Differentiating series and function limits

The functions sin(nx) divided by n converge uniformly to zero on the real line, since their magnitudes are at most one over n. Their derivatives are cos(nx), which do not converge at many fixed inputs. At x=π, the derivative values alternate between minus one and one, while the limiting zero function has derivative zero. The function sequence is even smooth at every stage. Uniform value convergence alone gives no permission to exchange derivative and limit.

This example expresses an error-budget issue: taking one derivative multiplies an oscillatory mode by its frequency. If the coefficients decay only as inverse frequency, derivative amplitudes remain constant. Taking more derivatives requires still more coefficient decay. The distinction is central to spectral approximations and scientific computing, where a visually small residual can have a large gradient. A resolution study of function values therefore cannot substitute for a derivative-specific norm estimate.

A useful clean theorem starts with continuously differentiable functions on a finite closed interval. Suppose their derivatives converge uniformly to a function g and the values at one fixed base point converge to c. Integrating the derivative identity from that point expresses each function as its base value plus an integral. Uniform derivative convergence passes through that finite integral, producing a uniform function limit equal to c plus the integral of g. Since g is continuous, the fundamental theorem gives the desired derivative.

The base-point assumption prevents arbitrary additive constants from drifting. Constant functions taking value n have derivatives identically zero, so even perfect derivative convergence cannot force the functions to converge. Uniform derivative control without an anchor determines shapes up to constants. Conversely, once an anchor is fixed, the interval length bounds the error propagated from slopes to values. This makes the theorem a practical stability estimate rather than simply a formal identity about infinite symbols.

For a series of differentiable terms, apply the theorem to finite partial sums. It is enough that the original series converges at one base point and the series of derivatives converges uniformly. The M-test often proves the latter by giving a summable sequence of bounds for derivative magnitudes. Convergence of the original series at every point is not an additional mystery: it follows from the anchor and derivative control. One should state the domain where the common derivative bound holds.

For power series, every compact interval strictly inside the radius supplies room to dominate derivative coefficients by a convergent geometric majorant. Repeating the argument permits derivatives of every finite order, while the radius remains the same. Boundary differentiation still needs separate justification. In the spectral lab, increasing p makes high-frequency amplitudes smaller, but a derivative loses one power of decay. Predict that extra requirement before raising N; the lab plots values and does not compute or certify an infinite differentiated series.