Chapter 1 Double Laplace Transform and Explicit Fractional Analogue of 2D Laplacian
S. Rogosin and M. Dubatovskaya
1.1 Introduction
Fractional modeling and fractional differential equations becomes recently very im- portant tool of applied mathematics (see, e.g. [GKMR], [KiSrTr], [SaKiMa]). By using fractional models and solving corresponding fractional differential equation one can understand the nature of nonstandard and anomalous processes in Physics, Engineering, Chemistry, Biology, Economics etc. (see, e.g., [Uch13]). One of the useful methods in the study of fractional differential equations is the integral trans- form method (see [DebBha], [Deb16]). The development of this method is an actual challenging problem. The aim of this paper is to study the properties of different type fractional analogs of 2D Laplacian by using double integral Laplace transform. The double Laplace transform has been investigated since late 1930s. An extended description of this transform and its properties was made by D.Bernstein (her PhD thesis has been published in [Bern41]). Some of the properties of the double Laplace transform are presented in the Sec. 1.2. Multi-dimensional fractional differential operators are not completely recognized by fractional society. An approach describing these operators in terms of Fourier images (see e.g. [SaKiMa, Ch. 5]) is more theoretical than practical. Recently, two attempts were made to find an explicit form of the fractional analog of 2D Laplace operator (see [Yak10], [DalYak]), namely, these analogs have the following form
α,β 1+α 1+β 4 u(x,y) = D0+,xu(x,y) + D0+,yu(x,y), 0 < α,β < 1, (1.1)
S. Rogosin The Belarusian State University, Minsk, Belarus, e-mail: [email protected] M. Dubatovskaya The Belarusian State University, Minsk, Belarus, e-mail: [email protected]
1 2 S. Rogosin and M. Dubatovskaya
α,β α α β β 4−+u(x,y) = −D1−,xD0+,xu(x,y) − D1−,yD0+,yu(x,y), 0 < α,β < 1. (1.2) We consider here operator (1.1) and the skewed modifications of operator (1.2) for the unit square and for the quarter-plane, respectively (introduced in [Lip11])1
α,β;S α β α β 4−+ u(x,y) = −D0+,xD1−,yu(x,y) − D0+,yD1−,xu(x,y), 0 < α + β ≤ 2, (1.3)
α,β;Q α β α β 4−+ u(x,y) = −D0+,xD−,yu(x,y) − D0+,yD−,xu(x,y), 0 < α + β ≤ 2. (1.4) γ γ γ Here (D0+,xu)(x,y), (D1−,xu)(x,y), (D−,xu)(x,y) are the left- and right-sided Riemann-Liouville fractional derivatives and the Liouville fractional derivative de- fined for arbitrary positive γ and n = [γ] + 1 (see, e.g., [KiSrTr]):
x 1 d n Z u(τ,y)dτ d n (Dγ u)(x,y) = = (In−γ u)(x,y), (1.5) 0+,x Γ (n − γ) dx (x − τ)γ−n+1 dx 0+,x 0
1 1 d n Z u(τ,y)dτ d n (Dγ u)(x,y) = − = − (In−γ u)(x,y), 1−,x Γ (n − γ) dx (τ − x)γ−n+1 dx 1−,x x (1.6) +∞ 1 d n Z u(τ,y)dτ d n (Dγ u)(x,y) = − = − (In−γ u)(x,y). −,x Γ (n − γ) dx (τ − x)γ−n+1 dx −,x x (1.7) Using the double Laplace transform applied to the above said fractional analogs of the 2D laplacian we show that the proper boundary conditions for the frac- tional Laplace equations have the form of the so-called Cauchy type conditions. This means that the proposed fractional analogs of the 2D laplacian are correct generalizations of the standard one-dimensional fractional differential operator (see [KiSrTr]). Main attention in this paper is paid to the operators (1.1) and (1.4). The study of another (and more cumbersome) form (1.3) is a subject of the forthcoming paper.
1.2 Double Laplace transforms. Definition and main properties
Double Laplace transforms is defined by the following relation (see, e.g. [DebBha, p. 274]2)
1 One of the reasons to considered the operator in skewed form, i.e. to have both derivatives in x and y in both terms, is to enrich the set of eigenfunctions of the operator. 2 Similar definition with extra normalization was proposed in [DitPru]. 1 Double Laplace Transform and Fractional Laplacian 3
+∞ +∞ Z Z ˆ −px−qy fˆ(p,q) = (L2 f )(p,q) = (L2 f (x,y))(p,q) = f (x,y)e dxdy. (1.8) 0 0 We present here the following main properties of the double Laplace transforms following [DebBha].
• (linearity) (L2(a f + bg))(p,q) = a(L2 f )(p,q) + b(L2g)(p,q), a,b ∈ R; • (scaling) 1 p q (L f (amx,bny))(p,q) = (L f (x,y)) , , 2 ambn 2 am bn
a,b > 0, m,n ∈ N; −ax−by • (first shifting property) L2e f (x,y) (p,q) = (L2 f )(p + a,q + b); • (second shifting property)
−ap−bq (L2 f (x − a,y − b)H(x − a)H(y − b))(p,q) = e (L2 f )(p,q), a,b > 0,
H is the Heaviside function; • (transform of functions with special dependence on the arguments)
1 (L2 f (x + y))(p,q) = p−q [(L f )(p) − (L f )(q)]; 1 (L2 f (x − y))(p,q) = p+q [(L f )(p) + (L f )(q)] f is even, 1 = p+q [(L f )(p) − (L f )(q)] f is odd;
• (transform of the truncation of functions)
1 (L2 f (x)H(x − y))(p,q) = q [(L f )(p) − (L f )(p + q)]; 1 (L2 f (x)H(y − x))(p,q) = q (L f )(p + q); 1 (L2 f (x)H(x + y))(p,q) = q (L f )(p);
• (differential properties)
∂ f L2 ∂x (x,y) (p,q) = p(L2 f (x,y))(p,q) − (Ly f (0,y))(q); ∂ f L2 ∂y (x,y) (p,q) = q(L2 f (x,y))(p,q) − (Lx f (x,0))(p); ∂ 2 f L2 ∂x∂y (x,y) (p,q) = p(L2 f (x,y))(p,q) + q(L2 f (x,y))(p,q) − (Ly f (0,y))(q) − (Lx f (x,0))(p) + f (0,0); 2 ∂ f (x,y) (p,q) = p2 ( f (x,y))(p,q) − p( f (0,y))(q) L2 ∂x2 L2 Ly ∂ f − Ly ∂x (0,y) (q) + f (0,0);
• (convolution) (L2 f ∗ g)(p,q) = (L2 f (x,y))(p,q) · (L2g(x,y))(p,q), where f ∗ x y g is the double convolution: ( f ∗ g)(x,y) = R R f (x −t,y − s)g(t,s)dt ds; 0 0 • (inverse transform) 4 S. Rogosin and M. Dubatovskaya
c+i∞ d+i∞ Z Z −1 ˆ 1 px 1 qy ˆ f (x,y) = L fˆ(p,q) (x,y) = e dp e fˆ(p,q)dq. (1.9) 2 2πi 2πi c−i∞ d−i∞
The last formula is valid if the direct double Laplace transform fˆ(p,q) is an analytic function in the direct product of two semi-planes, namely in the domain (p,q) ∈ C2 : Re p > c, Req > d , where c,d are some real numbers. All other for- mulas are valid provided existence of the both of their sides.
Remark 1. Double integral transform in the Laplace-Carson form (see [DitPru])
+∞ +∞ Z Z ¯ −px−qy f¯(p,q) = (C2 f )(p,q) = (C2 f (x,y))(p,q) = pq f (x,y)e dxdy (1.10) 0 0 has similar properties. In particular, the scaling property has the following form (which is suitable for the fractal theory, see [NigBal13]) p q (C f (amx,bny))(p,q) = (C f (x,y)) , , 2 2 am bn a,b > 0, m,n ∈ N. Few examples illustrating the above formulas. −kx−ly Example 1. f1(x,y) = e .
+∞ +∞ Z Z −(p+k)x −(q+l)y (L2 f1)(p,q) = e dx e dy = 0 0
1 +∞ 1 +∞ 1 1 − e−(p+k)x − e−(q+l)y = . p + k x=0 q + l y=0 p + k q + l
Example 2. f2(x,y) = H(x − y).
+∞ +∞ +∞ x Z Z Z Z −px −qy −px −qy (L2 f2)(p,q) = e dx e H(x − y)dy = e dx e dy = 0 0 0 0
+∞ 1 Z 1 1 1 +∞ 1 = e−px(1 − e−qx)dx = − e−px + e−(p+q)x = . q q p p + q x=0 p(p + q) 0 Remark 2. There exists a problem with application of the inverse double Laplace transform. To describe this problem let us consider a simple situation, namely, when the double Laplace image is a rational function of two complex variables p and q:
R(p,q) (L f )(p,q) = . 2 S(p,q) 1 Double Laplace Transform and Fractional Laplacian 5
Let the denominator is represented in the form
m n S(p,q) = C0 ∏(p − ak)∏(q − bl), (1.11) k=1 l=1 where ak, bl are certain complex numbers (not necessarily different). Then to re- cover an initial function f (x,y) one needs to apply Residue Theorem in two semi- planes of variables p and q, respectively. But, if the polynomial S(p,q) is irreducible, i.e. there exist no nontrivial poly- nomials S1(p,q), S2(p,q) of the above described form (1.11), such that S(p,q) = S1(p,q)S2(p,q), then such technique cannot be applied. Simplest example of the irreducible polynomial is S(p,q) = p2 + q2 − 1.
Let us repeat a simple example of application of the double Laplace transform to solving of an initial-boundary problem for a second order differential equation in the quarter-plane (see [DebBha, p. 278]). Example 3. Let us consider the wave equation
2 c uxx = utt , x ≥ 0, t > 0, (1.12) under the following initial conditions
u(x,0) = f (x), ut (x,0) = g(x), (1.13) the following boundary conditions
u(0,t) = 0, ux(0,t) = 0. (1.14)
Assume that the given functions f (x) and g(x) are suitable to apply the below tech- nique (e.g. are absolutely integrable in x ∈ R+). Let us apply the double Laplace transform uˆ(p,q) = (L2u(x,t))(p,q) to the equation (1.12):
2 c (L2uxx(x,t))(p,q) = (L2utt (x,t))(p,q).
By using the differential properties of the double Laplace transform we have
2 2 c p uˆ(p,q) − p(L2u(0,t))(q) − (L2ux(0,t))(q) + u(0,0) =
2 = p uˆ(p,q) − q(L2u(x,0))(p) − (L2ut (x,0))(p) + u(0,0) . Applying conditions (1.13), (1.14) we arrive at the following relation