A Fourier coefficient measures how much of one oscillation is present. A smoother function tends to have smaller high-frequency coefficients: integration by parts moves a derivative onto the function and produces a factor . At a jump, however, integration by parts also produces a boundary contribution. Keeping that contribution is what distinguishes a continuous corner from a discontinuity.
Summable coefficients make a series uniformly convergent, but differentiating the series multiplies its th coefficient by . The decay must be checked again after that multiplication. We work with complex coefficients throughout and use the earlier Fejér and Parseval theorems to identify the resulting series with the original function.
Periodic functions and their coefficients
For an integrable -periodic function, set
Definition 1.2.1
(Periodic functions)
A -periodic function is periodic (we also write periodically) if and its derivatives through order are continuous on the real line and -periodic. In particular their endpoint values over one period agree.
Parseval uses the normalized norm .
Theorem 1.2.2
(Integration by parts and jumps)
If is periodic , then for ,
where . More generally, suppose one period is split at finitely many points into intervals on which is with finite one-sided values and integrable derivative. Let include the periodic seam exactly once. Then
where is the piecewise derivative; point values at the cuts do not affect its integral. The values at a jump can be assigned arbitrarily without changing the coefficients.
Proof
(Retaining the boundary terms)
On one smooth interval , integrating by parts gives
Summing over a full period makes each cut contribute times . At the periodic seam the exponential agrees at endpoints, so the same rule applies. Divide by and rearrange. In the periodic case every jump vanishes; repeat the argument times and take absolute values inside the final integral.
The piecewise formula says only when the derivative is integrable. Its jump terms can cancel for some frequencies, but they cannot in general be discarded. For a continuous piecewise function, the jumps vanish; the Riemann–Lebesgue lemma applied to the integrable derivative gives . Absolute summability requires additional control; for example, suffices by the Parseval argument below.
When can the Fourier series be differentiated?
Theorem 1.2.3
(Absolute sums and smooth periodic functions)
Let be continuous and -periodic. If
for an integer (with when ), then the Fourier series and all its termwise derivatives through order converge absolutely and uniformly, and their limits are the corresponding derivatives of .
In particular, every periodic function satisfies this condition. Thus a periodic function has an absolutely and uniformly convergent Fourier series, even though the elementary coefficient bound alone would not prove that.
Proof
(Absolute convergence and identification)
For , the weighted condition bounds , since the finitely many modes are harmless. The Weierstrass test gives uniform convergence of the series of derivatives. The standard termwise differentiation theorem, applied successively on compact intervals, yields a periodic sum and identifies its derivatives with those uniform sums.
Uniform convergence permits termwise integration, so has Fourier coefficients . To identify , the Fejér means of are identically zero, while Fejér's theorem gives uniform convergence of those means to the continuous function . Thus . This use of Fejér uniqueness avoids assuming in advance that ordinary partial sums converge to .
Now suppose periodically and put . The integration-by-parts identity gives , including with the usual separate interpretation when . Since , Parseval and Cauchy–Schwarz give
The zero mode is finite. This proves the claimed weighted summability; for its contribution is zero.
Proof X-ray
Two separate uses of regularity
The weighted condition concerns , and integration by parts rewrites as with . Summing these absolutely costs one more derivative. Because is , Parseval applied to puts the sequence in , and Cauchy–Schwarz against converts square summability into absolute summability. This is exactly where rather than is used: the elementary bound alone has a divergent sum. The identification step rests on a different input. Uniform convergence lets the sum be integrated termwise, and Fejér uniqueness for the continuous difference closes the argument without any pointwise theorem for ordinary partial sums.
Worked examples
Worked example 1.2.4
(The square wave and its jumps)
On let , and on let , extended periodically. Its jumps are and , while its piecewise derivative is zero. Hence
Even modes vanish, but odd modes have magnitude , so the absolute coefficient sum diverges. The values assigned at and do not alter this calculation. The jump formula explains the decay without differentiating the discontinuous function as though its ordinary derivative captured the whole change.
Worked example 1.2.5
(A continuous corner and an absolutely convergent series)
Let on , extended periodically. Its derivative away from the corners is the square wave , and has no jumps. The jump-free integration-by-parts formula gives . Therefore
and Fejér uniqueness gives the uniformly convergent series
The corner produces rather than the square wave's coefficients. At the identity also yields .
Counterexample mode
Absolute convergence does not justify another derivative
The series converges absolutely and uniformly, yet its sum is not differentiable at . Formally differentiating it produces
The earlier piecewise- convergence theorem shows that this series converges pointwise to away from its jumps and to their midpoint, zero, at each jump. It cannot converge uniformly: the original partial sums are smooth, and a uniform limit of their derivatives together with convergence at one point would force their sum to be continuously differentiable. Thus one must verify the differentiated series separately. Absolute convergence of the undifferentiated coefficients is not enough.
Worked example 1.2.6
(A continuous function whose derivative jumps)
Let on , extended periodically. Its endpoint values agree, so is continuous with no jumps, but its piecewise derivative jumps by at the seam; thus is not periodic . Apply the jump formula to , whose piecewise derivative is the constant :
The jump-free formula for then gives , so
Each integration by parts gains a factor until a jump appears; the jump of fixes the rate . The coefficients are absolutely summable, so the case of the theorem gives uniform convergence to . The case fails, since , in agreement with . This is the edge case of the theorem: absolute summability holds although is not periodic , so the sufficient condition with is not necessary.
Common mistakes
Common mistake
(Dropping the jump terms)
The square wave has piecewise derivative zero. Discarding the jump terms in the integration-by-parts identity would give for every , so would equal its mean almost everywhere, which is false. The jumps and produce every nonzero coefficient.
Common mistake
(Checking regularity only inside one period)
Smoothness on says nothing about the seam. For on , extended periodically, inside the period but the seam jump is . Treating as periodic would give , hence for . The jump formula gives instead , that is,
which agrees with direct integration of and decays only like . Regularity must be checked for the periodic extension, endpoint values included.
Exercises
Quick check
(Suppose is periodic and . Bound for and prove its Fourier series converges uniformly to .)
Solution · Answer
Two integrations by parts yield . The coefficient series is absolutely summable, so the theorem gives uniform convergence to .
Quick check
(For the -periodic function on , find for using the seam jump. Why does the coefficient of the piecewise derivative vanish at nonzero frequencies?)
Solution · Answer
The periodic seam has jump , and the piecewise derivative is . Its nonzero Fourier coefficients vanish by orthogonality. Therefore , or . The assigned endpoint value does not matter.
Quick check
(Suppose is continuous and . What regularity and derivative series follow? Does this hypothesis require ?)
Solution · Answer
The theorem gives and uniform convergence of the original series and its first two termwise derivative series. It does not require : choose a smooth cutoff supported strictly inside and equal to one near zero, and extend periodically. It is but not at zero. Its third derivative is piecewise with a jump, so the jump formula gives ; three integrations by parts give . Thus the weighted sum converges.
Quick check
(Use Parseval on the square wave to compute independently of evaluating the series for at zero.)
Solution · Answer
. Parseval gives , so the sum is .