HAL Id: jpa-00247235
https://hal.archives-ouvertes.fr/jpa-00247235
Submitted on 1 Jan 1996
HAL is a multi-disciplinary open access archive for the deposit and dissemination of sci- entific research documents, whether they are pub- lished or not. The documents may come from teaching and research institutions in France or abroad, or from public or private research centers.
L’archive ouverte pluridisciplinaire HAL, est destinée au dépôt et à la diffusion de documents scientifiques de niveau recherche, publiés ou non, émanant des établissements d’enseignement et de recherche français ou étrangers, des laboratoires publics ou privés.
On the Large Order Behaviour of the Potts Model
Domenico Maria Carlucci
To cite this version:
Domenico Maria Carlucci. On the Large Order Behaviour of the Potts Model. Journal de Physique
I, EDP Sciences, 1996, 6 (8), pp.983-992. �10.1051/jp1:1996111�. �jpa-00247235�
J.
Phys.
I £Fonce 6(1996)
983-992 AUGUST1996, PAGE 983On the Large Order Behaviour of the Potts Model
Domenico Maria Carlucci
(*)
Scuola Normale
Superiore,
Piazza dei Cavalieri 7, 56100Pisa, Italy
(Received
8 March 1996, revised 22April
1996,accepted
2May1996)
PACS.05.20.-y
Statistical mechanics PACS.11.10.z Fieldtheory
PACS.11.10.Jj Asymptotic problems
andproperties
Abstract.
Following
thiworkby Houghton,
Reeve and Wallace about an alternative for-mulation of the n - 0 limit of the
In +1)
state Potts model in fieldtheory
for thelarge
order behaviour of theperturbative
expansion, wegeneralise
theirtechnique
to alln
by establishing
anequivalence
inperturbation theory
orderby
order with another bosonic fieldtheory. Restricting
ourselves to a cubic
interaction,
we obtain anexplicit
expression(in
terms ofn)
for thelarge
order behaviour of thepartition
function.R4sumk. Suivant le travail de
Houghton,
Reeve et Wallace concemant une formulationalternative de la limite n
- 0 du modAle de Potts en thdorie des
champs
pour le comportementaux
grands ordres,
nousg6n6raIisons
leurtechnique
h nquelconque
en 6tablissant une6quivaIence
ordre par ordre en thdorie de
perturbation
avec une autre thdorie dechamps
de bosons. En nousrestreignant
h une interactioncubique,
nous obtenons uneexpression explicite (en
fonction den)
pour le comportement auxgrands
ordres de la fonction departition.
Introduction
The Potts model [1] is one of the most
fascinating topics
in statistical mechanics. It can beregarded
as thegeneralisation
of theIsing
model [2] to ageneric
number q ofcomponents.
Originally
thisproblem
wasproposed
to Pottsby
his PhD Professor Domb[3],
whosuggested
him to
investigate
the criticalproperties
of asystem
ofinteracting spins
confined in aplane, pointing
in qequally spaced
directionsHi
= ~" i=
o, I,..
,q I
ii
q
with an Hamiltonian
depending only
on the relative orientation of twoneighbouring spins
of the form7i =
~j J(81, 8j) J(e~, 8j)
c~
cos(81 8j) (2)
lsPinsl
(* e-mail: carluccitlux2sns.sns.it
Q
Les(ditions
dePhysique1996
Actually,
theprevious
model bears the name ofplanar
Potts model.Nevertheless,
the most famous model is the one where twospins
interactonly
whenthey
arepresent
in the samestate,
that isJ(81,8j)
ccb([l~,. (3)
The interaction
(3)
can be re-written in such a way as to reflect afull-symmetry
in a(q11)-
dimensional space
by making
use of q vectors(q-1)-dimensional
e°pointing
in the qsymmetric
directions of a
hypertetraedron
in(q -1)
dimensions.Indeed,
the interaction(3)
can be re- written asb(81,8j)
=
[1
+(q -1)e~~ e~3). (4)
q
The
increasing
attention to the Potts model was due to the fact that it has proven to be related to alarge
class ofoutstanding problems, especially
whenparticular
values of q, even noninteger,
are taken. For
example,
thepercolation problem
can be formulated in terms if q= 1 Potts model
[4-6].
Also Fortuin andKasteleyn
[7] showed that the Kirchhoff's solution [8] for an ensemble of resistors isstrictly
related to q = o limit of the Potts model.Moreover,
the criticalbehaviour of the dilute
spin glasses
is well understood in terms of the q =)
limit[9,10]. Finally, Lubensky
and Isaacon[11]
showed thatinteresting
processes ingelation
and vulcanization fall in the same class ofuniversality
of o < q < 1 Potts model(For
acomplete
review of the Pottsmodel see Wu [12]
).
A convenient formulation of Potts model as a field
theory
was firstperformed by
Zia and Wallace[13],
whoinvestigated
the criticalproperties
in the framework of the renormalization group.Actually
in the fieldtheory
literature it iscustomary
to refer toin +1)
Potts modelin
+1 = q in theprevious notation).
Thesymmetry
of the model allows a trilinear interaction which is theleading
term in the renormalization group framework. Because of a pure cubicinteraction,
thehigh
order behaviour of theperturbative expansion
isclearly non-oscillating
and therefore non Borel-sommable.
Nevertheless,
whenparticular
limits are taken(e.g.
n - opercolation problem)
thelarge
order behaviour is found to beoscillating
and Borel-sommable [14] as obtainedby Houghton,
Reeve and Wallace(from
now on referred asHRW).
In Section1,
we recall the Hamiltonian of the
in +1)
Potts model and the alternativeapproach
to n = o limit usedby Houghton,
Reeve and Wallace. In Section 2 itsgeneralisation
to each value of n is showed. Section 3 will be devoted to the evaluation of thelarge
order behaviour for such analternative field
theory
and the results will be shown to agree with theprevious
ones.1. The
In +1)-State
Potts ModelWe consider the
in +1)
state Potts model in ddimension,
describedby
an Hamiltonian[13]
of the form
7i
[#]
=/ d~x
(T7#~ (T7#~ +
ii11
+ , p~jk 4~4j 4k (5)
2
2 3.
where sum over
repeated
indices isimplied
and runs from to n. Thesystem
possesses anhypertetrahedric symmetry, provided
that p~k is of the formn+i
p~k "
~ e)e)e( (6)
o=1
where the e's
satisfy
n+I n+I n+I
~j
ES= o
~j e$ej
=
In
+1)b~ ~j e$ef
=
In
+1)b°~
1ii)
a=I a=1 ~=l
N°8 ON THE LARGE ORDER BEHAVIOUR OF THE POTTS MODEL 985
Physical quantities
can be extracted from thepartition
function(the
discussion about theright
contour in the
path integral
for a pure cubictheory
can be found in[15] expanded
in powers of g:~n(9)
"
/ l~~e
~"
i~ ~K(n)9~~ (8)
~,_~
We are interested in the
asymptotic
behaviour ofZA-(n)
as K goes toinfinity.
We can extractZK In)
from(8) by
the usualdispersion
relation[16,17]
ZK(n)
=
/ Vi ) ~ dg~ )e~~l"1 (9)
m g
where the closed
integral
isperformed
in the cutcomplex plane g~.
In thelarge
Klimit,
theprevious integral
can be evaluatedby
thesteepest
descent method in the spaceIii, g)
and thesaddle
point equations
areT7~#1 =
e~k4j4k (10)
2
Equations (lo)
and(11)
can bedecoupled easily by setting
with the
~ti's satisfying
lli " fijklljllk.
(13)
Equation (13)
has n solutions ~t(~~, with r=
1, 2,..
,n,r
(r) jr) a
j~~)
~ " a §
~
~
i=1
where
a(r)
=
j15)
(n
+I)(n
+ 12r) Working
in d= 6
dimension,
the saddlepoint equations (lo, it)
then readV~4clx)
=
411x) l16)
~
~j~ti~ti / d~x#)(x)
=
-Kg~ Iii)
3
whose
general
solution is~~~~~
[12(x
~~~~
+ 1]~
~~~~
where I and To are constant
parameters.
From
equations (17, 18),
we obtain~~
~~~ /K I$~ ~~ ~~~~~~~~l~~~~
~~~ ~~~~~~~~~
~ ~, 5
~
~ ~ '~
j20)
3 28~3
minjrj (£~
ll)~~ll)~~For the
in
+I)
state Potts modelIn
>integer)
the minimum in(20)
is for r=
I, giving,
as
expected
for a pure cubictheory,
amalign
behaviour for the coefficientsZK
which do not oscillate[15, 16]
~~~~~
'~ ~~
~~~~.~~~~n ~~~~
~~~~
For n
= o
(corresponding
topercolation problem)
the saddlepoint
cannot be obtainedby
anaive
analytic
continuation of(20)
becauseevidently
r= I is not the
dominating
saddlepoint.
HRW [14] found an alternative formulation for the
percolation problem
andthey pointed
out that in the n - o limit the
Feynman
rules are the same as those of the Hamiltonian with two scalar fields~i14,#,9)
"/ d~x
I)jvi)~ jjv~l)~
+jji~ ~l~)
+j
g
ii
+l)~j j22)
provided
that ~ onekeeps only j-connected diagrams
as those shown inFigure
1.By
means ofpath integrals techniques, they
found the coefficients of theexpansion
of thetwo-point
correlation function
Gix
Y) -(~ GK 9~~ GK
+~ K<Al
~123)
which
fortunately
show anoscillating
behaviour.2. HRW Method for
Arbitrary
nAs mentioned in the
introduction,
it would be useful togeneralise
theprevious
treatment to each value of n. Our first task is to write down theequivalent
Hamiltonian inanalogy
with(22).
In order togain
someinsight,
let us compute thetwo-point
correlation function up to fourth order(~extPijk~i~j~kPlmn~l~m~nPabc~a~b~cPdef~d~e~f~ext) (24) by performing
all thepossible
Wick contractions. Because of theproperties (7),
we canget
terms as
L
n+I~l lli~
+i)b~~ ii lli~
+I)b~~ ii
lli~ +I)b~~ ii
a,#n,6=1
x
in
+I)b~~ lliin
+I)b~~ II f[ 125)
Inspired by
theprevious discussion,
wetry
tointerpret (n +1)b°~
and(-1)
aspropagators
of two different(possibly multi-component)
scalarfields, #
and# respectively.
The severalN°8 ON THE LARGE ORDER BEHAVIOUR OF THE POTTS MODEL 987
,- ,,
, ,
#' "
# I
la) 16)
,'h
,' ',
~'
iI '',
',
ic)
Fig.
1. Somediagrams
builtby
means of ifi(solid line)
and ifi(dotted line)
propagators.Only
thediagrams la)
and 16) are suitable because # connected.terms we obtain
by expanding
theproducts
in(25) correspond
in the secondtheory
to all the differentdiagrams
with the sametopology
as theoriginal
ones. As in the n - o caseonly diagrams
with external# legs
connectedby
aj's path
do contribute.In
Figure
1 we have drawn three of thepossible graph
builtby #
and ifipropagators. By requiring
that thediagrams
in the two theories have the same numericalcoefficients,
we obtain theFeynman
rules.Specifically,
from the terminvolving only #
fields(Fig. 1a)
weget
n+1 n+1
£ El in
+1)5b°~b°~b~~b~~b~~ e)
=in
+1)5 £ e)e) (26)
a,#,~,6"1 a=1
In these case, the presence of b's eliminate the sum over the four indices. If we associate a factor
in +1)
to each# propagator,
we obtain the same results withoutintroducing
anyweight
for the vertices.Let us now consider the second
graph (see Fig. lb)
n+I n+I
L ~l
li~ +i)~b~~b~~b~~b~~l~i) ~)
"
l~i)li~
+1)~ L ~l~) 127)
a,#,~,6 a=1
as we
said,
we associated a factor(-1)
for the ifipropagator
and we do not need any factor for the vertex#2#.
The situation is
quite
different when we deal withgraphs
where we have a#3
vertex as inFigure
1c which derives fromn+I n+I
n+I
l~i)l~i)l~i) ~ ~l
li~ +i)~b~~b~~~)
"l~i)
li~+1)~ L L l~)~
128)
a,#,~,6=1 ~=l a=1
We note that we are left with a sum over ~ which
gives
a factorin +1).
Soa1fi3
vertexrequires
a factor ofin
+1).
At
higher
order we also have non-zerodiagrams
with also#
connectedsub-graphs
as shown inFigure
2.,,--,,~
,' '~ l' ',
' , ' ,
# ' I
Fig.
2.High
orderdiagrams
witha # connected
subgraph.
The filledregion
means any non-zerodiagram.
It can be rather
simply
proven that eachj-connected subgraph gives
a factor ofin +1).
Summarising
theFeynman
rules forin +1) component
Potts model are1) Only j-connected graphs give
a non zero contribution.2)
Each# propagator gives
a factor ofin
+1).
3)
Each ifipropagator gives
a factor of(-1).
4)
Eachifi~vertex gives
a factor ofin
+1).
5)
Eachj-connected subgraph gives
a factor ofin
+1).
These
Feynman
rulescorrespond
to thefollowing
Hamiltoniani~ m
/ d6z lit jjiv<~)2
+iii)
~i
~)lV~)~
+-~fi~)
+) ~~ ~ (la
+~fi)~l129)
By following exactly
the sameprocedure
as HRW'S in order to extract thej-connected diagrams,
weget
for thetwo-point
correlation function/ Vi #(x)#(y) exp[-7i*(#)] e~~"
~~~
~~~~~~/ ~~ ~~~~
~~~~~~~~~~~~~~~
~~~~~~~~~~~~~~~
~~ )~~ ~~~~~ ~~~~~j
~~~~
where
7i*(#)
is the reduced Hamiltonian7~*ii,
~fi)=
/dx6 )j (iv«2 +12) jj (iv<)2
+ ~fi2) +
j ii
+~fi)~
131)
and
W*(ifi)
is definedby
exp
iw*tiff)i
=/ vi
exp -7i*
ii,
ifi)132)
N°8 ON THE LARGE ORDER BEHAVIOUR OF THE POTTS MODEL 989
3.
High
Order Behaviour~s in the n - o case, it can be shown that
only
the second term on the left hand side of(30)
contributes to the
high
order behaviour of theperturbative expansion
because of theparticular
form of the cubic interaction and we have to evaluate/ p~pj(i j~ j(2) j(i) j~) ~(2)
~)
l)
~~2
121ri
g2K
for large K.
Our
goal~V~#1
")9
(ifi +Ii
)~ i"
1,
2(34)
~
T7~4i + g(ifi +
Ii
)~ + g(ifi +d2)~
+Ii n)
~~~ = o(35)
il + 1 2 2
3~l
j / d~x j~
+
~i
)~ +lift
+42)~
+II
~l)~)
+ ~~" °
136)
It is
simple
to evaluate thepartial
derivative of W* withrespect
to g and# by steepest
descent method too and we obtainand
~~~
=
/d~x(ifi
+o)~) Ii
+O(g~))
fig
3'provided
thatlo obeys
to thefollowing equation
fi~~'°
=
j9140
+~)2
After the
re-scaling
#~ - 4l~ = #~ i
=
o, 1,
2 91 ifi ~
~Y"~ifi
the saddlepoint equation
arefinally
Ah = (Ah +
ill)~
i=
0,1,
2(37)
n + 1 2
T7~fll + (Ah +
ill)~
+(4l2
+ill)~
~~ ~~(4lo
+ill)~
= o(38)
2Kg2
=
j /d6x
j14~i
+4/)~ +14~~
+4/)~ ii -n)14~o
+4/)~j 139)
In order to solve the
equations (37, 38, 39),
let us make the ansatz4~i =
~liicix)
~P =uicix) 140)
where
#c(x)
is the solution of(16). Therefore,
we have reduced our differentialequations
toalgebraic
ones.(n 1)
~~ ~~ ~ ~~~~~ ~ ~ ~
(1 n) In
+1)
~In
+1)
~In
+1)
~°3, ~6y~3
Kg~
=
[(u
+ ~t1)~ +Iv
+ ~t2)~ +In I)(u
+~to)~j
5
Solving
the firstequation gives
two solution~~~~
l~ in l)
~in l)
~~~~~~
~ ~~~~~~ ~~~~Only
four choices arepossible 1)
~~ =~l+)
~~ = ~~ =~l-) 2)
~ti = ~t~~) ~t2 " ~to "~t(+)
4)
~to "~t(~)
~ti = ~t2 "~t(+)
From the
1)
and2)
we obtain twoequivalent
solutionsprovided
that n#
1~
in
+~~n
1)2
~~
~~~~The choices
3)
and4) gives
other twoequivalent
solutions for u when n#
3~ "
j~
/(
j~
~~~ n~
3j4~~
The
previous
two values for ugive
two saddlepoints
forg2
~ 1 3 231r~ n
~
~ j44)
~(~) K 5
in
+1)2 In 1)~
and
~ ~
9~2) "
(
~ ~j~ /)) j/~
~)2
~~
~(~~)
A this
point,
it isinteresting
to note that the saddlepoints
g(~~ and
g(~~ are
respectively
ther = 1 and the r
= 2
points
of(18). Anyway,
now, our final aim is to evaluate thelarge
order behaviour of theperturbative expansion,
that is the coefficientsZK
in theexpression (8).
SoN°8 ON THE LARGE ORDER BEHAVIOUR OF THE POTTS MODEL 991
a little care needs to understand which saddle
point
dominates at nfixed.
We can obtain the twofollowing
behavioursand
~
~~~
'~3
91r3
~~ ~~~ ~)
~~~~
~
~ ~~~~For n
= o
(percolation)
the second saddlepoint
dominate over the first one and we recover the HRW results.Actually
there exist other formulations for thepercolation problem disagreeing
with HRW
[18],
but it can be shown[19]
that the different result was due to anunappropriate
choice of anintegration
contour.Nevertheless,
the nature of thepercolation
behaviour stillremains controversial because there is another result
[20]
where the r = saddlepoint
seems to dominate. For n=
(Ising model)
we cannot choose the coefficientZ(~
andZ(~
isidentically
zero
according
to the fact that the trilinear interaction vanishes as n= 1. For n
integer
we have to take into account the(46)
and this is inagreement
with the calculationsperformed
inSection 1.
Conclusions
We
performed
thecomputation
of thelarge
order behaviour of theperturbative expansion
of theIn +1)
state Potts model with a cubic interaction for ageneric
value of n,by mapping
the
original problem
in asimpler
one. We showed that ourequivalent
fieldtheory
has two saddlepoints depending
on n and one has to choose the smallest one in order to extract the dominant term atlarge
order. A naturalquestion
can ariselooking
at the different number of the saddlepoints
in both the formulations. Indeed in Section 1 it has been shown that the Pottsmodel,
when studiedby
the usualtechniques,
has n saddlepoints
characterizedby
aninteger
r, whereas our alternative formulation
gives only
twoinequivalent
saddlepoints. Indeed,
thelarge
order behaviour is related to the cutdiscontinuity
of the correlation function for small value of thecoupling
constant, that is theimaginary part
of the Green functionsgenerated by
theTaylor expansions
around the saddlepoint.
For n=
2,3,..
theimaginary part
of the correlation functions is dominatedby
the r= 1 saddle
point.
When ananalytic
continuation in n isperformed
and someparticular
limit aretaken,
the r= 2 saddle
point
becomes theleading
term. All the other saddlepoints
aresub-leading
andthey
do not affect thelarge
order behaviour. Thisgeneralisation
canhelp
tocompute
criticalexponents
fortheory
where n is notinteger
as those mentioned in the introductioneventually by adding higher
order interactions.Acknowledgments
It is a
great pleasure
to thankGiorgio
Parisi forhaving suggested
andsupervised
this work and G.F. Bonini for scientific discussions.References
[1] Potts
R.B.,
Proc. Camb. Phil. Soc. 48(1952)
106.[2]
Ising E.,
Z.Phys.
21(1925)
613.[3] Domb
C.,
J.Phys.
A 7(1974)
1335.[4]
Kasteleyn
P.W. and FortuinC.M.,
J.Phys.
Soc.Jpn (S~tppi. )
26(1969)
11.[5] Giri
M.R., Stephen
M.J. and GrestG.S., Phys.
Rev. B16(1977)
4971.[6] Kunz H. and Wu
F.Y.,
J.Phys.
C11(1978)
L1.[7] Fortuin M.C. and
Kasteleyn P.W., Physica
57(1972)
536.[8]
KirchhoffiG.,
Ann.Phys.
Chem. 72(1847)
497.[9]
Aharony A.,
J.Phys.
C11(1978)
L457.[lo] Aharony
A. andPfeuty P.,
J.Phys.
C12(1979)
L125.[11] Lubensky
T.C. and IsaaconJ., Phys.
Rev. Lett. 41(1978)
829.[12]
WuF.Y.,
Rev. Mod.Phys.
54(1982)
235.[13]
Zia R.K. and WallaceD.,
J.Phys.
A 8(1975)
1495.[14] Houghton A.,
Reeve J.S. and WallaceD-J-, Phys.
Rev. B17(1978)
2956.[15]
McKaneA.J.,
Nuci.Phys.
B152(1979)
166.[16]
KleinertH.,
PathIntegrals
inQuantum Mechanics,
Statistics andPolymer Physics (World Scientific, 1990).
[17]
J.C. Le Guillou and J. Zinn JustinEds., Large
Order Behaviour of PerturbationTheory (North-Holland, 1990).
[18] McKane
A.J.,
J.Phys.
A 19(1986)453.
[19]
Caracciolo S. and CarlucciD.M.,
to bepublished.
[20] BrAzin E. and McKane