S. D. V. Stephens

notes / Studies in Real and Complex Analysis / chapter 5: Analysis in One Variable

20 Differentiation, Taylor’s Theorem and the Riemann Integral

· Denis Auroux

Original Professor Auroux’s handwritten notes (PDF), behind a password for now:

Differentiation in one variable

(Rudin Chapter 5 McMullen §5.)

Definition 20.1. is differentiable at if i.e. such that .

Proposition 20.2. differentiable at continuous at . (The converse is false, e.g. at .)

Proof. where the first factor as and the second as ; since multiplication is continuous, . ∎

The usual rules of calculation hold: derivatives of , , …; the chain rule (see Rudin p. 104–105).

Example 20.3.

  • (), . For , . is continuous but not differentiable at ().
Graph of x sin(1/x), oscillating faster and faster near the origin between the dashed lines y = x and y = -x.
  • , : is differentiable () but is not continuous at .
Graph of x^2 sin(1/x), oscillating near the origin between the dashed parabolas y = x^2 and y = -x^2.
  • is continuous (the series converges uniformly, since converges) and nowhere differentiable! (See also Rudin 7.18 for a related example.)

Theorem 20.4 (Mean value theorem). differentiable such that .

This follows logically from easier results:

  1. If has a local max (or min) at (i.e. a max of ) and is differentiable at , then . (Because is for and for : take the limit as from the left and from the right.)
  2. If is differentiable and , then such that . This is clear if is constant; otherwise look at the max or min of over and apply (1).
  3. The mean value theorem: apply (2) to .

Corollary 20.5 (Mean value inequality). If for all , then .

Generalization:

Theorem 20.6 (Taylor’s theorem). Let be times differentiable. The degree Taylor polynomial of at is Then such that

Proof.

  • Subtracting from both sides, we can reduce to the case (and ).
  • Let . Then , and we still have .
  • Now: the mean value theorem for : such that . The mean value theorem for : such that . And so on until such that , i.e.

∎

Remark.

  • We can compare to by applying the theorem to instead!
  • As with the mean value inequality: a bound gives a bound over .

Remark. There exist nonzero functions whose Taylor polynomials are all zero! E.g. , : (all derivatives exist) and for all , so the Taylor polynomials are all zero! The Taylor series of at converges, but ! (In other examples, it can also have , i.e. never be convergent for .)

Most functions aren’t analytic, i.e. can’t be written as power series.

Spaces of functions

Let

Theorem 20.7. If , pointwise, and uniformly, then and (and uniformly).

Proof. Fix . The mean value theorem gives The left-hand side as .

For the right-hand side: has a subsequence converging to some . Since is continuous, the uniform limit is continuous; we claim . Indeed: fix , let be such that , and let be such that and . Then for , Hence, returning to and taking the limit as : such that We now take the limit as : the right-hand side , using continuity of and the fact that (check this!). Hence is differentiable at and (and since is continuous, ).

Finally, the mean value inequality gives a uniform bound, so . ∎

Corollary 20.8. is a complete metric space.

Proof. Using completeness of (uniform topology): if is Cauchy in , then and are Cauchy in , so there exist uniform limits , and by the theorem and . Now and uniformly, so in . This proves the case . Repeat the same argument for successive derivatives for . ∎

Corollary 20.9. If is a power series with radius of convergence , then is over , and .

Proof. and have the same radius of convergence, so the partial sums for both converge uniformly over compact subsets of ; hence and . Repeat for successive derivatives (, so , …). ∎

Integration

(The Riemann integral; see Math 114 for the Lebesgue integral and much more.)

Notes not reached in this lecture, covered at the start of Lecture 21

The definite integral of continuous functions is a linear operator for each (linear: and ), satisfying the axioms:

  1. If then (so if then ).
  2. If then .
  3. .

In fact, such a linear map is unique; the difference between different theories of integration is in how much more general functions we allow ourselves to integrate.

The Riemann integral starts from step functions: such that such that is constant over each , (the values at the don’t matter). Then 2) and 3) suggest we must have This definition of the integral for step functions satisfies the required axioms.

Next: if for step functions, then In particular, if is bounded, then fixing , we can take and , giving the lower and upper Riemann sums of for the given partition of . Refining (i.e. subdividing further) gives better bounds on :

A function f (red) on [a,b] squeezed between a coarse upper step function S and a coarse lower step function s (black), with a finer lower step function s' (blue, dashed) lying between s and f.

The lower and upper Riemann integrals: For all bounded , .

Definition 20.10. is Riemann integrable, , if ; we then set .

Theorem 20.11. Continuous functions are Riemann integrable.

Proof. The key ingredient is uniform continuity: such that for , . (Recall: this is proved by applying the Lebesgue number lemma to the open cover : such that such that .)

Thus: given , take as in uniform continuity, and split such that for all . Then and (attained) satisfy for all , and on .

A continuous function f (red) on [a,b] with a partition into subintervals of width less than delta; on each subinterval the bar heights S_i (max) and lines s_i (min) differ by less than epsilon.

Let be the step functions taking the values on . Then on , so and ; moreover, for all , so . Hence ∎

Remark.

  • Piecewise continuous functions are also integrable, and so are some stranger functions (see Rudin, and see the homework). However, for example, is not Riemann integrable (, ).

    The Lebesgue integral allows more general decompositions into “measurable” subsets (rather than just sub-intervals) and allows more general functions to be integrated (including unbounded functions, which are never Riemann integrable). (E.g. for Riemann integration, only makes sense as an “improper integral”, i.e. , whereas Lebesgue can handle this and worse.)

  • In fact, Lebesgue gave a characterization of exactly which functions are Riemann integrable: iff is bounded on and the set of points where is discontinuous has Lebesgue measure 0, which means: there exists an at most countable collection of open intervals such that and .

  • It is easy to check (do it!) that is a vector space, and is linear and satisfies the above axioms.

Theorem 20.12 (Fundamental theorem of calculus). If is continuous on then is differentiable and .

Proof. using continuity of at to estimate the integral for . ∎