GEOPHYSICAL RESEARCH LETTERS, VOL. 26, NO. 16, PAGES 2617-2620, AUGUST 15, 1999 Robustness of collective behaviour in strongly driven avalanche models' magnetospheric implications N. W. Watkins•'2, S.C. Chapman 2, R. 0. Dendy3'2,G. Rowlands 2 Abstract. The hypothesisof self-organisedcritical- 1987;Lu, 1995]. Observationalmotivationis provided ity (SOC) predictsthat certainopen dissipativesys- by power law features of magnetosphericindex data, tems evolve to a critical state where all energy release notably AE which is an indicator of energy dissipated statistics display power law distributions for event oc- by the magnetosphere into the ionosphere.A significant currence, size and duration. This has motivated "sand- result [Chapmanet al., 1998]is that a classof sandpile pile" simulationsof magnetospheric energyconfinement andreleaseevents("avalanches"), previousexamples of which have taken the limit whereenergyinflow ("fuelling") is slow relative to dissipation,and either uniform or random. However the magnetosphericsystem has both slow and fast periods mixed together in observations,and naturally modulated fuelling. We have models [Dendyand Helander,1998],seealso [Pinho and Andfade,1998],yieldsystemwide avalanches where the statisticshavea well definedmean(intrinsicscale) whereas the internal avalanchestatistics are scale free; analogousto the differing characteristicsof internal and external magnetosphericenergyreleaseevents. A limitation of standard sandpile models is that their developedan avalanchemodel with variable, modulated fuelling, which would correspondin principle to magnefuelling rate. The power law form for the distribution tosphericloading, is assumedto be slow. This raisesa of energy releaseevents is the least ambiguouscurrent central question. Given that in reality magnetospheric indicator of SOC; we showthat this is preservedfor the loading can be strong, and fluctuate substantially, to large avalanchesin such a system under both constant what extent doesstandard sandpilephenomenologyperand varying loading and sosuchsystemsare remarkably sist? Does the sandpile(or magnetosphere) act as a efficient at eliminating small scale information about nonlinear filter, with the observedavalanchesretaining their fuelling. magnitude and frequency information contained in the fuelling(or loading)process? Tsurutaniet al. [1990]describeda brokenpowerlaw Introduction AE spectrum; this has been claimed as evidenceof SOC Most investigationsof the magnetosphericsubstorm [Consolini,1997; Uritsky and Pudovkin,1998], but a problem have focusedon single substorm episodesor power law spectrum for AE cannot be uniquely interevent groups. An alternative is to use long-term statis- preted as the output of an SOC system becauseother tics of magnetosphericenergyreleaseto infer constraints types of physical system can also produce scale free on the underlying physical processes. This has usu- spectra. Furthermore AE is a compound index in that ally required the definition of substorm "events" via a it mixes driven and unloading effects, a fact reflected thresholdingprocess. There is, however,growing in- both by its powerspectrum [Freemanet al., 1998]and terest in relating the characteristicsof global energy structurefunction [Takaloand Timonen,1998]. One transport in the solar wind-magnetosphere-ionosphere may alsodefineindividual energyreleaseevents,and exsystemto "sandpilemodels" [Consolini,1997;Dendy amine the statistics of their distribution; the signature and Helander, 1998; Chapman et al., 1998; Uritsky of an SOC system would be a "power law" probabil- and Purlorkin,1998],whichdissipateenergyby means o•f avalanches. This followsthe suggestion [Chang, 1992; Chang,1998a;Chang,1998b]that the magnetosphereis in a self organisedcritical (SOC) state. In ity densityfor this quantity. Consolini[personalcommunication,1998] usedAE data taken over a 10 year period to constructthe probabilitydensityD(s) of a burst measure s; the density obtained was fitted by an cutoffpowerlawD(s) ~ (s-•)e-•//• with sandpile models, the distribution of avalanchemagni- exponentially tude and frequencymay displayscalefree, inversepower c• = 1.23q-0.01 and/• = (1.17q-0.01)x 106,extendlaw statisticsthat are characteristicof SOC [Bak et al., ing the result obtained for the year 1978 by Consolini, [1997].A genericfeatureof thistypeof statisticalexper2Spaceand Astrophysics,University of Warwick, UK imental evidenceis that long runs of data are required, and in the magnetosphericsystemit is unavoidablethat 3EURATOM/UKAEA FusionAssociation,Culham,UK both the instantaneous XBritish Antarctic Survey, Cambridge, UK value and the recent mean of the loading rate will have strong variation. Another class of evidence is derived from in situ data. Here, long runs of data are not available, but a given region in the magnetospherecan be observedrepeatedly under Copyright1999by the AmericanGeophysical Union. Papernumber1999GL900586. 0094-8276/99/1999GL900586505.00 2617 2618 WATKINS ET AL.' ROBUST COLLECTIVE BEHAVIOUR IN STRONGLY DRIVEN SANDPILES different conditions. Of particular interest as evidence by O•j_l, the next nearestneighhouris multipliedby for avalanche-typeeventsare bursty bulk flows [Angelopoulos et al., 1996]in the geomagnetic tail. c•j_2c•j_•,and soon. Sincec•xc•2...c•k _• c•xc•...c•k_l•_ c•c•...c•_• and similarly for higher powers in c•i, it follows that the relative importance of • g • in depower law statistics still emergefrom sandpilemodels termining the size of the avalanchediminishesas the that have substantially varying loading, they remain scalej of the avalancheincreases.We consideronly the useful as a characteristicsignatureto test for in energy class of sandpile models which have scale free power releaseevent time series. The appearanceof power law law avalanche statistics in the limit of small but nonThis returns us to the central issue of robustness. If (scalefree) statisticsin suchdata wouldthen support zero • g • ("quiet time"). To give suchstatistics the hypothesisthat the underlying physicsremained II•__-•c•i • g • mustdecline sufficiently rapidly with governed by critical gradients leading to local redis- j for small oscillations in g (with • g • small) to be tribution generatingavalancheson all scales. A lim- effectivelydamped by the sandpile. iting caseof the sandpilealgorithm used in this paper [DendyandHelandev,1998]hasrecentlybeenshownto displayself-similarproperties [Helanderet al., 1999]. Sandpile with strong driving In general, the conditions necessaryfor SOC to oc- Let us considerthe effectsof rapid fuellingon a physi- cur constrainthe inflowrate (see [Lu, 1995]for some proposedcriteria); to emphasise this Jensen,[1998] cal systemin whichenergytransport and dissipationare has proposeda category of slowly-driven,interaction- causedonly by avalanches.If the systemdisplaysSOC over some dynamic range, the event statistics within dominated,thresholded(SDIDT) systems.The slowly driven condition is of particular interest in the magne- that range will be of power law form. There are three characteristic timescalesimplicit in any sandpile algotospheric context becausemost SOC simulationshave rithm for such a system' the relaxation time rr over been conducted in the limit where the rate of inflow is "slow" relative to dissipation. In this paper we pose which an avalanchetakes place; the average time rethequestion: howrobustaremagnetospherically rele- quired, followingan avalanche,for instability to recur vant sandpilemodelsto changesin the inflow ? at the first cell (we assumecentralfuellingfor simplicity), ru; and the time for oneiterationof the algorithm, Scale free behaviour under slow, finite loading At. The amount of sand added at cell 1 per timestep is again g, which may be constant, modulatedover time, or drawn from some random distribution; its ensemble averagedvalue is < g >. Followingthe physically In a typical sandpilealgorithm,redistributionof sand at cell j is triggeredif the local gradientzj exceedsa critical gradient zc. Redistributionmay in turn cause the neighbouringcell to becomeunstable, triggering further redistribution and, by iteration, an avalanche. Avalanchescan be classifiedas internal, meaning rearrangementsof sandwithin the pile, or systemwidewhere the entire sandpilereturns to its angleof reposeand excesssand actually leavesthe system. In the model con- sideredin [Chapmanet al., 1998]the valuesfor the critical gradientsat eachcell weredrawnfrom a probability motivatedargumentsof Dendyand Helander[1998] and Chapmanet a/.[1998]we also assumethat the critical gradient Zcrit for any grid cell and time step is drawn from some probability distributionP(zc) with mean < zc >. In the limit whereP(zc) is a delta function, < Zc > becomesthe "standard" uniform sandpile-wide Zcrit. It is clear that ru -•< Zc > / < g > and hence, for instantaneously relaxingsandpilemodels( •-• (( At ) we identifyslowand fast loadingregimes: (3) distribution. The sand added to trigger an avalancheis spread by the avalancheover j cells, at each of which fast ß•-• (( • • At (4) the averageheightof sandhas increased.Let hj,o and respectively. In the latter case, instability is likely to hj,x denotethe heightsat cellj immediatelybeforeand be triggered at eachtimestep. We reiterate, that in the after the avalanche respectively, and < g > the mean literature,the orderingof time scalesgivenby (3) has (ensembleaverage)fuellingrate. The amountof sand often been cited as a condition for SOC rather than one then transferred to the second cell can be written c•l(hl,0- h2,0-]-• g •) (1) of severalregimesin which aspectsof SOC behaviour may be observed. Normalising At to unity, that is one timestep per in unit time, the fast loading condition becomes• g •• Zc•. For a givenloadingrate, there will be an effective o•2{h2,o - h3,o-]--o•1(hl,O - h2,o--{-< # >)) (2) minimum avalanchelength neededto dissipatethe energy associatedwith the sand added in eachAt. In the is transferred to the third cell, where c•2 < 1, and so slow limit this correspondsto lessthan one cell, in the on. For the final transfer to the jth cell, it followsthat fast limit to many. If the mean fuelling rate • g • is inj-1 < g > is multipliedby IIi= xc•i, whereasthe nearest- creasedtowards and beyond • Zc •, the smallest scale neighboutheightdifferencehj-l,o- hj,o is multiplied avalancheswill be increasinglyeliminated. This will be where c•x< i by construction. Similarly, an amount WATKINS ET AL.' ROBUST COLLECTIVE BEHAVIOUR IN STRONGLY DRIVEN SANDPILES 2619 reflectedin the lower bound of any rangeof "powerlaw" event statistics for energy releasein the system. lO-1 Sandpile with variable driving o We now examine how far information concerning the modulation or fluctuation of g is transmitted in avalanches,and whether its consequencesmay become visiblein the eventstatistics("the sandpileas filter"). Insofar as fluctuations in g take the sandpile intermittently into the fast driving regime, the commentsin the precedingsection apply. In any event, there will exist a maximumavalanche scale(numberof cellsn/) above o • 10'2 o 00 e 0 0 • O0 lO-3 ß 0o 0o o oø 10-4 10' i 102 i 103 i 104 i 106 ** 106 ,•E which the effect of fluctuations in g becomesvanishingly small in the avalanchestatistics. We then have Figure 1. Normalisedhistogramsof probabilityof energy releaseof sizeAE for all (o) and systemwideonly two possiblecases. The first is that n! is sufficiently (*) avalanches for a 5000 cell sandpilewith constant large for there to be a detectablesignaturein the inter- fuelling g - 10. nal statistics(disruptionof the powerlaw), but is still muchsmallerthan the systemsize. The secondhasn! tem, with g: 0.01. Here we presentresultsfor a larger of order the system size, in which casethe event statis- grid of 5000 cells,with P(z•) a "top hat" distribution ticsof bothinternalandexternal(or systemwide) events covering[0.5,1.5] (seealso [Chapmanet al., 1999]). reveal a signaturereflectingboth the mean level and the The normalisedhistogramsof probability of energy refluctuation spectrum of g. In the latter casewe expect leaseevents of size AE in the plots shownhere are coma sandpilemodel to yield goodcorrelationbetweenfluc- prised of over 150,000 internal avalanchesand 20,000 tuationsin the inflowand in the external(systemwide) systemwide avalanches. In Figure i we show results timeseriesin the extremely strongly driven limit. for constant g: 10. The systemwide event statistics remain unaffected by the fast loading, illustrating the Results and discussion fact that large avalanchesare "well insulated" against the fuelling process;howeverthe internal avalanchedisAs a concreteexample of the preceedingarguments tributiondisplaysa dropoutin the numberof the smallwe modified the algorithm usedby Dendy and Helander est events. In this constant loading rate case, there is a [1998]and Chapmanet al. [1998]to incorporatevari- sharp cutoff at AE -• 100 and eventsbelow this size are able fuelling. We take a one dimensional,unit spaced of low probability. Above this, the emergent power law grid of N cells.Sandheightis givenby hi, sothe local form of the energy releasedistribution is preservedup gradientbecomeszj - hj - hj+l. The pile is assumed to AE of 105. This behaviourmay be comparedwith stableat the angleof repose.Both hj, zj arethenmea- Figure 2 in which the input fuelling rate has been varsured from this stable state. Each jth cell has a critical ied. The mean < g > is still 10 but each value of g is gradientzcj. In commonwith all otheravalanchemod- drawn from a uniform randomdistribution, in the range els the model of Dendy and Helander[1998]has a fu- [5, 15]. We seethat the effectis to "soften"the peak elling rule, an instability rule and a redistribution rule, at AE of 100, and to leave some avalanches down to which together form a repeating cycle. Fuelling occurs by adding sand at the first cell. We normaliseL and T to the averageloadingrate g so that unit volumesand is addedin unit time, (i.e. after 1/g iterations).The conservative redistributionrules [Dendyand Helander, 1998;Chapmanet al., 1998]aresuchthat "avalanches" i.e. redistribution events,occur on all scalesup to the size of the pile. lO-1 ooOO 0øø .. o o o o I•10-2 o o Each avalanchereleasesenergy(but not necessarily sand)from the sandpile o o o oo lO-3 N O' N o o o o j--1 j-1 OoO lO-4 lO' i i lOt lO3 , i i lO4 lO5 lO6 The extent to which the loading rate g is "fast" or "slow" can be measuredby comparingg to the mean Figure 2. Normalisedhistogramsof probability of valueof zcfrom (3) and(4). Wehavepreviously studied energyreleaseof sizeAE for all (o) andsystemwide only [Chapmanet al., 1998]avalanche statisticsunderslow (*) avalanches for a 5000 cell sandpilewith randomly (g << < zc >) constantloadingrate, for a 500cellsys- fluctuatingfuellingin the range[5,15] and < g >= 10. 2620 WATKINS ET AL.: ROBUST COLLECTIVE BEHAVIOUR IN STRONGLY DRIVEN SANDPILES AE = 10. We interpret this as being due to the range Chang T. S., Sporadic localized reconnectionsand mulriscale intermittent turbulence in the magnetotail, in of possiblefuelling rates, which means that the smallGeospaceMass and Energy Flow: Results[rom the Inest avalanchesare no longer automatically ruled out in ternational Solar Terrestrial Physics Program, ed. J. L. a giveniteration (i.e. for someiterations# is relatively small). Conclusions Horwitz, D. L. Gallagherand W. K. Peterson,Geophysical Monograph104, AmericanGeophysical Union, Washington, D.C., 193-199, 1998b. Chapman S.C., N. W. Watkins, R. O. Dendy, P. Helander and G. Rowlands, A simple avalanche model as an ana- loguefor magnetospheric activity, Geophys.Res. Lett., 25, Overlongperiods,the magnetospheric driver(that is, 2397-2400, 1998. the solarwind) is characterised by both strongvariabil- Chapman S.C., R. O. Dendy, and G. Rowlands,A sandpile model with dual scalingregimesfor laboratory,spaceand ity about the mean, and a large dynamic range of mean astrophysicalplasmas, Phys. Plasmas,in press,1999. energy input. Any statistical analysisof energyrelease Consolini G., Sandpile cellular automata and magnetoevents will therefore require long uninterrupted runs of spheric dynamics, in Proc. vol. 58, "CosmicPhysics in data. In order to use such data to test the hypothesis the Year 2000", S. Aiello, N. Iucci, G. Sironi, A. Treyes that the magnetospheremay be in a state of SOC, it and U. Villante (eds.),SocietaIralianadi Fisica,Bologna, is necessaryto examine how resilient the main observaItaly, 123-126, 1997. tional signaturesof SOC (notablyscaleinvariantpower Dendy R. O. and P. Helander, Sandpiles, silos and tokamak phenomenology:a brief review, Plasma Physicsand law statistics)will be under circumstances of varying Controlled Fusion, 39, 1947-1961, 1997. Dendy, R. O. and P. Helander, On the appearance and nonappearanceof self-organisedcriticality in sandpiles,Phys. Rev. E., 57, 3641-3644, 1998. Freeman, M.P., Rhodes, C. and Watkins, N. W., On the two-component nature of the auroral electrojets, AGU 1998 Fall Meeting, EOS Trans., 79 suppl., F793, 1998. Helander, P., Chapman S.C., R. O. Dendy, G. Rowlands, and N. W. Watkins, Exactly solvable sandpile with fractal avalanching, Phys. Rev. E, in press, 1999. JensenH. J., Self-OrganisedCriticality: Emergent Complex Behaviour in Physical and Biological Systems,Cambridge University Press, 1998. Lu E., Avalanches in continuum dissipative systems, Phys. events. Rev. Lett., 74{,2511-2514, 1995. Pinho S. T. R, and R. F. S. Andrade, An Abelian model for Acknowledgments. The authorswish to acknowledge rainfall, Physica A, 255, 483-495, 1998. valuable discussionswith T. S. Chang, G. Consolini, M.P. Takalo, J., and Timonen, J., Comparison of the dynamics Freeman, P. Helander, H. J. Jensen,A. Klimas, V. M. Uritof the AU and PC indices, Geophys.Res. Lett., 25, 2101sky, and Z. Voros. SCC would like to acknowledgea P PARC 2104, 1998. lecturer fellowship. Tsurutani B., M. Sugiura, T. Iyemori, B. E. Goldstein, W. D. Gonzalez, S.-I. Akasofu and E. J. Smith, The nonlinear response of AE to the IMF B• driver: a spectral break at References 5 hours, Geophys. Res. Lett., 17, 279-282, 1990. fuelling rate. Our model showsthat the crucial power law form of the burst size distribution of large events is surprisinglyrobust under fast loading. Accordingly, we expect inversepower law avalanchedistributionsto be a persistentfeature in long runs of data that include "fast" inflow conditions, if the underlying systemis governed by SOC. The relative stability of this distribution suggeststhat a long run of data includingperiodswith high driven activity will give a powerlaw, in an SOC magnetosphere.Furthermore restrictingdata to active times would give a power law with a tail-off at small AngelopoulosV., F. V. Coroniti, C. F. Kennel, M. G. Kivel- Uritsky V. M., and M. I. Pudovkin, Low frequency1/f-like fluctuations of the AE-index as a possible manifestation son, R. J. Walker, C. T. Russell, R. L. McPherron, E. of self-organisedcriticality in the magnetosphere,Ann. Sanchez, C. I. Meng, W. Baumjohann, G. D. Reeves,R. Geophys., 16, 1580-1588, 1998. D. Belian, N. Sato, E. Friis-Christensen,P. R. Sutcliffe, K. Yumoto and T. Harris, Multi-point analysis of a BBF event on April 11, 1985, J. Geophys.Res., 101, 4967, 1996. Bak P., C. Tang, and K. Weisenfeld, Self-organisedcriti- S.C. Chapman, G. Rowlands, Spaceand Astrophysics, Univ. of Warwick, Coventry, CV4 7AL, UK. (email: sandrac@astro.warwick.ac.uk; G.Rowlands@warwick.ac.uk) 381-384, 1987. R. O. Dendy, EURATOM/UKAEA FusionAssociation, Chang T. S., Low dimensional behaviour and symmetry breaking of stochasticsystemsnear criticality - can these Culham ScienceCentre, Oxon OX14 3DB, UK. (email: richard.dendy@ukaea.org.uk) effectsbe observedin spaceand in the laboratory?, IEEE N. W. Watkins, UASD, BAS, High Cross, Cambridge, Trans. Plasma Sci., 20, 691-694, 1992. CB3 0ET, UK. (email:nww@bas.ac.uk) Chang T. S., Multiscale intermittent turbulencein the magcality: An explanationof 1/f noise, Phys. Rev. Lett., 50, netotail, in Substorms-d,ed. S. Kokubun and Y. Kamide, Terra Scientific Publishing Company/Kluwer Academic Publishers, Tokyo, 431-436, 1998a. (ReceivedJune6, 1999;acceptedJuly 7, 1999.)