Insulation changes the equilibrium
Consider the heat equation on an interval with insulated ends. The solutions we work with are fixed first.
Definition 3.3.1
(Insulated heat problem and its solution class)
Let and . The insulated (zero-Neumann) heat problem is
A solution is a function continuous on such that are continuous on and the equation and boundary conditions hold there.
The zero Neumann conditions describe zero heat flux at both ends in this one-dimensional model. Unlike zero-temperature boundary conditions, insulation preserves the total heat.
Proposition 3.3.2
(Conservation of total heat)
For every solution in this class, for all . Equivalently, the spatial mean is constant in time.
Proof
(Differentiate the total heat)
For , continuity of on allows differentiation under the integral, and integration by parts gives
So the integral is constant on . Continuity of on the compact set makes as , which identifies the constant.
The constant mode cannot decay: it records a conserved quantity. This predicts that the long-time limit should be the initial mean, not necessarily zero.
Common mistake
(Dropping the zero mode)
With Dirichlet conditions every mode decays, and it is tempting to expect here too. But the derivative of a constant is zero, so an insulated boundary permits every constant equilibrium, and the coefficient below is exactly the mean. Omitting it forces every solution to have mean zero, which contradicts conservation for data such as .
Cosine modes and a solution
Separating gives with . Multiplying the eigenvalue equation by and integrating shows that any eigenvalue is real and nonnegative. Solving the cases , and shows that the eigenfunctions and eigenvalues are
For define
Theorem 3.3.3
(Cosine-series solution)
For , the series
converges absolutely and uniformly on , is smooth on , and is a solution of the insulated problem in the stated class.
Proof
(Coefficient decay, completeness and termwise differentiation)
Integrate twice by parts: the first boundary term vanishes because the sine is zero at both endpoints, and the remaining boundary terms and integral show . No endpoint compatibility is needed for this coefficient bound. The series therefore converges absolutely and uniformly on . Its initial sum is continuous and has the same cosine coefficients as ; completeness makes their difference zero in , and continuity makes it zero everywhere. For any , the exponential factor makes the series of all spatial and time derivatives converge uniformly on . Termwise differentiation then proves the PDE and zero-flux conditions for , and uniform convergence gives the initial condition.
We do not claim differentiability at the time-space corners without additional compatibility and regularity assumptions.
Worked example 3.3.4
(Finitely many cosine modes)
If , the coefficients are read off directly: , , , and all others vanish. Then
Its mean is for every . The second oscillatory mode decays four times as fast in its exponent as the first.
Worked example 3.3.5
(Data that violate the boundary condition)
Take , so . Then , and for integration by parts gives
which is for odd and for even . Hence
The coefficients are , as the proof predicts without compatibility. For every each term has zero -derivative at both ends, so , while : the derivative does not extend continuously to the corner . The temperature itself does, and the mean stays .
Energy, uniqueness, and rate of relaxation
The energy method gives uniqueness and a decay rate for every solution in the class, not only for the series.
Theorem 3.3.6
(Uniqueness and exponential relaxation to the mean)
A solution of the insulated problem in the stated class is uniquely determined by , and
The rate is sharp: gives equality.
Proof
(Energy identity and the Neumann Poincaré inequality)
Let . By conservation it has mean zero at every time, and it satisfies the same zero-flux boundaries. Its energy satisfies
The boundary term vanishes because . For uniqueness, apply the same identity to the difference of two solutions with the same initial data: its energy is nonnegative, nonincreasing, and by continuity at initially zero, so the difference vanishes.
For the rate, the cosine-series form of the Neumann Poincaré inequality says that every mean-zero obeys
If has cosine coefficients , integration by parts gives the sine coefficients of as ; the boundary term vanishes because the sine vanishes at both ends. Parseval for the complete sine and cosine systems then gives the inequality: the zero cosine coefficient is absent, and every other eigenvalue is at least . With this yields , and Grönwall's inequality gives , whose square root is the bound. For the solution is , which attains it.
The two inequalities in this proof do different work.
Proof X-ray
Where the mean-zero condition enters
The energy identity uses the boundary conditions once, to remove , and on its own shows only that energy does not increase. A rate needs a comparison of with , and that is where the mean-zero property is used: the Poincaré inequality fails for constants, which have but . Subtracting , which is legitimate because the mean is conserved, moves into the class whose lowest eigenvalue is rather than . Equality in the Poincaré inequality for the first cosine mode is what makes the rate sharp.
It is a norm bound, not a statement that each point of the solution moves monotonically toward the mean.
Counterexample mode
Conservation does not mean pointwise constancy
For the mean is zero at all times, while changes with time except at its nodal point . Conservation constrains the integral, not every temperature value.
Worked example 3.3.7
(How long until the deviation is one percent)
For , the mean is and . The theorem gives
The right side is at most one percent of its initial value once , that is, for . Doubling the length of the rod multiplies this time by four.
Practice and solutions
Quick check
(What is the insulated heat flow for constant initial data ?)
Solution · Answer
for all : every cosine coefficient with vanishes and .
Quick check
(For , find the mean and the solution.)
Solution · Answer
The mean is , and .
Quick check
(Why does the energy argument for uniqueness use the difference of two solutions?)
Solution · Answer
The difference satisfies the homogeneous PDE, homogeneous boundary conditions and zero initial data; its energy starts at zero and cannot increase, so it remains zero.
Quick check
(Why can an estimate with decay rate strictly faster than and prefactor one not hold for every mean-zero initial datum?)
Solution · Answer
The first cosine mode decays exactly at rate , so a faster universal bound would fail for .