a framework for scaling and renormalization in the

4
A framework for scaling and renormalization in the triangular lattice Dominique D´ es´ erable Laboratoire de M´ ecanique Appliqu´ ee, Automatique & G´ eom´ ecanique INSA – 35043 Rennes cedex – France http://www.insa-rennes.fr [email protected] Keywords: cellular automata, triangular lattice, hierarchi- cal Cayley graphs, arrowhead, scaling and renormalization. Abstract We expose the suitability of a hierarchical Cayley network underlying a cellular automaton to the renormalization meth- ods. The network is a hexagonal arrowhead torus generated on the triangular lattice. We show how the recursive topolo- gy and the symmetries of the arrowhead are likely to provide a convenient framework for these methods. 1 Introduction The renormalization methods appeared these last thirty years in the area of statistical physics and dynamical systems [8]. They apply quite well to the study of large scale homoge- neous systems, illustrated by the image of the “chessboard” whose side of length represents the macroscopic scale while the square of the chessboard, whose side is of length and where is the “correlation length”, represents the mesoscopic scale which defines a homogeneous subsys- tem. These systems are thus characterized by a translational invariance by a vector of modulus , namely, they are period- ic. The renormalization methods have especially proved their superiority in the study of critical phenomena where classi- cal methods had shown to be unsuccessful. A critical system is characterized by a scale invariance whose effect is a self- similar behavior and a divergence of the correlation length at the critical point of the control parameter. Usu- ally, the divergence follows a power law (or “scaling law”) of the form where is the critical exponent. The same goes for the correlation length as well as any macroscopic quantity expressed in terms of the de- viation bearing a specific critical exponent. It is up to the renormalization methods to deduce the value of these exponents, and thereby the behavior of the system in a criti- cal situation, from the only property of scale invariance. By relying upon the self-similarity of the system, the way is to split it into nested blocks and sub-blocks... and to yield lo- cal averages in each of them (Kadanoff construction). The long-range correlations remain unchanged after a sequence of transformations only if the system lies in a critical state. The operator defines a “renormalization group” – in the alge- braic meaning – and the sequence converges in that case to- wards a fixed point (Wilson & Kogut[12], Fisher[5], Ma[9], Stanley[11]). We focus here on the strictly topological aspect of the prob- lem of scale change underlying these methods. After a recall in Sect. 2 of the Kadanoff construction defined on an orthog- onal lattice, we present in Sect. 3 and Sect. 4 a new scheme iterated on the triangular lattice. As an example, this process will be applied in Sect. 5 to the determination of a represen- tative elementary volume (or “REV”) in a globally homoge- neous medium. 2 Kadanoff construction The Kadanoff construction [6, 12, 11, 7, 8], defined on an orthogonal lattice of spins for the Ising model, is displayed in Fig. 1 where the length separating two neighboring sites is the minimal scale. The validity of the construction comes from the fact that the interactions are short-range, that al- lows us to consider only the nearest-neighbor interactions. Each spin, associated to a site, has a proper energy ( ) or ( ) depending on the external magnetic field and each pair of neighbors, associated to a bond, has an interaction ener- gy ( ) or ( ) depending on that they point towards the same direction or towards two opposite directions. The calculation of the energy of the global configuration is renor- malized by forming square blocks of sites (Fig. 1a) and yielding a local average according to a majority rule (the block has an arbitrary size but should contain a small number of sites and size is usually chosen in the construction). Each block is then likened to a single spin at the upper scale of length and the lengths are contracted by a factor to maintain the initial density of sites (Fig. 1b). The construc- tion is thus iterated by forming blocks of size , ,..., until the correlation length of the system in critical state is reached. In the simple case of a globally homogeneous system, it is

Upload: others

Post on 15-May-2022

2 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: A framework for scaling and renormalization in the

A framework for scalingandrenormalizationin thetriangularlattice

DominiqueDeserableLaboratoiredeMecaniqueAppliquee,Automatique& Geomecanique

INSA – 35043Rennescedex – Francehttp://[email protected]

Keywords: cellularautomata,triangularlattice,hierarchi-cal Cayley graphs,arrowhead,scalingandrenormalization.

Abstract

We exposethe suitability of a hierarchicalCayley networkunderlyingacellularautomatonto therenormalizationmeth-ods. Thenetwork is a hexagonalarrowheadtorusgeneratedon thetriangularlattice. We show how therecursive topolo-gy andthesymmetriesof thearrowheadarelikely to providea convenientframework for thesemethods.

1 Introduction

Therenormalizationmethodsappearedtheselastthirty yearsin theareaof statisticalphysicsanddynamicalsystems[8].They apply quite well to the study of large scalehomoge-neoussystems, illustratedby the imageof the “chessboard”whoseside of length

�representsthe macroscopicscale

while the squareof the chessboard,whosesideis of length��� �andwhere

�is the “correlation length”, represents

themesoscopicscalewhich definesa homogeneoussubsys-tem.Thesesystemsarethuscharacterizedby a translationalinvarianceby avectorof modulus

�, namely, they areperiod-

ic. Therenormalizationmethodshaveespeciallyprovedtheirsuperiorityin the studyof critical phenomenawhereclassi-calmethodshadshown to beunsuccessful.A critical systemis characterizedby a scaleinvariancewhoseeffect is a self-similar behavior anda divergenceof the correlationlengthat the critical point ������ of the control parameter. Usu-ally, the divergencefollows a power law (or “scaling law”)of the form

�� � ��� ����� � ��� where ����� is the criticalexponent. The samegoesfor the correlationlengthaswellas any macroscopicquantityexpressedin termsof the de-viation ����� � bearinga specificcritical exponent. It is upto therenormalizationmethodsto deducethevalueof theseexponents,andtherebythebehavior of thesystemin a criti-cal situation,from theonly propertyof scaleinvariance.Byrelying uponthe self-similarityof the system,the way is tosplit it into nestedblocksandsub-blocks...andto yield lo-

cal averagesin eachof them(Kadanoff construction).Thelong-rangecorrelationsremainunchangedafter a sequenceof transformationsonly if the systemlies in a critical state.Theoperatordefinesa“renormalizationgroup”– in thealge-braicmeaning– andthesequenceconvergesin thatcaseto-wardsa fixedpoint (Wilson & Kogut[12], Fisher[5], Ma[9],Stanley[11]).

We focushereon thestrictly topological aspectof theprob-lemof scalechangeunderlyingthesemethods.After a recallin Sect.2 of theKadanoff constructiondefinedonanorthog-onal lattice,we presentin Sect.3 andSect.4 a new schemeiteratedon thetriangularlattice.As anexample,thisprocesswill beappliedin Sect.5 to thedeterminationof a represen-tative elementaryvolume(or “REV”) in a globally homoge-neousmedium.

2 Kadanoff construction

The Kadanoff construction[6, 12, 11, 7, 8], definedon anorthogonallattice of spinsfor the Ising model,is displayedin Fig. 1 wherethelength � separatingtwo neighboringsitesis theminimal scale.Thevalidity of theconstructioncomesfrom the fact that the interactionsare short-range,that al-lows us to consideronly the nearest-neighborinteractions.Eachspin, associatedto a site, hasa properenergy ( � ) or( � ) dependingon the externalmagneticfield andeachpairof neighbors,associatedto a bond,hasan interactionener-gy ( �! ) or ( !� ) dependingon that they point towardsthe samedirectionor towardstwo oppositedirections. Thecalculationof theenergy of theglobalconfigurationis renor-malizedby forming squareblocks of "$#�" sites(Fig. 1a)andyieldinga localaverageaccordingto amajority rule(theblockhasanarbitrarysizebut shouldcontainasmallnumberof sitesandsize "%#&" is usuallychosenin theconstruction).Eachblock is thenlikenedto a singlespinat theupperscaleof length "'� andthe lengthsarecontractedby a factor " tomaintainthe initial densityof sites(Fig. 1b). Theconstruc-tion is thusiteratedby forming blocksof size ( , (*) , . . . , (*+until thecorrelationlength

�of thesystemin critical stateis

reached.

In the simplecaseof a globally homogeneoussystem,it is

Page 2: A framework for scaling and renormalization in the

a

a

- a - - b -

Figure 1: Kadanoff constructionon the orthogonallattice:dottedlines standfor inner interactionswithin a block, fulllinesfor interactionsbetweensitesof neighboringblocks(afterA. Lesne[8])

clear that a similar nestedblock constructionwould reachthemesoscopicscaleof a REV asit will beseenin Sect.5.

3 The arrowhead torus

The topologyunderlyingour cellular automatanetwork [4]is an“arrowhead”torus,a hierarchicalCayley graphdefinedfrom groupsandpresentationsin thetriangular(or “hexava-lent”) grid [3]. Another family derived from the triangulargrid andwhich oftenappearsin theliteraturewasinvestigat-ed by Yebraet al. [13, 1, 2]: the grid is also a hexagonaltorusbut composedof circular ringsarrangedarounda cen-tral node. The relevant featureof the arrowheadis that thesymmetriesaremaximizedin theconstruction.

So,let , thedimensionalityof thearrowhead,denoted-/.1032and 4��5( 0 thenumberof sites(or by duality, thenumberofhexagonalcells in the associatedautomaton).The sitesarenumberedin theset6 0 �87:9�2;�<2=">2@?;?@?A2B( 0 ���<C (1)

accordingtoaschemeresultingfromarecursiveconstructionof thefinite toruswithin theinfinite triangularlattice.

The connectionschemein -/. 0 is defined as follows.Let D be any direction in the orderedset

4E2GFIHJ2GFIK1�A2ML@N 2 L;O>P 2 L:O*Q � = (1,2,3) be the set of associatedincre-mentsas shown in Fig. 2 and D be the oppositedirectionof D . Then RTSVU 9*�I� RTSVU 9*�I�W9 (2)

and X3,$YZ� R[S 0 (<\3�]�^(*\`_ L R (3)R[S 0 (<\3�]��( RTS 0badc \3�e�f_ L R (4)

where g S'h ji � denotestheneighborin direction k of asitei

in -/. h .

16

5

4

1

0

21

20

17

19

30

7

42

3

14

23

58

29

28

41

40

13

12

57

56

31

34

43

54

15

50

59

53

52

49

48

55

18

27

6

63

2

11

22

47

25

24

61

60

9

8

45

44

26

51

62

39

10

35

46

37

36

33

32

38

4

1

0

5

7

10

3

14

13

12

6

15

2

11

9

8

Figure2: Scalechangeby contractionof network -/.`0 in-to the network -1.10badc (for ,Z�ml ). The centersof blocksn 0bo c \p� aredisplayedin gray

As anillustration,knowing thewholeconfigurationof -1. ) ,let usexaminewhich the F , 4EK and 4$H neighborsof site(*9 arein -/.`q :O S q (*9*�r� ( O S ) �;9 �e�d_ L@N �5(ts;lu_W�v�w�:lNxQ S q (*9*�r� ( NyQ S ) �:9*�B�d_ L:ObP �^(ts;zv_�"T�Wl'(N{P S q (*9*�r� ( NuP S ) �:9*�e�|_ L O*Q �^(ts<�}_~lt���4 Construction on the triangular lattice

Starting from Kadanoff construction, Niemeijer & vanLeeuwengive an iteratedconstructionof spinblockson thetriangularlattice [10, 14]. Referingto the topologyof thearrowhead,we proposea new constructioninducedby thenumberingschemeof thecellsandwhich takesadvantageofthesymmetriesof thetorus.We definethesubsets6 0bo � �w7;\�� 6 0�� \��59 ����� ( � �AC 9��$���!,�� (5)

of thesitesof6 0 whoseindex is amultipleof ( � . For afixed� , a subset

6 0>o � definesa subdivision of thelatticeinto ( 0ba��blocksof size ( � denotedn 0bo � \3�]� 6 0bo �u_ 6 �1��7;\�_�� � \�� 6 0bo �/2|��� 6 �bC (6)

andwhosesites \ having their index in6 0bo � arecenters.For���w� in particular, weobtainacontractionof network -/.10

into a network -/.`0ba�c accordingto Fig. 2. Eachblock of( sitesnumberedin the set 7;(*\�2e(*\�_8�<2e(*\�_5">2e(*\�_5l>C isthuscontractedinto a singlesite \ where \$� 6 0ba�c'? Iterated� times, this schemeappliedto a triangularlattice inducesa renormalizationprocessanalogousto thatof theKadanoffconstruction. The advantageis that the symmetriesof thisframework seemto bebetterthanin themodelof Niemeijer& vanLeeuwen.

5 Application to the determination of a REVin a globally homogeneous medium

Let us considerthe trivial example of a scenerepresent-ing a compositematerialmadeup of two solid phases� c

Page 3: A framework for scaling and renormalization in the

Figure3: Determinationof a REV in a homogeneousmate-rial at the global scalewith concentrations�3cE���9 � and� ) �5l*9 � (phase��c appearsin light gray).Thepatterncon-tains (*9*l<9�� cells. Thehexagonalpolyhexe �[��o � 9 � contains"*�'� cellsanddefinesarepresentativevolume

and � ) , mixed accordingto concentrations�3c and � ) with�3c{_W� ) ���*? By applyingthe above constructionscheme,we planto determinatea mesoscopicscalereferredto a rep-resentative (elementary)volume. At this scale,the materialmaybeconsideredashomogeneous.

Let �;� betheoccurencefrequency of phase��c in a block ofsize ( � ( 9��!�J�Z, ). For a givensite,onehasin particular� U ��� if ��c is presentin this siteand � U �59 otherwise.Fora block of size ( �@ |c , the frequency �:�� fc follows from therecurrencerelation

�;�@ |c{�^¡ �;�<�]� �(q¢ £¤ U �|¥

£§¦� (7)

wherethe � ¥£¨¦� denotethefrequencieswithin eachof thefour

sub-blocks.Wheneverthemediumisgloballyhomogeneous,thereexistsa

�suchthat ©'� +  |c ��� + ©ª��« where« is apos-

itive realarbitrarily small. It follows from this thatsequence7�;� C is a Cauchysequencewhich convergestowardsthe fi-nite limit �d¬{�5¡ �d¬@�}�5�3cV? (8)

This iteratedconstructionleadsto partitionthenetwork into( 0ba + blocks of size ( + , eachof theseblocks having thepropertyof defining a representative volume of the wholenetwork.

Fig. 3 displaysa pattern1 of the material,generatedby theautomatonandcoveringabout ( 9T9<9<9 cells,with concentra-tions �3c~�­�9b� and � ) �®l*9 � , as well as a hexagonal

1Thenetwork hasdimensionalityT°&±

� U �*? 9*9<9*9<9� c 9�? �<9<9*9<9� ) 9�? �*z �<�'9� q 9�?²�>�;zb�'��� 9�?²�'9<l��:"�'³ 9�? �*´<´ "<"

Table 1: Convergenceof the iteratedsequenceof frequen-cies,attainedafter ( iterationswith accuracy «/�w��block � ��o � 9 � centredin cell 9 anddefininga representativevolumeof "*�'� cells. Table1 detailstheconvergenceof theiteratedsequence,ensuredfor

� ��( with an accuracy of«t���V��?6 Conclusion

Our contribution in this paperwasto tacklea topologicalas-pectof the scalingandrenormalizationprocessin the trian-gularlatticeandto provideaconvenientframework for thesemethods.While theKadanoff constructionmakesuseof thesymmetrieswithin the orthogonalgrid, we show how it isalso possibleto exploit a scalability propertyfor the trian-gular case. The way lies in the hierarchicalstructureof aCayley network: we claim that the symmetriesof the “ar-rowhead” shouldprovide a self-similar scheme,comparedwith theskewedconstructionof Niemeijer& vanLeeuwen.The last sectionis just a topologicalillustration of the self-similarity of our framework.

References

[1] M.S.Chen,K.G. Shin,D.D. Kandlur:Addressing, rout-ing, andbroadcastingin hexagonalmeshmultiproces-sors. IEEE Trans.Comp.39 (1) (1990)10–18

[2] A. Davis: Mayfly– a general-purpose, scalable, paral-lel processingarchitecture. J.LISPandSymbolicCom-putation5 (1992)7-47

[3] D. Deserable:A familyof Cayley graphsonthehexava-lentgrid. DiscreteAppliedMath.93 (1999)169-89

[4] D. Deserable:A versatiletwo-dimensionalcellular au-tomatanetworkfor granular flow (submittedto SIAMJ.AppliedMath.)

[5] M.E.Fisher:Therenormalizationgroupin thetheoryofcritical behavior. Rev. Mod. Phys.46 (4) (1974)597-616

Page 4: A framework for scaling and renormalization in the

[6] L.P. Kadanoff: Scalinglawsfor Ising modelsnear µd� .Physics2 (6) (1966)263-72

[7] S. Kirkpatrick, R.H. Swendsen:Statisticalmechanicsand disordered systems. Comm. ACM 28 (4) (1985)363-73

[8] A. Lesne:Methodesderenormalisation.Eyrolles(1996)

[9] S.Ma: Introductionto therenormalizationgroup.Rev. Mod. Phys.45 (4) (1973)589-614

[10] T. Niemeijer, J.M.J.van Leeuwen: Wilson theory forspinsystemsona triangular lattice. Phys.Rev. Lett. 31(23) (1973)1411-14

[11] H.E. Stanley: Introduction to phasetransitionsandcritical phenomena. Oxford Univ. Press, London(1971)

[12] K.G. Wilson,J.Kogut: Therenormalizationgroupandthe « expansion. PhysicsReports12 (2) (1974)75-200

[13] J.L. Yebra,M.A. Fiol, P. Morillo, I. Alegre: Thediam-eterof undirectedgraphsassociatedto planetessella-tions. Ars Combin.20B(1985)159–71

[14] J.M.Ziman: Modelsof disorder.CambridgeUniv. Press(1979)