HAL Id: jpa-00246828
https://hal.archives-ouvertes.fr/jpa-00246828
Submitted on 1 Jan 1993
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.
Boundary effects in a two-dimensional Abelian sandpile
J. Brankov, E. Ivashkevich, V. Priezzhev
To cite this version:
J. Brankov, E. Ivashkevich, V. Priezzhev. Boundary effects in a two-dimensional Abelian sandpile.
Journal de Physique I, EDP Sciences, 1993, 3 (8), pp.1729-1740. �10.1051/jp1:1993212�. �jpa-00246828�
J.
Phys.
I France 3 (1993) 1729-1740 AUGUST 1993, PAGE 1729Classification Physic-s Abstracts
05.40 05.60 46. lo 64.60
Boundary effects in
atwo-dimensional Abelian sandpile
J. G. Brankov
(*),
E. V. Ivashkevich and V. B. PriezzhevLaboratory
of TheoreticalPhysics,
Joint Institute for Nuclear Research, Dubna, 141980, Russia(Receii'ed 22 Ma;ch J993, accepted 25
April
J993)Abstract. We study boundary and finite-size effects in the Abelian
sandpile
model due to Bak, Tang and Wiesenfeld. In the case ofhalf-plane
geometry, theprobability
iPj (r) of a unit height at theboundary,
and at a distance r inside the sample is found for open and closed boundaryconditions. The
leading asymptotic
form of the correlation functions for the unitheights, tPj,
(r), in thestrip
andhalf-plane geometries
is obtained for different boundary conditions too.Our results confirm the
hypothesis
that the unitheight
behaves like the local energy operator in the zero-component limit of the Potts model.1. Introduction.
Since the
publication
of the worksby
Bak,Tang
and Wiesenfeld I], critical Abeliansandpile
models are in the focus ofcomprehensive studies,
see, e-g-[2, 3,
4,5].
The stochasticevolution of a
sandpile naturally
leads to a state ofself-organized criticality (SOC),
which is characterizedby
correlations withpower-law decay
in space and time. As thespatial aspects
of the SOC are similar to those of a critical state in statisticalmechanics,
the program forstudying
the structure of SOC states
essentially parallels
the one for the usual statistical models like theIsing
or the Potts models. Besides the evaluation of bulk critical exponents, animportant
item in this program is the deterrnination of different surface exponentsdescribing
the behaviour of correlations near aboundary.
In the two-dimensional case, finite-sizeanalysis [6]
and conformalfield-theory [7]
establishrelationships
between surface and bulkproperties
of the models, which lead to acomplete description
of thespatial
structure atcriticality. Thus, finding
acorrespondence
between the observables of asandpile
and anappropriate
model instatistical mechanics at different
boundary
conditions seems an attractive task.The number of distinct recurrent
configurations
of the Abeliansandpile
model on anarbitrary
lattice has beenexpressed through
the number ofspanning
trees on the same lattice[8, 9].
The statistics ofspanning
treesobeys
the Kirchhoff theorem and canbe,
in turn, related(*) On leave of absence from Institute of Mechanics and Biomechanics, Bulgarian
Academy
of Sciences, lll3Sofia, Bulgaria.to the q
=
0 limit of the q-component Potts model
[10, 11].
Indrawing
a furtheranalogy,
however, one encounters thedifficulty
ofidentifying
localsandpile
observables as site variables in the Potts model. The natural formulation of asandpile
model isgiven
in terms ofinteger height
variables z, at each lattice site I efif andtoppling
rules. The stableconfigurations
in the SOC state of the Abelian model are sets ofheights (z,,
I efit),
wherez~ e
(1, 2, 3, 4). Majumdar
and Dhar[12]
have calculated theprobability iJ'j
offinding
the value z, =I and the correlation function
Pi
j
(r)
for two sites of unitheight
at a distance r apart.In the two dimensional case, the result
iJ°j (r)
r~~',
with.<= 2, has been obtained for
large
r, which
provided grounds
forsuggesting [9]
thatiJ°jj(r)
is the counterpart of the energy- energy correlator in the zero-component Potts model. For the Pottsmodel,
besides the bulk exponent x = 2, the surface correlationsdecay
exponent is known too, xii = 2, in the case of freeboundary
conditions[13, 14].
The value of the surface exponent xii, in tum, is related to theamplitude
A of the inverse correlationlength
f~ '= AIL, which controls the
exponential decay
ofpair
correlationsalong
aninfinitely long strip
of width L.Namely,
theamplitude-
exponent relations
[15]
~
~~~j
,
~~~d~c
b-c.
follow from finite-size
scaling [6]
and the conformalproperties
of the model[14].
In this paper, we consider the behaviour of the
probability iJ°j
and the correlation function£Pjj (r)
near theboundary
of the Abelian two-dimensionalsandpile
model[8].
Thetoppling
rules are
specified by
the matrix A with elements A,~ for sites I,j
in the bulk of the latticegiven by
~4,
I =j
A,~ =
I, ii j
=
(2)
0
,
otherwise
Here
ii j
denotes the distance between sites andj.
To formulate the
toppling
rules at theboundary
3fif of thesandpile,
we use three standardboundary
valueproblems
for theLaplacian
on a finite lattice fit.I.
Open boundary
conditions(or
Dirichletboundary conditions),
whenA,,
=4 for I e3fif, and, therefore,
the sandparticles
are allowed to leave the systemthrough
theboundary.
2. Closed
boundary
conditions (or Neumannboundary conditions),
whenj,
= 3 forI e
3fif,
and the sandpanicles
cannot leave the systemthrough
3fif.3. Periodic
boundary
conditions, when A,~ =4 as in case I, but now the lattice is
wrapped
on a torus and
ii j
is the distancealong
the surface of the torus.Obviously,
if all the boundaries are closed orperiodic
nosteady
state of thesandpile
ispossible. Therefore,
inconsidering
thestrip
geometry, we start with arectangular shape
of thelattice
fit, impose
openboundary
conditions on the verticaledges
of therectangle
andboundary
conditions of one of the types 1, 2 or 3 on theopposite
horizontaledges.
Then, we let the horizontal size of therectangle
tend toinfinity.
The
following
results arereported
here.1. I IN THE CASE OF HALF-PLANE GEOMETRY. The
probability iJ'j (0
) of aheight
z= I at the
boundary
is obtained : for openboundary
conditionsiJ'j (0)
=
~ ~~
+
~~~ ~~~
= 0.10382
,
(3)
2 w 3
w 9 w
N° 8 BOUNDARY EFFECTS IN A TWO~DIMENSIONAL ABELIAN SANDPILE 1731
for closed
boundary
conditionsiJ'j(0)
=
~ ~
= 0.11338
(4)
On
moving
inside thesample,
theprobability iJ'j (I
at a distanceI
from theboundary
tends to its bulk valueiJ'j(~x~ )
=2/7r~ 4/7r~ according
to the lawtPj(ii=tPj(~x~i(i±j+. ), (51
where the upper
sign
is related to the openboundary
conditions and the bottomsign
to the closed ones.The
two-point
correlation functioniJ°jj(r)
at the surfacedecays according
to the law for openboundary
conditionsiJ°jj(r)
=
4
iJ°)(0) ~~
" ~~ ~~~ ~
+,
,
(6) (27
7r -108 7r+64)
rfor closed
boundary
conditions£Pjj (r)
=
£P)(o) ~/
~ +
(7)
7r r
1.2 IN THE CASE OF AN INFINITELY LONG STRIP OF WIDTH L. The
leading exponential decay
of the
pair
correlationsiJ°jj(r) along
the central line of thestrip
is obtained.a)
For both open and closedboundary
conditionsiP, (r
; L~4
=
iJ°)(L/2)
e~ ~ "'~~ +,
(8)
L~where the finite-size corrections to the
probability iPj(L/2)
in the middle of thestrip
are :2
iPj(L/2)
=
iJ°j(~x~)(1±)+. ). (9)
4L
Here the upper
sign
is related to the openboundary
conditions and bottomsign
to the closedones.
b) For
periodic- boundary
conditions thepair
correlations take the formiJ°jj (r
;L)
=
iP)(L/2)
~ "e~~
"'~~+
,
(lot
L~~
where
~
~s iJ°,(L/2)
=
£Pj(cc)~l
~+
(ll)
15(7r 2)L
The above results can be
compared
with thepredictions
of the conformalfield-theory.
Thus, from(6), (7)
it follows that xii= 2, and from
(8)
one has A=
2 7r, which is in agreement with
Ill-
Forperiodic boundary
conditionsequation (10) yields
A=
4 7r, and the bulk exponent
.< =
2 has been obtained
by Majumdar
and Dhar[12],
which isagain
inconformity
with theprediction (1).
Thus our results confirm the
hypothesis
about thecorrespondence
between the unitheight
in the Abeliansandpile
model and the energy in the zero-component Potts model. As far as otherobservables are
concerned,
e-g-heights 2, 3, 4,
sizes of avalanches, their duration andperimeters,
we lack at presentconvincing
evidences of their relation to observables in the Potts model. A short discussion of thatproblem
isgiven
in the final section.2. Correlations and Green functions.
The main tool for evaluation of
height probabilities
and correlations between them is amapping
of the set ofsandpile configurations
onto the set ofspanning
trees. Dhar[8]
has shown that the numberl"~
ofsandpile configurations
in the SOC state, as well as the number ofspanning
trees, isgiven by
thesimple expression
~V"~ = det A.
(12)
The different
height probabilities
on a lattice fit can be related also to the enumeration of certain types ofspanning
trees. Inparticular,
the number ofsandpile configurations
with z, =I at a
given
site I of fitequals
the number-Vi''
ofspanning
trees on the same lattice fitwhich have
just
one leaf attached to andpointing
in a fixed direction[9]. Following
Majumdar
and Dhar[12],
one may construct a new lattice fit ~'~, such that thespanning
treescovering
fit ~'~ contain the fixed leaf. To this end, onejust
cuts off all the bonds attached to site I infit,
but the onecontaining
thegiven
leaf. As aresult,
a newtoppling
matrix A~' for thesandpile
on fit ~'' is obtained. Note that the defect matrix B= A~' A has non-zero
elements
only
for the site and its nearestneighbours
but one. Thus theprobability
offinding
the value z, = I is
given by
the formulaiJ°j
= ~~~ ~~'~
=
det
(I
+ GB), (13)
where I is the unit matrix and G
= A~ is the lattice Green function. For
example,
if I is far away from the boundaries,B,,
=
3 and
B~~ = I,
B~~ = B~, =
I for
j being
the left,right
and lowerneighbour
of I. In this case iJ° isindependent
of the choice of site I inside the bulk of the lattice fit. Thepair
correlation functionPi (r
) can be defined in a similar wayby constructing
a new
toppling
matrix A~'J for thelattice,
fit ~'J with defects of the above described typeplaced
at a distance r apart ;
where Bjj
is
the compound defect atrix.Thus, the problems we have ormulated in
the
introduction can bereduced
to theof
the
lattice Green unctions for
different
boundaryeterminants
(13)
or(14)
for the specifieddefects.
The
eigenvalues and
eigenfunctions of thediscrete
Laplacian A on a finite lattice therectangularshape nder the consideredboundary
conditions
are wellknown. For
a
latticeof
size M
x L with open boundary onditions on the verticalor
eriodiconditions
on
the orizontal edges,G)~/(nl,
ml n2,m2)
"
~j ~j
V~~~(nj)
Vj~~(ml)
j~j jj V~~~(n2~
V~~~(~2). (15)
~,~~,
Ap +A~
N° 8 BOUNDARY EFFECTS IN A TWO-DIMENSIONAL ABELIAN SANDPILE 1733
Here
n=1,2,..,L
labels the row andm=1,2,...,M
labels the column of siteI
=
(n,
m)
e fitv~(n
are theeigenfunctions
and A~ are thecorresponding eigenvalues
of the one-dimensional discreteLaplacian
underboundary
conditions(T ),
T = I(open),
2(closed)
and 3~periodic).
For the vertical direction one has thefollowing explicit expressions
~p = 1, 2,
,
L ; n = 1, 2,
,
L)
:for open
boundary
conditionsAl'
= 2Ii
CDS~l~i vi' ~(n)
-
l~
i
Sin
Ill ('61
for closed
boundary
conditions~l
j_A)~~ =
2
(1
cosI
;v)~J(n
) ='~
~ ~ ~~
j~cos (7r~p-1)(n-~ )/Lj, p=2,. ,L;
L 2
(17)
forperiodic boundary
conditionsAi~'-
2Ii
CDS~
i~
;vi~>(ni
-
)
~xP~
t~~ ('81
Passing
to the limit of aninfinitely long strip,
M- ~x~, we first redefine the Green function
(15)
as follows.ii
Shift theorigin
of the column number m and the row number n to the center, I,e, for an oddintegers
M, L we set m=
m' + ~ ~
n =
n' + ~
where mj, nj labels the columns 2
and rows from the central ones
(with
mj= 0, nj
=
0).
ii)
Shift the value of the Green functionby
a constant term to remove thedivergency
which appears as M- ~x~, even at finite L when
periodic boundary
conditions areimposed
on the horizontaledges, Evidently,
such a shift does not affect the value of the determinants(13), (14),
It is convenient to set the shift termequal
to the value ofG)(
atcoinciding
sites in thecenter of the lattice, Then, in the limit M - ~x~ we obtain
G)~~(Jll, ~~
,
Jl2,
Rli)
"
~
~lffl
~~)i~(Jl~+~)~,~i+~)~,~i+~)~,Rli+~)~
M-oJ
G(Tj
L + I M + L + I M +~~' 2 2 ' 2 2
~(T)~~
jj(Tl(~
j ~~~(~i
~i j ~~(Tl
~ + l p(Tl ~ + lL j
j"
P I p 2 2 p ~ p ~=
~
dry,
(19)
~,=
"
o
A~~J+ 2(1
cos cyConsidering
correlationsalong
the central row of thestrip
it is convenient to represent theresulting
Green function as a sum of two parts an even, C )~~, and an odd,S)~
', with respect to each of the argumentsn(
andii(
=
C[~'(n(, n( m( m()
+S)~~(n(, n( m( m(), (20)
In the remainder,
referring
to thestrip
geometry, we shalldrop
theprime superscript
of therow and column numbers,
remembering
thatthey
are counted from the center,Explicitly
wehave
for open
boundary
conditionsall)(~
~ ~~y)
~(~~
l~ ~' ~' L+I ~
j,_~ 7r
j«
cos[7r (2 I
+ I) nil (L
+ I)]
cos[7r (2 I
+ I) n~/ (L
+ Iii
cos ma~
o 2 cos
[7r (2 I
+ Inil (L
+ Iii
cos cy ~~ ' ~~~~S)'~(nj,
n~m)
=
j IL 'V2 j
j«
sin[2 grinj/(L
+ I)]
sin[2 7rin~/(L
+ Iii
cos maL + I
~j,
7r ~ 2 cos
[2 7ril
(L + Iii
cos cy ~~ ' ~~~~for closed
boundary
conditions~~~~~" ~~'~~
2
iL ~
~)~ )s y~
~~ ~j IL -'V2
j"
cos(2 grinj/L)
cos(2 grin~/L)
cos ma~ L
~j,
7r ~ 2 cos
(2 7rilL )
cos cy ~~ ' ~~~~S)~~(iij,
n~,ml
=
i IL ">2
j«
sin[7r (2 I
Inj/L]
sin[7r (2 I ) n~/L
cos maL
~j~
7r ~ 2 cos
[7r (2
I I)IL
coscy
~~
'
~~~~
for
periodic boundary
conditions~~~ i L j
j«
sin(2 grinj/L)
sin(2 grin~/L )
cos ma~~
~'~" '~~' ~ 2 L~j,
gr ~ 2 cos (2
7rilL)
coscy
~~ ~~~~
In the case of a
half-plane
geometry, one may start withexpression (19)
and take the limit L- ~x~, thus
obtaining
up tounimportant
constant the Green function for ahalf-plane
with openboundary,
~, j
j« j«
sin njfl
sin n~ fl cos maG
'(nj,
n~ ;m)
= j dry
dfl
,
(27)
w ~ ~ 2 cos fl cos cy
and closed
boundary
~~~
i
In j«
cos(nj 1/2) fl
cos(n~ 1/2)
fl cos ma~ ~'~~' '~~ ~
w2
~ ~ 2 cos fl cos a
~~
~~
~~~~N° 8 BOUNDARY EFFECTS IN A TWO-DIMENSIONAL ABELIAN SANDPILE 1735
3. The
strip
geometry.We consider the correlations between two unit
heights,
one at the center of thestrip,
ni "
0,
m= 0, and the other at a distance r apart, at site n~ =
0,
m~= r, If we label the rows and columns of the 8 x 8 matrix B
(r),
associated with thecorresponding
defects, in the order (- 1,0), (0, Ii, (0, 0), (0, Ii, (-
I,r), (0,
r1), (0, r), (0,
r + I),
it takes the block-diagonal
formBjj
=
~~ (29)
where
Bj
is the 4 x 4 matrix0
1
~~
~~~~0 0
The restriction
©~(r)
of the lattice Green function(19) (for simplicity
of notation, in theremainder of this section we omit the
superscript (T) indicating
thespecific boundary
conditions)
to thecorresponding
set of defect sites is an 8 x 8 matrix with the block structure/j
~~~AL(0) AL(r)
~
AL(-
?')
L(0)~
'
(3')
where,
using
therepresentation (20), A~(r)
may be written as the 4 x 4 matrixC~(I,
; r) +S~(I,
Ir) C~(1,
0r I
) C~(1,
0 ;r) C~(1,
0 ; r +ll
~~~'~~
~~[
~j~~ ~~ ~~~~[~j~~-1) ~~~~[~j~~
~~~~~[~j~~~~~
C~(0,1;r-I) C~(0,0;r-2) C~(0,0;r-I) C~(0,0;r)
(32)
andA~(0)
isgiven by (32)
at r =0.
According
to(14),
thepair
correlation functionPi j(r
L) isiJ'jj
(r ;Li
= det 11 +
©~(r) Bj (rii iJ'I(L/2), (33)
where
iJ°)(L/2)
=
lim det
[1
+©~(r) Bjj (r)
,
(34)
- a~
is the square of the unit
height probability
at the central line of thestrip,
Thus theproblem
reduces to the evaluation of the determinant
~~~ ~ + GL
(?')
B(r ii
= det
~ ~
~L ~°
l ~' AL
(?'I
B~L(- ")
B I +AL(-
rBj ~~~)
Using
theexplicit
form ofBj
and thesymmetry properties
of the even and oddcomponents
of the Green function
(20),
after obvioustransformations,
we obtain theexpression
det
[1+ ©~(r)Bjj(r)]
=
6~ S~(q,
1;0)j~ S)(I,
Ir))
detf
~~~l)
,(36)
R
(r) Q
where
Q
andR(r)
are the 2 x 2 matrices(R~
is the transpose ofRI,
C~ (0,
0C~(0,
0 ;C~(0,
;~ II
C~(0,
0 ;1) C~ (0, 1)
C~(0,
0)
' ~~~~C~(0,0;r)-C~(0,1;r)
C~(0,0;r+I)-C~(0,1;r+I)j
~ ~~
C~(0,
0 r IC~(0,
r IC~(0,
0 rC~(0,
0) ~~~~The
I-dependent
terms inequations (36), (38)
arereadily
evaluated as r- ~x~ at fixed L » I ; theirleading asymptotic
formdepends
on theboundary
conditionsI)
For openboundary
conditionsS~(I,
I r=
~
) exp(-
2 ml-IL )(39)
L
C~(0,
0r) C~(0, r)
=
~
exp(-
grr/L;
(40)
2 L
2)
For closedboundary
conditionsS~(I,
Ir)
=
~ exp(- grilL) (41)
L
C~(0,
0r) C~(0,
r=
~ exp(-
2 7rr/L(42)
L
3)
Forperiodic boundary
conditionsS~(I,
I ; r=
~
)
exp
(-
2 7rr/L )(43)
L
C~(0,
0r) C~(0,
r=
~ exp(-
2 7rilL) (44)
L
By
substitution of(40)-(44)
inexpression (36),
see also(33),
one obtains that theexponential decay
of correlations isgoverned by
the correlationlength f
= L/2 7r in the casesof open and closed
boundary conditions,
and f=
L/4 7r in the case of
periodic boundary
conditions. The finite-size corrections to the unitheight probability iJ°j(L/2),
see(34),
iJ°j(L/2)
=
[1+ C~(0,
0 ;2)][1+
2C~(0,
1;Iii( S~(I,
1;0)) (45)
can be obtained
by using
standardtechniques
based on the Poisson summation formula, seee-g-
[16].
All the relevant terms can beexpressed through
fourL-dependent integrals F~(L),
k= 1,
,
4,
i ii
i i ii ii
ii il+~iii~i~ iii
N° 8 BOUNDARY EFFECTS IN A TWO-DIMENSIONAL ABELIAN SANDPILE 1737
where,
&L(~b =
I(sin
~b +~jL
I
j-
'(soy
Thus we obtain :
Ii
For openboundary
conditionsC~(0,0;2)=
-1+ ~-4sj(2L+2)-4 s~(2L+2) (51)
C~(0,
;1)
= + 2
s~(2
L +2) (52)
S~(I,1;0)= -2s~(2L+2)-2s~(2L+2), (53)
2)
For closedboundary
conditionsC~(0,
02)
= +
~
4
e~(2 L)
+ 4s~(2
L) (54)
C~(0,
; 1)=
2
s~(2
L(55)
S~(I,
1;0)
= + 2
Fj(2L)
+ 2s~(2 L). (56)
3)
Forperiodic boundary
conditionsC~(0,
0 ; 2)= +
~
4
s~(2
L) + 4E~(2
L(57)
C~(0,
;1)
= 2
s~(2
L(58)
S~(I,
1;0)
=
2
s~(2L)
2s~(2L). (59)
The final results for the
strip
geometry are summarized inequations (8)-(11),
see the introduction,4. The
half-plane geometry.
With the aid of
expressions (27)-(28)
for the Greenfunctions,
one can use the determinant formula(13)
to obtain the unitheight probability
at a site whichbelongs
to the latticeboundary
3fif. In this case, the modification of the lattice consists incutting
off the two bonds which are attached to I andbelong
to theboundary.
If we label the defect sites in the order (- 1, 0), (0,
0), (1,
0),
the matrix associated with the defect is the 3 x 3 matrix of the form : for an openboundary
0
B)'J
=
3
,
(60)
0for a closed
boundary
0 B(~~ =
l 2
(61)
0
A direct evaluation of the
corresponding
determinants leads toexpressions (3)
and(4), given
in the introduction.
Of course, one could construct a different defect matrix which leads to the same
results,
forexample, by preserving
one of theboundary
bonds andcutting off,
instead ofit,
the bond which starts from I andpoints
inside the lattice.To obtain a more detailed information about the
spatial
structure of thesandpile,
we have studied also theasymptotic
behaviour of the unitheight probability iJ°j (I)
atlarge
distancesfrom the
boundary, I
» I. In this case, the defect matrix is the same as in thebulk,
see[12].
The
asymptotic
behavior of theboundary
Green function atlarge separations,
for both openand closed
boundary conditions,
is mosteasily
obtained with the aid of the method ofreflections. Then, one needs to know
only
theasymptotic expansion
for the bulk Greenfunction at
large separations (r
»1)
:G~~j~(0,
0 ;r)
= ln r
$
~ ~~ ~+ +
~~
~
+
(62)
2 " 2 " 4 " 24
grr~
480 grrwhere y is the Euler constant, and
simple
calculationsgive
the result(5).
By using
the discreteLaplace equation
to generate a sequence of recurrentrelations,
onemay obtain the
asymptotic expansion
forG~~j~(0,n;r),
where n=1,2,... Next, by
subtracting (in
the case of an openboundary)
oradding (in
the case of a closedboundary)
the Green function for a fictitious source,placed symmetrically
with respect to theboundary,
onecan
easily
find theasymptotic expansion
for the Green function whichcorresponds
to the consideredboundary problem,
Thus for openboundary
conditions we obtainG~'~(0,0;r)=(-j+.
,
(63)
grr 2grr
and for closed ones
G~~~(0,0;r)=- lnr-1-~~~~-
~-
~~
~+. (64)
" " 2
gr 6 grr 240 grr
The defect matrix used in the calculation of the unit
height
correlations at theboundary,
in therepresentation
when the defect sites are labeled in the order:(-1,0), (0,0), (1,
0), (r
1, 0 ),(r, 0), (r
+ 1, 0),
has theblock-diagonal
forrnBii
~
~~
,
(65)
where the 3 x 3 matrix
Bj
isgiven by
one of theequations (60), (61).
The restriction of the
boundary
Green function to the set of defect sites can be written in the block formA~ (0 )
AL(ri
©L(ri (661
~
~~~- ii AL(01
N° 8 BOUNDARY EFFECTS IN A TWO-DIMENSIONAL ABELIAN SANDPILE 1739
Finally
the evaluation of the determinants(14)
leads us to the results(6), (7) presented
in the introduction.5. Conclusion.
In summary, we have calculated the
leading asymptotic
form of the correlation functions for the unitheights, iJ°j (r
), in thestrip
andhalf-plane geometries
of the Abeliansandpile
model,under three different
boundary
conditions. The results confirrn thesuggestion
that the unitheight
behaves like the local energy operator in the q-component Potts model at q =0.
Unfortunately,
the above conclusion cannot bedirectly
extended to the case ofheights
z, = 2,
3, 4, although correspondences
between theseheights
andspanning
trees still exist. In[17]
it has been shown that z,=
2
corresponds
to such trees(in
addition to the treesyielding
z, = I
),
in which none of thepaths, starting
from three nearestneighbours
of site I andending
at the open
boundary,
passthrough
I. In the case z, =3 this property must have the
paths starting
from two of the nearestneighbours
of I, and if z, = 4 from oneneighbouring
site,Thus,
theheights
z,=
2, 3, 4, being
localsandpile
observables,happen
to be connected with nonlocalproperties
of the trees. For the enumeration ofconfigurations
with theseheights
onehas to consider clusters of subtrees with
increasing
size. It has been established[12]
that suchcluster
expansions
areslowly
convergent, which makes difficult the evaluation of thecorrelation functions
P~~(r)
for z= 2, 3, 4. In
principle,
the above mentioned features may lead to a different fromr~~
power law ofdecay.
Finding
acorrespondence
between observables in the Potts model and characteristics of the avalanches is not an easier taskeither,
because of the nonlocal nature of the avalanches. The solution of theseproblems
needs furtheranalytical
and numerical studies.6.
Acknowledgements.
This work was
supported
in partby
theBulgarian
National Foundation for Scientific Research, Grant No. MM-74/91.References
[ii Bak P., Tang C., Wiesenfeld K., Phys. Rev. Len. 59 (1987) 381;
Phys.
Rev. A 38 (1988) 364.[2] Kadanoff L. P., Nagel S. R,, Zhou S. M., Phys. Rev. A 39 (1989) 6524.
[3] Hwa T., Kardar M,, Phys. Rev. Lett. 62 (1989) 1813.
[4] Manna S, S., J. Stat. Phys. 59 (1990) 509.
[5] Obukhov S. P., Random Fluctuations and Pattern Growth :
Experiments
and Models, H. E. Stanley, N. Ostrowsky Eds., NATO Advanced Study Institutes, Ser. C, Vol. 157 (Kluwer, Dordrecht,1988).
[6] Privman V., Finite Size Scaling and Numerical Simulations of Statistical Systems, V. Privman Ed.
(World Scientific,
Singapore,
1990).[7] Cardy J, L., Phase Transitions and Critical Phenomena, C. Domb, J. L. Lebowitz Eds., Vol.11 (Academic Press, London, 1987) p. 55.
[8] Dhar D., Phys. Rev. Leit. 64 (1990) 1613.
[9] Majumdar S, N., Dhar D., Physica A 185 (1992) 129.
[10] Wu F. Y., Ret'. Mod. Phys. 54 (1982) 235.
[I ii Fortuin C. M.,
Kasteleyn
P. W., Physica 57 (1972) 536.[12]
Majumdar
S. N., Dhar D., J. Phys. A 24 (1991) L357.[13] Cardy J. L., Nucl. Phys. B 240 (1984) 514.
[14] Dotsenko V. S., Nucl. Phys. B 235 (FS II) (1984) 54.
[15] Cardy J. L., J. Phys. A 17 (1984) L385.
[16]
Singh
S., Pathria R. K.. Phys. Ret,. B 31 (1985) 4483Physic-a
A 166 (1990) 1.[17] Priezzhev V. B., Preprint DIAS-STP-92-10, Dublin Institute for Advanced Studies (1992).