Sie sind hier: ICP » R. Hilfer » Publikationen

2.2 Mathematical Introduction to Fractional Derivatives

[22.1.1] The brief historical introduction has shown that fractional derivatives may be defined in numerous ways. [22.1.2] A natural and frequently used approach starts from repeated integration and extends it to fractional integrals. [22.1.3] Fractional derivatives are then defined either by continuation of fractional integrals to negative order (following Leibniz’ ideas [73]), or by integer order derivatives of fractional integrals (as suggested by Riemann [96]).

2.2.1 Fractional Integrals

2.2.1.1 Iterated Integrals

[22.2.1] Consider a locally integrable1 (This is a footnote:) 1 A function f:GR is called locally integrable if it is integrable on all compact subsets KG (see eq.(B.9)). real valued function f:GR whose domain of definition G=a,bR is an interval with -a<b. [22.2.2] Integrating

[page 23, §0]    n times gives the fundamental formula

Ia+nfx=axax1axn-1fxndxndx2dx1
=1n-1!axx-yn-1fydy,(2.27)

where a<x<b and nN. [23.0.1] This formula may be proved by induction. [23.0.2] It reduces n-fold integration to a single convolution integral (Faltung). [23.0.3] The subscript a+ indicates that the integration has a as its lower limit. [23.0.4] An analogous formula holds with lower limit x and upper limit a. [23.0.5] In that case the subscript a- will be used.

2.2.1.2 Riemann-Liouville Fractional Integrals

[23.1.1] Equation (2.27) for n-fold integration can be generalized to noninteger values of n using the relation n-1!=k=1n-1k=Γn where

Γz=01-logxz-1dx(2.28)

is Euler’s Γ-function defined for all zC.

Definition 2.1

[23.2.1] Let -a<x<b. [23.2.2] The Riemann-Liouville fractional integral of order α>0 with lower limit a is defined for locally integrable functions f:a,bR as

Ia+αfx=1Γαaxx-yα-1fydy(2.29)
for x>a. [23.2.3] The Riemann-Liouville fractional integral of order α>0 with upper limit b is defined as
Ib-αfx=1Γαxby-xα-1fydy

for x<b. [23.2.4] For α=0

Ia+0fx=Ib-0fx=fx(2.30)

completes the definition. [23.2.5] The definition may be generalized to αC with Reα>0.

[page 24, §1]    [24.1.1] Formula (2.29a) appears in [96, p.363] with a>- and in [76, p.8] with a=-. [24.1.2] The notation is not standardized. [24.1.3] Leibniz, Lagrange and Liouville used the symbol α [73, 22, 76], Grünwald wrote αdxαx=ax=x, while Riemann used x-α [96] and Most wrote da-α/dx-α [89]. [24.1.4] The notation in (2.29) is that of [99, 98, 54, 52]. [24.1.5] Modern authors also use fα [37], Iα [97], Ixαa [94], Ixα [23], Dx-αa [85, 102, 91], or d-α/dx-a-α [92] instead of Ia+α2 (This is a footnote:) 2 Some authors [97, 26, 92, 23, 85, 91] employ the derivative symbol D also for integrals, resp. I for derivatives, to emphasize the similarity between fractional integration and differentiation. If this is done, the choice of Riesz and Feller, namely I, seems superior in the sense that fractional derivatives, similar to integrals, are nonlocal operators, while integer derivatives are local operators..

[24.2.1] The fractional integral operators Ia+α,Ib-α are commonly called Riemann-Liouville fractional integrals [99, 98, 94] although sometimes this name is reserved for the case a=0 [85]. [24.2.2] Their domain of definition is typically chosen as DIa+α=L1a,b or DIa+α=Lloc1a,b [99, 98, 94]. [24.2.3] For the definition of Lebesgue spaces see the Appendix B. [24.2.4] If fL1a,b then Ia+αfL1a,b and Ia+αfx is finite for almost all x. [24.2.5] If fLpa,b with 1p and α>1/p then Ia+αfx is finite for all xa,b. [24.2.6] Analogous statements hold for Ib-αfx [98].

[24.3.1] A short table of Riemann-Liouville fractional integrals is given in Appendix A. [24.3.2] For a more extensive list of fractional integrals see [24].

2.2.1.3 Weyl Fractional Integrals

[24.4.1] Examples (2.5) and (2.6) or (A.3) and (A.5) show that Definition 2.1 is well suited for fractional integration of power series, but not for functions defined by Fourier series. [24.4.2] In fact, if fx is a periodic function with period 2π, and3 (This is a footnote:) 3The notation indicates that the sum does not need to converge, and, if it converges, does not need to converge to fx.

fxk=-ckeikx(2.31)

then the Riemann-Liouville fractional Ia+αf will in general not be periodic. [24.4.3] For this reason an alternative definition of fractional integrals was investigated by Weyl [124].

[24.5.1] Functions on the unit circle G=R/2πZ correspond to 2π-periodic functions on the real line. [24.5.2] Let fx be periodic with period 2π and such that the integral of f over the interval 0,2π vanishes, so that c0=0 in eq. (2.31). [24.5.3] Then the integral of f is itself a periodic function, and the constant of integration can be chosen such that the integral over 0,2π vanishes again. [24.5.4] Repeating the integration n times one finds using (2.6) and the integral representation

[page 25, §0]    ck=1/2π02πe-iksfsds of Fourier coefficients

k=-ckeikxikn=12π02πfyk=-k0eikx-yikndy(2.32)

with c0=0. [25.0.1] Recall the convolution formula [132, p.36]

f*gt=12π02πft-sgsds=k=-fkgkeikt(2.33)

for two periodic functions ftk=-fkeikt and gtk=-gkeikt. [25.0.2] Using eq. (2.33) and generalizing (2.32) to noninteger n suggests the following definition. [99, 94].

Definition 2.2

[25.1.1] Let fLpR/2πZ,1p< be periodic with period 2π and such that its integral over a period vanishes. [25.1.2] The Weyl fractional integral of order α is defined as

I±αfx=Ψ±α*fx=12π02πΨ±αx-yfydy,(2.34)

where

Ψ±αx=k=-k0eikx±ikα(2.35)

for 0<α<1.

[25.2.1] It can be shown that the series for Ψ±αx converges and that the Weyl definition coincides with the Riemann-Liouville definition [133]

I+αfx=1Γα-xx-yα-1fydy,(2.36)
respectively
I-αfx=1Γαxy-xα-1fydy

for 2π periodic functions whose integral over a period vanishes. [25.2.2] This is eq. (2.29) with a=- resp. b=. [25.2.3] For this reason the Riemann-Liouville

[page 26, §0]    fractional integrals with limits ±, I+αf=I-+αf and I-αf=I-αf, are often called Weyl fractional integrals [24, 99, 85, 94].

[26.1.1] The Weyl fractional integral may be rewritten as a convolution

I±αfx=K±α*fx,(2.37)

where the convolution product for functions on R is defined as4 (This is a footnote:) 4 If K,fL1R then K*ft exists for almost all tR and fL1R. [26.1.2] If KLpR, fLqR with 1<p,q< and 1/p+1/q=1 then K*fC0R, the space of continuous functions vanishing at infinity.

K*fx:=-Kx-yfydy(2.38)

and the convolution kernels are defined as

K±αx:=Θ±x±xα-1Γα(2.39)

for α>0. [26.1.3] Here

Θx=1,x>00,x0(2.40)

is the Heaviside unit step function, and xα=expαlogx with the convention that logx is real for x>0. [26.1.4] For α=0 the kernel

K+0x=K-0x=δx(2.41)

is the Dirac δ-function defined in (C.2) in Appendix C. [26.1.5] Note that K±αLloc1R for α>0.

2.2.1.4 Riesz Fractional Integrals

[26.2.1] Riemann-Liouville and Weyl fractional integrals have upper or lower limits of integration, and are sometimes called left-sided resp. right-sided integrals. [26.2.2] A more symmetric definition was advanced in [97].

Definition 2.3

[26.3.1] Let fLloc1R be locally integrable. [26.3.2] The Riesz fractional integral or Riesz potential of order α>0 is defined as the linear combination [99]

Iαfx=I+αfx+I-αfx2cosαπ/2
=12Γαcosαπ/2-fyx-y1-αdy(2.42)

[page 27, §0]    of right- and left-sided Weyl fractional integrals. [27.0.1] The conjugate Riesz potential is defined by

Iα~fx=I+αfx-I-αfx2sinαπ/2
=12Γαsinαπ/2-sgnx-yfyx-y1-αdy.(2.43)

[27.0.2] Of course, α2k+1,kZ in (2.42) and α2k,kZ in (2.43). [27.0.3] The definition is again completed with

I0fx=I0~fx=fx(2.44)

for α=0.

[27.1.1] Riesz fractional integration may be written as a convolution

Iαfx=Kα*fx(2.45a)
Iα~fx=K~α*fx(2.45b)

with the (one-dimensional) Riesz kernels

Kαx=K+αx+K-αx2cosαπ/2=xα-12cosαπ/2Γα(2.46)

for α2k+1,kZ, and

K~αx=K+αx-K-αx2sinαπ/2=xα-1sgnx2sinαπ/2Γα(2.47)

for α2k,kZ. [27.1.2] Subsequently, Feller introduced the generalized Riesz-Feller kernels [26]

Kα,βx=xα-1sinαπ/2+βsgnx2sinαπ/2Γα(2.48)

with parameter βR. [27.1.3] The corresponding generalized Riesz-Feller fractional integral of order α and type β is defined as

Iα,βfx=Kα,β*fx.(2.49)

[27.1.4] This formula interpolates continuously from the Weyl integral I-α=Iα,-π/2 for β=-π/2 through the Riesz integral Iα=Iα,0 for β=0 to the Weyl integral I+α=Iα,π/2 for β=π/2. [27.1.5] Due to their symmetry Riesz-Feller fractional integrals are readily generalized to higher dimensions.

[page 28, §1]

2.2.1.5 Fractional Integrals of Distributions

[28.1.1] Fractional integration can be extended to distributions using the convolution formula (2.37) above. [28.1.2] Distributions are generalized functions [105, 31]. [28.1.3] They are defined as linear functionals on a space X of conveniently chosen ‘‘test functions’’. [28.1.4] For every locally integrable function fLloc1R there exists a distribution Ff:XC defined by

Ffφ=f,φ=-fxφxdx,(2.50)

where φX is test function from a suitable space X of test functions. [28.1.5] By abuse of notation one often writes f for the associated distribution Ff. [28.1.6] Distributions that correspond to functions via (2.50) are called regular distributions. [28.1.7] Examples for regular distributions are the convolution kernels K±αLloc1R defined in (2.39). [28.1.8] They are locally integrable functions on R when α>0. [28.1.9] Distributions that are not regular are sometimes called singular. [28.1.10] An important example for a singular distribution is the Dirac δ-function. [28.1.11] It is defined as δ:XC

δxφxdx=φ0(2.51)

for every test function φX. [28.1.12] The test function space X is usually chosen as a subspace of CR, the space of infinitely differentiable functions. [28.1.13] A brief introduction to distributions is given in Appendix C.

[28.2.1] In order to generalize (2.37) to distributions one must define the convolution of two distributions. [28.2.2] To do so one multiplies eq. (2.38) on both sides with a smooth test function φCcR of compact support. [28.2.3] Integrating gives

K*f,φ=--Kx-yfyφxdydx
=--Kxfyφx+ydydx
=Kx,fy,φx+y,(2.52)

where the notation fy,φx+y means that the functional Ff is applied to the function φ(x+) for fixed x. [28.2.4] Explicitly, for fixed x

Ffφx=fy,φxy=fy,φx+y=-fyφx+ydx,(2.53)

[page 29, §0]    where φx()=φ(x+). [29.0.1] Equation (2.52) can be used as a definition for the convolution of distributions provided that the right hand side has meaning. [29.0.2] This is not always the case as the counterexample K=f=1 shows. [29.0.3] In general the convolution product is not associative (see eq. (2.113)). [29.0.4] However, associative and commutative convolution algebras exist [21]. [29.0.5] Equation (2.52) is always meaningful when suppK or suppf is compact [63]. [29.0.6] Another case is when K and f have support in R+. [29.0.7] This will be assumed in the following.

Definition 2.4

[29.1.1] Let f be a distribution fC0R with suppfR+. [29.1.2] Then its fractional integral is the distribution I0+αf defined as

I0+αf,φ=I+αf,φ=K+α*f,φ(2.54)

for Reα>0. [29.1.3] It has support in R+.

[29.2.1] If fC0R with suppfR+ then also I0+αfC0R with suppI0+αfR+.

2.2.1.6 Integral Transforms

[29.3.1] The Fourier transformation is defined as

Ffk=-e-ikxfxdx(2.55)

for functions fL1R. [29.3.2] Then

FI±αfk=±ik-αFfk(2.56)

holds for 0<α<1 by virtue of the convolution theorem. [29.3.3] The equation cannot be extended directly to α1 because the Fourier integral on the left hand side may not exist. [29.3.4] Consider e.g. α=1 and fCcR. [29.3.5] Then I+1fxconst as x and FI+1f does not exist [94]. [29.3.6] Equation (2.56) can be extended to all α with Reα>0 for functions in the so called Lizorkin space [99, p.148] defined as the space of functions fSR such that DmFf0=0 for all mN0.

[29.4.1] For the Riesz potentials one has

FIαfk=k-αFfk(2.57a)
FIα~fk=-isgnkk-αFfk(2.57b)

for functions in Lizorkin space.

[page 30, §1]    [30.1.1] The Laplace transform is defined as

Lfu=0e-uxfxdx(2.58)

for locally integrable functions f:R+C. [30.1.2] Now

LI0+αfu=u-αLfu(2.59)

by the convolution theorem for Laplace transforms. [30.1.3] The Laplace transform of I0-αf leads to a more complicated operator.

2.2.1.7 Fractional Integration by Parts

[30.2.1] If fxLpa,b,gLqa,b with 1/p+1/q1+α, p,q1 and p1, q1 for 1/p+1/q=1+α then the formula

abfxIa+αgxdx=abgxIb-αfxdx(2.60)

holds. [30.2.2] The formula is known as fractional integration by parts [99]. [30.2.3] For fxLpR,gLqR with p>1,q>1 and 1/p+1/q=1+α the analogous formula

-fxI+αgxdx=-gxI-αfxdx(2.61)

holds for Weyl fractional integrals.

[30.3.1] These formulae provide a second method of generalizing fractional integration to distributions. [30.3.2] Equation (2.60) may be read as

Ia+αf,φ=f,Ib-αφ(2.62)

for a distribution f and a test function φ. [30.3.3] It shows that right- and left-sided fractional integrals are adjoint operators. [30.3.4] The formula may be viewed as a definition of the fractional integral Ia+αf of a distribution provided that the operator Ib-α maps the test function space into itself.

2.2.1.8 Hardy-Littlewood Theorem

[30.4.1] The mapping properties of convolutions can be studied with the help of Youngs inequality. Let p,q,r obey 1p,q,r and 1/p+1/q=1+1/r. [30.4.2] If KLpR and fLqR then K*fLrR and Youngs inequality K*frKpfq holds. [30.4.3] It follows that K*fqCfp if

[page 31, §0]    1pq and KLrR with 1/r=1+1/q-1/p. [31.0.1] The Hardy-Littlewood theorem states that these estimates remain valid for K±α although these kernels do not belong to any LpR-space [37, 38]. [31.0.2] The theorem was generalized to higher dimensions by Sobolev in 1938, and is also known as the Hardy-Littlewood-Sobolev inequality (see [37, 38, 113, 63]).

Theorem 2.5

[31.1.1] Let 0<α<1, 1<p<1/α, -a<b. [31.1.2] Then Ia+α,Ib-α are bounded linear operators from Lpa,b to Lqa,b with 1/q=1/p-α,i.e. there exists a constant Cp,q independent of f such that Ia+αfqCfp.

2.2.1.9 Additivity

[31.2.1] The basic composition law for fractional integrals follows from

K+α*K+βx=0xK+α(x-y)K+β(y)dy=0xx-yα-1Γαyβ-1Γβdy
=xα-1Γαxβ-1Γβ011-zα-1zβ-1xdz
=xα+β-1Γα+β=K+α+β(x),(2.63)

where Euler’s Beta-function

ΓαΓβΓα+β=011-zα-1zβ-1dz=Bα,β(2.64)

was used. [31.2.2] This implies the semigroup law for exponents

Ia+αIa+β=Ia+α+β,(2.65)

also called additivity law. [31.2.3] It holds for Riemann-Liouville, Weyl and Riesz-Feller fractional integrals of functions.

2.2.2 Fractional Derivatives

2.2.2.1 Riemann-Liouville Fractional Derivatives

[31.3.1] Riemann [96, p.341] suggested to define fractional derivatives as integer order derivatives of fractional integrals.

Definition 2.6

[31.4.1] Let -a<x<b. [31.4.2] The Riemann-Liouville fractional derivative of order 0<α<1 with lower limit a (resp. upper limit b) is defined for

[page 32, §0]    functions such that fL1a,b and f*K1-αW1,1a,b as

Da±αfx=±ddxIa±1-αfx(2.66)

and Da±0fx=fx for α=0. [32.0.1] For α>1 the definition is extended for functions fL1a,b with f*Kn-αWn,1a,b as

Da±αfx=±1ndndxnIa±n-αfx,(2.67)

where5 (This is a footnote:) 5x is the largest integer smaller than x. n=Reα+1 is smallest integer larger than α.

[32.1.1] Here Wk,pG=fLpG:DkfLpG denotes a Sobolev space defined in (B.17). [32.1.2] For k=p=1 the space W1,1a,b=AC0a,b coincides with the space of absolutely continuous functions.

[32.2.1] The notation for fractional derivatives is not standardized6 (This is a footnote:) 6see footnote 2.1. [32.2.2] Leibniz and Euler used dα [73, 72, 25] Riemann wrote xα [96], Liouville preferred dα/dxα [76], Grünwald used dαf/dxαx=ax=x or Dαfx=ax=x [34], Marchaud wrote Daα, and Hardy-Littlewood used an index fα [37]. [32.2.3] The notation in (2.67) follows [99, 98, 54, 52]. Modern authors also use I-α [97], Ix-α [23], Dxαa [94, 85, 102], dα/dxα [129, 102], dα/dx-aα [92] instead of Da+α.

[32.3.1] Let fx be absolutely continuous on the finite interval a,b. [32.3.2] Then, its derivative f exists almost everywhere on a,b with fL1a,b, and the function f can be written as

fx=axfydy+fa=Ia+1fx+fa.(2.68)

Substituting this into Ia+αf gives

Ia+αfx=Ia+1Ia+αfx+faΓα+1x-aα,(2.69)

where commutativity of Ia+1 and Ia+α was used. [32.3.3] It follows that

DIa+αfx-Ia+αDfx=faΓαx-aα-1(2.70)

for 0<α<1. [32.3.4] Above, the notations

Dfx=dfxdx=fx(2.71)

were used for the first order derivative.

[page 33, §1]    [33.1.1] This observation suggests to introduce a modified Riemann-Liouville fractional derivative through

D~a+αfx:=Ia+n-αfnx=1Γn-αaxfnyx-yα-n+1dy,(2.72)

where n=Reα+1. [33.1.2] Note, that f must be at least n-times differentiable. [33.1.3] Formula (2.72) is due to Liouville [76, p.10] (see eq. (2.18) above), but nowadays sometimes named after Caputo [17].

[33.2.1] The relation between (2.72) and (2.67) is given by

Theorem 2.7

[33.2.2] For fACn-1a,b with n=Reα+1 the Riemann-Liouville fractional derivative Da+αfx exists almost everywhere for Reα0. [33.2.3] It can be written as

Da+αfx=D~a+αfx+k=0n-1x-ak-αΓk-α+1fka(2.73)

in terms of the Liouville(-Caputo) derivative defined in (2.72).

[33.3.1] The Riemann-Liouville fractional derivative is the left inverse of Riemann-Liouville fractional integrals. [33.3.2] More specifically, [99, p.44]

Theorem 2.8

[33.3.3] Let fL1a,b. [33.3.4] Then

Da+αIa+αfx=fx(2.74)

holds for all α with Reα0.

[33.4.1] For the right inverses of fractional integrals one finds

Theorem 2.9

[33.4.2] Let fL1a,b and Reα>0. [33.4.3] If in addition Ia+n-αfACna,b where n=Reα+1 then

Ia+αDa+αfx=fx-k=0n-1x-aα-k-1Γα-kDn-k-1Ia+n-αfa(2.75)

holds. [33.4.4] For 0<Reα<1 this becomes

Ia+αDa+αfx=fx-Ia+1-αfaΓαx-aα-1.(2.76)

[33.5.1] The last theorem implies that for fL1a,b and Reα>0 with n=Reα+1 the equality

Ia+αDa+αfx=fx(2.77)

[page 34, §0]    holds only if

Ia+n-αfACna,b(2.78)
and
DkIa+n-αfa=0

for all k=0,1,2,,n-1. [34.0.1] Note that the existence of gx=Da+αfx in eq. (2.77) does not imply that fx can be written as Ia+αgx for some integrable function g [99]. [34.0.2] This holds only if both conditions (2.78) are satisfied. [34.0.3] As an example where one of them fails, consider the function fx=x-aα-1 for 0<α<1. [34.0.4] Then Da+αx-aα-1=0 exists. [34.0.5] Now D0Ia+1-αx-aα-10 so that (2.78b) fails. [34.0.6] There does not exist an integrable g such that Ia+αg=x-aα-1. [34.0.7] In fact, g corresponds to the δ-distribution δx-a.

2.2.2.2 General Types of Fractional Derivatives

[34.1.1] Riemann-Liouville fractional derivatives have been generalized in [52, p.433] to fractional derivatives of different types.

Definition 2.10

[34.2.1] The generalized Riemann-Liouville fractional derivative of order 0<α<1 and type 0β1 with lower (resp. upper) limit a is defined as

Da±α,βfx=±Ia±β1-αddxIa±1-β1-αfx(2.79)

for functions such that the expression on the right hand side exists.

[34.3.1] The type β of a fractional derivative allows to interpolate continuously from Da±α=Da±α,0 to D~a±α=Da±α,1. [34.3.2] A relation between fractional derivatives of the same order but different types was given in [52, p.434].

2.2.2.3 Marchaud-Hadamard Fractional Derivatives

[34.4.1] Marchaud’s approach [78] is based on Hadamards finite parts of divergent integrals [36]. [34.4.2] The strategy is to define fractional derivatives as analytic continuation of fractional integrals to negative orders. [see [99, p.225]]

Definition 2.11

[34.5.1] Let -<a<b< and 0<α<1. [34.5.2] The Marchaud fractional derivative of order α with lower limit a is defined as

Ma+αfx=fxΓ1-αx-aα+αΓ1-αaxfx-fyx-yα+1dy(2.80)

[page 35, §0]    and the Marchaud fractional derivative of order α with upper limit b is defined as

Mb-αfx=fxΓ1-αb-xα+αΓ1-αxbfx-fyx-yα+1dy.(2.81)

[35.0.1] For a=- (resp. b=) the definition is

M±αfx=αΓ1-α0fx-fxyyα+1dy.(2.82)

[35.0.2] The definition is completed with M0f=f for all variants.

[35.1.1] The idea of Marchaud’s method is to extend the Riemann-Liouville integral from α>0 to α<0, and to define

I+-αfx=1Γ-α0y-α-1fx-ydy,(2.83)

where α>0. [35.1.2] However, this is not possible because the integral in (2.83) diverges. [35.1.3] The idea is to subtract the divergent part of the integral,

εy-α-1fxdy=fxαεα(2.84)

obtained by setting fx-yfx for y0. [35.1.4] Subtracting (2.83) from (2.84) for 0<α<1 suggests the definition

M+αfx=limε0+1Γ-αεfx-fx-yyα+1dy(2.85)

[35.1.5] Formal integration by parts leads to I+1-αfx, showing that this definition contains the Riemann-Liouville definition.

[35.2.1] The definition may be extended to α>1 in two ways. [35.2.2] The first consists in applying (2.85) to the n-th derivative dnf/dxn for n<α<n+1. [35.2.3] The second possibility is to regard fx-y-fx as a first order difference, and to generalize to n-th order differences. [35.2.4] The n-th order difference is

Δynfx=1-Tynfx=k=0n-1knkfx-ky,(2.86)

where 1fx=fx is the identity operator and

Thfx=fx-h(2.87)

[page 36, §0]    is the translation operator. [36.0.1] The Marchaud fractional derivative can then be extended to 0<α<n through [94, 98]

M+αfx=limε0+1Cα,nεΔynfxyα+1dy,(2.88)

where

Cα,n=01-e-ynyα+1dy,(2.89)

where the limit may be taken in the sense of pointwise or norm convergence.

[36.1.1] The Marchaud derivatives M±α are defined for a wider class of functions than Weyl derivatives D±α. [36.1.2] As an example consider the function fx=const.

[36.2.1] Let f be such that there exists a function gL1a,b with f=Ia+αg. [36.2.2] Then the Riemann-Liouville derivative and the Marchaud derivative coincide almost everywhere, i.e. Ma+αfx=Da+αfx for almost all x [99, p.228].

2.2.2.4 Weyl Fractional Derivatives

[36.3.1] There are two kinds of Weyl fractional derivatives for periodic functions. [36.3.2] The Weyl-Liouville fractional derivative is defined as [99, p.351],[94]

D±αfx=±ddxI±1-αfx(2.90)

for 0<α<1 where the Weyl integral ±I±αf was defined in (2.34). [36.3.3] The Weyl-Marchaud fractional derivative is defined as [99, p.352],[94]

W±αfx=12π02πfx-y-fxD1Ψ±1-αydy(2.91)

for 0<α<1 where Ψ±x is defined in eq. (2.35). [36.3.4] The Weyl derivatives are defined for periodic functions of with zero mean in CβR/2πZ where β>α. [36.3.5] In this space D±αfx=W±αfx, i.e. the Weyl-Liouville and Weyl-Marchaud form coincide [99]. [36.3.6] As for fractional integrals, it can be shown that the Weyl-Liouville derivative 0<α<1

D+αfx=1Γ1-α-xfyx-yαdy(2.92)

coincides with the Riemann-Liouville derivative with lower limit -. [36.3.7] In addition one has the equivalence D+αf=W+αf with the Marchaud-Hadamard fractional derivative in a suitable sense [99, p.357].

[page 37, §1]

2.2.2.5 Riesz Fractional Derivatives

[37.1.1] To define the Riesz fractional derivative as integer derivatives of Riesz potentials consider the Fourier transforms

FDI1-αfk=ikkα-1Ffk=isgnkkαFfk(2.93)
FDI1-α~fk=ik-isgnkkα-1Ffk=kαFfk(2.94)

for 0<α<1. [37.1.2] Comparing this to eq. (2.57) suggests to consider

ddxI1-α~fx=limh01hI1-α~fx+h-I1-α~fx(2.95)

as a candidate for the Riesz fractional derivative.

[37.2.1] Following [94] the strong Riesz fractional derivative of order α Rαf of a function fLpR, 1p<, is defined through the limit

limh01hf*Kh1-α-Rαfp=0,(2.96)

whenever it exists. [37.2.2] The convolution kernel defined as

Kh1-α=12Γ1-αsinαπ/2sgnx+hx+hα-sgnxxα(2.97)

is obtained from eq. (2.95). [37.2.3] Indeed, this definition is equivalent to eq. (2.94). [37.2.4] A function fLpR where 1p2 has a strong Riesz derivative of order α if and only if there exsists a function gLpR such that kαFfk=Fgk. [37.2.5] Then Rαf=g.

2.2.2.6 Grünwald-Letnikov Fractional Derivatives

[37.3.1] The basic idea of the Grünwald approach is to generalize finite difference quotients to noninteger order, and then take the limit to obtain a differential quotient. [37.3.2] The first order derivative is the limit

ddxfx=Dfx=limh0fx-fx-hh=limh01-Thhfx(2.98)

of a difference quotient. [37.3.3] In the last equality 1fx=fx is the identity operator, and

Thfx=fx-h(2.99)

is the translation operator. [37.3.4] Repeated application of T gives

Thnfx=fx-nh,(2.100)

[page 38, §0]    where nN. [38.0.1] The second order derivative can then be written as

d2dx2fx=D2fx=limh0fx-2fx-h+fx-2hh2
=limh01-Thh2fx,(2.101)

and the n-th derivative

dndxnfx=Dnfx=limh01hnk=0n-1knkfx-kh
=limh01-Thhnfx,(2.102)

which exhibits the similarity with the binomial formula. [38.0.2] The generalization to noninteger n gives rise to fractional difference quotients defined through

Δhαfx=k=0-1kαkfx-kh(2.103)

for α>0. [38.0.3] These are generally divergent for α<0. [38.0.4] For example, if fx=1, then

k=0N-1kαk=1Γ1-αΓN+1-αΓN+1(2.104)

diverges as N if α<0. [38.0.5] Fractional difference quotients were studied in [68]. Note that fractional differences obey [99]

ΔhαΔhβfx=Δhα+βfx.(2.105)
Definition 2.12

[38.1.1] The Grünwald-Letnikov fractional derivative of order α>0 is defined as the limit

G±αfx=limh0+1hαΔ±hαfx(2.106)

of fractional difference quotients whenever the limit exists. [38.1.2] The Grünwald Letnikov fractional derivative is called pointwise or strong depending on whether the limit is taken pointwise or in the norm of a suitable Banach space.

[38.2.1] For a definition of Banach spaces and their norms see e.g. [128].

[38.3.1] The Grünwald-Letnikov fractional derivative has been studied for periodic functions in LpR/2πZ with 1p< in [99, 94]. [38.3.2] It has the following properties.

[page 39, §1]   

Theorem 2.13

[39.1.1] Let fLpR/2πZ, 1p< and α>0. [39.1.2] Then the following statements are equivalent:

  1. G+αfLpR/2πZ

  2. [39.1.3] There exists a function gLpR/2πZ such that
    ikαFfxk=Fgxk where kZ.

  3. [39.1.4] There exists a function gLpR/2πZ such that
    fx-Ffx0=I+αgx holds for almost all x.

Theorem 2.14

[39.2.1] Let fLpR/2πZ, 1p< and α,β>0. [39.2.2] Then:

  1. G+αfLpR/2πZ implies G+βfLpR/2πZ for every 0<β<α.

  2. G+αG+βf=G+α+βf

  3. G+αI+αf=fx-Ff0

2.2.2.7 Fractional Derivatives of Distributions

[39.3.1] The basic idea for defining fractional differentiation of distributions is to extend the definition of fractional integration (2.54) to negative α. [39.3.2] However, for Reα<0 the distribution K+α becomes singular because xα-1 is not locally integrable in this case. [39.3.3] The extension of K+α to Reα<0 requires regularization [31, 128, 63]. [39.3.4] It turns out that the regularization exists and is essentially unique as long as -αN0.

Definition 2.15

[39.4.1] Let f be a distribution fC0R with suppfR+. [39.4.2] Then the fractional derivative of order α with lower limit 0 is the distribution D0+αf defined as

D0+αf,φ=D+αf,φ=K+-α*f,φ,(2.107)

where αC and

K+αx=Θxxα-1Γα,Reα>0dNdxNΘxxα+N-1Γα+N,Reα+N>0,NN(2.108)

is the kernel distribution. [39.4.3] For α=0 one finds K+0x=d/dxΘx=δx and D0+0=1 as the identity operator. [39.4.4] For the α=-k,kN one finds

K+-kx=δkx,(2.109)

where δk is the k-th derivative of the δ distribution.

[page 40, §1]    [40.1.1] The kernel distribution in (2.108) is

K+-αx=ddxΘxx-αΓ1-α=ddxK+1-αx(2.110)

for 0<α<1. [40.1.2] Its regularized action is

K+-αx,φx=ddxK+1-α(x),φ(x)=-K+1-α(x),φ(x)(2.111a)
=-1Γ1-αlimε0εx-αφxdx(2.111b)
=-limε0φx+CΓαxαε-εφx+CΓ-αx1+αdx(2.111c)
=0φx-φ0Γ-αx1+αdx,(2.111d)

where φ< was assumed in the last step and the arbitrary constant was chosen as C=-φ0. [40.1.3] This choice regularizes the divergent first term in (2.111c). [40.1.4] If this rule is used for the distributional convolution

K+-α*fx=1Γ-α0fx-fx-yyα+1dy=M+αfx(2.112)

then the Marchaud-Hadamard form is recovered with 0<α<1.

[40.2.1] It is now possible to show that the convolution of distributions is in general not associative. [40.2.2] A counterexample is

1*δ*Θ=1*Θ=0*Θ=01=1*δ=1*Θ=1*δ*Θ,(2.113)

where Θ is the Heaviside step function.

[40.3.1] D0+αf has support in R+. [40.3.2] The distributions in fC0R with suppfR+ form a convolution algebra [21] and one finds [31, 99]

Theorem 2.16

[40.3.3] If fC0R with suppfR+ then also I0+αfC0R with I0+αsuppfR+. [40.3.4] Moreover, for all α,βC

D0+αD0+βf=D0+α+βf(2.114)

with D0+αf=I0+-αf for Reα<0. [40.3.5] For each fC0R with suppfR+ there exists a unique distribution gC0R with suppgR+ such that f=I0+αg.

[page 41, §1]    [41.1.1] Note that

D0+αf=D0+α1f=K+-α*K+0*f=D0+αδ*f=δα*f(2.115)

for all αC. [41.2.1] Also, the differentiation rule

D0+αK+β=K+β-α(2.116)

holds for all α,βC. [41.2.2] It contains

DK+β=K+β-1(2.117)

for all βC as a special case.

2.2.2.8 Fractional Derivatives at Their Lower Limit

[41.3.1] All fractional derivatives defined above are nonlocal operators. [41.3.2] A local fractional derivative operator was introduced in [40, 41, 52].

Definition 2.17

[41.4.1] For -<a< the Riemann-Liouville fractional derivative of order 0<α<1 at the lower limit a is defined by

dαfdxαx=a=fαa=limxa±Da±αfx,(2.118)

whenever the two limits exist and are equal. [41.4.2] If fαa exists the function f is called fractionally differentiable at the limit a.

[41.5.1] These operators are useful for the analysis of singularities. [41.5.2] They were applied in [40, 41, 42, 44, 52] to the analysis of singularities in the theory of critical phenomena and to the generalization of Ehrenfests classification of phase transitions. [41.5.3] There is a close relationship to the theory of regularly varying functions [107] as evidenced by the following result [52].

Theorem 2.18

[41.6.1]   Let the function f:[0,[R be monotonously increasing with fx0 and f0=0, and such that D0+α,λfx with 0<α<1 and 0λ1 is also monotonously increasing on a neighbourhood 0,δ for small δ>0. [41.6.2] Let 0β<λ1-α+α, let C0 be a constant and Λx a slowly varying function for x0. [41.6.3] Then

limx0fxxβΛx=C(2.119)

holds if and only if

limx0D0+α,λfxxβ-αΛx=CΓβ+1Γβ-α+1(2.120)

holds.

[page 42, §1]

[42.1.1] A function f is called slowly varying at infinity if limxfbx/fx=1 for all b>0. [42.1.2] A function fx is called slowly varying at aR if f1/x-a is slowly varying at infinity.

2.2.2.9 Fractional Powers of Operators

[42.2.1] The spectral decomposition of selfadjoint operators is a familiar mathematical tool from quantum mechanics [116]. [42.2.2] Let A denote a selfadjoint operator with domain DA and spectral family Eλ on a Hilbert space X with scalar product (,). [42.2.3] Then

Au,v=σAλdEλu,v(2.121)

holds for all u,vDA. [42.2.4] Here σA is the spectrum of A. [42.2.5] It is then straightforward to define the fractional power Aαu by

Aαu,u=σAλαdEλu,u(2.122)

on the domain

DAα=uX:σAλαdEλu,u<.(2.123)

[42.2.6] Similarly, for any measurable function g:σAC the operator gA is defined with an integrand gλ in eq. (2.122). [42.2.7] This yields an operator calculus that allows to perform calculations with functions instead of operators.

[42.3.1] Fractional powers of the Laplacian as the generator of the diffusion semigroup were introduced by Bochner [13] and Feller [26] based on Riesz’ fractional potentials. [42.3.2] The fractional diffusion equation

ft=--Δα/2f(2.124)

was related by Feller to the Levy stable laws [74] using one dimensional fractional integrals I-α,β of order -α and type β [26]7 (This is a footnote:) 7Fellers motivation to introduce the type β was this relation.. [42.3.3] For α=2 eq. (2.124) reduces to the diffusion equation. [42.3.4] This type of fractional diffusion will be referred to as fractional diffusion of Bochner-Levy type (see Section 2.3.4 for more discussion). [42.3.5] Later, these ideas were extended to fractional powers of closed8 (This is a footnote:) 8 An operator A:BB on a Banach space B is called closed if the set of pairs x,Ax with xDA is closed in B×B. semigroup generators [4, 5, 69, 70]. [42.3.6] If -A is the infinitesimal generator of a

[page 43, §0]    semigroup Tt (see Section 2.3.3.2 for definitions of Tt and A) on a Banach space B then its fractional power is defined as

-Aαf=limε0+1-Γ-αεt-α-11-Ttfdt(2.125)

for every fB for which the limit exists in the norm of B [120, 121, 93, 123]. [43.0.1] This aproach is clearly inspired by the Marchaud form (2.82). Alternatively, one may use the Grünwald approach to define fractional powers of semigroup generators [122, 99].

2.2.2.10 Pseudodifferential Operators

[43.1.1] The calculus of pseudodifferential operators represents another generalization of the operator calculus in Hilbert spaces. [43.1.2] It has its roots in Hadamard’s ideas [36], Riesz potentials [97], Feller’s suggestion [26] and Calderon-Zygmund singular integrals [16]. [43.1.3] Later it was generalized and became a tool for treating elliptic partial differential operators with nonconstant coefficients.

Definition 2.19

[43.2.1] A (Kohn-Nirenberg) pseudodifferential operator of order αR σx,D:SRdSRd is defined as

σx,Dfx=12πdRdeixkσx,kFfkdk(2.126)

and the function σx,k is called its symbol. [43.2.2] The symbol is in the Kohn-Nirenberg symbol class Sα if it is in CR2d, and there exists a compact set KRd such that suppσK×Rd, and for any pair of multiindices β,γ there is a constant Cβ,γ such that

DkβDxγσx,kCβ,γ1+kα-β.(2.127)

[43.2.3] The Hörmander symbol class Sρ,δα is obtained by replacing the exponent α-β on the right hand side with α-ρβ+δγ where 0ρ,δ1.

[43.3.1] Pseudodifferential operators provide a unified approach to differential and integral or convolution operators that are ‘‘nearly’’ translation invariant. [43.3.2] They have a close relation with Weyl quantization in physics [116, 28]. However, they will not be discussed further because the traditional symbol classes do not contain the usual fractional derivative operators. [43.3.3] Fractional Riesz derivatives are not pseudodifferential operators in the sense above. [43.3.4] Their symbols do not fall into any of the standard Kohn-Nirenberg or Hörmander symbol classes due to lack of differentiability at the origin.

[page 44, §1]

2.2.3 Eigenfunctions

[44.1.1] The eigenfunctions of Riemann-Liouville fractional derivatives are defined as the solutions of the fractional differential equation

D0+αfx=λfx,(2.128)

where λ is the eigenvalue. [44.1.2] They are readily identifed using eq. (A.11) as

fx=x1-αEα,αλxα,(2.129)

where

Eα,β=k=0xkΓαk+β(2.130)

is the generalized Mittag-Leffler function [125, 126]. [44.1.3] More generally the eigenvalue equation for fractional derivatives of order α and type β reads

D0+α,βfx=λfx,(2.131)

and it is solved by [54, eq.124]

fx=x1-β1-αEα,α+β1-αλxα,(2.132)
Figure 2.1: Truncated real part of the generalized Mittag-Leffler function -3ReE0.8,0.9z3 for zC with -7Rez5 and -10Imz10. The solid line is defined by ReE0.8,0.9z=0.

[page 45, §0]    where the case β=0 corresponds to (2.128). [45.0.1] A second important special case is the equation

D0+α,1fx=λfx,(2.133)

with D0+α,1=D~0+α. [45.0.2] In this case the eigenfunction

fx=Eαλxα,(2.134)

where Eαx=Eα,1x is the Mittag-Leffler function [86]. [45.0.3] The Mittag-Leffler function plays a central role in fractional calculus. [45.0.4] It has only recently been calculated numerically in the full complex plane [108, 62]. [45.0.5] Figure 2.1 and 2.2 illustrate E0.8,0.9z for a rectangular region in the complex plane (see [108]). [45.1.1] The solid line in Figure 2.1 is the line ReE0.8,0.9z=0, in Figure 2.2 it is ImE0.8,0.9z=0.

Figure 2.2: Same as Fig. 2.1 for the imaginary part of E0.8,0.9z. The solid line is ImE0.8,0.9z=0.

[45.2.1] Note, that some authors are avoiding the operator D0+α,1 in fractional differential equations (see e.g. [112, 101, 84, 111, 7, 82] or chapters in this volume). [45.2.2] In their notation the eigenvalue equation (2.133) becomes (c.f.[112, eq.(22)])

ddxfx=λD0+1-αfx(2.135)

containing two derivative operators instead of one.