Shannon wavelets are used to define a method for the solution of integrodifferential equations.
This method is based on (1) the Galerking method, (2) the Shannon wavelet representation, (3) the decorrelation
of the generalized Shannon sampling theorem, and (4) the definition of connection coefficients.
The Shannon sampling theorem is considered in a more general approach suitable for analysing
functions ranging in multifrequency bands. This generalization coincides with the Shannon wavelet
reconstruction of L2(ℝ) functions. Shannon wavelets are C∞-functions and their any order derivatives can be analytically defined by some kind of a finite hypergeometric series (connection coefficients).
1. Introduction
In recent years wavelets have been successfully applied to the wavelet representation of integro-differential operators, thus giving rise to the so-called wavelet solutions of PDE and integral equations. While wavelet solutions of PDEs can be easily find in a large specific literature, the wavelet representation of integro-differential operators cannot be considered completely achieved and only few papers discuss in depth this question with particular regards to methods for the integral equations. Some of them refer to the Haar wavelets [1–3] to the harmonic wavelets [4–9] and to the spline-Shannon wavelets [10–13]. These methods are mainly based on the Petrov-Galerkin method with a suitable choice of the collocation points [14]. Alternatively to the collocation method, there has been also proposed, for the solution of PDEs, the evaluation of the differential operators on the wavelet basis, thus defining the so-called connection coefficients [6, 15–21].
Wavelets [22] are localized functions which are a useful tool in many different applications: signal analysis, data compression, operator analysis, PDE solving (see, e.g., [15, 23] and references therein), vibration analysis, and solid mechanics [23]. Very often wavelets have been used only as any other kind of orthogonal functions, without taking into consideration their fundamental properties. The main feature of wavelets is, in fact, their possibility to split objects into different scale components [22, 23] according to the multiscale resolution analysis. For the L2(ℝ) functions, that is, functions with decay to infinity, wavelets give the best approximation. When the function is localized in space, that is, the bottom length of the function is within a short interval (function with a compact support), such as pulses, any other reconstruction, but wavelets, leads towards undesirable problems such as the Gibbs phenomenon when the approximation is made in the Fourier basis. Wavelets are the most expedient basis for the analysis of impulse functions (pulses) [24, 25].
Among the many families of wavelets, Shannon wavelets [17] offer some more specific advantages, which are often missing in the others. In fact, Shannon wavelets
are analytically defined;
are infinitely differentiable;
are sharply bounded in the frequency domain, thus allowing a decomposition of frequencies in narrow bands;
enjoy a generalization of the Shannon sampling theorem, which extend to all range of frequencies [17]
give rise to the connection coefficients which can be analytically defined [15–17] for any order derivatives, while for the other wavelet families they were computed only numerically and only for the lower order derivatives [18, 19, 21].
The (Shannon wavelet) connection coefficients are obtained in [17] as a finite series (for any order derivatives). In Latto's method [18, 20, 21], instead, these coefficients were obtained only (for the Daubechies wavelets) by using the inclusion axiom but in approximated form and only for the first two-order derivatives. The knowledge of the derivatives of the basis enables us to approximate a function and its derivatives and it is an expedient tool for the projection of differential operators in the numerical computation of the solution of both partial and ordinary differential equations [6, 15, 23, 26].
The wavelet reconstruction by using Shannon wavelets is also a fundamental step in the analysis of functions-operators. In fact, due to their definition Shannon wavelets are box functions in the frequency domain, thus allowing a sharp decorrelation of frequencies, which is an important feature in many physical-engineering applications. In fact, the reconstruction by Shannon wavelets ranges in multifrequency bands. Comparing with the Shannon sampling theorem where the frequency band is only one, the reconstruction by Shannon wavelets can be done for functions ranging in all frequency bands (see, e.g., [17]). The Shannon sampling theorem [27], which plays a fundamental role in signal analysis and applications, will be generalized, so that under suitable hypotheses a few set of values (samples) and a preliminary chosen Shannon wavelet basis enable us to completely represent, by the wavelet coefficients, the continuous signal and its frequencies.
The Shannon wavelet solution of an integrodifferential equation (with functions localized in space and slow decay in frequency) will be computed by using the Petrov-Galerkin method and the connection coefficients. The wavelet coefficients enable to represent the solution in the frequency domain singling out the contribution to different frequencies.
This paper is organized as follows. Section 2 deals with some preliminary remarks and properties of Shannon wavelets also in frequency domain; the reconstruction of a function is given in Section 3 together with the generalization of the Shannon sampling theorem; the error of the wavelet approximation is computed. The wavelet reconstruction of the derivatives of the basis and the connection coefficients are given in Section 4. Section 5 deals with the Shannon wavelet solution of an integrodifferential equation and an example is given at last in Section 6.
2. Shannon Wavelets
Shannon wavelets theory (see, e.g., [16, 17, 28, 29]) is based on the scaling function φ(x) (also known as sinc function)
φ(x)=sincx=defsinπxπx=eπix-e-πix2πix,
and the corresponding wavelet [16, 17, 28, 29]
From these functions a multiscale analysis [22] can be derived. The dilated and translated instances, depending on the scaling parameter n and space shift k, are
2.1. Properties of the Shannon Scaling and Wavelet Functions
By a direct computation it can be easily seen that
φk0(h)=δkh,(h,k∈ℤ),
with δkh Kroneker symbol, so that
φk0(x)=0,x=h≠k(h,k∈ℤ),ψkn(x)=0,x=2-n(k+12±13),(n∈ℕ,k∈ℤ).
It is also
limx→2-n(h+1/2)ψkn(x)=-2n/2δhk.
Thus, according to (2.5), (2.8), for each fixed scale n, we can choose a set of points x:
x∈{h}∪{2-n(h+12±13)},(n∈ℕ,h∈ℤ),
where either the scaling functions or the wavelet vanishes, but it is important to notice that when the scaling function is zero, the wavelet is not and viceversa. As we shall see later, this property will simplify the numerical methods based on collocation point.
Since they belong to L2(ℝ), both families of scaling and wavelet functions have a (slow) decay to zero; in fact, according to their definition (2.3), (2.4)
limx→±∞φkn(x)=0,limx→±∞ψkn(x)=0,
it can be also easily checked that for a fixed x0
φk+1n(x0)<φkn(x0),φk+1n(x0)φkn(x0)=2nx-k2nx-k+1<1,ψk+1n(x0)ψkn(x0)=2n+1x-2k-12n+1x-2k-3×2sin(π(2nx-k))-12sin(π(2nx-k))+1.
Since
limx→∞2n+1x-2k-12n+1x-2k-3=1,2sin(π(2nx-k))-1<2sin(π(2nx-k))+1,
it is
The maximum and minimum values of these functions can be easily computed. The maximum value of the scaling function φk0(x) can be found in correspondence of x=k
max[φk0(xM)]=1,xM=k.
The min value of φk0(x) can be computed only numerically and it is
min[φk0(x)]≅φk0(xm)=sin2π2π,xm=k-1±2.
The minimum of the wavelet ψkn(x) can be found in correspondence of the middle point of the zeroes (2.7) so that
min[ψkn(xm)]=-2n/2,xm=2-n-1(2k+1),
and the max values of ψkn(x) are
2.2. Shannon Wavelets Theory in the Fourier Domain
Let
f̂(ω)=f(x)̂=def12π∫-∞∞f(x)e-iωxdx
be the Fourier transform of the function f(x)∈L2(ℝ), and
f(x)=2π∫-∞∞f̂(ω)eiωxdω
its inverse transform.
The Fourier transform of (2.1), (2.2) gives us
φ̂(ω)=12πχ(ω+3π)={12π,-π≤ω<π0,elsewhere,
and [17]
ψ̂(ω)=12πe-iω[χ(2ω)+χ(-2ω)]
with
χ(ω)={1,2π≤ω<4π,0,elsewhere.
Analogously for the dilated and translated instances of scaling/wavelet function, in the frequency domain, it is
φ̂kn(ω)=2-n/22πe-iωk/2nχ(ω2n+3π),ψ̂kn(ω)=-2-n/22πe-iω(k+1/2)/2n[χ(ω2n-1)+χ(-ω2n-1)].
It can be seen that
χ(ω+3π)[χ(ω2n-1)+χ(-ω2n-1)]=0
so that by using the function φ̂k0(ω) and ψ̂kn(ω) there is a decorrelation into different non-overlapping frequency bands.
For each f(x)∈L2(ℝ) and g(x)∈L2(ℝ), the inner product is defined as
〈f,g〉=def∫-∞∞f(x)g(x)̅dx,
which, according to the Parseval equality, can be expressed also as
〈f,g〉=def∫-∞∞f(x)g(x)̅dx=2π∫-∞∞f̂(ω)ĝ(ω)̅dω=2π〈f̂,ĝ〉,
where the bar stands for the complex conjugate.
With respect to the inner product (2.26). The following can be shown. [16, 17]
Theorem 2.1.
Shannon wavelets are orthonormal functions, in the sense that
〈ψkn(x),ψhm(x)〉=δnmδhk,
With δnm,δhk being the Kroenecker symbols.
For the proof see [17]. Moreover we have [16, 17].
Theorem 2.2.
The translated instances of the Shannon scaling functions φkn(x), at the level n=0, are orthogonal, in the sense that
〈φk0(x),φh0(x)〉=δkh,
being φk0(x)=defφ(x-k).
See the proof in [17].
The scalar product of the (Shannon) scaling functions with respect to the corresponding wavelets is characterized by the following [16, 17].
Theorem 2.3.
The translated instances of the Shannon scaling functions φkn(x), at the level n=0, are orthogonal to the Shannon wavelets, in the sense that
〈φk0(x),ψhm(x)〉=0,m≥0,
being φk0(x)=defφ(x-k).
Proof is in [17].
3. Reconstruction of a Function by Shannon Wavelets
Let f(x)∈L2(ℝ) be a function such that for any value of the parameters n,k∈ℤ, it is
|∫-∞∞f(x)φk0(x)dx|≤Ak<∞,|∫-∞∞f(x)ψkn(x)dx|≤Bkn<∞,
and ℬ⊂L2(ℝ) the Paley-Wiener space, that is, the space of band limited functions, that is,
suppf̂⊂[-b,b],b<∞.
According to the sampling theorem (see, e.g., [27] and references therein) we have the following.
Theorem 3.1 (Shannon).
If f(x)∈L2(ℝ) and suppf̂⊂[-π,π], the series
f(x)=∑k=-∞∞αkφk0(x)
uniformly converges to f(x), and
αk=f(k).
Proof.
In order to compute the values of the coefficients we have to evaluate the series in correspondence of the integer:
f(h)=∑k=-∞∞αkφk0(h)=(2.5)∑k=-∞∞αkδkh=αh,
having taken into account (2.5).
The convergence follows from the hypotheses on f(x). In particular, the importance of the band limited frequency can be easily seen by applying the Fourier transform to (3.3):
f̂(ω)=∑k=-∞∞f(k)φ̂k0(x)=(2.24)12π∑k=-∞∞f(k)e-iωkχ(ω+3π)=12πχ(ω+3π)∑k=-∞∞f(k)e-iωk
so that
f̂(ω)={12π∑k=-∞∞f(k)e-iωk,ω∈[-π,π]0,ω∉[-π,π].
In other words, if the function is band limited (i.e., with compact support in the frequency domain), it can be completely reconstructed by a discrete Fourier series. The Fourier coefficients are the values of the function f(x) sampled at the integers.
As a generalization of the Paley-Wiener space, and in order to generalize the Shannon theorem to unbounded intervals, we define the space ℬψ⊇ℬ of functions f(x) such that the integrals
αk=def〈f(x),φk0(x)〉=(2.27)∫-∞∞f(x)φk0(x)̅dx,βkn=def〈f(x),ψkn(x)〉=(2.27)∫-∞∞f(x)ψkn(x)̅dx
exist and are finite. According to (2.26), (2.27), it is in the Fourier domain that
αk=def∫-∞∞f(x)φk0(x)dx=(14)2π〈f(x)̂,φk0(x)̂〉=2π∫-∞∞f̂(ω)φk0(ω)̅dω=(2.24)2π∫-∞∞f̂(ω)12πeiωkχ(ω+3π)dω=(2.23)∫-ππf̂(ω)eiωkdω,βkn=def∫-∞∞f(x)ψkn(x)dx=(2.27)2π〈f(x)̂,ψkn(x)̂〉=(2.24)-2π∫-∞∞f̂(ω)2-n/22πeiω(k+1/2)/2n[χ(ω2n-1)+χ(-ω2n-1)]dω=(2.23)-2-n/2[∫2nπ2n+1πf̂(ω)eiω(k+1/2)/2ndω+∫-2n+1π-2nπf̂(ω)eiω(k+1/2)/2ndω],
so that
For the unbounded interval, let us prove the following.
Theorem 3.2 (Shannon generalized theorem).
If f(x)∈Bψ⊂L2(ℝ) and suppf̂⊆ℝ, the series
f(x)=∑h=-∞∞αhφh0(x)+∑n=0∞∑k=-∞∞βknψkn(x)
converges to f(x), with αh and βkn given by (3.8) and (3.10). In particular, when suppf̂⊆[-2N+1π,2N+1π], it is
f(x)=∑h=-∞∞αhφh0(x)+∑n=0N∑k=-∞∞βknψkn(x).
Proof.
The representation (3.11) follows from the orthogonality of the scaling and Shannon wavelets (Theorems 2.1, 2.2, and 2.3). The coefficients, which exist and are finite, are given by (3.8). The convergence of the series is a consequence of the wavelet axioms.
It should be noticed that
suppf̂=[-π,π]⋃n=0,…,∞[-2n+1π,-2nπ]∪[2nπ,2n+1π],
so that for a band limited frequency signal, that is, for a signal whose frequency belongs to the band [-π,π], this theorem reduces to the Shannon sampling theorem. More in general, the representation (3.11) takes into account more frequencies ranging in different bands. In this case we have some nontrivial contributions to the series coefficients from all bands, ranging from [-2Nπ,2Nπ]:
suppf̂=[-π,π]⋃n=0,…,N[-2n+1π,-2nπ]∪[2nπ,2n+1π].
In the frequency domain, (3.11) gives
f̂(ω)=∑h=-∞∞αhφ̂h0(ω)+∑n=0∞∑k=-∞∞βknψ̂kn(ω)f̂(ω)=(2.24)12π∑h=-∞∞αhe-iωhχ(ω+3π)-12π∑n=0∞∑k=-∞∞2-n/2βkne-iω(k+1/2)/2n[χ(ω2n-1)+χ(-ω2n-1)].
That is,
f̂(ω)=12πχ(ω+3π)∑h=-∞∞αhe-iωh-12πχ(ω2n-1)∑n=0∞∑k=-∞∞2-n/2βkne-iω(k+1/2)/2n-12πχ(-ω2n-1)∑n=0∞∑k=-∞∞2-n/2βkne-iω(k+1/2)/2n.
Moreover, taking into account (2.5), (2.7), we can write (3.11) as
f(x)=∑h=-∞∞f(h)φh0(x)-∑n=0∞∑k=-∞∞2-n/2fn(2-n(k+12))ψkn(x)
with
fn(x)=def∑k=-∞∞〈f(x),ψkn(x)〉ψkn(x).
3.1. Error of the Shannon Wavelet Approximation
Let us fix an upper bound for the series of (3.11) in a such way that we can only have the approximation
f(x)≅∑h=-KKαhφh0(x)+∑n=0N∑k=-SSβknψkn(x).
This approximation can be estimated by the following
Theorem 3.3 (Error of the Shannon wavelet approximation).
The error of the approximation (3.19) is given by
|f(x)-∑h=-KKαhφh0(x)+∑n=0N∑k=-SSβknψkn(x)|≤|f(-K-1)+f(K+1)-33π[f(2-N-1(-S-12))+f(2-N-1(S+32))]|.
Proof.
The error of the approximation (3.19) is defined as
f(x)-∑h=-KKαhφh0(x)+∑n=0N∑k=-SSβknψkn(x)=∑h=-∞-K-1αhφh0(x)+∑h=K+1∞αhφh0(x)+∑n=N+1∞[∑k=-∞-S-1βknψkn(x)+∑k=S+1∞βknψkn(x)].
Concerning the first part of the r.h.s, it is
∑h=-∞-K-1αhφh0(x)+∑h=K+1∞αhφh0(x)≤maxx∈ℝ[∑h=-∞-K-1αhφh0(x)+∑h=K+1∞αhφh0(x)]=∑h=-∞-K-1αhφh0(h)+∑h=K+1∞αhφh0(h)=(2.5)∑h=-∞-K-1αh+∑h=K+1∞αh=(3.3)∑h=-∞-K-1f(h)+∑h=K+1∞f(h),
and since f(x)∈L2(ℝ) is a decreasing function,
∑h=-∞-K-1αhφh0(x)+∑h=K+1∞αhφh0(x)≤f(-K-1)+f(K+1).
Analogously, it is
∑n=N+1∞[∑k=-∞-S-1βknψkn(x)+∑k=S+1∞βknψkn(x)]≤maxx∈ℝ∑n=N+1∞[∑k=-∞-S-1βknψkn(x)+∑k=S+1∞βknψkn(x)]=(2.18)∑n=N+1∞[∑k=-∞-S-1βknψkn(2-n-1(18k+7)3)+∑k=S+1∞βknψkn(2-n-1(18k+7)3)]=∑n=N+1∞[∑k=-∞-S-1βkn2n/233π+∑k=S+1∞βkn2n/233π]=33π∑n=N+1∞2n/2[∑k=-∞-S-1βkn+∑k=S+1∞βkn]=(3.17)-33π∑n=N+1∞2n/2[∑k=-∞-S-12-n/2f(2-n(k+12))+∑k=S+1∞2-n/2f(2-n(k+12))],
so that
∑n=N+1∞[∑k=-∞-S-1βknψkn(x)+∑k=S+1∞βknψkn(x)]≤-33π[f(2-N-1(-S-12))+f(2-N-1(S+32))]
from where (3.20) follows.
4. Reconstruction of the Derivatives
Let f(x)∈L2(ℝ) and let f(x) be a differentiable function f(x)∈Cp with p sufficiently high. The reconstruction of a function f(x) given by (3.11) enables us to compute also its derivatives in terms of the wavelet decomposition:
dℓdxℓf(x)=∑h=-∞∞αhdℓdxℓφh0(x)+∑n=0∞∑k=-∞∞βkndℓdxℓψkn(x),
so that, according to (3.11), the derivatives of f(x) are known when the derivatives
dℓdxℓφh0(x),dℓdxℓψkn(x)
are given.
Indeed, in order to represent differential operators in wavelet bases, we have to compute the wavelet decomposition of the derivatives:
dℓdxℓφh0(x)=∑k=-∞∞λhk(ℓ)φk0(x),dℓdxℓψhm(x)=∑n=0∞∑k=-∞∞γ(ℓ)hkmnψkn(x),
being
λkh(ℓ)=def〈dℓdxℓφk0(x),φh0(x)〉,γ(ℓ)khnm=def〈dℓdxℓψkn(x),ψhm(x)〉
the connection coefficients [15–21, 26, 29] (or refinable integrals).
Their computation can be easily performed in the Fourier domain, thanks to the equality (2.27). In fact, in the Fourier domain the ℓ-order derivative of the (scaling) wavelet functions is
dℓdxℓφkn(x)̂=(iω)ℓφ̂kn(ω),dℓdxℓψkn(x)̂=(iω)ℓψ̂kn(ω),
and according to (2.24),
Taking into account (2.27), we can easily compute the connection coefficients in the frequency domain
λkh(ℓ)=2π〈dℓdxℓφk0(x)̂,φh0(x)̂〉,γ(ℓ)khnm=2π〈dℓdxℓψkn(x)̂,ψhm(x)̂〉
with the derivatives given by (4.6).
If we define
μ(m)=sign(m)={1,m>0,-1,m<0,0,m=0,
the following has been shown [16, 17].
Theorem 4.1.
The any order connection coefficients (4.4)1 of the Shannon scaling functions φk0(x) are
λkh(ℓ)={(-1)k-hiℓ2π∑s=1ℓℓ!πss![i(k-h)]ℓ-s+1[(-1)s-1],k≠h,iℓπℓ+12π(ℓ+1)[1+(-1)ℓ],k=h,
Analogously for the connection coefficients (4.4)2 we have the following.
Theorem 4.2.
The any order connection coefficients (4.7)2 of the Shannon scaling wavelets ψkn(x) are
γ(ℓ)khnm=δnm{iℓ(1-|μ(h-k)|)πℓ2nℓ-1ℓ+1(2ℓ+1-1)(1+(-1)ℓ)+μ(h-k)∑s=1ℓ+1(-1)[1+μ(h-k)](2ℓ-s+1)/2ℓ!iℓ-sπℓ-s(ℓ-s+1)!|h-k|s(-1)-s-2(h+k)2nℓ-s-1×{2ℓ+1[(-1)4h+s+(-1)4k+ℓ]-2s[(-1)3k+h+ℓ+(-1)3h+k+s]}πℓ2nℓ-1ℓ+1},
respectively, for ℓ≥1, and γ(0)khnm=δkhδnm.
For the proof see [17].
Theorem 4.3.
The connection coefficients are recursively given by the matrix at the lowest scale level:
γ(ℓ)khnn=2ℓ(n-1)γ(ℓ)kh11.
Moreover it is
γ(2ℓ+1)khnn=-γ(2ℓ+1)hknn,γ(2ℓ)khnn=γ(2ℓ)hknn.
If we consider a dyadic discretisation of the x-axis such that
xk=2-n(k+12),k∈ℤ
according to (2.8), the (4.3)2 at dyadic points xk=2-n(k+1/2) becomes
φkn(2-n(k+12))=21+n/2π,k∈ℤ,
from where, in xk=(k+1/2), it is
[ddxφk0(x)]x=xk=2π∑h=-∞∞λkh.
5. Wavelet Solution of the Integrodifferential Equation
Let us consider the following linear integrodifferential equation:
Adudx=B∫-∞∞k(x,y)u(y)dy+u(x)+q(x)(A,B∈ℝ),
which includes as special cases the integral equation (A=0,B≠0) and the differential equation (A≠0,B=0). When A=B=0, there is the trivial solution u(x)=-q(x).
It is assumed that the kernel is in the form:
k(x,y)=f(x)g(y),
and the given functions f(x)∈L2(ℝ), g(x)∈L2(ℝ), q(x)∈L2(ℝ), so that, according to (3.11)
f(x)=∑h=-∞∞fhφh0(x)+∑n=0∞∑k=-∞∞fknψkn(x),g(x)=∑h=-∞∞ghφh0(x)+∑n=0∞∑k=-∞∞gknψkn(x),q(x)=∑h=-∞∞qhφh0(x)+∑n=0∞∑k=-∞∞qknψkn(x),
with the wavelet coefficients fh,fkn,gh,gkn,qh,qkn given by (3.8).
The analytical solution of (5.1) can be obtained as follows.
Theorem 5.1.
The solution of (5.1), in the degenerate case (5.2), in the Fourier domain is
û(ω)=2πB〈ĝ(ω),q̂(ω)/(Aiω-1)〉(1-2πB)〈ĝ(ω),f̂(ω)/(Aiω-1)〉f̂(ω)Aiω-1+q̂(ω)Aiω-1.
Proof.
The Fourier transform of (5.1), with kernel as (5.2), is
Adudx̂=Bf(x)̂∫-∞∞g(y)u(y)dy+u(x)̂+q(x)̂,Aiωû(ω)=2πBf̂(ω)〈ĝ(ω),û(ω)〉+û(ω)+q̂(ω),û(ω)=2πBf̂(ω)(Aiω-1)〈ĝ(ω),û(ω)〉+q̂(ω)(Aiω-1),
that is,
û(ω)=2πBf̂(ω)(Aiω-1)〈ĝ(ω),û(ω)〉+q̂(ω)(Aiω-1).
By the inner product with ĝ(ω) there follows
〈ĝ(ω),û(ω)〉=2πB〈ĝ(ω),f̂(ω)(Aiω-1)〉〈ĝ(ω),û(ω)〉+〈ĝ(ω),q̂(ω)(Aiω-1)〉,
so that
〈ĝ(ω),û(ω)〉=〈ĝ(ω),q̂(ω)/(Aiω-1)〉(1-2πB)〈ĝ(ω),f̂(ω)/(Aiω-1)〉.
If we put this equation into (5.6), we get (5.4).
Although the existence of solution is proven, the computation of the Fourier transform could not be easily performed. Therefore the numerical computation is searched in the wavelet approximation.
The wavelet solution of (5.1) can be obtained as follows: it is assumed that the unknown function and its derivative can be written as
u(x)=∑h=-∞∞αhφh0(x)+∑n=0∞∑k=-∞∞βknψkn(x),dudx=∑h=-∞∞αhddxφh0(x)+∑n=0∞∑k=-∞∞ddxβknψkn(x)=(4.3)∑h=-∞∞αh∑s=-∞∞λhs′φs0(x)+∑n=0∞∑k=-∞∞βkn∑m=0∞∑s=-∞∞γ'sknmψsm(x),
and the integral can be written as
∫-∞∞g(y)u(y)dy=〈g,u〉=∑h=-∞∞αhgh+∑n=0∞∑k=-∞∞βkngkn.
There follows the system
∑h=-∞∞αh∑s=-∞∞λhs′φs0(x)+∑n=0∞∑k=-∞∞βkn∑m=0∞∑s=-∞∞γ'sknmψsm(x)=∑h=-∞∞αhφh0(x)+∑n=0∞∑k=-∞∞βknψkn(x)+[∑h=-∞∞αhgh+∑n=0∞∑k=-∞∞βkngkn][∑h=-∞∞fhφh0(x)+∑n=0∞∑k=-∞∞fknψkn(x)]+∑h=-∞∞qhφh0(x)+∑n=0∞∑k=-∞∞qknψkn(x),
and, according to the definition of the connection coefficients,
∑h=-∞∞αh∑s=-∞∞λhs′φs0(x)+∑n=0∞∑k=-∞∞∑s=-∞∞βknγ'sknnψsn(x)=∑h=-∞∞αhφh0(x)+∑n=0∞∑k=-∞∞βknψkn(x)+[∑h=-∞∞αhgh+∑n=0∞∑k=-∞∞βkngkn][∑h=-∞∞fhφh0(x)+∑n=0∞∑k=-∞∞fknψkn(x)]+∑h=-∞∞qhφh0(x)+∑n=0∞∑k=-∞∞qknψkn(x).
By the inner product and taking into account the orthogonality conditions (Theorems 2.1, 2.2, and 2.3) it is
∑h=-∞∞αhλhk′=αk+[∑h=-∞∞αhgh+∑n=0∞∑h=-∞∞βhnghn]fk+qk,
or
∑n=0∞∑k=-∞∞βknγ'krnj=βrj+[∑h=-∞∞αhgh+∑n=0∞∑k=-∞∞βkngkn]frj+qrj
or, according to (4.11), and rearranging the indices
∑h=-∞∞βhn(γ′hknn-δhk)-fkn∑m=0∞∑h=-∞∞βhmghm=fkn∑h=-∞∞αhgh+qkn.
Thus the solution of (5.1) is (5.9)1 with the wavelet coefficients given by the algebraic system
∑h=-∞∞(λhk′-δhk-ghfk)αh=[∑n=0∞∑h=-∞∞βhnghn]fk+qk(k∈ℤ),∑h=-∞∞βhn(γ'hknn-δhk)-fkn∑m=0∞∑h=-∞∞βhmghm=fkn∑h=-∞∞αhgh+qkn(n∈ℕ,k∈ℤ)
and up to a fixed scale of approximation N,S:
dudx=∫-∞∞e-x2-|y|u(y)dy-x|x|u(x)-e-x2
with the condition
u(0)=1.
The analytical solution, as can be directly checked, is
u(x)=e-|x|.
Since
f(x)=e-x2,g(x)=e-|x|,q(x)=-e-x2
belong to L2(ℝ), let us find the wavelet approximation by assuming that also u(x) belongs to L2(ℝ), so that they can be represented according to (5.3), (5.9).
At the level of approximation N=0,S=0, from (5.3) we have
f(x)=e-x2≅0.97φ00(x),g(x)=e-|x|≅0.80φ00(x)+0.04ψ00(x),q(x)=-e-x2≅-0.97φ00(x),
so that
f0=0.97,f00=0,g0=0.80,g00=0.04,q0=-0.97,q00=0.
System (5.18) becomes
(λ00′-δ00-g0f0)α0=β00g00f0+q0,β00(γ′0000-δ00)-f00β00g00=f00α0g0+q00,
and, since λ00′=0 and γ'0000=0, according to (6.6) we have
-1-0.80×0.97α0=-0.97,-β00=0,
whose solution is
α0=0.548,β00=0,
so that
u(x)≅0.548φ00(x).
As expected, the approximation is very row (Figure 1(a)); in fact in order to get a satisfactory approximation we have to solve system (5.18) at least at the levels N=0,S=5 as shown in Figure 1(b).
Wavelet approximations (shaded) of the analytical solution (plain) of (6.1) obtained by solving (5.17).
7. Conclusion
In this paper the theory of Shannon wavelets combined with the connection coefficients methods and the Petrov-Galerkin method has been used to find the wavelet approximation of integrodifferential equations. Among the main advantages there is the decorrelation of frequencies, in the sense that the differential operator is splitted into its different frequency bands.
ShimH.-T.Park C.-H.An approximate solution of an integral equation by wavelets2005171-2-3709717LepikU.Numerical solution of evolution equations by the Haar wavelet method2007185169570410.1016/j.amc.2006.07.077MR2297840ZBL1110.65097LepikU.Solving fractional integral equations by the Haar wavelet method2009214246847810.1016/j.amc.2009.04.015MR2541683ZBL1170.65106CattaniC.KudreykoA.Application of periodized harmonic wavelets towards solution of egenvalue problems for integral equations20102010857013610.1155/2010/570136CattaniC.KudreykoA.Harmonic wavelet method towards solution of the Fredholm type integral equations of the second kind20102151241644171MuniandyS. V.MorozI. M.Galerkin modelling of the Burgers equation using harmonic wavelets1997235435235610.1016/S0375-9601(97)00639-7MR1482215ZBL1044.65511NewlandD. E.Harmonic wavelet analysis199344320322210.1098/rspa.1993.0140ZBL0793.42020XiaoJ.-Y.WenL.-H.ZhangD.Solving second kind Fredholm integral equations by periodic wavelet Galerkin method2006175150851810.1016/j.amc.2005.07.049MR2216359ZBL1088.65122YousefiS.BanifatemiA.Numerical solution of Fredholm integral equations by using CAS wavelets2006183145846310.1016/j.amc.2006.05.081MR2286207ZBL1109.65121MahmoudiY.Wavelet Galerkin method for numerical solution of nonlinear integral equation200516721119112910.1016/j.amc.2004.08.004MR2169755ZBL1082.65596MaleknejadK.LotfiT.Expansion method for linear integral equations by cardinal B-spline wavelet and Shannon wavelet as bases for obtain Galerkin system2006175134735510.1016/j.amc.2005.07.059MR2216346ZBL1088.65117MaleknejadK.RabbaniM.AghazadehN.KaramiM.A wavelet Petrov-Galerkin method for solving integro-differential equations20098691572159010.1080/00207160801923056MR2554602ZBL1170.65337MohsenA.El-GamelM.A sinc-collocation method for the linear Fredholm integro-differential equations200758338039010.1007/s00033-006-5124-5MR2320222ZBL1116.65131BellomoN.LodsB.RevelliR.RidolfiL.2008Boston, Mass, USABirkhäuserxii+194Modeling and Simulation in Science, Engineering and Technology10.1007/978-0-8176-4610-3MR2355054CattaniC.Harmonic wavelets towards the solution of nonlinear PDE2005508-91191121010.1016/j.camwa.2005.07.001MR2175583ZBL1118.65133CattaniC.Connection coefficients of Shannon wavelets2006112117132MR2231204ZBL1117.65179CattaniC.Shannon wavelets theory200820082416480810.1155/2008/164808MR2471410ZBL1162.42314LattoA.ResnikoffH. L.TenenbaumE.MadayY.The evaluation of connection coefficients of compactly supported waveletsProceedings of the French-USA Workshop on Wavelets and TurbulenceJune 1992New York, NY, USASpringer7689LinE. B.ZhouX.Connection coefficients on an interval and wavelet solutions of Burgers equation20011351637810.1016/S0377-0427(00)00562-8MR1854444ZBL0990.65096RestrepoJ.LeafG. K.Wavelet-Galerkin discretization of hyperbolic equations1995122111812810.1006/jcph.1995.1201MR1358525ZBL0838.65096RomineC. H.PeytonB. W.Computing connection coefficients of compactly supported wavelets on bounded intervals1997ORNL/TM-13413Computer Science and Mathematical Division, Mathematical Sciences Section, Oak Ridge National LaboratoryOak Ridgehttp://citeseer.ist.psu.edu/romine97computing.htmlDaubechiesI.199261Philadelphia, Pa, USASIAMxx+357CBMS-NSF Regional Conference Series in Applied MathematicsMR1162107CattaniC.RushchitskyJ.200774SingaporeWorld Scientificxii+458Series on Advances in Mathematics for Applied SciencesMR2353375BakhoumE.TomaC.Mathematical transform of travelling-wave equations and phase aspects of quantum interaction201020101569520810.1155/2010/695208TomaG.Specific differential equations for generating pulse sequences201020101132481810.1155/2010/324818CattaniC.Harmonic wavelet solutions of the schrödinger equation20035110UnserS.Sampling-50 years after Shannon200088456958710.1109/5.843002CattaniC.ShiY.Shannon wavelet analysis4488Proceedings of the International Conference on Computational Science (ICCS '07)May 2007Beijing, ChinaSpringer982989Lecture Notes in Computer Science Part IIDeriazE.Shannon wavelet approximation of linear differential operators2007676