The familiar sine functions vanish at both endpoints. They therefore cannot directly represent a temperature whose endpoint values change in time. One can first subtract a simple function with the prescribed endpoints, but that function itself changes in time. Its time derivative becomes a source in the equation for the remainder. Tracking that source is essential.
Lifting the boundary data
Fix and . Consider the real heat problem
Here and . Two objects are used throughout.
Definition 3.6.1
(Classical class and boundary lift)
A function belongs to the classical class for (1) if are continuous on the closed rectangle , the equation holds in its interior, and the stated data hold on its sides. The boundary lift of the endpoint data is
If , then has zero endpoints and satisfies
Conversely, any sufficiently regular solution of (2) gives a solution of (1) after adding . The minus sign follows directly from and .
Counterexample mode
A boundary lift is not automatically a solution
Take , and let . It has the desired endpoints, but , which is nonzero in the interior. An additional term is needed to cancel this source.
Constructing a classical solution
We give one concrete set of sufficient assumptions rather than claim the weakest possible regularity. Let , , and assume the corner conditions
The first pair makes the initial and boundary values agree. The second pair is necessary in this class: continuity of and the boundary data give and . Continuity of and identify the initial second derivative with ; the PDE then extends to the corners by continuity.
Let be the zero-endpoint solution of . Since is affine in , it is explicitly
Put . Then
The value and derivative conditions in (3) imply . The correction is useful because its time derivative vanishes at the endpoints. It makes the transformed forcing compatible with a twice-differentiable zero-endpoint sine expansion.
The assumption makes , involving and , continuous on the closed rectangle. For write , , and define
Our solution is , where
Theorem 3.6.2
(Classical solution under corner compatibility)
Let and . Every solution of (1) in the classical class satisfies the corner conditions (3). Conversely, if (3) holds, then , with from (4) and from (6), belongs to the classical class and solves (1).
The necessity of (3) was checked directly after (3). The construction is justified by coefficient decay.
Proof
(Uniform convergence of the lifted series)
Integrating the coefficient for by parts four times, using at the endpoints, gives . For , the function and its second derivative are respectively zero at the endpoints and affine in ; two integrations by parts followed by the explicit affine sine integral give . Thus the integral in (6) is bounded by uniformly on .
Consequently the series for , , and converge uniformly on the closed rectangle. Differentiating the th time coefficient gives ; the bounds above make uniformly convergent as well. Hence is represented by that series. Completeness of the sine system and continuity identify the initial series with and the source series with . The endpoint values vanish term by term, and . Thus (4)–(6) construct a solution with continuous on the closed rectangle. The two pairs in (3) are sufficient for this construction, under the stated smoothness assumptions.
Proof X-ray
Two corrections have different jobs
The linear lift puts the desired values at the endpoints, so has zero Dirichlet data. The polynomial does not alter those values; it absorbs into its second derivative. This is where each hypothesis enters. The value pair of (3) gives , and the derivative pair, through and , gives at both ends; together with they allow four boundary-free integrations by parts and . The forcing vanishes at the endpoints because does, so two integrations by parts give , and keep continuous. The factor produced by and is then beaten by these rates. Without the derivative pair, would in general only be , and at the series for would be the sine series of , which vanishes at the endpoints and so cannot converge uniformly to when ; the compatibility is a condition for regularity, not a new boundary condition on .
An explicit moving-end example
Worked example 3.6.3
(A rising left endpoint with one decaying mode)
Take , , and , where is a constant and
Now , so is just one decaying mode:
The endpoints are and . Also , which cancels the time derivative of the lift, while the sine term solves the homogeneous heat equation. At the formula gives the specified , which satisfies both pairs of (3). The fixed cubic term is a response to the rate at which the left endpoint rises, not an equilibrium for a fixed boundary temperature.
When both endpoints rise at the same rate, the correction becomes symmetric.
Worked example 3.6.4
(Both endpoints rising at the same rate)
Let and . Here , so . Formula (4) with gives
which has zero endpoints and . Since , the remainder data vanish, , and , so and . The corner conditions hold: and . Direct check: and .
Uniqueness and response to boundary changes
For a comparison of two solutions in the classical class let be the lift of their endpoint differences, the difference of initial data, the difference of the solutions, and put .
Theorem 3.6.5
(Uniqueness and boundary stability)
A solution of (1) in the classical class is unique. For two such solutions,
Proof
(Energy identity and the source estimate)
If two solutions in this classical class have the same initial and endpoint data, their difference solves the homogeneous heat equation with zero initial and endpoint values. Multiplying by and integrating gives
The nonnegative energy begins at zero and cannot increase, so . This proves uniqueness in the stated classical class.
For the estimate put . It has zero endpoints and solves with . The same multiplication and the zero-Dirichlet Poincaré inequality give
To avoid dividing by a possibly zero norm, apply this to . Since and , it gives ; integrate and let . The triangle inequality then yields (8).
Estimate (8) controls the solution in terms of the boundary values and their rates. It is one useful bound, not a claim that measuring boundary derivatives is necessary for every possible stability theorem.
When the corner conditions fail
Worked example 3.6.6
(Value compatibility without derivative compatibility)
Suppose , and . The values agree at : . At the right end , but at the left end . Therefore no solution can have both and continuous at the corner while satisfying the PDE and boundary data there: continuity would force and . This is a regularity obstruction, not a proof that no solution exists for positive times.
Common mistake
(Matching values at the corners is not enough)
Requiring only and is insufficient for the closed-rectangle regularity proved above. In the classical class the derivative pair , is necessary as well, as the example shows.
Common mistake
(Expanding the solution itself in sines)
Every partial sum of a sine series vanishes at and , so such a series cannot equal at an endpoint where or . The sine expansion is applied to the remainder , whose endpoint values are zero, and the time derivative of the lift appears as the source .
Practice with solutions
Quick check
(If , and , find the solution.)
Subtract the time-independent lift first.
Solution · Answer
The lift is constant in time and , so has initial data and no source. Hence .
Quick check
(With and , which of the corner conditions (3) does satisfy?)
Compute , and at both ends.
Solution · Answer
All four. and . Since , we get and . In fact from (4), so and the construction applies.
Quick check
(Suppose two solutions have identical endpoints, but initial data differing by . Find their difference and its norm.)
The difference solves the zero-Dirichlet problem.
Solution · Answer
Their difference is by uniqueness. Its norm is , so the first Dirichlet rate is attained and (8), with , holds with equality.