Motivation
A harmonic function describes a steady state with no interior source. Finding one with prescribed boundary values depends on both the shape of the domain and the way the data are prescribed. The Poisson kernel solves an upper-half-plane problem by averaging the boundary data. Conformal maps then turn some strip and wedge problems into that model, provided the map is globally one-to-one and the boundary correspondence is checked.
A formula valid inside a domain does not automatically have the intended value at every boundary point. Corners and discontinuous boundary data deserve particular care. They can also reduce regularity: a solution may have finite energy but fail to have square-integrable second derivatives, so the assumption used in a finite-element error estimate must be verified rather than guessed from the PDE.
Definitions
Definition 4.2.1
(Upper-half-plane Poisson integral)
For put . For bounded measurable , define
The integral is absolutely convergent. A boundary value at means a limit as from ; at discontinuities we state a specific limiting direction instead of assuming such a limit exists.
The kernel is positive and has integral one. Thus is a weighted average of boundary values; in particular . The width of the weight is proportional to . Moving toward the boundary concentrates the average near .
Concept lensAnalytic
The Poisson integral as an expectation
The function is the Cauchy probability density with location and scale . If is a standard Cauchy random variable, with density , then has exactly this density, so
In this description the bound is the bound for an expectation, and the jump average comes from . For the unit step it gives , the formula computed below. The Cauchy law has no mean, so averaging bounded data is essential; a linearly growing would make the expectation diverge.
Main results
Theorem 4.2.2
(Poisson construction and boundary convergence)
For bounded measurable , its Poisson integral is bounded and harmonic on the upper half-plane. If is continuous at , then as from above. If instead has finite left and right limits at , then the vertical limit is their average.
Proof
(Unit mass, harmonic kernel and concentrating tails)
Substituting shows . Direct differentiation gives . On every compact set with bounded away from zero, the kernel derivatives through second order have integrable bounds independent of ; differentiation under the integral is therefore justified and .
For a continuity point, write . Choose so the bracket is small when . When is close enough to , the remaining region has ; its kernel mass tends to zero as , since . Boundedness controls that part. This proves the full approach limit.
For the vertical limit at a jump, split the integral at and substitute . The even kernel assigns mass one half to each half-line. Dominated convergence then gives half the right limit plus half the left limit. A vertical midpoint limit does not imply a direction-independent limit at the jump.
Proof X-ray
Three kernel properties, three conclusions
The proof uses only three properties of , and each conclusion rests on a different one. Harmonicity of comes from harmonicity of the kernel, plus integrable derivative bounds away from ; boundedness of is what makes those bounds integrable against . The limit at a continuity point uses unit mass, to write as one integral, and vanishing tail mass . The key move is the split at distance : near continuity makes the bracket small, and far away boundedness of times the tail mass is small. Because the split is centered at while the kernel is centered at the moving point , the argument gives the full two-dimensional limit, not only the vertical one. At a jump the near part no longer becomes small. Evenness of the kernel then supplies the weights one half, and only along the vertical line, where the kernel stays centered on the jump.
Theorem 4.2.3
(Uniqueness with boundedness)
If and are bounded harmonic functions on the upper half-plane, continuous up to every finite real boundary point, and have the same boundary values there, then .
Proof
(Transfer to the disk and dominated convergence)
Map the disk to the upper half-plane by and let . It is bounded and harmonic on the disk. At every point of the unit circle except , its boundary limit is zero. Write the disk kernel as for , distinct from the upper-half-plane kernel . For any , the Poisson formula on the disk applied to gives
For each fixed , the kernel is bounded in , and the integrand is bounded uniformly in . As , except possibly at the single point . Dominated convergence makes the right side zero, while continuity makes the left side tend to . Hence .
The boundedness hypothesis is not a technical convenience: the dominated-convergence step uses it, and without it the conclusion fails.
Counterexample mode
Unbounded solutions with zero boundary values
Claim: two harmonic functions on the upper half-plane that are continuous up to the real axis and agree there must be equal. Take and . Both are harmonic, since ; both extend continuously to the real axis with value ; yet . The transported difference is unbounded near , so no dominating function exists. The repaired claim is the theorem above: agreement at every finite boundary point determines a solution within the bounded class.
Worked examples
Worked example 4.2.4
(A step on the real line)
Take for and for . Integration gives
For fixed the limit as is one; for it is zero. At the vertical value is . Along it is , showing directly that no direction-independent limit exists at the discontinuity. The value assigned to does not change the integral.
Concept lensGeometric
Poisson weights as visual angles
For an interval , integrating the kernel gives
where is the angle under which the segment from to is seen from . So for data equal to one on and zero elsewhere, is that visual angle divided by . For the step, the positive half-line is seen under the angle . Approaching an interior point of the interval opens the angle to , and the value tends to one; approaching an endpoint vertically gives the angle .
Worked example 4.2.5
(A strip via the exponential map)
Let and set . Its argument lies in , so is in the upper half-plane. The inverse uses the branch with argument in ; hence the map is bijective. The lower and upper edges map respectively to the positive and negative real axes.
The harmonic function has boundary values for real and is bounded on the closed upper half-plane. With and , its pullback is
Both strip edges have boundary value . As the solution tends to one, and as it tends to zero. These limiting behaviors also follow from the map's two ends, and .
Worked example 4.2.6
(A wedge and a corner singularity)
For , , choose with , and let . Its argument lies in and its radius is , so it is a bijection to the upper half-plane. The harmonic function pulls back to : it is zero on the lower ray and one on the upper ray. It is bounded throughout the wedge, but its limit at the vertex depends on direction, as expected from incompatible edge data.
Now set and on the truncated wedge . It is the imaginary part of , vanishes on both straight sides and takes on the outer arc. For a reentrant corner , . Near zero , whose square integrates as . The Hessian has squared Frobenius norm , so its integral contains . Thus this weak-energy solution is not near the vertex, even though the side data vanish. The standard -based first-order finite-element estimate cannot be invoked for it.
Methods and reasoning
Constructing and checking the maps
For a strip, periodicity of the exponential is not an obstacle inside a strip of height : its imaginary increment under is strictly between zero and , less than a full period. If , their imaginary coordinates differ by an integer multiple of ; both lie in , so that multiple is zero and the real coordinates agree. The derivative is nonzero. The inverse logarithm with argument in confirms both injectivity and surjectivity.
For a wedge, writing without naming a branch hides an essential choice. The branch is single-valued on this wedge, and multiplication by sends its angular interval bijectively to . The derivative is nonzero inside the wedge, where . The vertex is a boundary point, not a point at which the map is asserted to be conformal. For an arbitrary power, an angular interval of width greater than in the image creates overlap; even a nonzero derivative would then not by itself ensure global injectivity.
Once a map is chosen, translate data in the correct direction: if is prescribed on a boundary piece of , the corresponding upper-half-plane data are where that inverse boundary correspondence is defined. Form using the Poisson integral, then set . Harmonicity follows from the chain-rule identity . Check the limits at finite boundary points, endpoints and infinity separately; the interior identity alone does not settle them.
Data, energy, and uniqueness
The wedge step solution has . Its squared gradient has divergent integral times the angular width. Hence it is bounded and harmonic in the wedge but does not have finite energy in a neighborhood of the vertex. There is no conflict with the finite-energy example : that example uses compatible zero data on both rays. Even compatible boundary data do not automatically ensure regularity at a reentrant corner.
For bounded continuous real-line data, the first theorem constructs a bounded harmonic function with the specified values, and the second proves uniqueness in that class. If the data jump, the Poisson integral still gives a bounded harmonic function, but the uniqueness theorem as stated does not directly cover a solution that fails to extend continuously at that jump. One may impose an appropriate boundary convergence condition and prove a separate uniqueness result; simply citing the continuous-boundary theorem would skip a hypothesis. On bounded domains with continuous boundary data, the maximum principle gives the familiar uniqueness result, provided solutions are continuous on the closure.
A normal derivative is not copied unchanged
If and the boundary is smooth with a smooth conformal extension and nonzero derivative there, the chain rule multiplies the normal derivative by the local length scale , with a positive factor when matching inward normals (or matching outward normals); a sign change occurs only if the normal conventions are opposed. A conformal bijection with a smooth boundary extension sends the inward side to the inward side. Consequently Dirichlet values are composed with the inverse boundary map, whereas Neumann data need this scale factor as well. At a vertex the required smooth extension can fail. A numerical boundary condition should therefore be transformed from its definition, not copied as the same number onto the image boundary.
What the corner calculation says about a computation
The branch restriction in the wedge example is a global statement, not merely a choice of notation. An open angular interval of width exactly one full turn can still give an injective exponential: its two endpoints are excluded, as with the plane cut along a ray and an argument between zero and one full turn. If the image interval is wider, it contains two angles separated by a full turn, so the corresponding points with the same radius coincide. Checking only that the derivative is nonzero would miss that overlap.
The two wedge examples have different causes of difficulty. For incompatible constant values on the two sides, the solution remains between zero and one everywhere, but its gradient grows inversely with distance to the vertex. Integrating its squared gradient from an inner radius to a fixed outer radius produces a logarithm of the ratio of those radii. The energy therefore grows without bound as the inner radius tends to zero, even though the function values themselves stay bounded. A mesh that rounds off the corner can produce a finite discrete energy; refining the rounding scale may make that energy grow rather than converge to a finite limit.
In the compatible zero-side example, the singularity is weaker: the gradient energy near the vertex is finite, while the second-derivative integral diverges at a reentrant corner. This is precisely why a weak solution can exist without meeting the regularity assumption used to prove a particular finite-element rate. It does not by itself predict the numerical rate on a chosen mesh. To assess a computation, specify the boundary values at the vertex, the domain approximation and the norm being measured; then use an estimate whose hypotheses cover that geometry, or report numerical refinement as evidence rather than a proof.
Common misconceptions
Common mistake
(Writing a power without fixing its branch)
A holomorphic function with nonzero derivative is only locally conformal. For on a wedge, global injectivity depends on the branch: with the image angles fill , and this interval must not be wider than one full turn . The derivative cannot detect an overlap of two sheets.
Common mistake
(Expecting a value at a corner with conflicting data)
Harmonicity inside does not determine the value at a corner where the two sides carry different data. The wedge solution tends to along the ray , so different rays give different limits and there is no value to assign at the vertex.
Common mistake
(Treating finite energy as regularity)
A finite Dirichlet integral does not imply square-integrable second derivatives. At a reentrant corner, with has finite gradient energy and an infinite Hessian integral, so an -based finite-element estimate does not apply to it.
Boundedness is a genuine uniqueness hypothesis on the half-plane, as the counterexample after the uniqueness proof shows.
Quick check
Quick check
(Why does the wedge map use the exponent ?)
Track what the power does to arguments.
Solution · Answer
With the branch , the power multiplies every argument by . It sends the wedge's angular interval exactly onto , the angular interval of the upper half-plane, while radii become .
Exercises
Quick check
(For the step-data Poisson solution, find its value on the ray and its vertical limit at the origin.)
Solution · Answer
Substitution gives . At the value is for every , so the vertical limit is .
Quick check
(Map the strip to the upper half-plane and find the inverse with its branch specified.)
Solution · Answer
maps the imaginary interval to arguments . Its inverse is with . These formulas prove bijectivity, and the derivative is nonzero.
Quick check
(On the wedge , find a harmonic function equal to zero on the lower ray and one on the upper ray. Does it have a single limit at the vertex?)
Solution · Answer
The function is . It is the pullback of under with . Along the ray it equals for every , so its limit as depends on and there is no single vertex limit.
Quick check
(For on a truncated wedge of angle , determine whether the gradient and Hessian are square integrable near the vertex.)
Solution · Answer
Here . In polar coordinates the squared gradient integrates like , which is finite. The squared Hessian integrates like , which diverges; the factor is nonzero. Thus has finite local energy but is not there.
Further study
These explicit maps do not prove the Riemann mapping theorem. General boundary extension, distributional boundary data, Neumann problems, polygonal regularity theory and adaptive finite-element error estimates require further hypotheses and analysis.