Jackson estimates and smoothness

Take a real continuous $2\pi$-periodic function on the real line. Its periodic modulus of continuity is the largest difference between function values separated by a shift of magnitude at most h, with locations understood periodically. Compactness of one period makes this quantity tend to zero with h. A Lipschitz condition gives a linear bound in h, while a Hölder condition with exponent between zero and one gives a fractional-power bound. These are uniform conditions over every location, not average conditions.

Endpoint compatibility matters. A function smooth inside a finite interval may become discontinuous when its endpoints are identified to form a periodic target. A jump prevents uniform approximation by continuous trigonometric polynomials, so one cannot apply the periodic continuous theorem to that extension. Derivative matching similarly affects stronger periodic smoothness assumptions. Before reading a convergence rate, specify the domain, its boundary identification, and which norm measures the approximation error.

Let $E_n^{\mathrm{trig}}(f)$ be the least uniform error among real trigonometric polynomials using frequencies at most n. The first Jackson theorem gives a universal constant C, independent of f and n, such that this error is at most C times the modulus at scale 1/n for every n at least one. Consequently a uniform Hölder bound with exponent alpha yields an error of order n to the minus alpha. The statement guarantees an approximant; it does not name every Fourier partial sum as that approximant.

A construction idea is to convolve f with a normalized positive trigonometric kernel whose mass concentrates on shifts of size about 1/n. Subtract f inside the convolution and control shifted differences by the modulus. Jackson kernels use a suitable power of a Dirichlet-type expression and a bandwidth adjustment to obtain the required concentration. This is a proof sketch of the mechanism, not a complete kernel estimate. Higher-order Jackson results use higher differences or derivative smoothness with corresponding hypotheses.

If a particular series converges absolutely and its coefficients have a summable bound, the triangle inequality directly controls the uniform error of its truncation. This can prove a rate for a specific algorithm, whereas Jackson controls the best available error from the entire degree class. Neither direction should be reversed casually. An upper bound says the error is no larger than a rate; it does not establish a matching lower bound or show that the function lacks additional smoothness.

Analytic targets can have geometric approximation rates, which are stronger than the first-order estimate detects. Conversely, a continuous function with a slow modulus need not enjoy a useful algebraic rate under that information alone. In experiments, a straight line on a log-log plot over a few degrees is only evidence about those degrees. To turn it into a theorem, identify a function class, a norm, and constants or asymptotic quantifiers that are uniform over the intended range.