One-dimensional long-range percolation: a numerical study
OOne-dimensional long-range percolation: a numerical study
G. Gori, ∗ M. Michelangeli, N. Defenu,
2, 1 and A. Trombettoni
1, 2, 3 CNR-IOM DEMOCRITOS Simulation Center, Via Bonomea 265, I-34136 Trieste, Italy SISSA, Via Bonomea 265, I-34136 Trieste, Italy INFN, Sezione di Trieste, Via Bonomea 265, I-34136 Trieste, Italy
In this paper we study bond percolation on a one-dimensional chain with power-law bond prob-ability
C/r σ , where r is the distance length between distinct sites. We introduce and test anorder N Monte Carlo algorithm and we determine as a function of σ the critical value C c at whichpercolation occurs. The critical exponents in the range 0 < σ < ε -expansion results. Our analysis is in agreement, up to a numerical precision ≈ − , with the mean field result for the anomalous dimension η = 2 − σ , showing that there is nocorrection to η due to correlation effects. PACS numbers:
I. INTRODUCTION
Percolation is a paradigmatic model of statistical mechanics [1]. Despite its simplicity it is able to capture manyfeatures of real world phenomena ranging from coffee brewing to oil industry (see [2] for a recent review) and atthe same time it allows for various exact results [3, 4]. One of the most remarkable features of percolation is todisplay phase transition phenomena with nontrivial geometric properties [5] prototyping in a simplified and essentialstripped-to-the-bone setting features found in more sofisticated models.A significant bulk of the work regarding percolation deals with models where probability of connecting two elementsis zero beyond a given range, i.e. short-range (SR) models. The typical SR percolation refers to a non-vanishingprobability to connect nearest neighbour elements or sites of a lattice [3, 4]. In SR percolation the one-dimensionallimit is trivial: if one denotes by C the probability to connect with a bond two nearest neighbour sites one immediatelysees that in one dimension the critical value C c equals 1 (otherwise the two ends of the chain cannot be connected).This result parallels the well known results that for classical Ising and O(N) chains the critical temperature is vanishingfor SR interactions [6, 7]. However it is well known that adding long-range (LR) couplings in the one-dimensionalIsing model one may have a non vanishing critical temperature [8, 9]. Therefore in the literature it has been exploredthe percolation in presence of long-range (LR) interactions, for which one expects a percolative transition also in onedimension [10].Systems where the elementary constituents affect each other with laws that decay slowly as a function of theirmutual distance play a central role in physics including, among others, gravitational systems, unscreened plasmas and(di)polar systems [11]. A typical form of the interaction is ∝ /r d + σ , where r is the distance between elements orsites of the system having dimension d , and σ quantifies the range of the interactions.The inclusion of LR interactions into percolation models, with bond probability of the form ∝ /r d + σ , proves valu-able also for modeling of several phenomena belonging to the wide arena of complex systems, such as disease spreading[12], social aggregation [13] and financial transactions [14]. A very interesting example of such studies is provided byepidemic spreading processes [15]. In the presence of SR infection, when no cooperative effects are considered andthere is immunization or death after the infection, the epidemic process gives rise to ordinary percolation clusters [16].The generalization to cases where the dynamical process takes place in lattices where at least some of the infectionsare LR was considered [17–19]. Further interest in the study of LR percolation was also triggered by papers givingsome more exact results and its realization on finite graphs [20, 21].The consideration of LR systems presents in general many interesting features, with theoretical and numericalchallenges. Indeed in a renormalization group (RG) setting the SR Wilson-Fisher fixed point can be altered bysufficiently strong LR interactions, leading at criticality to different universality classes. In particular the effect ofLR interactions on a system in d dimensions has been related (even though the mapping should not be exact) to asystem living in an effective dimension d eff ≥ d .The existence of such effective dimension relation between LR and SR systems comes from the low energy behaviorof the critical propagator for LR interactions. Indeed using the conventional definition that the two point correlation ∗ Corresponding author: [email protected] a r X i v : . [ c ond - m a t . s t a t - m ec h ] O c t functions at criticality scales as G ( r ) ∼ r d − η , for a LR interacting system the anomalous dimension correction canbe rewritten as η = 2 − σ + δη , where 2 − σ is the result obtained in the mean-field approximation and δη is thecorrection due to correlation effects.The expected anomalous dimension correction δη was at the center of an long-lasting, intense debate. Classicalearly-day RG results on the Ising model conjectured δη = 0 to be valid at all orders [22]. Successive investigationscomplemented this result introducing the threshold value σ ∗ = 2 − η SR , where η SR is the anomalous dimension ofa SR model in the same dimension of the LR one under study [23]. At σ = σ ∗ LR interactions become irrelevantwith respect to SR ones [23]. These results, in agreement with Monte Carlo findings [24], have been questioned bynumerical simulations on the two-dimensional LR Ising model indicating δη (cid:54) = 0 and σ ∗ = 2 [25, 26]. We also observethat Monte Carlo simulations support the presence of SR behaviour into a finite amount of the region σ < D percolation [27], in agreement with the Sak’s result [23]. However recently presented results for susceptible-infected-removed epidemic processes with LR infection on a two-dimensional lattice appear to be in contraddictionwith the Sak estimate for σ ∗ , pointing out to the possibility that at least some of the critical exponents are differentfor all σ < δη = 0and σ ∗ = 2 − η SR [28], also indicating an extremely slow convergence as a function of the system size for the criticalamplitude of the standard Binder ratio of magnetization. Logarithmic corrections at the boundary value σ ∗ are apossible source of error in numerical approaches. The occurrence of such logarithmic corrections at the threshold value σ ∗ were confirmed by analytical results, which also indicated the results δη = 0 and σ ∗ = 2 − η SR [29, 30]. Numericalresults for the two-dimensional LR percolation were presented in [29]. The conformal invariance at criticality ofthe two-dimensional LR Ising model [31] and in quantum LR chains [32, 33] has been as well recently addressed.Moreover using both scaling arguments [34] and functional RG techniques [30] the results for the critical exponentsof a LR interacting model in dimension d with decay exponent σ were related to the ones of a SR model in dimension d eff = (2 − η SR ) d/σ . Such effective dimension relation was discussed in [34], finding very good agreement withnumerical Monte Carlo results. One can also see that the effective dimension is valid only at low approximation levelin the RG treatment, even though the error introduced by considering it exact has been quantified to be of < ∼
1% forthe critical exponent ν for the two-dimensional LR Ising models [30]. We observe that in one-dimensional LR Ising andpercolation models Sak’s results implies σ ∗ = 1 (since η SR = 1). A Berezinskii-Kosterlitz-Thouless (BKT) transitionmay be expected at σ = 1 for both one-dimensional Ising and percolation models, which has been deeeply investigatedin Ising chains [35] and also observed for susceptible-infected-removed epidemic processes with LR infection in onedimension [36].Despite the wide appeal of the LR percolation model some important information is still missing. Indeed, for theequilibrium properties, no estimate of the critical threshold and of critical exponents for different values of the decayparameter σ is available to date, at the best of our knowledge, for the LR one-dimensional percolation. The goal ofthe present paper is to give numerical estimates of these quantities. To this aim we develop a Monte Carlo algorithmallowig to efficiently simulate one-dimensional LR bond percolation. As a side result we can numerically check thevalidity of the Sak’s prediction [23] for the value of σ ∗ in the one-dimensional LR percolation.The paper is organized as follows: in Section II we present the model and briefly remind previous results relevant forour work. In Section III we present a novel order- N algorithm for the study of LR percolation models and discuss itsapplication to a LR version of Swendsen-Wang algorithm. In Sections V and VI we respectively present our findingsfor the critical thresholds and the critical exponents. In Section VII conclusions are drawn. II. THE MODEL
We will consider a bond percolation model on a one-dimensional lattice having N sites with occupation probability p i,j of the bond connecting sites i and j given by (with i (cid:54) = j ): p i,j = p | i − j | = C | i − j | σ (1)where C ≤
1. See Figure 1 for a graphical depiction of the model (1). When periodic boundary conditions are used,an appropriate periodic distance will be chosen, defined by the minimum distance among periodic images of the sites i and j .It is clear that if σ → ∞ then the SR bond percolation with a bond probability C between nearest neighbour sitesis obtained. Notice that in the SR limit it is C c = 1.It is known that for the bond probability (1) one has C c = 1 for σ ≥ p | i − j | decays for large FIG. 1: Pictorial representation of the LR bond percolation model. The lines represent the possible bonds with width decreasingas the connection probability decreases (left panel). An instance of the model toghether with the clusters generated (right panel). | i − j | as ∝ | i − j | σ , then one has C c = 1 for σ >
1, but it is not necessarily C c = 1 for σ = 1: in other words, theshort distance behaviour of the probability (1) may give rise to a percolation transition also at σ = 1 [10].For future convenience we also mention the exact result by Schulman [37]: comparing the LR percolation model toa related percolation model on the Bethe lattice a rigorous lower bound for the critical threshold C c was obtained C c ≥ C Bethe c = 12 ζ (1 + σ ) , (2)where ζ ( x ) = (cid:80) ∞ n =1 1 n x is the Riemann zeta function. III. THE ALGORITHM
The numerical study of percolation has largely benefitted from efficient clustering algorithms. Such algorithmsare able to track the cluster to which the local degrees of freedom belong to. Since the inception of the efficientHoshen-Kopelman [38] relabeling scheme these algorithms have been brought closer to the optimal rigorous boundspredicted for union and find algorithm class to which they belong.Simulation of LR models has always presented major challenges on its own. The fact that when we perform amove/update of a system degree of freedom we have to deal with the state of all the other degrees of freedom, unlikeSR systems, typically brings the time for a global update of the system from O ( N ) to O ( N ) severely reducingperformance.Recent advances in this field has been brought by the introduction of algorithms scaling linearly in the size ofthe system [39]. The efficiency of these algorithms relies upon the fact that only a small fraction of local degrees offreedom are actually interacting. Indeed even if every local variable in principle interacts with all the other degreesof freedom it effectively does it only with a O ( N ) fraction of them (in the case of non-frustrated interacting systemsthis condition often translates in the request for an extensive energy).Similarly, in our LR percolation model, the number of connected bonds, among the N available, is of the order O ( N ). More specifically consider the class of bonds of length l : the probability π ( n ) of having n connected bonds outof the N available (with periodic boundary conditions) is a binomial π l ( n ) = (cid:18) Nn (cid:19) p nl (1 − p l ) N − n (3)which average to N p l . If we sum all these probabilities we get a total number of connected links, in the thermodynamiclimit, of N Cζ (1 + σ ) for our model. The above consideration suggests the structure of an efficient O ( N ) algorithmthat is [40]: • extract, for each bond length l class, the connected bonds according to (3). This can be done in a rejection-freefashion by sampling the skip s l among two connected bonds which turn out to be distributed geometrically p ( s l ) ∝ (1 − p l ) s l . The number of random numbers to be extracted is on average 1 + p l • turn on the selected bond and cluster the sites according to the new connections. This can be done in constanttime.Note that the speedup can be traced back to having large classes of bonds with the same probability, which isthe case for translational invariant systems (also with open boundary conditions). The above algorithm, albeit itssimplicity, has not been discussed nor applied previously in the literature. Another remark is in order: the rejectionfree extraction of connected bonds plays a central role, since a more naive sampling (i.e. scanning all the elements)would spoil the overall order- N efficiency.Interestingly this algorithm for LR bond percolation lends itself to effectively implement the Swendsen-Wang algo-rithm [41] in LR Ising or similar models. Indeed in the Swendsen-Wang algorithm [41] the cluster construction canbe dealt with the above scheme. In contrast to what is usually done one should first extract a number of tentativelyconnected bonds of length l via our algorithm with the appropriate probability, that is p l = 1 − e − βJ l where J l is theinteraction among spins l site apart, and later connect them just if they have the same sign. The cluster flipping partwould be not altered. This provides an O ( N ) algorithm for LR Ising models. Although LR adaptations of Swendsen-Wang algorithms scaling as O ( N log N ) [39] and also O ( N ) [42] have already appeared in the literature, we believeour formulation to be very coincise and clear since the mentioned algorithms rely on binary search algorithms andWalker alias method respectively which, albeit well established, are more complicated to implement. Please noticehowever that our algorithm is well suited only for translational invariant systems, while other available algorithms,mentioned before, apply also to non translational invariant systems. IV. OBSERVABLES
Our estimates for the critical thresholds and critical exponents are based on a finite size scaling analysis of twoquantities: where possible the analysis is done with both.The first observable we have monitored consists of the average cluster size S : S = (cid:28) (cid:80) C C N (cid:29) . (4)This quantity is the analogue of the susceptibility of an Ising model in the paramagnetic phase. Indeed it is expectedto scale as S ∝ N γν (5)at the critical point [24]. Since according to scaling relations, γ/ν = 2 − η and η = 2 − σ + δη for a LR system, thescaling of S will be used to evaluate the correction δη to the power law decay of the correlation functions at criticality.In the following it will be shown that our results are in agreement with the widely accepted result δη = 0, which, asalready mentioned, has been recently confirmed by extensive numerical simulations and theoretical investigations onthe LR Ising model [28, 30, 34].We have also introduced and studied the ratio: Q G ≡ (cid:42) (cid:80) C C ( (cid:80) C C ) (cid:43) (6)where C is the size of cluster C , the sum runs over all clusters and the brackets denote average over the differentrealizations of percolation. Q G measures the spread in the cluster size since when they have a characteristic finitesize (in the non percolating phase) it tends to zero as we approach the thermodynamic limit; for C > C C , when amacroscopic (percolating) cluster develops, the ratio Q G tends to one thus behaving as the Binder cumulant [43], thatis widely used in magnetic systems. if we had considered the ratio of cluster size moments (cid:104) (cid:80) C C (cid:105) / (cid:104) (cid:0)(cid:80) C C (cid:1) (cid:105) asin [44] we would have not observed any crossing [40], as insted it occurs for Q G . Moreover since the value of Q G isruled by the correlation length of the system we can put forward the following scaling form in the vicinity of C c , fora finite system of size N : Q G = Q G ; c + a ( C − C c ) N /ν + a ( C − C c ) N /ν + b N ω + b N ω + . . . (7)Our numerical investigations showed that from the crossing of Q G it is possible to have a good estimate of the criticalthresholds C c ( σ ): to substantiate the use of Q G we report in Appendix A the results for Q G for the two-dimensionalSR percolation on a square lattice, showing that the correct result can be retrieved. Regarding the critical exponents,it turns out that using S gives good estimates for the critical exponents (we determined ν and η ) for all the range0 < σ <
1, while Q G can be used for σ close to 1, where by the way it gives more precise estimates than the onesobtained by using S . Both methods, i.e. using S and Q G give results in agreement between each others (where oneobtains reliable findings). Finally, we observe that using Q G one can conclude that C c = 1 for σ = 1 while using S one numerically confirms that C c = 0 for σ = 0, as it should be.The runs have scanned different values of σ , N and C . For each value of the parameters we used 4 · differentrealizations. The considered size of the system is up to N = 512 · . The random number generator chosen is theMT19973 [45]. The bulk of the experiments have been carried out with open boundary conditions, while for selectedvalues of the decay parameter σ we have also simulated periodic boundary conditions. C -1 S N σ C S N σ FIG. 2: (Left panel) SN σ as a function of C for σ = 0 .
25. The lines refer to increasing (small) sizes N = 10 , 2 · , 4 · ,8 · , and 16 · in the sense of the black arrow. Errors are smaller than the size of the lines. In the inset a zoom of thefigure around the crossing point is shown: the curves refer to sizes N = 10 , 2 · , 2 · , 2 · , . . . , 2 · with the curvesbecoming steeper as the system size is enlarged. (Right panel) Same as left panel with σ = 0. As it is apparent, no crossingoccurs. C Q G C Q G FIG. 3: (Left panel) Ratio Q as a function of C for σ = 0 .
5. The lines refer to increasing (small) sizes N = 10 , 2 · , 4 · ,8 · , and 16 · in the sense of the black arrow. Errors are smaller than the size of the lines. The inset shows the curves inthe vicinity of C = 1 for larger system sizes (the same of the inset of Figure 2) as used to locate the transition. (Right panel)Same as left panel with σ = 1. In the inset a zoom around the C = 1 point is reported: one can notice that the curves at thelevel of precision available do not show any crossing. V. CRITICAL THRESHOLDS
To determine the critical thresholds C C as a function of σ we study the quantity SN σ which should cross at C c (this is consistent with the fact that δη = 0 and therefore γ = σν , as it will be shown in the next Section). Indeed asone can see in Figure 2 the curves for different sizes meet at an almost N independent point.Another way to locate the phase transition is to use the cumulant Q G defined in (6). The behavior of the ratio Q G as a function of C is depicted in the left panel of Figure 3 for the value of σ = 0 .
5. Increasing the system size Q G approaches a step profile jumping at the critical value of C allowing for a precise determination of the critical point C c . In the right panel of Figure 3 we show the same plot for σ = 1. As we can see the curves, within numerical error(see inset), do not cross for C < C c = 1 for the threshold value σ = 1.Both methods devised agree for most values of σ but the Q G crossing fails for low σ because the crossing happensin parts of the curve with large curvature (at the sizes considered). For our best estimates of C c we will then rely oncrossings of SN σ . σ C c ν ν ( Q G -est.)0.05 0.0243486(7) 0.0497(2) -0.1 0.047685(8) 0.1006(5) -0.15 0.070482(2) 0.1503(3) -0.2 0.093211(16) 0.2016(7) -0.25 0.11638(3) 0.2474(7) -0.3 0.140546(17) 0.305(8) -1/3 0.15736(6) 0.340(4) -0.35 0.16610(5) 0.344(3) -0.4 0.193471(15) 0.355(9) -0.45 0.22293(6) 0.363(12) - σ C c ν ν ( Q G -est.)0.5 0.25482(5) 0.3503(12) -0.55 0.289410(13) 0.3532(13) -0.6 0.327098(6) 0.3512(15) -0.65 0.368333(13) 0.3495(5) -0.7 0.413752(14) 0.341(2) -0.75 0.464202(19) 0.3293(11) -0.8 0.521001(14) 0.319(2) 0.3161(4)0.85 0.586264(10) 0.299(2) 0.2953(7)0.9 0.66408(7) 0.258(4) 0.262(2)0.95 0.76501(9) 0.200(8) 0.207(7)TABLE I: Estimated critical thresholds C c different values of decay parameter σ . Values of the critical exponent ν are alsoreported. See text for details. σ C c FIG. 4: Critical thresholds C c as a function of σ . The dotted line is the rigorous lower bound (2). The results of the above analysis yield the critical threshold as a function of σ reported in the first column of TableI and for convenience in Figure 4 where we also report the Bethe lower bound 2. We notice that as σ approaches zerothe estimated C c follows C Bethe c closer and closer; this is especially true in the classical region σ < d/
3. On the otherside as σ → C c approaches the expected value of 1. VI. CRITICAL EXPONENTS
For the LR percolation model under exam we expect the existence of a SR effective dimension d eff such that theuniversal quantities of the LR model can be related to the ones of the corresponding SR in d eff dimensions [30, 34, 46].Such effective dimension relation reproduces standard scaling arguments and it reproduces correctly the qualitativefeatures of phase diagrams [30]. Also effective dimensional approach are very accurate close to the mean field regioneven if generally not exact.We have that for σ < d/ d eff > d = 1 for σ > d eff < δη and ν as a function of σ are reported in Figures 5 and 6 respectively. It is clear thatthe Sak prediction is rather clearly confirmed. The values of δη have been obtained by looking for crossing of thecurves SN − γ/ν = SN − σ + δη allowing for a nonzero value of δη . The best value has been chosen as the one giving themost constant value of the crossing value C for the largest sizes N ≥ · considered. Since we found values of δη consistent with zero up to 1 . η = 2 − σ in the subsequent analysis. Our best estimatesof ν (reported in the third column of table I) rely on the behaviour of SN − σ near C c . Standard scaling argumentslead us to expect the following scaling: SN − σ = const. + a ( C − C c ) N /ν + a ( C − C c ) N /ν + a ( C − C c ) N /ν . . . ;the analysis of the data indeed confirms this expectation. Notice that in contrast to the scaling for Q G (see (7)) nosubleading exponents ω i were required to explain the data. The region around the value σ = 1 / C − C c ) to be included in the fit in order to keep χ < . /ν . These data are given in the third column of Table I. Finally an independent estimate of 1 /ν hasbeen calculated via the more involved scaling (7) of Q G (reported in the fourth column of Table I). The values obtainedare reliable only in a small region around 0 . ≤ σ < S .In order to benchmark our numerical results for ν it would be useful to have analytical predictions for it. HoweverRG (and functional RG) calculations in LR percolation problems are less straightforward than the corresponding onesin LR Ising and O(N) models [30]. The derivation of perturbative or non-perturbative approximated formulas goesbeyond the scope of the present work. Nevertheless, it exists a shorthand to obtain approximated reference resultsto control the validity of our findings, relying on the effective dimension relation between LR and SR models. Indeedscaling arguments used to derive effective dimension relation d eff are based on the field theory description and arethus also valid for the percolation problem [47].In a very recent paper [49] ε -expansion formulas on φ theory up to order ε were derived and numerical estimates forthe critical exponents in integer dimensions of SR percolation were given. We then can use the previously mentionedLR-SR effective dimension relation. According to scaling arguments and low order RG approximations the criticalexponents of a LR model in dimension d with exponent σ are related to the ones of a SR model in dimension d eff = (2 − η SR ) dσ . (8)The latter result was successfully applied in the case of two-dimensional LR Ising [30, 34] and LR O(N) models [30]. Inthese works the correlation length exponent ν of the LR model was compared to the one of a SR model in dimension d eff using the relation ν = (2 − η SR ) σ ν SR ( d eff ) . (9)However we expect our case to be rather different. Indeed in the two-dimensional case both the LR and SR systemsare characterized by a spontaneous symmetry breaking of a Z symmetry. On the other hand the one-dimensionalLR Ising model shows spontaneous symmetry breaking only for σ <
1, while for σ = 1 it undergoes a topologicalphase transition of the BKT type [8], where kink excitations having logarithmic interactions [50] play a role similarto the vortices in the standard SR XY model [51].The presence of BKT type phase transition is normally associated with a divergence of the correlation lengthexponent ν and then we expect the correlation length exponent of the d = 1 LR Ising model to go to infinity as afunction of σ in the σ → O(N) models at d = 2 in the N → q → q → σ = 1. Then we expect our data to agree with the ones obtained from the SR resultsvia relation (9) only far from σ = 1 and close to the mean-field threshold σ = . We can then discard anomalousdimension corrections in the SR models and use the simplified mapping ν = 2 σ ν SR ( d eff ) . (10)with d eff = dσ which is the N → ∞ result [30].Therefore perturbative expressions for the critical exponents of LR percolation can be obtained imposing the relationbetween d eff and ε in the second order ε -expansion formulas obtained in [5] and contained in [49] (notice that in [49]it is d eff ≡ − ε ). Defining ˜ ε = σ − /
3, valid in the one-dimensional system, and expanding to second order in ˜ ε weobtain ν − = 13 + 27 ˜ ε − ε + O (˜ ε ) (11)which is plotted in Figure 6 as a gray line.Moreover in [49] results for the critical exponents are presented for integer dimensions using a Pad´e approximantof O ( ε ) expansion results constrained to reproduce the known exact values of the two-dimensional SR percolation.Interpolating such results a curve ν SR ( d ) for the correlation length exponent of the SR model as a function of thedimension d is obtained. This interpolation function is then transformed according to relation (10) with d eff = dσ yielding the results shown in Figure 6 (green line). We observe that if we instead use the relations (8) and (9) weobtain practically the same results for 1 / < ∼ σ < ∼ . σ = 1 for the reasons explainedabove, while as expected poor results are obtained if non-resummed series for ν are used.A reasonable agreement is found with the numerical results even if it does not match with the numerical errors.This issue was expected due to the presence of logarithmic corrections at σ = 1 /
3, which have not been includedin our scaling assumptions and then lead to lack of accuracy close to the mean-field region. It is worth noting thatlogarithmic effects rapidly vanish since for σ > ∼ . σ < ∼ . σ = 1 we can compare with the results of [49] which we generalized to the σ (cid:54) = 1 casefollowing the procedure of [53]. Thus we obtain an analytic result for the correlation length exponent ν close to theBKT point σ = 1: ν − = (cid:114) − σ q (cid:18) q − (cid:113) q + q + ( q − (cid:112) q (1 − σ ) (cid:19) , (12)where the limit q → ν are summarized in Figure 6. σ η σ − σ − η FIG. 5: Estimated values for η toghether with predicted values from the RG prediction η = 2 − σ [23]. Inset: deviation fromthe RG prediction η = 2 − σ . VII. CONCLUSIONS
In this paper we developed, tested and used an order- N algorithm to study one-dimensional LR bond percolation.As a function of the power σ of the decay of the bond probability we determined the critical threshold for percolationand the critical exponents η and ν . A precise estimate of the critical threshold is obtained by introducing a suitablegeometric cumulant Q G : comparison with exact results where available is reported. The value σ ∗ above which nopercolation occurs is retrieved to be 1. For the critical exponent η we compared our findings with the RG result η = 2 − σ by Sak [23], getting agreement at the level of 10 − precision. The exponent ν is compared with mean-field results and ε -expansion for the short-range percolation at an approximate effective dimension. For σ < / ν is found, while the ε -expansion provides a reasonable estimate in the region1 / < ∼ σ < ∼ .
5. Close to σ = 1, where a Berezinskii-Kosterlitz-Thouless (BKT) transition is expected, an expansionin 1 − σ has been derived.An interesting future work would be to explore more in detail the region close to σ = 1 and in particular to considerdifferent short-distance behaviour of the bond probability with non trivial percolation thresholds C c at σ = 1. Theextension of the results presented in this paper to two-dimensional LR percolation may provide a worthwhile subjectof future investigations: we expect that the critical exponents only depend on the ratio d/σ . Finally, we mentionthat with the algorithm discussed in this paper one can study the occurrence (or absence) of conformal invariance atcriticality in two-dimensional LR percolation and the corresponding problem in one dimension. σ / ν Monte Carloclassical region ε -expansion (2nd order) ε -expansion (4th order) (1 − σ ) -expansion FIG. 6: Estimated values for 1 /ν (circles) toghether with predicted values from RG in the classical region ( σ < /
3, bluecontinuous line) and the results for the nonclassical region (1 / < σ < ε -expansion results (11) taken from [49]. Acknowledgement:
We gratefully acknowledge discussions with Miguel Ib´anez Berganza, Luca Lepori, Stefano Ruffo,Slava Rychkov and Hirohiko Shimada. Hospitality in the Program ”Conformal Field Theories and RenormalizationGroup Flows in Dimensions d >
2” at the Galileo Galilei Institute for Theoretical Physics, Florence (Italy), where partof this work was performed, is gratefully acknowledged. GG and AT acknowledge funding support from ”ProgettoPremiale Anno 2012 ABNANOTECH - Atom-based technology”.
Appendix A: The ratio Q G for short-range two-dimensional percolation In order to assess the validity of the ratio Q G introduced in Section IV for characterizing a percolation transition weexamine its behavior for a well known percolation problem. Bond percolation on a square lattice is known to occur,due to symmetry arguments, at an occupation probability of p = p c = 1 /
2. In the left panel of figure 7 we plot Q G for different system sizes. As we can see the intersections approach the exact value. If we simply take the best valuefor our estimate of p c as the intersection of the two largest systems considered (i.e. 128 ×
128 and 256 × p c = 0 . SN − γ/ν instead, shown in the right panel of figure 7, withthe known value of γ/ν = 43 /
24 we obtain the slightly less precise estimate p c = 0 . [1] S. R. Broadbent and J. M Hammersley, Proc. Cambr. Phil. Soc. , 629 (1957).[2] N. A. M. Ara´ujo, P. Grassberger, B. Kahng, K. J. Schrenk, and R. M. Ziff, Eur. Phys. J. Spec. Top. , 2307 (2014).[3] G. R. Grimmett, Percolation (Berlin, Springer-Verlag, 1999).[4] B. Bollob´as and O. Riordan,
Percolation (Cambridge, Cambridge University Press, 2006).[5] J. W. Essam, Rep. Prog. Phys. , 833 (1980).[6] G. Mussardo, Statistical field theory: an introduction to exactly solved models in statistical physics (Oxford, Oxford Uni-versity Press, 2010).[7] H. Nishimori and G. Ortiz,
Elements of phase transitions and critical phenomena (Oxford, Oxford University Press, 2011).[8] D. J. Thouless, Phys. Rev. , 732 (1969).[9] F. J. Dyson, Commun.Math. Phys. , 269 (1971).[10] M. Aizenman, C.M. Newman, Commun. Math. Phys. , 611 (1986).[11] A. Campa, T. Dauxois, D. Fanelli, and S. Ruffo, Physics of long-range interacting systems , (Oxford, Oxford UniversityPress, 2014)[12] P. Grassberger, Math. Biosci. , 157 (1983).[13] S. Solomon, G. Weisbuch, L. de Arcangelis, N. Jan, and D. Stauffer, Physica A, , 239 (2000).[14] D. Stauffer, Advs. Complex Syst. , 19 (2001).[15] M. E. J. Newman, Phys. Rev. E , 016128 (2002). p Q G p -5 -4 -3 -2 -1 S N γ / ν FIG. 7: (Left panel) Ratio Q G as a function of C for σ = 1 / N = 32, 64, 128, and 256 in the sense of the black arrow. Errors are smaller than the size of the lines. (Rightpanel) SN γ/ν as a function of p for the same sizes in the left panel.[16] D. Mollison, J. R. Stat. Soc. B , 283 (1977).[17] M. Biskup, Ann. Prob. , 2938 (2004).[18] T. Emmerich, A. Bunde, S. Havlin, L. Guanlian, and L. Daqing, arXiv:1206.5710 [19] P. Grassberger, J. Stat. Phys. , 289 (2013).[20] I. Benjamini and N. Berger, Random Struct. Alg. , 102 (2001).[21] D. Coppersmith, D. Gamarnik, and M. Sviridenko, The Diameter of a Long-Range Percolation Graph , in Proceedingsof the 13th Annual ACM-SIAM Symposium on Discrete Algorithms (Philadelphia, Society for Industrial and AppliedMathematics, 2002).[22] M. E. Fisher, S.-K. Ma, and B. G. Nickel, Phys. Rev. Lett. , 917 (1972).[23] J. Sak, Phys. Rev. B , 281 (1973).[24] E. Luijten and H. W. J. Bl¨ote, Phys. Rev. Lett. , 2 (2002).[25] M. Picco, arXiv:1207.1018 [26] T. Blanchard, M. Picco, M. A. Rajabpour, Europhys. Lett. , 56003 (2013).[27] F. Linder, J. Tran-Gia, S. R. Dahmen, and H. Hinrichsen, J. Phys. A , 185005 (2008).[28] T. Horita, H. Suwa, and S. Todo, arXiv:1605.09496 [29] E. Brezin, G. Parisi, and F. Ricci-Tersenghi, J. Stat. Phys. , 855 (2014).[30] N. Defenu, A. Trombettoni, and A. Codello, Phys. Rev. E , 052113 (2015).[31] M. F. Paulos, S. Rychkov, B. C. van Rees, and B. Zan, Nucl. Phys. B , 246 (2016).[32] L. Lepori, D. Vodola, G. Pupillo, G. Gori, and A. Trombettoni, Ann. Phys. , 35 (2016).[33] L. Lepori, A. Trombettoni, and D. Vodola, arXiv:1607.05358 [34] M. C. Angelini, G. Parisi, and F.Ricci-Tersenghi, Phys. Rev. E , 062120 (2014).[35] E. Luijten and H. Meßingfeld, Phys. Rev. Lett. , 5305 (2001).[36] P. Grassberger, J. Stat. Mech P04004 (2013).[37] L. S. Schulman, J. Phys. A , L639 (1983).[38] J. Hoshen and R. Kopelman, Phys. Rev. B , 3438 (1976).[39] E. Luijten and H. W. J. Bl¨ote, Int. J. Mod. Phys. C , 359 (1995).[40] M. Michelangeli, Graduation Thesis, Politecnico di Torino & Universit´e Paris Diderot (Master in Complex Systems, 2015).[41] R. H. Swendsen and J.-S. Wang, Phys. Rev. Lett. , 86 (1987).[42] K. Fukui and S. Todo, J. Computat. Phys. , 2629 (2009).[43] K. Binder, Z. Phys B - Condensed Matter , 119 (1981).[44] Y. Deng, T. M. Garoni, and A. D. Sokal, Phys. Rev. Lett. , 030602 (2007).[45] M. Matsumoto and T. Nishimura, ACM Trans. Mod. Comp. Sim. , 3 (1998).[46] M. Ib´a˜nez Berganza and L. Leuzzi, Phys. Rev. B , 144104 (2013).[47] D. J. Amit, J. Phys. A , 1441 (1976).[48] Z. Q. Zhang, F. C. Pu, and B. Z. Li, J. Phys. A , L85 (1983).[49] J. A. Gracey, Phys. Rev. D , 025012 (2015).[50] J. L. Cardy, J. Phys. A , 1407 (1981).[51] J. M. Kosterlitz and D. J. Thouless, J. Phys. C , 1181 (1973).[52] A. Codello, N. Defenu, and G. D’Odorico, Phys. Rev. D , 105003 (2015).[53] J. M. Kosterlitz, Phys. Rev. Lett.37