Fractional diffusion equation in fractal media is an integropartial differential equation parametrized by fractal Hausdorff dimension and anomalous diffusion exponent. In this paper, the similarity solution of the fractional diffusion equation was considered. Through the invariants of the group of scaling transformations we derived the integro-ordinary differential equation for the similarity variable. Then by virtue of Mellin transform, the probability density function p(r,t), which is just the fundamental solution of the fractional diffusion equation, was expressed in terms of Fox functions.
1. Introduction
Standard diffusion in d-dimensional space, where d is a positive integer, is a process described by Gaussian distribution. A main feature of the process is the linear relation between the mean square displacement and time; namely, 〈r2(t)〉∝t. Some anomalous diffusion phenomena that take place in impure media, biological tissues, and porous media can be simulated by the diffusion model in fractals [1–6]. In recent years, the fractal theory has been developed rapidly, and it was found to be closely related to the anomalous diffusion phenomena [3–12].
In fractal media, the geometric obstacles existing on all length scales slow down the particle motion in a random walk. The mean square displacement behaves as [2]
(1)R2≡〈r2(t)〉∝t2/dw,
where dw(>2) is the anomalous diffusion exponent. The numerical simulation found that on a large class of fractal structures the general form of the probability density function p(r,t) that the walker is at distance r at time t from its starting point at time t=0 obeys asymptotically a non-Gaussian shape of the form [2, 3]
(2)p(r,t)~t-df/dwexp[-const.×(rR)u],rR≫1,
where u=dw/(dw-1) and df is the fractal Hausdorff dimension.
In order to simulate the diffusion phenomena in fractal media, some scholars have introduced fractional diffusion equations [4, 5, 11–13]. In this paper, we consider the fractional diffusion equation [5, 13]:
(3)∂γp(r,t)∂tγ=1rds-1∂∂r(rds-1∂p∂r),r>0,t>0,
where γ=2/dw, ds=2df/dw is the spectral dimension of the fractal, and the fractional time derivative on the left hand side of (3) is defined as the convolution integral [14–20]:
(4)∂γp(r,t)∂tγ=∂∂t∫0t(t-τ)-γΓ(1-γ)p(r,τ)dτ,0<γ<1,
where Γ(·) is Euler's gamma function. In the limit case, dw→2 and df→d, (3) reduces to the standard d-dimensional diffusion equation.
The fractional calculus has been applied to many fields in science and engineering, such as viscoelasticity, anomalous diffusion, biology, chemistry, and control theory [5, 11–13, 15, 19–22]. Researches on the fractional differential equations attract much attention [15, 23–28]. For linear fractional differential equations, the integral transforms, including the Laplace, Fourier, and Mellin transforms, are usually used to obtain analytic solutions.
In this paper using the similarity method [29] we solve (3) with the following initial and boundary conditions and the conservation condition:
(5)p(r,0)=0,r>0,p(∞,t)=0,t>0,ω(df)∫0∞p(r,t)rdf-1dr=1,
where ω(df) is a constant, which is defined as
(6)ω(df)=2πdf/2Γ(df/2).
We note that the probability density function p(r,t) is just the fundamental solution of the fractional diffusion equation. The similarity method was used by Gorenflo et al. [30], Wyss [31], and Buckwar and Luchko [32] for solving problems of time fractional partial differential equations in one-dimensional case.
2. Derivation of Similarity Solution
First we determine a symmetric group of scaling transformations
(7)Tα:r=αr-,t=αht-,p=αlp-,
where α>0 is a parameter and h, l are constants to be determined. Applying the group of scaling transformations (7), the fractional derivative is converted as follows:
(8)∂γp(r,t)∂tγ=αl∂∂t∫0t(t-τ)-γΓ(1-γ)p-(r-,α-hτ)dτ=αl-hγ∂∂t-∫0t-(t--τ′)-γΓ(1-γ)p-(r-,τ′)dτ′=αl-hγ∂γp-(r-,t-)∂t-γ,
where τ′=α-hτ. Hence the problem (3)–(5) is invariant under the group (7) if and only if
(9)h=2γ,l=-df.
So the symmetric group of scaling transformations is determined:
(10)Tα:r=αr-,t=α2/γt-,p=α-dfp-.
Eliminating the parameter α leads to two invariants:
(11)rt-γ/2=r-t--γ/2,rdfp=r-dfp-.
We denote the two invariants of the group of the scaling transformation Tα as
(12)η=rt-γ/2,F=rdfp.
Next we use the transformation
(13)p(r,t)=r-dfF(η),η=rt-γ/2
to determine the equations for the similarity solution of the problem (3)–(5). Calculating derivative we have
(14)∂p∂r=-dfr-df-1F(η)+r-dft-γ/2F′(η),(15)1rds-1∂∂r(rds-1∂p∂r)=r-dft-γF′′(η)+Δ1r-df-1t-γ/2F′(η)+Δ2r-df-2F(η),
where
(16)Δ1=ds-1-2df,Δ2=df(df-ds+2).
For the left hand side of (3), we introduce the new integral variable
(17)ξ=rτ-γ/2,
we obtain p(r,τ)=r-dfF(ξ), and
(18)∂γp(r,t)∂tγ=∂∂t∫0t(t-τ)-γΓ(1-γ)p(r,τ)dτ=-r2/γ-df-1t-γ/2-1ddη∫η+∞[(η/ξ)-2/γ-1]-γΓ(1-γ)aaaaaaaaaaaaaaaaaaaaaaa×F(ξ)ξ1-2/γdξ.
Letting
(19)g(w)={(w-2/γ-1)-γΓ(1-γ),0<w<1,0,w>1,
we rewrite (18) as
(20)∂γp(r,t)∂tγ=-r2/γ-df-1t-2/γ-1ddη∫0+∞g(ηξ)F(ξ)ξ1-2/γdξ.
From (15) and (20), we obtain the integro-ordinary differential equation for the similarity variables:
(21)-ddη∫0+∞g(ηξ)F(ξ)ξ-2/γ+1dξ=η-2/γ+1F′′(η)+Δ1η-2/γF′(η)+Δ2η-2/γ-1F(η).
The conditions (5) are converted to
(22)F(+∞)=0,ω(df)∫0+∞F(η)η-1dη=1.
Considering the integration in (21), we use Mellin transforms for the new problem (21) and (22). The Mellin transform of function f(x) is defined as [33]
(23)f^(s)=ℳ[f(x),s]=∫0+∞f(x)xs-1dx.
Applying Mellin transform with respect to η to both sides of (21), we get
(24)(s-1)g^(s-1)F^(s-2γ+1)=((s-2γ)(s-2γ-1)-Δ1(s-2γ-1)+Δ2)×F^(s-2γ-1).
Calculating integrations we obtain Mellin transform of the function g(w):
(25)g^(s)=Γ(γ+γs/2)sΓ(γs/2).
Inserting (25) into (24) and then replacing s by s+2/γ+1 we obtain the difference equation for the function F^(s):
(26)Γ(γ(s/2+1/γ+1))Γ(γ(s/2+1/γ))F^(s+2)=(s2+s-Δ1s+Δ2)F^(s).
In order to solve the difference equation, we introduce s=2q and F^(2q)=T(q), and rewrite (26) into
(27)T(q+1)T(q)=4(q+df2)(q+1+df-ds2)Γ(γq+1)Γ(γq+γ+1).
A particular solution of (27) is
(28)T(q)=C4qΓ(q+df/2)Γ(q+1+(df-ds)/2)Γ(γq+1),
where C is an arbitrary constant. For the solution of (27), we can multiply T(q) by any function Y(q) which satisfies Y(q+1)/Y(q)=1.
We notice that F^(s) is a Mellin transform defined only in some strip 0≤σ1<Re(s)<σ2 from the conditions (22). So (26) is valid only in the overlap of the two strips σ1<Re(s)<σ2 and σ1<Re(s+2)<σ2, and there is no such overlap unless σ1+2<σ2. Thus Y(q) cannot have poles; otherwise, it would have a row of poles separated exactly by one unit. In addition, Y(q) cannot grow faster than |q| as Im(q)→∞ in the inversion strip; otherwise the inversion integral would diverge. Thus Y(q) is a bounded entire function and equals a constant by Liouville's theorem.
Therefore, T(q) has only the form of (28) and we have
(29)F^(s)=T(s2)=C2sΓ(df/2+s/2)Γ(1+(df-ds)/2+s/2)Γ(1+γs/2).
It follows from (22) that F^(0)=1/ω(df). Thus we have
(30)C=1ω(df)Γ(df/2)Γ(1+(df-ds)/2).
The inverse Mellin transform of (29) is
(31)F(η)=C2πi∫c-i∞c+i∞2sΓ(df/2+s/2)Γ(1+(df-ds)/2+s/2)Γ(1+γs/2)aaaaaaaaaaaa×η-sds.
Replacing s by -s and using the definition of Fox functions we obtain [34, 35]
(32)F(η)=CH1,22,0(η2|(df/2,1/2),(1+(df-ds)/2,1/2)(1,γ/2)).
Inserting the expressions into (13) and using properties of Fox functions, we obtain the probability density function in terms of the Fox function:
(33)p(r,t)=dw2-dft-df/dwω(df)Γ(df/2)Γ(1+df/2-df/dw)×H1,22,0(rdw2dwt|(0,dw/2),(1-df/dw,dw/2)(1-df/dw,1)).
For a large class of fractal structures, the spectral dimension [2] satisfies ds<2; that is, df<dw. So the Fox function in (33) can be expanded into a series by using residue theorem on the simple poles:
(34)Pa={2kdw∣k=0,1,…}∪{2dw(1-dfdw+k)∣k=0,1,…}.
The series representation for the probability density is calculated to be
(35)p(r,t)=21-dft-df/dwω(df)Γ(df/2)Γ(1+df/2-df/dw)×∑k=0∞(-1)kk!×[Γ(1-df/dw-k)Γ(1-df/dw-2k/dw)(rdw2dwt)2k/dwaaaaaaaaaaaaaaaaaa+Γ(df/dw-1-k)Γ(1-df/dw-2/dw(1-df/dw+k))aaaaaaaaaaaaaaaaaa×(rdw2dwt)2/dw(1-df/dw+k)].
3. Discussions and Conclusions
In the limit case, dw→2 and df→d, (3) reduces to the d-dimensional standard diffusion equation, and the probability density (35) is simplified to the Gaussian distribution:
(36)p(r,t)=1(4πt)d/2exp(-r24t).
In Figures 1 and 2, we plot the curves of p(r,0.25) versus r and p(r,0.5) versus r, respectively, for df=1 and different values of dw. In Figures 3 and 4, we plot the curves of p(r,0.25) versus r and p(r,0.5) versus r, respectively, for df=1.5 and different values of dw. The figures display that, as the anomalous diffusion exponent dw increases, the peak value of the probability density function p(r,t) at r=0 decreases. In addition, as the fractal Hausdorff dimension df increases from 1 to 1.5, the peak value of p(r,t) at r=0 decreases.
Curves of p(r,0.25) versus r for df=1 and for dw=2 (solid line), dw=2.5 (dot line), dw=3 (dash line), and dw=3.5 (dot-dash line).
Curves of p(r,0.5) versus r for df=1 and for dw=2 (solid line), dw=2.5 (dot line), dw=3 (dash line), and dw=3.5 (dot-dash line).
Curves of p(r,0.25) versus r for df=1.5 and for dw=2 (solid line), dw=2.5 (dot line), dw=3 (dash line), and dw=3.5 (dot-dash line).
Curves of p(r,0.5) versus r for df=1.5 and for dw=2 (solid line), dw=2.5 (dot line), dw=3 (dash line), and dw=3.5 (dot-dash line).
Compared with the similarity method for classic partial differential equations, the similarity method for fractional diffusion equation involves the similarity integral variable ξ=rτ-γ/2, and the reduction equation is an integro-ordinary differential equation for the similarity solution. The obtained probability density p(r,t) is just the fundamental solution of the fractional diffusion equation.
Conflict of Interests
The authors declare that there is no conflict of interests regarding the publication of this paper.
Acknowledgments
This work was supported by the National Natural Science Foundation of China (11201308) and the Innovation Program of Shanghai Municipal Education Commission (14ZZ161).
MandelbrotB. B.1982New York, NY, USAW. H. FreemanMR665254HavlinS.Ben-AvrahamD.Diffusion in disordered media20025111872922-s2.0-003627369110.1080/00018730110116353LiuD.LiH.ChangF.LinL.Anomalous diffusion on the percolating networks1998621391442-s2.0-0343275908RenF.-Y.LiangJ.-R.WangX.-T.The determination of the diffusion kernel on fractals and fractional diffusion equation for transport phenomena in random media19992523-41411502-s2.0-0000532336ZengQ.LiH.Diffusion equation for disordered fractal media2000811171212-s2.0-0347998359CattaniC.Fractals and hidden symmetries in DNA20102010312-s2.0-7795452576210.1155/2010/507056507056MR2660045ZBL1189.92015LiM.ZhaoW.On bandlimitedness and lag-limitedness of fractional Gaussian noise201339291955196110.1016/j.physa.2012.12.035MR3038268LiM.A class of negatively fractal dimensional Gaussian random functions20112011182-s2.0-7925156472910.1155/2011/291028291028MR2754182ZBL1209.28015LiM.CattaniC.ChenS.-Y.Viewing sea level by a one-dimensional random function with long memory20112011132-s2.0-7865073597810.1155/2011/654284654284CattaniC.PierroG.On the fractal geometry of DNA by the binary image analysis20137591544157010.1007/s11538-013-9859-9MR3105523ZBL1272.92013MetzlerR.KlafterJ.The random walk's guide to anomalous diffusion: a fractional dynamics approach200033911772-s2.0-000264142110.1016/S0370-1573(00)00070-3MR1809268ZBL0984.82032GionaM.RomanH. E.Fractional diffusion equation for transport phenomena in random media19921851–487972-s2.0-44049116564MetzlerR.GlöckleW. G.NonnenmacherT. F.Fractional model equation for anomalous diffusion1994211113242-s2.0-43949160695SamkoS. G.KilbasA. A.MarichevO. I.1993Amsterdam, The NetherlandsGordon and BreachMR1347689PodlubnyI.1999San Diego, Calif, USAAcademic PressMR1658022OldhamK. B.SpanierJ.1974New York, NY, USAAcademic PressMR0361633BăleanuD.DiethelmK.ScalasE.TrujilloJ. J.2012Boston, Mass, USAWorld ScientificSeries on Complexity, Nonlinearity and Chaos10.1142/9789814355216MR2894576MillerK. S.RossB.1993New York, NY, USAJohn Wiley & SonsMR1219954KilbasA. A.SrivastavaH. M.TrujilloJ. J.2006Amsterdam, The NetherlandsElsevierMR2218073MainardiF.2010London, UKImperial College Press10.1142/9781848163300MR2676137XuM. Y.TanW. C.Theoretical analysis of the velocity field, stress field and vortex sheet of generalized second order fluid with fractional anomalous diffusion20014411138713992-s2.0-0035520475LiC. P.DengW. H.XuD.Chaos synchronization of the Chua system with a fractional order200636021711852-s2.0-2774455478110.1016/j.physa.2005.06.078MR2186261DuanJ.-S.Time- and space-fractional partial differential equations200546113504135112-s2.0-1704440706710.1063/1.1819524013504MR2113773ZBL1076.26006DuanJ.-S.The periodic solution of fractional oscillation equation with periodic input201320136869484MR311524210.1155/2013/869484DuanJ. S.RachR.BaleanuD.WazwazA. M.A review of the Adomian decomposition method and its applications to fractional differential equations2012327399LiuF.ZhuangP.AnhV.TurnerI.BurrageK.Stability and convergence of the difference methods for the space-time fractional advection-diffusion equation2007191112202-s2.0-3454767324410.1016/j.amc.2006.08.162MR2385498ZBL1193.76093WangZ. H.WangX.General solution of the Bagley-Torvik equation with fractional-order derivative2010155127912852-s2.0-7044912255910.1016/j.cnsns.2009.05.069MR2563928ZBL1221.34020YangA. M.CattaniC.JafariH.YangX. J.Analytical solutions of the onedimensional heat equations arising in fractal transient conduction with local fractional derivative20132013546253510.1155/2013/462535BlumanG. W.AncoS. C.2002New York, NY, USASpringerMR1914342GorenfloR.LuchkoY.MainardiF.Wright functions as scale-invariant solutions of the diffusion-wave equation20001181-21751912-s2.0-000010358910.1016/S0377-0427(00)00288-0MR1765948ZBL0973.35012WyssW.The fractional diffusion equation19862711278227852-s2.0-000948130310.1063/1.527251MR861345ZBL0632.35031BuckwarE.LuchkoY.Invariance of a partial differential equation of fractional order under the Lie group of scaling transformations1998227181972-s2.0-003983576810.1006/jmaa.1998.6078MR1652906ZBL0932.58038DaviesB.20023rdNew York, NY, USASpringer10.1007/978-1-4684-9283-5MR1867914MathaiA. M.SaxenaR. K.1978New Delhi, IndiaJohn Wiley & SonsMR513025SrivastavaH. M.GuptaK. C.GoyalS. P.1982New Delhi, IndiaSouth AsianMR691138