Renormalization
Group, Non-Gibbsian states, their
relation-ship
and
further
developments.
Aernout C.D.
van
ENTER1,
(1) Centrum
voor
theoretische natuurkunde R.U.G. Nijenborgh 4,$\mathit{9}7\mathit{4}7AG$, Groningen, the
Nether-lands.
Abstract:
We review whatwehave learnedaboutthe“Renormalization Grouppeculiarities” whichwerediscovered
morethan twentyfiveyears agobyGriffiths and Pearce. We discuss which ofthequestions they asked
have been answered and whichones arestill widelyopen.
Theproblems theyraised have ledto the studyof non-Gibbsian states(probability measures). We also
mention somefurther related developments, which find applications in nonequilibrium questions and disorderedmodels.
Keywords: Renormalization-Group$\mathrm{p}\mathrm{e}\mathrm{c}\mathrm{u}\mathrm{l}\mathrm{i}\mathrm{a}\mathfrak{l}\mathrm{r}\mathrm{i}\mathrm{t}\mathrm{i}\mathrm{a}\mathrm{e}$, non-Gibbsianmeasures.
1
Introduction
Morethantwentyfive yearsago, GriffthsandPearce$[30, 31]$discovered
some
unexpectedmathemat-ical difficulties in rigorously implementing
many
of the generally used real-space $\mathrm{R}\epsilon \mathrm{n}\mathrm{o}\mathrm{r}\mathrm{m}\mathrm{a}\mathrm{l}\mathrm{i}\mathrm{z}\mathrm{a}\mathrm{t}\mathrm{i}\mathrm{o}\mathrm{n}$ Grouptransformations as
mapson a
space of Hamiltonians.Inthisshort reviewI plan to
assess
whatwe
have learned about theseproblemssince then. The relevantobjectswhich
haveprovided themostinformation
are
theso-callednon-Gibbsian
measures.
Indeed, renormalizingGibbs
measures
associated tosome
Hamiltoniancan
result ina
non-Gibbsianmeasure
for whichno
“resonable” renormaliz\’eHamiltoniancan
befound.We will
see
that Renormalization-Group maps cannot be discontinuous, although they can beill-defined.
Moreover, in
any
region of the phase diagram the question whethera
particulartransforma-tion is well-defined or ill-defined turns out to be highly non-trivial. The ill-definedness of various
Renormalization-Groupmaps
can
beexpressedin the violation of the propertyof“quasilocality” inthe renormalized states. Thestudyand classification ofsuch non-quasilocalstates ($-\mathrm{n}\mathrm{o}\mathrm{n}$
-Gibbsian
measures–) has led also to various results of mathematical interest,
some
ofwhichwe
willmen-tion
further
on.
Some papers covering the area of non-Gibbsianness and Renormalization-Grouppeculiarities, also treatingfurtherrelated material
are
[104, 103, 98, 62, 112, 100, 21, 22, 23,83, 51,99, 48, 49, 38, 46] and referencestherein. The first systematic and extensive (an almost 300-page
paper...) follow-up study appeared in
1993
[103]. Althoughmany
further results have since beenproven, theconceptualpoint of view Ipresenthereis still essentially based
on
thatpaper.
BeyondRenormalizationGroup applications,non-Gibbsian
measures
have beenfound
in variousothercircumstances. Iwill shortlymention
some
of those. At the end of thepaper
Imention
some
furtheropenproblems.
2
Gibbs
measures
and
quasilocality
Inthis section
we
willdescribesome
definitions and factswe
will need about the theoryofGibbsWe will consider spin systems
on
a
lattice $Z^{d}$,
where in mostcases we
will takea
single-spinspace $\Omega_{0}$ which is finite. The configuration space ofthewhole system is $\Omega=\Omega_{0}^{Z^{d}}$
.
Configurationswill be denoted by small
Greek
letters suchas
$\sigma$or
$\omega$, and their coordinates at lattice site $\mathrm{i}$are
denoted by $\omega(i)$ or a(i). A (regular) interaction $\Phi$ isa
collection offunctions $\Phi_{X}$on
$\Omega_{0}^{X},$$X\in Z^{d}$whichis translation invariant and satisfy:
$\Sigma_{0\in X}|\Phi_{X}|_{\infty}<\infty$ (1)
FormallyHamiltonians
are
given by$H^{\Phi}=\Sigma_{X\in Z^{d}}\Phi_{X}$ (2)
Under the above regularity condition thesetype
of
expressions makemathematical
sense
if thesum
is takenover all subsets
havingnon-emptyintersections
witha
finite
volume A. For regularinteractions
one can
define Gibbs
measures
as
probabilitymeasures
on
$\Omega$havingconditional
probabilitieswhich
are
described intermsof appropriateBoltzmann-Gibbs factors:$\frac{\mu(\sigma_{\Lambda}^{1}|\omega_{\mathrm{A}^{\circ}})}{\mu(\sigma_{\Lambda}^{2}|\omega_{\Lambda^{\mathrm{c}}})}=\exp-(\Sigma_{X}[\Phi_{X}(\sigma_{\Lambda}^{1}\omega_{\Lambda^{\mathrm{c}}})-\Phi_{X}(\sigma_{\Lambda}^{2}\omega_{\Lambda^{\mathrm{c}}})])$ (3)
for eachvolume$\Lambda,$$\mu$-almosteveryboundarycondition$\omega_{\Lambda^{t}}$ outside A and eachpairof configurations $\sigma_{\Lambda}^{1}$ and$\sigma_{\Lambda}^{2}$inA.
As
longas
$\Omega_{0}$iscompact, there always exists at leastone
Gibbsmeasure
for every
regular interaction;the existence of
more
thanone Gibbs
measureisone
definition of theoccurrence
of a first-order phasetransition of
some
sort. Thus the map $\mathrm{h}\mathrm{o}\mathrm{m}$interactions tomeasures
isone
to$\mathrm{a}\mathrm{t}- \mathrm{l}\mathrm{e}\epsilon\epsilon \mathrm{t}$-one.
EveryGibbsmeasure
has the property that (forone
of itsversions) itsconditionalprobabilities
are
continuous functions of the boundary condition $\omega_{\Lambda^{\epsilon}}$,
in the product topology.It is
a
non-trivialfact
that this continuity, whichgoes
by thename
“quasilocality”or
“almostMarkovianness”, in fact characterizes the Gibbs
measures
$[95, 47]$,once
one
knows that all theconditional
probabilitiesare
boundedaway
fromzero
(that is, themeasure
is nonnullor
has thefinite
energy property). Insome
examples it turns out to be possible to che& this continuity(quasilocality) property quite explicitly. Ifa
measure
is aGibbsmeasure
for aregular interaction,this interactionis essentially uniquely determined. Thus the map ffom
measures
to interactions isone
toat-most-one.A second characterization of Gibbs
measures uses
the variational principle expressing that inequilibrium a system minimizes its
&ee
energy.
A probabilistic formulationof
this fact naturallyoccurs
in terms ofthe theory oflarge deviations. A (third level) large deviation rate function isup to
a
constant anda
sign equal toa
free energy density. To be precise, let $\mu$ bea
translationinvariant Gibbsmeasure, and let$\nu$be
an
arbitrarytranslationinvariantmeasure.
Then the relativeentropy density$i(\nu|\mu)$ canbe
defined as
the limit:$i( \nu|\mu)=\lim_{\Lambdaarrow Z^{d}}\frac{1}{|\Lambda|}I_{\Lambda}(\nu|\mu)$ (4)
where
$I_{\Lambda}( \nu|\mu)=\int log(\frac{d\nu_{\Lambda}}{d\mu_{\Lambda}})d\nu_{\Lambda}$ (S)
and $\mu_{\Lambda}$ and $\nu_{\Lambda}$
are
therestrictions of
$\mu$ and $\nu$ to $\Omega_{0}^{\Lambda}$.
It has theproperty that $i(\nu|\mu)=0$ if andonly ifthe
measure
$\nu$ isa
Gibbsmeasure
for thesame
interactionas
the basemeasure
$\mu$.
Wecan
use
this result inapplicationsifwe
knowfor
examplethata
knownmeasure
$\nu$ cannot bea Gibbs
measure
forthesame
interactionas some measure
$\mu$we
wantto investigate. Forexample, if$\nu$ isa
point measure,
or
if it is thecase
that $\nu$ isa
productmeasure
and$\mu$is not,we
can
conclude$\mathrm{h}\mathrm{o}\mathrm{m}$
the statement: $i(\nu|\mu)=0$, that $\mu$ lacksthe Gibbsproperty.
For another method ofproving that
a
measure
is non-Gibbsianbecauseofhaving the “wrong”3
Renormalization-Group
maps:
some
examples
Wewill mostly consider the standard nearestneighbor Isingmodelwith (formal) Hamiltonian
$H=\Sigma_{<i,j>}-\sigma(i)\sigma(j)-h\Sigma_{i}\sigma(i)$ (6)
at inversetemperature $\beta$
.
The magneticfield strength is $h$.
The dimension $d$ inwhat followswillbeat least 2.
Wewill considervariousreal-space Renormalization-Group
or
block-spintransformationswhichactonthe IsingGibbs
measures. Our
firstquestionis tofindthedomain of definitionof suchtrans-formations (that is: Is the first step ofthe renormalization
program
well-defined?). The question then is to find the renormalized interaction, that is theinteraction associated
to the transformedmeasure.
This thenshould definethetransformation atthe level of interactions(Hamiltonians).Although in applications the transformation needs to be iterated, and
one
would like to knowabout fixed points, domains of attraction etc, we will mostly restrict ourselves to considering a
single transformation. Existence ofthe first step is of
course
necessary but far from sufftcient forjustification of
an
iterative procedure. Forsome
recent work considers what happens after manyiterations,
see
[5, 6, 7]. We mentione.g.
that sometimes,even
ifthe first stepis ill-defined, afterrepeatedtransformations
a
transformed interactioncan
befound [78].In contrast to what
one
might atfirst
believe, the critical pointcan
beeither outsideor
insidethe domainofdefinition oftheRenormalizationGroup$\mathrm{o}\mathrm{p}\overline{\mathrm{e}}\mathrm{r}\mathrm{a}\mathrm{t}\mathrm{o}\mathrm{r}$, and does in generalnotplayany
special role.
We divide the lattice into
a
collection of non-overlapping blocks. A Renormalization-Grouptransformationdefinedat the levelof
measures
willbea
probabilitykernel$T(\omega’|\omega)=\mathrm{I}\mathrm{I}_{\mathrm{b}\mathrm{l}\mathrm{o}\mathrm{c}\mathrm{k}\mathrm{s}}T(\omega’(j)|\omega(i);i\in \mathrm{b}\mathrm{l}\mathrm{o}\mathrm{c}\mathrm{k}_{j})$ (7)
This
means
that the distribution ofa
block-spin depends onlyon
the spinsin thecorresponding block, inother words the transformation is local inrealspace. Thecase
ofa
deterministic transformationisincluded, by having
a
$\mathrm{T}$whichis
eitherzero
or one.
Renormalization-Group methods
are
widelyinuse
to study phasetransitions and in particularcritical phenomena ofvarious sorts (see for example [113, 68, 17, 26]), as
was
alostestified
by theworkshop.
Some
goodrecent referencesinwhich thetheoryis explained, mostly at aphysicallevelofrigour, but including
some
more
careful statements about what actually has and has not beenproven
are
[29, 2, 1].1)
One
class ofexampleswe
willconsider are(linear) block-averagetransformations. Thismeans
that the block-spins
are
theaveragespinsineachblock. Appliedto Ising systems theysuffer fromthefact that the renormalized system hasa differentsingle-spin spacefromthe original
one.
Despitethis objection, the linearity makes these maps mathematically rather attractive, and they have often
been considered. As long
as we are
not iteratingthe transformationwe
need notworry too muchabout the single-spin space changing, but
see
[5, 6, 7].2) Majorityrule and
Kadanoff transformations.
In the
case of
majority ruletransformations
[84] applied to blocks containgan
odd number ofsites, the block spin is just given by the sign of the majority of the spins in the block. These
transformations have been chosen often because of their numericaltractability.
The Kadanoff transformation is
a
softversion
(aproper
exampleofa
stochastic transformation)of themajorityrule:
$T(\sigma’(j)|\sigma(i);i\in \mathrm{b}\mathrm{l}\mathrm{o}\mathrm{c}\mathrm{k}_{j}))=C\exp M’(j)\Sigma_{i\in \mathrm{b}\mathrm{l}\mathrm{o}\mathrm{c}\mathrm{k}_{f}}\sigma(i)]$ (8)
In the limitin which$p$goesto
oo
theKadanoffmapreducestoa
majorityrule transformation.Once
block),
one
givesa
prescription what happens in such acase.
For example, apopular prescriptionis thento flip
a coin
to decide.Majorityruletransformations mapIsing systemsontoIsingsystems, and have at least inprinciple
thepossibility to beiterated, and tohavenontrivialfixed points.
$3)\mathrm{D}\mathrm{e}\mathrm{c}\mathrm{i}\mathrm{m}\mathrm{a}\mathrm{t}\mathrm{i}\mathrm{o}\mathrm{n}\mathrm{s}$ and projections.
Wewillcall
a
“decimation”takingthemarginalofameasure
restrictedto the spinson a
sublatticeofthe
same
dimensionas
the original system, (thus the block-spinsare
the spins insome
periodicsub-lattice).
A “projection” will
mean
takingthe marginal toalower-dimensionalsublattice. Projectionsare
notRenormalization-Groupmaps proper,but share
some
mathematical propertiesofRenormalization-Group maps. See [69, 70, 90].
Althoughdecimation transformationshave the advantages bothofbeing linearand ofpreserving
the single-spin
space,
infinite iteration hasthe disadvantage that criticalfixed points won’toccur.
However, this problemdoesnot show
up
aftera finite
number ofapplicationsof
these maps,so we
will here not too much
worry
about it.4
$\mathrm{P}\mathrm{e}\mathrm{c}\mathrm{u}\mathrm{l}\mathrm{i}\mathrm{a}\mathrm{r}\mathrm{i}\mathrm{t}\mathrm{i}\mathrm{e}\mathrm{s}:\mathrm{T}\mathrm{h}\mathrm{e}$investigations of
Grifflths and Pearce
Griffiths
and Pearce $[30, 31]$seem
to be the first investigatorswho looked seriously at the questionwhetherrenormalizedHamiltoniansexist in
a
precisemathematicalsense.
Theyfoundthatforsome
real-space
transformations
like decimationor
Kadanoff transformations in the low-density regime(that
means
strong magnetic fields in Ising language) the Renormalization-Groupmap maps theIsing Gibbs
measure on a
Gibbsmeasure
foran
in principle computable interaction. They alsofound, bothat phasetransitions and
near areas
ofthe phase diagram wherefirst order transitionsoccur, that there
are
regimes where the formal expression for renormalized interactions behavedin
a
peculiar way. These “peculiarities”were
found for decimation, Kadanoff, and majority ruletransformations.
The problem underlyingthe
peculiaritiesis
theoccurrence
of phasetransitions
inthe system,
once
it is constrained (or conditioned) by prescribingsome
particular, rather atypical,block-spinconfiguration. This
means
that therecan
exist for these block-spinconfigurationslong-range correlations in the presumed “short-wavelength degrees of haedom”-
or
“internal spins”-,which
are
to beintegratedoutina
Renormalization-Groupmap.Intheir
paper
GriffithsandPearcediscussvariouspossible explanations ofthese “peculiarities”:$\mathrm{P}\mathrm{l})\mathrm{T}\mathrm{h}\mathrm{e}$renormalized interactionmight not exist,
$\mathrm{P}2)\mathrm{i}\mathrm{t}$ mi$g\mathrm{h}\mathrm{t}$exist but be
a
singularfunctionof
theoriginal interaction, $\mathrm{P}3)\mathrm{i}\mathrm{t}$ might be non-quasilocal,or
$\mathrm{P}4)\mathrm{t}\mathrm{h}\mathrm{e}$ thermodynamic limit might beproblematic.
Shortly after, the problem
was
studied by Israel [42]. He obtained (very) high-temperatureexistenceresults, including approach to trivial fixedpoints,
as
wellas
an analysisof thedecimation transformationat lowtemperature,indicating strongly that inthatcase
therenormalized interaction
doesnot exist. Israel’s resultsconvinced Griffiths
thatinfact possibility$\mathrm{P}\mathrm{l}$)$-\mathrm{n}\mathrm{o}\mathrm{n}$-existence ofthe
renormalized interaction– applies [32]; however, it
seems
that most authorsaware
oftheirworkinterpret\’ethe
Griffiths-Pearce
peculiaritiesas
singular behaviouroftherenormalized interactions(possibility P2)$)$
.
See for example [39, 11, 41, 96]. Many authors did noteven
showawareness
that thedeterminationofrenormalized interactionspresentedproblems beyond
mere
computationaldifficulty.
See
theAppendix forsome
illustrationsofthis point.Deep inthe uniquenessregime theRenormalization-Group procedure appearedto be
well-defined
At that stage, various open questions formulated by Griffiths and Pearceand Israel
were
still left:Ql) What is thenature ofthe “peculiarities”?
Q2)
Can one say
anything about the criticalregime?Q3) Do different transformations exhibit similar behaviour? For example,
are
decimation,block-averaging, majority and momentum-space Renormalization-Grouptransformationssimilar regarding
the
occurrence
of “peculiarities”?Q4)Asthe“peculiarities”are dueto rather atypical spin-configurations,
can one
make the Renormalization-Groupenterprise work, by considering only typical configurations, and thuswork with appropriateapproximations? Or,
more
generally, is therea
framework in whichone
can
one
implement thewhole
Renormalization-Group machineryina
mathematicallycorrect way?5
Answered and unanswered
questions
About question Ql –the nature ofthe peculiarities–
we
have acquiredsome
more
insight. In[103] the Griffiths-Pearcestudy
was
takenup andfurtherpursued. Weprovideda
ratherextensivediscussion of the physical interpretation of these issues,
as
wellas
provinga
number of variedmathematical results. It was found for example, making
use
ofthe above-mentioned variationalcharacterization ofGibbs measures, that the peculiarities could not be due to discontinuities in
the Renormalization-Group maps. In fact the Renormalization Group map from interactions to
interactions is onetoat-most-one. The underlyingargument isthatif
one
transformstwomeasures,theirrelativeentropy tendstoshrink, for
a
wide class of transformations. Hencerenormalizingtwophases (different Gibbs measuresforthesameinteraction,having relativeentropy densityzerowith
respecttoeachother) cannotresult intwo Gibbs
measures
for differentinteractions,for which theserelative entropy densities needto be
non-zero.
However, in the “peculiar”
cases
considered byGriffiths
and Pearcethe renormalizedmeasures
allhave conditional probabilities with points-spinconfigurations-of essentialdiscontinuity. That
is, they
are
non-Gibbsian. See also [46]. Thus a renormalized Hamiltonian does not exist. Thisdespite many attempts to compute $\mathrm{t}\mathrm{h}\mathrm{e}\mathrm{s}\mathrm{e}-\mathrm{n}\mathrm{o}\mathrm{n}$-existent –renormalized Hamiltonians, and the
variousphysicallyplausibleand intuitively convincing conclusions, derived fromsuch approximate computations.
In fact, those constrained block-spin configurations, pointed out by Griffiths and Pearce, for
which the internal spins have phase transitions are precisely the points of discontinuity
–non-quasilocality-for
some
conditional probability. The observation that the “peculiarities”were
duetotheviolationof thequasilocalitycondition
was
inessence, thoughsomewhat ina
slightly implicitway, already made in Israel’s analysis.
At first the
non-Gibbsian
exampleswere
found ator near
phase transitions, at sufficiently low temperatures. Then itwas
found
[101] that decimation applied to many-statePotts
models givesnon-Gibbsian
measures
alsoabovethe transitiontemperature.Somewhatsurprisingly,it turned out that
even
deepinthe uniqueness regime,thenon-Gibbsianness
can
occur.
This happensforexampleat low temperatureforblock-average [103],and majorityrule[101] transformationsin strong external fields, and it iseven possibleto devise transformationsfor
which this happens at arbitrarily high temperatures [97].
On
the other hand, it turns out thatfor
Gibbs
measures
well inthe uniqueness regime,a
repeated application ofa
decimationtransfor-mation,
even
after composing with another Renormalization-Group map, leads again toa
Gibbsmeasure, although applying thedecimation onlya fewtimesmay resultin
a
non-Gibbsianmeasure
$[78, 79]$
.
Forrelated work in this directionsee
also [4]. As mentionedbefore, physicallythis meansAbout critical points (questionQ2), Haller and Kennedy [38] obtained thefirstresults, proving
both for
a
decimation anda Kadanoff
transformationexamplethata
single Renormalization-Groupmap
can
mapan area
includinga
critical point to a set ofrenormalizedinteractions.
Thereare
strongindications for similar behaviour for other transformations [45, 3, 12, 86]. The indications
are partly numerical,however, and fall short of
a
rigorous proof. See also the numerical work of[87].
On
the other hand, counterexamples wherea
transformed criticalmeasure
is non-Gibbsianalsoexist $[97, 98]$
Another criticalregimeresultis the observationof[18], thatmajorityrulescalinglimitsof critical
points abovethe upper critical dimension (when the critical behaviour is likethat of
a
Gaussian)are
non-Gibbsian.It
was
found thatnon-Gibbsian
measures can
alsooccur as a
result ofapplying momentum-spacetransformations
[100].Theconclusion
of
allthe above is thatdifferent transformations can
havevery
different behaviour.Thisis
a
sort ofanswer
toquestionQ3, althoughinprinciplenota
very
informativeone.
Infact
forapplicability, if not for existence, something like this
was
alreadyexpected (compare for exampleFisher’s [25] remarks
on
“aptness” andfocusability).Regarding question Q4) –to find the right setting for implementing Renormalization-Group
Theory–the issue is still essentiallyopen. One approachwhich
was
stimulated by thelate $\mathrm{R}.\mathrm{L}$,Dobrushin is related to the observation of
Griffiths
and Pearce that the block-spin configurationsresponsible for the peculiarities (the discontinuity points)
are
rather atypical. By removing themfrom configurationspace,
one
might hopeto be left witha
viabletheory. Such investigations haveled to the notions of “almost”, “intuitively weak”or “weak” Gibbs measures, the study ofwhose
properties isbein$g$ actively pursued [15, 16, 10, 69, 110, 23, 73, 71, 24, 92, 58, 77, 64, 5, 6, 7, 111].
This approach is somewhat alongthe lines ofGrifliths’ and Pearce’spossibility P3).
See
also thenext
section. Whether
it is possible to describe Renormalization-Group flows inspaces
of suchinteractions,whilekeeping
a
continuous connection betweeninteractions
describinga
positivelyand
a
negativelymagnetized state, is not at all clear. A warning signis that the variationalprinciplecan
beviolated fora
weaklyGibbsian
situation [51].As for projections, it isknownthat onthe phase-transition lineofthe2-dimensional Ising model
the projection to $Z$ of any Gibbs
measure
isnon-Gibbsian
[90]. In the whole uniqueness regime,except possibly at the critical point, this projection results in
a Gibbsian
measure
[74, 62, 63].In three dimensions the projected
measures
to two-dimensionalplanesare
again non-Gibbsianon
the transition line [24, 75, 70]; however, now, due to the presumably existing surface (layering)transition between different “Basuev states” (states with different layer thicknesses),
one
expectsthat theprojected
measures
also ina
smallfieldwill benon-Gibbsian $[62, 63]$.
The compositionof
a
projection anda
decimation in the phase transition region gives rise toa new
phenomenon, namely the possibility ofa state-dependent result. Thetransformedplus andminus
measures
are
Gibbsmeasures
for differentinteractions,while thetransformedmixedmeasures
are
non-Gibbsian [66]. Theirrelative entropy densitis with respect to each otherarepositive.6
Further results
on
non-Gibbsian
measures:
$\mathrm{F}\mathrm{K}$,
the
colors
of
H\"aggstr\"om, fuzzy Potts,
quenched disorder
and
non-equilibrium
Further investigations in which non-Gibbsian
measures were
displayed, have beendone about therandom-cluster models of Fortuin and Kasteleyn [88, 33, 8, 36, 92], the related coloring model
of H\"aggstr\"om, [35], about the Fuzzy Potts image analysis model [76, 37, 34], and about various
it is useful to note that
an
infinite-temperature Glauber dynamicscan
be viewed asa
single-sitestochastic “renormalization-group” map-. [94, 112, 80, 79, 99, 60, 72, 13, 59].
Also, joint quenched
measures
of disordered systems, have been shown sometimes to benon-Gibbsian [109, 107, 48, 49, 106], affecting the Morita approach to disordered systems $[82, 53]$
.
Inthis last case, the peculiarity
can
be so strong–and it actually is in the 3-dimensional randomfield Isingmodel-as to violatethevariationalprinciple. This
means
inparticularthat the (weaklyGibbsian) interactions belongingto the plus state and the minus state
are
different, despite theirrelative entropydensity being
zero.
Non-Gibbsianness
heremeans
that the quenchedmeasure
cannotbewritten as
an
annealedmeasure, thatisa
Gibbsmeasure
on
the joint space ofspinsand disorder variables forsome
“grand potential”, suchas
Moritaproposed.The
non-Gibbsian
character of the variousmeasures
consideredcomes
oftenas an
unwelcomesurprise. A description in terms
of
effective, coarse-grainedor
renormalized potentials is oftenconvenient, andeven
seems
essential forsome
applications. Thus, the fact that sucha
descriptionisnot $\mathrm{a}\mathrm{v}\dot{u}\mathrm{l}\mathrm{a}\mathrm{b}\mathrm{l}\mathrm{e}$can be a
severe
drawback. As remarked earlier there have been attempts to tamethe
non-Gibbsian
pathologies, and herewe
wantto givea
short comparisonofhowfarone
gets withsome
of thoseattempts.1)The
fact
that the constraints which actas
pointsofdiscontinuity often involveconfigurationswhich
are
veryuntypicalfor themeasure
underconsideration,suggesteda
notion of almostGibbsianor
weaklyGibbsian
measures.
Theseare measures
whose conditionalprobabilitiesare
eithercontin-uous
onlyon a
set of fullmeasure or can
be written in terms ofan
interactionwhichis summable onlyon
a
set offull
measure.
Intuitively, thedifference is thatinone case
the “good” configurationscan
shieldoff all influences from infinitely far away, and in the othercase
only almost all influences.Theweakly
Gibbsian
approachwas
firstsuggested by Dobrushintovariouspeople; his own versionwas
published only later[14, 15, 16]. An earlydefinition of almost Gibbsiannessappeared in printin [67],
see
also [24, 76, 77, 69, 111, 51, 83] for further developments.Some
examplesofmeasures
which
are
at the worst almost Gibbsianmeasures
in thissense are
decimatedor
projected Gibbsmeasures
inan
external field, random-clustermeasures
on
regular lattices, and low temperaturefuzzyPotts
measures.
In therandom-clustermeasures
one
can
actually identify explicitly all bondconfigurations which give rise to the non-quasilocality. They
are
preciselythose configurations inwhich (possibly after
a
local change)more
thanone
infiniteclustercoexist.Ona tree, because ofthe possible coexistenceofinfinitely many infixiteclusters with positive
probability, the random-cluster
measure can
violate theweak non-Gibbsianness condition and bestronglynon-Gibbsian [36].
Dobrushin [14, 77, 15, 16] showed that for
a
projected pure phaseon
the coexistence line ofthe2-dimensional Ising model itis possible to find
an
almost everywheredefinedinteraction, hencethese
measures are
weaklyGibbsian. Hisapproach, which is vialow-temperatureexpansions,pro-vides
a
way of obtaining good control for thenon-Gibbsian
projection. For similar ideas in a Renormalization-Group settingsee
[10,58, 65].For
some
Renormalization-Group examplesan
investigationvia lowtemperatureexpansionsintothe possibilityofrecognizing
non-Gibbsianness
was
started in [89].Anothersimple counter-example
of a
strongly non-quasilocal measure,where eachconfigurationcan
actas a
point ofdiscontinuity, isa
mixture oftwoGibbs
measures
fordifferent
interactions[108].
2) Stabilityunder decimation (andother transformations).
In $[78, 79]$ it
was
shown how decimating once-renormaliz\’enon-Gibbsian measures
results inGibbs
measures
againaftera
sufficiently large numberofiterations. Theseoften decimatedmeasures
are
inthehigh-temperatureregime, in which the usually applied Renormalization-Groupmapsare
well-defined
(thisdoes nothold true for all maps though, inviewofthe highly non-linear exampleOn
theother hand,inexampleswherethenon-Gibbsian
propertyis associated withlargedevia-tionproperties which
are
not compatible witha
Gibbsiancharacter(thisholdsforexampleforspin-projected Gaussians, invariant
measures
ofthe voterand theMartinelli-Scoppola mode1[56, 57, 80]$)$the
non-Gibbsian
propertysurvives all sort oftransformations$[79, 108]$. The argument is that whensome
obviously non-Gibbsianmeasure
hasa
rate functionzero
with respect to themeasure
underconsideration, thisproperty is generally preserved ([103]
Section
3.2and 3.3).Thefamilyofspin-projected
Gaussians
include scaling limits for majority-ruletransformationsin high dimensions [18]. The
transformation
ofthose scaling limits is, heuristically, interpretedas
making
a move
froma
fixed point in what is usuallycalleda
“redundant” direction (cf. Wegner’scontributionto [17]$)$ in
some
space of Hamiltonians. Here, of course, the wholepoint is that suchHamiltonians
do notexist.
3) The two criteria mentioned above
are
distinct. A simple one-dimensional example due toJ.
van
den Berg [65] givesa
one-dependentmeasure
which hasa
set ofdiscontinuitypoints offullmeasure, but due to the one-dependence the
measure
becomes after decimation independent andtherefore trivially Gibbsian. In the opposite direction, there exist examples of
measures
whosenon-Gibbsiannessis robust, although they
are
weakly, andevenalmost, Gibbsian[110].7
Conclusions and
some
further
open
problems
Wehave by
now
managed well tounderstandtheGriffiths-Pearcepeculiaritiesin thesense
thatwe
can
$\mathrm{i}\mathrm{d}\mathrm{e}\mathrm{n}\mathrm{t}\mathrm{i}\mathfrak{h}$ mathematically theirnature. However,how to get aroundthem,andmake Renormalization-Group theorymathematically respectable, is stilla
task requiringa
lotoffurther
work.Renormalization-Group ideas have
of
course
often beeninspirational,also for various rigorous analyses. Renormahzation-Groupimplementationson
spin models,whichboth for
numerical and pedagogical conveniencehaveoften been treated by Renormalization-Group methods,
run
intodifficulties
whichseem
hard to avoid.Animplementation of Renormalization-Group
ideas on
contour models looksmore
promising,at least for thedescriptionof
first-order
phasetransitions $[27, 9]$.
On
the otherhand,havingdifferentdescriptions anddifferent mathematicalobjectsat hightem-peratures (lattice spin interactions), at low temperatures (Peierls contour models), and yet again
somethingdifferent at criticaltemperatures (like possibly intwo dimensions SLE, –the
Schramm-Loewner Evolutions,
see
e.g. [93, 54, 91, 44] -)seems
nota
good way to describe the textbookRenormalization Group flowdiagrams, wherethese three regimes typically
are
all included.There-fore, although
we
understand the problems better, solutionsseem
still to be far out, and resultsof Renormalization Group in the Mathematical Sciences need to develop
a
lot further before the situation issatisfactory.Innon-equilibriumstatistical mechanicsthere
are
stillmanyopenquestionsabout theoccurrence
of non-Gibbsian
measures.
Whetherone can
ascribean
effectivetemperature ina
non-equilibriumsituation is
a
topic of considerable interest, (alsoin thephysics literature,see
e.g. [85]). Theterm non-Gibbsianor non-reversibleisoften used for invariantmeasures
insystems inwhich there isno
detailedbalance [61, 20, 19]. It is
an open
questiontowhatextent suchmeasures are
non-Gibbsian
in the
sense we
havedescribed here. It has been conjectured that suchmeasures
for which thereis
no
detailed balanceare
quite generally non-Gibbsianinsystems witha
stochasticdynamics,see
for example [57]
or
[20], Appendix 1;on
the other hand it has been predicted that non-Gibbsianmeasures are
ratherexceptional ([61], Open problem IV.7.5,p.224) at least fornon-reversiblespin-flipprocesses underthe assumptions of rateswhich
are
bounded awayfromzero.
Theexampleswe haveare
for themoment too fewto developa
goodintuitionon
this point, butsee
[59].We add the remark that sometimes a dynamics description is possible in terms of
a
Gibbsmeasure on
the space-time histories (in $d+1$ dimensions). In such cases, looking at the steadystatesis consideringthe$d$-dimensional projection ofsuchGibbs
measures.
Recently,
a
study of non-Gibbsianness in a mean-field setting has started. In this case, thecharacterization of Gibbs
measures as
having continuityproperies in the product topology breaks down. For thesedevelopments,see
[50, 37, 52].Another set ofquestionswherenon-Gibsiannessprobably plays
an
importantrole,but wherewe
don’t know much,is in Quantum
Statistical
Mechanics. A Gibbsstate here should bea
stateon
a$C^{*}$-algebra satisfying the KMS condition for a $C^{*}$-automorphism. It
seems
that the finiteenergypoperty corresponds with the state having
a
cyclic and separating vector, implying theTomita-Takesaki structure. This means it has the
KMS
structure, but on the level of thevon
Neumannalgebrawhich isthe weak closureofthe C’-algebra [105]. We do not have
a
good criterionas
yetofhow to establish that
a
given quantumstate isnon-Gibbsianin this sense, butone
properly quantummechanicalnon-Gibbsianexampleof a nonequilibrium steadystate hasbeen analyzed in [81].
Non-Gibbsian
measures
thusoccur
in quitedifferentareas
ofstatistical mechanics,besides renor-malization $\mathrm{g}\mathrm{r}o$up, andcan
have quite different properties. Bynow
itseems
somewhat surprisingthat it took
so
long to appreciate the fact that the Gibbs property is rather special, in particularinview of Israel’s [43] resultthat in theset ofalltranslationinvariant (ergodic, nonnull) measures,
Gibbs
measures
are
exceptional.8
Appendix:
Some
quotes
$\mathrm{h}\mathrm{o}\mathrm{m}$the
literature
As an illustrationthat the heuristic character ofRenormalization-Group theory, and inparticular
the need toconsider the existence problemof(approximately) local renormalized interactions
was
recognized byvariouspractitionersI mentionthe following quotes:
“Onecannotwrite arenormalization cookbook” (K.G.Wilson, citedbyNiemeijerand
van
Leeuwen[17]$)$
.
“The notionofrenormalizationgroupis not well-defined” ([2], opening sentence).
“Itis dangerous toproceedwithout thinking about the physics” [29].
“...the locality [of therenormalizedinteractions] is
a
non-trivial problemwhichwillnot be discussed further” [113].“A RenormalizationGroup for
a
space ofHamiltoniansshould satisfy the following [25]:A) Existence inthethermodynamic limit,... B) $\ldots\ldots$
C) Spatial locality...”
Or, in thewordsof Lebowitz: “On thecautionary side
one
should remember that thereare
stillsome
serious open problems concerning the nature oftheRG
transformation of Hamiltonians for statisticalmechanical systems, i.e. for criticalphenomena. A lot of mathematical work remains tobedonetomake it into
a
well-defined theoryofphasestransitions” [55].As an illustrationthat
on
the other hand theoccurrence
of the above-mentionedproblems hasnot been generallyrecognized, leadingto
some
incorrect orat least misleadin$g$statements, I quote:“Therenormalization-group operator... transforms
an
arbitrarysystem in the [interaction] space...
into another systemin thespace....” [96].“[$\mathrm{t}\mathrm{h}\mathrm{e}$setofcouplingconstantsgivesrise$\mathrm{t}\mathrm{o}$]$..\mathrm{t}\mathrm{h}\mathrm{e}$ most generalform of the Hamiltonian...” [40]. “Further
iterations
of therenormalizationgroupwillgeneratelong-range and multi-spininteractions“...the space of Ising Hamiltonians in
zero
fieldmay
be specified by the set of all possiblespin-coupling constants, and is closed [Fisher’s emphasis] under the decimation
transformation”
[26].Acknowledgments
I thank prof.
Keiichi
Ito for
inviting me, and fora
very
nice meetingin
Kyoto. My workon
non-Gibbsian
issues
has been to a large extenta collaborative effort.
Ithank
especially RobertoFern\’andezand
Alan
Sokal, and also TonnyDorlas,Fkank den Hollander,Roman
Koteck\’y,Christof
K\"ulske,
J\’ozsefL\"orinczi,
Christian Maes, Frank Redig,Roberto
Schonmann, SenyaShlosman
andEvgeny Verbitskiy, for all they taught me during these collaborations. I have also very much
benefited from conversationsand correspondence with manyothercolleagues.
References
[1] D. J. Amit and V. Martin-Mayor. Field Theory, The Renormalization Group and Criticol
Phenomena. World Scientific,Singapore (2005).
[2]
G.
Benfatto andG. Gallavotti. Renomalization group,
Physics Notes, Vol.1.Princeton
Uni-versityPress, Princeton (1995).
[3]
G.
Benfatto, E. Marinari and E. Olivieri.Some
numerical resultson
theblock
spintransfor-mation
for
the 2d Isingmodel
at the criticdpoint. J.Stat.
Phys. 78,731-757
(1995).[4] L.Bertini, E.CirilloandE.Olivieri. Renormalization-group
transformation
under strongmixingconditions: Gibbsianness and
convergence
of
renormalized interactions.J.
Stat.
Phys.97,831-815
(1999).[5] L. Bertini, E.
CiriUo
and E.Olivieri. Gibbsian
properties andconvergenoe
of
the iteratesfor
the block averaging
transformation.
MarkovProc. Rel. Fields 10,381-392
(2004).[6] L. Bertini, E. Cirillo and E.
Olivieri. Graded
cluster $e\varphi ansion$for
lattice systems.Comm.
Math. Phys. 258,
405-443
(2005).[7] L. Bertini, E.
Cirillo
and E. Olivieri.Renormalization grvup
in the uniquenes region:Weak
Gibbsianity and
convergence.
Comm.Math.Phys. 261,323-378
(2006).[8]
C.
Borgs and J. T. Chayes. Thecovanance
matrivof
the Potts model:a
random clusteranalysis. J.
Stat.
Phys. 82,1235-1297
(1996).[9] A. Bovier and C. K\"ulske. A $7\dot{\mathrm{t}}goroustenom\iota alization$grvup method
for
interfaces
in randommedia. Rev. Math. Phys. 6, 413-496 (1994).
[10] J. Bricmont, A.KupiainenandR. Lef\’evere. Renomalization grvuppathologies and the
defini-tion
of
Gibbsstates. Comm. Math. Phys.194,359-388
(1998).[11] T. W. Burkhardt and J. M. J.
van
Leeuwen. Progress andpmblems in real-spacerenormaliza-tion. InReal-spaceReno7malization, T.W. BurkhardtandJ.M. J.
van
Leeuwen, Eds.(Springer,1982).
[12] E. N. M.
Cirillo
andE.Olivieri.
Renormalization-grvup at criticality and complete analyticity[13] D.Dereudre andS.Roelly. Propagation
of
Gibbsiannessfor infinite-dimensional
gradientBrow-nian
diffusions.
J.Stat.
Phys. to appear (2006).[14]
R.
L. Dobrushin.Lecture
givenat
theworkshop “Probabilityand
Physics”, Renkum, Holland(1995).
[15] R. L. Dobrushinand S. B. Shlosman. G\’ibbsian representation
of
non-Gibbsian
fields.
Russian Math. Surveys25,285-299
(1997).[16] R. L.Dobrushinand
S.
B.Shlosman
‘Non-Gibbsian” states and theirGibbs
description.Comm.
Math. Phys. 200,
125-179
(1999).[17]
C.
Domb and M.S.
Green
(Eds.). Phase transitions and criticalphenomena, Vol.6. AcademicPress, NewYork, 1976.
[18] T.
C.
Dorlas and A. C. D. van Enter. Non-Gibbsian limitfor
large-block majority spintrans-formations.
J.Stat.
Phys. 55,171-181
(1989).[19] M. H. Ernst and H. J. Bussemaker. Algebraic spatial
corre
lations in latticegas automata
violating detailed balance. J.
Stat.
Phys. 81,515-536
(1995).[20]
G.
L. Eyink, J.L.
Lebowitz and H. Spohn. Hydrodynamics andfluctuations
outsideof
localequilibrium: driven
diffisive
systems. J. Stat. Phys. 83,385-472
(1996).[21] R.Fern\’andez. Measures
for
lattice systems. PhysicaA263,117-130
(1999).[22] R. Fern\’andez. Contribution to Proceedings 2005 les Houches school
on
statisticalmechanics,to appear.
[23] R. Fern\’andez, A. Le Ny and F. Redig. Variational principle and almost locality
for
some
renormalized
measures.
J. Stat. Phys. 111,465-477
(2003).[24] R. Fern\’andez and C.-Ed. Pfister. Global specifications and non-quasilocality
of
projectionsof
Gibbs
measures.
Ann. Prob. 25,1284-1315
(1997).[25] M. E. Fisher. Scaling, universality and
renormalization group
theory. InCritical
phenomena(Stellenbosch 1982), Ed. F.
J.
W. Hahne, Springer Lecture Notes in Physics186
(1983).[26] M. E. Fisher. Renormdization group theory: Its basis and
formulation
in statisticalphysics.Rev.Mod.Phys. 70,
653-681
(1998).[27] K. Gawedzki, R. Koteck\’y and A. Kupiainen. Coarse graining appmach to
first
order phasetransitions. J.
Stat.
Phys. 47, 701-724 (1987).[28] H.-O. Georgii. GibbsMeasures and Phase $\mathcal{I}\dagger \mathrm{u}nsitions$
.
Walter de Gruyter (deGruyterStudiesin Mathematics, Vol. 9), Berlin-NewYork (1988).
[29] N.
Goldenfeld.
Lectureson
phase transitions and the renormalizationgrvup.
Addison-Wesley,Rontiersin Physics85 (1992).
[30]
R.
B.Griffiths
and P.A.
Pearce.
Position-space renormalization-grouptransformations:
Some
proofs
and
some
problems. Phys.Rev.
Lett. 41,917-920
(1978).[31] R. B.
Griffiths
and P.A. Pearce.
Mathematicalpropertiesof
position-spacerenormalization-gm
uptransfornations.
J. Stat. Phys. 20,499-545
(1979).[32] R. B.
Griffiths.
Mathematicalprvpertiesof
renormdization-group
transformations.
PhysicaA[33] G. Grimmett. The stochastic random-cluster process and the uniqueness
of
random-clustermeasures.
Ann. Prob., 23,1461-1510 (1995), and The Random-cluster Model. Chapter 3 In:Probability,
Statistics
and Optimisation, Ed.F. P. Kelley, Wileyand SonsLtd. (1994).[34]
0.
H\"aggstr\"om. Is the hzzy Potts modei $Gibbsian^{q}$ Ann. Inst. Henri Poincar\’e, Prob.Stat.
39,891-917
(2003).[35] O. H\"aggstr\"om. Coloring percolation clusters at random. Stoch. Proc. and Appl. 96,
231-242
(2001).[36] O. H\"aggstr\"om. Almost
sure
quasilocalityfails for
the random-cluster modelon a
tree.J.Stat.
Phys. 84,
1351-1361
(1996).[37] O. H\"aggstr\"om and C. K\"ulske. Gibbs properties of fuzzy Potts models
on
trees and inmean
field. Markov Proc. Rel. Fields 10,477-506 (2004).
[38] K. Haller and T. Kennedy. Absence
of
reno
rmalizationgroup
pathologiesnear
the criticaltemperature-two examples. J.
Stat.
Phys. 85,607-638
(1996).[39] A.
Hasenhatz
and P. Hasenhatz. Singular renormalizationgroup
transfomations
andfirst
orderphase transitions (I). Nucl. Phys. B $295[\mathrm{F}\mathrm{S}21],$ $1- 20$ (1988).
[40] K. Huang.
Statistical
Mechanics,18.3, 2ndedition,Wiley (1987).[41]
O.
Hud\’ak.On
the characterof
peculiarities in the position-spacereno
malization-grouptrans-formations.
Phys. Lett.A
73,273-274
(1979).[42] R. B. Israel. Banach algebras and
Kadanoff transformations.
InJ.
HYitz,J.
L. Lebowitz, and D. Sz\’asz, editors, Random Fields (Esztergom, 1979), Vol. II,pages 593-608.
North-Holland,Amsterdam (1981).
[43] R. B. Israel. Private communication (1992), and Somegenericresults in
mathematical
physicsMarkovProc. Rel. Fields 10,
517-521
(2004).[44]
W.
Kager and B.Nienhuis.
A guidetostochastic
L\"ownerequation and itsapplications. J.Stat.
Phys. 115,
1149-1229
(2004).[45] T. Kennedy.
Some
rigorvus resultson
majority rule renormalizationgroup
transformations
near
the criticalpoint. J. Stat. Phys. 72,15-37
(1993).[46] T. Kennedy. Majority rule at low temperatures
on
the square and triangular lattices. J.Stat.
Phys. 86,
1089-1108
(1997).[47] O.K. Kozlov. Gibbs description
of
asystem ojrandom variables. Prob. Inf. TMransmission 10,258-265
(1974).[48] C. K\"ulske. $(Non-)Gibbsianness$ andphase transitions in random lattice spin models. Markov
Proc. Rel. Fields 5,
357-383
(1999).[49]
C.
K\"ulske. Weakly Gibbsianrepresentationsfor
jointmeasures
of
quenchedlattice spinmodels. Prob. Th. Rel. Fields 119,1-30
(2001).[50] C. K\"ulske. Analogues
of
non-Gibbsianness
in jointmeasures
of
disordered
mean
field
models.J.
Stat.
Phys. 112,1101-1130
(2003).[51] C. K\"ulske, A. Le Ny and F. Redig. Relative entropy and variational properties
of
generalized[52]
C.
K\"ulske andA.
Le Ny. Spin flip dynamicsof
the Cume-Weiss model: Lossof
Gibbsiannesswithpossibly broken symmetry. Preprint, $\mathrm{m}\mathrm{p}$-arc
05-537
(2005).[53] R. K\"uhn. Equilibrium ensemble apprvach to disordered systems. Z. Phys. B,100 $\mathrm{B}$,
231-242
(1996).
[54] G. F. Lawler,$0$
.
SchrammandW. Werner. The dimensionof
theplanarBrownianfrvntier
is$\frac{4}{3}$
.
Math. Res. Lett. 8,401-411
(2001) (as wellas
variousother papers).[55] J. L. Lebowitz.
Statistical
Physics. In The PhysicalReview: Thefirst
100
years (1995). [56] J. L.LebowitzandC. Maes. Theeffect of
an
extemalfield
on
an interface, entropic repulsion.J.
Stat.
Phys. 46, 39-49 (1987).[57] J. L.
Lebowitz
and R.H. Schonmann.
Pseudo-free
energies and large deviationsfor
non-Gibbsian $FKG$
measures.
Prob. Th. and Rel. Fields 77,49-64
(1988).[58] R. Lef\’evere. Variational principle
for
some
renomalizedmeasures.
J. Stat. Phys. 96,109-133
(1999).[59] R. Lef\’evere and H. $\mathrm{T}\mathrm{a}s\mathrm{a}\mathrm{k}\mathrm{i}$
.
High-temperature expansionfor
nonequlibrium steady states indriven lattice gases. Phys.
Rev.
Lett. 94, 200601 (2005).[60] A. Le Ny andF. Redig.
Short
time conservationof
Gibbsianness
under local stochasticevolu-tions. J. Stat.Phys. 109,
1073-1090
(2002).[61] T. M. Liggett. Interacting particle systems. Springer, Berlin(1985).
[62] J. L\"orinczi. On limits
of
the Gibbsianformalism
in themodynamics. Universityof Groningenthesis (1995).
[63] J. L\"orinczi. Quasilocality
of
projectedGibbs measures
through analyticity techniques. Helv. Phys.Acta68,605-626
(1995).[64] J. L\"orinczi and C. Maes. Weakly Gibbsian
measures
for
lattice systems. J. Stat. Phys. 89,561-579
(1997).[65] J. L\"orinczi,
C.
Maes and K. Vande Velde. $I$}$\mathfrak{n}nsfomations$of
Gibbsmeasures.
PTRF 112,121-147
(1998).[66] J. L\"orincziand K. VandeVelde. A note
on
theprojectionof
Gibbsmeasures.
J. Stat.Phys. 77,881-887
(1994).[67] J. L\"orincziand M. Winnink. Some remarks on almost
Gibbs
states.
In N. Boccara, E. Goles,S.
Martinez,andP. Picco, editors, Cellular Automata and CooperativeSystems,pages
423-432, Kluwer, Dordrecht (1993).[68]
S.
K. Ma. Modem theoryof
criticalphenomena. Benjamin, Reading, Massachusetts(1976).[69]
C.
Maes, A.van
Moffaert
and F. Redig. AlmostGibbsian
versus
WeaklyGibbsian
measures.
Stoch.
Proc.Appl. 79,1-15
(1999).[70]
C.
Maes, A.van
Moffaert and F. Redig. The restrictionof
the Ising model toa
layer. J.Stat.
Phys. 94,
893-912
(1994).[71]
C.
Maes, A.van
Moffaert, F. Redigand S. B. Shlosman. Percolation, path large deviationsand[72]
C.
Maes andK. Netocny. Spacetime expansionsfor
weakly coupled interactingparticle systems.J.Phys.A 35,
3053-3077
(2002).[73] C. Maes andS. B. Shlosman. Foeeringtransition in the Isingmodel
without
internal contours. Prob. TheoryRel. Fields 115,479-503
(1999).[74] C. MaesandK. Vande Velde. Defining relative energies
for
theprojected Isingmeasure.
Helv.Phys. Acta65,
1055-1068
(1992).[75] C. Maesand K. Vande Velde. The $(non-)Gibbsian$
nature
of
states
invariant under stochastictransfomations.
PhysicaA 206,587-603
(1994).[76]
C.
Maes and K. Vande Velde. The juzzy Potts model. J. Phys. $\mathrm{A}$, Math.Gen
28,4261-4271
(1995).
[77]
C.
Maes and K. Vande Velde. Relative energiesfor
non-Gibbsian
states.Comm.
Math. Phys.189,
277-286
(1997).[78] F. Martinelli and E.
Olivieri.
Some remarkson
pathologiesof
renomalization-grouptransfor-mations. J.
Stat.
Phys. 72,1169-1177 (1993).[79] F.MartinelliandE.Olivieri. Instability
of
renomalization-grouppathologiesunder
decimation.J. Stat.
Phys. 79,25-42
(1995).[80] F.Martinelli and E. Scoppola. A simplestochastic clusterdynamics: $f\dot{r}go\tau vus$results. J. Phys.
$\mathrm{A}$, Math.
Gen.
24,3135-3157(1991).[81] T. Matsui and Y. Ogata. Vartational principle
for
non-equilibrium steady statesof
the $XX$modd. Rev. Math. Phys. 15,905-923 (2003),
[82] T. Morita.
Statistical
mechanicsof
quenched solid solutions with applicationsto
magnet\’icallydilute alloys. J. Math. Phys. 5,
1402-1405
(1964).[83] Markov Processes and Related Fields 10, specialvolume
on Gibbs-non-Gibbs
(2004).[84] Th. Niemeijer and J. M. J.
van
Leeuwen. Renomalizaton Theoryfor
Ising-like spin systems.In Phase $\pi ansitions$ and CriticalPhenomena, Vol.6,
pp.
425-505,C.
Domb and M.S.
Green,editors,Academic Press,NewYork(1976).
[85] M. J.deOliveiraand A.Petri. Boltzmanntemeprature in out-of-equilibriumlattice
gas.
Preprint $\mathrm{a}\mathrm{r}\mathrm{X}\mathrm{i}\mathrm{v}$cond-mat/0511263 (2005).[86] M. Ould-Lemrabott.
Effect of
the block-spin configurationon
the locationof
$\beta_{c}$ intwo-dimensionalIsingmodels. J. Stat. Phys. 86, 1109-1115 (1997).
[87] K. Pinn. Zeroes in the complex betaplane
of
$\mathit{2}d$ Ising blockspin Boltzmannians. J. Phys. A31,
9219-9228
(1998).[88]
C.-Ed. Pfister
and K. Vande Velde.Almost
sure
quasilocalityin
therandom
cluster model. J.
Stat.
Phys. 79,765-774
(1995).[89] J. Salas. Low temperature series
for
renomalized operators: the ferrvmagneticsquare
Isingmodel. J.
Stat.
Phys. 80,1309-1326
(1995).[90] R. H. Schonmann. Projections
of
Gibbsmeasures
may be non-Gibbsian.Comm.
Math. Phys.[91] O. Schramm. Scalinglimits
of
loop-erased random walks andunifom
spanning trees. Isr. J. of Math. 118, 221-288 (2000).[92] T. Seppalainen. Entropy
for
translation-invariant random-cluster models. Ann. Prob. 26,1139-1178
(1998).[93]
S. Smirnov.
Criticalpercolation in the plane:Confomal
invariance, Cardy’s fomula, scalinglimits. C. R. Acad. Sci, ParisS\’er. I Math. 333, 239-244 (2001).
[94] E.R. Speer. The twospecies totally asymmetric simple exclt’-sion process.In Proceedings
NATO
$ASI$Leuven Workshop “On three levels”, M.Fannes,C. Maes andA.Verbeure,editors,91-103, Plenum Press (1994).
[95] W.
G. Sullivan.
Potentialsfor
almostMarkovian randomfields.
Comm.
Math. Phys.33,61-74
(1973).
[96] C.Thompson. Classicalequilibriumstatisticalmechanics, 7,4and7.5. ClarendonPress,Oxford
(1988).
[97] A.
C.
D.van
Enter.Ill-defined
block-spintransfomations
at arbitrarily high$t$emperatures. J.Stat.
Phys. 83,761-765
(1996).[98] A.
C.
D.van Enter. On thepossiblefailure of
the Gibbsprope$\mathrm{r}ty$for
measures on
latticesystems.MarkovProcessesand RelatedFields,2,209-224 (1996).
[99] A.
C.D. van
Enter, R. Fern\’andez, F. den Hollander and F. Redig. Possible loss and recoveryof
Gibbsianness during the stochastic evolutionof
Gibbs meosures. Comm. Math. Phys., 226,101-130 (2002).
[100]
A.
C.D.
van
Enter andR.
Fern\’andez.Problems with
thedefinition of
momentum-spacerenor-malization
transfomations.
Phys. Rev. $\mathrm{E}$ S9,5165-5171
(1999).[101] A.
C.
D.van
Enter, R. Fern\’andez, and R. Koteck\’y. Pathological behaviorof
renormdizationgroup maps at high
fields
and above the transition temperature. J. Stat. Phys. 79,969-992
(1995).[102] A.
C.
D. vanEnter, R. Fern\’andez, and A.D. Sokal. Renormdizationtransfomations
in thevicinity
of
first-order
phase transitions: What can and cannot go wrong. Phys. Rev. Lett. 66,3253-3256 (1991).
[103] A. C. D.
van
Enter, R. Fern\’andez, and A. D. Sokal. Regularity properties and pathologiesof
positio$n$-space renomalization-grvuptransfomations:
Scope and limitationsof
Gibbsiantheory. J.
Stat.
Phys. 72,879-1167
(1993).[104] A.
C.
D.van
Enter,R.
Fern\’andez, andA.
D.Sokal. Renomalization
transfomations:
Source
of
examples and prvblemsinprobability andstatistics.
Resenhas doInstituto
deMatematica
$\mathrm{e}$ EstatisticadaUniversidadedeSao
Paulo 1,233-262
(1994).[105] A. C. D.
van
Enter, R. FernAndez, and A. D. Sokal. Non-Gibbsian statesfor
Renomdiza-tion
group
transfomations
and beyond. In CellularAutomata
and Cooperative Systems,Eds.N.Boccara,E.Goles,
S.
MartinezandP.Picco,pp.141-152.
Proceedings1992
Nato AdvancedStudiesInstituteworkshop, les Houches. Kluwer,Dordrecht (1993).
[106] A.C. D.
van
Enter andC. K\"ulske. Two connections betweennon-Gibbsianness and disordered[107] A.
C.
D.van
Enter,C.
K\"ulskeandC.
Maes.Comment on
“Critical behavior
of
the
randomlyspin diluted$\mathit{2}D$ Ising model: A grand ensemble approach.” by R. K\"uhn. Phys.
Rev.
Lett. 84,6134
(2000).[108] A. C. D.
van
EnterandJ. L\"orinczi. Robustnessof
the non-Gibbsianproperty:some
examples.J. Phys. $\mathrm{A}$, Math. and Gen. 29,
2465-2473
(1996).[109] A. C. D.
van
Enter C.Maes, R. H. Schonmann andS.
B. Shlosman. TheGriffiths
singularityrandom
field.
On Dobrushin’sway:
Htom probability theory to statistical physics, pp. 51-58.Eds.R. A.Minlos, S. B. Sholsman andY. M. Suhov. AMShanslations, Providence RI(2000).
[110] A.
C.
D.van
Enter andS.
B.Shlosman.
(Almost) Gibbsian descriptionof
the signfield
of
$SOS$
fields.
J.
Stat. Phys. 92,353-368
(1998).[111]
A.
C.
D.van
Enterand
E.A.
Verbitskiy.On
the variational principlefor
generalized Gibbsmeasures.
MarkovProc.Rel.Fields, 10,411-434
(2004).[112] K. Vande Velde.
On
the $q\mathrm{u}$estionof
quasilocality in large systemsof
locally interactingcom-ponents. K.U. Leuven thesis (1995).
[113] K.