Boundary values can determine the other side
A holomorphic function on the upper half of a disk need not initially be defined below the real axis. If it extends continuously to the diameter and has real values there, complex conjugation supplies a candidate on the lower half. The remarkable point is that the candidate is holomorphic across the diameter, not just separately on either side.
Definition 3.6.1
(Upper half-disk and its diameter)
For and , put , and . The set is an interior segment of ; the circular arc is not part of the boundary data required below.
Theorem 3.6.2
(Schwarz reflection theorem)
Suppose is holomorphic on and continuous on , with for . Define
Then is holomorphic on all of and satisfies .
Proof
(Morera across the diameter)
The real boundary values ensure the two formulas meet continuously on . Away from , the lower formula is holomorphic: at a lower point , write the local Taylor series of around and conjugate its coefficients. It remains to check the diameter.
Use Morera's theorem. Let be a triangle whose closure lies in . Cut it along the real axis into its nonempty upper and lower polygonal pieces; a piece may touch the axis only at a vertex or an edge. Translate each entire piece a distance into its open half-disk, choosing smaller than the distance to . Cauchy's theorem gives zero boundary integral on each translated piece. Continuity of on makes these integrals converge to the boundary integrals of the original pieces as , so each original piece also has integral zero. Their common axis segment, if present, cancels with opposite orientations; hence . Morera now applies to the continuous function and gives holomorphy on . The reflection symmetry follows from the definition.
Proof X-ray
Continuity joins the pieces; Morera joins their derivatives
Conjugation of both the argument and the value produces a holomorphic function below the axis. Real boundary values make the pieces agree continuously. Neither fact alone proves complex differentiability on the axis; vanishing triangle integrals and Morera supply that missing step.
Corollary 3.6.3
(Uniqueness and real Taylor coefficients)
Under the hypotheses of the reflection theorem, is the only holomorphic function on that agrees with on , and its Taylor series about any has real coefficients.
Proof
(Identity theorem and conjugate coefficients)
If is holomorphic on and on , then vanishes on the nonempty open subset of the connected set , so by the identity theorem. Next write near . Since is real, on the same disk. The reflection identity says that this equals , and uniqueness of power-series coefficients gives for every .
Conversely, any holomorphic function with real Taylor coefficients about a real center is real on the portion of the real axis inside its convergence disk.
Worked examples
Worked example 3.6.4
(Recovering a polynomial)
On let . Its values on are real. For , the reflected formula gives . The theorem recovers the same polynomial below the axis.
Worked example 3.6.5
(Reflecting across a horizontal line)
Suppose is holomorphic above in a disk centered at a point on that line, continuous up to the line, and real there. Translate , apply the theorem in the -plane, then translate back. Below the line the extension is
The point is the mirror image of across . More generally, if the boundary values lie on an affine straight line in the target, first apply an affine rotation and translation sending that line to the real axis, reflect, and undo that transformation.
Worked example 3.6.6
(Purely imaginary boundary values)
Let , and on . On the values are purely imaginary, so the theorem does not apply to itself. It applies to , which is real on . Its extension below the axis is , and multiplying back by gives, for ,
Check: , so continues across . For imaginary boundary values the symmetry becomes , and the Taylor coefficients about real points are purely imaginary.
Counterexample mode
The boundary condition is doing work
The constant function is holomorphic on the upper half-disk and continuous on its diameter, but its boundary values are not real. The proposed lower formula is , so the two pieces have a jump and do not define a continuous extension. The function itself does extend holomorphically as the constant ; what fails is this particular reflection formula. Non-real boundary values do not by themselves rule out every extension.
Common mistake
(Holomorphy inside does not give continuity up to the axis)
Continuity up to the interval is also an assumption, not a consequence of holomorphy in the open half-disk. For example, is holomorphic in the upper half of a disk centered at zero but is unbounded near the boundary point zero. The reflection theorem cannot be applied across an interval containing that point.
Common mistake
(Conjugating only the argument)
The lower formula conjugates both the argument and the value. Using alone produces an antiholomorphic function: for it gives , whose real and imaginary parts and fail the Cauchy–Riemann equation at every point. Conjugating the value as well, , restores holomorphy.
Practice and solutions
Quick check
(What is the reflection of across the real axis, and what is the extension?)
Conjugate both the argument and the value.
Solution · Answer
For the reflected expression is . The extension is the polynomial itself.
Quick check
(Suppose satisfies the theorem and on a nonempty open subinterval of . What can be concluded?)
Apply the identity theorem to the extension.
Solution · Answer
The extension is holomorphic on the connected set and vanishes on a set with an accumulation point in . The identity theorem gives , hence on .
Quick check
(If is as in the theorem, what can be said about for ?)
Use the corollary on Taylor coefficients.
Solution · Answer
The Taylor coefficients of about are real, so is real. Equivalently, is the limit of the real difference quotients with real .
Quick check
(Why does not contradict the reflection theorem?)
Check each hypothesis.
Solution · Answer
Its boundary values on are not real, so the hypothesis of the theorem fails and the reflection formula is not claimed to work. The constant extension exists for a different reason.