized Floquet
|N, 0!. eigenstate
wave-packet,However,
restricted
F to the Hilbert subspace spanned 2 by the
t previously when
in another Fock basis |ñ , ñ ! where ñ and ñ = N − ñ1
the interacting dynamics onto the (22) we φ1are = (u1 + e −iωt/2
u2 )/ 2 , and√the other localized wave-
1 2 1 2
|n1 ,%ntwo nearly-degenerate
ized 0
wave-packets 0
described Fock
J $ states multiplied 1
by g0. In equation
2 ! and the bosonic field operator occupations of the first localized wave-packet,
1 − e N|U √u2−)/2U212, | respectively.
−iωt/2
ground statesψ̂(z,(i.e.
t)≈≈the
restricted−u1 ĉto aforementioned
the
† −iωt/2Hilbert † subspace (Udouble-minima)
−spanned by the previously packet, φ2 = (u > J, Then, the (23)
1 ĉ+1 ĉe1 − ĉ2 ĉu22 ĉ2+where ĉ2U
1 and 12 ) ĉ2 are standard φ1 = (u1 + e−iωt/2 u2 )/reads 2 , and√the other localized wave-
described
of the single particle 2 Fock
Floquet states |n
Hamiltonian, , n 4! and the bosonic
effectively fieldre- operator fieldmany-body
theories ground
are state
generically ergodic (the integrability of
"
annihilation operators. Such 1 2
a restriction #
is valid provided the it is energetically
packet, φ2 = (u1 − e favorable
−iωt/2 to collect all
u2 )/ 2 , respectively. Then, thebosons in a single
† 2 2 −iωt/2
u2 ĉ2 21ĉmodes.
(ĉ†2 )small 2ĉ†1 ĉcouplings
†
ducing the problemψ̂(z, t)×
interaction ≈energy
to 1 1ĉ)
u(ĉ
two 1+ ĉbosonic
e+very
2is 2+where
and 1 ĉ2ĉĉ12 and+ const.,
ĉ2the
to arecomple-standard
(22)
the isotropic,
localized undriven
wave-packet GPE
and
|N, being
consequently
0! + |0, N! a the fine-tuned
ground state excep-
of
many-body ground |ψ state reads √
0 ! ≈is a Schrödinger . cat-like state (24)
annihilation
mentary
These approximations Hilbertoperators.subspace
canarebeequal Such
can combined bea restriction
neglected. isIt valid
into will provided
be
three the theHowever,
the casetion). Hamiltonianthis (22) expectation 2 may not applywhich to the
where
interactionU and energyU to integrals
12 is very small and couplings to the comple- over time and space is clear if one writes such |N,a 0! + |0, N! Floquet eigenstate
many-body
because
cases. When both the interaction
approximations energy per
are particle
made will
(i.e. be of
tak- the order present situation. Due|ψ ! to
≈ the gravitational
√ . confining (24) po-
of products
mentary Hilbertof the probability
subspace densities of twowill evolving local-
case The in corresponding
another Focksingle basis particle
0
|ñ1 , ñ2 !probability ñ1 density,
J can thatbe is neglected.
extremely Itsmall be
as the where
2 and ñ2 = N − ñ1
ing N → ∞ ofand the tunneling
ized
becausealso projecting
wave-packets
theother
splitting
interaction multiplied to the by two-mode
g . In equation pic- com-
(22) tential,
we the dynamics are effectively confined to a zero-
pared to any energyenergy scale of perthe particle
0
system. will be of the order
Eigenstates of are occupations
The corresponding 1 (z, t) = near
ρregion of |ψ̂the
!ψ
single0 √ the
†
(z, t)first
particle ψ̂(z, localized
t)|ψ
probability 0 "of the
wave-packet,
density,
restricted
ture), the dynamics the exhibit to the a Hilbert
bifurcation subspace spanned
transition by smallthe
that previously dimensional minima double-well.
theofHamiltonian
tunneling (22)splitting
correspond J that to is extremely
many-body Floquet as com-
states φ1 = (u1 + e−iωt/2 uN2 )/! 2 , and√the other localized " wave-
described
pared to
produces period-doubling.(Sacha, anyFock other states
energy |n12015) , scale
n2 ! and of the system.
the
Intuitively, bosonic Eigenstatesfield
due operatorThis
of is in contrast t)
≈to the |φ Gross-Pitaevskii
†
t)| 2
t) |φ t)|ψ t)| 2 equation in an
,
that are also eigenstates of the time translation operator T T packet, φ 2 ρ=1 (z,
(u 1 − = e −iωt/2
!ψ 0 |1 u2 )/ 2 , respectively.
ψ̂(z,(z, +
ψ̂(z, 2 (z, 0 " Then, (25) the
ψ̂(z, ≈ u1 ĉ1 +
t)apparent
theinteraction,
Hamiltonian e−iωt/2
(22) u2realizes
correspond ĉ2 where tothat ĉ1there
many-body andis ĉ2min- are classes
Floquet standard
states
isotropic potential 2an array of coupled pendula, which
orstate
to the attractive
what is when the one Floquet energy are two many-body ground N !reads 2 2
"
|φ1 (z, t)|of+states. |φ2 (z, t)| ,
imized when of allannihilation
that
the are alsooperators.
eigenstates
particles of (22).Such
eigenstates
inhabit of the
Eigenstates
the a restriction
same timefrom well; isthe
translationvalid firstprovided
tunnel- operator
class areeach the
TT have an extensive
is plotted in figure 3 at different
≈ density
2 |N, 0! time moments. The discrete (25)
interaction
what isby
spanned apparentenergy
Fock whenis very
states onesmall
with realizes
only and even couplings
that there are
occupations to the twoof comple-
classes
the + |0, N!
ing between the wellsHilbert is thereby suppressed because all timewhile
So, translationthe number symmetry
|ψ0 ! ≈ ofisdegreespreserved
√ of in . time evoislution
freedom formally
(24)
mentary
of theu eigenstates subspace
of (22). can be neglected.
Eigenstates from It will beclass
the case
e−iωt/2 t) mode sequentially.
2 (z,tunnel while eigenstates from thethe other first are is plotted in figure 3 at different time
class byinfinite, of the itmany-body Floquet eigenstate 2 |ψ moments.
0 ! but still
Thestate
this discrete
is
N → ∞ bosons must
because
spanned the by interaction
Fock statesenergy with per onlyparticle will be of theoforder
even occupations the time translation symmetry is preserved in time evolution be-
may be that the KAM theorem applies
Fock states with the odd occupations only (Sacha 2015b). The
extremely
causeofthere corresponding
vulnerable
are effectively onlysingle
to any particle
perturbation.probability
a few After density,
a measure-
If one only considers
of the utunneling
eAssuming
−iωt/2 thet) mode
2 (z, attractive
second
splitting
while( g0 <
approximation that is extremely
Jeigenstates
0 ) interactions from the (i.e.
that
small
other fi- class
as com-
are veryby ment the
of amany-body
position x1Floquet of a single eigenstateparticle, |ψ0degrees
! but
the thisofstate
symmetry
freedom
is is
nite N withinweak,pared
the
Focki.e. to any
two
states mode other
with the energy
picture),
odd scalequantum
occupations of theonly system. fluctua-
(Sacha Eigenstates
2015b). of extremely
that are resonant ρ (z,
vulnerable with
t) = !ψ
to the|
anyψ̂ † periodic
(z, t) ψ̂(z,
perturbation. t)|ψ drive.
"After Thus,
a measure- KAM
N|U − 2U12 | < J , the ground state |ψ0 ! of (22) gone because the quantum state of the remaining particles 1 0 0
tions in the mode theAssuming
Hamiltonian
occupation (22)remain,
attractive correspond(g0 <and 0 )tointeractions
many-body
it was shown Floquet
that are states
very
mightment protect of a positionSτ B inx1this Nof! a scenario
single 2particle, as it the does " for
symmetry a single
is
that
weak,are i.e. also N|U eigenstates
− 2U12 | exponentially ofJthe , thetime ground translation
state |ψ operator T9T gone because the quantum ≈ (z, t)|of+the
|φ1state (z, t)|2 , particles
|φ2remaining (25)
that the tunneling time increases < with N 0 !, of (22) particle in a double-well 2 Floquet potential (Sec. III.B.4).
what is apparent when one realizes that there are two classes
τ ∝ eαN (Sacha, of 2015; Wang et
the eigenstates of (22).al., 2021b). EigenstatesThus, from the TTSB first class Interestingly,
are 9 is plotted ininfigure the absence of periodic driving, this is
3 at different time moments. The discrete
is achieved not in the limit
spanned by Fock of large statessystem with only size, evenbut rather of indeed
occupations the time thetranslation
case; specifically, it has been proven that the
symmetry is preserved in time evolution
in the limit of large
e−iωt/2 uparticle
2 (z, t) mode number. while However,
eigenstates from thethe two-
other class undriven
by GPE in a
of the many-body Floquet eigenstate double-well potential |ψ0 ! but breaks this state ergodic-
is
mode approximationFock states rules without the odd theoccupationsFloquet heating only (Sacha that 2015b). ity. Below extremely a critical
vulnerable energy,
to any aperturbation.condensate After becomes
a measure- “self-
is generally presumed Assuming occur in(ga0 <driven
to attractive 0 ) interactions many-body that are very trapped” ment in of aone position
of the of a single
x1 minima particle,ettheal.,
(Albiez symmetry
2005; Jack- is
system (D’Alessio weak, and i.e. N|U Rigol, − 2U2014; 12 | < J ,Else the ground et al.,state 2020b; |ψ0 ! of (22) son and gone Weinstein,
because the quantum 2004). state It isofperhaps the remaining thenparticlesplausible
Khemani et al., 2019). that9 the driven GPE can exhibit the same phenomena,
Finally, if one only considers the first approxima- but in the rotating frame of the 2:1 resonance [Fig. 8(b)].
tion (i.e. N → ∞ but without taking the two-mode Numerical simulations of the driven GPE support this
limit), Sτ B effectively requires that the periodically interpretation (Kuroś et al., 2020; Sacha, 2015).
driven Gross-Pitaevskii equation is non-ergodic. This is Finally, there is of course the question of what occurs
in contrast to the expectation that classical non-linear when one makes neither of the two aforementioned ap-
19
proximations. Exact answers are difficult to obtain in Sτ B is prethermal in this case, as could be evaluated in
this regime, as it is a full quantum many-body prob- the N → ∞ limit using the 2D GPE.
lem, but numerical results have been obtained within
time-dependent Bogoliubov theory (Kuroś et al., 2020)
and the truncated Wigner approximation (Wang et al., V. OPEN, PERIODICALLY-DRIVEN SYSTEMS AND
2021a). Both conclude that out to thousands of driving STOCHASTIC DYNAMICS
periods, the boson population remains almost entirely
within the two modes, with no evidence for heating. Our discussion thus far has focused on closed systems
These findings support the conclusion of the two-mode in which the dynamics are deterministic, x → Φ(x).
approximation (with finite N ), where τ ∝ eαN . Of course However, when a system is coupled to an environment,
such numerical simulations cannot definitively rule out it is often fruitful to model the effect of the environ-
exponentially slow depopulation and heating (Sec. IV.D), ment’s chaotic motion as noise (most-conveniently taken
as might occur in a prethermal time-crystal (Else et al., to be Markovian), so that the dynamics become effec-
2017; Machado et al., 2019; Sacha, 2020). However, if tively stochastic. Rather than focusing on a particu-
one accepts the conclusion that as N → ∞, the GPE can lar microstate, one instead considers a probability dis-
generically break ergodicity in this setting (as it does in tribution over microstates ρ(x, t) which evolves under a
a static double well potential), the possibility that Flo- “master equation,” for example the Fokker-Planck equa-
quet heating is absent out to infinite times even for the tion. Integrating the master equation over one Floquet
driven GPE, seems like a reasonable (albeit, still surpris- cycle then produces a discrete update of the distribu-
ing) possibility. tion, ρ(x, t + T ) = Φ[ρ(x, t)]. In classical mechanics this
Finally we comment on the contention that time- results in a probability distribution over canonical coor-
crystals in driven BECs are “effectively few-body” (Khe- dinates ρ(x = {p, q}, t)R which evolves under a Markov
mani et al., 2019). This is partly a matter of nomen- process, ρ(x0 , t + T ) = Φ(x0 |x)ρ(x, t)dx, where Φ(x0 |x)
clature. On the one hand, these systems are clearly not are the Markov transition probabilities. In the quantum
few-body as the interaction, g0 , and the limit N → ∞ case, we have a density matrix ρ̂(t) which evolves under a
are necessary for Sτ B. Furthermore, while not rigorously “quantum channel” ρ̂ → Φ[ρ̂]. One can also in principle
proven (much like Floquet MBL is also not rigorously consider non-Markovian baths, a point to which we will
proven), the authors of this Colloquium find it plausi- return.
ble that Sτ B could be stable up to times, τ ∝ eαN , even The definition of Sτ B given in Eq. (3) generalizes to
when the physics is treated as a full many-body problem. the open case by measuring the local observable, O, in ex-
Again, much like the situation with MBL, this “plausi- pectation, and stability can be defined by requiring Sτ B
bility” is bolstered by the fact that the undriven GPE in be robust to perturbations of the stochastic dynamics
a double-well potential does, in fact, break ergodicity. subject to locality and any other dynamical constraints
On the other hand, it seems that Sτ B is only stable one is interested in [Fig. 8(c)]. The environment is both
because of the reduced dimensionality induced by the good and bad for Sτ B. On the one hand, coupling to
gravitational confinement, which effectively ensures the an environment introduces friction: the energy and en-
accuracy of the two-mode approximation; no such picture tropy produced by the periodic drive can now be ab-
would be obtainable for the MBL setting (Sec. IV.B). sorbed by the bath, which can prevent the long-time
We venture that a more useful distinction is the nature Floquet heating which would otherwise destroy Sτ B in
of the thermodynamic limit being taken. In the case the absence of MBL. This tends to help stabilize sponta-
of MBL time-crystals (or the driven, open time-crystals neous time translation symmetry breaking. On the other
of Sec. V), true Sτ B is achieved in the thermodynamic hand, if the environment is at finite temperature, the in-
limit L → ∞, with τ ∼ eL/ξ , keeping intensive quan- evitable noise which results may occasionally conspire to
tities fixed. In the case of the driven BEC, due to the cause phase slips in the period-doubled motion. Roughly
confining gravitational potential, the system size L is not speaking, if noise nucleates phase slips at rate 1/τ , the
relevant (in this sense the problem is zero dimensional) Sτ B has a finite auto-correlation time τ and there is no
and the problem reduces to being effectively few-mode, true long-range order. The interplay of a periodic drive,
even though the physics is realized in a full, many-body interactions, dissipation, and noise results in a truly non-
system. In this setting, Sτ B is recovered in the limit equilibrium situation which is exceptionally rich – just
N → ∞, keeping g0 N fixed, with τ ∼ eαN . like the world around us (Booker et al., 2020; Buča and
An intriguing probe of this distinction would be to con- Jaksch, 2019; Buča et al., 2019; Chinzei and Ikeda, 2020;
sider a 2D generalization of the driven BEC: bosons are Dogra et al., 2019; Gambetta et al., 2019b; Gong et al.,
gravitationally confined in the z direction, but propagate 2018; Iemini et al., 2018; Keßler et al., 2020, 2019; Keßler
freely along x. In this case, the density of states is ex- et al., 2021; Kongkhambut et al., 2022; Lazarides et al.,
tensive in x, and domain walls between the two period- 2020; Lledó et al., 2019; Lledó and Szymańska, 2020; Sei-
doubled orbits can form. It is natural to suppose that bold et al., 2020; Tucker et al., 2018; Zhu et al., 2019).
NATURE PHYSICS ARTICLES 20
On the other hand, if the bath is in equilibrium at finite temperature T, j

the fluctuation-dissipation theorem implies that friction must come
with noise49. In classical systems, this noise can be captured as a
Langevin force, F B(t ) = − ηq ̇+ ξ(t ), on each coordinate q, where
Time
η is the strength of the friction and ξ(t) is a stochastic force with
variance, ξ(t )ξ(t ′) = 2ηTδ (t − t ′). Taking η > 0, T = 0 reduces to
the damped case where period-doubling is easily stabilized, while
a combination of finite T and driving results in a truly non-equilib-
rium situation. The question concerning the existence of ‘classical
discrete time crystals’ (CDTC) can then be posed as follows: in what
dimensions can a classical many-body system, coupled to an equi-
librium bath, exhibit rigid subharmonic entrainment for either the
closed case (η = 0), the zero-temperature case (η > 0, T = 0) or the
finite temperature case (η, T > 0)?
Because the T = 0 (no noise) case is already known to feature
FIG. 9 Arnoldrigid subharmonic
tongues of the entrainment,
damped in we focus on T > 0.
this work equation.
Matthieu
Furthermore, if a CDTC exists in d* dimensions, it will presumably
The x-axis is exist
the fordriveall d ≥ d*,2 so we focus on the most delicate
frequency expressed via the case: T > 0 in
dimen-
(2ω0 )2 /ω
sionless ratio aone=dimension , while the y-axis
(1D). Although an equilibrium phase
D is the drive
transition is
impossible inindicate
amplitude δ. Contours 1D, mightthe there nevertheless
critical drivebeamplitude
a non-equilibrium
dynamical when
required for resonance phase transition between of
the coefficient a period-doubled
friction takesCDTC a and
a symmetric
given value c. When c > phase?
0, the period-double solution (a = 1)
In this
onsets only at finite δ. work,
Figure weadapted
first argue,from basedPedersen,
on remarkable 1980.results due
Fig. 1 | Period-doubled dynamics ‘boil’ out of a uniform initial state. In the
to Gács50–52 and Toom53–55, that in principle true CDTCs, with an FIG.main 10 panel
Domain walls betweenview different period-doubled so-
infinite autocorrelation time, are possible in all dimensions d >lutions
we present a stroboscopic qj(2ntD), with time n running
0 of the parametrically driven 1D Frenkel-Kontorova
down vertically (0!<!n!<!1,200) and space j running horizontally over the
(ref. 56). However, in 1D the construction is so baroque that we
modelnosc!=!100 oscillators. The
(Eq.(12)) in colour scale shows
contact with qj!<!0aas red, qj!>!0 as blue
finite-temperature
cannot yet explicitly prove this conjecture theoretically or numer-
A. “Activated” time-crystals and qj!≈!0bath.
Langevin as white.Colors
Note that we strobe every
indicate the twoamplitude
driving periods,ofwhich
the j-th
ically. To this end, we instead investigate in detail a more physi-
is the frequency of theat
subharmonic response. Hence,times the displayed
qj (t phase
cal Hamiltonian, the parametrically driven Frenkel–Kontorova oscillator observed even stroboscopic = 2nT ),
of the oscillators varies slowly. A detail of a smaller region strobed at
A classical(FK) model. The basic
realization of thisidea isinterplay
simple: eachisnonlinear
given oscillator
by in
with time running vertically. Inset, the oscillations at strobo-
the driving frequency, qj(ntD), is shown in the inset. The period-doubled
the chain undergoes a 2:1 parametric resonance, and we couple scopic times q (nT ), which reveal the period doubling. Figure
Langevin dynamics.
the oscillatorsDuetogether
to the to trycoupling withthethe
and stabilize CDTC en-phase at oscillations qjj(ntD)!∝!(−1)n are now manifest. Strikingly, the correlations
adapted from (Yao et
are antiferromagnetic bothal., 2020).
in time and space, even though the oscillators
vironment, each finitedegree
temperature. We find that
of freedom (q,this
p)model exhibits anan
experiences intriguing
are coupled together ferromagnetically (ωD!=!1.958, g!=!0.065, δ!=!0.067,
line of first-order dynamical phase transitions, between an ‘acti-
additional Langevin force FL (t) = −η q̇ + f (t). Here η is η!=!0.003, T!=!0.004). In the final state, there is a finite density of π-domain
vated’ period-doubled CDTC and a symmetry-unbroken phase,
the friction coefficient and f (t) is a white-noise stochas- walls between the two different period-doubled solutions.
that terminates at a critical point (Figs. 1 and 2). The activated (Hayashi, 1953; Pedersen, 1980, 1935; Taylor and Naren-
0 an autocor-
tic force withCDTC is not a true time function
auto-correlation crystal; rather, it has(thas
hf (t)f )i = dra, 1969). As shown in Fig. 9, for γk > 0 the sub-
2ηT δ(t − t0 ), relation
wheretime T that diverges
is the exponentially as
temperature ofTthe → 0. Because
bath. of this
exponentially diverging timescale, it would be extremely difficult harmonic responseof acquires
the full complexity a finite
Gács’ construction, threshold
which valueis for
we now explain,
The total force is obtained
in experiments by adding
to distinguish FL antoactivated
between the time- CDTC andthe in fact required.
a drive amplitude δk , and modes outside the resonance
true, long-range-ordered
dependent Hamilton’s equations,CDTC; ṗ = indeed,
−∂q H(t) to detect
+ Fthis difference
L (t). tongues are damped. Absent noise, Sτ B is then real-
over many mil- Conjecture on the existence of classical time crystals. Although
This leads toweamust masterconductequation
careful numerical – theexperiments
Fokker-Planck izedour
as ultimate
a sharpinterestmany-body bifurcation transition
lions of Floquet cycles. is to understand the possibility of time as some
crys-
equation – of theBefore
general divingform ∂t ρ({p,
into the details,q}, = L[ρ], where
t) emphasize
let us the key charac- tals in openof
combination Hamiltonian systems, it is first
driving frequency, worth considering
amplitude or damping a
teristic of a true
L is the Fokker-Planck long-range-ordered
operator. Integrating CDTC,the namely, the existence more general class of dynamical systems, ‘probabilistic cellular
Fokker-
of period-doubling with an infinitely
strength is tuned into the 2:1 parametric
τ, automata’ (PCA). Recall that a deterministic cellular automata
resonance of a k-
Planck equation over one Floquet cycle oflong the autocorrelation
the drive timemode.
which is stable to small perturbations of the dynamics. More pre- (CA)For is a strongly
set of spinsviscous
{σi}, where σi ∈ {1, 2, waves,
Faraday ..., N}, withthea equations
discrete
then gives thecisely,
discrete-time
if one considers update
a model ρ(t
of + T ) = Φ[ρ(t)].
periodically driven oscillators update rule
of motion areσi → T[σi − 1, σi, σi more
somewhat + 1] . In complex,
58
a PCA, the systembut the is instead
conclu-
with position coordinates q
Note that we are assuming the environment remains i , the autocorrelation time τ of period- described by a probability distribution P[{σi }] over spin configu-
doubling can be quantified as ⟨q (nt )qi (0)⟩ ∝ (− 1) e n −ntD∕ τ sion are similar (Edwards and Fauve,
, where rations, which is updated by a local Markov process; that is, each
1994; Kumar and
in equilibrium at temperature T , andi soD instantaneously Tuckerman,
σ 1994).
tD is the period of the drive and 〈 〉 indicates averaging over noise i is updated with a probability distribution that depends only on
satisfies the fluctuation-dissipation
realizations. For low but finite theorem
temperatures,which the relates itself and itsnoise
period-doubling However, neighbours.
(i.e. One way to
finite obtain a PCA is toTstart
temperatures, > with
0) in-
the magnitudeweofobserve
the friction
in the drivenη andFK modelthe noise
exhibits2ηT only. anIf activated
H(t) auto- a deterministic CA and perturb it by stochastic errors; for example,
duces fluctuations between the different
with probability E ≪ 1, a spin can violate the transition rule and flip
cycles of the
correlation time,
were time-independent, this ≈ e Δeff ∕ Tensure
τ (T )would , where Δeff is andetailed-
(via effective activation Sτ B,intoleading
a randomto a (the
state finite
preciseauto-correlation
statement of such antime. Recall
error model
balance) thatbarrier.
the systemThus, inrelaxes
1D, the CDTC to theorder of the FKensem-
canonical model we study fromcanour be found in ref. 51).of prethermal systems (Sec. III.B.1
discussion
survives only to exponentially long, but not infinite times. Despite Given that the PCA update rule is discrete time-translation-
ble at long-times where Sτ B is forbidden (Bruno, 2013;
the activated behaviour of the period-doubling, we nevertheless and invariant,
Sec. IV.D) that we may transform to a rotating
one can then ask if the long-time probability distribu-
Nozières, 2013;find Watanabe and Oshikawa,
a first-order dynamical phase transition 2015).
where τ(T) Butdrops dis- frametionK(t):
can ever H(q, p, t)with
oscillate → H(P,period Q, two,t)thereby
= Heff (Q, P a) time
realizing + V (t)
when H(t) iscontinuously
periodically and period-doubling
driven, this isneed completelynot destroyed
be the(Fig. 2). crystal.the
in which For driving
a deterministic
V (t)CA, a time crystal
is weak. For ais trivially
single obtained
pendulum
In 1D, such a first-order phase transition would be impossible in by letting binary spins transform under the rule 0 ↔ 1 (this is anal-
case: a non-equilibrium
equilibrium57.
steady state can develop in which (Eq.ogous6), H takes the form of a double-well potential.
to eff
our previous discussion of fine-tuning, or introducing
energy and entropy This leaves thefrom
flows question the drive attoa peculiar
of CDTCs the bath viaalthough
point: Thefriction
noise but willno then
noise).leadHowever,to activated
any infinitesimal hoppingrate of between
errors
the system. we believe a true CDTC is in principle possible in 1D, the obvious will desynchronize
the minima at rate the 1/τspins= eand −∆/k theBautocorrelations
T
, where ∆ iswill thedecay
quasi-
candidate model is only exponentially close to one. It may be that over a timescale τ ~ 1/E.
We can now return to the question of period-doubling energy barrier between the two minima. Thus, at finite
in open classical many-body systems such as the Faraday- temperature, a single pendulum is no longer a true time-
NATURE PHYSICS | VOL 16 | APRIL 2020 | 438–447 | www.nature.com/naturephysics
wave instability and coupled pendula (see e.g. earlier crystal, but rather an “activated” time-crystal in 439
which
discussions surrounding Eq. (12) and Sec. III.A). For the autocorrelation time of the Sτ B diverges exponen-
small oscillations and weak damping, the friction re- tially with temperature.
sults in a damped Mathieu equation of the general form The problem is richer when coupling the pendula into
q̈k = − ωk2 + δk cos(ωD t) qk − γk q̇k . The sub-harmonic an array, where Heff approximately takes the form of an
responses of this model have been studied extensively Ising model (recall Sec. III.B.4 for e.g. the definition of
21
P∗ ), metry. This net force destroys the possibility of phase

coexistence. Thus, even in 2D and greater, the time-
X (Pi − P∗ )2

g̃ X
Heff ≈ + a cos(2Qi ) + (Qi − Qj )2 , crystalline response in this and similar models still ex-
i
2 2 hibits an activated auto-correlation time. Intriguingly,
hi,ji
(18) a closely related effect has been studied in the context
of probabilistic cellular automata, (Bennett et al., 1990)
where the parameters a depends on {δ, , ωD /ω0 } (from which are introduced in Sec. V.D.
Eq. 12) and g̃ is the transformed coupling strength. As η → 0, the magnitude of the aforementioned force
The low quasi-energy configurations are then local- goes to zero and the density of domain walls can then
ized domain walls between the different period-m orbits. undergo a transition analogous to the equilibrium Ising
These domain walls are easily observed when numerically transition. However, once η → 0, one needs to again
simulating a 1D coupled array of pendula [Eq. (18)] in worry about heating from the residual time-dependent
the presence of the Langevin noise, as shown in Fig. 10. part of the drive (this is precisely related to the prether-
Indeed, one observes a gas of domain walls undergoing mal situation discussed in Sec. IV.D).
Brownian motion due to the noise, occasionally nucle- In summary, the existence of Heff provides a unify-
ating or annihilating in pairs. (Yao et al., 2020) studied ing framework for three regimes of time-crystalline be-
the motion of these domain walls in detail and found that havior: in the closed cased (η = T = 0), one has the
it leads to a decay in the Sτ B autocorrelation function, prethermal scenario where heating destroys Sτ B after
hqi (t = nT )qi (t = 0)i ∼ (−1)n e−t/τ . The decay rate τ is an exponentially long time τ ∼ eωD /J ; in the open case
proportional to the density of domain walls, which is in (η > 0, T > 0), the environment nucleates domain walls
turn related to the temperature of the bath through an which destroy the Sτ B on an exponentially activated
Arrhenius law, τ ∼ e∆/kB T , where ∆ is now the quasi- time-scale τ ∼ e∆/kB T , where ∆ is a quasi-energy barrier;
energy activation barrier required to nucleate a domain and finally for the purely dissipative case η > 0, T = 0,
wall. there is true Sτ B conceptually analogous to that of the
(Yao et al., 2020) found numerically that in 1D, col- coupled map lattice (Sec. II.A).
lective effects can cause the barrier ∆ to drop discontin-
uously as the parameters (e.g. δ or ωD ) are varied, indi-
cating a non-equilibrium phase transition at which Sτ B B. Experimental realizations of activated time-crystals:
is completely destroyed. Without careful examination pendula, AC-driven charge density waves, and fractional
of the temperature dependence at low T , this transition Shapiro steps
might easily be mistaken as evidence for the existence of
a “true” Sτ B phase ensconced within a non-equilibrium Several experiments on open many-body systems have
phase transition, while in reality the low temperature be- observed subharmonic responses which may be inter-
havior is always activated. preted within the framework of activated Sτ B. One ob-
Naively, the existence of a finite critical temperature, vious example is a shaken pendulum treated as a macro-
Tc , for equilibrium Ising symmetry breaking in dimen- scopic object of ∼ 1023 particles. The coupled oscillator
sions two and greater suggests that the Sτ B domain walls array of Eq. (12), for example, can be understood as a
cannot proliferate for T < Tc , which would thereby sta- macroscopic pendulum composed of atoms, qi , when ac-
bilize time crystalline order to infinite times. However, counting only for its 1D width. The effect of noise on such
in the damped system this is not correct. To see why, systems is of practical interest; for example parametric
(Yao et al., 2020) evaluated the effect of the Langevin resonance of a mesoscopic mechanical oscillator such as
force in the rotating frame and found two contributions. an AFM tip (Rugar and Grütter, 1991) is a standard tool
First, the {Q, P } experience a Langevin force at the same for mass and force sensing. In perfect isolation, driv-
temperature T , which locally equilibriates Heff , explain- ing will generate stresses on the oscillator, causing the
ing the activated processes discussed above. Second, the phonon modes (qk6=0 ) to absorb a portion of the driving
effective Hamiltonian acquires
P a new contribution from power, heating up the system and (at long times) melt-
the friction η, of the form i ηP∗ Qi . Its physical origin ing the oscillator. This is the prethermal scenario, though
is that in the rotating frame, P encodes the amplitude some work remains to rigorously prove there is a regime
of the oscillation (cf. Eq. (7)), and because the friction in which the time-scale increases exponentially (Mori,
prefers to damp the amplitude, this equates to a net force 2018; Ye et al., 2021). Contact with a bath (e.g., air at
on Q. The net potential, V ≈ a cos(2Q) + ηP∗ Q, then bi- T = 293K) prevents the oscillator from melting, and it is
ases Q to “roll downhill”, which following Eq. (7), would helpful to estimate the resulting activation time. In par-
unlock the response from ωD /2. The bias leads to a net ticular, the nucleation of a “domain wall” corresponds to
force on the domain walls which causes nucleated islands ripping the oscillator in half, an energy scale so many or-
to grow (Yao et al., 2020), even though both sides of the ders of magnitude above T = 293K, that the time-scale is
domain wall are related by the Z2 time-translation sym- inconceivable. However, collisions with an unlikely con-
VOLUME 52, NUMBER 25 PHYSICAL REVIEW LETTERS 22 18 JUNE 1984
spiracy of air molecules can also cause the entire oscillator

NbSe~
to collectively hop between the two quasi-energy minima; T=42K
(aj V, (=SOmV
1/2
using Eq. (12), one can estimate the quasi-energy barrier 1/2 1/1
25-
for such a fluctuation as ∆ ∼ 2δ(2 − ωD /ω0 ).
At scales of kilograms, meters, and room-temperature, C?:
bQ 2/5 2/5 2/5
2/3
∆/kB T again results in an inconceivably long time-scale.
However, for mesoscopic resonators it can be highly rel-
evant. An application of this effect is found in nanoscale J,
15-
mass sensing based on parametric oscillation of an AFM 50=
C3
C:
tip. In order to weigh an object attached to the tip, one CD
CD
detects the sharp jump in the amplitude of oscillations as

the detuning 2 − ωD /ω0 is driven near parametric reso-
nance; the precise location of the jump is highly sensitive
20-
to the total mass. However, finite-T noise rounds out
the transition into a non-linear crossover, an effect which
is experimentally observable and limits the practical res- I5-
olution of parametric-resonance-based sensing (Prakash
et al., 2012; Zhang and Turner, 2004). -20 -10 0
I I I
10
I
20
Two more intrinsically mesoscopic examples, which sample voltage V (mV)
can be recast in the language of activated time-crystalline
FIG. 1. Differential resistance vs dc sample voltage (a) with and (b) without applied rf voltage at 25 MHz. The peaks
in (a) correspond to steps in the direct I- V curve. A few peaks are identified by p/q (see text).
order, are AC-driven charge density waves (CDW) and FIG. 11 Differential resistance measurements, dV /dI, in the
fractional Shapiro steps in Josephson junction arrays. charge density wave system NbSe3 . A bias voltage V (t) =
Both have received tremendous experimental attention; Vdc + V
sured with and without externally D t) is rfapplied,
rf cos(ωapplied voltage and the DC
quency locking component
between the of internal
the CDW (or
at co,„t/2n = 25 MHz.resistanceFor thedV rather/dIlow rfisvoltage
measured, Josephson)
holding frequency
V and rv;„,
ω and all harmonics pt0, „,
fixed.
since the two systems are conceptually equivalentused (Bohr dc
here, ET is not reduced from the value in the
dc rf D
of the applied rf field. The harmonics of m, „, are
et al., 1984; MacDonald and Plischke, 1983), absence we will of rf, butSharp peaks in the resistance correspond
there appears a whole family of
to plateaus in the
induced by the inherent nonlinearity of the system.
I-V curve.
peaks at impressed dc current such that The location of these peaksspectral
Direct can beanalysis attributed
of the to a oscillations'
current
largely focus on the former. In a charge density wave ma- motion of the CDW in which it shifts
in the by p/q wavevectors
of per
= absence applied rf reveals rich harmonic
terial such as NbSe3 (Grüner, 1988), the electron density p~ext q int~ (3)
driving period, resulting in a subharmonic intensity ofbetween
current, with response the harmonics qco;„, de-
spontaneously develops a charge density modulation where pat and q are integers. The ratio cu;„$2rrIcDw with q. Hence it is natural to interpret
the AC components of the voltagecaying
was found to be 30 MHz cm2/A in agreement with
and slowly
current. In the absence
the peaks as regions in which any harmonic of the
twice the Fermi wavevector, n(x) ∼ n0 cos(2kF x + θ(x)),
earlier studies. of ' theA fewAC drive
of the (panel
peaks b), the subharmonic
are identified internal frequency response locksistoabsent.
any harmonic of the
where θ is the slowly varying phase of the CDW. Treating
by p/q in the figure.Figure Theadapted fromwere
identifications (Browncon- et al., field. ' To the extent that mode locking
1984).
external
for several p/q by plotting Icnw vs co,„, and
θ as a single macroscopic degree of freedom, one firmedobtains
checking that the slopes were indeed p/q times that
within such regions is complete, the CDW is unable
to respond to changes in the applied dc voltage, and
a phenomenological equation of motion given by (Grüner
for the fundamental. The features with small p and the resistance rises to that of the normal electrons
et al., 1981): q are more conspicuous, Mostbeingnotably,both taller q 6=wid-
for and 1 (see alone.
for example
In the present the prominent
experiment, the steps have
er. Figure 1 is a typical trace. By varying V,f and lower resistance than the normal electrons, indicat-
either morepeak in Fig.
steps 11 can for p/q = 1/2), thetheresponse
locking is is not subhar-
θ̈ + (ω0 τ )−1 θ̇ + sin θ = E(t)/ET ~e„,, (19) or fewer be observed. ing that complete. The tendency
While the dependence monic:on the theseCDW shifts
parameters is by
not 1/qtoof a wavelength
lock is weaker for larger with each
p and q.
strong, substantially increasing or decreasing either It is largely agreed that complete mode locking
period of the AC-drive.
where sin θ accounts for the potential that pins theV,CDW,
ω0 τ is a relaxation time (note that time is rescaled
f or co,„, significantly reduces the number of steps
which canby be observed. While Eq. (19) treats θ as single
The steps with q =1, p ~ 1 have been studied
'
occurs for Eq. (2) only because of the inertial
term. ~'macroscopic
In NbSe3 thevariable,
inertial term is negligible
'
the natural oscillation frequency ω0 ) due to dissipation
previously " in the experiment
and are well known to arise from fre-
there are up to several hundred megahertz.
spatial fluctuations
based on the frequency-dependent
which Theconductivity
argument,
from the material, E(t) is the applied bias, and ET is can be accounted for by considering a 2D or 3D array
2278
the threshold bias for transport. This is the equation of of θ(r) coupled through an elastic stiffness. (Middleton
motion for a damped and driven pendulum. When E(t) et al., 1992) showed theoretically that the subharmonic
is larger than the threshold ET , the system enters into a response remains robust in this many-body setting. In
sliding state with h∂t θi = 6 0, generating a finite current. fact, in the overdamped regime with a purely sinusoidal
In the experiments of interest, a bias is applied with pinning potential, subharmonic responses are only found
both DC and AC components, E(t) = Edc +Eac cos(ωD t). when treating the problem as many-body one.
In the sliding state, the motion of the CDW over the pe- The phenomenological equation of motion [Eq. (19)],
riodic pinning potential defines a frequency scale ωdc ≡ which predicts perfect Sτ B, also neglects the effect of
hθ̇i ≈ ω0 τ EEdc
T
. When ωdc = pq ωD , commensuration ef- thermal fluctuations that are of course present in exper-
fects allow the AC-oscillations to assist the motion of iments such as Fig. (11). As discussed in (Thorne et al.,
θ over the potential. Since the current Idc is propor- 1987a,b), a careful examination of the experimental I-V
tional to ωdc , commensuration can lead to a plateau in curves reveals that some (but not all) of the subharmonic
Idc whenever ωdc ≈ pq ωD . This leads to the emergence plateaus are rounded-out in a sample dependent fashion.
of a “devil’s staircase” of plateaus (Brown et al., 1984; Complementary to this, the spectral distribution I(ω) (or
Thorne et al., 1987b; Zettl and Grüner, 1983) in the I- V (ω) in current biased mode) shows peaks at ω = pq ωD
V curve as Edc is swept at fixed Eac (see Fig. 11 which with finite width, rather than perfect Bragg peaks. This
depicts dV /dI versus the voltage V ). broadening is attributed to a distribution of velocities
23
∂t θ throughout the sample, which, in the language of ergodicity was introduced as a property of a measure-
time-crystalline order, implies the absence of perfect Sτ B. preserving deterministic system. The precise definition
Indeed, numerical simulations of driven 1D CDWs con- works differently in the stochastic case, but intuitively,
firm that finite temperature broadens the plateaus (Mali still captures whether a system inevitably “forgets” its
et al., 2012). Unfortunately, owing to thermal gradients initial condition. We direct the interested reader to
generated due to Ohmic heating, AC-driven CDWs do (Gielis and MacKay, 2000; Gray, 2001) for a discussion
not appear to be particularly well suited for quantita- of the technical aspects required to make the definitions
tively investigating the temperature dependence of the precise in the thermodynamic limit.
broadening at low T (see e.g. discussions in Sec. V.A). In the stochastic setting, a probability distribution ρs
Mathematically equivalent physics is found in the AC is an “invariant measure” of the dynamics if it is a fixed
Josephson effect (recall related discussions in Sec. I). In point of the stochastic update, Φ[ρs ] = ρs , i.e. it is a
the McCumber model (McCumber, 1968) for a resistively steady-state distribution. A stochastic system is said
and capacitively shunted Josephson junction (RCSJ), the to be ergodic if two properties hold: (1) the dynamics
superconducting phase difference φ across a Josephson have a unique steady state ρs and (2) the long-time be-
junction obeys the equation of motion havior of any initial state relaxes to this steady state,
limt→∞ ρ(t) = ρs . As in the closed, deterministic case,
ωp−2 φ̈ + ωc−1 φ̇ + sin φ = I(t)/Ic (20) Sτ B in the sense of Eq. (3) requires that Φ is non-ergodic.
p Otherwise, at long times, the m-possible orbits become
where ωp ≡ 2Ic /C, ωc ≡ 2Ic RN , and V /RN = φ̇/2RN indistinguishable and any oscillations will decay.
is the normal current across the junction; here, C is the Instead, a Sτ B phase will exhibit so-called “asymptotic
capacitance, I is the current across the Josephson junc- periodicity”(Lasota et al., 1984; Losson and Mackey,
tion, Ic is the current, and RN is the normal resistance. 1996). At long times the distribution relaxes to a convex
The situation considered in Sec. I occurs in experiments combinationPof m locally-distinguishable distributions
m
in which the junction is under-damped. The DC current ρp , ρ(t) → p=1 αp ρp+t which are cyclically permuted
is carried entirely by normal quasiparticles, IDC = In , under the evolution
P Φ[ρp ] = ρp+1 (with ρp+m = ρp ).
1
satisfying IDC /Ic = ωc−1 φ̇ or, simply, IDC = V /RN . An While ρs = m p ρp is a unique steady state, in the
AC supercurrent flows in response to the voltage drop generic case where the αp are unequal ρ(t) will continue
across the junction, IAC (t)/Ic = sin(2IDC RN t). How- to oscillate so that the limit limt→∞ ρ(t) = ρs fails to
ever, in this under-damped regime, the I − V curve is exist 3 .
hysteretic, and there is a branch in which the voltage The existence of true Sτ B in open dynamical systems
drop across the junction is zero so long as the DC cur- thus hinges on a far more fundamental question: can
rent is less than the critical current: IDC = Ic sin φ. This a locally interacting stochastic system generically break
vertical step in the IV curve is followed by a plateau at ergodicity? This line of questioning has a deep history in
I = Ic from V = 0 to V = Ic RN where it rejoins the the fields of non-linear dynamics, mathematical physics,
other branch. and computer science. The answer is intimately related
Returning to the general case of Eq. (20), we note to the stability of phase coexistence and phase transitions
that this equation of motion is formally equivalent to the in stochastic systems.
CDW [Eq. (19)], and by analogy one also expects steps We can illustrate the ideas at play by stripping away
in the I-V curve when the AC component of the current, the complexity of Hamiltonian dynamics and instead con-
I(t) = Idc + Iac cos(ωD t), is commensurate with rational sidering a classical spin system. For specificity we start in
harmonics of the DC voltage, ωJ = hφ̇i = 2eVdc . (The equilibrium with a classical Ising model HI [{σ}]. While
vertical step in the superconducting branch of the hys- HI itself does not define any dynamics, we may define
teretic I − V curve in the underdamped case is a trivial a “kinetic Ising model,” such as Glauber or Hasting-
version of such a step.) These “Shapiro steps” were first Metropolis rules, which update the spin configuration
observed at integer harmonics in single Josephson junc- σ → σ 0 according to a local conditional probability
tions (Shapiro, 1963). Later experiments (Benz et al.,
1990; Lee et al., 1991) on arrays of Josephson junctions
revealed subharmonic Shapiro steps. 3 In a deterministic system, P “ergodicity” governs the invariance
of Cesáro sums, limτ →∞ τ1 τt=1 O(Φ(t) (x)). In this case, Sτ B
implies that the m-fold iterated map Φ(m) is not ergodic, but
the map Φ itself generally is ergodic (though it is not mixing).
C. Ergodicity in open systems
In the stochastic case, however, ergodicity is usually defined to
imply the stronger
P form of convergence limτ →∞ = ρs , rather
At time scales t > e∆/kB T , an activated time crystal than limτ →∞ τ1 τt=1 Φ(t) [ρ0 ] = ρs . By this definition, Φ it-
1 Pm
loses memory of the initial condition which distinguishes self is non-ergodic. While the steady state ρs = m p=1 ρp
between the m-cycles of the Sτ B oscillation: its dynam- is unique, and one has limτ →∞ τ1 τt=1 Φ(t) [ρ0 ] = ρs , the limit
P
ics are ultimately ergodic. Earlier in this Colloquium, limτ →∞ ρ(t) does not exist.
24
Φ(σ 0 |σ) which obeys detailed balance with respect to implications for the generic behaviors obtainable by a
HI . Detailed balance ensures that the distribution ρs = classical computer perturbed by errors: e.g. Can reliable
e−βH /Z is a steady state. On a finite system at T > 0, systems emerge from unreliable components (Von Neu-
standard results guarantee that ρs is the unique steady mann, 1952)? For a review of these connections between
state with a finite relaxation time, so the process is er- theoretical computer science and more familiar notions
godic (Feller, 1957). However, the thermodynamic limit in non-equilibrium statistical physics, we refer the inter-
leads to richer possibilities. When h = 0 and T is be- ested reader to (Gray, 2001).
low the critical Ising temperature (e.g., when tuned to From the perspective of time crystals, the relevant
a first-order phase transition), the phase coexistence of question becomes the following: Do there exist CA which
the two magnetized phases implies the existence of two remain non-ergodic for generic small perturbations Φε ?
steady states, ρ(σ)+ = ρ(−σ)− , and the dynamics are And if so, might such a CA be the basis for sponta-
not ergodic. However, for any finite h phase coexistence neous time translation symmetry breaking? These ques-
is lost and the steady state is unique. This illustrates tions were answered in the affirmative in (Toom, 1980)
a general principle in equilibrium: ergodicity-breaking and (Bennett et al., 1990) Before diving into the first
requires tuning one or more parameters to a first-order question, we can sketch Bennett’s answer to the sec-
phase transition, and in this sense is fine-tuned (or re- ond. Suppose there does exist a local, non-ergodic CA
quires a symmetry beyond time-translation invariance). ΦCA with m-distinct steady states. For the simplicity of
We note that one route around this equilibrium no- discussion, we further assume that these states are ex-
go is to consider a spin Hamiltonian with power-law in- changed by a Zm symmetry, but (unlike a kinetic Ising
teraction of sufficiently slow fall-off, so that the energy or Potts model) we assume the non-ergodicity is stable
penalty for domain walls can be made to grow faster than to generic (possibly Zm breaking!) errors. We can then
the extensive energy of the symmetry-breaking external define a rotating version of the CA by composing ΦCA
field (Liggett, 2012). Or, taking this to the extreme, by with the action of Zm , Φ 2π m CA
≡ G2π/m ΦCA . By con-
considering models with all-to-all couplings in the spirit struction, ΦmCA = Φ m
2π ; thus, if ΦCA is proven to be
m CA
of a mean-field Hamiltonian, so that locality is lost en- stably non-ergodic, Φ 2π inherits this property and by
m CA
tirely (Lyu et al., 2020; Morita and Kaneko, 2006; Pizzi construction exhibits the “asymptotic periodicity” of a
et al., 2021b; Russomanno et al., 2017; Yang and Cai, time-crystal, cycling through the m-fixed steady states
2021). However, such models are not local in the usual of ΦCA .
sense: their thermodynamic properties are not extensive A particularly important point to emphasize is that
in system size, among other anomalies. So, while such (while convenient) it is not necessary to assume that ΦCA
models can be used as the basis for stabilizing even con- exhibits a Zm symmetry for stability. Indeed, if the non-
tinuous Sτ B (Kozin and Kyriienko, 2019), the starting ergodicity of ΦCA is stable to Zm -breaking perturbations,
point is rather different than the usual notion of locality one could just as well have started out with a perturbed
implied in the classification of phases. version of ΦCA which breaks the Zm symmetry. The con-
struction thus satisfies the important conceptual require-
ment that Sτ Bdoes not depend on any further internal
D. Probabilistic cellular automata symmetries.
Returning to the first question, the possibility of a lo-
Moving beyond equilibrium, one can consider more cal and generically non-ergodic PCA was a long-standing
general local Markov updates Φ(σ 0 |σ) of a discrete spin question for several decades (Dobruvsin et al., 1990;
system without reference to any particular H. Such Lebowitz et al., 1990; Toom, 1995) finally answered in
models are a type of probabilistic cellular automata the affirmative in the groundbreaking work of (Toom,
(PCA)(Dobruvsin et al., 1990); the continuous-time gen- 1980).
eralization of PCAs are dubbed “interacting particle sys-
tems” (Liggett, 2012). One particularly convenient way
to obtain a PCA is to start with a deterministic cellular E. An absolutely stable open time-crystal: the π-Toom
automata (CA) (Gutowitz, 1991) defined by a local up- model
date rule σ → ΦCA (σ), and then follow each CA update
with random spin flips at an “error rate” ε. For Marko- The Toom CA is a 2D binary spin model with a
vian errors we then obtain a Markov process which is a remarkably simple “north-east corner” (NEC) major-
perturbation of the deterministic CA, Φ = Φε ◦ ΦCA (we ity rule, defined as follows (Toom, 1980; ?). At each
note that mathematical results on PCA are even more step, each spin follows a majority vote amongst itself
general, allowing for non-Markovian errors). Such mod- and its two north and east neighbors: σx,y (t + 1) =
els have been of interest to the theoretical computer sci- maj(σx,y , σx+1,y , σx,y+1 ). This is a seemingly innocuous
ence community because CA are Turing complete, so the modification of the T = 0 nearest-neighbor kinetic Ising
study of the possible generic behaviors of a PCA have rule, in which each spin would follow a vote amongst its
YSICAL REVIEW LETTERS 12 AUGUST 1985 25
(demarcated 1 I I While the original Toom model exhibits m = 2 steady

rge islands of V) states, and hence is the basis for period m = 2 Sτ B, it
s shrink. A lg
or the vertical lQ two generalizes to any m > 2 (Bennett et al., 1990). Further-
tion of noise 0— phase more Toom-like models exist in all dimensions D > 1,
. Other runs region simply by stacking the 2D version. In fact, an extension
e p = q yield- in D = 3 provides the basis for error-corrected classical
nent describ- computing (Gács and Reif, 1988). The existence of stable
se amplitude -1 I
.2
0 Amplitude ergodicity breaking in 1D PCAs was only shown more re-
model differs FIG. 1. Phase diagram of the NEC system, for noise cently by Gács, Gács, 2001. Taken together, PCAs sup-
n having syn- FIG.parameters and q, ofwith
12 Phase pdiagram the amplitude = p + adapted
Toom model, from
q and bias port absolutely stable Sτ B of any period m ≥ 2 in all
= (p — q)/(p+ q).
ary runs in (Bennett and Grinstein, 1985). After each CA update, spins dimensions D ≥ 1.
ins are updat- randomly flip up or down with rate εp , εq respectively, with Is the full error-correcting capability of the Toom and
a fully asyn- “bias” (εp − εq )/(εp + εq ) and “amplitude” εp + εq . Within
We also
Gács models really necessary for stable Sτ B? For exam-
atively similarthe
phase diagram.region,
two-phase thereexplored
are two an analytically
distinct solvabledis-
steady-state
e, based on the recurrenceandrelation
tributions the dynamics are not ergodic. Unlike the Ising ple, in the construction Φ 2π m CA
≡ G2π/m ΦCA we may
m)2(1+m)+3(1 —q)(1
model, the— m)(1+can
system m) thus —q)(1+ m)one]/4,bit of informa-
+ (1“remember” instead take ΦCA to be the kinetic Ising model or its
t+1 as a function of that at time t. Here, too, noise.
tion even in the presence of biased the phase diagram was generalization to Zm . The ferromagnetic interactions of
ble zone, and the critical exponents were —, (for the two-phase region the Ising model are a mild form of error correction in that
they cause minority islands to shrink due surface tension.
iltonian, and four {north,
its free east,Wesouth,
energy. —
now west}
review neighbors. of an
Note, however,
the construction However, (Bennett et al., 1990) argued that this equilib-
an model for thatan arbitrary
due to the(in general the Toom ruleCA,
asymmetry, d-dimensional
spatial irreversible)
'2
cannot rium form of error correction is insufficient for stabilizing
e stable phasesbeover a set of finite measure in the parameter space of the
understood as minimizing the energy of any Hamil- m > 2 Sτ B.
X(1), . . . , X(t) of a d-dimensional CA can be viewed as configurations
tonian, and does not obey detailed
e boundary fixed at X(0), the initial state of the CA.balance (Grinstein
The probability The reason is that for m > 2, the chirality of the
et al., 1985).
ry may be expressed This asymmetry ensures that if an island of
as the product periodic evolution between steady states n → n + 1
P(X( t)/X(t the 1) ),minority—spin nucleates, the NEC-rule will cause the mod m implies that there is no symmetry which forbids
island to shrink linearly in time from the NE-direction
ditional probability for the CA to be in state X(l+ I) at time t+ I given
a stochastic force per unit length from acting on a do-
inward. This is far faster than the Ising model, where main wall between regions of n and n + 1 - type steady
the thermodynamic force on a domain wall depends on states (e.g. favoring n to grow at the expense of n + 1
)/X(i) ) ], its local curvature, and hence decays with the size of the mod m). Such a force is thus generic, and will drive mi-
island (Bennett
ar form of a Boltzmann factor (taking kT = 1): 1985).
and Grinstein, nority droplets to nucleate and grow. This force is coun-
The Toom PCA is then obtained by perturbing with teracted by the surface tension of the ferromagnetism,
X H(X (i + 1),biased
=0
),
X(i) errors in which, for example, spins flip up with which, however, decays with the local curvature. Thus
error rate εp , and down with error rate εq . The bias the former force will always win out for sufficiently large
Hamiltonian b = (εp − εq )/(εp + εq ) breaks the Ising symmetry. Re-
!
droplets, causing minority droplets to proliferate and de-
X(1), there
es. All prop- markably, etc. , represent finiteofvolume
time asteps
is nevertheless a synchronous
in {εp , εq }- stroy Sτ B at long times. The Toom model escapes this
as canonical- model, or discrete-time snapshots of an asynchronous
space in which two magnetized steady states persist de- reasoning because its error-correction effectively exerts a
onal system (master equation) model evolving in continuous time.
dimensional spiteHowever, the biasinthat prefers one case,
the synchronous over because
the other (Fig. 12).
transitions force on domain walls which is independent of their cur-
s free energy Rigorous results
at all sites in theprovesame that
time the
slice ergodicity breaking
are independent, the is vature, and can thus shrink minority droplets of any size.
initial condi- robust to essentially
equivalent Hamiltonian is of a generalized
any sufficiently small Ising
perturbation
form, (Bennett and Grinstein, 1985)
ows from the — even sum of local termscorrelated
being aa spatio-temporally —inP(y/x),
H(y, x) =one, or a noise However, the m = 2 case is an exception because there
ach X(i), the where P(y/x) is the conditional probability that a site
distribution which is not itself time-translation invari- is no handedness to the periodic evolution. Even if there
must be 1), will be in state y at time t+ 1, given that its neighbor-
dimensional ant hood (Berman
was inand Simon,
state 1988;t. Gacs,
x at time 2021; Toom, con-
The normalization 1980). is a force which favors (say) type n = 0 over n = 1, be-
namic limit We straint,
note thatviz. ,this
that last pointx, distinguishes
for each the sum over ythe “absolute
of P(y/x) cause the domains effectively switch type at each step,
be 1, restricts
system is a stability” of the the open GIM to a lower-dimensional
system π-Toom time crystal surface (to the net force on a domain wall averages to zero over one
ystem; there- be defined
in the parameter
in the next space of its coupling
sentence) from itsconstants,
closed systemon period, and islands can thus be shrunk via ferromagnetic
stable phases, which the free energy is zero.
Hamiltonian
counterparts (i.e. Sec. IV), where noise that breaks the
In the case of metastable phases the (d + 1)-
surface tension (Bennett et al., 1990). This implies the
underlying
dimensional free time
discrete translation
energy symmetry
per space-time would
site is not 0im- binary-spin π-Ising model can support Sτ B below the
ether but —ln(1
X(0), mediately destroy =I,
—I ) the time where crystal.
I is the nucleation rate Ising critical temperature, as was recently studied in de-
With Toom having done the difficult part, the π-Toom tail (Gambetta et al., 2019a).
659
model is then defined as the anti -majority NEC rule It is interesting to contrast this discussion with the
σx,y (t + 1) = anti-maj(σx,y , σx+1,y , σx,y+1 ). The π-Toom classical activated time crystal of Sec.V.A, which was not
model is a time-crystal, which is stable to arbitrary per- stable against the proliferation of minority islands even
turbations. Interestingly, the Sτ B of the π-Toom, model though m = 2. To see why, recall that the emergent
was already pointed out over two decades ago in (Ben- Ising degree of freedom was effectively embedded into a
nett et al., 1990), and later in Gielis and MacKay, 2000, larger continuous state space, e.g. σ = sgn(sin(Q)). A
where it was referred to as an example of a “type-G phase domain wall between σ = 1, −1 could thus be of two types
transition.” depending on whether Q wound clockwise or counter-
26
clockwise. Because this handedness is preserved under 1. Quantum many-body scars

the oscillation Q → Q + π, there is no symmetry which
forbids a domain wall from experiencing a net force which Interacting quantum systems can exhibit a weak break-
does not average to zero over m = 2 periods, even while down of thermalization, where certain initial conditions
the identities of the domains themselves are exchanged. exhibit persistent many-body revivals in time (Bernien
Does this imply that if the binary-spin π-Ising model is et al., 2017; Turner et al., 2018b). This phenomenon,
embedded into a model with additional degrees of free- dubbed quantum many-body scars, is associated with
dom, the Sτ B may not be generically stable? Or in- the presence of anomalous, non-thermal eigenstates; we
stead, may there be a parameter regime of models like the direct the interested reader to a recent review on the
Langevin-driven parametric oscillator arrays (12) which topic (Serbyn et al., 2021). From the perspective of time
exhibit true, rather than activated Sτ B? Only time will crystalline order, the presence of scar states can cause
tell. the system to undergo periodic entanglement and disen-
tanglement cycles following a quench (Ho et al., 2019;
Michailidis et al., 2020; Serbyn et al., 2021). In practice,
F. Open Hamiltonian and quantum dynamics however, quantum many-body scar states are quite frag-
ile since they rely upon the existence of a dynamically
While we have discussed the existence of stable time- disconnected subspace of non-thermalizing eigenstates;
crystals in the stripped-down setting of a PCA (Sec. V.D indeed, the presence of generic interactions is expected
and Sec. V.E), our original microscopic starting point to eventually lead to thermalization (Lin et al., 2020).
(Sec. V) was the periodically driven Langevin equation, The connection between quantum many-body scars
or its quantum analog, the Lindblad equation. An in- and time-crystalline behavior has recently been explored
triguing open question is the following: Do the rigor- in a Rydberg-based quantum simulation platform (Blu-
ous results on Sτ B in probabilistic cellular automata vstein et al., 2021; Maskara et al., 2021). In particular,
translate to these settings? Since PCA are motivated the experiments observed that the coherent revivals as-
as a coarse-grained description of the Hamiltonian world sociated with quantum many-body scars could be stabi-
around us, it certainly tempting to speculate that they lized by additional periodic driving. This periodic driv-
do. Two recent studies have shown how the π-Toom ing leads to both an increase in the lifetime of the scarred
model can be embedded into classical Langevin (Zhuang oscillations as well as the emergence of a period-doubled,
et al., 2021) or quantum Lindblad (McGinley et al., 2021) subharmonic response. While reminiscent of prethermal
dynamics, and in both cases there is strong numerical time crystalline order (Sec. IV.D), we note two impor-
evidence for true Sτ B. However, there is not yet a fully tant differences. First, the experiments are performed in
rigorous proof as in Toom and Gács work. While Toom- a regime where the driving frequency is of similar order as
like stability for a continuous-time version of a PCA was the local energy scales of the many-body system. Second,
proved in (Gács, 2001), the state space in that model is the subharmonic response exists only for Néel-like ini-
discrete. tial states (associated with the quantum scars) (Maskara
As we will discuss in the outlook, the answer to this et al., 2021). Looking forward, a number of doors are
question is a special case of a deeper one: Is error- opened by the possibility that periodic driving can en-
corrected computing physically realizable in this uni- hance the stability of non-ergodic dynamics. In addition
verse? to novel settings for observing time crystalline order, the
ability to parametrically control the lifetime of such order
could also lead to potential applications in areas such as
VI. OUTLOOK AND FUTURE DIRECTIONS quantum metrology and quantum information science.
A. New venues for time crystals

2. Stark time crystals
In Sec. III.C of this Colloquium, we introduced the
mantra: Where there is ergodicity breaking, there will be In the presence of an electric field, the wavefunction
time-crystals. This naturally suggests that new develop- of electrons in a material are localized to a region whose
ments in the ergodicity breaking of many-body systems size decreases as the field increases. In the absence of
— ranging from Hilbert space fragmentation (Sala et al., interactions, this phenomenon, which arises without dis-
2020; Yang et al., 2020) and shattering (Khemani et al., order, is known as Wannier-Stark localization (Wannier,
2020) to quantum scars (Serbyn et al., 2021; Turner et al., 1959). Recent theoretical and experimental investiga-
2018a) and Stark localization (Doggen et al., 2021; van tions have explored whether such localization can per-
Nieuwenburg et al., 2019; Schulz et al., 2019) — repre- sist in the presence of many-body interactions, leading
sents fertile ground for exploring novel formulations of to Stark MBL (Guo et al., 2020; Morong et al., 2021;
time-crystalline order. Scherg et al., 2021; Schulz et al., 2019).
ELSE, HO, and DUMITRESCU
27
1 statement t
(Kshetrimayum et al., 2020) discuss the possibility of <latexit sha1_base64="l9eImvYcFOKpzEDji/n9jPDeWb8=">AAAB6HicbVBNS8NAEJ3Ur1q/qh69LBbBU0lEqMeiF48t2A9oQ9lsJ+3azSbsboQS+gu8eFDEqz/Jm//GbZuDtj4YeLw3w8y8IBFcG9f9dgobm1vbO8Xd0t7+weFR+fikreNUMWyxWMSqG1CNgktsGW4EdhOFNAoEdoLJ3dzvPKHSPJYPZpqgH9GR5CFn1Fip6Q3KFbfqLkDWiZeTCuRoDMpf/WHM0gilYYJq3fPcxPgZVYYzgbNSP9WYUDahI+xZKmmE2s8Wh87IhVWGJIyVLWnIQv09kdFI62kU2M6ImrFe9ebif14vNeGNn3GZpAYlWy4KU0FMTOZfkyFXyIyYWkKZ4vZWwsZUUWZsNiUbgrf68jppX1U9t+o1ryv12zyOIpzBOVyCBzWowz00oAUMEJ7hFd6cR+fFeXc+lq0FJ585hT9wPn8Aep2MtQ==</latexit>
localized n
discrete time crystals protected from Floquet heating by
Local Observable hO(t)i

Prethermal regime What we
Stark many-body localization. They probe the existence 1 localization
Heff
of a Stark time crystal using numerical simulations of a Tr[Oe ]
<latexit sha1_base64="Av6/96qb5ouCpxMVwPgjCTwxhg0=">AAACHXicbVDJSgNBEO2JW4zbqEcvjUHwYpgRQY+il9yMkGgwM4aeTo026VnorhHDMD/ixV/x4kERD17Ev7GzCG4PCh7vVVFVL0il0Og4H1Zpanpmdq48X1lYXFpesVfXznSSKQ4tnshEtQOmQYoYWihQQjtVwKJAwnnQPx765zegtEjiJg5S8CN2FYtQcIZG6tp7XqgYz90ivyioh3CLKsqbqqCdE7jMd7wAkNF698uBMCwKv2tXnZozAv1L3AmpkgkaXfvN6yU8iyBGLpnWHddJ0c+ZQsElFBUv05Ay3mdX0DE0ZhFoPx99V9Ato/RomChTMdKR+n0iZ5HWgygwnRHDa/3bG4r/eZ0MwwM/F3GaIcR8vCjMJMWEDqOiPaGAoxwYwrgS5lbKr5mJC02gFROC+/vlv+Rst+Y6Nfd0r3p4NImjTDbIJtkmLtknh6ROGqRFOLkjD+SJPFv31qP1Yr2OW0vWZGad/ID1/gnPlaL7</latexit>
Z MBL,” alth
one-dimensional spin chain; focusing on two specific ini- the timesca
tial states (i.e. the ferromagnetic and anti-ferromagnetic quasiperiod
states), (Kshetrimayum et al., 2020) observe that for a perhaps ev
sufficiently strong linear potential, the states exhibit time future work
<latexit sha1_base64="x94/bDZzc8suBPYNlYtxnw5VJdc=">AAACJHicbVDLSgNBEJz1bXxFPXoZDIJewq4ICl5ELx4EFYwK2RB6J51kcGZ2mekVw+LHePFXvHjwgQcvfouTmIOvhoHq6i5qupJMSUdh+B6MjI6NT0xOTZdmZufmF8qLS+cuza3AmkhVai8TcKikwRpJUniZWQSdKLxIrg7684trtE6m5ox6GTY0dIxsSwHkqWZ5Nya8IauLo1SA4seJQ3sNXn3L467LQGARae0bBaajkB+v0waP7aBplithNRwU/wuiIaiwYZ00yy9xKxW5RkNCgXP1KMyoUYAlKbxjKc4dessr6GDdQwMaXaMYHHnL1zzT4u3U+meID9jvigK0cz2d+E0N1HW/Z33yv1k9p/ZOo5AmywmN+DJq54pTyvuJ8Za0KEj1PABhpf8rF12wIMjnWvIhRL9P/gvON6tRWI1Otyp7+8M4ptgKW2XrLGLbbI8dshNWY4LdsQf2xJ6D++AxeA3evlZHgqFmmf2o4OMTI0ylDA==</latexit>
crystalline order, which seems to be robust to π-pulse In this
imperfections (recall discussions from Sec. ??). How- phases of m
ever, when differences in the linear potential coincide above: pret
with integer multiples of the drive frequency, they ob- FIG. 132. Schematic
Prethermalization in quasiperiodically driventime
systems at lived) quas
FIG. depiction of a prethermal quasi-
serve the coherent self-destruction of time-crystalline be- high frequencies. Shown is a cartoon of the dynamics of a generic
crystal emerging from a quasiperiodically driven quantum sys-
havior. An advantage of the clean, disorder-free setting, traceless
tem. During localthe
observable
prethermal hOðtÞi under time
regime, localevolution
observablesby Eq. (6).
exhibit III. EM
is that one can imagine the possibility of order that spon- There are
discrete threequasicrystalline
time regimes. First, a brief
order,transient
whichregime, wheremelts
ultimately the BY
taneously breaks both spatial and time translation sym- local observable relaxes on a short timescale
at late times as the system Floquet heats to a featureless t r ∼ 1=J, where J is
the local energy scale of the system. Second, a prethermalization
metries. This theme has been studied in a number of infinite-temperature state. Figure adapted from (Else et al., Having
regime, where the system has locally equilibrated to a thermal
2020a).
contexts (Smits et al., 2018; Träger et al., 2021; Žlabys ensemble of an effective static Hamiltonian D, when viewed in outlined th
et al., 2021) and remains an active area of exploration. the rotating frame defined by PðtÞ. The evolution hOðtÞi in the analyze wh
laboratory frame shows a plateau around the prethermal value
π
these syste
Eq. 14)dashed
(black is adjusted
line), withtosmall
be time-quasiperiodic
close to a M -pulse. In cer-
oscillations of where a dir
B. Prethermalization beyond Floquet quantum systems amplitude
tain of ∼J=ω
scenarios, this(red dashednaturally
choice lines); here,leads
ω ¼ jωj is the
to an norm
effective This proce
of driving frequencies.
Hamiltonian, This regime
Heff , which exhibits lasts
aZupMtosymmetry
the long heating
(Pizzi Floquet en
ettime & ∼ exp½Cðω=JÞ
al., t2021c). Third, a final featureless infinite-
1. Classical prethermal discrete time crystals 1=ðmþϵÞ
When this %.ZM symmetry is spontaneously control inte
temperaturethe state, reached
willafter the systemat ahassubharmonic
fully heated. The
broken, system oscillate fre- the ground
In Sec. IV.D, we focused our attention on prethermal inset shows a zoom-in on the orange-shaded region in the
quency ∼ ωD /M throughout the prethermal regime. Us-
prethermal plateau.
is generate
time crystals in closed (i.e. unitarily evolving) quantum ing this strategy, Pizzi et al., 2021c construct a “phase from those
systems. However, as previously discussed, one of the diagram” for an ωD /4 PDTC in a three dimensional clas- Howeve
central features of Floquet prethermalization—i.e. expo- τ̃zi ≔spin
Pð0Þτ z †
sical i Pð0Þ , which is localized near site i. Under
system. high-freque
nentially slow heating—is also expected to occur in clas- “reverse
When one evolution”
utilizes defined as M in the strategy above,
a fractional long-time c
sical many-body systems (Hodson and Jarzynski, 2021; the dynamics can be even richer than the integer case. ible in so
†
Howell et al., 2019; Mori, 2018). This immediately begs Indeed, (Pizzi et al., τ̃zi ðtÞR ¼ UðtÞτ̃
2021a)
z
i UðtÞ this
explored ; scenario inð8Þthe equilibrium
the question: Can such systems also host prethermal time presence of long-range interactions (Huang et al., 2018; is effective
crystals (Hodson and Jarzynski, 2021; Howell et al., 2019; we obtain, using Eq. (6) and ½τzi ; D% ¼ 0, that significant
Kozin and Kyriienko, 2019; Russomanno et al., 2017),
Mori, 2018; Pizzi et al., 2021c; Ye et al., 2021)? tion. Preci
observingz signatures −itD of PDTC order at a variety of frac-
Recent work by Ye et al., 2021 and Pizzi et al., 2021c τ̃i ðtÞR ≈ PðtÞe τzi eitD PðtÞ† ¼ PðtÞτzi PðtÞ† : ð9Þ unitary fram
tional frequencies.
answer this question in the affirmative. A practical ad- time-evolu
vantage of the generalization to classical systems, is that This means that the motion of the τ̃zi ðtÞR is time quasi- the identity
it immediately enables the study of PDTCs in dimen- periodic, which in turn implies that there is an infinite are inheren
sions, d > 1. For example, Ye et al., 2021 study a near-
3.sequence
Prethermal in time
time quasi-crystals
whereby the operator returns arbitrarily regime.
neighbor-interacting classical Floquet spin model on the close (but never exactly) to the initial operator τ̃zi. Why In order
square lattice. Although the original Floquet evolution Building
should the upon
reversegeneralizations
evolution of the to l-bit τ̃ziPDTCs
higher-M
dressed be a twisted hig
does not exhibit any symmetries, Heff exhibits an emer- useful
and concept?PDTCs
fractional Consider the forward
(Matus and Sacha, Heisenberg
2019; timePizzi Floquet sy
al., 2021a,c;
etevolution Ye et al.,
of an operator 2021), anear
Oi localized natural Oi ðtÞ ≔to
question
site i via periodically
gent Z2 Ising symmetry. Unlike in the one dimensional
ask is† O
UðtÞ whether there can be non-trivial phases of mat- the context
case discussed in Sec. IV.D.2, in two dimensions, this i UðtÞ and ask how this operator spreads over time in
space.
ter In an ergodic
in isolated system,systems
many-body we expect thatthat areany localin
driven of matter c
Ising symmetry can be broken at finite temperatures. of TTSs in
a operator
way that spreads
is notgenerically
periodic ballistically,
in time. The or diffusively
focus of thisat
the slowest.
outlook In our present
subsection is on the case,closed
computing system, the overlap
prethermalof time-transl
2. Higher-M discrete time crystals Oi ðtÞ with
setting (Fig.the1)localized
where adressed τ̃zi , is replaced by
l-bitdrive
periodic allows us t
a quasiperiodic drive (Crowley et al., 2019; Gommers a simpler
TrðOet
et al., 2006; Ringot i ðtÞτ̃
al.,
z
¼ TrðORecent
i Þ 2000).
z
i τ̃i ðtÞR Þ;work has shown ð10Þ compared t
In both the quantum and classical settings, the emer-
in Sec. III C
gent symmetry in Heff is not restricted to just a Z2 sym- that such systems can host prethermal quasiperiodically-
reveals that this overlap varies quasiperiodically in time translation
metry and different subharmonic PDTCs (i.e. beyond pe- driven phases of matter, and in particular can give rise
rather than decaying to zero. We can interpret this as the limit in qu
riod doubling) can be realized (Giergiel et al., 2018, 2020; to discrete time quasi-crystals (Dumitrescu et al., 2018;
Pizzi et al., 2021a; Surace et al., 2019). The simplest Else et al., 2020a).
approach to doing this is to utilize a Floquet evolution Before jumping in however, we note that related phe-
where the “π-pulse” (see e.g. the discussion following nomena have been explored in non-linear dissipative021032-6 dy-
28
namical systems (see e.g. the coupled map lattice discus- 2019). As an example, (Giergiel et al., 2019) explore the
sions in Sec. II.A) driven at two incommensurate frequen- dynamics of an ultra-cold atomic ensemble bouncing be-
cies (Ding et al., 1989; Flicker, 2018; Held and Jeffries, tween two orthogonal harmonically oscillating mirrors.
1986; Romeiras and Ott, 1987; Sethna and Siggia, 1984). By tuning the bare frequencies of the unperturbed parti-
The spectral content of a periodic drive contains peaks cle motion, the system can reproduce fragments of the Fi-
only at a single drive frequency, ωD , and its harmonics bonacci sequence encoded via the bounces of the atomic
nωD , for integer n. By contrast, the spectral content of a ensemble off of the two mirrors.
quasiperiodic drive has peaks at integer linear combina-
tions: n1 ω1 + n2 ω2 + · · · + nm ωm , for some m > 1 num-
ber of incommensurate frequencies, {ω1 , · · · , ωm }. As in
Sec. IV.D, in order to discuss prethermal phases, it is C. Applications: from metrology to quantum information
necessary for the system to exhibit slow-heating (i.e. en- benchmarking
ergy absorption) from the drive. For quasiperiodic driv-
ing, this is strictly more challenging than the original
The periodic control underlying the discrete time crys-
Floquet setting, since the drive is (technically) able to
tal is, more generally, an indispensable tool in the con-
supply energy in arbitrary units. Nevertheless, an anal-
text of nuclear magnetic resonance spectroscopy (Bloem-
ogous slow-heating result was rigorously proven by Else
bergen et al., 1948, 1947; Waugh et al., 1968), quan-
et al., 2020a. Before heating occurs (Fig. 13), the dy-
tum information science (Biercuk et al., 2009; Khod-
namics of the quasiperiodically driven system are well-
jasteh and Lidar, 2005), and AMO-based quantum simu-
approximated (in a rotating frame) by an effective static
lation (Bloch et al., 2012; Goldman et al., 2014). Build-
Hamiltonian (in direct analogy to Sec. IV.D and Eqn. 16).
ing upon this connection, Choi et al., 2017b explored
In order to discuss the possibility of prethermal time
the possibility of engineering a Floquet system, where
quasicrystals (which are stable during the prethermal
quasienergy gaps can protect entangled states from static
time-scale, t < t∗ ), Else et al., 2020a begin by defin-
perturbations, while still ensuring their sensitivity to an
ing what an “order parameter” for such a phase would
oscillating signal. In some sense, this idea can be un-
look like. The subtlety is that a quasiperiodically driven
derstood as a generalization of spin echo spectroscopy,
system does not, strictly speaking, have any remaining
which utilizes many-body states; it is also related to the
time-translation symmetry to break. However, such a
conventional concept of using phases of matter with spon-
system can still exhibit a well-defined notion of a frac-
taneously broken symmetries for sensing (Bennett et al.,
tional or subharmonic frequency response (Fig. 13). In
2020; Frérot and Roscilde, 2018).
particular, an observable can respond in a quasiperiodic
manner with base frequencies: {e ω1 , · · · , ω
en }. When the Choi et al., 2017b investigated the use of a time-
ej are not harmonics of the original driving frequencies,
ω crystalline Floquet sequence to stabilize Schrodinger-cat
then the system is said to exhibit a fractional frequency states that are typically extremely fragile against local
response and discrete time quasicrystalline order (Else perturbations; they analyze a technique that allows for
et al., 2020a). Connecting to our previous discussions the enhancement of metrological bandwidth, while main-
about emergent symmetries in Heff , one can think of the taining the sensitivity and discuss an example in the con-
discrete time quasicrystal as emerging from the sponta- text of the precision measurement of AC magnetic fields.
neous breaking of a different finite Abelian group sym- More recently, these ideas have also inspired new many-
metry (i.e. which replaces the ZM symmetry discussed in body driving protocols aimed at circumventing the so-
Secs. III.B.2, VI.B.1). called interaction limit for quantum sensing (Zhou et al.,
Finally, let us end this outlook subsection by pointing 2020).
the reader to a number of related directions that fall out- Beyond metrology, the fact that signatures of time
side the prethermal context. Many of these connect to crystalline order have been observed in a diverse array of
the idea of quasiperiodic pattern formation in parametri- physical platforms (see e.g. Secs. IV.C.1, IV.C.2, IV.C.3),
cally driven systems. For example, in the context of two- suggests the possibility of cross platform benchmark-
frequency forcing, even when the frequencies are com- ing for the performance of near term quantum de-
mensurate (implying that they do not exhibit a quasiperi- vices (Preskill, 2018). This idea builds on the more
odic response in time), in certain scenarios, one can re- general thread that exploring non-equilibrium dynamical
alize stable Faraday-wave patterns that are analogous to phenomena may represent a particularly natural strat-
a two-dimensional quasicrystal (Besson et al., 1996; Ed- egy for verifying and validating noisy intermediate-scale
wards and Fauve, 1993; Silber et al., 2000). Various inquantum technology. Finally, suggestions for utilizing
carnations of a quasiperiodic response in time have also time crystalline order as a frequency standard or for
been explored in both theoretical proposals and experi- beyond-SQL (standard quantum limit) quantum sensing
ments on cold atomic systems (Autti et al., 2018; Chinzei have also been discussed (Lyu et al., 2020), although ex-
and Ikeda, 2020; Giergiel et al., 2019, 2018; Pizzi et al., plorations along this direction remain relatively nascent.
29
D. Concluding Remarks (Sec. IV.D) to the Floquet-MBL discrete time crystal

(Sec. IV.B); one expects a similar transition from the
As we have seen in this Colloquium, recent evidence activated DTC to the Toom-type time crystal, as one
points to (at least) two venues where infinitely-long- tunes a uniform anti-majority-vote rule toward the anti-
lived time crystals can exist: (i) Sτ B which is stable majority NEC rule.
to time-translation invariant perturbations can occur in One possibility is that the exponentially long-lived
periodically-driven closed 1D quantum systems in the time crystals’ lifetimes — e.g. scaling as ∼ AeωD /J in
presence of strong disorder (Sec. IV.B) and (ii) Sτ B the prethermal case (Sec. IV.D) or as ∼ Ae∆/T in the
which is stable to arbitrary perturbations can occur in activated case (recall that ∆ is the quasi-energy barrier
periodically-driven D ≥ 1 open systems, such as the defined in Sec. V.A) — exhibit a diverging prefactor. For
π-Toom model, whose dynamics effectively implement a example, as the transition is approached, A could diverge
form of error correction (Sec. V.E). These two flavors of as A ∼ |g − gc |−y , where g is the tuning parameter. An
time crystals evade ergodicity by vastly different means, alternate possibility is that the effective energy scale 1/J
either by strongly localizing the degrees of freedom or by or ∆ might diverge at the transition.
actively shrinking potentially ergodicity-generating fluc- A second set of open questions is whether a direct tran-
tuations. Between these two extremes lie two regimes sition is possible from a Floquet-MBL discrete time crys-
of time-crystalline behavior that are exponentially long- tal to a π-Toom or Gács-type time crystal (or something
lived: prethermal time crystals in closed systems (both continuously connected to it). To the authors of this
classical and quantum) and activated time crystals in Colloquium, this possibility seems a bit unlikely since
open systems. Although ergodicity is deferred in both such a transition would have to occur precisely when the
cases, it is inevitable in the long run (Palmer, 1982; Pe- coupling to a bath is zero, where the system goes from
tersen, 1989; Walters, 2000). closed to open. On the other hand, a Floquet-MBL dis-
The possibility of Sτ B in driven open systems can be crete time crystal is what we would create with a per-
understood as one consequence of a more radical form of fect quantum computer executing precisely the same set
ergodicity breaking: fault-tolerant computation. There of gates at each time step. Meanwhile, a π-Toom time
is a long history of understanding computation as a fun- crystal is what we might create with a noisy but fault-
damentally physical process, and the subsequent con- tolerant quantum computer in which error correction was
straints which arise from thermodynamics: “Comput- governed by a probabilistic cellular automaton. But since
ers may be thought of as engines for transforming free the former could be simulated by the latter, one might
energy into waste heat and mathematical work” (???). also logically expect them to be separated by no more
In this point of view, a time-crystal can be understood than a second-order phase transition at which the rate of
as a physical realization of a simple computer program: entropy production rises from zero with a discontinuous
“while true, apply a global NOT-gate.” If the program derivative.
can execute perfectly despite faulty (noisy) gates, and
with a physical implementation that relies only on local
interactions, the execution of such a program can be un- REFERENCES
derstood as a non-equilibrium “phase of matter,” and the
error threshold for fault-tolerance as a non-equilibrium Abanin, Dmitry, Wojciech De Roeck, Wen Wei Ho, and
François Huveneers (2017a), “A rigorous theory of many-
phase transition into a time-crystalline phase.
body prethermalization for periodically driven and closed
The work of (Gács, 2001; Gács and Reif, 1988) has quantum systems,” Communications in Mathematical
shown that fault-tolerant classical computation can in- Physics 354 (3), 809–827.
deed be realized as a locally-interacting autonomous pro- Abanin, Dmitry A, Ehud Altman, Immanuel Bloch, and
cess in the thermodynamic limit, and this work was later Maksym Serbyn (2019), “Colloquium: Many-body localiza-
used to prove that fault-tolerant quantum computing can tion, thermalization, and entanglement,” Reviews of Mod-
be as well (??). One could thus realize a time-crystalline ern Physics 91 (2), 021001.
Abanin, Dmitry A, Wojciech De Roeck, Wen Wei Ho,
phase by repeatedly running the program “NOT” on such
and François Huveneers (2017b), “Effective hamiltonians,
an error-corrected computer, either quantum or classi- prethermalization, and slow energy absorption in peri-
cal. In this sense, the existence of time-crystals in open odically driven many-body systems,” Physical Review B
systems is an elementary application of deeper results 95 (1), 014112.
regarding the physical realizability of error-correction in Abanin, Dmitry A, Wojciech De Roeck, and François Hu-
autonomous, locally interacting systems. veneers (2015), “Exponentially slow heating in periodi-
One open question is to understand how the transition cally driven many-body systems,” Physical review letters
115 (25), 256803.
from exponentially to infinitely long-lived time crystals Abanin, Dmitry A, Wojciech De Roeck, and François Huve-
precisely occurs. Indeed, as disorder is increased in closed neers (2016), “Theory of many-body localization in period-
1D quantum systems, one expects a sharp phase transi- ically driven systems,” Annals of Physics 372, 1–11.
tion to occur from the prethermal discrete time crystal Abanin, Dmitry A, and Zlatko Papić (2017), “Recent progress
30
in many-body localization,” Annalen der Physik 529 (7), (1994), “Lagrange equilibrium points in celestial mechanics
1700169. and nonspreading wave packets for strongly driven rydberg
Abobeih, MH, J Randall, CE Bradley, HP Bartling, electrons,” Physical review letters 73 (13), 1777.
MA Bakker, MJ Degen, M Markham, DJ Twitchen, Biercuk, Michael J, Hermann Uys, Aaron P VanDevender,
and TH Taminiau (2019), “Atomic-scale imaging of a Nobuyasu Shiga, Wayne M Itano, and John J Bollinger
27-nuclear-spin cluster using a quantum sensor,” Nature (2009), “Optimized dynamical decoupling in a model quan-
576 (7787), 411–415. tum memory,” Nature 458 (7241), 996–1000.
Albiez, Michael, Rudolf Gati, Jonas Fölling, Stefan Hun- Birkhoff, George D (1931), “Proof of the ergodic theorem,”
smann, Matteo Cristiani, and Markus K Oberthaler Proceedings of the National Academy of Sciences 17 (12),
(2005), “Direct observation of tunneling and nonlinear self- 656–660.
trapping in a single bosonic josephson junction,” Physical Birkhoff, George David (1927), Dynamical systems, Vol. 9
review letters 95 (1), 010402. (American Mathematical Soc.).
Alekseev, Anton, Jin Dai, and Antti J Niemi (2020), “Prove- Bloch, Immanuel, Jean Dalibard, and Sylvain Nascimbene
nance of classical hamiltonian time crystals,” Journal of (2012), “Quantum simulations with ultracold quantum
High Energy Physics 2020 (8), 1–21. gases,” Nature Physics 8 (4), 267–276.
Arute, Frank, Kunal Arya, Ryan Babbush, Dave Bacon, Bloembergen, Nicolaas, Edward Mills Purcell, and Robert V
Joseph C Bardin, Rami Barends, Rupak Biswas, Sergio Pound (1948), “Relaxation effects in nuclear magnetic res-
Boixo, Fernando GSL Brandao, David A Buell, et al. onance absorption,” Physical review 73 (7), 679.
(2019), “Quantum supremacy using a programmable su- Bloembergen, Nicolaas, EM Purcell, and RV Pound (1947),
perconducting processor,” Nature 574 (7779), 505–510. “Nuclear magnetic relaxation,” Nature 160 (4066), 475–
Autti, S, VB Eltsov, and GE Volovik (2018), “Observation of 476.
a time quasicrystal and its transition to a superfluid time Bluvstein, Dolev, Ahmed Omran, Harry Levine, Alexan-
crystal,” Physical review letters 120 (21), 215301. der Keesling, Giulia Semeghini, Sepehr Ebadi, Tout T
Autti, Samuli, Petri J Heikkinen, Jere T Mäkinen, Grig- Wang, Alexios A Michailidis, Nishad Maskara, Wen Wei
ori E Volovik, Vladislav V Zavjalov, and Vladimir B Eltsov Ho, et al. (2021), “Controlling quantum many-body dy-
(2021), “Ac josephson effect between two superfluid time namics in driven rydberg atom arrays,” Science 371 (6536),
crystals,” Nature Materials 20 (2), 171–174. 1355–1359.
Benjamin, Thomas Brooke, and Fritz Joseph Ursell (1954), Bohr, Tomas, Per Bak, and Mogens Hφgh Jensen (1984),
“The stability of the plane free surface of a liquid in ver- “Transition to chaos by interaction of resonances in dissipa-
tical periodic motion,” Proceedings of the Royal Society tive systems. ii. josephson junctions, charge-density waves,
of London. Series A. Mathematical and Physical Sciences and standard maps,” Physical review A 30 (4), 1970.
225 (1163), 505–515. Booker, Cameron, Berislav Buča, and Dieter Jaksch (2020),
Bennett, Charles H, G. Grinstein, Yu He, C. Jayaprakash, and “Non-stationarity and dissipative time crystals: spectral
David Mukamel (1990), “Stability of temporally periodic properties and finite-size effects,” New Journal of Physics
states of classical many-body systems,” Phys. Rev. A 41, 22 (8), 085007.
1932–1935. Bordia, Pranjal, Henrik Lüschen, Ulrich Schneider, Michael
Bennett, Charles H, and Geoffrey Grinstein (1985), “Role of Knap, and Immanuel Bloch (2017), “Periodically driving
irreversibility in stabilizing complex and nonergodic behav- a many-body localized quantum system,” Nature Physics
ior in locally interacting discrete systems,” Physical review 13 (5), 460–464.
letters 55 (7), 657. Brown, Stuart E, George Mozurkewich, and George Grüner
Bennett, Robert, David Steinbrecht, Yaroslav Gorbachev, (1984), “Subharmonic shapiro steps and devil’s-staircase
and Stefan Yoshi Buhmann (2020), “Symmetry breaking behavior in driven charge-density-wave systems,” Physical
in a condensate of light and its use as a quantum sensor,” review letters 52 (25), 2277.
Phys. Rev. Applied 13, 044031. Bruno, Patrick (2013), “Impossibility of spontaneously rotat-
Benz, SP, MS Rzchowski, M Tinkham, and CJ Lobb (1990), ing time crystals: a no-go theorem,” Physical review letters
“Fractional giant shapiro steps and spatially correlated 111 (7), 070402.
phase motion in 2d josephson arrays,” Physical review let- Buča, Berislav, and Dieter Jaksch (2019), “Dissipation in-
ters 64 (6), 693. duced nonstationarity in a quantum gas,” Physical review
Berges, Jürgen, Sz Borsányi, and Christof Wetterich (2004), letters 123 (26), 260401.
“Prethermalization,” Physical review letters 93 (14), Buča, Berislav, Joseph Tindall, and Dieter Jaksch (2019),
142002. “Non-stationary coherent quantum many-body dynamics
Berman, Piotr, and J’anos Simon (1988), “Investigations of through dissipation,” Nature Communications 10 (1), 1–6.
fault-tolerant networks of computers,” in Proceedings of the Buchleitner, Andreas, and Dominique Delande (1995),
twentieth annual ACM symposium on Theory of computing, “Nondispersive electronic wave packets in multiphoton pro-
pp. 66–77. cesses,” Physical review letters 75 (8), 1487.
Bernien, Hannes, Sylvain Schwartz, Alexander Keesling, Buchleitner, Andreas, Dominique Delande, and Jakub Za-
Harry Levine, Ahmed Omran, Hannes Pichler, Soonwon krzewski (2002), “Non-dispersive wave packets in period-
Choi, Alexander S Zibrov, Manuel Endres, Markus Greiner, ically driven quantum systems,” Physics reports 368 (5),
et al. (2017), “Probing many-body dynamics on a 51-atom 409–547.
quantum simulator,” Nature 551 (7682), 579–584. Bukov, Marin, Luca D’Alessio, and Anatoli Polkovnikov
Besson, Thomas, W Stuart Edwards, and Laurette S Tucker- (2015), “Universal high-frequency behavior of periodically
man (1996), “Two-frequency parametric excitation of sur- driven systems: from dynamical stabilization to floquet en-
face waves,” Physical Review E 54 (1), 507. gineering,” Advances in Physics 64 (2), 139–226.
Bialynicki-Birula, Iwo, Maciej Kaliński, and JH Eberly Bunimovich, Leonid A, and Ya G Sinai (1988), “Spacetime
31
chaos in coupled map lattices,” Nonlinearity 1 (4), 491. hilbert-space shattering,” Physical Review B 103 (10),
Chi, Huan-Hang, and Hong-Jian He (2014), “Single-valued L100202.
hamiltonian via legendre–fenchel transformation and time Dogra, Nishant, Manuele Landini, Katrin Kroeger, Lorenz
translation symmetry,” Nuclear Physics B 885, 448–458. Hruby, Tobias Donner, and Tilman Esslinger (2019),
Chinzei, Koki, and Tatsuhiko N Ikeda (2020), “Time crys- “Dissipation-induced structural instability and chiral dy-
tals protected by floquet dynamical symmetry in hubbard namics in a quantum gas,” Science 366 (6472), 1496–1499.
models,” Physical Review Letters 125 (6), 060601. Dumitrescu, Philipp T, Romain Vasseur, and Andrew C Pot-
Chirikov, Boris V (1979), “A universal instability of many- ter (2018), “Logarithmically slow relaxation in quasiperiod-
dimensional oscillator systems,” Physics reports 52 (5), ically driven random spin chains,” Physical review letters
263–379. 120 (7), 070602.
Choi, Jae-yoon, Sebastian Hild, Johannes Zeiher, Peter D’Alessio, Luca, and Anatoli Polkovnikov (2013), “Many-
Schauß, Antonio Rubio-Abadal, Tarik Yefsah, Vedika Khe- body energy localization transition in periodically driven
mani, David A Huse, Immanuel Bloch, and Christian Gross systems,” Annals of Physics 333, 19–33.
(2016), “Exploring the many-body localization transition in Eckardt, André, and Egidijus Anisimovas (2015), “High-
two dimensions,” Science 352 (6293), 1547–1552. frequency approximation for periodically driven quantum
Choi, Joonhee, Hengyun Zhou, Helena S Knowles, Renate systems from a floquet-space perspective,” New journal of
Landig, Soonwon Choi, and Mikhail D Lukin (2020), “Ro- physics 17 (9), 093039.
bust dynamic hamiltonian engineering of many-body spin Edwards, W Stuart, and S Fauve (1994), “Patterns and quasi-
systems,” Physical Review X 10 (3), 031002. patterns in the faraday experiment,” Journal of Fluid Me-
Choi, Soonwon, Joonhee Choi, Renate Landig, Georg Kucsko, chanics 278, 123–148.
Hengyun Zhou, Junichi Isoya, Fedor Jelezko, Shinobu On- Edwards, W Stuart, and Stéphane Fauve (1993), “Parametri-
oda, Hitoshi Sumiya, Vedika Khemani, et al. (2017a), “Ob- cally excited quasicrystalline surface waves,” Physical Re-
servation of discrete time-crystalline order in a disordered view E 47 (2), R788.
dipolar many-body system,” Nature 543 (7644), 221–225. Else, Dominic V, Bela Bauer, and Chetan Nayak (2016), “Flo-
Choi, Soonwon, Norman Y Yao, and Mikhail D Lukin quet time crystals,” Physical review letters 117 (9), 090402.
(2017b), “Quantum metrology based on strongly correlated Else, Dominic V, Bela Bauer, and Chetan Nayak (2017),
matter,” arXiv preprint arXiv:1801.00042. “Prethermal phases of matter protected by time-
Choudhury, A Ghose, and Partha Guha (2019), “Branched translation symmetry,” Physical Review X 7 (1), 011026.
hamiltonians and time translation symmetry breaking in Else, Dominic V, Wen Wei Ho, and Philipp T Dumitrescu
equations of the liénard type,” Modern Physics Letters A (2020a), “Long-lived interacting phases of matter protected
34 (32), 1950263. by multiple time-translation symmetries in quasiperiodi-
Cornfeld, Isaac P, Sergej V Fomin, and Yakov Grigorevich cally driven systems,” Physical Review X 10 (2), 021032.
Sinai (2012), Ergodic theory, Vol. 245 (Springer Science & Else, Dominic V, Christopher Monroe, Chetan Nayak, and
Business Media). Norman Y Yao (2020b), “Discrete time crystals,” Annual
Crowley, Philip JD, Ivar Martin, and Anushya Chan- Review of Condensed Matter Physics 11, 467–499.
dran (2019), “Topological classification of quasiperiodi- Fan, Chu-hui, D Rossini, Han-Xiao Zhang, Jin-Hui Wu, M Ar-
cally driven quantum systems,” Physical Review B 99 (6), toni, and GC La Rocca (2020), “Discrete time crystal in a
064306. finite chain of rydberg atoms without disorder,” Physical
Dai, Jin, Antti J Niemi, and Xubiao Peng (2020), “Classical Review A 101 (1), 013417.
hamiltonian time crystals–general theory and simple exam- Faraday, Michael (1831), “Xvii. on a peculiar class of acousti-
ples,” New Journal of Physics 22 (8), 085006. cal figures; and on certain forms assumed by groups of par-
Dai, Jin, Antti J Niemi, Xubiao Peng, and Frank Wilczek ticles upon vibrating elastic surfaces,” Philosophical trans-
(2019), “Truncated dynamics, ring molecules, and mechan- actions of the Royal Society of London (121), 299–340.
ical time crystals,” Physical Review A 99 (2), 023425. Feller, William (1957), “An introduction to probability theory
D’Alessio, Luca, and Marcos Rigol (2014), “Long-time Be- and its applications,” .
havior of Isolated Periodically Driven Interacting Lattice Fisher, Michael E, Shang-keng Ma, and BG Nickel (1972),
Systems,” Physical Review X 4 (4), 041048. “Critical exponents for long-range interactions,” Physical
De Roeck, Wojciech, and François Huveneers (2017), “Sta- Review Letters 29 (14), 917.
bility and instability towards delocalization in many-body Flatté, Michael E, and Martin Holthaus (1996), “Classical
localization systems,” Physical Review B 95 (15), 155129. and quantum dynamics of a periodically driven particle in
De Roeck, Wojciech, Francois Huveneers, Markus Müller, and a triangular well,” annals of physics 245 (1), 113–146.
Mauro Schiulaz (2016), “Absence of many-body mobility Flicker, Felix (2018), “Time quasilattices in dissipative dy-
edges,” Physical Review B 93 (1), 014203. namical systems,” SciPost Physics 5 (1).
Deutsch, J M (1991), “Quantum statistical mechanics in a Floquet, G (1883), “Sur les équations différentielles linéaires
closed system,” Phys. Rev. A 43 (4), 2046–2049. à coefficients périodiques,” Annales scientifiques de l’École
Ding, Mingzhou, Celso Grebogi, and Edward Ott (1989), Normale Supérieure 2e série, 12, 47–88.
“Evolution of attractors in quasiperiodically forced sys- Frérot, Irénée, and Tommaso Roscilde (2018), “Quantum crit-
tems: From quasiperiodic to strange nonchaotic to ical metrology,” Phys. Rev. Lett. 121, 020402.
chaotic,” Physical review A 39 (5), 2593. Frey, Philipp, and Stephan Rachel (2022), “Realization
Dobruvsin, Roland L, RL Dobrushin, VI Kriukov, and of a discrete time crystal on 57 qubits of a quan-
AL Toom (1990), Stochastic cellular systems: ergodicity, tum computer,” Science Advances 8 (9), eabm7652,
memory, morphogenesis (Manchester University Press). https://www.science.org/doi/pdf/10.1126/sciadv.abm7652.
Doggen, Elmer VH, Igor V Gornyi, and Dmitry G Polyakov Fulton, William, and Joe Harris (2013), Representation the-
(2021), “Stark many-body localization: Evidence for
32
ory: a first course, Vol. 129 (Springer Science & Business physics in time crystals,” New Journal of Physics 22 (7),
Media). 075003.
Gács, Peter (2001), “Reliable cellular automata with self- Guo, Qiujiang, Chen Cheng, Hekang Li, Shibo Xu, Pengfei
organization,” Journal of Statistical Physics 103 (1), 45– Zhang, Zhen Wang, Chao Song, Wuxin Liu, Wenhui Ren,
267. Hang Dong, et al. (2020), “Stark many-body localization
Gacs, Peter (2021), “A new version of toom’s proof,” arXiv on a superconducting quantum processor,” arXiv preprint
preprint arXiv:2105.05968. arXiv:2011.13895.
Gács, Peter, and John Reif (1988), “A simple three- Gutowitz, Howard (1991), Cellular automata: Theory and ex-
dimensional real-time reliable cellular array,” Journal of periment (MIT press).
Computer and System Sciences 36 (2), 125–147. Hannaford, Peter andSacha, Krzysztof (2022), “Condensed
Gambetta, FM, F Carollo, Achilleas Lazarides, I Lesanovsky, matter physics in big discrete time crystals,” arXiv preprint
and JP Garrahan (2019a), “Classical stochastic discrete arXiv:2202.05544.
time crystals,” Physical Review E 100 (6), 060105. Hayashi, Chihiro (1953), “Subharmonic oscillations in nonlin-
Gambetta, FM, F Carollo, M Marcuzzi, JP Garrahan, and ear systems,” Journal of Applied Physics 24 (5), 521–529.
I Lesanovsky (2019b), “Discrete time crystals in the ab- Held, GA, and Carson Jeffries (1986), “Quasiperiodic tran-
sence of manifest symmetries or disorder in open quantum sitions to chaos of instabilities in an electron-hole plasma
systems,” Physical review letters 122 (1), 015701. excited by ac perturbations at one and at two frequencies,”
Gielis, G, and RS MacKay (2000), “Coupled map lattices with Physical review letters 56 (11), 1183.
phase transition,” Nonlinearity 13 (3), 867. Ho, Wen Wei, Soonwon Choi, Mikhail D Lukin, and Dmitry A
Giergiel, Krzysztof, Arkadiusz Kuroś, and Krzysztof Sacha Abanin (2017), “Critical time crystals in dipolar systems,”
(2019), “Discrete time quasicrystals,” Physical Review B Physical review letters 119 (1), 010602.
99 (22), 220303. Ho, Wen Wei, Soonwon Choi, Hannes Pichler, and Mikhail D
Giergiel, Krzysztof, Artur Miroszewski, and Krzysztof Sacha Lukin (2019), “Periodic orbits, entanglement, and quantum
(2018), “Time crystal platform: from quasicrystal struc- many-body scars in constrained models: Matrix product
tures in time to systems with exotic interactions,” Physical state approach,” Physical review letters 122 (4), 040603.
Review Letters 120 (14), 140401. Hodson, Wade, and Christopher Jarzynski (2021), “Energy
Giergiel, Krzysztof, Tien Tran, Ali Zaheer, Arpana Singh, An- diffusion and absorption in chaotic systems with rapid pe-
drei Sidorov, Krzysztof Sacha, and Peter Hannaford (2020), riodic driving,” Physical Review Research 3 (1), 013219.
“Creating big time crystals with ultracold atoms,” New Holthaus, Martin (1995), “On the classical-quantum cor-
Journal of Physics 22 (8), 085004. respondence for periodically time dependent systems,”
Givental, Alexander B, Boris A. Khesin, Jerrold E. Mars- Chaos, Solitons & Fractals 5 (7), 1143–1167.
den, Alexander N. Varchenko, Victor A. Vassiliev, Oleg Ya. Holthaus, Martin, and Michael E Flatté (1994), “Subhar-
Viro, and Vladimir M. Zakalyukin, Eds. (2009), “Proof of monic generation in quantum systems,” Physics Letters A
a theorem of a. n. kolmogorov on the invariance of quasi- 187 (2), 151–156.
periodic motions under small perturbations of the hamil- Howell, Owen, Phillip Weinberg, Dries Sels, Anatoli
tonian,” in Collected Works: Representations of Func- Polkovnikov, and Marin Bukov (2019), “Asymptotic
tions, Celestial Mechanics and KAM Theory, 1957–1965 prethermalization in periodically driven classical spin
(Springer Berlin Heidelberg, Berlin, Heidelberg) pp. 267– chains,” Phys. Rev. Lett. 122 (1), 010602.
294. Huang, Yi, Tongcang Li, and Zhang-qi Yin (2018),
Goldman, Nathan, G Juzeliūnas, Patrik Öhberg, and Ian B “Symmetry-breaking dynamics of the finite-size lipkin-
Spielman (2014), “Light-induced gauge fields for ultracold meshkov-glick model near ground state,” Physical Review
atoms,” Reports on Progress in Physics 77 (12), 126401. A 97 (1), 012115.
Gommers, R, S Denisov, and F Renzoni (2006), “Quasiperi- Iemini, F, A Russomanno, J Keeling, M Schirò, M Dalmonte,
odically driven ratchets for cold atoms,” Physical review and R Fazio (2018), “Boundary time crystals,” Physical
letters 96 (24), 240604. review letters 121 (3), 035301.
Gong, Zongping, Ryusuke Hamazaki, and Masahito Ueda Imbrie, John Z (2016), “On many-body localization for quan-
(2018), “Discrete time-crystalline order in cavity and cir- tum spin chains,” Journal of Statistical Physics 163 (5),
cuit qed systems,” Physical review letters 120 (4), 040404. 998–1048.
Gray, Lawrence F (2001), “A reader’s guide to gacs’s “positive Inoue, J-I, and A. Tanaka (2010), “Photoinduced Transition
rates” paper,” Journal of Statistical Physics 103 (1), 1–44. between Conventional and Topological Insulators in Two-
Gring, Michael, Maximilian Kuhnert, Tim Langen, Takuya Dimensional Electronic Systems,” Physical Review Letters
Kitagawa, Bernhard Rauer, Matthias Schreitl, Igor Mazets, 105 (1), 017401, arXiv:1006.5283 [cond-mat.mtrl-sci].
D Adu Smith, Eugene Demler, and Jörg Schmiedmayer Ippoliti, Matteo, Kostyantyn Kechedzhi, Roderich Moess-
(2012), “Relaxation and prethermalization in an isolated ner, SL Sondhi, and Vedika Khemani (2020), “Many-body
quantum system,” Science 337 (6100), 1318–1322. physics in the nisq era: quantum programming a discrete
Grinstein, G, C Jayaprakash, and Yu He (1985), “Statisti- time crystal,” arXiv preprint arXiv:2007.11602.
cal mechanics of probabilistic cellular automata,” Physical Jackson, Russell K, and Michael I Weinstein (2004), “Geo-
review letters 55 (23), 2527. metric analysis of bifurcation and symmetry breaking in a
Grüner, G, A Zawadowski, and PM Chaikin (1981), “Non- gross—pitaevskii equation,” Journal of Statistical Physics
linear conductivity and noise due to charge-density-wave 116 (1), 881–905.
depinning in nb se 3,” Physical Review Letters 46 (7), 511. Jiang, L, T. Kitagawa, J. Alicea, A. R. Akhmerov, D. Pekker,
Grüner, George (1988), “The dynamics of charge-density G. Refael, J. I. Cirac, E. Demler, M. D. Lukin, and
waves,” Reviews of modern physics 60 (4), 1129. P. Zoller (2011), “Majorana Fermions in Equilibrium and in
Guo, Lingzhen, and Pengfei Liang (2020), “Condensed matter Driven Cold-Atom Quantum Wires,” Physical Review Let-
33
ters 106 (22), 220402, arXiv:1102.5367 [cond-mat.quant- Kjaergaard, Morten, Mollie E Schwartz, Jochen Braumüller,
gas]. Philip Krantz, Joel I-J Wang, Simon Gustavsson, and
Kagan, Yu, and MI Klinger (1974), “Theory of quantum dif- William D Oliver (2020), “Superconducting qubits: Cur-
fusion of atoms in crystals,” Journal of Physics C: Solid rent state of play,” Annual Review of Condensed Matter
State Physics 7 (16), 2791. Physics 11, 369–395.
Kaneko, Kunihiko (1984), “Period-doubling of kink-antikink Kolmogorov, Andrey Nikolaevich (1954), “On conservation of
patterns, quasiperiodicity in antiferro-like structures and conditionally periodic motions for a small change in hamil-
spatial intermittency in coupled logistic lattice: Towards a ton’s function,” in Dokl. Akad. Nauk SSSR, Vol. 98, pp.
prelude of a “field theory of chaos”,” Progress of Theoreti- 527–530.
cal Physics 72 (3), 480–486. Kongkhambut, Phatthamon, Jim Skulte, Ludwig Mathey,
Kaneko, Kunihiko (1992), “Overview of coupled map lat- Jayson G Cosme, Andreas Hemmerich, and Hans Keßler
tices,” Chaos: An Interdisciplinary Journal of Nonlinear (2022), “Observation of a continuous time crystal,” Science
Science 2 (3), 279–282. , eabo3382.
Kaneko, Kunihiko, and Tetsuro Konishi (1987), “Transition, Kontorova, T, and Jl Frenkel (1938), “On the theory of plastic
ergodicity and lyapunov spectra of hamiltonian dynamical deformation and twinning. ii.” Zh. Eksp. Teor. Fiz. 8, 1340–
systems,” Journal of the Physical Society of Japan 56 (9), 1348.
2993–2996. Kosterlitz, JM (1976), “Phase transitions in long-range ferro-
Kapral, Raymond (1985), “Pattern formation in two- magnetic chains,” Physical Review Letters 37 (23), 1577.
dimensional arrays of coupled, discrete-time oscillators,” Kozin, Valerii K, and Oleksandr Kyriienko (2019), “Quantum
Physical Review A 31 (6), 3868. time crystals from hamiltonians with long-range interac-
Katok, Anatole, and Boris Hasselblatt (1997), Introduction tions,” Physical review letters 123 (21), 210602.
to the modern theory of dynamical systems, 54 (Cambridge Kshetrimayum, A, J Eisert, and DM Kennes (2020), “Stark
university press). time crystals: Symmetry breaking in space and time,”
Keßler, Hans, Jayson G Cosme, Christoph Georges, Ludwig Physical Review B 102 (19), 195116.
Mathey, and Andreas Hemmerich (2020), “From a contin- Kucsko, Georg, Soonwon Choi, Joonhee Choi, Peter C Mau-
uous to a discrete time crystal in a dissipative atom-cavity rer, Hengyun Zhou, Renate Landig, Hitoshi Sumiya, Shi-
system,” New Journal of Physics 22 (8), 085002. nobu Onoda, Junichi Isoya, Fedor Jelezko, et al. (2018),
Keßler, Hans, Jayson G Cosme, Michal Hemmerling, Lud- “Critical thermalization of a disordered dipolar spin sys-
wig Mathey, and Andreas Hemmerich (2019), “Emergent tem in diamond,” Physical review letters 121 (2), 023601.
limit cycles and time crystal dynamics in an atom-cavity Kumar, Krishna, and Laurette S Tuckerman (1994), “Para-
system,” Physical Review A 99 (5), 053605. metric instability of the interface between two fluids,” Jour-
Keßler, Hans, Phatthamon Kongkhambut, Christoph nal of Fluid Mechanics 279, 49–68.
Georges, Ludwig Mathey, Jayson G. Cosme, and Andreas Kuroś, Arkadiusz, Rick Mukherjee, Weronika Golletz, Fred-
Hemmerich (2021), “Observation of a dissipative time crys- eric Sauvage, Krzysztof Giergiel, Florian Mintert, and
tal,” Phys. Rev. Lett. 127, 043602. Krzysztof Sacha (2020), “Phase diagram and optimal con-
von Keyserlingk, Curt W, Vedika Khemani, and Shivaji L trol for n-tupling discrete time crystal,” New Journal of
Sondhi (2016), “Absolute stability and spatiotemporal Physics 22 (9), 095001.
long-range order in floquet systems,” Physical Review B Kuwahara, Tomotaka, Takashi Mori, and Keiji Saito (2016),
94 (8), 085112. “Floquet–magnus theory and generic transient dynamics in
von Keyserlingk, Curt W, and Shivaji L Sondhi (2016a), periodically driven many-body quantum systems,” Annals
“Phase structure of one-dimensional interacting flo- of Physics 367, 96–124.
quet systems. i. abelian symmetry-protected topological Kyprianidis, Antonis, Francisco Machado, William Morong,
phases,” Physical Review B 93 (24), 245145. Patrick Becker, Kate S Collins, Dominic V Else, Lei Feng,
von Keyserlingk, Curt W, and Shivaji L Sondhi (2016b), Paul W Hess, Chetan Nayak, Guido Pagano, et al. (2021),
“Phase structure of one-dimensional interacting floquet “Observation of a prethermal discrete time crystal,” Sci-
systems. ii. symmetry-broken phases,” Physical Review B ence 372 (6547), 1192–1196.
93 (24), 245146. Lasota, A, T-Y Li, and JA Yorke (1984), “Asymptotic peri-
Khemani, Vedika, Michael Hermele, and Rahul Nandk- odicity of the iterates of markov operators,” Transactions
ishore (2020), “Localization from hilbert space shattering: of the American Mathematical Society 286 (2), 751–764.
From theory to physical realizations,” Physical Review B Lasota, Andrzej, and Michael C Mackey (2013), Chaos, frac-
101 (17), 174204. tals, and noise: stochastic aspects of dynamics, Vol. 97
Khemani, Vedika, Achilleas Lazarides, Roderich Moess- (Springer Science & Business Media).
ner, and Shivaji L Sondhi (2016), “Phase structure of Lazarides, Achilleas, Arnab Das, and Roderich Moessner
driven quantum systems,” Physical review letters 116 (25), (2015), “Fate of many-body localization under periodic
250401. driving,” Physical review letters 115 (3), 030402.
Khemani, Vedika, Roderich Moessner, and SL Sondhi Lazarides, Achilleas, Sthitadhi Roy, Francesco Piazza, and
(2019), “A brief history of time crystals,” arXiv preprint Roderich Moessner (2020), “Time crystallinity in dissipa-
arXiv:1910.10745. tive floquet systems,” Physical Review Research 2 (2),
Khemani, Vedika, Roderich Moessner, and SL Sondhi (2021), 022002.
“A comment on” discrete time crystals: rigidity, criticality, Lebowitz, Joel L, Christian Maes, and Eugene R Speer (1990),
and realizations”,” arXiv preprint arXiv:2109.00551. “Statistical mechanics of probabilistic cellular automata,”
Khodjasteh, Kaveh, and Daniel A Lidar (2005), “Fault- Journal of statistical physics 59 (1), 117–170.
tolerant quantum dynamical decoupling,” Physical review Lee, Hyun C, RS Newrock, DB Mast, SE Hebboul, JC Gar-
letters 95 (18), 180501. land, and CJ Lobb (1991), “Subharmonic shapiro steps in
34
josephson-junction arrays,” Physical Review B 44 (2), 921. junctions,” Journal of Applied Physics 39 (7), 3113–3118.
Liao, Lei, Jasper Smits, Peter van der Straten, and HTC Stoof McGinley, Max, Sthitadhi Roy, and SA Parameswaran (2021),
(2019), “Dynamics of a space-time crystal in an atomic “Absolutely stable spatiotemporal order in noisy quantum
bose-einstein condensate,” Physical Review A 99 (1), systems,” arXiv preprint arXiv:2111.02499.
013625. McLachlan, NW (1947), “Mathieu functions of fractional or-
Liggett, Thomas Milton (2012), Interacting particle systems, der,” Journal of Mathematics and Physics 26 (1-4), 29–41.
Vol. 276. Mi, Xiao, Matteo Ippoliti, Chris Quintana, Amy Greene,
Lin, Cheng-Ju, Anushya Chandran, and Olexei I Motrunich Zijun Chen, Jonathan Gross, Frank Arute, Kunal Arya,
(2020), “Slow thermalization of exact quantum many-body Juan Atalaya, Ryan Babbush, et al. (2021), “Observation
scar states under perturbations,” Physical Review Research of time-crystalline eigenstate order on a quantum proces-
2 (3), 033044. sor,” arXiv preprint arXiv:2107.13571.
Lindner, Netanel H, Gil Refael, and Victor Galitski (2011), Michailidis, AA, CJ Turner, Z Papić, DA Abanin, and
“Floquet topological insulator in semiconductor quantum Maksym Serbyn (2020), “Stabilizing two-dimensional
wells,” Nat. Phys. 7, 490–495. quantum scars by deformation and synchronization,” Phys-
Lledó, Cristóbal, Th K Mavrogordatos, and MH Szymańska ical Review Research 2 (2), 022065.
(2019), “Driven bose-hubbard dimer under nonlocal dis- Middleton, A Alan, Ofer Biham, PB Littlewood, and P Sibani
sipation: A bistable time crystal,” Physical Review B (1992), “Complete mode locking in models of charge-
100 (5), 054303. density waves,” Physical review letters 68 (10), 1586.
Lledó, Cristóbal, and Marzena H Szymańska (2020), “A dissi- Monroe, C, W. C. Campbell, L.-M. Duan, Z.-X. Gong, A. V.
pative time crystal with or without z2 symmetry breaking,” Gorshkov, P. W. Hess, R. Islam, K. Kim, N. M. Linke,
New Journal of Physics 22 (7), 075002. G. Pagano, P. Richerme, C. Senko, and N. Y. Yao (2021),
Losson, Jérôme, and Michael C Mackey (1996), “Thermody- “Programmable quantum simulations of spin systems with
namic properties of coupled map lattices,” Stochastic and trapped ions,” Rev. Mod. Phys. 93, 025001.
spatial structures of dynamical systems , 41–69. Mori, Takashi (2018), “Floquet prethermalization in periodi-
Losson, Jérôme, John Milton, and Michael C Mackey (1995), cally driven classical spin systems,” Phys. Rev. B 98 (10),
“Phase transitions in networks of chaotic elements with 104303.
short and long range interactions,” Physica D: Nonlinear Mori, Takashi, Tomotaka Kuwahara, and Keiji Saito (2016),
Phenomena 81 (1-2), 177–203. “Rigorous bound on energy absorption and generic relax-
Luitz, David J, François Huveneers, and Wojciech De Roeck ation in periodically driven quantum systems,” Physical
(2017), “How a small quantum bath can thermalize long review letters 116 (12), 120401.
localized chains,” Physical review letters 119 (15), 150602. Morita, Hidetoshi, and Kunihiko Kaneko (2006), “Collective
Luitz, David J, Roderich Moessner, SL Sondhi, and Vedika oscillation in a hamiltonian system,” Physical review letters
Khemani (2020), “Prethermalization without tempera- 96 (5), 050602.
ture,” Physical Review X 10 (2), 021046. Morong, William, Fangli Liu, Patrick Becker, KS Collins, Lei
Lyu, Changyuan, Sayan Choudhury, Chenwei Lv, Yangqian Feng, Antonis Kyprianidis, Guido Pagano, Tianyu You,
Yan, and Qi Zhou (2020), “Eternal discrete time crystal AV Gorshkov, and Christopher Monroe (2021), “Obser-
beating the heisenberg limit,” Physical Review Research vation of stark many-body localization without disorder,”
2 (3), 033070. arXiv preprint arXiv:2102.07250.
MacDonald, AH, and Michael Plischke (1983), “Study of the Möser, J (1962), “On invariant curves of area-preserving map-
driven damped pendulum: Application to josephson junc- pings of an annulus,” Nachr. Akad. Wiss. Göttingen, II ,
tions and charge-density-wave systems,” Physical Review 1–20.
B 27 (1), 201. Moudgalya, Sanjay, David A Huse, and Vedika Khemani
Machado, Francisco, Dominic V Else, Gregory D (2020), “Perturbative instability towards delocalization at
Kahanamoku-Meyer, Chetan Nayak, and Norman Y phase transitions between mbl phases,” arXiv preprint
Yao (2020), “Long-range prethermal phases of nonequilib- arXiv:2008.09113.
rium matter,” Physical Review X 10 (1), 011043. Nandkishore, Rahul, and David A Huse (2015), “Many-body
Machado, Francisco, Gregory D Kahanamoku-Meyer, Do- localization and thermalization in quantum statistical me-
minic V Else, Chetan Nayak, and Norman Y Yao (2019), chanics,” Annu. Rev. Condens. Matter Phys. 6 (1), 15–38.
“Exponentially slow heating in short and long-range inter- van Nieuwenburg, Evert, Yuval Baum, and Gil Refael
acting floquet systems,” Physical Review Research 1 (3), (2019), “From bloch oscillations to many-body local-
033202. ization in clean interacting systems,” Proceedings of
Mali, Petar, Jasmina Tekić, Zoran Ivić, and Milan Pantić the National Academy of Sciences 116 (19), 9269–9274,
(2012), “Effects of noise on interference phenomena in the https://www.pnas.org/content/116/19/9269.full.pdf.
presence of subharmonic shapiro steps,” Physical Review Nozières, Philippe (2013), “Time crystals: Can diamagnetic
E 86 (4), 046209. currents drive a charge density wave into rotation?” EPL
Maskara, N, A. A. Michailidis, W. W. Ho, D. Bluvstein, (Europhysics Letters) 103 (5), 57008.
S. Choi, M. D. Lukin, and M. Serbyn (2021), “Discrete Olver, Peter J (1982), “A nonlinear hamiltonian structure for
time-crystalline order enabled by quantum many-body the euler equations,” Journal of Mathematical Analysis and
scars: Entanglement steering via periodic driving,” Phys. Applications 89 (1), 233–250.
Rev. Lett. 127, 090602. Pal, Soham, Naveen Nishad, TS Mahesh, and GJ Sreejith
Matus, Pawel, and Krzysztof Sacha (2019), “Fractional time (2018), “Temporal order in periodically driven spins in star-
crystals,” Physical Review A 99 (3), 033626. shaped clusters,” Physical review letters 120 (18), 180602.
McCumber, DE (1968), “Effect of ac impedance on dc Palmer, RG (1982), “Broken ergodicity,” Advances in Physics
voltage-current characteristics of superconductor weak-link 31 (6), 669–735.
35
Pedersen, PAULI (1980), “Stability of the solutions to zine and Journal of Science 15 (94), 229–235.
mathieu-hill equations with damping,” Ingenieur-Archiv Ringot, Jean, Pascal Szriftgiser, Jean-Claude Garreau, and
49 (1), 15–29. Dominique Delande (2000), “Experimental evidence of dy-
Pedersen, PO (1935), “Subharmonics in forced oscillations in namical localization and delocalization in a quasiperiodic
dissipative systems. part i,” The Journal of the Acoustical driven system,” Physical review letters 85 (13), 2741.
Society of America 6 (4), 227–238. Romeiras, Filipe J, and Edward Ott (1987), “Strange non-
Peng, Pai, Chao Yin, Xiaoyang Huang, Chandrasekhar Ra- chaotic attractors of the damped pendulum with quasiperi-
manathan, and Paola Cappellaro (2021), “Floquet prether- odic forcing,” Physical Review A 35 (10), 4404.
malization in dipolar spin chains,” Nature Physics 17 (4), Rovny, Jared, Robert L Blum, and Sean E Barrett (2018a),
444–447. “Observation of discrete-time-crystal signatures in an or-
Petersen, Karl E (1989), Ergodic theory, Vol. 2 (Cambridge dered dipolar many-body system,” Physical review letters
University Press). 120 (18), 180603.
Pizzi, Andrea, Johannes Knolle, and Andreas Nunnenkamp Rovny, Jared, Robert L Blum, and Sean E Barrett (2018b),
(2019), “Period-n discrete time crystals and quasicrystals “P 31 nmr study of discrete time-crystalline signatures in
with ultracold bosons,” Physical review letters 123 (15), an ordered crystal of ammonium dihydrogen phosphate,”
150601. Physical Review B 97 (18), 184301.
Pizzi, Andrea, Johannes Knolle, and Andreas Nunnenkamp Roy, Rahul, and Fenner Harper (2016), “Abelian floquet
(2021a), “Higher-order and fractional discrete time crystals symmetry-protected topological phases in one dimension,”
in clean long-range interacting systems,” Nature communi- Physical Review B 94 (12), 125105.
cations 12 (1), 1–7. Roy, Rahul, and Fenner Harper (2017), “Periodic table for
Pizzi, Andrea, Andreas Nunnenkamp, and Johannes Knolle floquet topological insulators,” Physical Review B 96 (15),
(2021b), “Bistability and time crystals in long-ranged di- 155118.
rected percolation,” Nature communications 12 (1), 1–8. Rubio-Abadal, Antonio, Matteo Ippoliti, Simon Hollerith,
Pizzi, Andrea, Andreas Nunnenkamp, and Johannes Knolle David Wei, Jun Rui, SL Sondhi, Vedika Khemani, Chris-
(2021c), “(classical) prethermal phases of matter,” arXiv tian Gross, and Immanuel Bloch (2020), “Floquet prether-
preprint arXiv:2104.13928. malization in a bose-hubbard system,” Physical Review X
Ponte, Pedro, Zlatko Papić, François Huveneers, and 10 (2), 021044.
Dmitry A Abanin (2015), “Many-body localization in peri- Rugar, D, and P Grütter (1991), “Mechanical parametric am-
odically driven systems,” Physical review letters 114 (14), plification and thermomechanical noise squeezing,” Physi-
140401. cal Review Letters 67 (6), 699.
Potirniche, I-D, Andrew C Potter, Monika Schleier-Smith, Russomanno, Angelo, Fernando Iemini, Marcello Dalmonte,
Ashvin Vishwanath, and Norman Y Yao (2017), “Floquet and Rosario Fazio (2017), “Floquet time crystal in the
symmetry-protected topological phases in cold-atom sys- lipkin-meshkov-glick model,” Physical Review B 95 (21),
tems,” Physical review letters 119 (12), 123601. 214307.
Potter, Andrew C, Takahiro Morimoto, and Ashvin Vish- Sacha, Krzysztof (2015), “Modeling spontaneous breaking of
wanath (2016), “Classification of interacting topological time-translation symmetry,” Physical Review A 91 (3),
floquet phases in one dimension,” Physical Review X 6 (4), 033617.
041001. Sacha, Krzysztof (2020), Time crystals (Springer).
Prakash, Gyan, Arvind Raman, Jeffrey Rhoads, and Sacha, Krzysztof, and Jakub Zakrzewski (2017), “Time crys-
Ronald G Reifenberger (2012), “Parametric noise squeezing tals: a review,” Reports on Progress in Physics 81 (1),
and parametric resonance of microcantilevers in air and liq- 016401.
uid environments,” Review of Scientific Instruments 83 (6), Sahay, Rahul, Francisco Machado, Bingtian Ye, Chris R Lau-
065109. mann, and Norman Y Yao (2021), “Emergent ergodicity at
Preskill, John (2018), “Quantum computing in the nisq era the transition between many-body localized phases,” Phys-
and beyond,” Quantum 2, 79. ical Review Letters 126 (10), 100604.
Prosen, T (1998), “Time Evolution of a Quantum Many-Body Sala, Pablo, Tibor Rakovszky, Ruben Verresen, Michael
System: Transition from Integrability to Ergodicity in the Knap, and Frank Pollmann (2020), “Ergodicity break-
Thermodynamic Limit,” Physical Review Letters 80, 1808– ing arising from hilbert space fragmentation in dipole-
1811, cond-mat/9707180. conserving hamiltonians,” Physical Review X 10 (1),
Rahav, Saar, Ido Gilary, and Shmuel Fishman (2003), “Time 011047.
independent description of rapidly oscillating potentials,” Scherg, Sebastian, Thomas Kohlert, Pablo Sala, Frank
Physical review letters 91 (11), 110404. Pollmann, Bharath Hebbe Madhusudhana, Immanuel
Randall, J, CE Bradley, FV van der Gronden, A Galicia, Bloch, and Monika Aidelsburger (2021), “Observing non-
MH Abobeih, M Markham, DJ Twitchen, F Machado, ergodicity due to kinetic constraints in tilted fermi-hubbard
NY Yao, and TH Taminiau (2021), “Observation of a chains,” Nature Communications 12 (1), 1–8.
many-body-localized discrete time crystal with a pro- Schulz, Maximilian, CA Hooley, Roderich Moessner, and
grammable spin-based quantum simulator,” arXiv preprint F Pollmann (2019), “Stark many-body localization,” Phys-
arXiv:2107.00736. ical review letters 122 (4), 040606.
Rayleigh, Lord (1883a), “Vii. on the crispations of fluid rest- Seibold, Kilian, Riccardo Rota, and Vincenzo Savona (2020),
ing upon a vibrating support,” The London, Edinburgh, “Dissipative time crystal in an asymmetric nonlinear pho-
and Dublin Philosophical Magazine and Journal of Science tonic dimer,” Physical Review A 101 (3), 033839.
16 (97), 50–58. Sels, Dries, and Anatoli Polkovnikov (2021), “Dynamical ob-
Rayleigh, Lord (1883b), “Xxxiii. on maintained vibrations,” struction to localization in a disordered spin chain,” Phys-
The London, Edinburgh, and Dublin Philosophical Maga- ical Review E 104 (5), 054105.
36
Serbyn, Maksym, Dmitry A Abanin, and Zlatko Papić (2021), Thiery, Thimothée, François Huveneers, Markus Müller, and
“Quantum many-body scars and weak breaking of ergod- Wojciech De Roeck (2018), “Many-body delocalization as
icity,” Nature Physics 17 (6), 675–685. a quantum avalanche,” Physical review letters 121 (14),
Sethna, James P, and Eric D Siggia (1984), “Universal transi- 140601.
tion in a dynamical system forced at two incommensurate Thorne, RE, WG Lyons, Joseph W Lyding, JR Tucker, and
frequencies,” Physica D: Nonlinear Phenomena 11 (1-2), John Bardeen (1987a), “Charge-density-wave transport in
193–211. quasi-one-dimensional conductors. i. current oscillations,”
Shapere, Alfred, and Frank Wilczek (2012a), “Branched Physical Review B 35 (12), 6348.
quantization,” Physical review letters 109 (20), 200402. Thorne, RE, WG Lyons, JW Lyding, JR Tucker, and John
Shapere, Alfred, and Frank Wilczek (2012b), “Classical time Bardeen (1987b), “Charge-density-wave transport in quasi-
crystals,” Physical review letters 109 (16), 160402. one-dimensional conductors. ii. ac-dc interference phenom-
Shapere, Alfred D, and Frank Wilczek (2019), “Regulariza- ena,” Physical Review B 35 (12), 6360.
tions of time-crystal dynamics,” Proceedings of the Na- Thouless, DJ (1969), “Long-range order in one-dimensional
tional Academy of Sciences 116 (38), 18772–18776. ising systems,” Physical Review 187 (2), 732.
Shapiro, Sidney (1963), “Josephson currents in superconduct- Toom, Andrei (1980), “Stable and attractive trajectories
ing tunneling: The effect of microwaves and other observa- in multicomponent systems,” in Multicomponent Random
tions,” Physical Review Letters 11 (2), 80. Systems, Vol. 6 (Marcel Dekker) pp. 549–575.
Shirley, Jon H (1965), “Solution of the schrödinger equa- Toom, Andrei (1995), “Cellular automata with errors: prob-
tion with a hamiltonian periodic in time,” Physical Review lems for students of probability,” Topics in contemporary
138 (4B), B979. probability and its applications , 117–157.
Silber, Mary, Chad M Topaz, and Anne C Skeldon (2000), Träger, Nick, Pawel Gruszecki, Filip Lisiecki, Felix Groß, Jo-
“Two-frequency forced faraday waves: weakly damped hannes Förster, Markus Weigand, Hubert Glowiński, Pi-
modes and pattern selection,” Physica D: nonlinear phe- otr Kuświk, Janusz Dubowik, Gisela Schütz, et al. (2021),
nomena 143 (1-4), 205–225. “Real-space observation of magnon interaction with driven
Sinai, Yaha G (1959), “On the notion of entropy of a dynam- space-time crystals,” Physical Review Letters 126 (5),
ical system,” in Doklady of Russian Academy of Sciences, 057201.
Vol. 124, pp. 768–771. Tucker, Kristopher, Bihui Zhu, Robert J Lewis-Swan, Jamir
Smith, Jacob, Aaron Lee, Philip Richerme, Brian Neyenhuis, Marino, Felix Jimenez, Juan G Restrepo, and Ana Maria
Paul W Hess, Philipp Hauke, Markus Heyl, David A Huse, Rey (2018), “Shattered time: can a dissipative time crystal
and Christopher Monroe (2016), “Many-body localization survive many-body correlations?” New Journal of Physics
in a quantum simulator with programmable random disor- 20 (12), 123003.
der,” Nature Physics 12 (10), 907–911. Turner, Christopher J, Alexios A Michailidis, Dmitry A
Smits, J, HTC Stoof, and P van der Straten (2020), “On the Abanin, Maksym Serbyn, and Zlatko Papić (2018a), “Weak
long-term stability of space-time crystals,” New Journal of ergodicity breaking from quantum many-body scars,” Na-
Physics 22 (10), 105001. ture Physics 14 (7), 745–749.
Smits, J, HTC Stoof, and P Van Der Straten (2021), “Spon- Turner, CJ, AA Michailidis, DA Abanin, Maksym Serbyn,
taneous breaking of a discrete time-translation symmetry,” and Z Papic (2018b), “Quantum scarred eigenstates in a
Physical Review A 104 (2), 023318. rydberg atom chain: Entanglement, breakdown of thermal-
Smits, Jasper, Lei Liao, HTC Stoof, and Peter van der Straten ization, and stability to perturbations,” Physical Review B
(2018), “Observation of a space-time crystal in a superfluid 98 (15), 155134.
quantum gas,” Physical review letters 121 (18), 185301. Urbina, C, JF Jacquinot, and M Goldman (1982), “Rotat-
Sørensen, Anders, and Klaus Mølmer (1999), “Quantum com- ing transverse helical nuclear magnetic ordering,” Physical
putation with ions in thermal motion,” Physical review let- Review Letters 48 (3), 206.
ters 82 (9), 1971. Von Neumann, John (1952), “Probabilistic logics and the syn-
Srednicki, Mark (1994), “Chaos and quantum thermal- thesis of reliable organisms from unreliable components,”
ization,” Phys. Rev. E 50 (2), 888–901, arXiv:cond- in Automata Studies.(AM-34), Volume 34 (Princeton Uni-
mat/9403051. versity Press) pp. 43–98.
Strogatz, Steven H (2018), Nonlinear dynamics and chaos Von Plato, Jan (1991), “Boltzmann’s ergodic hypothesis,”
with student solutions manual: With applications to Archive for History of Exact Sciences 42 (1), 71–89.
physics, biology, chemistry, and engineering (CRC press). Walters, Peter (2000), An introduction to ergodic theory,
Šuntajs, Jan, Janez Bonča, Tomaž Prosen, and Lev Vidmar Vol. 79 (Springer Science & Business Media).
(2020), “Quantum chaos challenges many-body localiza- Wang, Jia, Peter Hannaford, and Bryan J Dalton (2021a),
tion,” Physical Review E 102 (6), 062144. “Many-body effects and quantum fluctuations for discrete
Surace, Federica Maria, Angelo Russomanno, Marcello Dal- time crystals in bose–einstein condensates,” New Journal
monte, Alessandro Silva, Rosario Fazio, and Fernando Iem- of Physics 23 (6), 063012.
ini (2019), “Floquet time crystals in clock models,” Physi- Wang, Jia, Krzysztof Sacha, Peter Hannaford, and Bryan J
cal Review B 99 (10), 104303. Dalton (2021b), “Discrete time crystals in bose-einstein
Taylor, James H, and Kumpati S Narendra (1969), “Stability condensates and the symmetry-breaking edge in a simple
regions for the damped mathieu equation,” SIAM Journal two-mode theory,” Physical Review A 104 (5), 053327.
on Applied Mathematics 17 (2), 343–352. Wang, Y H, H. Steinberg, P. Jarillo-Herrero, and N. Gedik
Thakurathi, M, A. A. Patel, D. Sen, and A. Dutta (2013), “Observation of Floquet-Bloch States on the Sur-
(2013), “Floquet generation of Majorana end modes and face of a Topological Insulator,” Science 342, 453–457,
topological invariants,” Phys. Rev. B 88 (15), 155133, arXiv:1310.7563 [cond-mat.mes-hall].
arXiv:1303.2300 [cond-mat.mes-hall]. Wannier, Gregory H (1959), Elements of solid state theory
37
(CUP Archive). discrete time crystals: Rigidity criticality and realiza-

Watanabe, Haruki, and Masaki Oshikawa (2015), “Absence of tions”,” arXiv preprint arXiv:2109.07485.
quantum time crystals,” Physical review letters 114 (25), Ye, Bingtian, Francisco Machado, and Norman Y. Yao
251603. (2021), “Classical prethermal discrete time crystals,”
Waugh, JS, LM Huber, and U Haeberlen (1968), “Approach arXiv:2104.13927 [quant-ph].
to high-resolution nmr in solids,” Physical Review Letters Zel’Dovich, Ya B (1973), “Scattering and emission of a quan-
20 (5), 180. tum system in a strong electromagnetic wave,” Soviet
Weidinger, Simon A, and Michael Knap (2017), “Floquet Physics Uspekhi 16 (3), 427.
prethermalization and regimes of heating in a periodi- Zettl, A, and G Grüner (1983), “Observation of shapiro steps
cally driven, interacting quantum system,” Scientific re- in the charge-density-wave state of nbse3,” Solid state com-
ports 7 (1), 1–10. munications 46 (7), 501–504.
Wilczek, Frank (2012), “Quantum time crystals,” Physical Zhang, Jiehang, PW Hess, A Kyprianidis, P Becker, A Lee,
review letters 109 (16), 160401. J Smith, G Pagano, I-D Potirniche, Andrew C Potter,
Xu, Huikai, Jingning Zhang, Jiaxiu Han, Zhiyuan Li, A Vishwanath, et al. (2017), “Observation of a discrete
Guangming Xue, Weiyang Liu, Yirong Jin, and Haifeng time crystal,” Nature 543 (7644), 217–220.
Yu (2021), “Realizing discrete time crystal in an one- Zhang, Wenhua, and Kimberly L Turner (2004), “Noise anal-
dimensional superconducting qubit chain,” arXiv preprint ysis in parametric resonance based mass sensing,” in ASME
arXiv:2108.00942. International Mechanical Engineering Congress and Expo-
Yang, Xiaoqin, and Zi Cai (2021), “Dynamical transitions sition, Vol. 47144, pp. 363–367.
and critical behavior between discrete time crystal phases,” Zhao, Liu, Pengfei Yu, and Wei Xu (2013), “Hamiltonian de-
Physical Review Letters 126 (2), 020602. scription of singular lagrangian systems with spontaneously
Yang, Zhi-Cheng, Fangli Liu, Alexey V Gorshkov, and broken time translation symmetry,” Modern Physics Let-
Thomas Iadecola (2020), “Hilbert-space fragmentation ters A 28 (05), 1350002.
from strict confinement,” Physical review letters 124 (20), Zhou, Hengyun, Joonhee Choi, Soonwon Choi, Renate
207602. Landig, Alexander M Douglas, Junichi Isoya, Fedor
Yao, Norman Y, Chris R Laumann, Sarang Gopalakrishnan, Jelezko, Shinobu Onoda, Hitoshi Sumiya, Paola Cappel-
Michael Knap, Markus Mueller, Eugene A Demler, and laro, et al. (2020), “Quantum metrology with strongly in-
Mikhail D Lukin (2014), “Many-body localization in dipo- teracting spin systems,” Physical review X 10 (3), 031003.
lar systems,” Physical review letters 113 (24), 243002. Zhu, Bihui, Jamir Marino, Norman Y Yao, Mikhail D
Yao, Norman Y, and Chetan Nayak (2018), “Time crystals in Lukin, and Eugene A Demler (2019), “Dicke time crystals
periodically driven systems,” Physics Today 71 (9), 40–47, in driven-dissipative quantum many-body systems,” New
https://doi.org/10.1063/PT.3.4020. Journal of Physics 21 (7), 073028.
Yao, Norman Y, Chetan Nayak, Leon Balents, and Michael P Zhuang, Quntao, Francisco Machado, Norman Y Yao, and
Zaletel (2020), “Classical discrete time crystals,” Nature Michael P Zaletel (2021), “An absolutely stable open time
Physics 16 (4), 438–447. crystal,” arXiv preprint arXiv:2110.00585.
Yao, Norman Y, Andrew C Potter, I-D Potirniche, and Žlabys, Giedrius, Chu-hui Fan, Egidijus Anisimovas, and
Ashvin Vishwanath (2017), “Discrete time crystals: Rigid- Krzysztof Sacha (2021), “Six-dimensional time-space crys-
ity, criticality, and realizations,” Physical review letters talline structures,” Physical Review B 103 (10), L100301.
118 (3), 030401. Zounes, Randolph S, and Richard H. Rand (2002), “Subhar-
Yao, Norman Y, Andrew C Potter, Ionut-Dragos Potirniche, monic resonance in the non-linear mathieu equation,” In-
and Ashvin Vishwanath (2021), “Reply to comment on” ternational Journal of Non-Linear Mechanics 37 (1), 43–73.

@2023-Quantum and Classical Discrete Time Crystals

Uploaded by

Document Information

Original Description:

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

@2023-Quantum and Classical Discrete Time Crystals

Uploaded by

Copyright:

Available Formats

Colloquium: Quantum and Classical Discrete Time Crystals

(Dated: May 17, 2023)

CONTENTS crystalline and magnetic ordering to superfluidity, su-

Few-body Systems Spontaneous Floquet

Downloaded from http://science.sciencemag.org/ on July 23, 2021

Wilczek proposed the intriguing concept of a “time crystal”—

(TTSB) [6–8] and ultimately, led to a proof of the “absence of

E.g. parametric resonance of a single non-linear (equilibrium)During quantum

equilibrium setting, and recent work [10,11] has demonstrated

Open / Stochastic Systems

Temperature, T = 0 Temperature, T > 0

Dissipative Dynamical Systems Thermally Activated Toom / Gacs Models

E.g. coupled logistic maps as a realization

Time crystalline order survives to ⌧ !1

Time =crystalline order persists for ⌧ ⇠ e

to a many-body model (a “coupled map lattice”) by height q,

(a) (b) “Floquet Hamiltonian”: HF (Buchleitner et al., 2002;

K(t) = Kmagnus (t) ◦ K0 (t), again with K(mT ) = K(0).

MARKO RADUNOVIC ©123RF.COM

of a discrete time crystal and to draw

through a time-dependent canonical transformation given ap-

ourselves to ground-state properties,

Peak height variance (a.u.)

Peak height (a.u.)

resulting from the effective

intensity, |S( )|2

as evinced other words, thereH 2are locally-indistinguishable states of

the B rotated states jψðtÞi

experimentally with its corresponding 0

crystal phase, the same correlation ψ(ð2n

tire spectrum is established. The

Downloaded from http://science.sciencemag.org/ on July 13, 2021

Downloaded from http://science.sciencemag.org/ on July 13, 2021

(t*) and magnetization (tPDTC) times at

ues are perturbed away from π → π + 1/τ . Intuitively, (a) (b)

these configurations slowly precess into each other over

On the other hand, if the bath is in equilibrium at finite temperature T, j

P∗ ), metry. This net force destroys the possibility of phase

spiracy of air molecules can also cause the entire oscillator

detects the sharp jump in the amplitude of oscillations as

(demarcated 1 I I While the original Toom model exhibits m = 2 steady

clockwise. Because this handedness is preserved under 1. Quantum many-body scars

A. New venues for time crystals

Local Observable hO(t)i

D. Concluding Remarks (Sec. IV.D) to the Floquet-MBL discrete time crystal

(CUP Archive). discrete time crystals: Rigidity criticality and realiza-

You might also like