A separated harmonic mode on a rectangle is easy to write down. A boundary temperature is usually a function rather than one mode, so we must also ask whether the sum converges, whether it satisfies the equation, and what happens at the corners. We work under a concrete sufficient smoothness assumption and prove each of these steps.
One side prescribed
Let and .
Definition 5.3.1
(The one-side Dirichlet problem)
Given with , find a real function with in and
The endpoint values are not optional: the top and vertical sides meet at the same corners, and a continuous function must have one value at each corner.
For a separated solution , the zero vertical sides lead to with , . The zero bottom value selects . Normalizing at the top gives the mode
A general boundary function suggests
The coefficients come from sine orthogonality, but a formal expansion alone is not yet a solution.
Theorem 5.3.2
(A classical interior solution with continuous boundary values)
Under the assumptions above, (3) converges uniformly on , defines a continuous function there, is twice continuously differentiable and harmonic in , and takes all four prescribed boundary values. It is the unique harmonic function in with those boundary values.
Proof
(Convergence, the equation and uniqueness)
Convergence and the boundary. Since vanishes at the endpoints, two integrations by parts yield
Thus for a constant depending on and . For ,
The Weierstrass M-test makes (3) uniformly convergent on the closed rectangle. It is therefore continuous, and the bottom and side values are zero term by term. At the top, the uniformly convergent sine series has the same sine coefficients as . Completeness of the sine system gives equality in ; continuity makes it pointwise equality on . This proves (1), including the corners. The hypothesis is sufficient for this argument, not claimed to be optimal.
The equation. On any region , , (5) gives exponential decay in . The first and second -derivatives have factors and . The -derivatives have the same exponential control, using for : for example
The corresponding differentiated series converge uniformly on such a region. Since the original series converges and its differentiated series converge uniformly on compact interior neighborhoods, the standard termwise differentiation theorem applies. Since , each term has zero Laplacian, so the sum is harmonic in . We have not asserted that second derivatives extend continuously through every corner.
Uniqueness. If two solutions have the same boundary values, their difference is harmonic and zero on the boundary. For , has Laplacian and therefore cannot have an interior maximum, so it is bounded by its boundary maximum ; letting gives . Applying the same argument to proves .
Proof X-ray
Two convergence tasks, not one
The proof uses two different sources of decay, and they do different jobs. On the closed rectangle only the coefficients help, since : the hypotheses and enter exactly once, in the integration by parts (4), and give the summable bound that yields continuity up to the top and at the corners. Away from the top the factor in (5) takes over, and it beats any power produced by differentiating twice; this is why harmonicity needs no smoothness of at all. The argument for the equation is therefore local to the interior and does not assert uniform second-derivative convergence at the corners. Uniqueness is separate again: it uses only the maximum principle and not the series.
Explicit boundary profiles
Two concrete profiles show how the coefficients and the vertical factors behave.
Worked example 5.3.3
(A two-mode top profile)
For , orthogonality leaves exactly two terms:
At the center the second sine vanishes, since , and
by . The higher mode is not simply multiplied by the same vertical factor as the first; its separation constant is different.
Worked example 5.3.4
(A parabolic top profile)
For an example with infinitely many nonzero modes, take . It is and vanishes at both ends, so the theorem applies. Formula (4) with gives , that is,
Thus the even coefficients vanish and the odd coefficients are . Substitution into (3) gives a convergent harmonic solution.
Common mistake
(Extending the boundary polynomial into the rectangle)
The polynomial is the boundary profile, not the solution. Its Laplacian is , so it is not harmonic, and it does not vanish on the bottom side. The harmonic extension is the series built from (6), with each mode damped by its own factor .
Four nonzero sides
Suppose each closed side is given a function of its natural coordinate and the adjacent values agree at all four corners. Let the prescribed corner values be and define the bilinear lift
Theorem 5.3.5
(Dirichlet data on all four sides)
If the four side profiles are in their natural coordinates and agree at the four corners, the Dirichlet problem on has a unique solution in , given by plus four one-side series.
Proof
(Lift the corners, then solve four one-side problems)
Each term of (7) is affine in and in separately, so . On each side is the linear interpolation of the two corner values, so subtracting from the side data makes all four residual profiles vanish at their endpoints, while keeping them . Apply the one-side construction to each residual profile: reflect vertically for the bottom, or interchange with and reflect horizontally for the left and right. Their sum plus is harmonic, continuous on the closure, and assumes all four side values. The maximum principle proves uniqueness. This construction does not require derivatives of the side profiles to match across a corner, and it does not claim regularity on the closed rectangle.
The hypothesis that adjacent values agree at the corners cannot be dropped.
Counterexample mode
A boundary function must assign one value to each corner
Suppose the three lower/vertical sides are prescribed as zero, while the top is prescribed as all the way to its endpoints. A continuous function on the closed rectangle cannot satisfy both: at the top-left corner the top requests and the left side requests . A series may still describe an interior harmonic function with nonmatching side limits, but it cannot solve this continuous closed-boundary problem.
A discontinuous corner: what the series actually solves
The incompatible data in the counterexample still describe a useful interior temperature: take the top side to be for and the other three open sides to be . The sine coefficient of is , so (3) suggests
This formula must not be justified by the earlier theorem, which assumed matching endpoint values. Instead, for every the exponential bound (5) makes the series and all derivatives locally uniformly convergent where . Each term is harmonic, so is harmonic inside. The same bound gives the value on each open vertical side and on the bottom, by continuity of the series away from the top corners.
To see the top value, put . On each interval , partial sums of are uniformly bounded, and summation by parts shows that converges uniformly there. Its sum is : for the corresponding power series is the imaginary part of ; its limit as is , and Abel's theorem identifies it with the ordinary sum.
Moreover, for , decreases with and tends to for each fixed as . Indeed is increasing for , so differentiating in gives a negative derivative. Summation by parts therefore bounds every weighted tail by a constant times , uniformly in and in away from the endpoints. Passing the finite initial terms to the limit proves uniformly on each closed subinterval of the open top side. Thus all four open side limits are attained, even though there is no continuous extension to the two upper corners. In particular, approaching an upper corner along the top and vertical sides gives different limits. No value assigned at the corner can cure that mismatch.
Worked example 5.3.6
(Estimating the center value with one mode)
At the first mode of (8) is . Set . Since by (5) and , the absolute error from keeping only that first mode is at most
For a square, : the first mode is , and the bound is about . As a check, the exact center value for a square is : the four rotated copies of this problem add up to the solution with all sides equal to , which is the constant . The actual error, about , is within the bound. The bound is safe but need not be sharp: it ignores cancellation between the alternating signs . This interior approximation does not imply uniform convergence on the closed rectangle.
Stability and filtering by distance
Proposition 5.3.7
(Maximum-principle stability)
If and are harmonic in and continuous on , with boundary data and on the whole boundary, then
The proof is one line: are harmonic, so each is bounded above by its boundary maximum, which is at most . For the one-side problem, the zero sides contribute no difference, so the right-hand side is simply the sup norm of the difference of the top profiles. This is a stability statement for boundary perturbations, not a proof that every arbitrary discontinuous boundary prescription is admissible.
Equation (5) also quantifies how high spatial frequencies penetrate from the top. The assumed belongs to , so Parseval applies. If , then Parseval and (5) imply
Each mode actually carries its own factor , so high frequencies are more strongly attenuated at a fixed positive distance below the top. This statement does not say that the boundary profile itself is smooth merely because its interior extension is.
Common mistake
(Giving every mode the same vertical factor)
It is tempting to write or to damp every mode by . Each mode has its own separation constant , so the -th coefficient is multiplied by , which is much smaller than for large . Only a single-mode profile is reproduced by one vertical factor.
Practice with solutions
Quick check
(For top data and the other three sides zero, find and its value at .)
One mode is present.
Solution · Answer
. At the center the sine is and , giving .
Quick check
(Derive (6) for and identify which coefficients vanish.)
Use (4) with .
Solution · Answer
Formula (4) with gives , which is (6) after . Even give ; odd give .
Quick check
(Suppose all four boundary values equal the same constant . Use the four-side construction to find .)
Start with the corner values.
Solution · Answer
All four corner values are , so (7) gives and all four residual profiles are zero. Thus ; uniqueness follows from the maximum principle.
Quick check
(Can a function continuous on the closed rectangle have top value and all three other sides zero? Explain the role of the corners.)
Look at an upper corner.
Solution · Answer
No. At either upper corner the top and vertical side would demand different values. This rules out the stated continuous closed-boundary problem, not every possible interior or weak formulation with discontinuous boundary data.
Quick check
(If the top temperature in (8) is a constant instead of while the other open sides remain at zero, what changes?)
The problem is linear.
Solution · Answer
Replace by . It is harmonic, has top limit and zero limits on the other open sides; the first-mode error bound is multiplied by . When the upper-corner mismatch remains.