Abstract
We consider the Dirichlet, Neumann and Zaremba problems for harmonic functions in a bounded plane domain with nonsmooth boundary.
Purpose
We wish to construct explicit formulas for solutions of these problems when the boundary curve belongs to one of the following three classes: sectorial curves, logarithmic spirals and spirals of power type.
Methods
To study the problem, we apply the familiar Vekua-Muskhelishvili method which consists in the use of conformal mapping of the unit disk onto the domain to pull back the problem to a boundary problem for harmonic functions in the disk. This in turn later reduces to a Toeplitz operator equation on the unit circle with symbol-bearing discontinuities of the second kind.
Results
We develop a constructive invertibility theory for Toeplitz operators and thus derive solvability conditions as well as explicit formulas for solutions.
Conclusions
Our results raise Fredholm theory for boundary value problems in domains with singularities which are not necessarily rectifiable.
Similar content being viewed by others
Avoid common mistakes on your manuscript.
Introduction
Elliptic partial differential equations are known to appear in many applied areas of mathematical physics, to name a few, mechanics of solid media, diffraction theory, hydrodynamics, gravity theory and quantum field theory.
In this paper, we focus on boundary value problems for the Laplace equation in plane domains bounded by non-smooth curves . We are primarily interested in domains in which boundaries have a finite number of singular points of the oscillating type. By this, we mean that the curve may be parametrised in a neighbourhood of a singular point z0 by z(r) = z0 + r exp(ı φ(r)) for r ∈ (0,r0], where r is the distance between z and z0 and where φ(r) is a real-valued function which tends to infinity as r → 0 or is bounded while its derivative is unbounded at r = 0. Furthermore, φ(r) and φ′(r) are allowed to tend to infinity rather quickly, and our study encompasses domains with a non-rectifiable boundary as well.
There is a huge literature devoted to boundary value problems for elliptic equations in domains with non-smooth boundary (cf.[1–3] and the references given there). In most papers, one treats piecewise smooth curves with corner points or cusps (cf.[4–9]). One paper [10] is of particular importance for it gives a characterisation of Fredholm boundary value problems in domains with weakly oscillating cuspidal edges on the boundary.
There have been essentially fewer works dealing with more complicated curves . They mostly focus on qualitative properties, such as existence, uniqueness and stability of solutions, with respect to small perturbations (see for instance [11]). The present paper deals not only with qualitative investigations of boundary value problems in domains whose boundaries strongly oscillate at singular points, but also with constructive solutions of such problems. Hence, it sheds some new light on the operator calculus which lies behind the problems.
Our results gain in interest if we realize that the general theory of elliptic boundary value problems in domains with singular points on the boundary has made no essential progress since Kondratiev wrote his seminal paper [12]. The Fredholm property is proven to be equivalent to the invertibility of certain operator-valued symbols, where the problem is as immense as the original one. In order to get rid of operator-valued symbols, one has to carefully analyse the classical problems of potential theory.
Background on Toeplitz operators
Statement of the problem
We restrict ourselves to the Dirichlet and Neumann problems for the following Laplace equation:
in a simply connected domain with boundary in the plane of variables . The boundary data are as follows:
and
on , respectively, where (∂ / ∂ ν)u means the derivative of u in the unit outward the normal vector to .
We also treat the so-called Zaremba problem (see [13]) where the solution to (1) is required to satisfy the mixed conditions on , which contain both (2) and (3). More precisely, let be a non-empty arc on , and then the mixed condition in question reads as follows:
where u0 and u1 are given functions on and , respectively. Although being model for us, problem (4) proves to be of great importance in mathematical physics.
Instead of the normal derivative on , one can consider an oblique derivative, which can be tangent to the boundary of . This gives rise to the Sturm-Liouville problems with boundary conditions having discontinuities of the first or second kind. Moreover, the boundary curve itself is allowed to bear singularities at the points of . One question that is still unanswered is whether the eigen- and root functions of such problems are complete in the space L2 in the domain (cf.[14, 15] and references therein). The Sturm-Liouville problems in domains with piecewise smooth boundary are also of great interest in multidimensional case.
General description of the method
Our approach to the study of elliptic problems in domains with non-smooth boundary goes back to at least as far as the study of Khuskivadze and Paatashvili [16, 17]. It consists in reducing the problem in to a singular integral equation on the unit circle by means of a conformal mapping of the unit disk onto . The coefficients of the singular integral equation obtained in this way fail in general to be continuous for they are intimately connected with the derivative of boundary values of the conformal mapping. This method was successfully used for solving problems in domains with piecewise smooth boundary, where the singular points are corner points or cusps (see [4–7, 18]). In this case, the coefficients of the mentioned singular integral equation have discontinuities of the first kind. Since the theory of such equations is well elaborated, one has succeeded in constructing a sufficiently complete theory of boundary value problems for a number of elliptic equations in domains with piecewise smooth boundary. Note that by now, the theory of singular integral equations (or, in other terms, the theory of Toeplitz operators) with oscillating coefficients is well elaborated, too. In particular, in a previous study [19–21], a constructive theory of normal solvability (left and right invertibility) is elaborated in the case of coefficients with rather strong discontinuities (see also the monograph [22]).
The present paper deals with main boundary value problems for the Laplace equation for three classes of boundary curves , namely sectorial curves, logarithmic and power spirals. A sectorial curve is a plane curve , such that the angle at which the tangent of intersects the real axis is a bounded function in a punctured neighbourhood of any vorticity point of (see Definition 3 for more details). As distinct from corner or cuspidal points, the angle need not possess finite one-sided limits at a singular point and may, in general, undergo discontinuities of the second kind. For example, the arc z(t) = t + ı t2 sin(1 / t), where |t| < ε is a part of the sectorial curve with singular point z(0) = 0. The main result for sectorial curves is Theorem 8 which reduces the Dirichlet problem with data at a sectorial curve to a Toeplitz operator with sectorial symbol. The theory of such operators is well understood. A logarithmic spiral is a curve of the form z(r) = r exp(ı δ lnr), where r ∈ (0, r0) and δ are fixed real number. Note that in [23], Fredholm theory was developed for potential-type operators on slowly oscillating curves, a typical example being a logarithmic spiral. In the present paper, we not only elaborate the theory of Fredholm boundary value problems with data on logarithmic spirals, but also construct explicit formulas for solution. Finally, by a power spiral, we mean a curve of the form z(r) = r exp(ı c r - 1 / δ), where δ > 0. Notice that for δ < 1, the curve is not rectifiable. However, our method allows one not only to develop a Fredholm theory for the corresponding boundary value problems, but also to obtain formulas for solutions in a closed form.
Reduction of the Dirichlet problem
The Dirichlet problem is the most frequently encountered elliptic boundary value problem. This is not only because the Dirichlet problem is of great interest in applications in electrostatics, gravity theory, incompressible fluid theory, etc., but also since it is a good model where one tests approaches to other, more complicated, problems.
Let be a simply connected, bounded domain in the plane of real variables (x,y). The boundary of is a closed Jordan curve which we denote by . Consider the Dirichlet problems (1) and (2) in with data u0 on . As usual, we introduce a complex structure in by z = x + ı y and pick a conformal mapping of the unit disk onto the domain , cf. Riemann mapping theorem. Throught the paper, we make a standing assumption on the mappings under consideration, namely
Problems (1) and (2) can then be reformulated as follows:
where and .
For 1 ≤ p < ∞, we denote by the Hardy space on the unit disk (for the properties of Hardy spaces and conformal mappings, we refer the reader to the classical book [24]). By conformal mapping , the space is transported to the so-called Hardy-Smirnov space of functions on . A holomorphic function f on is said to belong to if
where is the push-forward of the circle |ζ| = r by . It is easy to see that if and only if
It is then a familiar property of the functions of Hardy class that the function has finite non-tangential limit values almost everywhere on the unit circle .
If is a rectifiable curve, then the function is continuous on the closed unit disk , absolutely continuous on the unit circle and almost everywhere on . It follows from (7) that f(z) has finite non-tangential limit values almost everywhere on , and
If fails to be rectifiable and there is a function with nonzero non-tangential limit values almost everywhere on , then relation (7) yields that the derivative also possesses finite non-tangential limit values almost everywhere on . Moreover, for , the limit on the left-hand side of (8) exists; hence, we are able to interpret integrals over the boundary curve like that on the right-hand side.
We will also study boundary value problems in Hardy-Smirnov spaces with weights .
Set
for , where {ζ1, …, ζ n } are pairwise different points on the unit circle. Here, μ1, …, μ n are real numbers in the interval (-1 / q, 1 / p), p and q being conjugate exponents, i.e. 1 / p + 1 / q = 1. The weight functions of form (9) are called power weights. The advantage of using such weight functions lies certainly in the fact that they are holomorphic in . A holomorphic function f in is said to lie in if
It is well known that for each harmonic function u(x, y) in , there is an analytic function f(z) in whose real part is u. We therefore look for a solution u for problems (1) and (2), which have the form with . The boundary condition u = u0 is understood in the sense of non-tangential limit values of u almost everywhere on .
Definition 1
Given any Dirichlet data u 0 on of class in the sense that
we shall say that problems (1) and (2) possess a solution inif there is a harmonic function u in, such thatfor someand u = u0on.
If w ≡ 1 (i.e. all μ k vanish), then we recover the Hardy-Smirnov spaces and , respectively.
We proceed to reduce the Dirichlet problem. By the previously mentioned data, we can look for f of the following form:
for , where h+ is an analytic function of Hardy class .
By Theorem 4 in[24], the conformal mapping is bijective and continuous on the closed unit disk. Hence, the function has finite non-tangential limit values almost everywhere on , and in this way, U(ζ) = U0(ζ) is understood on the unit circle . This enables us to rewrite problem (6) in the following form:
for , where h+ is an analytic function of Hardy class . The latter problem can in turn be reformulated as follows:
for , where
can be specified within analytic functions of Hardy class Hp in the complement of the closed unit disk.
More precisely,
belongs to the Hardy class Hp in the complement of up to an additive complex constant if the functions of Hardy class Hp in are assumed to vanish at infinity.
Recall the definition of Hardy spaces Hp± on the unit circle. Let , where 1 ≤ p ≤ ∞. We parametrise the points of by ζ = exp(ı t) with t ∈ [0, 2π]. Let
be the Fourier series of h, the coefficients being
Then, h ∈ Hp+, if c j (h) = 0 for all integers j < 0, and h ∈ Hp-, if c j (h) = 0 for all integers j ≥ 0.
The functions of Hp+ are non-tangential limit values on of functions of Hardy class . The functions of Hp- are non-tangential limit values on of functions of Hardy class Hp in , which vanish at the point at infinity. Moreover, for 1 < p < ∞, the space splits into the topological direct sum Hp+ ⊕ Hp-, as is well known.
Finally, we transform the Dirichlet problem to the following:
for , where
andIt is well known from the theory of conformal mappings that
for , where is the angle at which the tangent of at the point intersects the real axis. Note that .
Now, let
stand for the singular Cauchy integral. If 1 < p < ∞, then is a bounded operator in , and the operators
prove to be continuous projections in called analytic projections. They are intimately related with the classical decomposition of into the direct sum of traces on of Hardy class Hp functions in and , respectively.
More precisely, we get
whence and .
On applying to both sides of equality (10) and taking into account that and , we get
for , where is a Toeplitz operator with symbol a on and for .
We thus arrived at the following result.
Theorem 1
-
(1)
If with is a solution of the Dirichlet problem in , then is a solution of Equation 11.
-
(2)
If h + ∈ H p + is a solution of (11) and the kernel of T(a) is zero, then the function is a solution of the Dirichlet problem in .
Proof
(1) has already been proven; it remains to show (2). Let h + ∈ H p + satisfy (11). This equality transforms to the following:
whereSince
as is easy to check, we deduce from (12) that
Applying to both sides of this equality yields the following:
Comparing this with (11), we get whence for all , for the kernel of T(a) is zero.
Equality (12) can then be rewritten as follows:
for . Since the function is real-valued, the last equality just amounts to saying that for , where
The function is harmonic in , and it has non-tangential limit values almost everywhere on , which coincide with U0(ζ). Moreover, is of weighted Hardy-Smirnov class , and is a solution of the Dirichlet problem in , as desired. □
Corollary 1
If the operator T(a)is invertible on the space Hp+and
for, then the Dirichlet problem inhas a unique solution of the following form:
with, where
Proof
Applying the operator T(a)-1 to (11) yields the following:
for all . Since both sides of the equality extend to holomorphic functions in the disk, we can set ζ = 0. By (13), we get (T(a)-11)(0) = 1; hence,
We thus conclude that the general solution of (11) has the following form:
where c is an arbitrary real constant. From (13), it follows that
is a real number. Therefore, is actually independent of c, establishing the corollary. □
Remark 1
Condition (13) is actually fulfilled in all cases to be treated in this work (see Remark 3 below).
If Equation 11 has many solutions, then we must specify among them those solutions which give rise to solutions of the Dirichlet problem in weighted Hardy-Smirnov spaces.
Factorisation of symbols
The results of this section with detailed explanations, proofs and corresponding references can be found elsewhere [25–27].
Let be the space of all essentially bounded functions on the unit circle , H∞± the Hardy spaces on which consist of the restrictions to of bounded analytic functions in and , respectively, and the space of all continuous functions on .
A bounded linear operator A on a Hilbert space H is said to be normally solvable if its range im A is closed. A normally solvable operator is called Fredholm if its kernel and cokernel are finite dimensional. In this case, the index of A is introduced as follows:
where α(A) = dim kerA and β(A) = dim coker A.
The symbol a(ζ) of a Toeplitz operator T(a) is said to admit a p -factorisation, with 1 < p < ∞, if it can be represented in the following form:
where κ is an integer number,
p and q are conjugate exponents (i.e. 1 / p + 1 / q = 1), and is a bounded operator on .
The functions a+ and a- in (14) are determined uniquely up to a constant factor. On putting the additional condition a-(∞) = 1, one determines the factorisation uniquely.
Remark 2
As proven in[27], the factorisation of the symbol a with property (15) only is also unique up to a multiplicative constant.
Theorem 2
An operator T(a) is Fredholm in the space Hp+if and only if the symbol a(ζ) admits a p -factorisation. If T(a) is Fredholm, then ind T(a) = -κ.
Theorem 3
Letand a (ζ) ≠ 0 almost everywhere on; then at least one of the numbers α(T(a)) and β(T(a)) is equal to zero.
Combining Theorems 2 and 3, we get a criterion of invertibility for Toeplitz operators.
Corollary 2
An operator T(a) is invertible on Hp+if and only if the symbol a(ζ) admits a p -factorisation with κ = 0. In this case,
Proof
If κ = 0, then α (T(a)) = β(T(a)); thus, both α(T(a)) and β(T(a)) vanish. Hence, it follows that T(a) is invertible on Hp+.
We now establish the formula for the inverse operator (T(a))-1. Let f ∈ Hp+, and then
and similarly
Here, we have used the familiar equalities and which are valid for all h- ∈ Hq- ⊕ {c} and h+ ∈ Hq+. □
Given a non-vanishing function , we denote by the winding number of the curve about the origin or the index of the origin with respect to .
Theorem 4
Suppose, then the operator T(a) is Fredholm in the space Hp+if and only if a(ζ) ≠ 0 for all. Under this condition, the index of T(a) is given by the following:
We now introduce the concept of sectoriality which is of crucial importance in this paper.
Definition 2
A functionis called p -sectorial if ess inf |a(ζ)| > 0 and if there is a real number φ0such that
for all.
A function is said to be locally p -sectorial if ess inf |a(ζ)| > 0, and for any , there is an open arc containing ζ0, such that (16) is satisfied for all ζ in the arc with some depending on ζ0. Each p -sectorial curve is obviously locally p -sectorial.
Theorem 5
-
(1)
If a(ζ) is a p -sectorial symbol, then the operator T(a) is invertible in the space H p +.
-
(2)
If a(ζ) is a locally p -sectorial symbol, then T(a) is a Fredholm operator in H p +.
Suppose h ∈ G H∞+, that is, h ∈ H∞+ and 1 / h ∈ H∞+, then the operator T(h) is invertible in Hp+. Indeed, it is easy to check that (T(h))-1 = T(1 / h). Analogously, if h ∈ G H∞-, then the operator T(h) is invertible in Hp+ and (T(h))-1 = T(1 / h).
Theorem 6
-
(1)
Let a(ζ) = h(ζ) a 0(ζ), where h ∈ G H ∞± and , and then the operator T(a) is Fredholm in the space H p+ if and only if the operator T(a 0) is Fredholm, in which case ind T(a) = ind T(a 0).
-
(2)
Let a(ζ) = c(ζ)a 0(ζ), where and , and then T(a) is Fredholm in H p + if and only if c(ζ) vanishes at no point of and T(a 0) is Fredholm, in which case
Proof
This is a straightforward consequence of Theorems 2 and 4 and Corollary 2. □
In conclusion, we give a brief summary on Toeplitz operators with symbols having discontinuities of the first kind (the reader is referred to Chapter 5 of [26]). Let P C stand for the space of all piecewise continuous functions on which have at most finitely many jumps. Suppose a(ζ) ∈ P C and ζ1, …, ζ n are the points of discontinuity of a. Given any and , we introduce a function by the following:
for , where argz ∈ (-π, π]. It is easily seen that has at most one point of discontinuity at ζ = ζ0, with jump
If a(eı 0±ζ k ) ≠ 0 for all k = 1, …, n, then there are complex numbers with the property that
so
where .
Theorem 7
Let a(ζ) ∈ P C, and then the operator T(a) is Fredholm in Hp+if and only if the following are met:
-
(1)
a(e ı 0± ζ) ≠ 0 for all ,
-
(2)
There are integer numbers κ k such that
Under conditions (1) and (2), the index of the operator T(a) in Hp+ is actually given by the following:
which is due to (17).
Remark 3
If the operator T(a) with symbol of (10) is invertible and a(ζ) admit a p -factorisation a = a+a-with
then condition (13) holds true.
Indeed, by Corollary 2, we get the following:
as desired.
Conformal reduction of Dirichlet problems
Sectorial curves
In this section, we consider a simply connected domain whose boundary is smooth away from a finite number of points. By this, it is meant that is a Jordan curve of the following form:
where is an arc with initial point zk-1 and endpoint z k which are located after each other in the positive direction on , and z n = z0. Moreover, (zk-1, z k ) is smooth for all k.
Definition 3
The curveis called p -sectorial if, for each k = 1, …, n, there is a neighbourhoodof z k onand a real number φ k , such that
where α(z) is the angle at which the tangent ofat the point z intersects the real axis.
If z k is a conical point of , then the angle at which the tangent of at z intersects the real axis has jump j k < π when z passes through z k . It follows that (19) is fulfilled at z k with a suitable φ k , if p ≥ 2, and is fulfilled if, moreover, j k < (p - 1)π, if 1 < p < 2. If z k is a cuspidal point of , then the angle has jump j k = π when z passes through z k . Hence, condition (19) is violated, i.e. cuspidal points are prohibited for sectorial curves.
Example 1
Letbe a curve parametrised in a neighbourhood of the singular point z(0) = 0 by z(t) = t + ı y(t) with |t| ≤ ε, where y(t) is a continuous function on the interval [-ε, ε] whose derivative is continuous away from zero in (-ε, ε) and bounded in this interval. The angle at which the tangent ofat the point z(t) then intersects the real axis is argz′(t) = arg(1 + ı y′(t)). Thus, for p ≥ 2, the curve z = z(t) is a part of a sectorial curve. If 1 < p < 2, then z = z(t) is a part of a sectorial curve provided that
In particular, for p ≥ 2, the curves z(t) = t + ı|t|1 + ϱ sin|t|ϱwith ϱ > 0 and z(t) = t + ı t sin ln|t|, the parameter t varying over [-ε, ε], are parts of sectorial curves.
Example 2
Consider the curve z(t) = t + ı t sint-1, where |t| ≤ ε. Here, we readily get
whence
for all δ > 0. Hence, the discontinuity at point z(0) = 0 is not of the sectorial type (the curve oscillates rapidly at 0).
Dirichlet data on sectorial curves
Theorem 8
Supposeis p -sectorial for 1 < p < ∞ and w(ζ) ≡ 1, then the Toeplitz operator T(a) corresponding to the Dirichlet problem is invertible.
Proof
Recall that the symbol of the Toeplitz operator in question is as follows:
where for .
The idea of the proof is to represent the symbol in the form a(ζ) = c(ζ) a0(ζ), where a0 is p -sectorial and is such that . To this end, we first choose a continuous branch of the function on , where for k = 1, …, n. Consider an arc (ζ1, ζ 1+) on and take the branch of such that (19) holds for k = 1. Hence, it follows that the argument of a(ζ) satisfies the following:
where
for k = 1, …, n.
We then extend to a continuous function on the arc (ζ1, ζ2). Note that the right-hand side of (20) can be written as follows:
for all 1 < p < ∞.
It is easy to see that there is an integer number j2 with the property that
where ψ2 is defined previously.
Choose the continuous branch of on such that (21) is still valid with replaced by
We now extend to a continuous function on the arc (ζ2, ζ3), and so on. On proceeding in this fashion, we get a continuous branch of on all of (ζ n , ζ1) satisfying the following:
with some integer j1.
The task is now to show that j1 = 0, so the inequality (20) actually holds with replaced by ∖{ζ1}. For this purpose, we link any two points and together by a smooth curve , such that
-
1.
is a smooth closed curve which bounds a simply connected domain .
-
2.
The angle at which the tangent of at the point z intersects the real axis satisfies (19).
Consider a conformal mapping of onto . By the very construction, holds for all z ∈ Suppose (22) is valid with j1 ≠ 0, then, in particular,
where . From this, we deduce that the function has a nonzero increment (equal to 2π j1) when the point ζ makes one turn along the unit circle starting from the point with . Hence, it follows, by the argument principle, that the function has zeros in , which contradicts the conformality of . Thus, j1 = 0 in (22).
We have thus chosen a continuous branch of the function on the set , satisfying
for all k = 1, …, n, where j k is integer and j1 = 0. This allows one to construct the desired factorisation of a(ζ).
We first define c(ζ) away from the arcs which encompass singular points ζ k of . Namely, we set
for
To define c(ζ) in any arc with k = 1, …, n, we pick an ε k > 0 small enough so that The symbol c(ζ) is then defined by the following:
if ,
if , and
if . Here,
Obviously, c(ζ) is a non-vanishing continuous function of .
From (23) it follows that . Put
for , and then
for all . Moreover, if the numbers ε1, …, ε n are small enough, then
for all k = 1, …, n. Hence, a0(ζ) is a p -sectorial symbol, which yields the desired factorisation.
By Theorem 5(1), we conclude that the Toeplitz operator T(a0) is invertible in the space Hp+. Moreover, Theorem 6(2) shows that T(a) is Fredholm of index zero. Finally, Theorem 3 implies that the operator T(a) is actually invertible, as desired. □
Corollary 1 gives the solution of the Dirichlet problem in via the inverse operator T(a)-1. If a(ζ) admits a p -factorisation, then Corollary 2 yields an explicit formula for T(a)-1. In case the boundary of is a sectorial curve, it is possible to construct a p -factorisation of a(ζ) with the help of conformal mapping .
Theorem 9
Letbe a p -sectorial curve; then the Dirichlet problem has a unique solutionwith, and this solution is of the following form:
for.
Proof
According to Theorems 8 and 2, a p -factorisation of the symbol of Toeplitz operator corresponding to the Dirichlet problem in a domain with p -sectorial boundary, if there is any, looks like a(ζ) = a+(ζ)a-(ζ). We begin with the following representation:
for , cf. (10).
In the case of p -sectorial curves, the angle α(z) is bounded, so the curve is rectifiable. By a well-known result (see for instance [24]), the derivative belongs to H1+, whence and . Comparing this with a(ζ) = a+(ζ) a-(ζ) we get the following:
By (15), the left-hand side of this equality belongs to H1+; the right-hand side, to H1- ⊕ {c}. Hence, it follows that
where c is a complex constant. The factorisation a(ζ) = a+(ζ)a-(ζ) with
satisfies (15), and which is due to Corollary 2. This establishes the theorem when combined with the formula of Corollary 1. We fill in details.
We first observe that, according to Remark 3, condition (13) is fulfilled. Hence, we may use the formula of Corollary 1. Set
for . An easy computation shows that
holds for almost all . On writing the projection as the Cauchy integral, we get the following:
for all . Since
the proof is complete. □
Dirichlet data on logarithmic spirals
Let Σ a be a horizontal half-strip of the form with a being a positive number. Consider the mapping
of the half-strip into the complex plane , where .
A direct computation shows that the mapping is conformal if and only if a < 2π cosφ.
For υ ∈ [0, a], set then the curve is a spiral. Indeed, if r = |z| and ϑ = arg z, then any point is characterised by the following:
Hence, it follows that can be described by the following equation:
where ϑ runs over (-υ cosφ, +∞), if φ ∈ (-π / 2, 0), and over (-∞, -υ cosφ), if φ ∈ (0, π / 2).
Denote by the image of Σ a by ℓ φ . This is a domain in the z -plane whose boundary is the following composite curve:
where and are given by (24) and the arc ba,φ by z = exp(-eıφı υ) with υ ∈ [0, a]. Thus, is a conformal mapping of Σ a onto which transforms the boundary of Σ a onto .
It is easily seen that is a rectifiable curve. Indeed, the arc length of can be evaluated by the following formula:
We assume for definiteness that φ∈(-π / 2,0), and then
is finite, as desired.
Our next objective is to find a conformal mapping of onto . To this end, we compose three well-known conformal mappings:
-
1.
maps the unit disk conformally onto the upper half-plane .
-
2.
maps conformally onto the complement in of the closed unit disk , the branch of being chosen according to the condition .
-
3.
maps conformally onto the strip Σ a .
In this way, we arrive at the conformal mapping of onto given by the following:
With η = η(ζ), we get the following:
for all . Note that η(ı) = -1, η(-ı) = 1 and
for all different from 1, where the equality is understood modulo entire multiples of 2π.
Since η(ζ) runs over (-∞, -1], if ζ∈(1, ı], over [-1, 1], if ζ ∈ [ı, -ı], and over [1,∞], if ζ ∈ [-ı, 1), it follows that
The second equality is due to the fact that holds for all ζ ∈ [ı, -ı].
Having disposed of this preliminary step, we introduce the Toeplitz operator T(aa,φ) with the following symbol:
cf. (10). This operator is responsible for the solvability of the Dirichlet problem in the space .
Theorem 10
The operator T(aa,φ)in Hp+with 1 < p < ∞ is Fredholm if and only if
Ifthen T(aa,φ)is invertible in the space Hp+. Ifthen ind T(aa,φ) = 1.
Proof
The following function is introduced:
which obviously belongs to G H∞+. Indeed,
and ha,φ(ζ) is continuous at each point of the arc . Let us consider the following quotient:
for .
An easy computation shows that
Hence, a0(ζ) is a P C function with discontinuity points {1, ı, -ı}. One verifies readily that
We thus conclude that a0(ζ) possesses a representation (17) with ζ1 = 1, ζ2 = ı, ζ3 = -ı and
Observe that
for all p ∈ (1, ∞). Therefore, we may apply Theorem 7 with κ2 = κ3 = 0. Since we always have
the following cases may occur:
-
1.
If
then κ1 = 0. By Theorem 7, the operator T(a0) is Fredholm, so the operator T(aa,φ) is Fredholm, too, which is due to Theorem 6(1). The inequality can be rewritten in the following form:
which just amounts to .
-
2.
If
then κ1 = -1. Since a < 2π cosφ, the left inequality is automatically fulfilled, so the entire inequality reduces to the following:
i.e.
-
3.
Obviously,
if and only if . In this case, there is no entire number κ1 with the property that
By Theorem 7, T(a0) is not Fredholm. From it follows that T(aa,φ) is not Fredholm.
We now suppose that T(a0) is Fredholm. For the symbol a0, we get then a representation (17) with ζ1 = 1, ζ2 = ı and ζ3 = -ı and specified previously. More precisely,
where . Moreover, the winding number of the cycle with respect to the origin vanishes. Hence, Theorem 7 applies to the operator T(a0); in particular, the index of T(a0) is evaluated by formula (18). To complete the proof, it suffices to use Theorem 6. □
Theorem 10 allows one to construct explicit formulas for solutions of the Dirichlet problem in domains bounded by logarithmic spirals.
Theorem 11
Letbe the conformal mapping ofontogiven by (25).
-
(1)
If then the Dirichlet problems (1) and (2) have a unique solution in given by the following:
for.
-
(2)
If then the Dirichlet problems (1) and (2) have infinitely many solutions in given by the following:
for, where c is an arbitrary complex constant.
Proof
Let By Theorem 2 ensures the existence of p -factorisation of the following form:
where a±(ζ) bear properties (15). On the other hand, we have the following:
cf. (10).
Rewrite in the following form:
for . Close to either of the points 1, ı and -ı, one derives easily the following asymptotic relations:
On taking into account the estimates a < 2π cosφ and p > (a / π) cosφ, one sees that the factorisation of aa,φ given by (27) satisfies (15). By Remark 2, equality (27) is actually a p -factorisation of aa,φ with κ = 0.
Analysis similar to that in the proof of Theorem 9 completes now the proof of part (1).
Suppose By Theorem 10, the p -factorisation of the symbol aa,φ is of the following form:
where a±(ζ) satisfy (15). Therefore, (27) fails to be a p -factorisation of this symbol. We correct it in the following manner. Set
then
is a p -factorisation of aa,φ. To show this, it suffices to establish (15) in a neighbourhood of the point ζ = 1.
From (28), it follows that
where Since a < 2π cosφ, we get the following:
whence . On the other hand, from we deduce that
whence . Similarly, we obtain the following:
Consider Equation 11:
for , cf. (11). Find all solutions of this equation and choose among them those solutions which give rise to solutions of our Dirichlet problem. In the case under study, the operator T(aa,φ) has a right inverse of the following form:
see Theorem 2.1 in [22] and elsewhere. Moreover, the general solution of (11) proves to be as follows:
where c is an arbitrary complex constant. Put ζ = 0 in (29). Since the right inverse (T(aa,φ))r - 1 maps Hp+ to functions which vanish at the origin, we immediately get the following:
Thus, (29) yields the following:
Since
is the limit value for almost all of the Cauchy integral
it follows that
for all .
Now, let . A trivial verification shows that
for any complex constant c, provided that ζ ≠ 1. Moreover, since the function u0 is real-valued, we obtain the following:
Hence, it follows that (30) is a general solution of Dirichlet problems (1) and (2), as desired. □
The inverse mapping of is given explicitly by the following:
where
The condition a < 2π cosφ implies whence for all p ≥ 2 cos2φ. In particular, the condition is satisfied if p ≥ 2 or if
Remark 4
If p = (a / π) cosφ, then the operator T(aa,φ) is not Fredholm. One can show that it has zero null space and dense range in this case. What is still lacking is an explicit description of the range.
Dirichlet data on spirals of power type
In this section, we consider the Dirichlet problem in Hardy-Smirnov spaces with weights .
Let a > 0 and γ > 0. Consider the domain in the plane of complex variable θ that is bounded by the following curves:
The boundary of is thus the composite curve , with each arc being smooth.
Given any δ > 0, we define the following:
for . This function maps onto a domain .
Consider the curve in the z -plane. Introduce the polar coordinates r = |z| and ϑ = argz, and then the parametric representation of the curve just amounts to the following:
so the equation of reduces to r = ϑ-δ with ϑ ≥ 1. In this way, we obtain what will be referred to as the power spiral.
Note that the curve is rectifiable if and only if δ > 1. Indeed, the integral
is finite if and only if δ > 1.
Theorem 12
Assume γ ∈ (0, 1 / 2]and δ > 0. If a > 0 is small enough, thenis a conformal mapping.
Proof
It is sufficient to show that the curve has no self-intersections. The arcs , and h δ (ba,γ) have no self-intersections, which is easy to check. Our next goal is to show that the arcs and do not meet each other.
Suppose
then and for some und . If z1 = z2, then a trivial verification shows that
where and . The last equality is equivalent to the couple of real equalities:
with j being an integer number. In this way, we arrive at an equation relative to the real part of , namely
If a → 0, then
On substituting these asymptotic formulas into (31), we get the following:
or
This equation is in turn equivalent to the following:
i.e.
If a is small enough, then
for any . Hence, it follows that (31) has no solutions for any j, so the equation z1 = z2 is not possible, as desired. □
We now construct a conformal mapping of the unit disk onto . To this end, we compose four well-known conformal mappings:
-
1.
maps the unit disk conformally onto the upper half-plane .
-
2.
maps conformally onto the strip Σ a .
-
3.
translates the strip Σ a horizontally.
-
4.
maps Σ a +1 conformally onto the domain .
In this way, we obtain the conformal mapping of onto given by the following:
with η = η(ζ). Recall that the continuous branches of the multi-valued functions under consideration are chosen in such a manner that lnθ = ln|θ| + ı arg θ with arg θ ∈ [0, 2π), , and more generally, with .
Introduce the function of . A direct computation yields the following:
where We rewrite this as follows:
For the analysis of the Dirichlet problem in , we employ the Hardy-Smirnov spaces with weight w(ζ) = (1 - ζ) - μ, where -1 / q < μ < 1 / p.
Theorem 13
Let γ ∈ (0, 1 / 2], δ > 0 and letbe the conformal mapping ofontogiven by (32). The Toeplitz operator T(aa,γ,δ) is then Fredholm if and only if μ ≠ 0. Moreover,
-
(1)
If then the operator T(a a,γ,δ) is invertible.
-
(2)
If then the index of T(a a,γ,δ) is equal to -1.
Proof
Consider the Toeplitz operator T(aa,γ,δ) with the following symbol:
cf. (10). Set
for . It is easily seen that (ha,γ,δ(ζ))±1 ∈ H∞+, so we introduce the following function:
of .
Note that
for all real γ. Hence, it follows that a0(ζ) ∈ P C with discontinuities at the points 1 and ±ı.
On taking into account that
with argζ ∈ (0, 2π), we readily deduce that
Hence, representation (17) holds for the symbol a0(ζ) with ζ1 = 1, ζ2 = ı, ζ3 = -ı, and
and a continuous function c(ζ), such that the winding number of the curve about the origin vanishes. Thus, according to Theorem 7, the operator T(aa,γ,δ) is Fredholm if and only if μ ≠ 0.
Suppose that -1 / q < μ < 0, then all the quantities , and belong to the following interval:
In this case, Theorem 7 applies with κ1 = κ2 = κ3 = 0. Combining Theorems 7 and 3 yields the invertibility of T(a0) in the space Hp+. By Theorem 6(1), the operator T(aa,γ,δ) is invertible in Hp+, too, for aa,γ,δ = ha,γ,δa0. According to Theorem 2, the symbol aa,γ,δ admits a p -factorisation with κ = 0. Hence, the operator T(aa,γ,δ) is invertible.
Now, let then
for all p ∈ (1, ∞). In this case, Theorem 7 applies with κ1 = 1 and κ2 = κ3 = 0, according to which the operator T(a0) in Hp+ is Fredholm and its index just amounts to -1. By Theorem 2, the symbol aa,γ,δ admits a p -factorisation with κ = 1 whence indT(aa,γ,δ) = -1. □
The advantage of our method lies in the fact that we construct explicitly the p -factorisations in question.
Theorem 14
Let γ ∈ (0, 1 / 2], δ > 0 and letbe the conformal mapping ofontogiven by (32).
-
(1)
If then the Dirichlet problems (1) and (2) have a unique solution in given by the following:
for.
-
(2)
If then the Dirichlet problems (1) and (2) have a unique solution in given by the following:
for.
Proof
Suppose
By Theorem 13, the symbol aa,γ,δ(ζ) admits a p -factorisation with κ = 0. Reasoning similar to that in the proof of Theorem 11 shows that this factorisation has the form aa,γ,δ = a+a- with
Corollary 2 now implies
Our next task is to prove that condition (13) is fulfilled. To this end, we observe that
as desired.
Thus, we can use the representation of Corollary 1. In this way we immediately obtain the following:
with . This establishes the formula of part (1).
We now assume that Put
then
is a p -factorisation of aa,γ,δ. Show that condition (15) holds. For this purpose, it suffices to consider the behaviour of the factors in a neighbourhood of the point ζ = 1.
From (33), it follows that
Since μ belongs to the interval we have the following:
whence and . Analogously, we get the following:
Consider the folllowing equation:
for , cf. (11). In the case under study, the operator T(aa,γ,δ) has no inverse operator defined on all of Hp+ (but a left inverse). We introduce the auxiliary operator . This operator is invertible, so applying its inverse to both sides of the equation, we rewrite it equivalently as follows:
that is
for .
Assume that (36) possesses a solution. On putting ζ = 0 and taking into account that
we get the following:
As previously mentioned, lies in Hq- ⊕ {c}, so is independent of . An easy computation shows that
for all whence
which is due to (36). Combining this equality with (37) gives the following:
for .
On the other hand, putting ζ = 0 in (38), we obtain the following:
Since U0(ζ) is real-valued, it follows that
which coincides with (37).
We have thus proven that each solution of (35) has necessarily form (38), i.e. the solution is unique. Moreover, the function h+(ζ) given by (38) satisfies Equation 35.
To complete the proof, it suffices to apply part (2) of Theorem 1. Equation 34 shows that the symbol aa,γ,δ admits a p -factorisation with κ = 1. By Theorem 2, the index of the Toeplitz operator T(aa,γ,δ) is equal to -1. Hence, it follows by Theorem 3 that the kernel of T(aa,γ,δ) is zero. We thus conclude that Theorem 1 is applicable.
If , then
showing part (2). □
Remark 5
In the case of μ = 0, the operator T(aa,γ,δ) fails to be Fredholm. This motivates the introduction of spaces with weights.
Conformal reduction of Neumann problems
The Neumann problem
Let be a simply connected, bounded domain in the plane of real variables (x, y). Denote by the boundary of which is a Jordan curve. Given a function u1 on , we consider the problem of finding a harmonic function u in whose outward normal derivative on coincides with u1. In this way, we obtain what has been formulated in (1) and (3).
As usual, we give the plane a complex structure by z = x + ı y and pick a conformal mapping of onto satisfying (5). We continue to use power weight functions w(ζ) introduced in (9). A function u1 on is said to belong to if
Definition 4
Given any, the Neumann problem is said to possess a solution of classif there exists an analytic function f in, such thatand the harmonic functionsatisfies (∂ / ∂ ν)u = u1on.
If the curve is rectifiable and then also . However, f no longer needs to belong to if fails to be rectifiable.
Our next concern will be to reduce the Neumann problem to Toeplitz operator equations on the unit circle. According to [4, 16], if with , then
for almost all , where β(z) is the angle at which the outward normal of at the point z intersects the real axis. Introduce the following function:
of , which is obviously analytic in the unit disk. Moreover, h+ can be specified within the Hardy space Hp+ on the unit circle, as is easy to check. Put for . Using (39), we rewrite the equation (∂ / ∂ ν)u = u1 equivalently as follows:
for , where .
Note that
for , where α(z) is the angle at which the tangent of at the point z intersects the real axis. As mentioned,
for . Hence, it follows that
where so (40) just amounts to the following:
for , with
On applying the projection to both sides of the last equality, we derive the Toeplitz equation:
on , where . We have used the equality which is easily verified.
Suppose now that h+(ζ) is a solution of Equation 42 considered in the space Hp+ and that the pair h+(ζ) and satisfies equation 40, put
then
for all , where z0 is an arbitrary fixed point of and the integral is over any path in connecting z0 and z. It is easily seen that solution u is determined uniquely up to a real constant.
Theorem 15
Let.
-
(1)
If is a solution of the Neumann problem in , then the function is a solution of Toeplitz equation (42).
-
(2)
If h + is a unique solution of Toeplitz equation (42) in H p +, then the function u given by (43) is a solution of the Neumann problem in the space .
Proof
This theorem can be proven in much the same way as Theorem 1. □
If the operator T(b) is invertible in the space Hp+, then the Neumann problem possesses a solution in if and only if (T(b)-1g+)(0) = 0. Indeed, applying T(b)-1 to both sides of (42) yields ζ h+(ζ) = (T(b)-1g+)(ζ). On substituting ζ = 0, we obtain (T(b)-1g+)(0) = 0. Conversely, if the latter condition is fulfilled, then
is of Hardy class Hp+, so (43) gives us the general solution of the Neumann problem.
Corollary 3
Suppose the symbol b(ζ)has a p -factorisation of the form b = b+b-with
then
-
(1)
For the Neumann problem to possess a solution in , it is necessary and sufficient that
-
(2)
If this condition holds, then any solution of the Neumann problem in the space has the following form:
for, whereis a fixed point and the integration is over any path connecting z0and z.
Proof
Using the p -factorisation b = b+b-, we get the following:
showing part (1) of the corollary.
Furthermore, on writing the analytic projection as the Cauchy integral for , we deduce the following:
which establishes part (2). □
Neumann data on sectorial curves
As in section ‘Dirichlet data on sectorial curves’, we consider a simply connected domain whose boundary is smooth everywhere except for a finite number of points. More precisely, is a Jordan curve of the following form:
where is an arc with initial point zk-1 and endpoint z k which are located after each other in positive direction on , and z n = z0. Moreover, (zk-1,z k ) is smooth for all k.
We first notice that the Neumann problem behaves well for sectorial boundary curves with the sectoriality index different from that for the Dirichlet problem. To make it more precise, we observe that the right-hand side of estimate (19) for the variation of the inclination of the tangent of close to a singular point z k can be written as min{p - 1, 1} π / 2 for all 1 < p < ∞. This has been referred to as p -sectoriality. The right-hand side of the estimate of variation that we might allow in the case of Neumann problem looks like min{1/(p - 1),1} π / 2 = min{q - 1, 1} π / 2, which corresponds to the q -sectoriality.
Definition 5
The curveis called q -sectorial if, for each k = 1, …, n, there is a neighbourhoodof z k onand a real number φ k , such that
where α(z) is the angle at which the tangent ofat the point z intersects the real axis.
If z k is a conical point of , then the angle at which the tangent of at z intersects the real axis has jump j k < π when z passes through z k . Hence, the estimate is fulfilled at z k with a suitable φ k , if 1 < p < 2, and is fulfilled if moreover j k < 1 / (p - 1) π, if p ≥ 2.
Theorem 16
Let the boundary ofbe a q -sectorial curve, and the Neumann problem has then a solution inif and only if the condition of Corollary 3(1) is satisfied. In this case, any solution of the problem has the form of Corollary 3(2).
Proof
With the reasoning similar to that in the proof of Theorem 8, it shows that b=c b0 where b0 is p -sectorial and is such that . Combining Theorems 5, 6 and 3, we conclude that the operator T(b) is invertible and so the symbol b admits a p -factorisation of the form b = b+b-.
On the other hand, (41) yields the following:
whence
Since and 1 / b+ ∈ Hp+, the left-hand side of this equality belongs to H1+ (see for instance [24]). In the same manner, one sees that the right-hand side belongs to H1- ⊕ {c}. Therefore,
where c is a complex constant. Thus, equality (44) is actually a p -factorisation, so Corollary 3 applies to complete the proof. □
Neumann data on logarithmic spirals
Let be a simply connected domain bounded by two logarithmic spirals and an auxiliary curve as described in the ‘Dirichlet data on logarithmic spirals’ section. According to formula (41), the Toeplitz operator corresponding to the Neumann problem in the space has the following form:
with the argument of evaluated in (26).
Since the symbol aa,φ being given before Theorem 10, we obtain the following:
with
belonging to G H∞+ and belonging to P C. Moreover, b0(ζ) has representation (17) with ζ1 = 1, ζ2 = ı and ζ3 = -ı,
and a continuous function c(ζ) which is different from zero on and satisfies .
We are now in a position to employ Theorem 3 for studying the Neumann problem in .
Theorem 17
As previously defined, the operator T(ba,φ)is Fredholm in the space Hp+if and only if
Proof
Since a < 2π cosφ, it follows that
whence
By Theorem 7, the operator T(ba,φ) is Fredholm if and only if as desired. □
Theorem 18
Assume that
then
-
(1)
The operator T(b a,φ) is invertible in H p +, and the symbol has a p -factorisation b a,φ = b + b - with
-
(2)
For the Neumann problem to possess a solution in , it is necessary and sufficient that
-
(3)
Under this condition, any solution of the Neumann problem in has the following form:
for, whereis an arbitrary fixed point and the integration is over any curve inconnecting z0and z.
Proof
Inequality (45) just amounts to saying that . In this case, Theorem 7 applies immediately to conclude that the operator T(ba,φ) is invertible in the space Hp+. The p -factorisation is now derived analogously to that in the proof of Theorem 3. □
Theorem 19
If
then T(ba,φ)is a Fredholm operator in Hp+of index -1, and the symbol has a p -factorisation ba,φ = -b+ζ b-with
Proof
Note that the assumption is equivalent to the following:
Hence, from Theorem 7, it follows that the operator T(ba,φ) in Hp+ is Fredholm and its index equals -1. The p -factorisation is established similarly to that in the proof of Theorem 10. □
We now turn to the Neumann problem provided that (46) is satisfied. In our case, Equation 42 looks like the following:
for . The operator T(b+b-) is easily seen to be invertible, and its inverse is given by the following:
Applying the inverse to both sides of (47) leads us to the following:
Thus, h+ ∈ Hp+ if and only if the right-hand side of the last equality has a second-order zero at the point ζ = 0, and in this case, the solution is unique. Since does not vanish at the origin, we get the following:
Corollary 4
Let (46) be satisfied, then the condition
for j = 0, 1, is necessary and sufficient in order that the Neumann problem might have a solution in. Under this condition, all solutions inhave the following form:
for, whereis fixed and the outer integral is over any path inconnecting z0and z.
Proof
It remains to establish the formula for solutions. The latter follows readily from the following:
for all . □
Neumann data on spirals of power type
Let be a simply connected domain bounded by two power-like spirals and an auxiliary curve as described in the ‘Dirichlet data on spirals of power type’ section. According to formula (41), the Toeplitz operator corresponding to the Neumann problem in has the following form:
where w(ζ) = (1 - ζ)-μ.
Arguments similar to those in the proof of Theorem 13 show that the symbol ba,γ,δ factorises as follows:
with
belonging to G H∞+ and b0 ∈ P C having discontinuities at the points 1 and ±ı. Moreover, b0(ζ) admits a representation (17) with ζ1 = 1, ζ2 = ı and ζ3 = -ı,
and a continuous function c(ζ) which does not vanish on the circle and satisfies . Thus, we may apply Theorem 3 to develop the Neumann problem in . The exposition is much the same as that for data on logarithmic spirals at the end of the ‘Neumann data on logarithmic spirals’ section.
Theorem 20
As defined previously, the operator T(ba,γ,δ) is Fredholm in the space Hp+if and only if μ ≠ 0.
Proof
Since , we get the following:
By Theorem 7, the operator T(ba,γ,δ) is Fredholm if and only if which is equivalent to μ ≠ 0. □
Theorem 21
Let
then
-
(1)
The operators T(b a,γ,δ)is invertible in the space H p +, and its symbol has a p -factorisation b a,γ,δ = b + b - with
-
(2)
In order that the Neumann problem might have a solution in , it is necessary and sufficient that
-
(3)
Under this condition, any solution of the Neumann problem in is of the following form:
for, whereis an arbitrary fixed point and the integration is over any curve inconnecting z0 and z.
Proof
If then
In this case, Theorem 7 implies readily that the operator T(ba,γ,δ) is invertible in the space Hp+. The p -factorisation is now derived analogously to that in the proof of Theorem 3. □
Theorem 22
If
then T(ba,γ,δ)is a Fredholm operator in Hp+of index 1, and the symbol has a p -factorisation ba,φ = -b+ζ-1b-with
Proof
If then
Hence, from Theorem 7 it follows that the operator T(ba,γ,δ) in Hp+ is Fredholm and its index equals 1. By Theorem 2, the symbol ba,γ,δ admits a p -factorisation ba,γ,δ = -b+ζ-1b-. Arguing as in the proof of Theorem 11, we deduce that the factors b± have the desired form. □
We now turn to the Neumann problem provided that (49) is satisfied. In our case, Equation 42 looks like the following:
for . The operator T(b+b-) is invertible, so Equation 50 has a unique solution in Hp+ given by the following:
for .
Corollary 5
Let (49) be satisfied, then for each, the Neumann problem has a solution in. The general solution of this problem has the following form:
for, whereis fixed and the outer integral is over any path inconnecting z0and z.
Proof
It suffices to use the formula , where f is determined from the following equation:
for . □
Conformal reduction of Zaremba problems
The Zaremba problem
Let be a Jordan curve in the plane and the bounded domain in whose boundary is . We assume that is smooth almost everywhere and denote by τ = τ(z) the unit tangent vector of at a point . As defined previously, by ν = ν(z), it is meant that the unit outward normal vector of at z which exists for almost all points .
Moreover, let be a non-empty open arc on . Given any functions u0 and u1 on and , respectively, we consider the problem of finding a harmonic function u in such that
cf. (4). Zaremba wrote in [13] that it was Wirtinger who pointed out to him the great practical importance of this mixed boundary problem.
Our standing assumption is that u0 has a derivative along the arc almost everywhere on , that is,
for , where α(z) is the angle at which the tangent of at z intersects the real axis. Introduce the following function:
Our next goal is to reduce the mixed boundary value problem to a Toeplitz operator equation. To this end, we notice that if , where f is an analytic function in , then the Dirichlet condition on can be rewritten equivalently up to a constant function as follows:
for almost all . On the other hand, the Neumann condition on just amounts to the following:
for almost all , where β(z) is the angle at which the outward normal of at z intersects the real axis (see (39)).
Pick a conformal mapping of the unit disk onto , such that . Introduce the following weight function:
for , where and define
On substituting these expressions into (52) and (53), we get a system of two real equations for the unknown complex-valued function h+ of Hardy class on , namely
In the section ‘The Neumann problem’, we have proven that
where On taking into account that for , we can rewrite (54) in the following form:
where for , with
and
Multiplying the first equation by ı, we reduce system (54) finally to the following single equation:
on , where
and (see (51) for the definition of u0,1).
Use the following representation of the function . Write A for the initial point of and E for the end point. Let
then
for with
The functions and are analytic in the unit disk and in the complement of , respectively, and the branches of these functions are chosen in such a way that
Notice that (56) fails to be a p -factorisation of , for condition (15) is violated.
We now rewrite Equation 55 in the following form:
Set
Applying to both sides of this equality the analytic projection leads us to the following Toeplitz equation:
on . It is quite natural to look for a solution of this equation in the Hardy space Hp+, so we assume that
These preliminary considerations suggest a functional theoretic setting to treat (1) and (4).
Definition 6
Given any Zaremba data (u0, u1) onof classin the sense that
we shall say that the Zaremba problem possesses a solution inif there is a harmonic function u in, such thatfor someand u satisfies (4) on.
If is a solution of (57), then
and
for all , where
cf. (43), , the integral is over any path in connecting the points z0 and z, and c an arbitrary real constant. Each function u of family (58) satisfies the following conditions:
but not necessarily the first condition of (4). The latter is satisfied by the one and only function of family (58). For finding the corresponding constant c, we observe that the function u0 is continuous almost everywhere on . Suppose u s is continuous up to at least one continuity point of u0, then u0(z1) = u s (z1) + c, implying c = u0(z1) - u s (z1).
Theorem 23
Let.
-
(1)
If is a solution of the Zaremba problem in , then the function is a solution of the Toeplitz equation (57) in H p +.
-
(2)
If is a unique solution of the Toeplitz equation (57) in H p +, then the function u given by (58) is a solution of the Zaremba problem in the space .
Proof
This theorem just summarises the reduction of the Zaremba problem to a Toeplitz operator equation, as stated previously. The proof is analogous to that of Theorem 1. □
If the operator T(b) is invertible in the space Hp+, then for the Zaremba problem to possess a solution in , it is necessary and sufficient that the condition
be fulfilled. Indeed, applying T(b)-1 to both sides of (57) yields the following:
On substituting ζ = 0, we obtain (59). Conversely, if the latter condition is fulfilled, then
is of Hardy class Hp+, so (58) gives us the general solution of the Zaremba problem.
Corollary 6
Assume that the symbol b(ζ)has a p -factorisation of the form b = b+b-with
then
-
(1)
For the Zaremba problem to possess a solution in , it is necessary and sufficient that
-
(2)
Under this condition, the problem has a unique solution in given by u = u s + c with
for, where, the outer integral is over any curve connecting z0and z, and c = u0(z1) - u s (z1).
Proof
Using the p -factorisation b = b+b-, we get the following:
where . Here, we have used the equality
and the fact that This proves part (1) of the corollary.
Furthermore, on writing the analytic projection as the Cauchy integral for , we deduce the following:
which establishes part (2). □
Zaremba data on sectorial curves
Corollary 6 readily applies to the Zaremba problem in domains bounded by q -sectorial curves.
Theorem 24
Letbe a q -sectorial curve and (u0, u1)Zaremba data onof class, then the Zaremba problem has a solution inif and only if
If the solvability condition is fulfilled, then the Zaremba problem has actually a unique solution in the space . This solution is of the form u = u s + c with
for , where , the outer integral is over any curve connecting z0 and z, and c = u0(z1) - u s (z1), z1 being an arbitrary point on the smooth part of .
Zaremba data on logarithmic spirals
Suppose is a simply connected domain in the plane bounded by two logarithmic spirals and an auxiliary curve as described in the ‘Dirichlet data on logarithmic spirals’ section. As defined previously, the symbol of the Toeplitz operator corresponding to the Zaremba problem in is as follows:
the argument of being given by (26).
The arguments given at the beginning of the ‘Neumann data on logarithmic spirals’ section still hold for ba,φ. To study the Zaremba problem in the domain , we employ Theorem 6. Our standing assumption on the Zaremba data (u0, u1) on the logarithmic spiral is .
Theorem 25
Suppose that
then
-
(1)
For the Zaremba problem to possess a solution in , it is necessary and sufficient that
-
(2)
If this solvability condition is fulfilled, then the Zaremba problem has actually a unique solution in the space . The solution is of the form u = u s + c with
for, where, the outer integral is over any path connecting z0and z, and c = u0(z1) - u s (z1), z1being an arbitrary point on the smooth part of.
Proof
Indeed, under the assumption of the theorem, the operator T(ba,φ) is invertible in the space Hp+, and the symbol has a p -factorisation of the form ba,φ = b+b- with
which is due to Theorem 6, we get the remaining part of the theorem. □
Theorem 26
Let
then
-
(1)
The Zaremba problem has a solution in the space if and only if
for j = 0, 1.
-
(2)
Under these conditions, the solution is unique and has the following form: u = u s + c with
for, where, the outer integral is over any path inconnecting z0and z, and c = u0(z1) - u s (z1), z1being an arbitrary point on the smooth part of.
Proof
Indeed, under the assumption of the theorem, T(ba,φ) is a Fredholm operator in Hp+ of index -1, and its symbol has a p -factorisation of the form ba,φ = -b+ζ b- with
(see Theorem 19), so Equation (57) is as follows:
for . The operator T(b+b-) is invertible. Applying the inverse to both sides of the last equality, it yields the following:
The function determined from this equality belongs to Hp+ if and only if the right-hand side of the equality has zero of multiplicity two at the point ζ = 0. Since does not vanish at the origin, familiar reasoning completes the proof. □
Remark 6
We do not consider, in which case, the operator T(ba,φ) fails to be Fredholm in Hp+.
Zaremba data on spirals of power type
Consider the Zaremba problem in a simply connected domain bounded by two power-like spirals and an auxiliary curve as described in the ‘Dirichlet data on spirals of power type’ section. The functional theoretic setting is suggested by the particular method we use for the study and consists of weighted Hardy-Smirnov spaces , where w(ζ) = (1 - ζ)-μ with -1 / q < μ < 1 / p. From what has been shown in section ‘The Zaremba problem’, it follows that the Toeplitz operator corresponding to the Zaremba problem has the following symbol:
for .
Our standing assumption on the Zaremba data (u0, u1) on power-like spirals is .
Theorem 27
If
then
-
(1)
For the Zaremba problem to possess a solution in , it is necessary and sufficient that
-
(2)
If this solvability condition is fulfilled, then the Zaremba problem has a unique solution in the space . The solution is of the form u = u s + c with
for, where, the outer integral is over any path connecting z0and z, and c = u0(z1) - u s (z1), z1being an arbitrary point on the smooth part of.
Proof
Indeed, under the assumption of the theorem, the operator T(ba,γ,δ) is invertible in the space Hp+, and the symbol has a p -factorisation of the form ba,γ,δ = b+b- with
which is due to Theorem 6 gives to us all statements of the theorem. □
Theorem 28
If
then, given any data of class, the Zaremba problem possesses a unique solution in the space. This solution is given by u = u s + c with
for, where, the outer integral is over any curve inconnecting z0and z, and c = u0(z1) - u s (z1), z1being an arbitrary point on the smooth part of.
Proof
By Theorem 22, T(ba,γ,δ) is a Fredholm operator in Hp+ of index 1, and its symbol has a p -factorisation of the form ba,φ = -b+ζ-1b- with
Thus, Equation 57 is as follows:
for . The operator T(b+b-) is invertible. Applying the inverse to both sides of the last equality yields the following unique solution:
The function defined by this equality belongs to Hp+, as is easy to check. We thus obtain , where f is a holomorphic function in the domain , satisfying the following:
for all . On arguing as in (58), we derive the desired formula for the solution u. □
Remark 7
We do not consider μ = 0, in which case, the operator T(ba,γ,δ) fails to be Fredholm in Hp+.
Conclusions
We investigated the main boundary value problems for harmonic functions in a simply connected plane domain with strong singularities on the boundary. We developed a Fredholm theory of such problems in weighted function spaces when the boundary curve is of one of the following three classes: sectorial curves, logarithmic spirals and spirals of power type. Moreover, we elaborated a constructive invertibility theory for Toeplitz operators and thus derive explicit solvability conditions as well as formulas for solutions.
References
Kondratiev VA, Landis EM: Qualitative theory of second order linear partial differential equations. Egorov, YuV, Shubin, MA (eds) Encyclopaedia of Mathematical Sciences, vol. 32, 87–192, Springer-Verlag, Berlin 1991.
Kozlov VA, Maz’ya VG, Rossmann J: Spectral Problems Associated with Corner Singularities of Solutions to Elliptic Equations. Providence: AMS; 2000.
Maz’ya V, Nazarov S, Plamenevskij B: Asymptotic Theory of Elliptic Boundary Value Problems in Singularly Perturbed Domains. vol. 1–2, Birkhäuser, Basel. 2000.
Duduchava R, Silbermann B: Boundary value problems in domains with peaks. Meth. Diff. Equ. Math. Phys 2000, 21: 3–122.
Khuskivadze G, Kokilashvili V, Paatashvili V: Boundary value problems for analytic and harmonic functions in domains with non-smooth boundary: applications to conformal mappings. Meth. Diff. Equ. Math. Phys 1998, 14: 3–195.
Khuskivadze G, Paatashvili V: Zaremba’s problem in Smirnov class of harmonic functions in domains with piecewise Lyapunov boundary. Proc. A. Razmadze Math. Inst 2007, 143: 79–94.
Khuskivadze G, Paatashvili V: Zaremba’s boundary problem in Smirnov class of harmonic functions in domains with piecewise smooth boundary. Memoirs Diff. Equat. Math. Phys 2010, 51: 73–91.
Kilpäinen T, Malý J: The Wiener test for potential estimates for quasilinear elliptic equations. Acta Math 1994, 172: 137–161. 10.1007/BF02392793
Maz’ya V, Solov’ev A: The Dirichlet problem for harmonic functions in domains with boundary of Zygmund class. Mat. Sb 1989, 180: 1211–1233.
Rabinovich V, Schulze B-W, Tarkhanov N: Boundary value problems in oscillating cuspidal wedges. Rocky Mountain J. Math 2004,34(3):73.
Keldysh MV: On the solvability and stability of the Dirichlet problem. Amer. Math. Soc. Transl 1966, 51: 1–73.
Kondratiev VA: Boundary value problems for elliptic equations in domains with conical points. Trudy Mosk. Mat. Obshch 1967, 16: 209–292.
Zaremba S: Sur un problème mixte relatif à l’équation de Laplace, Bulletin de l’Academie des sciences de Cracovie. Classe des sciences mathématiques et naturelles, série A, 313–344 1910.
Agmon A: On the eigenfunctions and on the eigenvalues of general elliptic boundary value problems. Comm. Pure Appl. Math 1962, 15: 119–147. 10.1002/cpa.3160150203
Agranovich MS: Mixed problems in a Lipschitz domain for strongly elliptic second order systems. Funct. Anal. Appl 2011,45(2):81–98. 10.1007/s10688-011-0011-z
Muskhelishvili N: Singular Integral Equations, 3rd edn.. Nauka: Moscow; 1968.
Vekua I: On a linear boundary problem of Riemann. Trans. of Razmadze Math. Inst. Tbilisi 1942, 11: 109–139.
Kokilashvili V, Paatashvili V: The Dirichlet problem for harmonic functions in domains with boundary of Zygmund classes. Georgian Math. J 2003,10(3):531–542.
Grudsky SM: Toeplitz operators and the modelling of oscillating discontinuities with the help of Blaschke product. Elschner, J, Gohberg, I, Silbermann, I (eds) Operator Theory. Advances and Applications, vol. 121, pp. 162193. Birkhuser, Basel 2001.
Grudsky SM, Dybin VB: The Riemann boundary value problem with discontinuities of almost periodic type in its coefficient. Soviet Math. Dokl 1977,18(6):1383–1387.
Grudsky SM, Dybin VB: The Riemann boundary value problem with discontinuities of almost periodic type in its coefficient in Lp(Γ,ϱ). Mat. Issled 1980, 34: 36–49.
Grudsky SM, Dybin VB: Introduction to the Theory of Toeplitz Operators with Infinite Index. Birkhäuser: Basel; 2002.
Rabinovich V: Potential type operators on curves with vorticity points. J. Anal. Appl 1999,18(4):1065–1081.
Golusin GM: Geometric Theory of Functions of Complex Variable. Providence: AMS; 1969.
Gokhberg I, Krupnik N: One-dimensional Linear Singular Integral Equations. vol. 1–2. Birkhäuser: Basel; 1992.
Böttcher A, Silbermann B: Analysis of Toeplitz Operators. Berlin: Springer-Verlag; 1990.
Litvinchuk GS, Spitkovsky I: Factorization of Measurable Matrix Functions. Birkhäuser: Basel; 1987.
Acknowledgments
The authors acknowledge the financial support of the DFG (German Research Society), grant TA 289/8-1, and the CONACYT (National Society of Science and Technology), grant 102800. We are grateful to the referees who pointed out some errors in an earlier version and made helpful comments.
Author information
Authors and Affiliations
Corresponding author
Additional information
Competing interests
Both authors declare that they have no competing interest.
Authors’ contributions
Both SG and NT worked in close connection over all range of problems considered in the paper. Both authors read and approved the final manuscript.
Rights and permissions
Open Access This article is distributed under the terms of the Creative Commons Attribution 2.0 International License (https://creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.
About this article
Cite this article
Grudsky, S., Tarkhanov, N. Conformal reduction of boundary problems for harmonic functions in a plane domain with strong singularities on the boundary. Math Sci 7, 14 (2013). https://doi.org/10.1186/2251-7456-7-14
Received:
Accepted:
Published:
DOI: https://doi.org/10.1186/2251-7456-7-14