Chapter VII · Additional theorems in the differential and integral calculus

Taylor's theorem: the higher mean value theorems

§§150–151 40 min

How well nn derivatives at one point predict a function nearby, with the error given exactly (Lagrange) or as o(hn)o(h^n).

Why this matters

The mean value theorem of §126 says f(a+h)=f(a)+hf′(ξ)f(a + h) = f(a) + h f'(\xi) for some ξ\xi between aa and a+ha + h. It is the tangent-line approximation with an honest account of its error. This lesson asks what happens if we know more derivatives: can f(a),f′(a),…,f(n−1)(a)f(a), f'(a), \dots, f^{(n-1)}(a) predict f(a+h)f(a + h), and how large is the error? The answer, Taylor’s theorem, is the main tool of the whole first half of the chapter: power series, maxima and minima, limits of the form 0/00/0 and the contact of curves all come from it.

The idea

Start from the mean value theorem. It says that the gap between f(a+h)f(a + h) and the constant prediction f(a)f(a) is hh times some value of f′f' between aa and a+ha + h. We do not know which value, but if we know the largest and smallest values of f′f' there, we know a range in which f(a+h)f(a + h) must lie.

Now use one more derivative. The natural prediction is the tangent line f(a)+hf′(a)f(a) + h f'(a). Its error turns out to be 12h2\tfrac12 h^2 times some value of f′′f'' between aa and a+ha + h. With nn derivatives, the prediction is the Taylor polynomial

Pn−1(a+h)=f(a)+hf′(a)+h22!f′′(a)+⋯+hn−1(n−1)!f(n−1)(a),P_{n-1}(a + h) = f(a) + h f'(a) + \frac{h^2}{2!} f''(a) + \dots + \frac{h^{n-1}}{(n-1)!} f^{(n-1)}(a),

which is the polynomial of degree n−1n - 1 that has the same value and the same first n−1n - 1 derivatives as ff at aa. Its error, the remainder RnR_n, is hn/n!h^n/n! times some value of f(n)f^{(n)} between aa and a+ha + h. So f(a+h)f(a + h) is trapped in a band whose width is controlled by how much f(n)f^{(n)} varies.

Explore The band in which ff must lie

Try this. Start with n=1n = 1: the band is f(a)+h [min⁡f′,max⁡f′]f(a) + h\,[\min f', \max f'], the mean value theorem. Raise nn to 22, 33, 44 and watch the band close in around ff. Then drag hh and read off where ξ\xi sits and the ratio θn=(ξ−a)/h\theta_n = (\xi - a)/h.

f(x)=exf(x) = e^x, centre a=0a = 0, polynomial P0P_{0} (degree 0), remainder term of order n=1n = 1
xy
  • f(x)
  • Taylor polynomial P₀
  • band allowed by Lagrange's remainder
  • remainder R₁ at x = a + h
  • ξ (on the x-axis)
At x=a+h=1.5x = a + h = 1.5: f(x)=4.48169f(x) = 4.48169, P0(x)=1P_{0}(x) = 1, remainder R1=3.482R_{1} = 3.482.
The band at this xx runs from 2.52.5 to 7.72257.7225, and f(x)f(x) lies in it. Lagrange: R1=h11!f(1)(ξ)R_{1} = \frac{h^{1}}{1!} f^{(1)}(\xi) with ξ=0.84205\xi = 0.84205, so θ1=(ξ−a)/h=0.5614\theta_{1} = (\xi - a)/h = 0.5614 (compare 1/(n+1)=0.51/(n+1) = 0.5).
§151 form: 1! R1h1=2.32113=f(1)(a)+η\frac{1!\,R_{1}}{h^{1}} = 2.32113 = f^{(1)}(a) + \eta with η=1.32\eta = 1.32.
What to notice

However you set things, the blue curve never leaves the green band: that is the theorem. For small hh the band is very thin, because its width is a multiple of hnh^n. And ξ\xi is a genuine point between aa and a+ha + h; it moves as hh changes, so the theorem tells us it exists but not where.

Two cautions are already visible. The point ξ\xi depends on hh (and on nn and ff): it is not a fixed number. And the formula with f(n)(ξ)f^{(n)}(\xi) is useful only when we can say something about f(n)f^{(n)} on the whole interval, usually a bound.

Precise statements

Hardy first proves the case of two derivatives, which he calls the mean value theorem of the second order, and then the general case.

Theorem Mean value theorem of the second order Hardy: §150 (2)

Suppose that f′f' is continuous on [a,a+h][a, a + h] and that f′′f'' exists on the open interval between aa and a+ha + h. Then

f(a+h)=f(a)+hf′(a)+12h2f′′(a+θ2h)for some 0<θ2<1.f(a + h) = f(a) + h f'(a) + \tfrac12 h^2 f''(a + \theta_2 h) \qquad \text{for some } 0 < \theta_2 < 1.
Theorem Taylor's theorem, with Lagrange's remainder Hardy: §150, the general mean value theorem

Suppose that f(n−1)f^{(n-1)} is continuous on the closed interval with end points aa and a+ha + h, and that f(n)f^{(n)} exists on the open interval. Then

f(a+h)=f(a)+hf′(a)+⋯+hn−1(n−1)!f(n−1)(a)+hnn!f(n)(a+θnh)f(a + h) = f(a) + h f'(a) + \dots + \frac{h^{n-1}}{(n-1)!} f^{(n-1)}(a) + \frac{h^n}{n!} f^{(n)}(a + \theta_n h)

for some θn\theta_n with 0<θn<10 < \theta_n < 1. Equivalently, with b=a+hb = a + h, the last term is (b−a)nn!f(n)(ξ)\dfrac{(b - a)^n}{n!} f^{(n)}(\xi) with ξ\xi strictly between aa and bb.

For n=1n = 1 this is the mean value theorem itself. When ff is only defined on [a,b][a, b], the derivatives at the end points are understood one-sidedly, as Hardy remarks.

§151 proves a second form that assumes much less: derivatives up to order nn at the single point aa (which forces f(n−1)f^{(n-1)} to exist near aa, but says nothing about f(n)f^{(n)} anywhere else). In exchange it says less: only that the error is small compared with hnh^n.

Theorem Taylor's theorem, with the remainder o(hn)o(h^n) Hardy: §151 (1), (2)

If ff has an nn-th derivative at x=ax = a, then

f(a+h)=f(a)+hf′(a)+⋯+hn−1(n−1)!f(n−1)(a)+hnn!{f(n)(a)+η},f(a + h) = f(a) + h f'(a) + \dots + \frac{h^{n-1}}{(n-1)!} f^{(n-1)}(a) + \frac{h^n}{n!}\bigl\{f^{(n)}(a) + \eta\bigr\},

where η→0\eta \to 0 as h→0h \to 0. In the order notation of §98, f(a+h)=∑k=0nhkk!f(k)(a)+o(hn)f(a + h) = \sum_{k=0}^{n} \dfrac{h^k}{k!} f^{(k)}(a) + o(h^n).

The two forms answer different questions. Lagrange’s form is a statement at a fixed hh, and with a bound on f(n)f^{(n)} it gives a numerical error bound. The o(hn)o(h^n) form is a statement about h→0h \to 0 only, and it is exactly what the applications in §§153–155 need.

Proofs

The proof of Taylor’s theorem is the proof of the mean value theorem with a cleverer auxiliary function. Write b=a+hb = a + h.

Proof Taylor's theorem with Lagrange's remainder
Strategy

Build a function ϕ\phi on [a,b][a, b] that vanishes at both ends, so that Rolle’s theorem gives a point where ϕ′=0\phi' = 0, and arrange that ϕ′(ξ)=0\phi'(\xi) = 0 says exactly what we want. The key is to expand about the moving point xx and predict the fixed value f(b)f(b).

Let Fn(x)F_n(x) be the error in predicting f(b)f(b) from xx

Define

Fn(x)=f(b)−f(x)−(b−x)f′(x)−⋯−(b−x)n−1(n−1)!f(n−1)(x).F_n(x) = f(b) - f(x) - (b - x) f'(x) - \dots - \frac{(b - x)^{n-1}}{(n-1)!} f^{(n-1)}(x).

Then Fn(b)=0F_n(b) = 0, and Fn(a)F_n(a) is exactly the remainder RnR_n that we want to describe.

Its derivative collapses to one term

Differentiate the general term: ddx{(b−x)kk!f(k)(x)}=(b−x)kk!f(k+1)(x)−(b−x)k−1(k−1)!f(k)(x)\dfrac{d}{dx}\Bigl\{\dfrac{(b-x)^k}{k!} f^{(k)}(x)\Bigr\} = \dfrac{(b-x)^k}{k!} f^{(k+1)}(x) - \dfrac{(b-x)^{k-1}}{(k-1)!} f^{(k)}(x). Summing over k=0,…,n−1k = 0, \dots, n-1, the terms cancel in pairs (the sum telescopes), leaving

Fn′(x)=−(b−x)n−1(n−1)!f(n)(x)(a<x<b).F_n'(x) = -\frac{(b - x)^{n-1}}{(n-1)!} f^{(n)}(x) \qquad (a < x < b).

Correct FnF_n so that it also vanishes at aa

Let ϕ(x)=Fn(x)−(b−xb−a)nFn(a)\phi(x) = F_n(x) - \Bigl(\dfrac{b - x}{b - a}\Bigr)^n F_n(a). Then ϕ(a)=Fn(a)−Fn(a)=0\phi(a) = F_n(a) - F_n(a) = 0 and ϕ(b)=0−0=0\phi(b) = 0 - 0 = 0. The hypotheses make ϕ\phi continuous on [a,b][a, b] (it involves f,…,f(n−1)f, \dots, f^{(n-1)}) and differentiable on (a,b)(a, b) (it involves f(n)f^{(n)} there).

Apply Rolle's theorem

There is a ξ\xi with a<ξ<ba < \xi < b and

0=ϕ′(ξ)=−(b−ξ)n−1(n−1)!f(n)(ξ)+n(b−ξ)n−1(b−a)nFn(a).0 = \phi'(\xi) = -\frac{(b - \xi)^{n-1}}{(n-1)!} f^{(n)}(\xi) + \frac{n (b - \xi)^{n-1}}{(b - a)^n} F_n(a).

Solve for the remainder

Since b−ξ≠0b - \xi \ne 0 we may divide by (b−ξ)n−1(b - \xi)^{n-1}, which gives Fn(a)=(b−a)nn!f(n)(ξ)F_n(a) = \dfrac{(b - a)^n}{n!} f^{(n)}(\xi). Writing b=a+hb = a + h and ξ=a+θnh\xi = a + \theta_n h gives the theorem. If h<0h < 0, the same argument works on [a+h,a][a + h, a].

For the o(hn)o(h^n) form we cannot use Rolle’s theorem on f(n)f^{(n)}, because f(n)f^{(n)} may exist only at aa. Hardy’s idea is to push the information about the nn-th derivative at one point down through the lower derivatives, one at a time.

Proof Taylor's theorem with remainder o(hn)o(h^n)
Strategy

Subtract the whole Taylor polynomial of degree nn. The difference gg has g(0)=g′(0)=⋯=g(n)(0)=0g(0) = g'(0) = \dots = g^{(n)}(0) = 0. We show that such a gg satisfies ∣g(h)∣≤ε∣h∣n|g(h)| \le \varepsilon |h|^n near 00, for any ε>0\varepsilon > 0, by applying the mean value theorem n−1n - 1 times.

Set up gg

Let g(h)=f(a+h)−∑k=0nhkk!f(k)(a)g(h) = f(a + h) - \sum_{k=0}^{n} \dfrac{h^k}{k!} f^{(k)}(a). Then g(k)(0)=0g^{(k)}(0) = 0 for k=0,1,…,nk = 0, 1, \dots, n. Because g(n)(0)g^{(n)}(0) exists, g(n−1)g^{(n-1)} exists on some interval around 00, and so do g,g′,…,g(n−2)g, g', \dots, g^{(n-2)}, each continuous there. We must show g(h)/hn→0g(h)/h^n \to 0.

The top derivative is small compared with hh

By the definition of the derivative, g(n−1)(h)/h=(g(n−1)(h)−g(n−1)(0))/h→g(n)(0)=0g^{(n-1)}(h)/h = \bigl(g^{(n-1)}(h) - g^{(n-1)}(0)\bigr)/h \to g^{(n)}(0) = 0. So, given ε>0\varepsilon > 0, there is a δ>0\delta > 0 with ∣g(n−1)(t)∣≤ε∣t∣|g^{(n-1)}(t)| \le \varepsilon |t| whenever ∣t∣<δ|t| < \delta.

Pass the bound down, one derivative at a time

Let ∣h∣<δ|h| < \delta. Since g(n−2)(0)=0g^{(n-2)}(0) = 0, the mean value theorem gives g(n−2)(h)=h g(n−1)(t)g^{(n-2)}(h) = h\, g^{(n-1)}(t) with ∣t∣<∣h∣|t| < |h|, so ∣g(n−2)(h)∣≤∣h∣⋅ε∣h∣=ε∣h∣2|g^{(n-2)}(h)| \le |h| \cdot \varepsilon |h| = \varepsilon |h|^2. Repeating with g(n−3),…,gg^{(n-3)}, \dots, g (each vanishes at 00) gives ∣g(n−3)(h)∣≤ε∣h∣3|g^{(n-3)}(h)| \le \varepsilon |h|^3, …, ∣g(h)∣≤ε∣h∣n|g(h)| \le \varepsilon |h|^n.

Conclude

So ∣g(h)/hn∣≤ε|g(h)/h^n| \le \varepsilon for 0<∣h∣<δ0 < |h| < \delta, and ε\varepsilon was arbitrary: g(h)=o(hn)g(h) = o(h^n). Setting η=n! g(h)/hn\eta = n!\,g(h)/h^n gives exactly the form in the theorem, with η→0\eta \to 0.

Hardy’s own version reaches the same conclusion with one-sided inequalities: he compares Fn(h)F_n(h) with hnn!{f(n)(a)±δ}\tfrac{h^n}{n!}\{f^{(n)}(a) \pm \delta\} and uses the fact that a function with positive derivative at 00 is increasing there (§122). The idea, information flowing down from f(n)(a)f^{(n)}(a) to ff, is the same.

Explore Where is ξ\xi as h→0h \to 0?

Try this. With n=2n = 2, drag hh slowly towards 00 and watch θ2\theta_2 and η\eta in the read-out. Repeat with n=3n = 3. Then move the centre to a=0a = 0, set n=1n = 1, and shrink hh again.

f(x)=sin⁡xf(x) = \sin x, centre a=0.6a = 0.6, polynomial P1P_{1} (degree 1), remainder term of order n=2n = 2
xy
  • f(x)
  • Taylor polynomial P₁
  • band allowed by Lagrange's remainder
  • remainder R₂ at x = a + h
  • ξ (on the x-axis)
At x=a+h=1.8x = a + h = 1.8: f(x)=0.973848f(x) = 0.973848, P1(x)=1.55505P_{1}(x) = 1.55505, remainder R2=−0.5812R_{2} = -0.5812.
The band at this xx runs from 0.835050.83505 to 1.14851.1485, and f(x)f(x) lies in it. Lagrange: R2=h22!f(2)(ξ)R_{2} = \frac{h^{2}}{2!} f^{(2)}(\xi) with ξ=0.93943\xi = 0.93943, so θ2=(ξ−a)/h=0.2829\theta_{2} = (\xi - a)/h = 0.2829 (compare 1/(n+1)=0.33331/(n+1) = 0.3333).
§151 form: 2! R2h2=−0.807219=f(2)(a)+η\frac{2!\,R_{2}}{h^{2}} = -0.807219 = f^{(2)}(a) + \eta with η=−0.243\eta = -0.243.
What to notice

For small hh, θn\theta_n settles near 1/(n+1)1/(n+1), and η\eta (the §151 error) tends to 00. At a=0a = 0 with n=1n = 1 the rule fails: θ1\theta_1 tends to 1/3≈0.5771/\sqrt3 \approx 0.577, because f′′(0)=−sin⁡0=0f''(0) = -\sin 0 = 0. The exercise below explains both.

Worked examples and exercises

Exercise The sine with a remainder ●○○ Hardy: Examples LV, 3

Write down Taylor’s theorem for sin⁡(x+h)\sin(x + h) with the remainder after the term in h2n−1h^{2n-1}, and deduce that the polynomial part approximates sin⁡(x+h)\sin(x + h) with error at most ∣h∣2n/(2n)!|h|^{2n}/(2n)!.

Hint
The derivatives of sin⁡\sin cycle: sin⁡,cos⁡,−sin⁡,−cos⁡\sin, \cos, -\sin, -\cos. In general sin⁡(k)x=sin⁡(x+12kπ)\sin^{(k)} x = \sin(x + \tfrac12 k\pi).

Solution

Apply the theorem with 2n2n in place of nn:

sin⁡(x+h)=sin⁡x+hcos⁡x−h22!sin⁡x−⋯+(−1)n−1h2n−1(2n−1)!cos⁡x+(−1)nh2n(2n)!sin⁡(x+θ2nh).\sin(x + h) = \sin x + h \cos x - \frac{h^2}{2!}\sin x - \dots + (-1)^{n-1}\frac{h^{2n-1}}{(2n-1)!}\cos x + (-1)^n \frac{h^{2n}}{(2n)!} \sin(x + \theta_{2n} h).

The remainder is (−1)nh2n(2n)!sin⁡(x+θ2nh)(-1)^n \dfrac{h^{2n}}{(2n)!}\sin(x + \theta_{2n}h), and ∣sin⁡∣≤1|\sin| \le 1, so the error is at most ∣h∣2n/(2n)!|h|^{2n}/(2n)!. The same works for cos⁡(x+h)\cos(x + h) with cos⁡(x+θh)\cos(x + \theta h) in the remainder.

Exercise Taylor's theorem for 1/x1/x ●●○ Hardy: Examples LV, 2

Let xx and x+hx + h be positive. Show that

1x+h=1x−hx2+⋯+(−1)n−1hn−1xn+(−1)nhn(x+θnh)n+1,0<θn<1,\frac{1}{x + h} = \frac1x - \frac{h}{x^2} + \dots + \frac{(-1)^{n-1} h^{n-1}}{x^n} + \frac{(-1)^n h^n}{(x + \theta_n h)^{n+1}}, \qquad 0 < \theta_n < 1,

and check the result directly, without Taylor’s theorem.

Hint
The kk-th derivative of 1/x1/x is (−1)kk!/xk+1(-1)^k k!/x^{k+1}. That gives the formula at once.
Hint
For the direct check, sum the geometric progression, and ask whether xn(x+h)x^n(x + h) lies between xn+1x^{n+1} and (x+h)n+1(x + h)^{n+1}.

Solution

From the theorem. With f(t)=1/tf(t) = 1/t we have f(k)(t)/k!=(−1)k/tk+1f^{(k)}(t)/k! = (-1)^k/t^{k+1}, so the terms are (−1)khk/xk+1(-1)^k h^k/x^{k+1} and the remainder is (−1)nhn/(x+θnh)n+1(-1)^n h^n/(x + \theta_n h)^{n+1}. The hypotheses hold because 1/t1/t has derivatives of every order on the interval between xx and x+hx + h, which does not contain 00.

Directly. The sum of the first nn terms is a geometric progression with ratio −h/x-h/x:

∑k=0n−1(−h)kxk+1=1−(−h/x)nx+h,so1x+h−∑k=0n−1(−h)kxk+1=(−1)nhnxn(x+h).\sum_{k=0}^{n-1} \frac{(-h)^k}{x^{k+1}} = \frac{1 - (-h/x)^n}{x + h}, \qquad\text{so}\qquad \frac{1}{x+h} - \sum_{k=0}^{n-1} \frac{(-h)^k}{x^{k+1}} = \frac{(-1)^n h^n}{x^n (x + h)}.

We need xn(x+h)=(x+θh)n+1x^n(x + h) = (x + \theta h)^{n+1} for some 0<θ<10 < \theta < 1. If h>0h > 0 then xn+1<xn(x+h)<(x+h)n+1x^{n+1} < x^n(x + h) < (x + h)^{n+1}; if h<0h < 0 the inequalities reverse. The function θ↦(x+θh)n+1\theta \mapsto (x + \theta h)^{n+1} is continuous and runs from xn+1x^{n+1} to (x+h)n+1(x + h)^{n+1}, so by the intermediate value theorem it takes the value xn(x+h)x^n(x + h) at some θ\theta strictly between 00 and 11.

Exercise Newton's method ●●○ Hardy: Examples LV, 7–8

Let ξ\xi be an approximation to a root ξ+h\xi + h of f(x)=0f(x) = 0, with f′(ξ)≠0f'(\xi) \ne 0. Use the second-order mean value theorem to show that ξ1=ξ−f(ξ)/f′(ξ)\xi_1 = \xi - f(\xi)/f'(\xi) is in error by O(h2)O(h^2). Apply this to x2=2x^2 = 2 starting from ξ=32\xi = \tfrac32, twice.

Hint
0=f(ξ+h)=f(ξ)+hf′(ξ)+12h2f′′(ξ+θh)0 = f(\xi + h) = f(\xi) + h f'(\xi) + \tfrac12 h^2 f''(\xi + \theta h). Divide by f′(ξ)f'(\xi).
Hint
For f(x)=x2−2f(x) = x^2 - 2 the step is ξ1=ξ−(ξ2−2)/(2ξ)=12(ξ+2/ξ)\xi_1 = \xi - (\xi^2 - 2)/(2\xi) = \tfrac12(\xi + 2/\xi).

Solution

Dividing the hint’s equation by f′(ξ)f'(\xi) gives ξ+h=ξ−f(ξ)/f′(ξ)−12h2f′′(ξ+θh)/f′(ξ)\xi + h = \xi - f(\xi)/f'(\xi) - \tfrac12 h^2 f''(\xi + \theta h)/f'(\xi), so the root differs from ξ1\xi_1 by 12h2f′′(ξ+θh)/f′(ξ)\tfrac12 h^2 f''(\xi + \theta h)/f'(\xi). If ∣f′′∣≤M|f''| \le M and ∣f′∣≥K>0|f'| \ge K > 0 near a simple root, this is at most Mh2/(2K)Mh^2/(2K): the error is squared at each step (up to a constant), and the next steps have errors O(h4)O(h^4), O(h8)O(h^8), ….

For x2=2x^2 = 2: ξ1=12(32+43)=1712=1.41666…\xi_1 = \tfrac12(\tfrac32 + \tfrac43) = \tfrac{17}{12} = 1.41666\ldots, and ξ2=12(1712+2417)=577408=1.4142157…\xi_2 = \tfrac12\bigl(\tfrac{17}{12} + \tfrac{24}{17}\bigr) = \tfrac{577}{408} = 1.4142157\ldots. Since 2=1.4142136…\sqrt2 = 1.4142136\ldots, ξ2\xi_2 is correct to five decimal places, although the start 32\tfrac32 was correct to none.

Exercise Difference quotients ●●○ Hardy: Examples LVI, 2–3

Show that if f′(a)f'(a) exists then f(a+h)−f(a−h)2h→f′(a)\dfrac{f(a + h) - f(a - h)}{2h} \to f'(a), and that if f′′(a)f''(a) exists then f(a+h)−2f(a)+f(a−h)h2→f′′(a)\dfrac{f(a + h) - 2f(a) + f(a - h)}{h^2} \to f''(a), as h→0h \to 0. Do the converses hold?

Hint
Use the §151 form with n=1n = 1, and then with n=2n = 2, once for +h+h and once for −h-h.

Solution

With n=1n = 1: f(a±h)=f(a)±h{f′(a)+η±}f(a \pm h) = f(a) \pm h\{f'(a) + \eta_\pm\} where η±→0\eta_\pm \to 0. Subtracting, f(a+h)−f(a−h)=2hf′(a)+h(η++η−)f(a + h) - f(a - h) = 2h f'(a) + h(\eta_+ + \eta_-); divide by 2h2h and let h→0h \to 0.

With n=2n = 2: f(a±h)=f(a)±hf′(a)+12h2{f′′(a)+η±}f(a \pm h) = f(a) \pm h f'(a) + \tfrac12 h^2 \{f''(a) + \eta_\pm\}. Adding, the f′(a)f'(a) terms cancel: f(a+h)−2f(a)+f(a−h)=h2f′′(a)+12h2(η++η−)f(a + h) - 2f(a) + f(a - h) = h^2 f''(a) + \tfrac12 h^2(\eta_+ + \eta_-); divide by h2h^2.

The converses fail. For f(x)=∣x∣f(x) = |x| at a=0a = 0 the first quotient is 00 for every hh, but f′(0)f'(0) does not exist. For f(x)=x∣x∣f(x) = x|x|, which has f′(0)=0f'(0) = 0 but no f′′(0)f''(0), the second quotient is 00 for every hh.

Exercise Where ξ\xi goes as h→0h \to 0 ●●● Hardy: Examples LV, 12

Suppose that f(n+1)f^{(n+1)} is continuous near aa and f(n+1)(a)≠0f^{(n+1)}(a) \ne 0. Show that the θn\theta_n of Lagrange’s form tends to 1/(n+1)1/(n+1) as h→0h \to 0.

Hint
Write f(a+h)f(a + h) with Lagrange’s remainder of order nn and of order n+1n + 1, and compare: f(n)(a+θnh)−f(n)(a)=hn+1f(n+1)(a+θn+1h)f^{(n)}(a + \theta_n h) - f^{(n)}(a) = \dfrac{h}{n+1} f^{(n+1)}(a + \theta_{n+1} h).
Hint
Apply the mean value theorem to the left side: it is θnh f(n+1)(a+θθnh)\theta_n h\, f^{(n+1)}(a + \theta\theta_n h).

Solution

Both forms describe the same number, so

hnn!f(n)(a+θnh)=hnn!f(n)(a)+hn+1(n+1)!f(n+1)(a+θn+1h).\frac{h^n}{n!} f^{(n)}(a + \theta_n h) = \frac{h^n}{n!} f^{(n)}(a) + \frac{h^{n+1}}{(n+1)!} f^{(n+1)}(a + \theta_{n+1} h).

Hence f(n)(a+θnh)−f(n)(a)=hn+1f(n+1)(a+θn+1h)f^{(n)}(a + \theta_n h) - f^{(n)}(a) = \dfrac{h}{n+1} f^{(n+1)}(a + \theta_{n+1}h). By the mean value theorem the left side is θnh f(n+1)(a+θθnh)\theta_n h\, f^{(n+1)}(a + \theta\theta_n h) with 0<θ<10 < \theta < 1. Cancelling hh,

θn=1n+1⋅f(n+1)(a+θn+1h)f(n+1)(a+θθnh).\theta_n = \frac{1}{n+1}\cdot\frac{f^{(n+1)}(a + \theta_{n+1}h)}{f^{(n+1)}(a + \theta\theta_n h)}.

As h→0h \to 0 both arguments tend to aa, and by continuity numerator and denominator tend to the same non-zero number f(n+1)(a)f^{(n+1)}(a). So θn→1/(n+1)\theta_n \to 1/(n+1).

The condition f(n+1)(a)≠0f^{(n+1)}(a) \ne 0 is needed (Hardy leaves it implicit). For sin⁡\sin at 00 with n=1n = 1, sin⁡h=hcos⁡(θ1h)\sin h = h\cos(\theta_1 h) gives cos⁡(θ1h)=(sin⁡h)/h=1−h2/6+…\cos(\theta_1 h) = (\sin h)/h = 1 - h^2/6 + \dots, and comparing with 1−θ12h2/21 - \theta_1^2 h^2/2 shows θ1→1/3\theta_1 \to 1/\sqrt3, as the second exploration shows.

Check yourself

Check yourself Choose one

In f(a+h)=f(a)+hf′(a)+12h2f′′(ξ)f(a + h) = f(a) + h f'(a) + \tfrac12 h^2 f''(\xi), what is ξ\xi?

Why

Rolle’s theorem only guarantees that a suitable ξ\xi exists in the open interval. It moves with hh (drag hh in the widget), and is the midpoint only by accident; for small hh it is near a+h/3a + h/3 when f′′′(a)≠0f'''(a) \ne 0. It is not where f′′f'' is largest: the remainder uses a value between the smallest and the largest.

Check yourself True or false?

Taylor’s theorem says that the remainder after nn terms is hnn!f(n)(a)\dfrac{h^n}{n!} f^{(n)}(a).

Why

False. That would make ff equal to a polynomial of degree nn. Lagrange’s form has f(n)f^{(n)} at an unknown point a+θnha + \theta_n h; the §151 form has f(n)(a)+ηf^{(n)}(a) + \eta with η→0\eta \to 0. Both say “approximately hnn!f(n)(a)\frac{h^n}{n!} f^{(n)}(a) for small hh”, which is a different statement.

Check yourself Enter a number

Approximate e1/2e^{1/2} by 1+12+181 + \tfrac12 + \tfrac18 (Taylor’s theorem for exe^x about 00 with n=3n = 3). The remainder is h33!eξ\tfrac{h^3}{3!} e^{\xi} with h=12h = \tfrac12 and 0<ξ<120 < \xi < \tfrac12. Using only eξ<2e^\xi < 2, what upper bound for the error does this give?

Why

(1/2)36⋅2=124≈0.0417\tfrac{(1/2)^3}{6} \cdot 2 = \tfrac{1}{24} \approx 0.0417. The true error is e1/2−1.625=0.0237…e^{1/2} - 1.625 = 0.0237\ldots, comfortably below the bound. A bound on f(n)f^{(n)} over the interval is exactly what Lagrange’s form needs.

Check yourself Choose one

Which hypothesis is enough for f(a+h)=∑k=0nhkk!f(k)(a)+o(hn)f(a + h) = \sum_{k=0}^{n} \frac{h^k}{k!} f^{(k)}(a) + o(h^n) as h→0h \to 0?

Why

That is the point of §151: one derivative of order nn at one point suffices. Continuity alone gives nothing beyond f(a+h)=f(a)+o(1)f(a + h) = f(a) + o(1). The stronger hypotheses also work, but they are not needed; and even infinitely many derivatives do not make the series converge to ff (next lesson).

Check yourself True or false?

Taylor’s theorem with Lagrange’s remainder, in the case n=1n = 1, is the mean value theorem.

Why

True. With n=1n = 1 the polynomial part is just f(a)f(a) and the remainder is hf′(a+θ1h)h f'(a + \theta_1 h). The proof also reduces to the usual one: ϕ(x)=f(b)−f(x)−b−xb−a{f(b)−f(a)}\phi(x) = f(b) - f(x) - \frac{b - x}{b - a}\{f(b) - f(a)\}.

Hardy’s language

But we do not assume even the existence of the nth derivative at any point except x = a.

— G. H. Hardy, §151

Summary

Summary
  • Taylor polynomial: Pn−1(a+h)=∑k=0n−1hkk!f(k)(a)P_{n-1}(a + h) = \sum_{k=0}^{n-1} \frac{h^k}{k!} f^{(k)}(a), the polynomial matching ff and its first n−1n-1 derivatives at aa.
  • Lagrange: f(a+h)=Pn−1(a+h)+hnn!f(n)(a+θnh)f(a + h) = P_{n-1}(a + h) + \frac{h^n}{n!} f^{(n)}(a + \theta_n h), 0<θn<10 < \theta_n < 1. A bound on f(n)f^{(n)} gives a bound on the error.
  • Proof: Rolle’s theorem applied to Fn(x)−(b−xb−a)nFn(a)F_n(x) - \bigl(\frac{b-x}{b-a}\bigr)^n F_n(a), where Fn(x)F_n(x) is the error in predicting f(b)f(b) from xx.
  • With only f(n)(a)f^{(n)}(a) assumed: f(a+h)=Pn(a+h)+o(hn)f(a + h) = P_n(a + h) + o(h^n), proved by passing the smallness down one derivative at a time.
  • The point ξ\xi moves with hh; if f(n+1)(a)≠0f^{(n+1)}(a) \ne 0 it approaches a+h/(n+1)a + h/(n+1).