Alet Cargese

Introduction to Many-Body Localization
Physics in the middle of the spectrum
Fabien Alet
Laboratoire de Physique Théorique
Toulouse
Ecole Méso 2016 Matière topologique,

interactions et couplage matière-lumière,
Cargèse, November 2016
Outline
• Introduction
• A toy model for MBL
• Toy model of the toy model
• Experiments
• (If time allows) More features
• Open issues & Conclusions

What is Many-Body Localization ?
«Many-body localization» (MBL)
A new distinct dynamical phase of matter, which does not self-thermalize
• Characteristics:
• Zero DC conductivity at finite temperature

• Low entanglement
• Anderson localization + dephasing
• Key ingredients : Disorder + Interactions, Isolated system
Interactions
In general, expect interactions to induce transport and to thermalize an isolated localized system
Why Many-Body Localization?
• Why is it an interesting problem? Mostly fundamental questions:
• Does a closed quantum system self-thermalize?

• Is there a perfect insulator at finite temperature?
• What happens to Anderson localization in presence of interactions?
• …
• Why is it a difficult problem?
• All the tough ingredients are there : Quantum Many-Body interactions,

disorder, out-of-equilibrium
• Absence of thermalization: can’t use thermodynamic ensembles!
• Usual condensed matter methods geared towards low-energy properties
• Too many papers !!

PREQUEL
Eigenstate Thermalization Hypothesis

Thermalization of isolated systems
• Quench protocol: Evolve initial state with a (many-body) Hamiltonian | (t)i = exp( iHt)| 0i
Q. : Does the system reach thermal equilibrium ?

X X
• Expand | 0i = an |ni in eigenbasis of H = En |nihn|
i n
• Time-evolved observable (generic Hamiltonian)
X t!1
X
hO(t)i = a⇤n0 an e i(En0 En )t
Onn0 ! |an |2 Onn
n,n0 n ‘Diagonal ensemble’
• Eigenstate thermalization hypothesis (ETH) Deutsch, Srednicki,

… Rigol
hn|O|ni ' hn0 |O|n0 i = O(E) |ni, |n0 i in the same energy shell
hn|O|n0 i vanish in the thermodynamic limit and for few-body observables
• ETH implies thermalization
hO(t ! 1)i = O(E) = O(T ) E=h 0 |H| 0i
E = hHiT
Consequences & Exceptions
• Each eigenstate is thermal, «knows» equilibrium
⇢(0) = |nihn| = ⇢(t) = ⇢eq (Tn ) En = hHiTn
• Memory of initial conditions is lost
• ETH is a «justification» of the microcanonical ensemble at the individual eigenstate level
• Exceptions to ETH : Integrable systems

Anderson (single-particle) localization
t!1
Many-Body Localized (MBL) systems
Old problem revived by an enormous amount of contributions!
Anderson, Fleishmann, ‘Old’ Reviews (2015): Nandkishore & Huse, Altman & Vosk
Shepelyansky... New reviews upcoming (Annalen der Physik)
Season 1
Let’s start with a toy model …

Toy model to understand MBL
• XXZ Spin 1/2 chain in a random magnetic field
X X 6= 0
H= Six Si+1
x
+ Siy Si+1
y
+ Siz Si+1
z
hi Siz
i i
hi 2 [ h, h]
• Infinite disorder : eigenstates are fully localized product-states, no entanglement
• : non-interacting case. Maps to 1d Anderson model

• Branch small interaction : perturbative calculations Gornyi et al., Basko et al. indicate that
thermalization does not occur: states keep localized, no spin or energy transport
• Beyond perturbation (numerics): localization can survive interactions

0 Thermal, «ergodic» MBL
h
Eigenstates look all the same (~ Random Matrix Theory) Eigenstates all different
For a given energy density (say middle of spectrum)

ETH states versus MBL states
Ergodic states MBL states
Follow ETH Violate stat. mech.
Observables are the same Observables differ from

within the same energy shell eigenstate to eigenstate
ETH MBL
100
n + 1|Siz |n + 1 |
=0.5
Difference of local
magnetization between 10 1
consecutive eigenstates
L = 12
10 2 L = 14
| n|Siz |n
L = 16
L = 18
L = 20
L = 22
3
10
0 1 2 3 4 5
h

Random matrix statistics Integrable (Poisson) statistics
ETH MBL
Gap ratio (avoids unfolding)
0.54
rGOE
0.52
0.50
gn = |En En 1| 0.55
0.48 =0.5
r
0.50
0.46
0.44 0.45
hc =3.72(6)
0.42 0.40
=0.91(7)
0.40 rPoisson
80 40 0 40
0.38
12 14 15 16 17 18 19 20 22


Eigenstates occupy all configuration space No delocalization
X
Localization of a wave-function in a basis |ni = ni |ii pi = |hn|ii|2 {|ii} = {S z } basis
i
X 1 X
Participation entropies S1p = pi ln(pi ) Sqp = ln pqi = ln (IPR)
1 q i
i
limited system sizes do not allow to se
Perhaps more accessible to experiments, bipartite fluc-
tuations F of subsystem magnetization (taken here to be possibilities).
a half-chain L/2) have a similar behavior. Being sim- Discussions and conclusions— Using
Ergodic statesconstant of the subsystem, we also ex-
ply the Curie for the MBL transition, MBL ourstates
large-scal
Follow
pect ETH
thermal extensivity (subextensive response) in the ED results indicate
Violate thestat.
existence
mech. of an
ergodic (localized) regime. This is clearly checked in body mobility edge in the excitation s
Observables
Fig. 4 for ✏ =are0.3the sameF/L has a crossing point at the
where Observables
of the random differ from
field Heisenberg chain.
within the same energy
disorder-induced shell
MBL transition. A data collapse (inset eigenstate
show that the ergodic to eigenstate
regime has full fe
of Fig. 4) is also possible for F/L = g[L1/⌫ (h hc )]), lic phase (with aq = 1 and GOE stati
Random
giving hmatrix statistics
c = 3.09(7) and ⌫ = 0.77(4), consistent with es-
Integrable
ergy levels and the (Poisson) statistics
wavefunction coeffi
timates from other quantities (Fig. 1). Finally, we also the localized many-body states do no
Eigenstates
performed occupy
an analysisall configuration space
of the dynamic fraction f of an No delocalization
Hilbert-space localization for configura
initial spin polarization [28], and obtained similar consis- dim H ⇠ 7·105 [61]. Our detailed finite-s
tent scaling (see Supp. Mat. and Fig. 1).X sis (Sup. Mat.) provides a consistent e
1 lengthX q
acteristic
p
ln diverging
pi = lnas(IPR) |h hc |
The disordered many-body system S1p = can be mapped
pi ln(p i) Sq =
1 theq fulli phase diagram. This e
onto a single particle problem on the complex graph through
i
ponent ⌫ appears to violate the Harri
14
criterion ⌫ p 2/d (see p also Ref. [32]) w
a1 =1.00±0.02, l1 =-0.56±0.69 ETH: Sq This
sizes used. = aqisS0quite...intriguing g
12 a2
a1
=1.00±0.01,
=0.07±0.09,
l2
l1
=-1.42±1.16
=1.61±0.79
P same size range, the location a ' 1cri
ofq the
S1
10 a2 =0.00±0.07, l2 =1.66±0.60
P
Eigenstates
sistent forare
all completely delocalized
various estimates used (
8 = 0.4 S2 opens new questions on the finite-size sc
SqP
6 rections to scaling at the MBL transitio

.8
4
h=1 MBL: aq ⌧
follow [27, 28] 1 or Sqp =
standard lq ln(Sop ) + ...
forms.
S1P h = 4.8 Besides these results for the particula
2
S2P
believe that
Eigenstates are the numericaldelocalized
‘marginally’ techniques
0 lel energy-resolved diagonalisation) and
7 8 9 10 11 12 13 14
the ergodic-localized transition (eigens
S0P
Volume Law for entanglement Area law for entanglement
ETH : Entanglement entropy of eigenstates is extensive : Volume law

B
⇢A = TrB ⇢ = ⇢eq A
A (Tn ) NA
SA = TrA ⇢A log ⇢A / NA if Tn 6= 0
System is its own bath: B acts a thermal bath for A
MBL : states have low entanglement SA /NA ! 0 : Area law
NB1 : Same entanglement structure as ground-states

Abanin, Vidal et al.
NB2: MBL states efficiently represented as matrix-product states Eisert et al. Pekker, Clark
- ETH states versus MBL states
eErgodic states MBL states
-
r within the same energy shell eigenstate to eigenstate
of Random matrix statistics Integrable (Poisson) statistics
t
e
r Volume law
e 5
o ETH 4
0.1
S /L
3
h
E
2
e h Area law 1
- 0
0.01 MBL
) 12 16 20 24
L
e
In a nutshell: MBL states do not follow ETH

MBL phase is not thermal
Crucial to work in the « eigenstate ensemble »
Each eigenstate has its own life
Dynamics
Ergodic phase MBL phase
‘Metal’ Not Anderson insulator either!
No memory of initial state Memory, Revivals
Quench procedure : start from an initial CDW state (Néel),

follow the imbalance (staggered magnetization)
0.8
MBL: memory of the initial state
0.6
even at infinite time
0.4
0.2
0
0 50 Metal: no memory
100
(converges to thermal ensemble)

|⟨n|j |m⟩|2 δ(ω − ϵm + ϵn ), (3)
Dynamics
the general features are very similar in both cases [6,36,38],
m e.g., the locations of the maxima are at ω ∼ 1. Essential
− − + differences occur for low ω ≪ 1. While for " = 0 there is
S − S S ) is the (spin) current
Ergodic phase a clear drop towards σ0 = 0 for all W , for " = 1 we find a
MBL phase
+1 i i+1 i
mber of MB states. For calculations
rather broad regime in which σ (ω) follows the low-ω behavior
(ω) and its STS fluctuations we use
in Eq. (1), with σ0 > 0 and α ∼ 1 [36,39]. This behavior will
‘Metal’
act diagonalization (ED). In both of
be elaborated further on. Not Anderson insulator either!
ysis to the system without a uniform
In order to estimate the possible influence of finite-size
The first one is the full ED, allowing
f NNo memory of initial stateeffects, we present in Fig. 2 the direct comparison of the
d ∼ 100 samples with random hi ,
Memory, Revivals
results for σ (ω) for fixed " = 1 but various W = 2,4,6, as
pecial case " = 0, Eq. (2) transforms
obtained for different sizes L = 16–28. Here, for L = 16 we
del Transport
of NI spinless fermions [3–5], No transport
use the ED, while for larger L = 20–28 we use the MCLM.
e ED within the single-particle basis
It is rather obvious that deviations are hardly visible (taking
y L ∼ 1.6 × 104 .
e results are obtained via the
method (MCLM) [36,39,40], best
ntities at elevated T > J . Its compu- 0.06
essentially equivalent to the ones for
ED, but with an increased number of
to improve the frequency resolution
Barisic et al.
W=2
to obtain results for L = 28, Nst ∼
0.04
with typical δω ∼ 2 × 10−3 > δϵ. δϵ
evel separation (e.g., δϵ ∼ 10−3 for
σ(ω)
for LMetal : Finite

= 28). Spectra areDC
broadened
erized by conductivity
the frequency width η. W=4
a macroscopic meaning providing 0.02
er η results involve finite-size and L = 16
L = 20
g., for ω " δϵ, any level repulsion L = 24
ncy dependencies of σ (ω) and STS L = 28
W=6
0
0 0.2 0.8 1
URES AND FLUCTUATIONS OF ω
HE SPECTRA
MBL : Zero DC conductivity, FIG. 2. σ (ω) compared for different system sizes L and three
e delicate issues,
even we
atpresent in Fig.
infinite 1
temperature
values of W at fixed " = 1. L = 16 system is calculated via the ED,
veraged σ (ω) for the NI (" = 0) and whereas for L = 20,24,28 we employ MCLM.
Dynamics
Ergodic phase MBL phase
‘Metal’ Not Anderson insulator either!
No memory of initial state Memory, Revivals
Transport No transport
Entanglement spreads fast Entanglement spreads, but slowly
Quench from an initial product state : follow entanglement growth S(t) = Tr⇢(t) ln ⇢(t)
ETH : Ballistic growth MBL: Logarithmic

ETH MBL h spreadfully diagonaliz
101 0.4 typically L =
a) L = 28 b) L = 20 variational
Znidaric etapp
al.
0.3 formalism [32,
Bardarson et al.
where the entan
S(t)
S(t)
0.2
10 0 the MBL phase
h =1.0
0.1 phase due to th
h =1.6 h =7.0 Anderson :
h =2.0 h =8.0
In order to add
0.0 bounded
of the algorithm
100
Out-of-time ordered correlators
• OTOC quantifies ‘scrambling’ of quantum information ( ~ probability of attempting to

recover information via local operations) 2
i- and j-
|ni leads
(4)
ns results
(5)
few com-
xpanded Fan et al.
bsence of
FIG. 1: The calculation of the von Neumann EE, the second
TOC de- Rényi EE and the OTOC for the MBL and the AL cases in
his shows
PREVIOUSLY
From Nandkishore & Huse
ETH Anderson MBL

Thermal phase Single-particle localized Many-body localized
Memory of initial conditions Some memory of local initial Some memory of local initial
‘hidden’ in global operators conditions preserved in local conditions preserved in local
at long times observables at long times observables at long times.
ETH true ETH false ETH false
May have non-zero DC conductivity Zero DC conductivity Zero DC conductivity
Continuous local spectrum Discrete local spectrum Discrete local spectrum
Eigenstates with Eigenstates with Eigenstates with

volume-law entanglement area-law entanglement area-law entanglement
Power-law spreading of entanglement No spreading of entanglement Logarithmic spreading of entanglement

from non-entangled initial condition from non-entangled initial condition
Dephasing and dissipation No dephasing, no dissipation Dephasing but no dissipation
TABLE I: A list of some properties of the many-body-localized phase, contrasted with properties
of the thermal and the single-particle-localized phases. The spreading of entanglement is discussed
further in Sec.IV-C. Local spectra are discussed further in Sec.IV-D.
fermions in a random potential. An analogous argument can be constructed for objects

with more than two states, but we stick to this two-state example for specificity. Let us
SEASON 2
Toy model of the toy model

Phenomenology : ‘Fixed point’ of MBL
• Exact results Imbrie , phenomenology Explicit , perturbative
Huse etconstruction
al.,
results,
of local conserved strongmany-body
operators in disordered
Ros et al.
disordersystems
RG:
Abanin et al. 1
T. E. O’Brien,1 Dmitry A. Abanin,2 Guifre Vidal,3 and Z. Papić4 Altman & Vosk
Instituut-Lorentz, Universiteit Leiden, P.O. Box 9506, 2300 RA Leiden, The Netherlands
2
Department of Theoretical Physics, University of Geneva,
24 quai Ernest-Ansermet, 1211 Geneva, Switzerland
Quasi-local unitary transform can «diagonalize» the Hamiltonian 4
3
Perimeter Institute for Theoretical Physics, Waterloo, Ontario N2L 2Y5, Canada and
School of Physics and Astronomy, University of Leeds, Leeds, LS2 9JT, United Kingdom
The presence and character of local integrals of motion – quasi-local operators that commute
with the Hamiltonian – encode valuable information about the dynamics of a quantum system.
In particular, strongly disordered many-body systems can generically avoid thermalisation when
rXiv:1608.03296v1 [cond-mat.dis-nn] 10 Aug 2016

there are extensively many such operators. In this work, we explicitly construct local conserved
operators in 1D spin chains by directly minimising their commutator with the Hamiltonian. We
demonstrate the existence of an extensively large set of local integrals of motion in the many-body
localised phase of the disordered XXZ spin chain. These operators are shown to have exponentially
decaying tails, in contrast to the ergodic phase where the decay is (at best) polynomial in the size
Exponential decay of coupling of the subsystem. We study the algebraic properties of localised operators, and confirm that in the
many-body localised phase they are well-described by “dressed” spin operators.
I. INTRODUCTION
• Localized bits (l-bits) Isolated quantum systems (i.e., those disconnected

from any thermal bath) evolve according to unitary evo-
lution. Recently there has been much interest in under-
U ⌧iz U † form a complete set of local integral of motions

standing and classifying the possible outcomes of such
evolution for generic systems, e.g., those containing many
particles, subject to local potentials and interactions, and
in the presence of quenched disorder degrees of freedom.
MBL ~ integrable system
For simplicity, we consider systems defined on a lattice,
and therefore with a local Hilbert space of finite dimen-
sion (e.g., two in the case of spin-1/2 qubits). The total
Hilbert space, being a tensor product of local Hilbert
spaces, is exponentially large in lattice size L, which
makes the problem still very difficult to treat in general. FIG. 1: (Color online) A typical integral of motion ⌧i in th
many-body localised phase of the disordered XXZ model.
• Existence of l-bits now often seen as the definition of MBL One possible outcome that is relatively well understood
is ergodic evolution: a system, prepared in an arbitrary
decays exponentially away from the central site i. The ope
ator ⌧i is determined using the method described in Sec. IV
initial state, evolves towards local thermal equilibrium and its spatial profile is obtained by plotting the disorde
at long times. This process is a result of the “eigenstate averaged logarithm of the operator density, Eq. (14). Th
• Almost all features of MBL can be tested / understood in the l-bits language thermalisation hypothesis” (ETH), which governs the un-
derlying structure of individual (many-body) eigenvec-
system is XXZ spin chain with interaction V = 0.5t and di
order = 5t in Eq. (5).
tors of ergodic systems1–3 . More recently, there has been
a surge of activity focused on understanding a distinct
• Various way of constructing l-bits: perturbation, time-averaged observables, real-space
dephasing between RG
di↵erent parts …system
class of systems which undergo non-ergodic dynamics. A
of the 8–11
. O
well-known example of such systems is the Anderson in- the other hand, it was shown that quantum informatio
sulator4 . Anderson localisation is a generic property of can be recovered in MBL systems using spin-echo tech
low-dimensional systems which is not sensitive to a par- Imbrie, Ros, Scardicchio
niques12 , and there are proposals that exotic types
ticular type of lattice or disorder, but only applies to order can be stabilised using the MBL mechanism13–1
Dynamics & Difference with Anderson localization
• Anderson localized and MBL phases share similar features : no transport, integrable (Poisson)
statistics, area law entanglement …
• Difference is dephasing: can be traced back to the exponential decaying interactions and tails
• Dynamics of l-bits in MBL phase: precession around z axis ( ⌧ z conserved) with a rate due to
interactions with other l-bits.
• For generic initial unentangled states : off-diagonal elements of reduced density matrix decay
as a power law with time : dephasing Serbyn et al.
• No « spin-flip » term in the l-bits Hamiltonian: « no dissipation »
• Explains differences between Anderson & MBL : log growth of entanglement, more complex
Out-of-Time-Order Correlations …
• Helps designing spin-echo protocols to detect / manipulate MBL

Serbyn et al.
SEASON 3
Experiments
Experiments : Fermions
X⇣ in 1d ⌘ Schreiber et al., Science (2015)
• Cold-atomic gas realization of interacting Aubry-André model:

X⇣ † ⌘ X †
X
Ĥ = J ĉi, ĉi+1, + h.c. + cos(2⇡ i + )ĉi, ĉi, + U n̂i," n̂i,# .
i, i, i
X
Ne No
• (Non-)Equilibration of a quenched initial state measured by imbalance I =
Ne + No
,
initial state
B 3
AA localized
e o e o e o e o
non-ergodic localized
/J 2
/J
AA extended
ergodic delocalized
U/J =4.7(1)
0.8 , U/J =10.3(1)
/J =8
0 0.6
0 5 10 15 20
Imbalance
U/J 0.4
/J =3
0.2
/J =0
0
0 10 20 30
Time ( )
One-dimensional trapped ions Smith et al., Nat. Phys. (2016)
• Effective S=1/2 quantum Ising model for 10 trapped ions with ‘programmable’ interactions
Long-range interactions Randomness
• For strong disorder, initial magnetization doesn’t decay : MBL 3 d

perimental
culate the di
and find them
a MBL state
Before sea
tem’s time e
the measured
der. We incr
ditions that
two orthogon
their therma
• Reconstructed level spacing Figure 2a
distribution follows Poisson statistics B = 4Jmax a
Néel ordered
This configur
Experiments: Bosons in 2d Choi et al., Science (2016)
• Trapped cold-atomic bosonic gases with optical lattice, a
interactions
order
and random disorder 1.0
Optical lattice
laser beams
Vacuum window
60
X UX X 0
Ĥ = J â†i âj + n̂i (n̂i 1) + ( i + Vi )n̂i . 0.8 3
40
2 4 20
ts
hi,ji i i Measured
8
disorder potential
13
0.15
0.6Objective lens
Here â†i(âj ) is the bosonic creation (annihilation) op-
Imbalance,
Probability
0.10
0.05
† DMD
• Follow dynamics
erator, n̂ i = â i âi ofthealocal density densitydomain wall on site i =
operator light potential
z
0.00
Disorder, δi
(ix , iy ) and the first sum includes all neighboring sites. A

y
0.4
harmonic trapping ETHpotential Vi =MBL m(!x2 i2x + !y2 i2y )/2 with
b =0 c = 13 3
frequencies (!x , !y ) = 2⇡ ⇥ (54, 60) Hz in x and y direc- 0.2
tion confines the atoms around the trap minimum. The
order 1.0
nearest-neighbor
y
hopping strength at a lattice depth of 60
X † UX X 0.0
12 Er is Ĥ J/h x= = tJ τ â Hz
= 024.8
i âj +
corresponding
n̂i (n̂i 1) + a( tunneling
to i + Vi )n̂i .
time 0.8
0
3
40
of ⌧ = h/2⇡J =hi,ji6.4 ms, 2and i longer range hopping terms 4 20
ts
i 8
0 13 100 200
are neglected† as they are exponentially suppressed. The 0.6
0 5 10 15
Here âi (âj ) isstrength the bosonic creation (annihilation) op- Time, t
Imbalance,
onsite interaction τ †
is U = 24.4 J and i denotes
erator, n̂i = âi âi the local density operator on site i =
93
the onsite disorder potential. For these parameters, in
(ix , iy ) and the first sum includes all neighboring sites. A 0.4
the absence
harmonic oftrapping
disorder, the system’s
potential Vi = m(!x2 i2xground
+ !y2 i2y )/2state
with is in FIG. 2. Relaxation dynamics of a d
the Mott insulating
frequencies (!x , !phase,
y ) = 2⇡ ⇥ however,
(54, 60) Hzwithin x andstrong particle
y direc- wall. The evolution of the imbalance I sh
0.2
tion confines
hole fluctuations 187 τ the
[44].atoms around the trap minimum. The ferent disorder strengths /J = 0 (dark gr
nearest-neighbor hopping strength at a lattice depth of green), 4 (light green), 8 (light blue) and 13
For reference,
12 Er is J/h =we 24.8first tracked the
Hz corresponding to aevolution
tunneling time without 0.0
plays a saturation behavior towards a quasi
any disorder
of ⌧ = h/2⇡J potential= 6.4 ms, applied.
and longerAlready fromterms
range hopping the bare all disorder 100
strengths. For
0 200 low disorders
300 (g
imagesare shownneglected in as
249 τ
Fig. they1b areitexponentially
becomes suppressed.
apparent The that the asymptotic value oft the imbalance is vanishi
Time,
initiallyonsite
preparedinteraction strength is U = 24.4 J and i denotes
Single image
the onsite disorder
density
Averaged image step
potential.
isFor
Single smeared
image Averagedout
these image after a few
parameters, in imbalance remains for higher disorder (blue)
tens of thetunneling
absence oftimes disorder, ⌧ and after longer
the system’s ground time
state isno in infor-
FIG. 2. are fits to thedynamics
Relaxation data withof Ia =
density
I0 expdomain
( t/ts )+
Superconducting films Ovadia et al., Sc. Rep. (2015)
• Thin films of InO can exhibit a superconducting-insulator quantum phase

transition induced by a field
• Close to the critical field, at finite temperature, conductivity measurements suggest

a perfect insulator at non-zero temperature : MBL ?
-4
10
(b)
S1aHiR
-6
10
)
-8
-1
10
(
-10
10 B=0.75 T
Arrhenius fit
Eq.2 fit
B=12 T
-12 ES fit
10
0 0.05 0.10 0.15 0.20 0.25 0.30

T (K)
• Existence
Mapping of a finite-temperature
T dependence of R and . (a)transition implies existence of a many-body
pping ofmobility edge (see
T dependence of R.later)
R (in log scale)
n of T 1/2 : At B = 12 T (in red), including
BEHIND THE SCENES
MBL = Technical challenge

Technicalities
The MBL problem is very difficult to study …
• Theory : full analytical treatment of interactions + disorder impossible at finite energy
density, even in 1d
• Numerics not easier : exact methods limited to small systems, standard iterative
methods target extrema of spectrum, not middle! Stochastic methods (Monte Carlo)
can’t be applied : MBL phase is NOT thermal
• Experiments : bath is always present!

… but it expedited the creation of new methods !
• Extensions of real space renormalization group (RSRG-X) and DMRG (DMRG-X,

possible because of area law) to target high-energy states
• Improved exact diagonalization schemes (shift-invert) to target middle of spectrum
• Dynamics can be followed using Time-Evolving Block Decimation techniques

(because of slow growth entanglement) or in a non-equilibrium Lindblad
framework using Matrix-Product States/Operators
• Race for more isolated experiments with disorder : Cold-atomic systems maybe
easier to isolate and control than traditional condensed-matter ones
SEASON 4
Some new / more stuﬀ

ETH / MBL Phase transition
0 Thermal, «ergodic»
hc MBL
h
‘Quantum phase transition’ (?) implying an
exponential number of states
• Crossover ? Phase transition : first order (?), continuous ?
• If transition, can’t be seen in thermodynamics
• No clean understanding on the nature of the phase transition : Phenomenological RG suggest

a new infinite disorder fixed point in 1d, with dynamical critical exponent
Potter et al., Vosk et al.
• Numerics see a transition, but exponent violate the (applicable?) Harris bound
Luitz et al., Kjall et al.
agnetic field. In order to access properties at varying energy densities across
Many-body mobility edge

r system sizes up to L = 22 spins, we use a spectral transformation which can
vely parallel fashion. Our results allow for an energy-resolved interpretation
alization transition including the existence of a many-body mobility edge.
well characterized by Gaussian orthogonal ensemble statistics, volume-law
• First suggested
ull delocalization by perturbative
in the Hilbert calculations
space. Conversely, : For
the localized fixed disorder, nature of states may
(non-ergodic) depend
on their
n statistics, energy.
area-law Mobilityand
entanglement edge
signsatofextensive energy
multifractality in theisHilbert
possible
calization never occurs. We perform finite size scaling to extract the critical
the localization length divergence.
Basko, Aleiner, Altshuler  
• Numerics see a mobility edge in XXZ and other models

interactions in quan-Energy
intriguing phenomena,density Numerics L=22
y-body localization has
t years. Following pre- Luitz, Laflorencie, FA
co-workers have estab-
approach that the cele-
an survive interactions,
der, many-body eigen-
nse to be precised later)
ommonly referred to as
phase.
t for this topic over the
ed to the fact that the
foundations of quan-
Disorder
to striking theoretical Figure 1. Disorder (h) — Energy (✏) phase diagram of the
7, 8]. Several key fea- disordered Heisenberg chain Eq. Many body
(1). The mobility
ergodic phase edge
highlighted as follows. (dark region with a participation entropy volume law co-
eigenstate thermaliza- efficient a1 ' 1) is separated from the localized regime
• Existence of mobility
a closed system in the
(brightedge debated
region with a:1 ‘bubbles’ argument
⌧ 1). Various symbolsDe Roeck,
(see leg- Huveneers, Müller
end) show the energy-resolved MBL transition points ex-
solely following its own
• ‘Localized-bits’
e of a many-body mo-
model
tractedNOT
from suited in
finite size presence
scaling of a
performed mobility
over systemedge
sizes
L 2 {14, 15, 16, 17, 18, 19, 20, 22}. Red squares correspond to
MBL can host «forbidden» order
• (Discrete) quantum order can be protected by disorder Huse et al., Bauer & Nayak
X X Y
Random Ising chain H = hi ix Ji iz i+1
z
P= x
i
i i i
p
• No disorder, J h : Ground-states have ferro. LRO |0, ±i = (|""""iz ± |####iz )/ 2
Excited states are disordered (domain-walls)

MBL can host «forbidden» order
• (Discrete) quantum order can be protected by disorder Huse et al., Bauer & Nayak
X X Y
Random Ising chain H = hi ix Ji iz i+1
z
P= x
i
i i i
• Adding disorder can localize domain-walls

p
Excited states can now order (spin-glass fashion) |n, ±i = (|""#"iz ± |##"#iz )/ 2
2
tended states, with a dispersion proportional to h, and there-

fore destroy the order at any nonzero temperature (energy den-
sity above the ground state). The model is one-dimensional,
Mermin-Wagner theorem does not apply:
consequently any bond disorder ( J > 0) localizes the non-
Long-range
interacting domain orderandeven
wall excitations in 1dforms
the system ! an
Anderson insulator. The next-nearest neighbor coupling J2
introduces a repulsive interaction between domain walls on
adjacent bonds, and breaks the integrability of the model in
the absence of disorder. In this work, we are primarily in-
terested in the regime of repulsive interactions in the ferro-
magnetic phase. For all the numerical results presented in this
paper, we use the parameters J = 1 and h/2 = J2 = 0.3.
Numerics in 1d conclusions
Our qualitative model: Kjall do et
notal.
depend on the exact values
of these parameters.
At a fixed nonzero interaction strength a MBL transi-
tion is expected at a finite critical disorder strength JcMBL ,
which generally depends on the energy density. An intuitive
• FIG. 1. Phase diagram of the Ising model Eq. (1) in the h/2 = J2 =
Same argument for topological order
0.3 plane with ✏ = 2(E Emin )/(Emax Emin ) being the energy
schematic picture of the nature of the different phases in terms
density relative to the total band width. The axes give the energy
of domain walls is given at the top of Fig 1. In the ther-
• MBL states
density above could bestate
the ground usedand to
the build
disorderquantum
strength. Thememories
col- mal phase domain walls are extended over the whole system,
while in the MBL phase domain walls are localized. Various
quantum quenches [30], the latter addresses
in the same class as Ls interacting fermionic
Precursor effect of MBL in 1d : subdiffusion
are ever increasing technological ca-1 recipes [18, 21]
disordered potential
isolated quantum systems through
[20–23].
versaling
/2, thus
Exact
doubling
the MBL transition)
entanglement
(which should
diago-
the
ing
findstate
growth
be timequestion
valid
a subdiffusive
lapse
in some
is lost phase
before
of
regionfor
how
withfortime,
saturation.
observing
preced-the memory
energyand The
uni- of an sub
com-to make
allows
init
studies
nd, have
recently, through clearly
coupled, identified
con- anda spin
MBL transi-
transport, with continuously changing subdiffusion T
•
,qubits
In 1d,
24], [2].
MBL
andWhile
a many
phase
there is can
body
induce
a com-bination
mobility
anomalous
of
exponents. open
edgeNumerical
slowing
boundaries down in
nection the
and
in onecalculations (on small systems) find ei-
regular
large
with ETHsystem
recent phase (Griffiths
sizes
experiments is cru-effect)
on interact
tical handle on capturing ground state cial to thercapture
(i) a subdiffusive the regimein
asymptotic closea 1D to thequasi-random
regime
MBL transition for the
pre- optical
spreadinglattice
Agarwal [31].
et MB
al.
[18, 24], in contrast with single particle Ander-
ms •[3],Sub-diffusion
understanding their dynamical by
characterized ceded,
varying at smaller disorder, by
power-laws even a transition
deepphase intothe diffusion
ETH [19],
phase or sho
ation. However the
way from the ground state, is fraught precise of entanglement.
nature of the tran- In the ETHOur
(ii) a subdiffusive regime all the way to zero disorder without exact of
numerical the random-field
2 results for the time
ains challenges.
erical elusive despite ETHtentative Heisenberg finite
MBL size
a sharp chain
transitionscaling
h model
in between Eq.
these
[20]. (1) two at small
quantities disorder provide
a) strength
a strong cas
sup
fully diagonalizing H is restricted to small system sizes, L = 20
ent years we have

practically 10 witnessed a change
limited 1
to the h, we
small In
see
rangethisin Letter
Fig.
of
0.4 we 2 resolve
a
typically
avail- L =the 16 issue
sub-ballistic
absence for Eq.by (1).clearly
of growth demonstrat-
Time evolution
di↵usive
variational approaches based on matrix-product states
1.0
in
using
regimetime
L = 24
of
in
L = 28 the
most of cri
the
ing the existence offormalism
both a [32, diffusive
33] are and a subdiffusive regime
mofroleisolated quantum systems, in par-
a) L = 28 b) L = 20
sizes L  22 [18]. entanglement 0.3
entropy, Entanglement ETH
which phase.
particularly
entropy
follows Instead,
successful in cases
a sub-ballistic tim e
that disorder plays in such sys- within the 0.2 ergodic phase; in so doing we identify the tran-
where the entanglement entropy remains small, e.g.
0.8 in
S(t)
S(t)
y,came twoaboutanalytical
from
10 0
the studyphenomenological
of An- sition point renormal-
(its
the MBL phase,
value
phase being
due to
growthbut rapidly
different
the fast
is
from
entanglement
clearly
break down in the
previous
growth
observed
ergodic
claims),
(see below).
for the von-Neum bet
1/z
h =1.0
0.1
1/z 1/z
in interacting,have
proaches many-bodybeen quantum proposed and
h =1.6
h =2.0 [25, provide26]0.0an
h =7.0
h =8.0 forexplanation
In order toS(t)
the for
S(t)
address the/ finite-size
the/ t
ETH t ,
regime,, effects
with in
a0.6terms
disorder-dependent
we take advantage
of the algorithm first proposed in Ref. 29 which is based
(4) Th
exp
vation that disorder
transition 10MBL and quantum
0
— ETH ef- in ofone a mechanism
dimension. for equilibration
on a projection The ofHamiltonian
the conserved
of the relaxation to the of
quantities
Krylov anspace initial spin densitypow im
n
t (of energy, charge or spin) even at
di↵erent
c)
ingredients,
L = 28
with
both
d) of
L
a
= the
20
studies
integrable
1.0
disorder-dependent
neverthe-
clean
K = model
span Imbalance
(| 0 [23].
i, H|
displays
0 i,Which
.n
.
dynamical
.0H
a
| regime
i) using
power-law
occurs
the
exponent
0.4 de-
Lanczos
expo- behavior, z 1.
as it com dec
nd in the presence of interactions [6] pends both0.9on thenential anisotropy
algorithm and
parameter
calculation
in the orthonormal Krylov
of
of the
the clean
(small) matrix
system
⇣ space basis. Here, we
I(t)
I(t)
I(t)
d comparable
phenomenology of aconclusions
so-called many- Theregarding
time
and on windowthethe
disorder
0.8
criti-
useover
strength. I(t)
which
The phase
the implementation /ofsub-ballistic
t diagram
the SLEPc with a[34]
summarizing
package non-universal
entanglement
which
0.2
exponent ing ⇣
hase
. One exhibiting many unique
interesting commonand in- aspect these points
is that is shownslow in Fig.
calculates the 1.
bymatrix exponential in the Krylov basis0by
sub-dominant oscillatory
z able to reach 1 terms.
2 3 These
4 tw
10
ow growth of entanglement
1
spreading
[7,10 8],
is visible
We study spin transport
0.7
grows
a simple as
eigendecomposition.
byforcoupling
(s 1 L)
We are
thestrength
, which
system(uptotoa combi-
is
large clearly h ap-
system sizes any disorder L = 28
is predicted on 10
], protection of symmetries
the 0
10
delocalized
1
t [10], parent
and
2 1
side
0 1
10 10 10 10 10 10
in
2
of 3
the
Fig. 2 as
4
tran-
sites)plateaus governing
in the intermediate of the
time the
regimelocal entropy
(up to t(in
2
' 10 time) growth expo-and the d
t Luitz, Laflorencie, FA Tim
erpreted
of eigenstatesasFigure
Griffiths
[11–13] regions
are aaveraged
fewnent of [27].ofobtained
1/z Signatures from
saturates
imbalance
duesliding
to finite-system fits are
for the largest systems) before the entanglement entropy
to
sizes.
continuously
Figure
Asthe 2. b), d) and varying
form
we previously Eq. with
f) Disorder averaged
4 (see
three values of disorder. c),hal
1. Disorder time evolution the entangle-
2 γ sizesh and e) a
fomalously
this newly identified
slow phase;
dynamics see re- on the ergodic side strength and both vanish at the MBL trans
•
showed [18] that the critical disorder strength of the
Calculation with simple
ment entropy S(t) [panels a) and b)] for the half-system in c
hensive list. The

an open chain
implications
d)], all
and spin
measured after a of caption
thefrom a randomof
density
quench
imbalance I(t)
initialFig.
[panels
product 2).
c) and
These
MBL transition dependslocal
1.5on the energypower of by law
power
eigenstates, fits
law
it fits provide
over 8 points in time,
can
local baths
(unentangled)
g inherently robust, on large
state having
are far reaching, systems
an
anshowaverage energy
estimate
1.5
in the middle
ofer-how
is crucial to specify the energy of the initialwith
the exponent
=
changes
state. growing
diffusive
To this extent in terms of system
if the
s
1/z,fit win-as a functio
end, we calculate
sub-diffusive for 1
all disordered samples the average
l physics confirm
of the spectrum.
to the
godic sub-diffusion
theory
ETH phase, of
Left panels
quantum
where
the behavior in the
the entanglement entropy grows as energy density ✏ = (h (0) |H| (0)i
0 E 1
the
) /
0 (E
exponent
E ),
displayed of
• disorder strength as the exponent decrea
0.5
of which havecays
a powerlaw / t
already been
algebraically
until dow
1/z
saturationis
/ texperi-⇣
and displaced
∆ imbalance
the 1
at intermediate times (EDhresults
de- and ac2plateau
the sample,
1
for random
indicates
with Eh (E ) the groundstate (maximal) energy of
0
basis
Subdiffusion
a
= super-diffusive
states | (0)i
real
until we
powershould law NOT be of
0
c1 find one whose energy density is closethere in d>1
regime. As
for L = 28 sites). Right panels display the dynamical be-
havior in the MBL phase, where the entanglement entropy the
0.5saturates at
observed
diffusive domains
desired target density.
of = ballistic we focus local expo-
In the following,
constant
enough to
infi
the
region (in one dimensional systems) Ls /2, thus doubling the time lapse
Znidaric, Scardicchio, Varma grow with system
grows logarithmically in time and the imbalance
nents
1
are located size, we ofconclude (✏that0.5). in the ther-
on initial states with total zero magnetization that
a nonzero constant (ED results for L = 20 sites). Here, we
in the middle the spectrum versal = entanglement growth beforepar
sat
h remains to be said about the con-
have averaged over 103 disorder configurations.
We average our results over at least=1000 disorder real-
ransition to it. Although both aspects modynamic 0 limit the entanglement entropy
bination
localized
grows indeed and large
of open boundaries
izations, choosing a di↵erent initial state for each sample. the
0 0.2 of 0.4 0.6 0.8 1 cial to capture the asymptotic regime
cus on characterizing the conducting
MBL regime, we recover the slow logarithmic growth
entanglement, while theas a power law.
density For sizesWeLfirstofdiscuss
memory of initial spin h the entanglement
Sub-ballistic systemgrowth— 16theaccessible
entanglement. In the to wito
ETH phase
SCENES FROM NEXT EPISODE
Open questions
Current (open) issues in the field
• Existence of mobility edge : numerics and bubble arguments in contradiction, what
about local integral of motions ?
• Nature of the phase transition ? RG vs numerics ? Field theory ??
Does MBL exist in d>1 ? In the continuum ?

Can we constraint existence of MBL due to What about experiments in d=2??
symmetry (à la Mermin-Wagner) ?
Potter, Vasseur
Does MBL really need disorder ? Suggestions of translation-invariant MBL

(including in Josephson Junctions). Do they pass the numerical tests? De Roeck, Huveneers,
Pino, Ioﬀe, Altschuler Müller
Influence on the metallic phase :

Relation to Anderson problem
subdiffusion limited to 1d, generic?
(e.g. on a random graph) ?
MBL and driven / Floquet
Quantitative description of
systems: avoid heating!
coupling to a bath ?
Nandkishore, Gopalakrishnan
Search for MBL in other condensed-matter
systems (Josephson junctions, DNP) ??
Conclusions & outlooks
• Message 1: MBL is an active interesting field! Revisits usual stat-mech,
connections to different fields (many-body quantum physics, quantum
chaos, quantum information, cold-atoms…)
• Message 2 : MBL is a technical challenge for theoreticians

Field theory? Numerics are hard! New Methods?
& experimentalists
Realization in condensed matter / mesoscopic physics setups ?
• Message 3 : Many physics question still open!

References
• Previous reviews in Ann. Rev. Cond. Mat. Phys. (2015)
Focus on static properties: R. Nandkishore, D. Huse, ibid 6, 15 (2015) [ arXiv:1404.0686 ]
Dynamical properties: E. Altman & R. Vosk, ibid 6, 383 (2015) [ arXiv:1408.2834 ]
Coupling to a bath: R. Nandkishore & S. Gopalakrishnan, arXiv:1606.08465 (intro contains review)
• Special edition of Annalen der Physik (2017) contains reviews
Local integral of motions: J. Imbrie, V. Ros & A. Scardicchio, arXiv:1609.08076

L. Rademaker, M. Ortuno & A. Somoza, arXiv:1610.06238
Sub-diffusion in the ergodic side & numerics: D. Luitz & Y. Bar Lev, arXiv:1610.08993
Transitions and dynamics: S. Parameswaran, A. Potter & R. Vasseur, arXiv:1610.03078
Griffiths effects: K. Agarwal et al., arXiv:1611.00770
and more to come …

Alet Cargese

Uploaded by

Document Information

Original Title

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

Alet Cargese

Uploaded by

Copyright:

Available Formats

Introduction to Many-Body Localization

Physics in the middle of the spectrum

Ecole Méso 2016 Matière topologique,

• Open issues & Conclusions

A new distinct dynamical phase of matter, which does not self-thermalize

• Zero DC conductivity at finite temperature

• Key ingredients : Disorder + Interactions, Isolated system

• Does a closed quantum system self-thermalize?

• Why is it a difficult problem?

• All the tough ingredients are there : Quantum Many-Body interactions,

• Absence of thermalization: can’t use thermodynamic ensembles!

• Usual condensed matter methods geared towards low-energy properties

• Too many papers !!

Eigenstate Thermalization Hypothesis

Q. : Does the system reach thermal equilibrium ?

• Eigenstate thermalization hypothesis (ETH) Deutsch, Srednicki,

hn|O|n0 i vanish in the thermodynamic limit and for few-body observables

• ETH implies thermalization

hO(t ! 1)i = O(E) = O(T ) E=h 0 |H| 0i

• ETH is a «justification» of the microcanonical ensemble at the individual eigenstate level

• Exceptions to ETH : Integrable systems

Many-Body Localized (MBL) systems

Old problem revived by an enormous amount of contributions!

Let’s start with a toy model …

• : non-interacting case. Maps to 1d Anderson model

• Beyond perturbation (numerics): localization can survive interactions

For a given energy density (say middle of spectrum)

Observables are the same Observables differ from

Observables are the same Observables differ from

Random matrix statistics Integrable (Poisson) statistics

Observables are the same Observables differ from

Random matrix statistics Integrable (Poisson) statistics

6 rections to scaling at the MBL transitio

ETH : Entanglement entropy of eigenstates is extensive : Volume law

MBL : states have low entanglement SA /NA ! 0 : Area law

NB1 : Same entanglement structure as ground-states

In a nutshell: MBL states do not follow ETH

Quench procedure : start from an initial CDW state (Néel),

(converges to thermal ensemble)

for LMetal : Finite

ETH : Ballistic growth MBL: Logarithmic

• OTOC quantifies ‘scrambling’ of quantum information ( ~ probability of attempting to

ETH Anderson MBL

ETH true ETH false ETH false

May have non-zero DC conductivity Zero DC conductivity Zero DC conductivity

Continuous local spectrum Discrete local spectrum Discrete local spectrum

Eigenstates with Eigenstates with Eigenstates with

Power-law spreading of entanglement No spreading of entanglement Logarithmic spreading of entanglement

Dephasing and dissipation No dephasing, no dissipation Dephasing but no dissipation

fermions in a random potential. An analogous argument can be constructed for objects

Toy model of the toy model

rXiv:1608.03296v1 [cond-mat.dis-nn] 10 Aug 2016

• Localized bits (l-bits) Isolated quantum systems (i.e., those disconnected

U ⌧iz U † form a complete set of local integral of motions

• No « spin-flip » term in the l-bits Hamiltonian: « no dissipation »

• Helps designing spin-echo protocols to detect / manipulate MBL

• Cold-atomic gas realization of interacting Aubry-André model:

Long-range interactions Randomness

• For strong disorder, initial magnetization doesn’t decay : MBL 3 d

• Trapped cold-atomic bosonic gases with optical lattice, a

(ix , iy ) and the first sum includes all neighboring sites. A

of ⌧ = h/2⇡J =hi,ji6.4 ms, 2and i longer range hopping terms 4 20

• Thin films of InO can exhibit a superconducting-insulator quantum phase