Extreme and smooth gradient percolation Bernard Sapoval Ecole polytechnique Subtitle: Fractal exponents (like 7/4) without fractals and without SLE and critical fluctuations without critical parameter Agnès Desolneux (MAP 5, Université Paris 5) Andrea Baldassarri (Università La Sapienza, Roma) Diffuse distribution of particles on a lattice: gradient percolation Particles source Lg Lg+1 lines Particles sink L columns Smooth gradient percolation: At each occupied site a continuous gaussian distribution with a variance s2 is attached. F(x,y) =(1/2s2) occup. sites i exp-{[(x-xi)2 +(y-yi)2] / 2s2} Then the contributions of the occupied sites are summed: The contributions of the occupied sites are summed: 500 x 500 s=2 Find the constant level lines: Whatever the level, one observes similar fluctuations. Critical fluctuations but around an arbitrary level Physical problems at the origin of gradient percolation: • Diffusion fronts: geometry of diffuse contacts and soldering. • Structure of fuzzy images. • Corrosion fronts in the etching of random solids. • Erosion fronts: sea-coasts geometry. DIFFUSE STRUCTURE: xf Lg 2 The front is characterized by its position xf , its width f and its length Nf. A source of particles is kept at a constant concentration =1 At time t > 0, the particle concentration at a distance x is given by the complementary error function: p(x) 1 where lD 2(Dt) 1 2 2 1 2 u2 e du x lD 0 is the diffusion length at time t If nf(x) is the mean number per unit horizontal length of points of the front: Diffusion fronts: geometry of diffuse contacts and soldering. What is gradient percolation? Random distribution of particles with a gradient of concentration. The gradient percolation front is the frontier of the infinite cluster Mathematical aspects: P. Nolin ≈ Lg4/7 Numerical observations The average front position is such that, for large Lg, p(xf) is close to pc pc = 0.59280 ± 10-5 Rosso, Gouyet, BS(1985) pc = 0.592745 ± 2. 10-6 Ziff, BS (1986) pc = 0.5927460 ± 5. 10-7 Ziff and Stell (1988) First measurements of the front fractal dimension… Df = 1.76 ± 0.02. (1984) 1984 The front width f followed a power law f Lgwith ≈ 0.57… The front length Nf followed a power law Nf Lg with ≈ 0.42… An island is a finite cluster because it is situated at a position x where p(x) < pc f K. (x f f ) (x) 0 p(x) pc K p(x ) p f 0 f f c dp K 0 . f (x f ) dx f K 0 Lg = 4/3 (den Nijs, 1983) 4 f Lg with 1 7 = 0.5714… Front length Nf: The front width being a correlation length: xf Lg 2 Nf is of order (L / ).Df Nf ≈ L. (Df-1) ≈ L. Lg = (Df -1)/(1 + ) ≈ 0.426 The fact that is the horizontal correlation length has been shown recently by Pierre Nolin (arXiv:math/0610682) but one had +≈1 /(1 + ) + (Df -1)/(1 + ) = 1 Df / (1+ ) ≈ 1 Percolation cluster hull Df = (1+ )/ Df = 7/4 ??? conjecture (1984) But now we know (Duplantier, 1987, Smirnov, 2001) that for sure Df = 7/4: The number of particles in a box of lateral width is : (Nf /L). ≈ Df ≈ Lg Df . / (1+ ) ≈ Lg • But (Nf /L). is the number of particles in a box of size where is the statistical width of the frontier. This width is defined independently of the fractal character of the frontier. Fractality does not appear in this statement For diffusion, it means that the correlated surface is of the order of the diffusion length at time t: lD(t) = 1/2 2(Dt) What if the gradient is so large that the frontier is no more fractal? Small Lg values? The same power laws are observed Is there a conservation law? ? ?? Other situation with gradient percolation: fuzzy image a: original object b: fuzzy photograph c, d: gradient percolation Df = 7/4 Df = 4/3 e, f: smoothing and filtering B. Sapoval And M. Rosso, Fractals, 3, 23-31 (1995) Statistique des fractures: empirique ou théorique? Attaque corrosif d’une couche mince d'aluminium plongé dans une solution Expériences: L.Balazs (1996) PMC Ecole Polytechnique. CCD Camera Aluminium Solution Verre Lumière Other situation with gradient percolation: pit corrosion of an aluminum film. L. Balasz (Ecole polytechnique,1997) Time evolution of the corrosion picture: • The first circular pit grows with time. • It roughens progressively and slows down. • It finally stops on a fractal frontier with dimension 4/3. QuickTime™ et un décompresseur TIFF (non compressé) sont requis pour visionner cette image. The corrosion model: •Andrea Baldassari •Andrea Gabrielli Width of the front as a function of the gradient but where is the gradient? • EROSION OF ROCKY COASTS: Rocky coast-line erosion has marine and atmospheric causes which act on random ‘lithologic’units: random rocks Random means that the ‘mechano-chemical’ properties of the rocks (due to structure and composition) are unknown and exhibit some dispersion. Sea-coasts could be fractal because their geometry damps the sea-waves (and currents …) in such manner that, for a given ‘sea power ’, the erosion is minimized. In that sense it is not only the coast which is eroded but the effective erosion force of the sea which is diminished by the geometry of the coast. And this is why one observes fractal sea-coasts ???? Phys. Rev. Lett. 2004. http://www.nature.com/nsu/031124/031124-4.html The sea power and erosion force is a decreasing function of the coast perimeter, for example: 1 f (t) g Lp (t) 1 Lp (t 0) Empirical breakwater construction recipe: The ‘resisting’ earth is represented by a square lattice where each site presents a random lithology or resistance to the sea. x11 x12 x13 x14 x15 x6 x7 x8 x9 x10 x1 x2 x3 x4 x5 Model representation of the erosion process x11 x12 x13 x14 x15 x6 x7 x8 x9 x10 x1 x2 x3 x4 x5 5 , f(t=0) Lp = L 57p ,=f(t=1) = f(t=2) < f(t=0) The weaker sites are eroded and, at the same time: 1- new sites (strong or weak) are uncovered. 2- the coastal length is modified. Time evolution: fractal morphology is a statistical geometrical attractor North Sardinia real and numerical Df, geo. = 1.33 Df, num. = 4/3 Box-counting determination of the fractal dimension Df = 4/3 The value 4/3 is the dimension of the percolation cluster accessible perimeter. (Grossman, Aharony, 1985) (Duplantier et al. 1999) Scaling behavior of the coast width 4/7 is related to gradient percolation Final coast morphologies The erosion model in itself is more general. It could apply to rough but non-fractal coastline: What if the gradient is so large that the frontier is no more fractal? Small Lg values? The same power laws are observed Extreme gradient percolation or fractal exponents without fractals A. DESOLNEUX, B. SAPOVAL, and A. BALDASSARRI, Self-Organised Percolation Power Laws with and without Fractal Geometry in the Etching of Random Solids, in “Fractal Geometry and Applications” Proceedings of Symposia in Pure Mathematics (PSPUM). American Mathematical Society In print. See cond-mat/0302072. Note that if percolation had been defined in the first place through gradient percolation : …. pc = 1/2 Extreme gradients: is f 7/4 proportionnal to Lg ? Lg = 2 Remark: for small Lg One expects a law of the form 7/4 = a(Lg + b) And Nf7/3 = c(Lg + d) Is b = -1? Numerical test of the exponent: Is 7/4 the best exponent? 1- Choose arbitrary 1.6 < <1.9. 2- Find the best fit values of the law a(Lg + b). 3- Then for each ,compute the distance: 2 1 50 d( ) f (Lg ) a (Lg b ) 47 l 4 2 g Result: d()2 For very extreme gradients, one can determine exact values: DISCUSSION: Comparison between the exact values and the values extrapolated from the numerics for Lg between 4 and 50. Discussion: Comparison between the exact values and the values extrapolated from the numerics for Lg = 4 and Lg = 5 for which one “knows” standard deviations Extrapolation from small to large Concluding remarks • Exact values for Lg = 2 and 3 enters the confidence interval deduced from Lg = 4 and 5. • There exists a unique power law relating the width and length of the gradient percolation front to the gradient length for Lg between 2 and infinity. It does not apply for Lg= 1. • The same type of results are obtained for the triangular lattice. • There exists an implicit relation between width and length of the type 7/4 Nf7/3. You may measure the interface properties and find if it was created, in the past, by a gradient percolation mechanism. • Here, 7/4 is not related to conformal invariance or SLE nor to fractality … • Its origin is combinatory … • There is a conservation law in diffusion on a lattice: The flux per column is proportional to the correlated surface… • However, nature does not live on a lattice and the question remains of how to apply those results to rough coatslines for example… Smooth gradient percolation in 2D: Other broadening functions creates the same type of fluctuations: Gaussian Exponential Yukawa potential … Smooth gradient percolation in 2D: Gaussian broadening s=2 s=3 threshold lines = 0.1; 0.3; 0.5; 0.7; 0.9 Ordinary percolation on a lattice is defined by the occupation or not of sites or bonds with probability p. In continuous percolation a Poisson distribution of points is occupied by circles. The macroscopic manifestations of these elementary events appear only in a narrow region near pc. In smooth percolation, critical fluctuations are found around an arbitrary threshold. Invariance of the geometry with respect to the arbitrary threshold apparent fractal dimension QuickTime™ et un décompresseur TIFF (non compressé) sont requis pour visionner cette image. threshold Gaussian width Exponential broadening = exp(-r/) (1s wave functions) =3 =2 levels lines = 0.1; 0.3; 0.5; 0.7; 0.9 Yukawa potential or screened Coulomb potential = [(1/(r+1)] exp(-r/) is the Debye-Huckel length =5 Yukawa potential or screened Coulomb potential = [(1/(r+1)] exp(-r/) =10 Examples of Yukawa potentials 1- Distribution of radio frequencies antennas: lines of constant electric field should critical geometrical fluctuations. 2- Equipotential lines due to a charge concentration distribution of ions in an electrolyte. Electrochemical reaction paths? 1D Gradient percolation is simple but not trivial 1D Percolation dc= distance for the first empty site dc k k 0 pc 1 p kp (1 p) 1 p k 1 Gradient percolation in 1D Starting at the origin, where do we find the first empty site? The size of the fluctuation zone is Lg - 2E(T) of order Lg Example: Lg =100; s=3 Characteristic function C(x)= 1 if F(x) ≥ with =0.6 Average of Fs,(x) Smooth gradient percolation in 1D At each occupied site a continuous gaussian distribution with a variance s2 is attached. Then the contributions of the occupied sites are summed: 1 (xk)2 /2s2 F(x) e s 2 occupied k Example: Lg =1000; s=3 Characteristic function C(x)= 1 if F(x) ≥ with =0.6