1: % Template article for preprint document class `elsart'

2: % SP 2001/01/05

3: 

4: \documentclass{elsart}

5: 

6: % Use the option doublespacing or reviewcopy to obtain double line spacing

7: % \documentclass[doublespacing]{elsart}

8: 

9: % if you use PostScript figures in your article

10: % use the graphics package for simple commands

11: % \usepackage{graphics}

12: % or use the graphicx package for more complicated commands

13: \usepackage{graphicx}

14: % or use the epsfig package if you prefer to use the old commands

15: \usepackage{epsfig}

16: 

17: % The amssymb package provides various useful mathematical symbols

18: \usepackage{amssymb}

19: 

20: \begin{document}

21: 

22: \begin{frontmatter}

23: 

24: 

25: 

26: 

27: \title{Power laws in surface physics: \\

28: The deep, the shallow and the useful}

29: 

30: \author{Joachim Krug}

31: 

32: \address{Institut f\"ur Theoretische Physik, Universit\"at zu 

33: K\"oln \\ Z\"ulpicher Strasse 77, 50937 K\"oln, Germany}

34: 

35: \begin{abstract}

36: The growth and dynamics of solid surfaces displays a multitude

37: of power law relationships, which are often associated with geometric

38: self-similarity. In many cases the mechanisms behind

39: these power laws are comparatively trivial, and require little more

40: than dimensional analysis for their derivation. The information of 

41: interest to surface physicists then resides in the prefactors. This

42: point will be illustrated by recent experimental and theoretical work

43: on the growth-induced roughening of thin films and step fluctuations

44: on vicinal surfaces. The conventional distinction between trivial and nontrivial power

45: laws will be critically examined in general, and specifically in the context

46: of persistence of step fluctuations.  

47: 

48: \end{abstract}

49: 

50: \begin{keyword} Power laws \sep scale invariance \sep

51: self-affine scaling \sep

52: kinetic roughening \sep thin film growth \sep step fluctuations \sep 

53: persistence

54: 

55: \PACS 89.75.Da \sep 05.40.-a \sep 68.35.Ct \sep 81.15.Aa

56: 

57: \end{keyword}

58: \end{frontmatter}

59: 

60: \small 

61: \emph{Empirical science is apt to cloud the sight, and, by the very knowledge

62: of functions and processes, to bereave the student of the manly contemplation

63: of the whole. The savant becomes unpoetic.} \hfill Ralph Waldo Emerson

64: 

65: \normalsize

66: 

67: Per Bak's theory of self-organized criticality is based on the

68: observation that power laws are ubiquitous in nature, and the understanding

69: that this observation

70: \emph{requires a (general) scientific explanation}.

71: The latter point is often taken for granted in statistical physics,

72: but it is less self-evident in other disciplines. In fact, it could be

73: argued that the predelection of statistical physicists for power

74: law relationships is a professional deformation, which originates

75: in the success story of the theory of equilibrium critical phenomena.

76: In that context, power laws are indeed anomalies which require the

77: ingenious machinery of the renormalization group for their explanation.

78: 

79: However, not every power law carries a deep message. Statistical physicists

80: account for this by distinguishing between \emph{trivial} and

81: \emph{nontrivial} power laws. Though deeply rooted in our jargon, this

82: distinction hard to make precise. Most people would agree that the

83: relation 

84: \begin{equation}

85: \label{RW}

86: \langle \Delta x^2 \rangle \sim t

87: \end{equation}

88: for the mean square

89: displacement of a random walker (really a manifestation of the central

90: limit theorem) is a trivial power law, whereas, say, the Onsager exponents

91: for the two-dimensional Ising model are nontrivial. But consider

92: Kolmogorov's 1941 theory of fully developed turbulence \cite{Frisch}.

93: The derivation of the $k^{-5/3}$-energy spectrum is trivial, in the

94: sense that it uses only dimensional analysis, but it requires

95: the higly nontrivial physical insight that energy dissipation is

96: constant across scales. Clearly the demarcation line between the 

97: trivial and the nontrivial also evolves with the progress of science.

98: 

99: Since the discovery of the fractal nature

100: of diffusion-limited aggregation (DLA) more than two decades ago 

101: \cite{Witten81}, scaling concepts have become one of the

102: main tools in the study of growth processes in condensed

103: matter \cite{Barabasi95,Krug97b,Meakin98,Michely03}. While much of the

104: theoretical activity has been driven by the 

105: (still unfinished) quest of understanding the two most prominent

106: nontrivial power laws in the field --  the fractal dimension of

107: DLA, and the strong coupling exponents of the Kardar-Parisi-Zhang (KPZ)

108: equation \cite{Kardar86} in dimensions larger than one 

109: \cite{Krug97b,Krug91,Halpin95} -- experimentalists have begun

110: to routinely employ the concepts of scaling and self-similarity

111: to analyse topographic data in thin film and crystal growth.    

112: In the following I will describe some recent applications of scaling

113: ideas in surface physics. I will argue that, more often than not,

114: power laws that, by the standards of statistical physicists,

115: are quite trivial, have been of most use in interpreting experimental 

116: data and gaining insight into kinetic processes at real surfaces.

117: The discussion will include 

118: the currently popular concept of persistence of a stochastic process 

119: \cite{Majumdar}, which provides an interesting perspective on the

120: distinction between trivial and nontrivial power laws.

121: 

122: \begin{figure}

123: \centerline{\includegraphics[width=0.8\textwidth]{krug1.eps}}

124: \caption{Experimental data for the surface roughenss $\sigma$

125: as a function of DIP film thickness. Full data points

126: were obtained from atomic force microscopy, and open data

127: points from x-ray diffraction. Inset shows a typical

128: fit of the measured scattering intensity to the form

129: expected for a rough surface (from \cite{Duerr03})}

130: \label{Duerr}

131: \end{figure}

132: 

133: Our first example is the growth-induced roughening of thin films

134: of an organic semiconductor, diindenoperylene (DIP) \cite{Duerr03}.

135: The relevant scaling law is the relationship between the 

136: root mean square surface roughness $\sigma$ (the standard deviation 

137: of the film height distribution) and the film thickness $D$,

138: \begin{equation}

139: \label{sigma}

140: \sigma(D) \sim D^\beta,

141: \end{equation}

142: which defines the roughening exponent $\beta$. The experimental data for

143: $\sigma(D)$ shown in Fig.\ref{Duerr} are remarkable in several respects.

144: First, they cover more than two decades in film thickness, and they

145: contain consistent results obtained using two completely different methods,

146: x-ray diffraction and atomic force microscopy. The second remarkable

147: feature is the experimentally determined value $\beta = 0.75 \pm 0.05$

148: of the roughening exponent. This value \emph{exceeds} the random deposition

149: (RD) limit $\beta_{\mathrm{RD}} = 1/2$, which applies in the absence of

150: any mass transport between different layers of the growing film 

151: \cite{Barabasi95,Michely03}. In this limit the height follows 

152: a Poisson distribution, and (\ref{sigma}) becomes equivalent to 

153: the random walk relation (\ref{RW}). Random deposition can be approximately

154: realized in the growth of crystalline metal films,

155: where transport between atomic layers is inhibited by step 

156: edge barriers (see below).   

157: Since mass transport processes along the growing surface are driven

158: by differences in bonding energy, they will generally tend to smoothen

159: the film. The random deposition roughness 

160: $\sigma_{\mathrm{RD}}$ would therefore be expected

161: to define an upper bound on the roughness that random fluctuations

162: can induce during the growth of a thin film; in particular, 

163: $\beta \leq \beta_{\mathrm{RD}} = 1/2$. In this sense systems

164: with $\beta > 1/2$ are anomalous, and constitute examples 

165: for the (largely unexplained) phenomenon of \emph{rapid roughening}

166: \cite{Krug97b,Duerr03}.  

167: 

168: But do the DIP data in 

169: Fig.\ref{Duerr} really violate the RD bound? To answer this question,

170: it is essential to include also the prefactor of the power law

171: relationship (\ref{sigma}) in the analysis. Denoting by $d$ the

172: thickness of a single atomic layer, the random deposition 

173: roughness is $\sigma_{\mathrm{RD}} = d \sqrt{D/d}$, which is depicted

174: as a dotted line in Fig.\ref{Duerr}. The comparison shows that the

175: experimental roughness data remain \emph{below} the random deposition

176: limit even for the largest film thicknesses. It is conceivable

177: (though perhaps not very likely) that the roughening exponent crosses

178: over to a value below 1/2 before the critical thickness $D_\zeta$

179: in Fig.\ref{Duerr} is reached, and that the early time value 

180: $\beta \approx 0.75$ is merely a transient\footnote{Similar transients

181: have been observed in simulations of metal epitaxy \cite{Caspersen02}.}; 

182: unfortunately films with thicknesses $D > D_\zeta$ could not be grown for

183: technical reasons. If, on the other hand,

184: the measured value $\beta \approx 0.75$ is truly asymptotic, then

185: the observed roughness cannot be due to the random fluctuations in the

186: deposition beam. In \cite{Duerr03} it was conjectured that quenched

187: in-plane disorder arising from the boundaries of the tilt domains

188: in the organic thin film could be responsible for the rapid roughening

189: behavior. Such disorder is known to induce sublinear roughening behavior

190: of the form $\sigma \sim D/\ln(D)^\psi$, which can mimic a power law with

191: exponent $1/2 < \beta < 1$ over extended time scales \cite{Krug95}. 

192: Unfortunately, the available 

193: experimental information about the growth process

194: on the molecular level is insufficient at present to validate or refute

195: this hypothesis. 

196:   

197: If our understanding of DIP film growth, as described above,

198: seems unsatisfactory, it is nevertheless fairly representative of the field as a whole.

199: Although the power law relationship (\ref{sigma}) has been, and is currently

200: being reported for a host of growth systems \cite{Barabasi95,Meakin98}, including metallic as well

201: as semiconductor materials and crystalline as well as amorphous films, and 

202: although the measured values of $\beta$ tend to cluster around numbers

203: that can be derived from theoretical models \cite{Krim95}, a clear-cut example, where

204: a well-understood growth mechanism gives rise to a definite prediction for

205: $\beta$ which is quantitatively confirmed by experiment, is so far missing.

206: In particular, 2+1--dimensional KPZ scaling, a mathematical object under passionate

207: theoretical pursuit for close to twenty years, is still to emerge in any real growth

208: system\footnote{In 1+1 dimensions KPZ-scaling with the (trivial or nontrivial?) exponent $\beta = 1/3$

209: has been demonstrated for one-dimensional slow combustion fronts in paper \cite{Maunuksela97}.}.

210: 

211: 

212: Perhaps the only exception to this statement

213: is mound formation in homoepitaxial crystal growth \cite{Michely03},

214: where the exponent $\beta$ describes the roughening of a morphology that, rather than

215: being scale-invariant, displays a distinct lateral length scale.

216: The growth of ``wedding cakes'' under conditions of 

217: strongly inhibited interlayer transport is particularly simple

218: \cite{Krug97,Kalff99}. In this limit the roughening exponent takes on the random deposition

219: value $\beta = \beta_{\mathrm{RD}} = 1/2$, while deviations of the height distribution

220: from the ideal Poisson form show up in the prefactor. 

221: The Poisson distribution is cut off at large heights because a new layer can be nucleated

222: on top of a mound only when the top terrace has reached a critical size

223: \cite{Krug00}. This leads 

224: to the expression       

225: \begin{equation}

226: \label{wed}

227: \sigma/d = \sqrt{(1 - \theta_c) D/d} 

228: \end{equation}

229: for the surface roughness, where $\theta_c$ is the coverage corresponding to 

230: the critical top terrace size \cite{Krug02}. Equation (\ref{wed}) is our first example

231: of a trivial power law where the nontrivial information (the specifics of the 

232: interlayer transport processes that determine $\theta_c$ \cite{Krug00,Krug02}) resides

233: in the prefactor.  

234: 

235: In the remainder of the paper we will be concerned with the roughening

236: of \emph{one-dimensional} objects (lines). Rough lines appear naturally

237: as atomic steps on

238: vicinal crystal surfaces \cite{Jeong99,Giesen01}. A vicinal surface is obtained

239: by cutting the crystal in a direction close to (in the \emph{vicinity} of)

240: a high symmetry plane, and it consists of high symmetry terraces separated

241: by steps of monoatomic height. 

242: In thermal equilibrium, the steps are roughened by thermal fluctuations.

243: We describe the step at time $t$ by a function $y(x,t)$,

244: where the $x$-axis is taken along the mean step direction and the $y$-direction

245: is perpendicular to the step (in the direction of vicinality). Up to the 

246: length scale where collisions with neighboring steps become important

247: the static step conformation is the graph of a one-dimensional random walk,

248: $\langle (y(x,t) - y(x',t) )^2 \rangle \sim \vert x- x' \vert$.  

249: Time-dependent fluctuations are well described by a

250: linear Langevin equation of the form 

251: \begin{equation}

252: \label{Langevin}

253: \frac{\partial}{\partial t} y(x,t) = - K \left( - \frac{\partial^2}{\partial x^2} 

254: \right)^{z/2} y(x,t) + \eta(x,t)

255: \end{equation}

256: where $K$ is a positive constant, $\eta(x,t)$ is white noise, and the 

257: \emph{dynamic exponent} $z$ depends on the dominant kinetic pathway through

258: which the step fluctuations relax to equilibrium. The most important

259: cases are $z=2$, corresponding to fast mass exchange between the step and

260: the terrace (\emph{nonconserved} kinetics) and $z=4$ corresponding to 

261: mass transport only along the step (\emph{conserved} kinetics). It is straightforward

262: to show from (\ref{Langevin}) that the temporal step correlations scale as

263: \begin{equation}

264: \label{C}

265: C(t) = \langle (y(x,s) - y(x,s+t))^2 \rangle \sim t^{1/z}.

266: \end{equation} 

267: By analogy with (\ref{sigma}) we can say that 

268: the roughening exponent of the step is $\beta = 1/2z$. 

269: 

270: 

271: The scaling law (\ref{C}) has been observed in many experiments, on both metal

272: and semiconductor surfaces \cite{Jeong99,Giesen01}. The focus in these experiments

273: has usually not been on the power laws as such, which (in the sense of 

274: our introductory discussion) are rather trivial manifestations of the simple

275: linear dynamics (\ref{Langevin}). The relation

276: (\ref{C}) is useful mainly as a classification

277: tool, which serves to identify the dominant step relaxation processes.

278: The nontrivial (and materials-specific) information lies instead in the

279: temperature dependence of the prefactor, which gives insight into the

280: energy barriers governing the atomic processes at the step edge.   

281: Examples of such processes are shown in the left panel of Fig.\ref{Jouni}.

282: The right panel shows simulation data for the correlation function 

283: $C(t)$, which illustrate the decrease of the prefactor of the $t^{1/4}$-law

284: as the kink rounding barrier $E_{\mathrm{kr}}$ is increased. As shown by the

285: detailed analysis in \cite{Kallunki03}, the activation energy of the prefactor

286: depends linearly on $E_{\mathrm{kr}}$, provided this quantity is larger

287: than the kink energy (the energy cost for the formation of a kink), and

288: therefore a temperature-dependent measurement of $C(t)$ can be used to 

289: experimentally determine $E_{\mathrm{kr}}$.

290:   

291: 

292: \begin{figure}

293: \includegraphics[width=0.5\textwidth]{krug2a.eps}\hspace*{1.cm}\includegraphics[width=0.35\textwidth]{krug2b.eps}

294: \caption{Left panel: Schematic of atomic processes at a step edge. The 

295: diffusion of single atoms along straight segments of the step is governed

296: by an energy barrier $E_{\mathrm{st}}$. The detachment from a kink requires

297: the additional energy $E_{\mathrm{det}}$, and the rounding of the kink 

298: the energy $E_{\mathrm{kr}}$. The distance between kinks 

299: is $\ell_{\mathrm{k}}$. Right panel: Temporal step correlation function

300: $C(t)$ obtained from Monte Carlo simulations for different values of 

301: $E_{\mathrm{kr}}$. Dashed lines are fits of the form $C(t) = A t^{1/4}$

302: (from \cite{Kallunki03})}  

303: \label{Jouni}

304: \end{figure}

305: 

306: It would be premature to conclude, however, that the kinetics of step fluctuations

307: can be reduced completely to ``trivial'' power laws like (\ref{C}). Although the

308: Langevin equation (\ref{Langevin}) is linear, it is still considerably more

309: complex than a simple random walk, because the step is a spatially extended

310: object. As a consequence, the step position $y(x,t)$ at fixed $x$ is a 

311: \emph{non-Markovian} process in $t$. The non-Markovian character of a stochastic

312: process manifests itself clearly when considering the \emph{persistence

313: probability}, i.e. the probability that the process does not cross a

314: particular value     

315: in a specified time interval. 

316: The computation of this quantity requires the knowledge of temporal correlations

317: of arbitrary order. For a Gaussian process, such as the solution of the

318: Langevin equation (\ref{Langevin}), all higher order correlation functions are,

319: in principle, encoded in the two-point function (\ref{C}), but in practice

320: the calculation of the persistence probability for a general Gaussian process

321: is a hard, unsolved problem \cite{Majumdar}.  

322: 

323: 

324: \begin{figure}\includegraphics[width=0.5\textwidth]{krug3a.eps}

325: \includegraphics[width=0.5\textwidth]{krug3b.eps}

326: \caption{Left panel: Transient persistence exponent $\theta_0$ as a function

327: of the roughening exponent $\beta$ for the linear Langevin equation (\ref{Langevin}).

328: The figure shows numerical estimates obtained from 

329: discrete solid-on-solid models (squares),

330: direct simulation of the Langevin equation (triangles), simulation of an

331: equivalent stationary Gaussian process (diamonds), a perturbation expansion

332: around the Markovian case $\beta = 1/2$ (bold line) and rigorous bounds

333: derived by comparison with related Markovian processes (gray area)

334: (from \cite{Krug97c}). Right panel: Experimentally determined persistence

335: probability for steps on a Si(111)-surface which has undergone a 

336: reconstruction by deposition of a fraction of a monolayer of Al. Symbols

337: show measurements at 970 K (squares), 870 K (circles) and 770 K (triangles).

338: The data sets have been shifted vertically for greater clarity; in fact   

339: the data show no systematic temperature dependence 

340: (from \cite{Dougherty02})}

341: \label{Pers}

342: \end{figure}

343: 

344: To be concrete, we define

345: the step persistence probability 

346: \begin{equation}

347: \label{Persistence}

348: P(t_0, t) = \mathrm{Prob}[y(s) \neq y(t_0) \vert t_0 \leq s \leq t_0 + t],

349: \end{equation}

350: where it is understood that the step is completly straight at time 

351: $t=0$ ($y(x,0) \equiv 0$). Note that here the specified level that the step should not

352: cross is set by its position at the beginning of the time interval. 

353: The analytic and numerical investigations in \cite{Krug97c} have shown

354: that the quantity (\ref{Persistence})

355: displays two distinct scaling regimes. In the \emph{transient}

356: regime $t_0 \ll t$ the persistence probability decays as $P \sim t^{-\theta_0}$,

357: while in the \emph{steady state} regime $t_0 \gg t$ it decays as

358: $P \sim t^{-\theta_S}$. The two persistence exponents $\theta_0$ and 

359: $\theta_S$ are different. Moreover, while the steady state exponent is simply

360: related to the roughening exponent $\beta$ by 

361: \begin{equation}

362: \label{thetas}

363: \theta_S = 1 - \beta,

364: \end{equation} 

365: the 

366: transient exponent is truly ``nontrivial'', in the sense that it can be computed

367: only approximately using a range of fairly sophisticated methods 

368: (see Fig.\ref{Pers}). 

369: 

370: The relationship (\ref{thetas}) provides a good

371: illustration of the fallibility of the distinction between trivial and 

372: nontrivial scaling exponents. On the one hand it would seem that $\theta_S$, being

373: simply related to the arguably trivial exponent $\beta$, should also be

374: classified as trivial. On the other hand the relation (\ref{thetas}), while

375: well supported by simulations and plausible scaling arguments,

376: is not rigorously established; in fact, in the context of the general

377: problem of determining the persistence probability of a Gaussian process

378: with known correlator, there is no reason why 

379: $\theta_S$ should be simpler to compute, and hence less nontrivial,

380: than $\theta_0$. The scaling relation (\ref{thetas}) holds because the 

381: stochastic process $y(x,t)$ at fixed $x$, in the steady state regime, combines

382: two invariance properties: It is scale invariant \emph{as well as} translationally

383: invariant in time\footnote{Together these two properties define a

384: Gaussian process known as \emph{fractional Brownian motion} (fBm) \cite{Mandelbrot68}, 

385: a generalization of the Wiener process, the continuum limit of the simple random walk.

386: For another recent application of fBm to a persistence problem

387: see \cite{Majumdar03}.}\cite{Krug98}. 

388: Because it is based on these general principles,

389: (\ref{thetas}) appears to hold also for non-Gaussian interface fluctuations, such as

390: those of the KPZ universality class, where not even the temporal two-point correlations

391: (let alone correlations of higher order) are explicitly known \cite{Kallabis99}.   

392: 

393: Several groups have recently met the challenge of experimentally measuring the 

394: persistence probability of one-dimensional interfaces, for steps on vicinal surfaces 

395: \cite{Dougherty02,Dougherty03,Constantin03} as well as for slow

396: combustion fronts in paper \cite{Merikoski03}. In all cases the steady state persistence

397: exponent $\theta_S$ was measured, and the scaling relation (\ref{thetas}) was confirmed,

398: with $\beta = 1/4$ \cite{Dougherty02}, $\beta = 1/8$ \cite{Dougherty03} and 

399: $\beta = 1/3$ \cite{Merikoski03}, respectively.

400: The transient persistence exponent (which is of greater theoretical interest) has so far

401: not been experimentally accessible, because of the difficulty of preparing the special

402: (flat) initial condition that it refers to. In addition to the persistence probability

403: defined by (\ref{Persistence}), in \cite{Dougherty02} also the probability for the

404: step not to cross the \emph{mean} step position (determined as an average over the

405: measured time series) was considered. This quantity turns out \emph{not} to follow

406: a power law. This is because in the thermodynamic limit the step position starts,

407: with probability one, infinitely far away from the mean (the variance of $y$ diverges

408: with diverging step length); hence the persistence probability remains at unity

409: and decays exponentially only on a much larger time scale set by the finite length

410: of the step \cite{Dasgupta03,Satya04}. 

411: 

412: Experimental step persistence

413: data from \cite{Dougherty02} are displayed

414: in the right panel of Fig.\ref{Pers}. In contrast to the step correlation function in Fig.

415: \ref{Jouni}, the persistence data show no systematic temperature dependence; the

416: prefactor of the observed power law depends only on the sampling time \cite{Williams}.

417: This is because the property of a height fluctuation to return (or not) to its

418: initial value in a prescribed time interval does not depend on the overall 

419: amplitude of the fluctuation; the persistence probability is a functional

420: only of the \emph{shape} of the (suitably normalized)

421: correlation function \cite{Krug97}. As a consequence, persistence measurements

422: cannot be used to extract energy barriers for specific microscopic processes.

423: The benefit of these measurements is of a more fundamental nature --

424: they prove that the theoretical description of step fluctuations through the Langevin

425: equation (\ref{Langevin}), and the underlying picture

426: of ``universality classes'' encoded by the dynamic exponent $z$, 

427: extends beyond the two-point function to correlations of arbitrary order.   

428: As the persistence probability is known to be extremely sensitive to hidden

429: temporal correlations affecting the interface fluctuations \cite{Kallabis99},

430: it may be particularly useful in cases where the assignment of the universality

431: class (the value of $z$) through more conventional measurements is ambiguous.

432: 

433: 

434: Where does this leave us with regard to the general reflections on power laws, the

435: trivial and the nontrivial, the deep and the shallow, which introduced this paper?

436: From discussions in the early 1990's, when SOC was challenged by the competing

437: concept of Generic Scale Invariance 

438: \cite{Grinstein90}, I recall that Per Bak had a distinct,

439: and rather unfavorable opinion of simple power laws generated by simple equations

440: such as (\ref{Langevin}): He referred to them as systems operating by the 

441: \emph{garbage in, garbage out} principle, because they merely perform a (linear, or,

442: more generally, nonlinear) transformation of the driving white noise into correlated

443: fluctuations. Per viewed self-organization as an essential part of SOC. He emphasized

444: the capability of self-organizing systems to develop new, emergent levels of 

445: structure, and to undergo a history which is open to contingent influences.

446: Is a rough surface or a fractal cluster grown by DLA self-organized in this sense? 

447: Probably not\footnote{Nevertheless it is true that scale-invariant structures

448: need sufficient time to develop; more precisely, the time $T$ required to grow a structure

449: of size $R$ typically scales as $T \sim R^z$ with $z > 1$. This is probably

450: the reason why most spatial power laws in Nature extend over less than two

451: orders of magnitude \cite{Avnir98}: Under typical physical conditions, there is

452: insufficient time for scale-invariant correlations to develop further.}. 

453: Still, we must accept that many of the power laws in the world

454: that surrounds us have rather humble origins. As theoretical physicists, our task

455: it to explain these origins; but as ``experimental philosophers'' (Per Bak) we would

456: also like to know what they \emph{mean}. The latter question 

457: is of course not one to be answered, but one that is to be constantly 

458: clarified (and reobscured) in the ongoing discourse of the community. 

459: In these discussions Per will be sorely missed, and 

460: gratefully remembered for a long time to come. 

461: 

462: 

463: 

464: 

465: \section*{Acknowledgements}

466: I wish to thank Satya Majumdar for helpful correspondence,

467: and Arndt D\"urr and Ellen Williams

468: for the permission to reproduce experimental data.

469: The work reported here was conducted at the University of Duisburg-Essen

470: with partial support of DFG within SFB 237.

471:  

472: 

473: \begin{thebibliography}{00}

474: 

475: \bibitem{Frisch} U. Frisch: \emph{Turbulence} (Cambridge University Press, 1995)

476: 

477: \bibitem{Witten81} T.A. Witten, L.M. Sander: Phys. Rev. Lett.

478: \textbf{47}, 1400 (1981)

479: 

480: \bibitem{Barabasi95} A.--L. Barab\`asi, H.E. Stanley:

481: \emph{Fractal Concepts in Surface Growth}, (Cambridge University

482: Press, 1995)

483: 

484: \bibitem{Krug97b} J. Krug: Adv. Phys. \textbf{46}, 139 (1997)

485: 

486: \bibitem{Meakin98} P. Meakin: \emph{Fractals,

487: Scaling and Growth far from

488: Equilibrium} (Cambridge University Press, 1998)

489: 

490: 

491: \bibitem{Michely03} T. Michely and J. Krug,   

492: \emph{Islands, Mounds and Atoms. Patterns and Processes in Crystal

493: Growth Far from Equilibrium} (Springer, Berlin 2003)

494: 

495: \bibitem{Kardar86} M. Kardar, G. Parisi, Y.-Z. Zhang: 

496: Phys. Rev. Lett. \textbf{56}, 889 (1986)

497: 

498: \bibitem{Krug91} J. Krug and H. Spohn, in: \emph{Solids far from Equilibrium},

499: ed. by C. Godr\`eche (Cambridge University Press, Cambridge 1991)

500: pp.~479--582

501: 

502: \bibitem{Halpin95} T. Halpin-Healy, Y.C. Zhang: Phys. Rep.

503: \textbf{254}, 215 (1995)

504: 

505: \bibitem{Majumdar} S.N. Majumdar: Curr. Sci. \textbf{77}, 370 (1999)

506: 

507: \bibitem{Duerr03} A.C. D\"urr, F. Schreiber, K.A. Ritley, V. Kruppa,

508: J. Krug, H. Dosch, B. Struth: Phys. Rev. Lett. \textbf{90}, 016104 (2003)

509: 

510: \bibitem{Caspersen02} K.J.~Caspersen, A.R.~Layson, C.R.~Stoldt,

511: V.~Fournee, P.A.~Thiel, J.W.~Evans: Phys. Rev. B \textbf{65},

512: 193407 (2002)

513: 

514: \bibitem{Krug95} J. Krug: Phys. Rev. Lett. \textbf{75}, 1795 (1995)

515: 

516: \bibitem{Krim95} J. Krim, G. Palasantzas: Int. J. mod. Phys. B

517: \textbf{9}, 599 (1995)

518: 

519: 

520: \bibitem{Maunuksela97} J. Maunuksela, M. Myllys, 

521: O.-P. K\"ahk\"onen, J. Timonen, N. Provatas, 

522: M. J. Alava, T. Ala-Nissila:

523: Phys. Rev. Lett. \textbf{79}, 1515 (1997)

524: 

525: \bibitem{Krug97} J. Krug: J. Stat. Phys. \textbf{87}, 505

526: (1997)

527: 

528: \bibitem{Kalff99} M. Kalff, P. \v{S}milauer, G. Comsa, T.

529: Michely: Surf. Sci. \textbf{426}, L447 (1999)

530: 

531: \bibitem{Krug00} J. Krug, P. Politi, T. Michely: Phys. Rev. B

532: \textbf{61}, 14037 (2000)

533: 

534: \bibitem{Krug02} J. Krug, P. Kuhn, in: \emph{Atomistic

535: Aspects of Epitaxial Growth}, ed. by M. Kotrla, N.I. Papanicolaou,

536: D.D. Vvedensky, L.T. Wille (Kluwer, Dordrecht 2002) pp.~145--163

537: 

538: 

539: \bibitem{Jeong99} H.-C. Jeong, E.D. Williams: Surf. Sci. Rep.

540: \textbf{34}, 171 (1999)

541: 

542: \bibitem{Giesen01} M. Giesen: Prog. Surf. Sci. \textbf{68}, 1

543: (2001)

544: 

545: \bibitem{Kallunki03} J. Kallunki, J. Krug: Surf. Sci. Lett.

546: \textbf{523}, L53 (2003)

547: 

548: \bibitem{Krug97c} J. Krug, H. Kallabis, S.N. Majumdar, S. Cornell,

549: A.J. Bray, C. Sire: Phys. Rev. E \textbf{56}, 2792 (1997)

550: 

551: \bibitem{Mandelbrot68} B.B. Mandelbrot, J.W. van Ness: SIAM Rev. \textbf{10},

552: 422 (1968)

553: 

554: \bibitem{Majumdar03} S.N. Majumdar: Phys. Rev. E \textbf{68}, 050101(R) (2003)

555: 

556: \bibitem{Krug98} J. Krug: Markov Proc. Rel. Fields \textbf{4}, 509 (1998)

557: 

558: \bibitem{Kallabis99} H. Kallabis, J. Krug: Europhys. Lett. \textbf{45}, 20 (1999)

559: 

560: \bibitem{Dougherty02} D.B. Dougherty, I. Lyubinetsky, E.D. Williams,

561: M. Constantin, C. Dasgupta, S. Das Sarma:

562: Phys. Rev. Lett. \textbf{89}, 136102 (2002)

563: 

564: \bibitem{Dougherty03} D.B. Dougherty, O. Bondarchuk, M. Degawa,

565: E.D. Williams:

566: Surf. Sci. Lett. \textbf{527}, L213 (2003)

567: 

568: \bibitem{Constantin03} M. Constantin, S. Das Sarma, C. Dasgupta, O. Bondarchuk,

569: D.B. Dougherty, E.D. Williams: Phys. Rev. Lett. \textbf{91}, 086103 (2003)

570: 

571: \bibitem{Merikoski03} J. Merikoski, J. Maunuksela, M. Myllys, J. Timonen:

572: Phys. Rev. Lett. \textbf{90}, 024501 (2003)

573: 

574: \bibitem{Dasgupta03} C. Dasgupta, M. Constantin, S. Das Sarma, S.N. Majumdar:

575: Preprint ({\tt cond-mat/0307086})

576: 

577: \bibitem{Satya04} S.N. Majumdar (private communication)

578: 

579: \bibitem{Williams} E.D. Williams (private communication)

580: 

581: \bibitem{Avnir98} D.Avnir, O. Biham, D. Lidar, O. Malcai:

582: Science \textbf{279}, 39 (1998)

583: 

584: \bibitem{Grinstein90} G. Grinstein, D.-H. Lee, S. Sachdev:

585: Phys. Rev. Lett. {\bf 64}, 1927 (1990)

586: 

587: 

588: \end{thebibliography}

589: 

590: \end{document}

591: 

592: 

593: 

594: % \bibitem{label}

595: % Text of bibliographic item

596: 

597: % notes:

598: % \bibitem{label} \note

599: 

600: % subbibitems:

601: % \begin{subbibitems}{label}

602: % \bibitem{label1}

603: % \bibitem{label2}

604: % If there is a note, it should come last:

605: % \bibitem{label3} \note

606: % \end{subbibitems}

607: 

608: 

609: