Chapter VII · Additional theorems in the differential and integral calculus

Chapter VII problems

Miscellaneous examples 120 min

Eleven of the Miscellaneous Examples on Chapter VII, on Taylor expansions, functions of several variables and definite integrals, with graded hints and full solutions.

Hardy’s Miscellaneous Examples on Chapter VII are long and varied. They follow the chapter’s two halves: Taylor expansions and functions of several variables (Examples 1–30), then definite integrals (Examples 31–48). Eleven are collected here, in three groups, roughly in order of difficulty within each group. Try each one before opening the first hint; the hints go from a nudge to an outline, and every solution is complete. As in Hardy, assume that all the derivatives that occur are continuous.

Taylor expansions

Exercise The series for tan⁡x\tan x and sec⁡x\sec x ●●○ Hardy: Misc. Examples VII, 1 (1), (2)

Verify the first terms tan⁡x=x+13x3+215x5+…\tan x = x + \tfrac13x^3 + \tfrac{2}{15}x^5 + \dots and sec⁡x=1+12x2+524x4+…\sec x = 1 + \tfrac12x^2 + \tfrac{5}{24}x^4 + \dots.

Hint
Computing five derivatives of tan⁡x\tan x is tedious. Instead use tan⁡xcos⁡x=sin⁡x\tan x\cos x = \sin x and sec⁡xcos⁡x=1\sec x\cos x = 1, with the known series for sin⁡\sin and cos⁡\cos.
Hint
tan⁡\tan is odd and sec⁡\sec is even, so write tan⁡x=x+ax3+bx5+o(x5)\tan x = x + ax^3 + bx^5 + o(x^5) and sec⁡x=1+cx2+dx4+o(x4)\sec x = 1 + cx^2 + dx^4 + o(x^4) (Taylor’s theorem, §151), multiply by 1−12x2+124x4+o(x5)1 - \tfrac12x^2 + \tfrac1{24}x^4 + o(x^5), and compare coefficients.

Solution

By Taylor’s theorem in the form of §151, tan⁡x=x+ax3+bx5+o(x5)\tan x = x + ax^3 + bx^5 + o(x^5) for some numbers a,ba, b: the even coefficients vanish because tan⁡\tan is odd, and the coefficient of xx is tan⁡′(0)=1\tan'(0) = 1. Also cos⁡x=1−12x2+124x4+o(x5)\cos x = 1 - \tfrac12x^2 + \tfrac1{24}x^4 + o(x^5) and sin⁡x=x−16x3+1120x5+o(x5)\sin x = x - \tfrac16x^3 + \tfrac1{120}x^5 + o(x^5). Multiplying,

tan⁡xcos⁡x=x+(a−12)x3+(b−12a+124)x5+o(x5).\tan x\cos x = x + \left(a - \tfrac12\right)x^3 + \left(b - \tfrac12a + \tfrac1{24}\right)x^5 + o(x^5).

This equals sin⁡x\sin x, and two expansions of the same function in powers of xx up to o(x5)o(x^5) have the same coefficients (subtract and divide by successive powers of xx). So a−12=−16a - \tfrac12 = -\tfrac16, giving a=13a = \tfrac13; and b−16+124=1120b - \tfrac16 + \tfrac1{24} = \tfrac1{120}, giving b=1−5+20120=215b = \tfrac{1 - 5 + 20}{120} = \tfrac{2}{15}.

In the same way sec⁡xcos⁡x=1+(c−12)x2+(d−12c+124)x4+o(x4)=1\sec x\cos x = 1 + (c - \tfrac12)x^2 + (d - \tfrac12c + \tfrac1{24})x^4 + o(x^4) = 1 gives c=12c = \tfrac12 and d=14−124=524d = \tfrac14 - \tfrac1{24} = \tfrac5{24}.

Exercise A second difference at infinity ●●○ Hardy: Misc. Examples VII, 6

Prove that xx{x−2x+1+x+2}→−14x\sqrt x\left\{\sqrt x - 2\sqrt{x+1} + \sqrt{x+2}\right\} \to -\tfrac14 as x→∞x \to \infty.

Hint
The bracket is a second difference ϕ(x)−2ϕ(x+1)+ϕ(x+2)\phi(x) - 2\phi(x+1) + \phi(x+2) with ϕ(t)=t\phi(t) = \sqrt t. Examples LXVII, 1 (in Approximate integration: Simpson's rule) turns a second difference into a second derivative.
Hint
Apply that result at the centre x+1x + 1 with h=1h = 1: the bracket is ϕ′′(ξ)\phi''(\xi) for some ξ\xi between xx and x+2x + 2.

Solution

By Examples LXVII, 1 with centre x+1x + 1 and h=1h = 1, ϕ(x+2)−2ϕ(x+1)+ϕ(x)=ϕ′′(ξ)\phi(x+2) - 2\phi(x+1) + \phi(x) = \phi''(\xi) for some ξ\xi in (x,x+2)(x, x+2). With ϕ(t)=t1/2\phi(t) = t^{1/2}, ϕ′′(t)=−14t−3/2\phi''(t) = -\tfrac14t^{-3/2}. So the expression equals

x3/2⋅(−14ξ−3/2)=−14(xξ)3/2.x^{3/2}\cdot\left(-\tfrac14\xi^{-3/2}\right) = -\tfrac14\left(\frac{x}{\xi}\right)^{3/2}.

Since x<ξ<x+2x < \xi < x + 2, we have 1>x/ξ>x/(x+2)→11 > x/\xi > x/(x+2) \to 1, so the expression tends to −14-\tfrac14.

(Alternatively, expand: x+k=x(1+k2x−k28x2+O(x−3))\sqrt{x + k} = \sqrt x\left(1 + \frac{k}{2x} - \frac{k^2}{8x^2} + O(x^{-3})\right), and in the bracket the terms in 11 and 1/x1/x cancel, leaving x⋅2−48x2=−14x3/2\sqrt x\cdot\frac{2 - 4}{8x^2} = -\frac{1}{4x^{3/2}} plus smaller terms.)

Functions of several variables

Exercise Euler’s theorem on homogeneous functions ●●○ Hardy: Misc. Examples VII, 12

A function u(x,y,z,… )u(x, y, z, \dots) is homogeneous of degree nn if u(λx,λy,λz,… )=λnu(x,y,z,… )u(\lambda x, \lambda y, \lambda z, \dots) = \lambda^n u(x, y, z, \dots) for all λ>0\lambda > 0. Prove that then

x∂u∂x+y∂u∂y+z∂u∂z+⋯=nu.x\frac{\partial u}{\partial x} + y\frac{\partial u}{\partial y} + z\frac{\partial u}{\partial z} + \dots = nu .
Hint
Fix the point (x,y,z,… )(x, y, z, \dots) and regard both sides of the defining identity as functions of λ\lambda alone.
Hint
Differentiate with respect to λ\lambda by the chain rule of §157, then put λ=1\lambda = 1.
Solution

Fix (x,y,z,… )(x, y, z, \dots) and let g(λ)=u(λx,λy,λz,… )g(\lambda) = u(\lambda x, \lambda y, \lambda z, \dots). The arguments are functions of λ\lambda with derivatives x,y,z,…x, y, z, \dots, so by the chain rule (valid since the partial derivatives are continuous)

g′(λ)=x ux(λx,λy,… )+y uy(λx,λy,… )+z uz(λx,λy,… )+… .g'(\lambda) = x\,u_x(\lambda x, \lambda y, \dots) + y\,u_y(\lambda x, \lambda y, \dots) + z\,u_z(\lambda x, \lambda y, \dots) + \dots .

On the other hand g(λ)=λnu(x,y,z,… )g(\lambda) = \lambda^n u(x, y, z, \dots), so g′(λ)=nλn−1u(x,y,z,… )g'(\lambda) = n\lambda^{n-1}u(x, y, z, \dots). Putting λ=1\lambda = 1 gives xux+yuy+zuz+⋯=nuxu_x + yu_y + zu_z + \dots = nu. Check: u=x2y+y3u = x^2y + y^3 is homogeneous of degree 33, and x⋅2xy+y(x2+3y2)=3x2y+3y3=3ux\cdot 2xy + y(x^2 + 3y^2) = 3x^2y + 3y^3 = 3u.

Exercise The functional equation of the logarithm ●●○ Hardy: Misc. Examples VII, 18

Let f′(x)=1/xf'(x) = 1/x for x>0x > 0 and f(1)=0f(1) = 0. With u=f(x)+f(y)u = f(x) + f(y) and v=xyv = xy, show that uxvy−uyvx=0u_xv_y - u_yv_x = 0. Hardy concludes that uu is a function of vv; show by putting y=1y = 1 that f(x)+f(y)=f(xy)f(x) + f(y) = f(xy). Then give a direct proof that does not rely on the (unproved) sufficiency of the Jacobian condition.

Hint
ux=1/xu_x = 1/x, uy=1/yu_y = 1/y, vx=yv_x = y, vy=xv_y = x.
Hint
Direct proof: fix yy and differentiate g(x)=f(xy)−f(x)−f(y)g(x) = f(xy) - f(x) - f(y) with respect to xx.

Solution

The Jacobian. uxvy−uyvx=1x⋅x−1y⋅y=0u_xv_y - u_yv_x = \frac1x\cdot x - \frac1y\cdot y = 0. If, as Hardy argues (Misc. Ex. 15), this forces a relation u=F(v)u = F(v), then putting y=1y = 1 gives u=f(x)+f(1)=f(x)u = f(x) + f(1) = f(x) and v=xv = x, so F(x)=f(x)F(x) = f(x) for every x>0x > 0, and f(x)+f(y)=u=F(v)=f(xy)f(x) + f(y) = u = F(v) = f(xy).

Directly. Fix y>0y > 0 and let g(x)=f(xy)−f(x)−f(y)g(x) = f(xy) - f(x) - f(y). By the chain rule g′(x)=y⋅1xy−1x=0g'(x) = y\cdot\frac{1}{xy} - \frac1x = 0, so gg is constant (§126), and g(1)=f(y)−f(1)−f(y)=0g(1) = f(y) - f(1) - f(y) = 0. Hence f(xy)=f(x)+f(y)f(xy) = f(x) + f(y). Since f(x)=∫1xdt/tf(x) = \int_1^x dt/t is such a function (§162), this is the functional equation of log⁡\log. (The same direct argument, with g(x)=f ⁣(x+y1−xy)−f(x)−f(y)g(x) = f\!\left(\frac{x+y}{1-xy}\right) - f(x) - f(y), proves Hardy’s companion result f(x)+f(y)=f ⁣(x+y1−xy)f(x) + f(y) = f\!\left(\frac{x+y}{1-xy}\right) for f′=1/(1+x2)f' = 1/(1+x^2), f(0)=0f(0) = 0, when xy<1xy < 1.)

Exercise Derivatives of implicitly related variables ●●○ Hardy: Misc. Examples VII, 25

Three variables x,y,zx, y, z are connected by a relation which makes zz a function of xx and yy, with derivatives zx,zyz_x, z_y, and also makes xx a function of yy and zz, with derivatives xy,xzx_y, x_z. Prove that xy=−zy/zxx_y = -z_y/z_x and xz=1/zxx_z = 1/z_x.

Hint
Substituting x=x(y,z)x = x(y, z) into z=z(x,y)z = z(x, y) gives an identity z=z{x(y,z),y}z = z\{x(y, z), y\} in the two independent variables yy and zz.
Hint
Differentiate that identity once with respect to zz and once with respect to yy.

Solution

For all (y,z)(y, z) in the relevant region, z{x(y,z),y}=zz\{x(y, z), y\} = z. Differentiate with respect to zz (with yy fixed), using the chain rule: zx xz=1z_x\,x_z = 1. Differentiate with respect to yy (with zz fixed): zx xy+zy=0z_x\,x_y + z_y = 0. The first shows zx≠0z_x \ne 0, so xz=1/zxx_z = 1/z_x and xy=−zy/zxx_y = -z_y/z_x.

This is Hardy’s computation with differentials (dz=zx dx+zy dydz = z_x\,dx + z_y\,dy and dx=xy dy+xz dzdx = x_y\,dy + x_z\,dz, with dydy, dzdz independent) made explicit. Note the minus sign: xyx_y is not 1/yx1/y_x in disguise. With the third relation yz=−xz/xyy_z = -x_z/x_y one obtains the well-known xy yz zx=−1x_y\,y_z\,z_x = -1. Check with x+y+z=0x + y + z = 0: zx=zy=−1z_x = z_y = -1, xy=−1=−zy/zxx_y = -1 = -z_y/z_x, xz=−1=1/zxx_z = -1 = 1/z_x.

Definite integrals

Exercise An integral that jumps ●●○ Hardy: Misc. Examples VII, 32

Evaluate I(α)=∫−11sin⁡α dx1−2xcos⁡α+x2I(\alpha) = \displaystyle\int_{-1}^{1}\frac{\sin\alpha\,dx}{1 - 2x\cos\alpha + x^2}. For what values of α\alpha is II a discontinuous function of α\alpha?

Hint
If sin⁡α≠0\sin\alpha \ne 0, 1−2xcos⁡α+x2=(x−cos⁡α)2+sin⁡2α1 - 2x\cos\alpha + x^2 = (x - \cos\alpha)^2 + \sin^2\alpha, and arctan⁡x−cos⁡αsin⁡α\arctan\dfrac{x - \cos\alpha}{\sin\alpha} is an antiderivative.
Hint
1−cos⁡αsin⁡α=tan⁡12α\dfrac{1 - \cos\alpha}{\sin\alpha} = \tan\tfrac12\alpha and 1+cos⁡αsin⁡α=cot⁡12α\dfrac{1 + \cos\alpha}{\sin\alpha} = \cot\tfrac12\alpha. Work first with 0<α<π0 < \alpha < \pi, then −π<α<0-\pi < \alpha < 0, then use periodicity.

Solution

If sin⁡α=0\sin\alpha = 0 the integrand is 00 (for x≠±1x \ne \pm1), so I=0I = 0. Otherwise the denominator is positive on [−1,1][-1, 1], and differentiating confirms the antiderivative in the first hint. So

I(α)=arctan⁡(tan⁡12α)+arctan⁡(cot⁡12α),I(\alpha) = \arctan\left(\tan\tfrac12\alpha\right) + \arctan\left(\cot\tfrac12\alpha\right),

using the oddness of arctan⁡\arctan for the lower limit. For 0<α<π0 < \alpha < \pi, 12α\tfrac12\alpha is in (0,12π)(0, \tfrac12\pi): the first term is 12α\tfrac12\alpha and the second 12π−12α\tfrac12\pi - \tfrac12\alpha, so I=12πI = \tfrac12\pi. For −π<α<0-\pi < \alpha < 0 the first term is 12α\tfrac12\alpha and the second is −12π−12α-\tfrac12\pi - \tfrac12\alpha (the value of arctan⁡\arctan must lie in (−12π,0)(-\tfrac12\pi, 0)), so I=−12πI = -\tfrac12\pi. The integrand has period 2π2\pi in α\alpha, hence

I(α)=12π  on (2nπ,(2n+1)π),−12π  on ((2n−1)π,2nπ),0  at α=nπ.I(\alpha) = \tfrac12\pi \ \text{ on } (2n\pi, (2n+1)\pi), \qquad -\tfrac12\pi \ \text{ on } ((2n-1)\pi, 2n\pi), \qquad 0 \ \text{ at } \alpha = n\pi .

II is discontinuous exactly at the multiples of π\pi. The integrand is a perfectly good continuous function of xx for each fixed α\alpha, but as α→0\alpha \to 0 it concentrates near x=1x = 1 into a spike of area 12π\tfrac12\pi; this is why “the integral of a function continuous in each variable” need not be continuous in the parameter. (Near α=π\alpha = \pi the spike is at x=−1x = -1.)

Exercise ●●○ Hardy: Misc. Examples VII, 35

If a>1a > 1, prove that ∫−111−x2a−x dx=π{a−a2−1}\displaystyle\int_{-1}^{1}\frac{\sqrt{1-x^2}}{a - x}\,dx = \pi\left\{a - \sqrt{a^2-1}\right\}.

Hint
Put x=cos⁡θx = \cos\theta, to obtain ∫0πsin⁡2θa−cos⁡θ dθ\displaystyle\int_0^\pi\frac{\sin^2\theta}{a - \cos\theta}\,d\theta.
Hint
Write sin⁡2θ=1−cos⁡2θ=(a−cos⁡θ)(a+cos⁡θ)−(a2−1)\sin^2\theta = 1 - \cos^2\theta = (a - \cos\theta)(a + \cos\theta) - (a^2 - 1), and use ∫0πdθa+bcos⁡θ=πa2−b2\displaystyle\int_0^\pi\frac{d\theta}{a + b\cos\theta} = \frac{\pi}{\sqrt{a^2 - b^2}} for a>∣b∣a > |b| (Examples LXIII, 7).

Solution

With x=cos⁡θx = \cos\theta (θ\theta from π\pi to 00, dx=−sin⁡θ dθdx = -\sin\theta\,d\theta, 1−x2=sin⁡θ≥0\sqrt{1-x^2} = \sin\theta \ge 0), the integral becomes ∫0πsin⁡2θa−cos⁡θ dθ\int_0^\pi\frac{\sin^2\theta}{a - \cos\theta}\,d\theta. By the identity in the second hint,

sin⁡2θa−cos⁡θ=a+cos⁡θ−a2−1a−cos⁡θ.\frac{\sin^2\theta}{a - \cos\theta} = a + \cos\theta - \frac{a^2 - 1}{a - \cos\theta}.

Integrating over [0,π][0, \pi]: ∫a=aπ\int a = a\pi, ∫cos⁡θ=0\int\cos\theta = 0, and by Examples LXIII, 7 with b=−1b = -1, ∫0πdθa−cos⁡θ=πa2−1\int_0^\pi\frac{d\theta}{a - \cos\theta} = \frac{\pi}{\sqrt{a^2-1}}. So the integral is aπ−(a2−1)πa2−1=π{a−a2−1}a\pi - (a^2 - 1)\frac{\pi}{\sqrt{a^2-1}} = \pi\{a - \sqrt{a^2-1}\}. All integrands are continuous, since a>1≥cos⁡θa > 1 \ge \cos\theta. (Check: a=2a = 2 gives π(2−3)=0.8418\pi(2 - \sqrt3) = 0.8418.)

Exercise A substitution that does all the work ●●● Hardy: Misc. Examples VII, 42

Prove, by the substitution 1+x4=(1+x2)cos⁡ϕ\sqrt{1+x^4} = (1+x^2)\cos\phi, that ∫011−x21+x2 dx1+x4=π42\displaystyle\int_0^1\frac{1-x^2}{1+x^2}\,\frac{dx}{\sqrt{1+x^4}} = \frac{\pi}{4\sqrt2}.

Hint
From cos⁡2ϕ=1+x4(1+x2)2\cos^2\phi = \dfrac{1+x^4}{(1+x^2)^2} find sin⁡ϕ\sin\phi as a function of xx. As xx goes from 00 to 11, where does ϕ\phi go?
Hint
Differentiate sin⁡ϕ=2 x1+x2\sin\phi = \dfrac{\sqrt2\,x}{1+x^2} with respect to xx and solve for dϕ/dxd\phi/dx.

Solution

sin⁡2ϕ=1−1+x4(1+x2)2=2x2(1+x2)2\sin^2\phi = 1 - \dfrac{1+x^4}{(1+x^2)^2} = \dfrac{2x^2}{(1+x^2)^2}, so with ϕ\phi in [0,12π)[0, \tfrac12\pi), sin⁡ϕ=2 x1+x2\sin\phi = \dfrac{\sqrt2\,x}{1+x^2}. On [0,1][0, 1] this increases from 00 to 12\tfrac{1}{\sqrt2}, so ϕ=arcsin⁡2 x1+x2\phi = \arcsin\dfrac{\sqrt2\,x}{1+x^2} increases from 00 to 14π\tfrac14\pi, with a continuous derivative. Differentiating sin⁡ϕ\sin\phi,

cos⁡ϕ dϕdx=2 (1+x2)−2x2(1+x2)2=2 (1−x2)(1+x2)2,\cos\phi\,\frac{d\phi}{dx} = \sqrt2\,\frac{(1+x^2) - 2x^2}{(1+x^2)^2} = \frac{\sqrt2\,(1-x^2)}{(1+x^2)^2},

and since cos⁡ϕ=1+x4/(1+x2)\cos\phi = \sqrt{1+x^4}/(1+x^2),

dϕdx=2 1−x2(1+x2)1+x4.\frac{d\phi}{dx} = \sqrt2\,\frac{1-x^2}{(1+x^2)\sqrt{1+x^4}} .

So the integrand is exactly 12dϕdx\frac{1}{\sqrt2}\frac{d\phi}{dx}, and by the substitution rule (§166) the integral is 12{ϕ(1)−ϕ(0)}=12⋅π4\frac{1}{\sqrt2}\{\phi(1) - \phi(0)\} = \frac{1}{\sqrt2}\cdot\frac{\pi}{4}. (Numerically 0.55540.5554.)

Exercise Schwarz’s inequality ●●○ Hardy: Misc. Examples VII, 44

For continuous ϕ\phi, ψ\psi on [a,b][a, b], prove that (∫abϕψ dx)2≤∫abϕ2 dx∫abψ2 dx\left(\displaystyle\int_a^b\phi\psi\,dx\right)^2 \le \displaystyle\int_a^b\phi^2\,dx\int_a^b\psi^2\,dx.

Hint
For all real λ\lambda, μ\mu: ∫ab(λϕ+μψ)2 dx≥0\int_a^b(\lambda\phi + \mu\psi)^2\,dx \ge 0.
Hint
A quadratic form Aλ2+2Bλμ+Cμ2A\lambda^2 + 2B\lambda\mu + C\mu^2 that is never negative has B2≤ACB^2 \le AC.

Solution

Let A=∫ϕ2A = \int\phi^2, B=∫ϕψB = \int\phi\psi, C=∫ψ2C = \int\psi^2. Expanding by linearity, 0≤∫(λϕ+μψ)2=Aλ2+2Bλμ+Cμ20 \le \int(\lambda\phi + \mu\psi)^2 = A\lambda^2 + 2B\lambda\mu + C\mu^2 for all real λ,μ\lambda, \mu. If A=0A = 0 then ϕ≡0\phi \equiv 0 (a continuous non-negative ϕ2\phi^2 with zero integral, Misc. Ex. 43), so B=0B = 0 and the inequality holds. If A>0A > 0, take μ=1\mu = 1 and λ=−B/A\lambda = -B/A: then 0≤−B2/A+C0 \le -B^2/A + C, that is, B2≤ACB^2 \le AC. It is the continuous version of Cauchy’s inequality (∑arbr)2≤∑ar2∑br2(\sum a_rb_r)^2 \le \sum a_r^2\sum b_r^2 (Misc. Examples I, 10), to which Riemann sums reduce it. Equality holds only if λϕ+μψ≡0\lambda\phi + \mu\psi \equiv 0 for some (λ,μ)≠(0,0)(\lambda, \mu) \ne (0, 0), that is, if ϕ\phi and ψ\psi are proportional.

Exercise Legendre’s polynomials ●●● Hardy: Misc. Examples VII, 45–46

Let Pn(x)=1(β−α)nn!dndxn{(x−α)(β−x)}nP_n(x) = \dfrac{1}{(\beta-\alpha)^n n!}\dfrac{d^n}{dx^n}\{(x-\alpha)(\beta-x)\}^n. Show that PnP_n is a polynomial of degree nn and that ∫αβPn(x)θ(x) dx=0\int_\alpha^\beta P_n(x)\theta(x)\,dx = 0 for every polynomial θ\theta of degree less than nn. Deduce that ∫αβPmPn dx=0\int_\alpha^\beta P_mP_n\,dx = 0 for m≠nm \ne n.

Hint
Let w(x)={(x−α)(β−x)}nw(x) = \{(x-\alpha)(\beta-x)\}^n, of degree 2n2n with zeros of order nn at α\alpha and β\beta. Then w,w′,…,w(n−1)w, w', \dots, w^{(n-1)} all vanish at α\alpha and β\beta.
Hint
Integrate ∫w(n)θ\int w^{(n)}\theta by parts repeatedly, moving derivatives from ww onto θ\theta. If θ\theta has degree m<nm < n, stop after m+1m + 1 steps.

Solution

ww has degree 2n2n, so w(n)w^{(n)} has degree nn (its leading coefficient is (−1)n(2n)!/n!(-1)^n(2n)!/n!, not 00), and PnP_n is a polynomial of degree nn. Near α\alpha, w(x)=(x−α)nq(x)w(x) = (x - \alpha)^n q(x) with qq a polynomial; by Leibniz’s rule each of w′,…,w(n−1)w', \dots, w^{(n-1)} still contains a factor (x−α)(x - \alpha), so they all vanish at α\alpha, and likewise at β\beta.

Let θ\theta have degree m<nm < n. Integrating by parts once,

∫αβw(n)θ=[w(n−1)θ]αβ−∫αβw(n−1)θ′=−∫αβw(n−1)θ′,\int_\alpha^\beta w^{(n)}\theta = \Big[w^{(n-1)}\theta\Big]_\alpha^\beta - \int_\alpha^\beta w^{(n-1)}\theta' = -\int_\alpha^\beta w^{(n-1)}\theta' ,

since w(n−1)(α)=w(n−1)(β)=0w^{(n-1)}(\alpha) = w^{(n-1)}(\beta) = 0. Repeating m+1m + 1 times (possible since m+1≤nm + 1 \le n, and every bracket vanishes because it contains w(n−k)w^{(n-k)} with 1≤k≤n1 \le k \le n) gives ±∫αβw(n−m−1)θ(m+1)=0\pm\int_\alpha^\beta w^{(n-m-1)}\theta^{(m+1)} = 0, as θ(m+1)=0\theta^{(m+1)} = 0. Dividing by the constant (β−α)nn!(\beta - \alpha)^n n!, ∫αβPnθ=0\int_\alpha^\beta P_n\theta = 0.

If m<nm < n, PmP_m is a polynomial of degree less than nn, so ∫PmPn=0\int P_mP_n = 0; by symmetry the same holds if m>nm > n. (For [α,β]=[0,1][\alpha, \beta] = [0, 1]: P1(x)=1−2xP_1(x) = 1 - 2x, and indeed ∫01(1−2x) dx=0\int_0^1(1-2x)\,dx = 0.)

Exercise A three-point rule exact for quintics ●●○ Hardy: Misc. Examples VII, 48

Let α\alpha, β\beta be the roots of x2−x+110=0x^2 - x + \tfrac1{10} = 0. Prove that for every polynomial ϕ\phi of degree at most 55,

∫01ϕ(x) dx=118{5ϕ(α)+8ϕ(12)+5ϕ(β)}.\int_0^1\phi(x)\,dx = \tfrac1{18}\left\{5\phi(\alpha) + 8\phi(\tfrac12) + 5\phi(\beta)\right\}.
Hint
Both sides are linear in ϕ\phi, so it is enough to check the six polynomials 1,t,t2,…,t51, t, t^2, \dots, t^5, where t=x−12t = x - \tfrac12 (every quintic in xx is a quintic in tt).
Hint
α,β=12∓c\alpha, \beta = \tfrac12 \mp c with c2=14−110=320c^2 = \tfrac14 - \tfrac1{10} = \tfrac3{20}. Odd powers of tt take care of themselves by symmetry.
Solution

In terms of t=x−12t = x - \tfrac12 the interval is [−12,12][-\tfrac12, \tfrac12], and the three points are t=−c,0,ct = -c, 0, c with c2=320c^2 = \tfrac3{20}. Both sides are linear in ϕ\phi, so check ϕ=tk\phi = t^k, k=0,…,5k = 0, \dots, 5.

Odd kk. ∫−1/21/2tk dt=0\int_{-1/2}^{1/2}t^k\,dt = 0, and the rule gives 118{5(−c)k+0+5ck}=0\tfrac1{18}\{5(-c)^k + 0 + 5c^k\} = 0.

k=0k = 0. ∫1=1\int 1 = 1 and 118(5+8+5)=1\tfrac1{18}(5 + 8 + 5) = 1.

k=2k = 2. ∫t2=2⋅(1/2)33=112\int t^2 = 2\cdot\frac{(1/2)^3}{3} = \frac1{12}, and 10c218=10⋅3/2018=112\frac{10c^2}{18} = \frac{10\cdot 3/20}{18} = \frac{1}{12}.

k=4k = 4. ∫t4=2⋅(1/2)55=180\int t^4 = 2\cdot\frac{(1/2)^5}{5} = \frac1{80}, and 10c418=10⋅9/40018=180\frac{10c^4}{18} = \frac{10\cdot 9/400}{18} = \frac{1}{80}.

So the rule is exact for every quintic. It uses only three ordinates, like Simpson’s rule (§169), but is exact for polynomials of two degrees higher, because the outer points are placed cleverly rather than at the ends. It is the three-point Gauss–Legendre rule, and α\alpha, β\beta, 12\tfrac12 are the zeros of the Legendre polynomial P3P_3 of the previous problem on [0,1][0, 1]. It fails for t6t^6: ∫t6=1448\int t^6 = \frac1{448}, while the rule gives 10c618=31600\frac{10c^6}{18} = \frac{3}{1600}.