ARTICLES

Entanglement and the foundations of

statistical mechanics

Published online: 29 October 2006; doi:10.1038/nphys444

SANDU POPESCU1,2 , ANTHONY J. SHORT1 * AND ANDREAS WINTER3

H. H. Wills Physics Laboratory, University of Bristol, Tyndall Avenue, Bristol BS8 1TL, UK
Hewlett-Packard Laboratories, Stoke Gifford, Bristol BS12 6QZ, UK
3
Published
online:
29 October
2006;
doi:10.1038/nphys444
Department of
Mathematics,
University
of Bristol,
University Walk, Bristol BS8 1TW, UK
* e-mail: tony.short@bristol.ac.uk
1

Statistical mechanics is one of the most successful areas

he gre
physic
of physics. Yet, almost 150 years since its inception,
and e
properties
its foundations and basic postulates are still the subject
Here w
he great
conceptual
puzzle
statistical
mechanics isof how
a
Statistical mechanics is one of the most successful areas of debate.
statistica
Here
we suggest
thatof the
main postulate
physical system, despite always being in some definite
state,
about
twen
that
results
from
considering
each
state
of
the
universe
to
be
equally
of physics. Yet, almost 150 years since its inception, of statistical
mechanics,
the equal acan
priori
probability
subjective
and evolving
deterministically,
exhibit
thermodynamical
probable.
To make
sensetoofstatistical
this image,
assumptions
ergodicity
are 1not
. re
properties
pertinent
averages,
such as of
the
entropy
postulate,
should
be
abandoned
as
misleading
and
its foundations and basic postulates are still the subject are needed to ensure that the universe explores all the available
is, the sys
Here we consider an alternative approach to the foundations
is in a q
space
equally, and
ofargue
coursethat
thisitdoes
not solve
the problem
what
unnecessary.
We
should
be replaced
by
aofAharonov
of
statistical
mechanics,
suggested
to
one
of
us
by
Yakir
E
of debate. Here we suggest that the main postulate the state of the subsystem is at a given time.
thermaliza
about canonical
twenty years
ago. Inwhose
this approach
the usualis devices
of
the environ
general
principle,
physical
content
What
we
showed
here
is
that
these
averages
are
not
necessary.
2
of
statistical mechanics, the equal a priori probability subjective randomness, ensemble-averaging or time-averaging
,
the univer
Published online: 29 October 2006; doi:10.1038/nphys444
Rather,
(almost)different
any individual
state
of the universe
is such
that
fundamentally
from
the
postulate
it
replaces:
it
similar
line
are not required. We show that, although the universe
(that
any
suciently
small
subsystem
behaves
as
if
the
universe
was
in
postulate, should be abandoned as misleading and is, the system together with a suciently large environment)
and Gemm
refers
to
individual
states,
rather
than
to
ensemble
or
he great conceptual
puzzle
of statistical mechanics
how aThis is due to massive entanglement
the
equiprobable
average isstate.
Statistical mechanics is one of the most successful areas
We for
is
in
a
quantum
pure
state
subject
to
a
global
constraint,
physical
system,
despite
always
being
in
some
definite
state,
S argue that it should be replaced by a between the subsystem and
unnecessary.
We
the
rest of
universe,
which states
is a that
time
averages.
Furthermore,
whereas
thethe
original
postulate
of physics. Yet, almost 150 years since its inception,
and evolving deterministically,
can exhibit
thermodynamical
thermalization
results
from
entanglement
between
the
system
state)and
for
1
genericaverages,
featuresuch
ofasthe
vast majority
of states. To obtain this result,
.
properties pertinent to statistical
the entropy
its
foundations
and
basic
postulates
are
still
the
subject
the
This
leads to
athe
finite
entropy
ofpropose
the
system,
despiteq
general canonical principle, whose physicalHere
content
is
an environment.
unprovable
assumption,
principle
we
rigorous
we considerisan
alternative
approach to the
foundations
we
have
introduced
measures
ofthe
eective
size
of the
system,
to one of usitself
by Yakirhaving
Aharonovzero entropy. Significant results
the
universe
alongth
e ordinary
of debate. Here we suggest that the main postulate of statistical mechanics,dsuggested
,
and
its
environment
(that
is,
the
rest
of
the
universe),
d
,
and
SIn
is
mathematically
proven.
The
key
element
in
this
proof
E
3
4
about
twenty
years
ago.
this
approach
the
usual
devices
of
fundamentally different from the postulate it replaces: it similar lines have been obtained
constraints
,
Lloyd
by
Bocchieri
and
Loinger
2
of statistical mechanics, the equal a priori probability subjective randomness,showed
that
the
average
distance
between
the
individual
reduced
,
ensemble-averaging or time-averaging
5
determine
B
Gemmer
et
al.
;
see
also
very
recent
work
by dGoldstein
et al6 .
isand
the
quantum
entanglement
between
the
system
and
e
are not required. We states
show
that,
although
the
universe
(that
/d
.
Levys
and
the
canonical
state
is
directly
related
to
S
refers
individual
states,as rather
than
or
state for th
E
postulate,toshould
be abandoned
misleading
and tois,ensemble
the system together with a We
suciently
large environment)
formulate
and
prove
a
general
canonical
principle
, which
lemma
is then
invoked
to concludeseparates
that all but
an
exponentially
constraints
its
environment.
Our
approach
the
issue
of
is
in
a
quantum
pure
state
subject
to
a
global
constraint,
unnecessary. We argue that it should be replaced by a
states
that between
the
system
will
be thermalized
(that state.
is, in the canonical
time
averages. Furthermore, whereas the original
postulate
thermalization
results from
entanglement
the system
generalized
small
fraction
of all
states
areand
close
to the canonical
Figure
3
Example:
A
system
of
spins.
As
a
concrete
application
of
our
theorem,
finding
the
canonical
state
from
finding
out
how
close
a
toThe
a finite
entropy
of the system,
despite
for
almost
all
pure
states
of the
universe,
provide
general canonical principle, whose physical content is the environment. This leads state)
Our res
main
message
of our
paper
is that
averages
are notand
needed
the
universe
itself
having
zero
entropy.
Significant
results
along
consider
of n spin-1/2 systems
in an external magnetic
field B, where
a
is
ana set
unprovable
assumption,
principle
we propose
consider
an
rigorous
In fact,
we the
actually
beyond
3 bounds.
fundamentally
different from
the postulate the
it replaces:
it similar
tosystem
justify
thetoquantitative
canonical
state
ofto4 ago
system
inbeyond
contact
with
thego
rest
of
is
it,
allowing
even
usual
, us
Lloyd
lines have been obtained
by Bocchieri
and
Loinger
particular subset of k spins form the system S, and the remaining nand
kGemmer
spins form
almost
all
thermalization:
statistical
setting,
energy
et al.5 ; seethe
alsoordinary
very recent work
by Goldstein
et
al6in
. the standard
universealmost
any
individual
state
of
the
universe
is
enough
refers
to
individual
states,
rather
than
to
ensemble
or
mathematically
proven.
The
element
proof
in which
theisenvironment
E. We consider
a restriction
to thekey
energy
eigenspacein
HRthis
We
formulate
and prove
a general canonical
principle
, whichon the state of the universe,thermalize
boltzmannian
situation.
constraints
are
imposed
which
to
lead
to
the
canonical
state. In eect, we propose to replace
any initial
that the system will be thermalized (that is, in the canonical
thewith
original
postulate
nptime
spinsaverages.
are in theFurthermore,
excited state whereas
|1 (aligned
the field)
and thestates
remaining
a corresponding
thermal
canonical
priori temperature
probabilities and
by the
principle
of the s
thedetermine
postulate
equalanda provide
state)
for almostand
all pure
states
of the of
universe,
is p)
the
quantum
entanglement
between
the
system
about
n(1
spins
are
in
the
ground
state
|0
(opposite
to
the
field).
With
this
setup,
is an unprovable assumption, the principle we propose rigorous quantitative bounds.
fact,
we actually
go In
beyond
stateIn for
the
system.
contrast, we
allow
completely
arbitrary
a priori
probabilities,
which
states
that as far
apparently
equal
weas
might e
ordinary
thermalization:
in
the
standard
statistical
setting,
energy
= 2kenvironment.
and d R 2nH(p) , where
H(p)
=
p
log
(p)

(1
p)log
(1
p)
is
the
d Sits
2
2
Our approach separates the issue of constraints, which leads to the system being in a corresponding
increase. It
2

ARTICLES

is mathematically proven. The key element in this proof

theonsystem
constraints are imposed
the stateis ofconcerned
the universe,almost
which

every state of the universe seems

LETTERS

Thermalization and its mechanism for generic

isolated
quantum
systems
Thermalization
and its mechanism
for generic

Marcos Rigol1,2, Vanja Dunjko1,2 & Maxim Olshanii2 Vol 452 | 17 April 2008 | doi:10.1038/nature06838

LETTERS
isolated quantum systems

dependent on it14, seemingly implying that conservation of e

An understanding of the temporal evolution of isolated many1,2
1,2
2
body
quantum
has Dunjko
long been
elusive.
Recently,
meaningis the only independent constraint. On the other hand, even
Marcos
Rigolsystems
, Vanja
& Maxim
Olshanii
1,2
ful
experimental
studies
operators are functionally dependent, their expectation val
the problem
Marcos
Rigol1,2, Vanja
Dunjko1,2 of
& Maxim
Olshanii2have become possible,
37
considered as functionals of statesgenerally are not: for exa
stimulating theoretical interest . In generic isolated systems,
8,9
two states may have
the same mean energies but different me
non-equilibrium dynamics is expected to result in thermaliza14 of energy
dependent
on
it
An
understanding
of
the
temporal
evolution
of
isolated
many,
seemingly
implying
that
conservation
dependent
on it For
An understanding
of the temporal
evolution
of of
isolated
many- squared
, seemingly
implyinghamiltonians
that conservation
of e
non-degenerate
a maximal
tion:
a relaxation
in which
values
macroscopic
body quantum
systems hasto
longstates
been elusive.
Recently,the
meaningis the only
independent constraint.
On the energies.
other hand, even
when
body
quantum
systems
has
long
elusive.
Recently,
meaningis the their
onlyexpectation
independent
constraint.
On the
other hand,expec
even
ful experimental
operators
are
functionally dependent,
values
of the problem
havebeen
become
possible,
constants
of
motion
with
functionally
independent
quantities
arestudies
stationary,
universal
with
respect
to
widely
differing
considered
aspossible,
functionals of statesgenerally
are
not:functionally
for example,
stimulating
theoretical
interest
.1,2
In generic isolated systems,
ful
experimental
studies
operators
are
dependent,
their
expectation
val
become
1,2
1,2 of the problem
2 have
values
is
as
large
as
the
dimension
of
the
Hilbert
space;
exa
initial
conditions,
and
predictable
using
statistical
mechanics.
Marcos
Rigol
,
Vanja
Dunjko
&
Maxim
Olshanii
two
states
may
have
the
same
mean
energies
but
different
means
of
non-equilibrium dynamics is expected to result
37 in thermalizaconsidered
functionals^aof
not: for
exa
stimulating
theoretical
interest
.of In
generic
isolated quantum
systems,
include
theasprojectors
However,
it is
what
of many-body
~statesgenerally
energy
eigens
jY a ihY a j to the are
squared energies.
For non-degenerate
hamiltonians
a maximal set ofP
tion: a relaxation
to not
statesobvious
in which the
valuesfeature
macroscopic
8,9
5
constants
of motion
with functionally
independent
expectation
two the
states
may have
theofsame
mean energies
quantities are stationary,
universal withthermalization
respect
to widely differing
non-equilibrium
dynamics
is expected
topossible
result
in
thermalizamechanics
makes
quantum
in
a
sense
anaand
integer
powers
the
hamiltonian
. but different me
values is as large as the dimension of the Hilbert space; examples
initial conditions, and predictable using statistical mechanics.
squared
energies.eigenstates
non-degenerate
a maximal
dependent
on itthermaAn understanding
ofin
thewhich
temporal
evolution
of isolated
many, seemingly
implying
that conservation
ofFor
energy
tion:
a itrelaxation
to
states
in
which
the
values
of macroscopic
logous
dynamical
chaos
makes
classical
The
and analytichamiltonians
evidence from
the stu
include
the projectors
P^a ~jY a ihY
However, to
isthat
not obvious
what feature
of many-body
quantum
the energynumerical
a j tocurrent
body quantum systems has
long
been
elusive.
Recently,
meaningis
the
only
independent
constraint.
On
the
other
hand,
even
when
10 thermalization
constants
of motion
with that
functionally
independent
expec
quantities
are
stationary,
universal
with
respect
to
differing
mechanics
makes
quantum
possible
in a sense
ana- chaos
andwidely
theitself
integer
powers of the
hamiltonian
lization
possible
.ofFor
example,
dynamical
cannot
integrable
systems
suggests
there exists
a minimal set
of
ful
experimental
studies
operators
are functionally
dependent,
their. expectation
values
the problem
have
become
possible,
A
B
logous to conditions,
that
in which interest
dynamical
chaos
makesisolated
classical
thermaThe current
numericalofand
analytic
evidence
from
the
study
of dimension of the Hilbert space;observ
values
is
as
large
as
the
exa
initial
and
predictable
using
statistical
mechanics.
considered
as
functionals
statesgenerally
are
not:
for
example,
stimulating
theoretical
.
In
generic
systems,
occur
an isolated
quantum
system,
in which
timesystems
evolution
pendent
constraints
size of which is much less than the dime
lization in
possible
. For example,
dynamical
chaos
itself
cannot the
integrable
suggests
there
existsbut
a minimal
set
ofthe
indewith
two
states may
have
the samethat
mean
energies
different
means
of P
non-equilibrium
dynamics
is expected
to result
in11
thermaliza^a ~jY a ihY a j to the energy eigens
4,5
include
the
projectors
However,
it
is
not
obvious
what
feature
of
many-body
quantum
occur
in
an
isolated
quantum
system,
in
which
the
time
evolution
Thermal
state
pendent
constraints
the
size
of
which
is
much
less
than
the
dimension
system
is
linear
and the
spectrum
. Some recent
studies
of the
Hilberta maximal
space set
but
I
squared
energies. Foreven
non-degenerate
hamiltonians
of may still be much 5greater than one.
tion:
a relaxation
to states
in which is
thediscrete
values of macroscopic
is
linear
and
the
spectrum
is
discrete
.
Some
recent
studies
even
of
the
Hilbert
space
but
may
still
be
much
greater
than
one.
In
pre3
mechanics
makes
quantum
thermalization
in apredictions
and the
integer
powers
theanhamiltonian
.
averag
constants
ofsense
motionanawith functionally
independent
expectation
quantities are
stationary,
universal
with respect
to widely
differing
suggest
that
statistical
mechanics
maypredictions
givepossible
incorrect
vious
work
we
showed of
that
isolated integrable
one-dimen
suggest
that
statistical
mechanics
may
give
incorrect
vious
work
we
showed
that
an
isolated
integrable
one-dimensional
is as large
as the dimension of
the Hilbert
space;
examples
initial conditions,
using statistical
mechanics.
logous
to
thatand
in predictable
which
dynamical
chaos
makesvalues
classical
thermaThe
current
numerical
analytic
evidence
from the stu
for the
relaxation
infeature
such
systems.
Here
we
demonsystem
ofwe
lattice
hard-core
relaxes
to
an
equilibrium
charac- and
for
theoutcomes
of
relaxation
in
such
systems.
Here
demonsystem
of
lattice
hard-core
bosons
relaxes
to an equilibrium
ch
^a ~bosons
include
the
projectors
P
However,
itoutcomes
is notof
obvious
what
of
many-body
quantum
Y
Y
to
the
energy
eigenstates
j
i
h
j
a
a
10
strate
that
a
generic
isolated
quantum
many-body
system
does
terized
not by powers
the
usual
ensemble
but
bysystems
a generalized
Gibbs
lization
possible
.thermalization
For
example,
dynamical
cannot
integrable
suggests
thatensemble
there exists
set of
mechanics
makes
possible
in a sensemany-body
ana- chaos
and
the itself
integer
of Gibbs
the hamiltonian
. not
strate
that
aquantum
generic
isolated
quantum
system
does
terized
by
the usual
Gibbs
but abyminimal
a generalized
relax
to
a
state
well
described
by
the
standard
statistical-mechanical
ensemble.
Instead
of
just
the
energy,
the
Gibbs
exponent
contains
logous toin
that
in isolated
which dynamical
chaos makes
classical
thermaThetime
current
numerical and analytic
evidence
from the study
of size of which is much less than the dime
occur
an
quantum
system,
in
which
the
evolution
pendent
constraints
the
relax
to
a
state
well
described
by
the
standard
statistical-mechanical
prescription.
Moreover,
we
show
that
time
evolution
itself
plays
ensemble.
Instead
of
just
the energy, the Gibbs exponent
co
a
linear
combination
of
conserved
quantitiesthe
occupation
The ETH
lization possible . For example, dynamical chaos itself
cannot
integrable systems4,5suggests that there exists a minimal set of inde11
a
merely
auxiliary
role
in
relaxation,
and
that
thermalization
numbers
of
the
eigenstates
of
the
corresponding
JordanWigner
feris
linear
and
the
spectrum
is
discrete
.
Some
recent
studies
even
of
the
Hilbert
space
but
may
still
be
much
greater
than
one.
occur in an isolatedMoreover,
quantum system,
in which
thethat
time evolution
pendent constraints
the size of which
islinear
much lesscombination
than the dimension of conserved quantitiesthe
have occup
very Id
prescription.
we
show
time
evolution
itself
plays
a
instead
at the level
of individual
eigenstates,
as firsteven
pro- ofmionsthe
number
of may
which
is still
only agreater
tiny fraction
of the
3than one.
is
linear happens
andthat
the spectrum
is discrete
. Some
recentmay
studies
the Hilbert
space but
still
be much
Indimenpre- that an isolated integrable one-dimen
present ther
statistical
mechanics
give
incorrect
predictions
vious
work
we eigenstates
showed
asuggest
merely
auxiliary
role
in
relaxation,
and
that
thermalization
numbers
of
the
posed
by
Deutsch
and
Srednicki
.
A
striking
consequence
of
this
sion
of
the
Hilbert
space.
Yet
this
ensemble
works,
although
the
usual of the corresponding JordanWign
suggest that statistical mechanics may give incorrect predictions
vious work we showed that an isolated integrable one-dimensional
Initial state
the ETH, so
for
the
outcomes
relaxation
infor
such
systems.
Here
we
demonsystem
of
lattice
to anfraction
equilibrium
ch
eigenstate-thermalization
scenario,
confirmed
our
is system
one
does
not,first
for
a wide
ofrelaxes
initial to
conditions
as hard-core
well
asofforwhich
a bosons
for
the
outcomes
of relaxation
inlevel
such
systems.
Here
wesystem,
demonof as
lattice
hard-core
bosonsmionsthe
an equilibrium
characinstead
happens
atof
the
of individual
eigenstates,
pro-variety
number
is stillrelaxes
only a tiny
of the
d
tems.
To
beg
that knowledge
ageneric
single
eigenstate
sufficient
fermionic
system
; does
it also
explains
a recent
experimental
strate
that
a generic
isolated
quantum
many-body
doesto terized
not
by
the usual
Gibbs
ensemble
butnot
by
a generalized
GibbstheGibbs ensemble but by a generalized
12 many-body
13 issystem
strate
that
aofaveragesany
isolated
quantum
many-body
system
terized
by thegasresult,
usual
posed
by
Deutsch
and
Srednicki
.
A
striking
consequence
of
this
sion
of
the
Hilbert
space.
the
in the case
compute
thermal
eigenstate
in
the
microcanonical
absence
of
thermalization
in
the
TonksGirardeau
. Thus,Yet this ensemble works, although
relax to a state well described by the standard statistical-mechanical
ensemble. Instead of just the energy, the Gibbs exponent contains
15by a single
relax
a state
byevolution
thesame
standard
statistical-mechanical
ensemble.
Instead
just
the energy,
theconditions
Gibbs exponent
energy to
window
will well
do, because
give
the
result.plays
although
atsystem,
least someofis
constraints
other
than
the for
conservation
ofvariety
prescription.
Moreover,
wedescribed
showthey
thatall
time
itself
afor
linear
conserved
quantitiesthe
occupation
eigenstate-thermalization
scenario,
confirmed
ourcombination
one
does
not,
a of
wide
of initial
as wellcoa
energy must
kept,
it turns
outcorresponding
only a relatively
limited
If we pierce
an inflated
inside
a vacuum
chamber,
very
a merely
auxiliary
role inballoon
relaxation,
and
that that
thermalization
numbers
ofitself
thebeeigenstates
of the
JordanWigner
fer- of conserved quantitiesthe
16 number
Next,occup
nucle
prescription.
Moreover,
we many-body
show
time
evolution
plays
athatlinear
combination
that
knowledge
of
a
single
eigenstate
is
sufficient
to
fermionic
system
;
it
also
explains
a
recent
experimental
resu
of
additional
conserved
quantities
with
functionally
independent
soon wehappens
find thatatthe
has uniformly
filled the
enclosure
instead
thereleased
level ofair
individual
eigenstates,
as first
promionsthe number of which is still only
a 2:
tinyThermalization
fraction of the dimenFIG.
in
classical
vs
quantum
mechanics.
a,
In
vidual
wave
1
aposed
auxiliary
and
thermalization
numbers
of
the
eigenstates
ofin
thethe
corresponding
JordanWign
expectation
values
are needed;
adding
further
ones
is redundant.
andmerely
that
air
molecules
haverole
attained
therelaxation,
Maxwell
velocityof
districompute
thermal
averagesany
eigenstate
inthat
the
microcanonical
absence
of
thermalization
TonksGirardeau
gas
.
by the
Deutsch
and Srednicki
. Ain
striking
consequence
this
sion
of
the Hilbert
space.
Yet this ensemble
works,
although
the
usual
classical mechanics, time evolution constructs the thermal state from
[20]. Then
bution,
the
width
of
which
depends
only
on
the
total
number
and
As
it
is
not
clear
which
sets
of
conserved
quantitiesand
some
are
eigenstate-thermalization
confirmed
for our system,
is
does not,
for
a wide
variety ofmionsthe
initial
conditions
as well
as for
instead
happenswill
atscenario,
the
of individual
eigenstates,
as
first
pronumber
ofawhich
still only
tiny
fraction
of the d
an initial
state
that
generally
bears
no is
resemblance
toathe
former.
energy
window
do,level
because
they
all give
theone
same
result.
although
at do
least
constraints
other
than
the conservati
quantum sy
energy
of the air of
molecules.
balloon
shapes,
always presentconstrain
relaxation
and experimental
which
not,result,
itsome
becomes
that
knowledge
a single
many-body
eigenstate
sufficient toor fermionic
system ; it also explains
ab,recent
the
12Different
13 isplacements,
In
mechanics,
according
to theensemble
eigenstate thermalizaposed
by
Deutsch
and
Srednicki
. Adistributions.
striking
consequence
ofto
this
sion
ofquantum
the
Hilbert
space.
Yetout
this
although
the
satisfy the
piercing
all averagesany
leadan
to the
sameeigenstate
spatial
and
even chamber,
more
urgent
determine
whether
or not
generic
isolated
energy
must
be
kept,
it
turns
that
only a works,
relatively
limited
nuE
If wepoints
pierce
inflated
balloon
inside
a vacuum
very
compute
thermal
invelocity
the microcanonical
absence
of thermalization
in the
TonksGirardeau
gas
.
Thus,
tion hypothesis, every eigenstate of the Hamiltonian always implic- 15
Classical
physicswill
explains
this thermodynamical
universality
quantum
relaxconstraints
usual
thermal
state.
Thisa question
More
genera
energy
window
do, because
they allscenario,
giveair
the has
sameconfirmed
result. as fol- filled
eigenstate-thermalization
for
our
system,
isto the of
although
atsystems
least
some
other
than not,
the
conservation
of
one
does
for
wide
ofbetween
initial
asindepe
well
additional
conserved
quantities
with
functionally
soon
we
find
that
the
released
uniformly
the
enclosure
itly
contains
a thermal
state.
Thevariety
coherence
theconditions
eigenstates
lows
:
almost
all
particle
trajectories
quickly
begin
to
look
alike,
even
has
received
increased
theoretical
attention
recently,
because
of
the
energy must be kept, it turns out that only a relatively limited16number
If we pierce an inflated balloon inside a vacuum chamber, very
regime,resu
the
initially hides
it,
but time;are
dynamics
revealsadding
it through
dephasing.
that
knowledge
of
aairsingle
many-body
eigenstate
isvelocity
sufficient
to
fermionic
system
it also
explains
a recent
experimental
values
needed;
further
ones iswhich
redunda
and
that
air
molecules
have
attained
the
Maxwell
distriif their
initial
points
very different,
because
nonlinear
equations
levels ofconserved
isolation
andexpectation
control
possible
in
experiments
ofhigh
additional
quantities
with functionally
independent
soon
we
findthe
that
theare
released
has uniformly
filled the enclosure
in1tur
compute
thermal
averagesany
eigenstate
in
the
microcanonical
absence
of
thermalization
in
the
TonksGirardeau
gas
.
drive
them
to
explore
the
constant-energy
manifold
ergodically,
covwith
ultracold
quantum
gases.
However,
despite
numerous
studies
of
expectation
values
are
needed;
adding
further
ones
is
redundant.
and
that
the
air
molecules
have
attained
the
Maxwell
velocity
distribution,
the width
of which
depends
only on
the specific
total number
and
As it isonnot
clear
which
sets of conserved quantitiesand
som
chaotic syst
ering
it
uniformly
with
respect
to
precisely
the
microcanonical
meamodels,
there
is
not
yet
consensus
how
or
even
whether
bution,
the
width
of
which
depends
only
on
the
total
number
and
As
it
is
not
clear
which
sets
of
conserved
quantitiesand
some
are
energy
window
will
do, because
theyplacements,
allquantities
giveorshapes,
thealways
same
result.
although
atenergy
least
somesys-constraints
than
the
integrability
energy
of
the
airsystem
molecules.
Different
balloon
placements,
always
presentconstrain
relaxation
and
do conservat
not,
it bec
Discussion)
centered
at E0 , other
and
thewhich
normalsure. However,
ifmolecules.
the
possesses
further
conserved
relaxation
to the usual or
thermal
values
occurs
fordonon-integrable
energy
of the
air
Different
balloon
shapes,
presentconstrain
relaxation
and
which
not, it window
becomes
observables
energy
be
kept,
it
turns
out that
only a with
relatively
limited
nu
pierce
an
inflated
balloon
inside
aother,
vacuum
chamber,
thatIfarewe
functionally
independent
of
the
hamiltonian
and
each
piercing
points
all lead
to the
sameto
spatial
and
velocity
distributions.
temsmore
. distributions.
Theurgent
conventional
wisdom
isization
thatmore
itmust
does
, but
some
recent
even
tovery
determine
whether
orN
not, E
generic
is
theisolated
number
of energy
eigenstates
E
piercing
points
all
lead
the
same
spatial
and
velocity
even
urgent
to
determine
whether
or ennot generic
is

Thermalization and its mechanism for generic

isolated quantum systems
14

1,2

37

8,9

14

14

1,2

37

10

thermal

thermal

thermal

8,9

11

4,5

coherence

time

time

Thermal state

14

dephasing

chaos

11

12

4,5

13

thermal

thermal

thermal

15

16

coherence

10

EIGENSTATE THERMALIZATION

12

13

15

16

10

1,2,17

18,19

8,9

Initial state