A Fourier mode under the heat equation decays, but under the wave equation it oscillates. On a circle these oscillations have integer frequencies, so their phases can return together. One mode needs separate attention: the spatial mean has no restoring force and can drift linearly in time.
The Cauchy problem in the energy space
Let have normalized measure ; all norms use , and functions may be real- or complex-valued. The integral in the d’Alembert formula below uses ordinary Lebesgue .
Definition 1.10.1
(Fourier coefficients and the space )
For write . The space consists of those with finite norm . For such the distributional derivative lies in and has coefficients .
Let and , and consider
The equation below is understood in distributions on ; initial values are taken in the spaces stated in the theorem.
Theorem 1.10.2
(Wave evolution in the energy space)
There is exactly one solution in of (1). Its Fourier coefficients are
For every its energy is conserved:
Moreover and . If has zero mean, the full state is -periodic in time.
Proof
(Each frequency is an oscillator)
A finite Fourier sum reduces (1) to ; its solutions give (2), including the linear zero mode. For , differentiation yields
The real rotation preserves the squared norm even for complex entries. Thus the nonzero-mode contribution to is independent of , and the zero-mode contribution is .
Approximate in and in by Fourier polynomials. For a fixed bounded time interval, (4) bounds the and velocity tails uniformly: for the squared position tail is at most twice the conserved tail , while the zero position mode is an explicit affine function of time. The finite solutions are therefore Cauchy in for every compact time interval , with the usual norm including the derivative. The limit has the stated regularity, initial values and coefficients; passing to the limit against smooth compactly time-supported periodic test functions gives the distributional equation. Passing Parseval's identity to the limit proves (3). The same bounds and the affine mean give the stability estimate ; only the zero mode can cause the factor growing in .
For uniqueness, a distributional solution in the stated class has each Fourier coefficient satisfying as a one-dimensional distribution, with the prescribed initial value and derivative. Since its right side is continuous in time, that coefficient is twice continuously differentiable and obeys the ordinary equation, which has the unique solution (2). Finally, integer makes the trigonometric factors -periodic, while the zero mode changes by .
Proof X-ray
Energy does not control an unconstrained mean
The construction rests on one choice of variables. Pairing with turns each nonzero mode into the rotation (4), so every tail of the energy is conserved, not merely the total; that is what makes the finite approximations Cauchy uniformly on compact time intervals. The zero mode escapes the rotation. Its velocity enters the energy, but its displacement does not, which is why the stability bound carries the factor and why recurrence needs . Uniqueness uses a separate idea: testing against reduces a distributional solution to one ordinary differential equation per frequency, and such an equation is determined by its two initial values.
Traveling waves and finite propagation
Theorem 1.10.3
(The periodic d'Alembert formula)
Take the continuous representative of and periodic representatives of on . Then the solution is represented by
The integral is signed when . If and vanish almost everywhere on a lifted open interval , with , then whenever for that choice of lift.
Proof
(Pass from smooth data to energy data)
For smooth periodic data, differentiating (5) proves the equation and the initial conditions; uniqueness for the Fourier-mode equations identifies it with (2). The formula is periodic in even when has nonzero mean. For general data, Fourier polynomials approximate in and in . The one-dimensional Fourier estimate gives uniform convergence of the approximating . On bounded time intervals, Cauchy–Schwarz on the lifted integration interval bounds the integral of the -error uniformly in ; periodicity reduces it to finitely many periods. The same -to-uniform estimate applied to the left side identifies its continuous representative with this uniform limit. Therefore (5) passes to the energy solution constructed above.
If vanishes almost everywhere on the open interval, its continuous representative vanishes on the closed interval as well. When , both endpoints lie in that closed interval and the integration interval lies inside it; is zero almost everywhere there. Formula (5) then gives throughout the cone, including its boundary. This is a local statement on the lifted circle; it does not say that waves disappear after traveling around it.
Worked example 1.10.4
(Two frequencies return together)
For and , (2) gives
The conserved energy is . Both modes and their velocities return after ; no decay is implied by conservation of energy.
Worked example 1.10.5
(A corner that travels without smoothing)
Let be the periodic extension of from . Then and , the square wave equal to on and on . Choose , which lies in and has mean zero. Since is absolutely continuous, , and (5) gives
The corner at moves to the right with speed one and keeps its shape: the evolution transports the corner and does not smooth it. Formula (2) agrees, because gives , the coefficient of . At every time and each have squared norm , so .
Counterexample mode
Conserved energy can coexist with growing displacement
For and , the solution is . Its energy is always , yet . Thus energy conservation alone cannot bound the mean displacement or imply time recurrence. The missing assumption is zero mean velocity.
Worked example 1.10.6
(A localized velocity pulse meets itself)
Fix , let , and let be the periodic function equal to on and to on the rest of . By (5), is half the length of the part of covered by the lifted support . For only the pieces centered at and can meet this interval, and
The first line is finite propagation: the data vanish on , an interval of radius about , so the d'Alembert theorem gives for . At the two fronts, which have traveled around the circle in opposite directions, arrive together, and each contributes slope . The long-time behavior is the edge case: , so the state does not recur. Indeed , as the first theorem requires.
Common mistakes
Common mistake
(Treating the zero mode like the others)
The expression has no value at . The zero mode obeys , so it equals , which is also the limit of as . Dropping it, or giving it an oscillating form, loses the drift: for and the solution is , whose mean grows linearly.
Common mistake
(Expecting the wave equation to smooth the data)
Unlike heat evolution, the wave equation need not smooth energy data. For example, take and . The mean of is zero, so ; this jump discontinuity is not in . There is no general gain of one spatial derivative in the velocity at positive time; in particular, this example is not in .
Practice
Quick check
(Solve (1) for and .)
Solution · Answer
. Its velocity is .
Quick check
(Solve (1) for and , and compute the energy.)
Solution · Answer
Only the zero mode is present: . The energy is .
Quick check
(If and , find and its energy.)
Solution · Answer
and .
Quick check
(Suppose . What is the spatial mean of for all , and what is the state after time ?)
Solution · Answer
The mean remains , and both and return to their values at time zero after . Some particular states can have a smaller period.
Quick check
(If and vanish almost everywhere on , what can be concluded in the lifted cone ?)
Solution · Answer
The traveling-wave formula gives throughout that cone, including its boundary. Values can be affected once their characteristic interval reaches nonzero initial data; the conclusion makes no claim outside the cone.