ebook img

Distribution of the area enclosed by a 2D random walk in a disordered medium PDF

4 Pages·0.12 MB·English
Save to my drive
Quick download
Download
Most books are stored in the elastic cloud where traffic is expensive. For this reason, we have a limit on daily download.

Preview Distribution of the area enclosed by a 2D random walk in a disordered medium

Distribution of the area enclosed by a 2D random walk in a disordered medium K.V. Samokhin∗ Cavendish Laboratory, University of Cambridge, Madingley Road, Cambridge CB3 0HE, UK (February 1, 2008) 9 The asymptotic probability distribution for a Brownian particle wandering in a 2D plane with 9 random trapsto enclose thealgebraic area Aby timet is calculated usingtheinstanton technique. 9 1 PACS numbers: 02.50.-r, 05.40.Fb, 05.40.Jc n a J 8 It is well known that the properties of random walks B,andusingtheWienerpath-integralformulafortheso- 1 are dramatically changed by the presence of quenched lution of Eq. (1), we arrive at the following expression: disorder. In particular, if a walker can be irreversibly ] h trapped at some randomly distributed sites, then the ∞ r(t)=r (A,tU)= dpeiBA r(τ)e−S[r(τ)], (2) c asymptoticprobabilitytoreturntothestartingpoint(or e P | Z−∞ Zr(0)=r′ D the totalprobabilityto surviveuntil time t) is givenind m dimensionsbyp(t) exp( ctd/(d+2))[1],whilethemean where - square displacemen∼t decre−ases compared to a pure diffu- t a sion: r2 t2/(d+2) [2]. On another hand, if a walker t 1 i t h i ∼ S = dτ r˙2(τ)+U(r(τ))+ Br2(τ)θ˙(τ) . s isadvectedbyarandomforce,thenthe totalprobability −Z (cid:18)2D 2 (cid:19) . is conserved, but the diffusion coefficient and the mean 0 t a square displacement acquire logarithmic corrections [3]. It is easy to see that the path integralon the right-hand m Further examples can be found, for instance, in Ref. [4]. side of Eq. (2) represents the Euclidean Green func- - Much less, however, is known about the influence of a tion (r,t;r′,0)ofafictitiousquantumparticleofmass nd quenched disorder on the topological properties of ran- m =G(B2D)−1 moving in the random potential U(r) and o dom walks, which became a subject of theoretical inves- in the uniform magnetic field B (we choose the units in c tigation very recently [5]. By “topological” properties which h¯ = e = c = 1). Let us now assume that the [ we mean such characteristics as the winding number of trajectory is closed (r = r′) and average the enclosed 1 a Brownian particle, or the linking number of a closed area distribution over all positions of the starting point: v polymer,orthe algebraicareaenclosedbythe trajectory (...) r = (1/Ω) d2r(...) (Ω is the system volume). We h i 5 of a random walker. In this Rapid Communication, we have: R 6 concentrate on the latter case and calculate the asymp- 1 totic distribution of the area swept by a planar random (A,tU)= ∞ dB eiBAZ(t,B U), 1 walk wandering in the presence of traps. P | Z | 0 −∞ The probability distribution for a random walk start- 9 ing at some point r′ at t = 0 to end at a point r after where Z(t,B) is the partition function at inverse “tem- 9 / time t obeys the diffusion equation perature” 1/T = t. Averaging now over the positions of at traps, we finally obtain: ∂P m =D 2P U(r)P. (1) ∂t ∇ − ∞ ∞ - (A,t)= dB dE eiBAe−EtN(E,B), (3) ond HReir)eisDthisetrhaenddoiffmus“iponotceonetffiiacl”ie,nwthainchdrUe(prr)es=enUts0tPheitδr(arp−- P Z−∞ Z0 ping probability per unit time (U > 0). The positions whereN(E,B)istheaveragedensityofstatesofaquan- c 0 : Rioftrapsaredistributeduniformlyinaplaneaccording tum particle described by the Hamiltonian v to the Poissonlaw with meandensity ρ. The probability i H =D( i A(r))2+U(r). (4) X for a random walk of “length” t to enclose the algebraic − ∇− r area A can be obtained by averaging the corresponding a δ-function constraint over the solutions of Eq. (1) in a We choose the cylindricalgauge for the vector potential: A =Br/2. given distribution of traps: θ In an ideal case (i.e. in the absence of traps), the 1 t eigenvalues of the Hamiltonian (4) are the Landau lev- (A,tU) δ A dτ r2(τ)θ˙(τ) , P | ≡(cid:28) (cid:18) − 2Z0 (cid:19)(cid:29)P(r,t;r′,0|U) els En = ωc(n+1/2), where ωc = 2DB is the cyclotron frequency. The density of states is then given by the set where θ(t) is the angle between the radius-vector r(t) of of equidistant δ-function peaks, and the partition func- the particle and some fixed direction in the plane. Writ- tion can be easily evaluated, resulting in the following ingtheδ-functionasanintegraloveranauxiliaryvariable expression for the enclosed area distribution: 1 1 A (A,t) , x . (5) GR(r,r′;E,B)= i lim 2Φ(r)ϕ(r)ϕ¯(r′) P ∼ cosh2x ∼ Dt − η→+0Z D exp i d2r Φ¯(E H +iη)Φ , (6) This result was first obtained by Levy quite a while ago × (cid:18) Z − (cid:19) [6]. The form of the dimensionless scaling variable x is quite natural, since the only characteristic scale with di- where the two-component superfields mensionalityofareainacleansystemisthemeansquare ϕ(r) displacement r2 Dt. Φ(r)= , Φ¯(r)=(ϕ∗(r),ψ¯(r)) h i∼ (cid:18)ψ(r)(cid:19) In order to calculate the density of states in the pres- ence of disorder, let us first estimate what characteristic are composed of a Bose field ϕ and a Grassmanian field scalesofenergyandmagneticfielddeterminetheasymp- ψ, and 2Φ = (1/π) (Reϕ) (Imϕ) ψ¯ ψ. The prob- totic behavior of (A,t). We are interested in calcula- D D D D D P ability of having N impurities located at the points tion of the probability distribution at large but fixed t R ,...,R in the area Ω in the plane is given by the 1 N and A . As seen from (3), this limit corresponds to →∞ Poissonlaw: E 0, B 0, and ω E. Therefore, it looks natural c → → ≪ tostartwiththecaseofB =0andmakesurethatweare e−ρΩ P (R ,...,R )= (ρΩ)N. able to treat the magnetic field as a small perturbation N 1 N N! in the relevant range of parameters. Using this expressionto average(6)overthe positionsof At B = 0, the low-energy behavior of the density traps, we end up with the following effective action: of states is determined by the rare fluctuations of the concentration of impurities creating the large areas free of traps which are able to sustain the eigenstates with iS[Φ]= dDr iΦ¯ E D( i A)2+iη Φ E 0. TheasymptoticexpressionisgivenbyN(E,B = Z n (cid:0) − − ∇− (cid:1) 0)→exp( constρD/E) at E E , where E = ρU is ρ 1 e−iU0Φ¯Φ . (7) 0 0 0 the∼mean−value of the random≪potential [7]. Quantita- − (cid:16) − (cid:17)o tively, such exponentially small “tails” are determined Animportantpropertyoftheaction(7)isthatitisin- by the contributions of instantons, i.e. spatially local- variantundertheglobalsupersymmetrytransformations, ized solutions of the saddle-point equations of the effec- mixing the boson and fermion sectors: tive field theory. The typical size of a clean area, or the instanton diameter, grows in the time representation as Φ Φ˜ =TΦ, (8) l (Dt/ρ)1/4 [2]. If an external magnetic field is im- → inst ∼ posedontheinstanton,thenitseffectontheenergyspec- where T is a 2 2 unitary supermatrix [9]. Due to this × trumcanbecalculatedperturbativelyaslongasthemag- symmetry,theproblemoffindingthesaddlepointsof(7) neticlengthl = 1/Bconsiderablyexceedstheinstan- can be considerably simplified. Indeed, since the saddle B ton dimension l p, which is the case if A (Dt/ρ)1/2. pointmanifoldisinvariantunderthetransformations(8), inst It is this condition that determines the limi≫ts of applica- weareabletoseektheinstantonsolutioninthefollowing bility of our theory. Note also that if we were interested form: aintcmaolcduelraattioenAo,ftthheeninwteercmouedldiauteseatshyemepxtaoctitcsexopfrPes(sAio,nt)s Φ (r)= ϕinst(r) , (9) inst (cid:18) 0 (cid:19) for the density of states at E ω /2 ω [8] (since the c c − ≪ random potential is positive everywhere, the density of whereϕ (r)=ϕ (r)isacylindricallysymmetricspa- inst inst states vanishes at E <ω /2). c tially localized function. If to write the boson fields as Let us now calculate the effect of an external mag- ϕ=ϕ +iϕ , ϕ∗ =ϕ iϕ , where ϕ are real on the 1 2 1 2 1,2 netic field on instantons explicitly. The density of states initialfunctionalintegra−tioncontour,thentheimaginary canbeexpressedintermsoftheretardedGreenfunction partofthe Greenfunctionis determinedbyanon-trivial of the Schr¨odinger equation with the Hamiltonian (4): saddle point of the action in the complex plane of ϕ 1,2 N(E,B) = (πΩ)−1 dDr Im GR(r,r;E,B) U, where [10,11], and GR(r,r′;E,B−)= r(ER H+i0h)−1 r′ . TheGireenfunc- h | − | i tioncan,inturn,becalculatedbystandardmeansofthe N(E,B) e−Sinst(E,B) (10) quantumfieldtheory. Inordertocarryouttheaveraging ∼ overthepositionsoftraps,weresorttothesupersymme- withtheexponentialaccuracy. Thefermionsectorcanbe try approach, in which the cancellation of denominators neglected as long as we are not interested in calculation is achieved by doubling the degrees of freedom and in- of the pre-exponential factor in (10). troducing the commuting and anticommuting fields on In order to make Eq. (7) in the boson sector real, we equalfooting(see,forinstance,Ref.[9]). Beforedisorder rotate the integration contour: ϕ e−iπ/4ϕ . As 1,2 1,2 → averaging,theretardedGreenfunctioncanbewrittenas resultof this, the exponentonthe right-handside of Eq. the following functional integral: (6) changes: iS[Φ] S[Φ], where the action S is →− 2 S = d2r ϕ E+D( i A)2 ϕ and (10), the Lifshitz result N(E) exp( constρD/E) i i ∼ − Z n (cid:0)− − ∇− (cid:1) is recovered. +ρ 1 e−U0ϕ2i (11) In principle, one can find the exact solutions of the (cid:16) − (cid:17)o Schr¨odinger equations inside and outside the potential well at B =0 (they are expressedin terms of the conflu- (i=1,2). Introducing the dimensionless variables: 6 ent hypergeometric functions), match them at the point x = x and finally end up with a transcendental equa- r =ξx, ϕ (r)= U f(x), 1 inst 0 tionforx (α,β),whichcanbesolvedatβ 0. However, 1 p → we prefer not to follow this procedure here, because the whereξ = D/E,weobtain,fromEq. (11),anon-linear sameresultscanbeobtainedusingmuchmorephysically differentialpequation for the saddle-point solution f: apparent reasoning in the spirit of the Lifshitz’s original derivation [7]. First, we note that, in the limit α , 1 d x d f +β2x2f +α2e−f2f =f, (12) the instanton solution satisfies the Schr¨odinger equ→ati∞on − xdx(cid:18) dx(cid:19) foraparticleconfinedinthepotentialwellwithinfinitely high walls in a magnetic field. The ground state energy where α = E /E 1 and β = ω /4E 1. Similar equations wpitho0ut m≫agnetic field werce obta≪ined in Refs. ǫ0 is equalto unity (inthe units ofE). Inthe absenceof magnetic field this condition fixes the radius of the well [12,13], using different techniques. at x (0) = a. At β = 0, the lowest order perturbative Due to complexity of the equation (12), we are able 1 6 correction to the ground state energy is to find only an approximate solution. To this end, we replace the “potential” V(f) = α2e−f2 in Eq. (12) by a a β2 dxx3f2(x) piecewise constant potential: Z δǫ = 0 , (18) 0 a α2 , at f <1, dxxf2(x) Veff(f)=(cid:26)0 , at f >1. (13) Z0 where f(x) J (x) is the unperturbed ground state Then, Eq. (12) becomes effectively linear andreduces to ∼ 0 wave function (16). Calculating the integrals with the a couple of the Schr¨odinger equations, whose solutions Bessel functions, we obtain: δǫ =cβ2, where c 1.261. f (x) satisfy the following conditions: 0 ≃ 1,2 To keep the ground state energy fixed, this correction f (x )=f (x )=1, f′(x 0)=f′(x +0), (14) should be compensated by the correspondingincrease in 1 1 2 1 1 1− 2 1 the radius of the potential well: x2(β) = x2(0)(1+δǫ ). 1 1 0 where x (α,β) is the position of the discontinuity in Substituting this in (15), we finally obtain: 1 the effective potential, which is to be determined self- πρDa2 cD2 consistently. Going back to the dimensional variables, it S (E,B) 1+ B2 . (19) inst ≃ E (cid:18) 4E2 (cid:19) is easy to see from (11) and (12) that the instanton ac- tion is proportional to the area of the effective potential This expression is valid at E 0, B 0, ω /E 0. well: → → c → Theasymptoticprobabilitydistributionoftheenclosed S =πρξ2x2. (15) areacannow be obtainedfrom(3), (10)and(19)by cal- inst 1 culating the integrals by the steepest descent method: Let us start with the case of B = 0, i.e. β = 0. The ∞ ∞ solution of the linearized saddle-point equations can be (A,t) dB dE eiBAe−Ete−Sinst(E,B) P ∼Z Z written in form −∞ 0 a√π √ρA2 f (x)=C J (x), at 0<x<x , exp( πρDa2t)exp . (20) f(x)= 1 1 0 1 (16) ∼ − (cid:26)− c (Dt)3/2(cid:27) (cid:26)f2(x)=C2K0(αx), at x1 <x, p The first exponential on the right-hand side is noth- where J0(x) and K0(x) are the Bessel functions of real ing but the asymptotic “tail” of the total probability andimaginaryarguments,respectively. Substitutingthis p(t) = dA (A,t) for a random walker not to be solution in the matching conditions (14), we obtain the trapped [R1]. TPhe second exponential thus represents the equation conditionalprobability to enclose the areaA, provideda walker has survived until time t. J (x) K (αx) 1 =α 1 . (17) Weseethattheasymptoticbehaviorof (A,t)isdras- J (x) K (αx) P 0 0 tically changedby the presence of disorder,comparedto theLevy’sresult(5). ThedistributionbecomesGaussian In the limit α , assuming that x 1 and using → ∞ 1 ∼ (A,t) exp( x2), with the scaling variable the asymptotic expansions of the Bessel functions [14], P ∼ − fiwrestobzetraoinoJf0th(xe1f)u=nct0i,oin.eJ. x(x1)=. Aa,ftwerhesurebsati≈tu2ti.o4n05inis(t1h5e) x A (ρDt)1/4, (21) 0 ∼ Dt 3 so that the standard deviation now growsslower than in ∗ Permanent address: L.D.Landau Institutefor Theoret- the clean case: A2 1/2 t3/4 (the mean value A is, of ical Physics, Kosygina Str.2, 117940 Moscow, Russia. h i ∼ h i course,zero). Suchadifferentformofthescalingvariable can be related to the presence of an extra length scale [1] B.Ya.BalagurovandV.G.Vaks,Zh.Eksp.Teor.Fiz.65, r 1/ρ in the system, which depends on the concen- 1939 (1973) [Sov.Phys. – JETP, 38, 968 (1974)]. ρ tra∼tiopn of traps (nothing depends on the absolute value [2] P.GrassbergerandI.Procaccia,J.Chem.Phys.77,6281 U of the random potential). Note also that a similar (1982). G0aussian distribution was obtained in Ref. [15] for a 2D [3] D.S. Fisher, Phys.Rev.A 30, 960 (1984). [4] J.P. Bouchaud and A. Georges, Phys. Rep. 195, 127 random walk in a box of a finite size L. In this case, the (1990). fictitious magnetic field was also treated perturbatively, giving rise to A2 1/2 L(Dt)1/2. Qualitatively, such a [5] K.V.Samokhin,J.Phys.A31,8789(1998);ibid31,9455 h i ∼ (1998). differencebetweenthetwosystemsisduetothefactthat [6] P.Levy,Processus Stochastique et Mouvement Brownien in our case the size of the effective potential well is not (Gauthier-Villars, Paris, 1948). constant, but grows with time as L(t)=l (t) t1/4. inst ∼ [7] I.M. Lifshitz, Usp. Fiz. Nauk 83, 617 (1964) [Sov. Phys. In conclusion, we studied the asymptotic probability – Uspekhi7, 549 (1965)]. distribution of the algebraic area enclosed by a planar [8] E. Brezin, D.Gross and C. Itzykson,Nucl.Phys. B 235 random walk in the presence of immobile random traps. [FS11], 24 (1984). It is shown that this probability is directly related to [9] K.B. Efetov, Supersymmetry in Disorder and Chaos “the Lifshitz tail” in the density of states of a quantum (Cambridge University Press, 1997). particleinaPoissondisorderanduniformmagneticfield. [10] J.S. Langer, Ann.Phys. 41, 108 (1967). In contrast to the case of an ideal random walk, the en- [11] C.CallanandS.Coleman,Phys.Rev.D16(1977)1762. closed area distribution turns out to be Gaussian with [12] R.FriedbergandJ.M. Luttinger,Phys.Rev.B 12, 4460 the standard deviation growing as t3/4. (1975). [13] T.C. Lubensky,Phys.Rev.A 30, 2657 (1984). This work was financially supported by the Engineer- [14] M.AbramowitzandI.A.Stegun,HandbookofMathemat- ing and Physical Sciences Research Council (UK). ical Functions (Dover,New York,1965). [15] J. Desbois and A.Comtet, J. Phys. A 25, 3097 (1992). 4

See more

The list of books you might like

Most books are stored in the elastic cloud where traffic is expensive. For this reason, we have a limit on daily download.