ebook img

Site percolation and random walks on d-dimensional Kagome lattices PDF

11 Pages·0.17 MB·English
Save to my drive
Quick download
Download
Most books are stored in the elastic cloud where traffic is expensive. For this reason, we have a limit on daily download.

Preview Site percolation and random walks on d-dimensional Kagome lattices

8 9 9 1 Site percolation and random walks on d-dimensional Kagom´e n a lattices J 3 1 Steven C. van der Marck SIEP Research and Technical Services ] h P.O. Box 60, 2280 AB Rijswijk, The Netherlands c e February 1, 2008 m - t a Abstract t s . The site percolation problem is studied on d-dimensional generalisations of the Kagom´e t a lattice. These lattices are isotropic and have the same coordination number q as the hyper- m cubiclattices in d dimensions, namely q=2d. The site percolation thresholds are calculated - numericallyford=3,4,5,and6. Thescalingofthesethresholdsasafunctionofdimensiond, d oralternativelyq,isdifferentthanforhypercubiclattices: pc∼2/q insteadofpc ∼1/(q−1). n ThelatteristheBetheapproximation,whichisusuallyassumedtoholdforalllatticesinhigh o dimensions. A series expansion is calculated, in order to understand the different behaviour c [ of the Kagom´e lattice. The return probability of a random walker on these lattices is also shown to scale as 2/q. For bond percolation on d-dimensional diamond lattices these results 1 imply pc ∼1/(q−1). v 2 1 1 Introduction 1 1 The Kagom´e lattice is one of the most interesting lattices in two dimensions. It is one of the 0 8 eleven Archimedean tiling lattices, where all the vertices are of the same type (see e.g. Weisstein 9 1997). In the case of the Kagom´e lattice each vertex touches a triangle, hexagon, triangle, and / a hexagon. All these polygons are regular. Moreover, the Kagom´e lattice is closely related to t a the other lattices in two dimensions. The sites of the Kagom´e lattice correspond to the bonds m of the honeycomb lattice, which in turn is the dual of the triangular lattice. Therefore, since - the bond percolation threshold of the honeycomb lattice is 1 2sin(π/18), the site percolation d − n threshold of the Kagom´e lattice is 1 2sin(π/18) = 0.6527036... too (Sykes and Essam 1964). − o The bond percolation threshold is not known exactly, but has been calculated numerically with c high precision to be 0.5244053 0.0000005(Ziff and Suding 1997). : ± v Although these percolation thresholds have been known for quite some time, it is not clear i why this site percolation threshold is high, compared to other lattices. For instance, it is much X higher than the threshold 0.5927460 0.0000005 for the square lattice (Ziff and Sapoval 1986, r ± a Ziff 1992), although its coordination number q =4 is equal to that of the Kagom´e lattice. What is more striking is that eventhe pentagonallattice, which has a low averagecoordinationnumber of q = 31, has a lower site percolation threshold 0.6471 0.0006 than the Kagom´e lattice (Van 3 ± der Marck 1997). In other words, the site percolation thresholds are not ordered according to the coordination number q. This runs contrary to common intuition, which leads one to expect that a lattice with a higher connectivity has a lower percolation threshold. If one searches for general formulas that correlate percolation thresholds with dimension and coordination number, the Kagom´e lattice therefore poses a problem. Galam and Mauger (1996) introduced different classes of lattices to avoid this problem, and they used different correlations for these classes. Although this enabled them to derive good correlations, it prompts the question why certain lattices belong to one class and others to another class. 0 An analogue in three dimensions was found recently: there is a lattice with q = 6 and perco- lation threshold 0.3898 0.0008 (Van der Marck 1997b). Compared to the simple cubic lattice, ± which also has q =6, but a threshold of 0.3114 0.0004,this threshold is much higher. It is even ± higher than the threshold for several lattices with coordination number q = 5 (Van der Marck 1997). In this paper, a generalisation of the Kagom´e lattice to higher dimensions is given (Sec. 2), and numerical calculations of the site percolation thresholds for 3, 4, 5, and 6 dimensions are presented (Sec. 3). The scaling of these thresholds as a function of dimension appears to be different than for hypercubic lattices. For the latter, Gaunt, Sykes and Ruskin (1976) calculated a series expansion in 1/(2d 1), where d is the number of dimensions. The leading term in their − series is p (d) = 1/(2d 1), which is the so-called Bethe approximation. This approximation c − holds exactly for Bathe lattices (see e.g. Stauffer and Aharony 1992). For hypercubic lattices, the approximation underestimates the percolation threshold in low dimensions, but improves in accuracyinhigherdimensions. InSec.4theseriesexpansionforthed-dimensionalKagom´elattices is studied. It is suggested that in this case the leading term in the series is 1/d, not 1/(2d 1). − The return probability of a random walker on d-dimensional Kagom´e lattices is discussed in Sec. 5. Ishioka and Koiwa (1978) conjectured that this probability is a good estimator for the percolation threshold. Indeed it is shown here that this return probability also scales as 1/d. Section 6 contains a discussion of the results. These results give some insight into the problem why the 2-dimensional Kagom´elattice has a high site percolation threshold. 2 Kagom´e lattices in d dimensions The Kagom´e lattice can be defined in d dimensions as follows. The lattice has a (d+1)-point basis, and these pointsformaregulard-dimensionalpolytope. Allthepointsofthisbasisaredirectneighbours ofeachother. Letusdenotethebasispointsbyb ,fori=0,...,d. Thelatticecanbeconstructed i by translation of the basis in 1d(d+1) directions. These translations are given by the vectors 2 2(b b ) for j =i. This is a dependent set of vectors. One can select a minimal set of d vectors j i − 6 by setting e.g. i=0, j =1,...,d. A site of the lattice can be identified by its number i within the base polytope (runs from 0 to d), and the translation x with respect to a reference position. Consider a site i,x . This site has d neighbours in the same polytope, j = i,x , and another d neighbours{in a}djacent polytopes, j = i,x+2(b b ) . Therefore{the6 coord}ination number of this lattice is q = 2d. i j { 6 − } Also,becausethereisnopreferentialdirectionintheconstruction,allthedirectionsareequivalent for this lattice, i.e. it is an isotropic lattice. So the Kagom´e lattice resembles the cubic lattice in the sense that both are isotropic d- dimensionallatticeswithcoordinationnumberq =2d. However,wealreadyknowthatintwoand three dimensions, the site percolation thresholds of these lattices are distinctly different. IntwodimensionsthesitepercolationproblemontheKagom´elatticeisequivalenttothebond percolationproblemonthehoneycomblattice(SykesandEssam1964). Thiswasshownbymeans of the star-triangle transformation. Analogous to the star-triangle transformation, one can use a ‘star-tetrahedron’ transformation in three dimensions, see Fig. 2. The site percolation problem on the tetrahedron ABCD, i.e. the three-dimensional Kagom´e lattice, is equivalent to the bond percolation problem on the dashed lattice. This dashed lattice is the diamond lattice, for which p = 0.3893 0.0003 has been calculated (Van der Marck 1997). Generalising to d dimensions, c,b ± the site percolation on the d-dimensionalKagom´elattice is equivalent to bond percolation on the d-dimensional diamond lattice, which has q =d+1. The Kagom´e lattice defines a tiling of d-dimensional space. In two dimensions, the Kagom´e lattice defines a tiling of the plane in terms of a regular triangle and a regular hexagon. One can construct this hexagon from the base triangle: take three neighbouring base triangles, as in Fig 1b. These three form a larger regular triangle. The hexagon appears when one truncates the larger triangle by taking away the three smaller triangles. In two dimensions this is a rather 1 complicateddescriptionofthe tiling,butthe advantageisthatonecanuseanidenticalprocedure inddimensions. Onecanstartwiththeregularbasepolytopeof(d+1)pointsinddimensions,b i (fori=0,...,d). Thispolytopeisshiftedinddirectionsbythevectors2(b b )fori=1,...,d. i 0 − This defines a larger regular polytope, which is then truncated by taking away the (d+1) small polytopes. In three dimensions, e.g., the resulting polyhedron is the truncated tetrahedron (see e.g. Weisstein 1997). The d-dimensional space is filled with the base regular polytope and the larger truncated polytope. In view of the considerations in sections 4 and 5, it is helpful to discuss one more important property of the d-dimensional Kagom´e lattice: two adjacent sites have d 1 common neighbours. − This is demonstrated by the following arguments. When the two sites are within the same base polytope, characterised by i,x and j,x (j = i), all the other d 1 sites of that polytope { } { } 6 − are common neighbours. When the two sites are not within the same base polytope, they are characterisedby i,x and j,x+2(b b ) . Note that j cannot be equal to i for two adjacent i j sites. The d 1 c{omm}onne{ighbours oft−hese}two sites are k,x+2(b b ) with the restriction i k that k = i,j−. This is because they are one step b b a{way from i−,x , a}nd one step b b i k k j from j6,x+2(b b ) . − { } − i j { − } 3 Percolation thresholds The site percolation thresholds of the Kagom´e lattice in 3, 4, 5, and 6 dimensions, are show in Table 1 as afunction of the lattice size. I usedthe method givenby Stauffer andAharony(1992). This method is simple to program,but in its simplest formit has the drawbackthat one needs to haveallN =(d+1)Ldsitesofanetworkresidentinmemory. Thereforethecalculationisrestricted to relatively low values of L, the linear size of the lattice. Especially in higher dimensions this is rather restrictive: the highest L-value used was L = 14 for the 6-dimensional Kagom´e lattice. One can check in a scaling plot, Fig. 3,that the used lattices areindeed largeenoughto be in the scaling regime, and hence allow extrapolation to infinite lattice sizes. For each of the lattices, percolation in only one direction, say the x -direction, was checked. 1 The lattice was defined to be percolating whenever the two boundaries i = 0,x = 1 and 1 i = d,x = L were connected. The boundary conditions in the othe{r directions wer}e not 1 { } periodic. The percolation thresholds of a lattice of size N obey the scaling relation p (N) p ( ) N−1/(νd). (1) c c | − ∞ |∼ Herethecriticalexponentν is0.88inthreedimensionsand0.68,0.57,0.5in4,5,and6dimensions respectively (Stauffer andAharony 1992). The results quoted inthe rowmarked in table 1 are ∞ fits of p (N) to this scaling relation, using the last three points in the table. c The data and the fits are shown in Fig. 3 for d = 3 and d = 6. From the figure it can be concluded that the input values for the fits are indeed within the scaling regime, and that the fits are thereforesensible. For d=6 the difference p (N) p ( ) is largerthanfor lowerdimensions, c c − ∞ because the number of lattice sites that are near to a boundary increases with dimension. Also shown in Table 1 are results for the percolation thresholds of the cubic lattice. These numbers agree with the literature (Stauffer and Aharony 1992) and can be seen as a check that the programs used in the present work are correct. The difference between the percolation threshold for the Kagom´e lattices and the (hyper-) cubic ones increases with dimension. This is illustrated in Fig. 4. In the next section a series expansion is studied, in order to understand this behaviour. 4 Series expansion One could conjecture, based on the numerical estimates of the percolation thresholds, that the scaling of these thresholds as a function of dimension is different for the Kagom´e lattice than for 2 the cubic lattice. The scaling behaviour of the cubic lattice was studied by Gaunt et al. (1976). Their result for the percolation threshold was 1 3/2 15/4 83/4 p (d)= 1+ + + +... . (2) c 2d 1 2d 1 (2d 1)2 (2d 1)3 − (cid:18) − − − (cid:19) The leadingtermis the percolationthresholdofaBethe lattice withcoordinationnumberq =2d. Equation 2 agrees well with the known numerical estimates for the cubic lattices for d 3, as is ≥ clear from Fig. 4. Gauntet al. (1976)usedclustercountingtocalculateaseriesexpansionforthemeansizeS of clusters at low probabilities p. Since the mean cluster size diverges at the percolation threshold, theystudiedtheradiusofconvergenceoftheseriesexpansionofS. Theyarguedthatthisradiusof convergencewillbedeterminedpredominantlybythesingularityatp=p ,whichenabledthemto c deriveEq.(2). Inthis sectionthese techniques areusedto study the percolationthresholds ofthe Kagom´elatticeinhigherdimensions. Atthebasisoftheapproacharethe‘perimeterpolynomials’, as described by Sykes and Glen (1976). Denoting the mean number per lattice site of clusters of s sites by n¯ , we have e.g. s n¯ = pq2d , 1 n¯ = dp2 q3d−1 . (3) 2 Herep isthe probabilitythata siteis occupied,andq =1 pthe probabilitythatasite is empty. − Onecaninterpretp(s)=sn¯ astheprobabilitythatasiteisoccupiedbyaclusterofsizes. When s one sums p(s) overallcluster sizes,one gets the probabilitythat a site is occupiedby any cluster, i.e. p: ∞ ∞ sn¯ = p(s)=p. (4) s s=1 s=1 X X The power 3d 1 of q in the expressionfor n¯ indicates that all clusters of 2 sites are surrounded 2 by3d 1neigh−bours. Alreadyweseeadifferencewiththecubic lattice,whichhasn¯ =dp2q4d−2. 2 − So eventhough a site on the Kagom´elattice has as many neighboursas a site onthe cubic lattice (namely 2d),a2-siteclusterhaslessneighboursonaKagom´elattice thanonacubic lattice. This is because on the Kagom´e lattice two adjacent sites have d 1 common neighbours, whereas on − the cubic lattice they have none. A relativelysimple counting procedure by hand revealsthe next few perimeter polynomials to be n¯ = 1d(4d 1)p3 q4d−2, (5) 3 3 − n¯ = 1 d(5d 1)(5d 2)p4 q5d−3, 4 12 − − n¯ = 1 d(6d 1)(6d 2)(6d 3) 2d(d 1) p5 q6d−4 5 60 − − − − − +(cid:2) 2d(d 1)p5 q6d−5. (cid:3) − In Figs. 5, 6 and 7 the contributing graphs are depicted in two dimensions. In the expressions for n¯ ,...,n¯ a clear pattern seems to emerge. The coefficient of n¯ has factors (s+1)d 1, 1 4 s − (s+1)d 2, ..., and the power of q is always (s+1)d (s 1). − − − However, this trend is broken in n¯ . Although the sum of the two coefficients appearing here 5 still has the factors (s+1)d 1, etc., there are now terms with different powers of q. In other − words, in n¯ we encounter for the first time that there are clusters with different numbers of 5 neighbours; clusters with 6d 4 neighbours and clusters with 6d 5 neighbours. The latter ones − − aredepicted inFig.8. They arethe type ofcluster that ‘bites itself inthe tail’. Atfirstsight,one would think that this would occur already for clusters of two sites, but in those cases the number of neighbours is not affected. This number is only affected for the clusters depicted in Fig. 8. On the cubic lattice this occurs for the first time for clusters of 3 sites. Another difference is that for the cubic lattice the leading power of q is 2sd 2(s 1), compared to (s+1)d (s 1) for the − − − − Kagom´e lattice. 3 The identity (4) can be exploited to determine one more mean cluster number, namely n¯ , 6 albeit in the limit of q 1. Following Sykes and Glen (1976) I substitute Eqs. (3) and (5) into → the identity (4), and set q = 1 p. This yields a power series in p, of which the coefficient of − p is 1, and the coefficients of p2,...,p5 vanish identically. The higher order coefficients do not vanish, because the number of clusters of six or more sites were not included. Because n¯ is the 6 only missing term that can contribute terms of order p6, we can conclude 1 5 n¯ = d(7d 1)(7d 2)(7d 3)(7d 4) d(d 1) p6. (6) 6|q→1 360 − − − − − 3 − (cid:20) (cid:21) The mean size of clusters at low probabilities, S, is defined as sp(s) 1 S = s = s2 n¯ . (7) s p(s) p P s s X P Again Eqs. (3) and (5) can be used, and q =1 p, to derive a power series − S(p) = 1+2(dp)+2(dp)2+2(dp)3+2(dp)4 10 1 +2(dp)5 1+ 1 + p6 . (8) d3 − d O (cid:20) (cid:18) (cid:19)(cid:21) (cid:0) (cid:1) The first few terms in this low density expansion of S(p) are remarkably simple. It is almost a geometric series, until the fifth power. The extra terms in the coefficient of (dp)5 are due to the clusters that ‘bite in their own tail’. Althoughthe resemblanceofS(p)toageometricseriesis notexact,itdoes suggestthat1/dis the obvious candidate for expansion parameter. This in contrast to the cubic lattice, where it is 1/(2d 1), see Eq. (2). In fact, the resemblance to a geometric series suggests that a singularity − ofS(p) shouldoccur in the vicinity ofdp=1. Therefore one canexpectthe percolationthreshold to scale as 1 p . (9) c ∼ d In Fig. 4 the relation p =1/d is shown with a dashed line. The percolation thresholds for d=5 c and 6 are already reasonably well approximated by this relation. In an attempt to calculate the percolation threshold for d=8, I computed 0.1059 0.0005 for L=5, 0.1086 0.0006 for L=6, ± ± and 0.1117 0.0005for L=7. Based on the last two points one can fit the percolation threshold ± to be 0.120 0.003, which is close to the value 0.125 one would expect on the basis of Eq. (9). ± However,thepointforL=5doesnotlieonthesamefit,indicatingthattheselatticesizesarenot yet large enough. Therefore there is probably also a small systematic error in the determination of the value 0.120. WhenonewantstorefinethescalingbehaviourgivenbyEq.(9)withtermsoftheorderof1/d2 and further, analogous to Eq. (2), more terms in the series expansion would be required, plus a careful mathematical analysis of the radius of convergenceof the series. This is beyond the scope of the present paper. 5 Random walks ItisinterestingtostudythereturnprobabilityP ofarandomwalkeronad-dimensionalKagom´e r lattice. Ishioka and Koiwa (1978) suggested that P is an upper bound for the site percolation r threshold on any lattice: P p . For Bethe lattices, the equal sign holds (Hughes and Sahimi r c ≥ 1982). The arguments given by Ishioka and Koiwa to support their conjecture are not exact, but therelationappearstoworkfairlywell. P andp lieclosertogether,themoreconnectedalattice r c is. AlsoSahimiet al. (1983)studiedtherelationbetweenarandomwalker(notself-avoiding)and percolation, albeit bond percolation. The return probability of a random walker can be calculated numerically with a simple com- puter program. One can let N walkers perform at maximum N steps, and count the number w s 4 of walkers that have re-visited the site they started from. Alternatively, one can let N walkers w perform steps until they are either back at the origin or further away from the origin than a certain predefined distance. I have used both methods to estimate P with an estimated inaccu- r racy of about 0.001. Note that there are two sources of inaccuracy. The first one is a statistical uncertainty, which scales as 1/√N . The second one is a systematic error, because each walker w is stopped at a certain moment (after N steps or at a given distance from the origin). For each s of these walkers there is a finite probability that they would have reached the origin, when given enough time. As a result, the numerical estimates have a systematic error to the downside. This bias can be made smaller by using a large number of steps, or a large cut-off distance. In the calculation of the numbers quoted in Table 1, I used N = 106 and higher, and N = 105. The s w results for the cubic lattices are consistent with Finch (1997) and Flajolet (1995). ThenumericalvaluesforP areclosetothepercolationthresholds,especiallyford 4. Itlooks r ≥ as if the return probability of a random walker obeys the same scaling relation as the percolation threshold. In the remainder of this section I therefore calculate a crude approximation for the return probability using simple arguments. This approximation corroborates the scaling of the return probability as P 1/d. r ∼ Consider a random walker on the d-dimensional Kagom´e lattice. Since all directions of the lattice are equivalent, it does not make a difference which step the walker makes first. Assume, withoutlossofgenerality,thatthewalkerremainswithinthebasepolytope i,0 . Assumefurther, { } asafirstapproximation,thathestayswithinthispolytopeforanumberofsteps,andthenreturns to the origin. The walker can take an arbitrary number of steps within this base polytope, with a probability (d 1)/(2d) (the walker can choose from 2d directions, d of which are outside the − base polytope, and one of the 2d is the origin). After a number of steps the walker should step back to the origin, which happens with a probability 1/(2d). This approximationyields ∞ s−2 1 d 1 1 P = − = . (10) r,1 2d 2d 1+d s=2 (cid:18) (cid:19) X Thisapproximationalreadyrevealsaninterestingpoint: inhighdimensions,thereturnprobability scales as P 1/d. Since the probability 1/(1+d) is the exact probability of return via a few ∼ selected paths, we also know that the exact probability on return via any path will be higher than 1/(1+d). It is therefore impossible that it is as low as 1/(2d 1), as is the case for cubic − lattices. One can improve on the above approximation by allowing the walker to step outside the base polytope occasionally. Choose for instance i sites x from which the walker steps outside. If the i walker makes s steps within the base polytope before returning to the origin, there are (s−1) i possible choices (because the last of the s steps is to the origin): ∞ 1 d 1 s−2 s−1 s 1 1 i Pr,2 = 2d 2−d × −i 4d . s=2 (cid:18) (cid:19) i=0(cid:18) (cid:19)(cid:18) (cid:19) X X The factor 1/(4d) emerges because the walker has a probability of 1/2 to step outside the base polytope, and a probability 1/(2d) to immediately step back to x . One can also allow the walker i to make a number ofadditionalsteps, as long as he is only one step awayfromx andoutside the i base polytope. There are (d 1) possibilities out of a total of 2d to make such a step. − ∞ s−2 s−1 ∞ k i 1 d 1 s 1 1 d 1 Pr,2 = s=2 2d(cid:18) 2−d (cid:19) ×i=0(cid:18) −i (cid:19)"4dk=0(cid:18) 2−d (cid:19) # X X X 1 1+3/(2d) = . (11) d · 1+3/(2d)+3/(2d2) This type of reasoning can be taken a step further, yielding 1 1+2/d+9/(4d2) P = . (12) r,3 d · 1+2/d+11/(4d2)+9/(4d3) 5 Allthepathsthatareincludedinthiswayarepathswherethewalkerreturnstotheoriginthrough the base polytope he started in. The first contributions from paths that return from the opposite side are paths of six steps, see Fig. 8. As there are 2d(d 1) of such paths, the probability to − return from the opposite side in six steps is (d 1)/(2d)5, which is of the order 1/d4. Thus the − conclusion that the return probability scales as 1/d remains unchanged. As a numerical check on this scaling behaviour, I calculated the return probability for the 8- dimensionalKagom´elattice: P =0.126 0.001. Thiscompareswellwiththeapproximation0.124 r ± from Eq. (12). 6 Discussion The Kagom´elattice has rather different properties in d dimensions than the cubic lattice: it does not behave as a Bethe lattice in high dimensions, although one usually assumes that all lattices do. The series expansion of Section 4 provides some insight into why this is the case. One clear difference with the cubic lattice is that the leading power of q in the expression for the number of clusters of s sites is (s+1)d (s 1) for the Kagom´e lattice, but 2sd 2(s 1) for the cubic − − − − lattice. Thecoefficientofsddiffersbyafactor2here. Thesignificanceofthis powerofq isthatit is the ‘coordinationnumber’of acluster ofs sites, just asq is the coordinationnumber ofa single site. The underlying reasonfor the different powers ofq is that two adjacentsites on the Kagom´e lattice alwayshaved 1commonneighbours,whereastwoadjacentsitesonthe cubic latticehave − none. Inhighdimensionsthisdifferencedoesnotdisappear,butinsteadbecomesmoreimportant. The same difference also explains why the return probability is different for these lattices. Since adjacent sites have d 1 common neighbours, there is, in high dimensions, an increasing − number ofwaystoreturnto the originin threesteps. OnaBethe lattice, suchpaths donotexist: here a walker can only return to the origin by retracing his steps. Therefore the Kagom´e lattice does not behave as a Bethe lattice in high dimensions. Going back to the 2-dimensional Kagom´e lattice, one could say that one of the reasons why it has a much higher site percolation threshold than the square lattice, is because of its ‘common neighbours’. We can apply the same reasoning to bond percolation. The bond percolation threshold of the Kagom´elattice is alsohigher than that ofthe squarelattice, but the difference is smaller than for the site percolation threshold. Two adjacent bonds on the square lattice always have 2 common neighbours. OntheKagom´elattice,twoadjacentbondshave,dependingontheirrelativeposition, 2 or 3 commonneighbours. The former happens in 2 of the situations, the latter in 1. Therefore, 3 3 for bond percolation the Kagom´e lattice has more resemblance to the square lattice, but still the two lattices are different. This is consistent with the difference in bond percolation thresholds being smaller, but not negligible. Finally, let us consider the relation to the diamond lattice in d dimensions. Since the site percolation threshold of the Kagom´e lattice scales as p 1/d = 2/q, this holds for the bond c ∼ percolation threshold of the diamond lattice as well. But for the diamond lattice q = d+1, so p 1/(q 1), which is the familiar Bethe result once more. This highlights a more general c,dia ∼ − property. Eachbondproblemcanbemappedontoasiteproblem(seee.g.Kesten1982). Whenthe bond problem is on a lattice with q =q , the lattice for the corresponding site problem will have b q =2(q 1). Asaconsequence,foreachclassoflatticesforwhichthebondpercolationthresholds s b − scale as p f(q), there is another class of lattices for which the site percolation thresholds scale c ∼ as p f(1q+1). It would therefore be more consistent to describe bond percolationin terms of c ∼ 2 the number of neighbours a bond has, which is 2(q 1). − Acknowledgments I would like to thank Ed Stephens for critically reading the manuscript and Shell International Exploration & Production for permission to publish this paper. 6 References Finch S 1997 http://www.mathsoft.com/asolve/constant/polya/polya.html Flajolet P 1995 sci.math.researchnewsgroup, July 10. Galam S and Mauger A 1996 Phys. Rev. E 53 2177–2181 Gaunt D S, Sykes M F, and Ruskin H 1976 J. Phys. A 9 1899–1911 Gaunt D S and Ruskin H 1978 J. Phys. A 11 1369–1380 Hughes B D and Sahimi M 1982 J. Stat. Phys. 29 781–794 Ishioka S and Koiwa M 1978 Phil. Mag. A 37 517–533 Kesten H 1982 Percolation Theory for Mathematicians (Boston: Birkh¨auser) Sahimi M, Hughes B D, Scriven L E, and Davis H T 1983 J. Phys. A 16 L67–71 Stauffer D andAharony A 1992Introduction to Percolation Theory (London: Taylorand Fran- cis) Sykes M F and Essam J W 1964 J. Math. Phys. 5 1117–1127 Sykes M F and Glen M 1976 J. Phys. A 9 87–95 Van der Marck S C 1997 Phys. Rev. E 55 1514–1517;Erratum 56 3732 — 1997b Phys. Rev. E 55 6593–6397 Weisstein E W 1997 http://www.astro.virginia.edu/ eww6n/math/, to be published as ∼ Treasure Trove of Mathematics (Boca Raton, Florida, CRC Press Inc) Ziff R M 1992 Phys. Rev. Lett. 69 2670–2673 Ziff R M and Sapoval B 1986 J. Phys. A 19 L1169–1172 Ziff R M and Suding P N 1997 J. Phys. A 30 5351 Table 1: The site percolation thresholds of the cubic and Kagom´e lattices in 3, 4, 5 and 6 dimensions, as a function of the linear lattice size L. The values for the cubic lattices given by Stauffer and Aharony (1992) are 0.3116, 0.197, 0.141, and 0.107. In the last row, marked ‘rw’, the values for the return probability of a random walker are given. The estimated error margins concerning the last digits are indicated between brackets. d=3 d=4 d=5 d=6 L cubic Kagom´e L cubic Kagom´e L cubic Kagom´e L cubic Kagom´e 8 0.2088(2) 0.2787(2) 8 0.1425(2) 0.2080(3) 6 0.1043(2) 0.1630(2) 16 0.3233(2) 0.3952(2) 12 0.2037(2) 0.2753(2) 12 0.1413(2) 0.2076(2) 8 0.1049(2) 0.1643(3) 32 0.3171(2) 0.3924(2) 16 0.2014(2) 0.2736(2) 16 0.1412(2) 0.2078(2) 10 0.1060(2) 0.1652(3) 64 0.3139(2) 0.3908(2) 25 0.1989(2) 0.2726(2) 20 0.1407(2) 0.2080(2) 12 0.1064(2) 0.1660(3) 128 0.3125(2) 0.3903(2) 32 0.1987(2) 0.2724(2) 24 0.1406(2) 0.2081(2) 14 0.1067(2) 0.1664(3) 250 0.3119(2) 0.3897(2) 50 0.1974(2) 0.2719(2) 32 0.1406(2) 16 0.1070(2) ∞ 0.3114(2) 0.3895(2) ∞ 0.1967(3) 0.2715(3) ∞ 0.1407(3) 0.2084(4) ∞ 0.1079(5) 0.1677(7) rw 0.343(1) 0.417(1) rw 0.195(1) 0.274(1) rw 0.136(1) 0.208(1) rw 0.105(1) 0.170(1) 7 a b Figure 1: (a) A description of the Kagom´e lattice as a lattice with a 3-point basis. The basis points form a (regular)triangle. (b) Three base trianglesdefine a largertriangle, which truncates to a hexagon. A X B D C Figure 2: The star-tetrahedrontransformation. The dashed lines AX,...,DX, that come together in the point X, form the diamond lattice. 8 10-2 d=6 ∞) d=3 (c p N) - (c p 10-3 10-4 104 105 106 107 108 N Figure 3: The scaling of the percolation threshold with network size N =(d+1)Ld. 1 data Kagome′ data cubic Eq. (2) 1/d 0.5 c Kagome′ p 0.2 cubic 0.1 0.05 2 3 4 5 6 7 8 9 10 dimension d Figure 4: The percolation thresholds as a function of dimension d. Eq. (2) is the result of Gaunt et al. (1976) for the thresholds of the cubic lattices. The thresholds for the Kagom´elattices scale as 1/d. (cid:18) (cid:19) 2 2 d d 3 2 Figure 5: The clusters of 3 sites. 9

See more

The list of books you might like

Most books are stored in the elastic cloud where traffic is expensive. For this reason, we have a limit on daily download.