Heat diffusion, vibrating strings, and equilibrium potentials have different time behavior, but boundary conditions and spatial eigenfunctions connect them. Separation of variables discovers candidate solutions; convergence and differentiability must still be justified before an infinite series becomes a classical solution. Unless stated otherwise, the functions in these basic models are real-valued.
Heat flow and sine modes
From conservation to the heat equation
Why does the second spatial derivative measure heating? Take a uniform rod and write all rates per unit cross-sectional area. Let be temperature, the heat flux in the positive direction, and the heat supplied per unit volume per unit time. With constant density and specific heat , the heat capacity per unit volume is . For smooth functions, conservation of energy on any subinterval says
Positive enters the interval, whereas positive leaves it. Writing the flux difference as and using the arbitrariness of gives the local balance . Conservation alone does not determine the flux; the constitutive assumption known as Fourier's law is , with constant conductivity , and the minus sign expresses flow from hotter to colder locations. Hence
Without a source this becomes . At the outward flux is , while at it is . Fixing an endpoint temperature specifies ; insulation specifies zero outward flux and therefore when . A zero temperature is not the same condition as zero flux.
Common mistake
(A variable conductivity stays inside the derivative)
If varies smoothly, the balance law gives , not . For and , the profile has but , so it is not steady, although it would be steady for constant conductivity. The lost term is the whole difference.
Where the permitted modes and coefficients come from
Consider on , , with zero endpoint values. Writing and dividing by where it is nonzero gives with . Division is only a way to discover candidates; the resulting modes are checked in the original equation, including at their zeros. The spatial problem is the following.
Definition 2.1.1
(Dirichlet eigenvalue problem on )
A number is a Dirichlet eigenvalue of on if there is a nonzero , possibly complex-valued, with
Such an is an eigenfunction, and is an eigenpair.
Proposition 2.1.2
(Dirichlet eigenpairs of an interval)
The Dirichlet eigenvalues on are exactly , , and the eigenfunctions for are the nonzero multiples of . Moreover
Proof
(Integration by parts, then the two endpoint conditions)
Multiply by and integrate by parts; the boundary term vanishes because , so
Since , the left integral is positive, so is real and nonnegative. If , then , and both endpoint conditions force . Hence with and . The condition at zero gives ; then requires , so with . Conversely each is an eigenfunction. The orthogonality relation follows from : for both cosines integrate to zero over , and for only the constant survives.
The time equation gives the factor . For suitable initial data , the formal solution is
For a finite sum, multiplying by the th sine and integrating eliminates every other term and gives the coefficient formula. For an infinite expansion the same calculation is valid only once convergence permits it; orthogonality alone does not prove that the expansion exists or recovers . If the are bounded, the exponential factors make every differentiated series converge uniformly on for each , so the formula is classical for positive time. Continuity on the closed space-time boundary also requires the compatibility .
Worked example 2.1.3
(A curved initial profile)
Let and . Integrating by parts twice, with and , gives
Even coefficients vanish, and , . Therefore
The bound makes the series converge uniformly even at , and the sine-series convergence theorem recovers this continuous, endpoint-compatible profile. The profile is symmetric, but that does not mean only the first mode is present: every odd coefficient is nonzero. The mode decays at nine times the exponential rate of the first, so diffusion smooths oscillations rather than translating them.
Quick check
(What is the coefficient of in the solution with , and why?)
Look at the factor , or at symmetry about .
Solution · Answer
It is zero: the formula contains . Equivalently, is symmetric about , whereas is antisymmetric about , so their product integrates to zero.
A time-dependent source: from the PDE to coefficient equations
A source need not be steady, and different spatial modes may be supplied at different rates. Consider, for and ,
Put , and write , and for the th sine coefficients of , and .
Proposition 2.1.4
(Modal coefficients of a forced heat solution)
Suppose are continuous on and solves the problem above there. Then for every ,
and consequently
Proof
(Two integrations by parts and an integrating factor)
Continuity of on the compact rectangle permits differentiation under the integral, so . Integrate the term by parts twice:
The first bracket vanishes because the sine is zero at both ends; the second vanishes because is zero there. This gives the coefficient equation, and continuity of at gives . Multiplying by turns the left side into ; integrating from to yields (H).
The first term of (H) remembers the initial temperature; a contribution supplied at time has only the remaining time in which to decay. Formula (H) is a necessary identity for every sufficiently smooth solution. It does not by itself prove that the infinite sine series can be differentiated twice. If and are finite sine sums and the coefficients are chosen by (H), direct substitution gives a classical solution, and the energy theorem in the section on heat uniqueness identifies it with any other solution in the same class.
Worked example 2.1.5
(Forcing that decays at the natural modal rate)
Let , , , and . Only two modes occur, with , , , . Formula (H) yields
So . The factor appears because the forcing of the first mode decays exactly at that mode's own rate; it does not mean unbounded growth, since . Both endpoint values vanish, the initial value is , and direct differentiation verifies the source.
Quick check
(With , , zero initial data and , find the solution.)
Only one coefficient is nonzero.
Solution · Answer
Only is forced, and (H) gives . Hence ; it satisfies and starts at zero.
Fixed nonzero temperatures: subtract the equilibrium
Sine modes vanish at both ends, so they cannot by themselves represent a rod held at two nonzero temperatures. For constant endpoint temperatures and , first solve the steady problem , , , whose solution is . Then satisfies with zero endpoint values and initial value : it is this difference, not , that is expanded in sines. The argument uses ; a time-dependent boundary lift generally introduces a source term.
Worked example 2.1.6
(A rod held at two temperatures)
Let , , and prescribe
The equilibrium is , leaving . The previous example with gives , so
The bound gives uniform convergence on , so the formula attains the initial profile continuously; for termwise differentiation is permitted, so it is classical for positive time, and uniformly as . This is an edge case for the energy theorem below: the corners are only continuous (see the next warning). Applying the energy calculation on and letting still proves uniqueness among solutions continuous on the closed region and smooth up to the ends for .
Common mistake
(Matching values at a corner is not matching derivatives)
In the last example the initial profile agrees with both endpoint temperatures, so the solution is continuous at the corners. But . If and were continuous up to a corner, the constant boundary temperature would give there and the equation would force . In general, a solution of the forced problem with zero ends whose and are continuous up to the corners needs in addition to . Writing sine coefficients does not repair an incompatible profile, though a positive-time solution may still exist.
Quick check
(For , , endpoints , and , find and its limiting profile.)
Subtract the steady part first.
Solution · Answer
The steady part is , and the remainder already consists of two sine modes, so
Both exponentials decay, so uniformly; the factor in the second rate is the square of the spatial frequency .
A rectangular plate: two spatial frequencies
For a thin homogeneous plate with , consider with on all four sides. For positive integers put
Each factor vanishes on its pair of sides, and , so is a mode. The decay rate is times the sum of both squared frequencies. Finite combinations are solutions by linearity; convergence of double series is a separate question. The energy argument in the section on heat uniqueness, with Green's identity in place of one-dimensional integration by parts, gives uniqueness for sufficiently smooth solutions.
Worked example 2.1.7
(Two modes on a square plate)
On with zero side temperatures, let . The eigenvalues are and , so
At this is the prescribed profile, each side value is zero, and term by term. The second mode decays at twice the exponential rate of the first, even though its initial coefficient is larger.
Quick check
(On the same square with zero side temperatures, find the solution with initial profile and its decay rate.)
Add the two squared frequencies.
Solution · Answer
The eigenvalue is , so with exponential rate .
Heat uniqueness from energy
Theorem 2.1.8
(Energy uniqueness and stability for the heat equation)
Let solve , , with the same boundary values, and let . Then for . In particular, equal initial data give .
Proof
(Differentiate the energy of the difference)
The difference satisfies with . Differentiating under the integral and integrating by parts gives
So the energy is nonincreasing, which is the stated bound. If , the nonnegative energy starts at zero and remains zero; continuity of then gives everywhere.
Proof X-ray
One identity, two conclusions, no existence proof
Three hypotheses enter at three different places. The regularity class justifies differentiating under the integral and integrating by parts. Equal boundary values make vanish at the ends and kill the flux term ; with different boundary data this term has no sign. The sign turns into a dissipation term. Only after these steps does the argument split: zero initial difference gives uniqueness, nonzero initial difference gives stability. Neither conclusion constructs a solution, so existence must come from elsewhere, for example the modal formulas above.
The theorem compares two solutions through their difference, and this is essential.
Common mistake
(Each energy decreasing does not make two solutions equal)
For and , and both satisfy the heat equation with zero endpoints, and both energies decrease, yet . The proof applies the estimate to , whose initial energy is zero only when the initial data agree.
Wave propagation
From force balance to the string equation
Let be the transverse displacement of a string whose material point labeled moves vertically, and let be mass per unit length. In the ideal small-slope model, neglect bending stiffness and air resistance, and take the horizontal component of tension to be a positive constant ; these are modeling assumptions. The tangent is parallel to , so the vertical tension component is . With an external transverse force per unit length, Newton's second law on a segment gives
Writing the tension difference as and using continuity and the arbitrariness of gives
Without external force, . The curvature term is the difference of vertical pulls at the segment ends: a straight tilted segment feels no net transverse force. The tension magnitude is approximately constant only for small slopes, so this is a linear model. A smoothly varying tension gives , for the same reason as the variable conductivity above.
Worked example 2.1.9
(Equilibrium plus vibration under a constant load)
Fix and apply a constant load . A static profile must satisfy with zero ends; integrating twice gives . Since balances the load, solves the unforced equation with zero ends. For initial displacement , , and zero initial velocity,
The second term has , starts at with zero velocity, and vanishes at both ends. A sufficient small-slope condition is . A curved static string under load is therefore no counterexample to the fact that an unforced static string has .
Quick check
(Keep , , and the initial displacement of the last example, but quadruple the mass density. What happens to , the period, and the equilibrium?)
Which quantities contain ?
Solution · Answer
The speed becomes . The angular frequency halves, so the period doubles. The equilibrium is unchanged, since does not contain .
Why the formula has this shape
On the whole line, both the displacement and the velocity are prescribed, and the solution is given explicitly.
Theorem 2.1.10
(d’Alembert’s formula)
Let , and . The problem on , , , has exactly one solution , namely
In particular depends only on the data on the interval with endpoints and .
Proof
(Characteristic coordinates)
Set , , and write ; the change is invertible, with and . The chain rule gives
So is independent of , and integration gives for functions . The data require
Differentiating the first equation and solving gives and . Integrating from and choosing ,
and adding the profiles at and yields the formula. Every solution has this representation, which proves uniqueness. Conversely, for and the formula is , and direct differentiation verifies the equation and both data.
Proof X-ray
Two data determine two traveling profiles
The key move is the change to characteristic coordinates, after which the equation says only that a mixed derivative vanishes. That leaves two unknown profiles, and the two data supply exactly two equations: displacement fixes , velocity fixes . The equations determine and but not the separate constants of integration; shifting a constant from to leaves unchanged, so this freedom produces no second solution. The integrals have finite endpoints, so no decay at infinity is used, and dropping the velocity datum would leave , hence , undetermined.
The domain-of-dependence statement is local, as the next example shows.
Worked example 2.1.11
(A periodic profile with a uniform push)
Let with and constant. The addition formula gives , and the integral term is . Hence
The oscillation is a superposition of oppositely traveling waves, and the uniform initial velocity adds a uniform displacement. Finite propagation speed does not make solutions compactly supported: these data already fill the line. It says that changing data outside cannot change .
How the fixed-string coefficients are found
On a string with , a separated solution leads to the same Dirichlet problem as for heat, so , and now . With ,
These modes oscillate rather than decay. For a finite sine sum, or whenever the expansion and its time derivative can be identified with the data, orthogonality gives
The factor appears because equals at . These formulas find coefficients; they do not alone prove convergence to a classical solution.
Worked example 2.1.12
(Displacement in one mode, velocity in another)
Let and . Then , , and all other coefficients vanish:
At the first term gives the displacement and has zero velocity; the second starts at zero but its derivative supplies . Both terms vanish at the endpoints for every time.
Quick check
(With zero initial displacement and initial velocity , what is the solution, and why can the factor not be omitted?)
Differentiate your answer at .
Solution · Answer
The solution is . Without the factor, has initial velocity , which is correct only when .
Uniqueness for the fixed string again comes from an energy, now one that includes the velocity.
Theorem 2.1.13
(Energy conservation for the fixed string)
If , , solves with , then
is constant. Consequently such a solution is determined by its initial displacement and velocity.
Proof
(The boundary power vanishes)
Differentiate, substitute and integrate by parts:
Fixed endpoint values give there, so . For the difference of two solutions with the same displacement and velocity, , so everywhere; is constant and vanishes at the ends, hence .
When the endpoints move
Suppose instead that and . Separate the prescribed boundary motion from the interior response by
Since , the function has zero endpoint values but is generally forced:
Replacing the boundary values by zero without adding this forcing would solve a different problem. If and are continuous at the corners, then necessarily , , and ; these conditions alone are not an existence theorem. The moving supports can also do work: the computation in the last proof gives , whose sign is not fixed.
Laplace and Poisson equations
Definition 2.1.14
(Harmonic function and Poisson equation)
Let be open. A function is harmonic if in . For a given , Poisson's equation is .
Theorem 2.1.15
(Weak maximum principle and Dirichlet uniqueness)
Let be bounded and open, and let be harmonic. Then and . Consequently two such harmonic functions satisfy , and the Dirichlet problem for Poisson's equation has at most one solution in .
Proof
(Perturb to a strict subsolution)
For put , so . At an interior maximum point of the Hessian would be negative semidefinite, giving ; hence the maximum of over the compact set is attained on . With ,
Let ; applying this to gives the minimum. For two solutions of with equal boundary values, the difference is harmonic with zero boundary values, so it is both and .
Boundedness enters twice: compactness of guarantees that attains a maximum, and is finite.
Counterexample mode
An unbounded domain loses Dirichlet uniqueness
Claim: a harmonic function continuous up to the boundary and zero on the boundary is zero. On the half-plane take . Then , is continuous on , and on , yet . Both and solve the same Dirichlet problem, and while . The repaired statement assumes bounded, as in the theorem, or adds a growth condition at infinity.
Existence is a separate question; on a rectangle, separation produces explicit solutions.
Worked example 2.1.16
(A rectangle with one nonzero side)
On with zero data on three sides and , try . Then with and , so
The sine vanishes at , the hyperbolic sine at , and the quotient is at ; by direct substitution. By the theorem, this is the only solution continuous on the closed rectangle.
Neumann data behave differently. On a bounded connected domain with smooth boundary, two classical solutions of with the same normal derivative differ by a harmonic with , and Green's identity gives , so they differ by a constant. Moreover, if and , the divergence theorem gives the necessary compatibility .
Practice with solutions
Quick check
(A whole-line string has and . Find and check its initial velocity.)
Use d’Alembert’s formula.
Solution · Answer
The integral in d’Alembert’s formula is over an interval of length , so , and .
Exercise 2.1.17
(Can a function satisfy in a nonempty bounded smooth domain and everywhere on ?)
Core
Hint(1)
Integrate the equation over .
Solution(3)
- 1
Integrate the equation
Because: The divergence theorem applies to .
- 2
Compare with the source
Key step
The left side equals , which is positive because is a nonempty open set.
- 3
Conclude
No: the compatibility condition fails, since .