Why an anchor is needed
A derivative measures changes but loses the starting height: and have the same derivative. To reconstruct a limit of functions from limits of derivatives, we must also control one function value. A single convergent value is enough because every other value can be reached by integrating the derivative from that point.
Definition 5.3.1
(Uniform convergence on an interval)
Let . We say uniformly on if as . Equivalently, for every there is one index such that for every and every in the interval, not an index chosen separately for each point.
Here means continuously differentiable on the interval, with endpoint derivatives interpreted one-sidedly.
Theorem 5.3.2
(Anchored uniform derivative limit)
Let and let . Suppose that for some , the numbers converge to , and that converges uniformly on to . Then converges uniformly to a function , with and .
The assertion about at endpoints has the same one-sided interpretation.
Proof: separate height from change
Proof
(Height at the anchor plus accumulated change)
Since each is continuous and their convergence is uniform, is continuous. Define
The fundamental theorem of calculus gives and , so , and . Thus, for every ,
The same bound works when : reversing the integral changes its sign but not the estimate. Both terms on the right tend to zero, independently of . Taking the supremum over proves uniform convergence of to .
Proof X-ray
Where each hypothesis is spent
The argument makes one non-obvious move: it compares and not through their values but through their integral representations from the same base point . The difference then splits into the anchor gap plus the integral of from to .
Convergence of is used once, for the first term. Uniform convergence of is used twice: first to make continuous, so that is defined and by the fundamental theorem; then to bound the integral of by times a supremum that does not depend on . That independence is exactly what makes the conclusion uniform. The argument closes because both terms are numbers indexed by alone. Without the anchor the first term need not be a number tending to zero; without uniform control of the derivatives the second term cannot be bounded by one supremum.
Worked examples
Worked example 5.3.3
(A differentiable limit built from a series)
For define
Each is and , so the anchor converges. The derivative is
For every , , and the numerical series converges. The Weierstrass M-test therefore gives uniform convergence of the derivative series. More concretely, after term its absolute tail is bounded uniformly by
The identity follows by writing and summing the geometric series and its weighted companion. The theorem now proves that the limit is and
In particular . This is a justified limit of derivatives; we did not infer it merely from convergence of the original series. The tail estimate also bounds the derivative error at every , not only at .
Worked example 5.3.4
(An endpoint where the derivative limit fails)
On let . The anchor converges: . The functions converge uniformly to , because . The derivatives are , which converge pointwise to on but equal at for every . The limit function is therefore discontinuous, so the convergence of is not uniform, and the theorem does not apply. The conclusion fails exactly at the point where uniformity fails: , whereas .
Worked example 5.3.5
(The theorem's bound versus the actual error)
On let and anchor at . Then , and , so . The theorem gives with and the estimate
Here the true error is , much smaller than the bound. The bound uses only the uniform derivative error and the anchor gap, so it holds for every sequence with those two sizes; it need not be sharp for a particular one.
What fails without each hypothesis
Counterexample mode
Derivatives alone cannot fix a height
Claim: if uniformly, then converges. Take on . Every equals , so the derivatives converge uniformly with zero error. Yet diverges, and hence has no finite pointwise limit at the anchor. The theorem does not apply. Knowing the slope perfectly cannot prevent a whole graph drifting upward; the repair is the hypothesis that converges for some .
The second hypothesis, uniform convergence of the derivatives, cannot be replaced by pointwise convergence even when everything else converges nicely.
Counterexample mode
Pointwise derivative convergence is too weak
Claim: if and pointwise, then . On set . The anchor tends to zero. Also
For nonzero fixed , the denominator tends to ; at zero every derivative is zero. This derivative limit is discontinuous, so convergence cannot be uniform: a uniform limit of continuous functions on the interval is continuous. In fact the functions themselves converge uniformly to , because
But has left derivative and right derivative at zero, so it has no derivative there. Even a convergent anchor and uniform convergence of the functions do not repair the missing uniform control of their derivatives.
Common mistake
(Uniform convergence of the functions does not control the derivatives)
The theorem runs from derivatives to functions, not back. Let on . Then , so uniformly, with . But , and at this equals , so the derivatives do not even converge pointwise. Fast, uniform convergence of functions says nothing about the slopes.
Checks
Quick check
(Suppose , , and . Find the limit and a uniform error bound.)
Use the anchor and .
Solution · Answer
Define . The proof above gives . The derivative limit is ; the value at 1 fixes the otherwise arbitrary additive constant.
Quick check
(Let on . Do the derivatives converge uniformly? Does the theorem apply to ?)
Compute and .
Solution · Answer
Since for every , the derivatives converge uniformly (with zero error) to . But does not converge, so no anchor exists and the theorem does not apply. Indeed has no limit at any point: it alternates between and .
Quick check
(Each is on with and , and pointwise. Must the limit of be differentiable with derivative ?)
Modify the example so that it vanishes at .
Solution · Answer
No. Take . Then , , and converges pointwise, but uniformly, since lies in (because ). The limit is not differentiable at . Bounded derivatives with pointwise convergence do not replace uniform convergence.