Integrating by parts gives the remainder in Taylor's theorem as an exact integral, from which Lagrange's and Cauchy's forms both follow by the mean value theorems for integrals.
In §150 Taylor’s theorem was proved with Rolle’s theorem, and the error after n terms came out as n!hnf(n)(a+θh) for some unknown θ between 0 and 1 (Lagrange’s form). §152 showed that this is not always good enough: for (1+x)m with x near −1, knowing only 0<θ<1 we cannot prove that the error tends to 0. §167 gives a second proof, by integration by parts, which produces the error exactly, as an integral with no unknown θ in it. From that integral, Lagrange’s form and a new one, Cauchy’s form, both follow; and Cauchy’s form settles the binomial series in the next lesson.
The idea
Taylor’s theorem is the fundamental theorem of the calculus, f(a+h)=f(a)+∫aa+hf′(x)dx, improved step by step. Integrate by parts once, moving a derivative onto f′, and a term hf′(a) splits off, leaving an integral of f′′. Again, and 21h2f′′(a) splits off. After n−1 steps,
So the remainder is the area under a definite curve on [0,1]. An area can be written as “length times a height reached by the curve” in more than one way, and the two classical forms of the remainder are two such ways.
Cauchy: replace the whole integrand by its value at one point t=θ. By the first mean value theorem the area is that value times 1.
Lagrange: keep the factor (1−t)n−1, which is never negative, as a weight, and replace only f(n)(a+th) by a value f(n)(a+θh). By the weighted mean value theorem this is allowed, and ∫01(1−t)n−1dt=1/n gives n!hnf(n)(a+θh).
ExploreOne remainder, two ways to average it
Try this.With log(1+x), n=3 and h near −0.8, compare the two numbers in the read-out (they agree), and the two θ’s. Then increase n, and move h to positive values.
f(x)=log(1+x), remainder after n=3 terms at h=−0.8Rn=(n−1)!hn∫01(1−t)n−1f(n)(th)dt, the area of the blue region (counted negative below the axis)
t
integrand; its shaded area is exactly Rₙ
Cauchy: rectangle of width 1, same area (height = integrand at θ)
Lagrange: the weight (1 − t)ⁿ⁻¹ scaled by f⁽ⁿ⁾(θh), same area; it crosses the integrand at θ
Function
f(h)−∑k<nk!f(k)(0)hk=−0.489438; area under the curve =−0.489438.
Cauchy’s form holds with θ=0.7601; Lagrange’s with θ=0.3702. Same remainder, different points.
What to notice
The blue area is the remainder itself: the read-out computes it also as f(h) minus the Taylor polynomial, and the two agree. The ochre rectangle and the dashed purple curve both have that same area, but they touch the integrand at different points, so Cauchy’s θ and Lagrange’s θ are different numbers. For negative h near −1 and large n, f(n)(th) is huge near t=1 while the factor (1−t)n−1 is tiny there, and their product stays small. Lagrange’s form separates the two factors (and, knowing only 0<θ<1, must allow for the huge values of f(n)); Cauchy’s form keeps them together at one point, which is what it exploits.
Precise statements
TheoremTaylor’s theorem with integral remainderHardy: §167 (1), (2)
Let f have continuous derivatives up to order n on the interval between a and a+h. Then
CorollarySchlömilch’s form; Lagrange’s and Cauchy’s formsHardy: §167 (3), (4)
Under the same hypotheses, for every integer p with 1≤p≤n there is a θ with 0<θ<1 such that
Rn=p(n−1)!(1−θ)n−pf(n)(a+θh)hn.
With p=n this is Lagrange’s formRn=n!hnf(n)(a+θh); with p=1 it is Cauchy’s form
Rn=(n−1)!(1−θ)n−1f(n)(a+θh)hn.
The θ depends on p (and on n and h).
Two remarks from Hardy. The integral form is exact: no undetermined number appears in it. On the other hand, as a proof of Lagrange’s form it is slightly weaker than the proof of §150, because it assumes that f(n) is continuous, while Rolle’s theorem needs only that f(n) exists. (The Rolle argument can be adapted to give the general forms too.) The name “Schlömilch’s form” for general p is the modern one; Hardy does not name it.
Proofs
ProofThe integral remainder
Strategy
Fix the end point b=a+h and consider the error of the Taylor polynomial about the moving point x, as a function Fn(x). It vanishes at x=b, and its derivative telescopes to a single term. Then Fn(a)=Fn(b)−∫abFn′, which is the remainder.
Then Fn(a)=Rn (it is the error of the expansion about a), and Fn(b)=0 (every term with a factor b−x vanishes, and f(b)−f(b)=0).
Differentiate: the sum telescopes
For k≥1, dxdk!(b−x)kf(k)(x)=k!(b−x)kf(k+1)(x)−(k−1)!(b−x)k−1f(k)(x). Adding these for k=0,…,n−1, each positive term cancels the negative term of the next, and
Fn′(x)=−(n−1)!(b−x)n−1f(n)(x).
Integrate
Fn′ is continuous (this is where the continuity of f(n) is used), so Fn(a)=Fn(b)−∫abFn′(x)dx=(n−1)!1∫ab(b−x)n−1f(n)(x)dx.
Change the variable
Put x=a+th, dx=hdt, so b−x=h(1−t), and t runs from 0 to 1: Rn=(n−1)!hn∫01(1−t)n−1f(n)(a+th)dt. (The argument works equally for h<0.)
∎
The telescoping derivative is integration by parts in disguise: differentiating Fn is the same as integrating by parts n−1 times, as in the idea above.
ProofSchlömilch’s form, hence Lagrange’s and Cauchy’s
Strategy
Split the factor (1−t)n−1 as (1−t)n−p⋅(1−t)p−1. Treat (1−t)p−1≥0 as a weight, and apply the generalised mean value theorem of §165 to the rest.
Choose the weight
Let 1≤p≤n. On [0,1] put ϕ(t)=(1−t)p−1≥0 and g(t)=(1−t)n−pf(n)(a+th), which is continuous. Then the integral in Rn is ∫01gϕ.
Generalised mean value theorem
By (9) of §165, ∫01gϕ=g(θ)∫01ϕ for some θ in [0,1], and ∫01(1−t)p−1dt=1/p. So
∫01(1−t)n−1f(n)(a+th)dt=p(1−θ)n−pf(n)(a+θh).
Multiplying by hn/(n−1)! gives the formula. (Hardy states 0<θ<1; the theorem as proved gives 0≤θ≤1, and the end points can be excluded by a slightly finer argument; nothing below depends on it.)
The two extreme cases
p=n: the weight is all of (1−t)n−1, and n(n−1)!1=n!1 gives Lagrange’s form. p=1: the weight is 1, nothing is kept, and the first mean value theorem gives Cauchy’s form.
∎
Worked examples and exercises
ExerciseTwo values of θ●○○Hardy: §167
For f(x)=ex, a=0, h=1 and n=2, the remainder is e−2. Find the θ in Lagrange’s form and the θ in Cauchy’s form (numerically for the second). Check with the widget.
Hint
Lagrange: e−2=21eθ. Cauchy: e−2=(1−θ)eθ.
Solution
Lagrange.eθ=2(e−2), so θ=log{2(e−2)}=log1.43656…=0.3623.
Cauchy.(1−θ)eθ=e−2=0.71828. The left side decreases from 1 to 0 on [0,1] (its derivative is −θeθ), so there is exactly one solution; bisection gives θ=0.6095. Same remainder, different points: the θ is not a property of the function but of the form chosen.
ExerciseThe logarithmic series at x=1●●○Hardy: §167; cf. §220
Using the integral form of the remainder, show that for 0<h≤1,
log(1+h)=h−2h2+3h3−…and in particularlog2=1−21+31−….Hint
f(n)(x)=(−1)n−1(n−1)!/(1+x)n, so Rn=(−1)n−1hn∫01(1+th)n(1−t)n−1dt.
Hint
For h>0, (1+th)n≥1.
Solution
The derivatives give f(k)(0)/k!=(−1)k−1/k for k≥1, so the Taylor polynomial is h−21h2+⋯+(−1)nhn−1/(n−1). By the first hint and the second, for 0<h≤1,
∣Rn∣≤hn∫01(1−t)n−1dt=nhn≤n1→0.
So the series converges to log(1+h), including at h=1. (Lagrange’s form would also work here: ∣Rn∣=n(1+θh)nhn≤nhn.)
ExerciseThe logarithmic series for −1<x<0●●○Hardy: §167; cf. §168
Show that the series of the previous exercise also represents log(1+h) when −1<h<0, using Cauchy’s form. Why does Lagrange’s form fail to prove this for −1<h<−21?
Hint
Cauchy’s form: Rn=(−1)n−1hn(1+θh)n(1−θ)n−1. Write it as (−1)n−1hn(1+θh1−θ)n−11+θh1.
Hint
For −1<h<0, show 1+θh1−θ<1, that is, 1−θ<1+θh, that is, −θ<θh.
Solution
Let −1<h<0 and put k=∣h∣<1. Since 0<θ<1, we have θh=−θk>−θ, so 1+θh>1−θ and 1+θh1−θ<1. Also 1+θh>1−k. Hence
∣Rn∣≤kn⋅1⋅1−k1→0,
a bound that does not involve θ. So the series converges to log(1+h).
Lagrange’s form gives only ∣Rn∣=n(1−θk)nkn, and all we know is 1−θk>1−k, so the best bound is n1(1−kk)n. When k>21, 1−kk>1, and this bound tends to infinity. The remainder does tend to 0; Lagrange’s form simply cannot see it, because it throws away the factor (1−θ)n−1 that Cauchy’s form keeps.
ExerciseBounds from bounds on the n-th derivative●●○Hardy: Misc. Examples VII, 4
Let F have continuous derivatives up to order n on [0,h], with F(0)=F′(0)=⋯=F(n−1)(0)=0 and A≤F(n)(x)≤B there. Show that An!xn≤F(x)≤Bn!xn for 0≤x≤h, and deduce Taylor’s theorem in Lagrange’s form.
Hint
The integral form with a=0 and the vanishing derivatives gives F(x)=(n−1)!1∫0x(x−t)n−1F(n)(t)dt.
Hint
For the deduction, apply the result to F(x)=f(x)−f(0)−xf′(0)−⋯−(n−1)!xn−1f(n−1)(0), with A and B the least and greatest values of f(n) on [0,h].
Solution
By the integral form, since all the Taylor terms of F at 0 vanish, F(x)=(n−1)!1∫0x(x−t)n−1F(n)(t)dt. The weight (x−t)n−1 is ≥0 on [0,x], so by (9) of §165 the integral lies between A∫0x(x−t)n−1dt and B∫0x(x−t)n−1dt, and ∫0x(x−t)n−1dt=xn/n. Dividing by (n−1)! gives Axn/n!≤F(x)≤Bxn/n!.
For Taylor’s theorem, take F as in the second hint: its first n−1 derivatives vanish at 0 and F(n)=f(n). At x=h, F(h) lies between Ahn/n! and Bhn/n!, so F(h)=ηhn/n! with A≤η≤B. Since f(n) is continuous it takes the value η at some ξ=θh, and F(h)=n!hnf(n)(θh), which is Lagrange’s form.
Check yourself
Check yourselfTrue or false?
For given f, a, h and n, the θ in Lagrange’s form and the θ in Cauchy’s form are the same number.
Why
False. Both forms equal the same remainder, but they are different expressions in θ, so they are satisfied at different points. For ex with h=1, n=2: Lagrange’s θ≈0.362, Cauchy’s θ≈0.610. Only for n=1 do the two forms coincide.
Check yourselfChoose one
In the general form Rn=p(n−1)!(1−θ)n−pf(n)(a+θh)hn, which value of p gives Cauchy’s form?
Why
With p=1 the weight (1−t)p−1 is 1, the whole integrand is replaced by its value at θ, and the factor (1−θ)n−1 survives in the answer. p=n gives Lagrange’s form. p=0 is not allowed: the weight (1−t)−1 has no finite integral.
Check yourselfChoose one
Compared with the proof of Taylor’s theorem by Rolle’s theorem (§150), what extra hypothesis does the integration-by-parts proof use?
Why
The integral ∫01(1−t)n−1f(n)(a+th)dt is defined (in the sense of §161) for continuous integrands, and the step Fn(a)=−∫abFn′ needs Fn′ continuous. Rolle’s theorem needs only that f(n) exists. In return, the integral proof gives an exact formula. It works for h<0 as well.
Check yourselfEnter a number
For f(x)=ex, a=0, h=1 and n=1, find θ with e−1=eθ (the forms coincide when n=1).
Why
θ=log(e−1)=log1.71828≈0.5413. This is just the mean value theorem f(1)−f(0)=f′(θ).
Hardy’s language
This proof of Taylor’s theorem has the advantage of leading to an exact formula for Rn, viz. (2), which does not involve an undetermined number θ.
— G. H. Hardy, §167
Summary
Summary
If f(n) is continuous, the remainder after n terms is exactly Rn=(n−1)!hn∫01(1−t)n−1f(n)(a+th)dt.
The proof: the error of the expansion about a moving point x has a telescoping derivative; equivalently, repeated integration by parts.
Splitting (1−t)n−1 into a weight (1−t)p−1 and the rest, the weighted mean value theorem gives a form for each p: p=n is Lagrange’s, p=1 is Cauchy’s.
Cauchy’s form keeps the factor (1−θ)n−1, which beats a large 1/(1+θh)n when h is negative; this is why it succeeds where Lagrange’s fails.