25.04.2013 Views

On Some Discrete Equations in a Half-Space

On Some Discrete Equations in a Half-Space

On Some Discrete Equations in a Half-Space

SHOW MORE
SHOW LESS

Create successful ePaper yourself

Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.

<strong>On</strong> <strong>Some</strong> <strong>Discrete</strong> <strong>Equations</strong> <strong>in</strong> a<br />

<strong>Half</strong>-<strong>Space</strong><br />

Alexander V. Vasilyev ∗ , Vladimir B. Vasilyev ∗∗<br />

∗ Belgorod State University, Belgorod 308015, Russia (e-mail:<br />

alexvassel@gmail.com).<br />

∗∗ Lipetsk State Technical University, Lipetsk 398600, Russia (e-mail:<br />

vbv57@<strong>in</strong>box.ru)<br />

Abstract: We consider discrete multi-dimensional s<strong>in</strong>gular <strong>in</strong>tegral equation with Calderon-<br />

Zygmund kernel <strong>in</strong> a discrete half-space. Tak<strong>in</strong>g <strong>in</strong>to account the Fourier transform properties<br />

and the correspond<strong>in</strong>g properties of Calderon-Zygmund operators we study the solvability for<br />

such equations.<br />

Keywords: Calderon-Zygmund operator, discrete convolution, solvability<br />

1. INTRODUCTION<br />

We consider discrete operator generated by Calderon-<br />

Ztgmund kernel K(x), with is def<strong>in</strong>ed for functions of<br />

discrete variable uh(˜x), ˜x ∈ Z m h , where Zm h is <strong>in</strong>teger<br />

lattice (modulo h) <strong>in</strong> Rm , and correspond<strong>in</strong>g equation<br />

auh(˜x) + <br />

K(˜x − ˜y)uh(˜y)h m = vh(˜x), ˜x ∈ Z m h,+,(1)<br />

˜y∈Z m<br />

h,+<br />

<strong>in</strong> the discrete half-space Zm h,+ = {˜x ∈ Zm h<br />

uh, vh ∈ L2(Zm h,+ ) ≡ l2 h .<br />

: ˜ xm > 0} ,<br />

By def<strong>in</strong>ition we let K(0) = 0, and the symbol of operator<br />

uh(˜x) ↦→ au(˜x) + <br />

K(˜x − ˜y)uh(˜y)h m , ˜x ∈ Z m h ,<br />

˜y∈Z m<br />

h<br />

is periodic function<br />

σh(ξ) = a + <br />

˜x∈Z m<br />

h<br />

e −iξ˜x K(˜x)h m<br />

with basic cube period [−πh −1 ; πh −1 ] m .<br />

The sum <strong>in</strong> (2) is def<strong>in</strong>ed as a limit of partial sums for<br />

cubes QN<br />

lim<br />

<br />

e −iξ˜x K(˜x)h m ,<br />

N→∞<br />

˜x∈QN<br />

<br />

QN = ˜x ∈ Z m h : |˜x| ≤ N, |˜x| = max<br />

1≤k≤m |˜xk|<br />

<br />

.<br />

Rem<strong>in</strong>d the symbol of classical Calderon-Zygmund operator<br />

Mikhl<strong>in</strong> et al. (1986) is def<strong>in</strong>ed by Fourier transform<br />

of the kernel K(x) <strong>in</strong> pr<strong>in</strong>cipal value sense<br />

<br />

σ(ξ) = lim K(x)e<br />

N→∞<br />

iξx dx.<br />

ε→0<br />

ε


(1992) one can assert that (4) holds at least for cont<strong>in</strong>uous<br />

symbol σ(ξ) on sphere S m−1 and transmission condition<br />

σ(0; +1) = σ(0; −1).<br />

3. DISCRETE CONVOLUTIONS ON A HALF-AXIS<br />

A convolution of two functions f, g on a straight l<strong>in</strong>e is<br />

def<strong>in</strong>ed by the <strong>in</strong>tegral<br />

+∞<br />

(f ⋆ g)(x) = f(x − y)g(y)dy,<br />

−∞<br />

which exists always if f, g ∈ L2(R). It is a cont<strong>in</strong>ual<br />

convolution. A discrete convolution is def<strong>in</strong>ed analogously.<br />

If f, g are functions of discrete variable, i.e. sequences, then<br />

(f ⋆ g)(n) ≡ <br />

f(n − k)g(k) ≡ <br />

fn−kgk, (5)<br />

k∈Z<br />

k∈Z<br />

fk ≡ fk, g(k) ≡ gk, k ∈ Z,<br />

which exists for f, g ∈ l2.<br />

The Fourier transform of discrete function is def<strong>in</strong>ed by<br />

the formula<br />

(F f)(ξ) ≡ ˜ f(ξ) = <br />

fke −ikξ , ξ ∈ [−π, π].<br />

k∈Z<br />

Application of Fourier transform to (5) leads to the standard<br />

formula<br />

F (f ⋆ g) = ˜ f · ˜g,<br />

which implies immediately a solvability criteria for discrete<br />

convolution equation<br />

au(n) + <br />

M(n − k)u(k) = v(n), (6)<br />

k∈Z<br />

a is constant, M, v are given discrete functions, u is sought.<br />

The function a + ˜ M(ξ), ξ ∈ [−π, π], is called symbol of<br />

the equation (6).<br />

So, the equation (6) is uniquely solvable iff its symbol never<br />

vanishes, M, v ∈ l2.<br />

A situation is essentially complicated, if one considers<br />

the equation (6) not on whole lattice Z, but on Z+ =<br />

{0, 1, 2, ...},i.e.<br />

au(n) + <br />

M(n − k)u(k) = v(n), n ∈ Z+; (7)<br />

k∈Z+<br />

here the discrete function M is def<strong>in</strong>ed on whole Z, but<br />

given function v (and sought function u) on Z+ only.<br />

Let’s <strong>in</strong>troduce two projectors<br />

<br />

u(n), n ≥ 0<br />

0, n ≥ 0<br />

(P+u)(n) =<br />

(P−u)(n) =<br />

0, n < 0,<br />

u(n), n < 0,<br />

and discrete convolution operator M : u(n) ↦−→ au(n) +<br />

M(n − k)u(k), then the equation (7) can be written<br />

k∈Z+<br />

<strong>in</strong> the form<br />

P+Mu+ = f+, (8)<br />

where the functions u+(sought), f+ (given) are def<strong>in</strong>ed on<br />

Z+. It’s looked easily, that the equation (8) is equivalent<br />

from solvability viewpo<strong>in</strong>t to so-called paired equation<br />

(M1P+ + M2P−)U = F (9)<br />

on whole lattice Z, M2 = I (identity operator).<br />

Formal us<strong>in</strong>g the discrete Fourier transform leads to summation<br />

of divergence series<br />

<br />

e −ikξ , (10)<br />

k∈Z+<br />

and to destroy this divergence one adds the factor e is and<br />

then tends to limit under s → 0. This activity gives the<br />

form for the operator P+ as Fourier image.<br />

Write<br />

<br />

k∈Z+<br />

e −ikξ e iks = <br />

k∈Z+<br />

e −ik(ξ+is) = <br />

k∈Z+<br />

e −ikζ , ζ = ξ + is.<br />

The corrected series is absolutely converges, and its sum<br />

is <br />

e −ikζ = 1/2 − i/2 cot ζ/2.<br />

Thus,<br />

k∈Z+<br />

(F P+u) = 1/2ũ(ξ) − i/2 lim<br />

π<br />

s→0+<br />

−π<br />

cot<br />

ζ − τ<br />

ũ(τ)dτ.<br />

2<br />

Let’s note a similar <strong>in</strong>tegral (as a pr<strong>in</strong>cipal value) arises<br />

under summation of the series (10) by standard method<br />

(the Dirichlet kernel and transfer to limit <strong>in</strong> partial sums<br />

Edwards (1979)), and leads to the periodic variant of the<br />

Hilbert transform<br />

<br />

(Hu)(x) = v.p.<br />

π<br />

−π<br />

cot<br />

x − t<br />

2 u(t)dt.<br />

If we deal with projector P−, then the sum (9) takes the<br />

form<br />

= i/2 + i/2 cot(ζ/2),<br />

and we have the formula<br />

(F P−u) = −1/2ũ(ξ) + i/2 lim<br />

π<br />

s→0+<br />

−π<br />

cot<br />

ζ − τ<br />

ũ(τ)dτ.<br />

2<br />

4. PERIODIC RIEMANN BOUNDARY PROBLEM<br />

Let’s <strong>in</strong>troduce the function<br />

Φ(ζ) = 1<br />

π<br />

cot<br />

4πi<br />

−π<br />

ζ − t<br />

2 φ(t)dt,<br />

and assume that φ(t) satisfies the Hölder condition on<br />

[−π, π],<br />

|φ(t1) − φ(t2)| ≤ c|t1 − t2| α ,<br />

∀t1, t2 ∈ [−π, π], 0 < α ≤ 1, φ(−π) = φ(π).<br />

The limit values (s → ±0) can be calculated by transfer<br />

from [−π, π] to unit circle and apply<strong>in</strong>g classical Plemelj-<br />

Sokhotskii formulas, and then we have<br />

Theorem 1. The formulas<br />

Φ ± (ξ) = ± φ(t)<br />

2<br />

+ 1<br />

2πi v.p.<br />

π<br />

−π<br />

cot<br />

ξ − t<br />

φ(t)dt, (11)<br />

2<br />

hold, Φ ± (ξ) denotes limit values Φ ± (ζ) under s → ±0.


These formulas lead to the follow<strong>in</strong>g statement for periodic<br />

Riemann boundary problem: f<strong>in</strong>d<strong>in</strong>g two functions Φ ± (z),<br />

which are holomorphic <strong>in</strong> the half-strips<br />

Π± = {z ∈ C : z = t + is, t ∈ [−π, π], ± s > 0},<br />

and their boundary values under s → 0± satisfy l<strong>in</strong>ear<br />

relation<br />

Φ + (t) = G(t)Φ − (t) + g(t), t ∈ [−π, π],<br />

where G(t), g(t) are given functions on [−π, π].<br />

If we suppose, that G(t) ∈ C[−π, π], G(−π) = G(π), then<br />

the <strong>in</strong>dex of function G on the segment [−π, π] is called<br />

variation of the arg G(t) divided 2π, under t varies from<br />

−π to π . It is entire number denoted by κ.<br />

Theorem 2. If G(t) satisfies the Hölder condition, κ=0,<br />

then periodic Riemann boundary problem has unique<br />

solution Φ ± (t) ∈ L2[−π, π], which is constructed with the<br />

help of fuction Φ(ζ).<br />

5. EQUATIONS IN CONTINUAL CASE AND THE<br />

CLASSICAL RIEMANN BOUNDARY PROBLEM<br />

5.1 A half-axis case<br />

The reduction of (9) to so-called characteristic s<strong>in</strong>gular<br />

<strong>in</strong>tegral equation is constructed with the help of special<br />

Hilbert transform Gakhov (1990), Gohberg et al.<br />

(1992), K<strong>in</strong>g (2009)<br />

(Hu)(x) ≡ 1<br />

πi v.p.<br />

+∞<br />

u(s)<br />

s − x ds<br />

≡ 1<br />

πi<br />

lim<br />

N→+∞<br />

ε→0+<br />

<br />

(<br />

x−ε<br />

−N<br />

+<br />

−∞<br />

N<br />

x+ε<br />

) u(s)<br />

s − x ds.<br />

The properties of this operator are studied, and particularly<br />

the operator H : L2(R) → L2(R) is l<strong>in</strong>ear bounded<br />

operator, its spectra is ±1, and H 2 = I.<br />

Besides, the follow<strong>in</strong>g two operators<br />

P = 1/2(I + H), Q = 1/2(I − H)<br />

are projectors on subspace A(R) ⊂ L2(R) of functions<br />

admitt<strong>in</strong>g analytic cont<strong>in</strong>uation <strong>in</strong>to upper complex halfplane<br />

C+, and B(R) ⊂ L2(R) of functions admitt<strong>in</strong>g<br />

analytic cont<strong>in</strong>uation <strong>in</strong>to lower complex half-plane C−<br />

so that<br />

A(R) ⊕ B(R) = L2(R).<br />

The follow<strong>in</strong>g identities hold<br />

P 2 = P, Q = I − P, Q 2 = Q, P Q = QP = 0.<br />

If we will denote as above P+, P− the restriction operators<br />

on positive and negative half-axis, then it’s easily verified,<br />

that Gohberg et al. (1992)<br />

F P+ = QF, F P− = P F. (12)<br />

Further, apply<strong>in</strong>g the Fourier transform to one-dimensional<br />

equation (9) we obta<strong>in</strong> the follow<strong>in</strong>g<br />

1<br />

2 σM1(ξ)(I<br />

1<br />

− H) Ũ(ξ) +<br />

2 σM2(ξ)(I + H) Ũ(ξ) = ˜ F (ξ),<br />

where σM1 , σM2 are symbols of operators M1, M2. After<br />

summands collection one can write the full equation as<br />

follow<strong>in</strong>g<br />

σM1 (ξ) + σM2 (ξ)<br />

Ũ(ξ)+<br />

+ σM1(ξ) + σM2(ξ)<br />

v.p.<br />

2πi<br />

2<br />

+∞<br />

−∞<br />

Ũ(η)<br />

η − ξ dη = ˜ F (ξ). (13)<br />

The equation (13) is well-known <strong>in</strong> the theory of s<strong>in</strong>gular<br />

<strong>in</strong>tegral equations Gakhov (1990), and is called<br />

characteristic s<strong>in</strong>gular <strong>in</strong>tegral equation. Its solution is<br />

related closely with classical boundary Riemann problem<br />

for upper and lower half-plane C±. It is formulated by the<br />

follow<strong>in</strong>g way: f<strong>in</strong>d<strong>in</strong>g two functions Φ ± (t) def<strong>in</strong>ed on R,<br />

which admit an analytic cont<strong>in</strong>uation <strong>in</strong>to C± and satisfy<br />

on straight l<strong>in</strong>e R the l<strong>in</strong>ear relation<br />

Φ + (t) = G(t)Φ−(t) + g(t), (14)<br />

where G(t), g(t) are given on R functions. If we will denote<br />

σM1 (t) + σM2 (t) σM1 (t) − σM2 (t)<br />

a(t) = , b(t) = ,<br />

2<br />

2<br />

then we will see that equation (13) <strong>in</strong> the spase L2(R) and<br />

the problem (14) for Φ ± ∈ L2(R) are equivalent Gakhov<br />

(1990), so, the coefficient G(t) and right hand side g(t) are<br />

easily def<strong>in</strong>ed by a and b:<br />

a(t) + b(t)<br />

˜F (t)<br />

G(t) = , g(t) =<br />

a(t) − b(t) a(t) − b(t) ,<br />

and vice versa, given problem (14) corresponds to related<br />

characteristic s<strong>in</strong>gular <strong>in</strong>tegral equation (13). It was shown<br />

[3] the solvability condition for the equation (13) is determ<strong>in</strong>ed<br />

by certa<strong>in</strong> topological <strong>in</strong>variant, whis is called <strong>in</strong>dex.<br />

Let’s note <strong>in</strong> our case<br />

G(t) = σM1 (t)σ−1 (t). (15)<br />

M2<br />

We suppose the follow<strong>in</strong>g with respect to the function<br />

(15). Let’s denote R one-po<strong>in</strong>t compactification of R and<br />

assume that G(t) is cont<strong>in</strong>uous on R and never vanish<strong>in</strong>g.<br />

The variation of argument of G(t) divided 2π, when t<br />

varies from −∞ to +∞, is called <strong>in</strong>dex æ of this function.<br />

If æ = 0, then the solution of the equation (13) is unique,<br />

and it can be written by exact formula with the help of<br />

the Hilbert transform Gakhov (1990).<br />

5.2 A half-space case<br />

Let’s go back to the equation (9) assum<strong>in</strong>g M1, M2 are<br />

Calderon-Zygmund operators (like <strong>in</strong> equation (3)), and<br />

P+, P− we mean the restriction operators on half-space<br />

R m + = {x = (x1, ..., xm), ± xm > 0}.<br />

Obviously, the previous constructions will be valid with<br />

some little supplement. If we as earlier denote the Fourier<br />

tranmsform by F , then we have the follow<strong>in</strong>g relations<br />

F P+ = Qξ ′F, F P− = Pξ ′F,<br />

P = 1/2(I + Hξ ′), Q = 1/2(I − Hξ ′),<br />

where Hξ ′ is Hilbert transform on variable ξm, ξ ′ =<br />

(ξ1, ..., ξm−1) is fixed:<br />

(Hξ ′u)(ξ′ , ξm) ≡ 1<br />

πi v.p.<br />

+∞<br />

−∞<br />

u(ξ ′ , τ)<br />

dτ.<br />

τ − ξm


The equation (13) for this case will take the form of the<br />

follow<strong>in</strong>g equation with the parameter ξ ′ :<br />

σM1 (ξ′ , ξm) + σM2 (ξ′ , ξm)<br />

Ũ(ξ)+<br />

+ σM1 (ξ′ , ξm) + σM2 (ξ′ , ξm)<br />

v.p.<br />

2πi<br />

2<br />

+∞<br />

−∞<br />

Ũ(ξ ′ , η)<br />

dη =<br />

η − ξm<br />

˜ F (ξ).(16)<br />

It corresponds to Riemann boundary problem (with parameter<br />

ξ ′ also) with coefficient<br />

G(ξ ′ , ξm) = σM1(ξ ′ , ξm)σ −1<br />

M2 (ξ′ , ξm). (17)<br />

For unique solvability of the equation (16) we need the<br />

<strong>in</strong>dex of G(ξ ′ , ξm) on variable ξm is equal to 0.<br />

The symbol of the Calderon-Zygmund operator is very<br />

specific, it is a function homogeneous of order 0, i.e.<br />

<strong>in</strong>deed it is def<strong>in</strong>ed on unit sphere S m−1 . Let m ≥ 3. Fix<br />

ξ ′ ∈ S m−2 and suppose G(0, −1) = G(0, +1).Vary<strong>in</strong>g ξm<br />

from −∞ to +∞ the function G(ξ) will vary along the<br />

arc of big half-circle across south pole (0, −1) and north<br />

pole (0, +1). At the same time the symbol’s values will go<br />

along closed curve <strong>in</strong> a complex plane. These all curves<br />

for different ξ ′ will be homotopic, and therefore they have<br />

the same entire-valued <strong>in</strong>dex æ with respect to 0. The<br />

condition æ = 0 implies the unique solvability for the<br />

equation (16).<br />

6. BACK TO A DISCRETE CASE<br />

Here we return aga<strong>in</strong> to the equations <strong>in</strong> a discrete context<br />

assum<strong>in</strong>g that <strong>in</strong> equation (9) P± are restriction operators<br />

on Z m h,± , M1, M2 are discrete Calderon-Zygmund<br />

operators generated by kernels M1(x), M2(x), which are<br />

bounded <strong>in</strong> the space L2(Z m h ).<br />

For functions of discrete variable def<strong>in</strong>ed on the lattice Z m h<br />

its discrete Fourier transform is given by the formula<br />

u(˜x) ↦−→ 1<br />

(2π) m<br />

<br />

u(˜x)e −i˜x·ξ h m ≡ ũ(ξ), ξ ∈ [−π, π] m .<br />

˜x∈Z m<br />

h<br />

Such Fourier transform has the same properties as usual<br />

Fourier transform Sobolev et al. (1997).<br />

Accord<strong>in</strong>g to theorem 1 and Sec.5 we <strong>in</strong>troduce periodic<br />

analogue of the Hilbert transform on variable ξm (ξ ∈<br />

[−π, π] m , ξ ′ is fixed) by the formula<br />

(H per<br />

ξ ′ u)(ξm) = 1<br />

2πi<br />

πh −1<br />

<br />

−πh −1<br />

u(t) cot<br />

and periodic analogues of projectors (12)<br />

P per<br />

ξ ′<br />

= 1/2(I + H per<br />

ξ ′ ), Q per<br />

ξ ′<br />

h(t − ξm)<br />

dt (18)<br />

2<br />

= 1/2(I − H per<br />

ξ ′ ).<br />

Insted of the equation (16) we obta<strong>in</strong> its periodical analogue<br />

σ1,h(ξ ′ , ξm) + σ2,h(ξ ′ , ξm)<br />

Ũ(ξ)+<br />

2<br />

+ σ1,h(ξ ′ , ξm) + σ2,h(ξ ′ , ξm)<br />

×<br />

4πi<br />

×v.p.<br />

πh −1<br />

<br />

−πh −1<br />

Ũ(ξ ′ , η) cot<br />

h(η − ξm)<br />

dη =<br />

2<br />

˜ F (ξ). (19)<br />

where σ1,h, σ2,h are symbols (2) of discrete operators<br />

M1, M2. Naturally, the equation (19) will be related to<br />

correspond<strong>in</strong>g periodic Riemann bounary problem, for<br />

which its unique solvability conditions are <strong>in</strong> theorem 2.<br />

In our case it is<br />

Ind σ1,h(·, ξm)σ −1<br />

2,h (·, ξm) = 0.<br />

7. LIMIT TRANSFER FROM A DISCRETE CASE TO<br />

CONTINUOUS ONES<br />

First here we have remember that images of symbols σ and<br />

σh are the same [9]. Moreover, the <strong>in</strong>dex is entire-valued<br />

characteristic both <strong>in</strong> cont<strong>in</strong>ual case (if the transmission<br />

condition σ(0, −1) = σ(0, +1) holds)and <strong>in</strong> discrete (periodic)<br />

ones. Look<strong>in</strong>g for the variation σh(·, ξm) along arcs<br />

of big half-circles on S m−1 and tak<strong>in</strong>g <strong>in</strong>to account that<br />

lim<br />

h→0 σh(ξ) = σ(ξ), ∀ξ ∈ S m−1 ,<br />

we conclude, that for transmission condition we have<br />

Theorem 3. The equations (1) and (3) are solvable or<br />

unsolvable at the same time.<br />

REFERENCES<br />

Edwards, R.E. (1979). Fourier Series. A Modern Introduction.<br />

V. 1,2. Spr<strong>in</strong>ger-Verlag, New York-Heidelberg-<br />

Berl<strong>in</strong>.<br />

Esk<strong>in</strong>, G. (1981). Boundary Problems for Elliptic Pseudo<br />

Differential <strong>Equations</strong> . AMS, Providence, RI.<br />

Gakhov, F.D. (1990). Boundary Value Problems . Dover<br />

Publications, New York.<br />

Gohberg, I., Krupnik N. (1992). <strong>On</strong>e-Dimensional S<strong>in</strong>gular<br />

Integral <strong>Equations</strong>, Vol.1: Introduction . Birkhäuser,<br />

Basel.<br />

K<strong>in</strong>g, F.W. (2009). Hilbert Transforms, V. 1, 2. Cambridge<br />

University Press, Cambridge.<br />

Mikhl<strong>in</strong> S.G., Prossdorf S. (1986). S<strong>in</strong>gular Integral<br />

Operators . Akademie-Verlag, Berl<strong>in</strong>.<br />

Sobolev, S.L., Vaskevich V.L. (1997). The Theory of<br />

Cubature Formulas . Kluwer Academic Publishers,<br />

Dordrecht.<br />

Vasilyev V.B. (2008). <strong>Discrete</strong> convolutions and difference<br />

equations. In G.S. Ladde, N.G. Medh<strong>in</strong>, Chuang Peng,<br />

M. Sambandham (ed.), Proceed<strong>in</strong>gs of Dynamic Systems<br />

and Applications , volume 5, 474–480. Dynamic<br />

Publishers, USA.<br />

Vasilyev, V.B. (2011). Elliptic equations and boundary<br />

value problems <strong>in</strong> non-smooth doma<strong>in</strong>s. Operator<br />

Theory: Advances and Applications , 213, 105–121.

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!