Changing the variable of integration, seen as a picture and as the chain rule; and the strategy of choosing a new variable that turns an algebraical integrand into a rational one.
Rational functions can always be integrated (§133). The next class, algebraical functions such as ax2+2bx+c, cannot in general, and here Hardy introduces the main tool of practical integration: substitution. Put x=f(t) for a cleverly chosen f, and a hard integral in x may become an easy one in t. The most important special case is rationalisation: choosing t so that the new integrand is a rational function, which §133 then finishes off. The next two lessons apply this idea to square roots of quadratics.
The idea
Take ∫1+xdx. The square root is the obstacle. If we write x=t2 (with t≥0), then x=t, and the integrand becomes 1/(1+t), which is rational. But the integral is not simply ∫dt/(1+t): changing the variable changes the scale along the axis. A small step dt in t corresponds to a step dx=2tdt in x, so the integrand must be multiplied by 2t:
∫1+xdx=∫1+t2tdt=2t−2log(1+t)=2x−2log(1+x).
Why the factor f′(t)? Think of the integral as an area, as Hardy will justify in §148. Cut the t-axis into equal strips. The map x=f(t) sends them to strips of the x-axis that are not equal: each is stretched by roughly f′(t). For the areas to match, the height in the t-picture must be the height in the x-picture times that stretch.
ExploreSubstitution as matching strips
Try this.The lower picture is ∫dx/(1+x) from x=0 to 4; the upper is the same integral after x=t2. Move the highlighted strip k from left to right and read the stretch factor. Increase n. Then try the example 1−x2 with x=sint.
∫1+x1dx with x=t2 becomes ∫1+t2tdt
tψ(f(t)) f′(t)
xψ(x)
strips (alternate colours)
…and their partners
strip k in both pictures
Strip 1: t from 0 to 0.333 (width 0.333) becomes x from 0 to 0.111 (width 0.111, stretched by 0.333). Its area is 0.0913 in both pictures (computed separately: 0.0913). Totals: ∫1+t2tdt=1.8028 and ∫1+x1dx=1.8028.
What to notice
Equal strips in t become wider and wider strips in x (stretch ≈2t), and the upper curve 2t/(1+t) is correspondingly taller than 1/(1+x) at the matching place. Each pair of matching strips has exactly the same area, so the totals agree. With x=sint the stretch is cost, which shrinks towards t=21π, where the strips in x crowd together.
Precise statements
TheoremIntegration by substitutionHardy: §136 (1)
Let ϕ be an integral of ψ, and let f be differentiable, with values where ψ is defined. Then
∫ψ{f(t)}f′(t)dt=ϕ{f(t)}.
If moreover f is strictly monotonic on an interval, with inverse t=g(x), then ϕ(x) is recovered as the integral in t evaluated at t=g(x).
Hardy states the practical rule like this: put x=f(t), multiply by f′(t), integrate the resulting function of t if you can, and express the result in terms of x. The theorem is the same rule as the chain rule backwards of §133, read with the roles of the variables exchanged.
Now the strategy that gives the method its power. Hardy considers integrals ∫R(x,y)dx, where R is a rational function of two variables and y is an algebraical function of x (§135). For example
px+q−ax2+2bx+cpx+q+ax2+2bx+c
is a rational function of x and y=ax2+2bx+c. This form is no more general than “an algebraical function of x” (a rational function of x and y is itself algebraical), but it is far more convenient.
TheoremRationalisationHardy: §136
If there is a variable t such that both x and y are rational functions of t, say x=R1(t) and y=R2(t), then
∫R(x,y)dx=∫R{R1(t),R2(t)}R1′(t)dt,
and the right-hand side is the integral of a rational function of t, which can be found by partial fractions.
For example, ∫R(x,x)dx is rationalised by x=t2, x=t, and becomes ∫2tR(t2,t)dt. When can such a t be found? In general that is a deep question (it leads to the genus of an algebraic curve) and Hardy does not pursue it. He treats the most useful case: y2 a quadratic in x, where the graph of y is a conic. That is the next lesson.
Proofs
ProofIntegration by substitution
Strategy
Differentiate ϕ{f(t)} with respect to t. For the second part, substitute t=g(x) back and use the chain rule once more.
Chain rule
dtdϕ{f(t)}=ϕ′{f(t)}f′(t)=ψ{f(t)}f′(t), so ϕ{f(t)} is an integral of ψ{f(t)}f′(t).
Any integral in t has this form
If Θ(t) is any integral of ψ{f(t)}f′(t) on the interval, then Θ(t)=ϕ{f(t)}+C, by the uniqueness theorem of §130.
Return to x
If f is strictly monotonic with inverse g, then f{g(x)}=x, so Θ{g(x)}=ϕ(x)+C, an integral of ψ. This is why the substitution must be invertible on the interval used: otherwise we cannot express the result in terms of x.
∎
The proof of the rationalisation theorem is the same with ψ(x)=R(x,y(x)) and f=R1, since then ψ{R1(t)}=R{R1(t),R2(t)}.
Worked examples and exercises
ExerciseA first rationalisation●○○Hardy: §136
Find ∫xx−1dx for x>1 by putting x=1+t2 with t>0. (This is the third example in the widget.)
Hint
x−1=t and dx=2tdt.
Solution
The integral becomes ∫(1+t2)t2tdt=∫1+t22dt=2arctant. Since t=x−1,
∫xx−1dx=2arctanx−1.
Check: dxd2arctanx−1=1+(x−1)2⋅2x−11=xx−11.
ExerciseAn nth root of a linear function●○○Hardy: Examples XLIX, 9
Show that ∫R{x,nax+b}dx, with R rational and a=0, is reduced to the integral of a rational function by the substitution ax+b=yn.
Hint
Express x and dx/dy in terms of y.
Solution
With ax+b=yn we have nax+b=y, x=(yn−b)/a and dx/dy=nyn−1/a. All three are rational in y, so
∫R{x,nax+b}dx=∫R{ayn−b,y}anyn−1dy,
the integral of a rational function of y. (For even n we take y≥0 and ax+b≥0, so that the substitution is invertible.)
ExerciseTwo different roots●●○Hardy: §136
Find ∫x+3xdx for x>0.
Hint
Choose x=tk so that both x and 3x become powers of t: k must be divisible by 2 and by 3.
Hint
With x=t6 the integrand becomes t3+t26t5=t+16t3. Divide.
With t=x1/6: the answer is 2x−33x+66x−6log(1+6x).
ExerciseThe binomial integral ∫(1+x)pxqdx●●●Hardy: Examples XLIX, 11
Let p,q be rational. Show that ∫(1+x)pxqdx can be rationalised if q is an integer (put 1+x=ts, where s is the denominator of p), and if p+q is an integer (put 1+x=xts).
Hint
In the first case (1+x)p=tsp and sp is an integer.
Hint
In the second case solve for x: x=1/(ts−1). Then 1+x=ts/(ts−1).
Solution
q an integer. With 1+x=ts: (1+x)p=tsp with sp an integer, xq=(ts−1)q is rational in t because q is an integer, and dx=sts−1dt. The integrand becomes stsp+s−1(ts−1)q, a rational function.
p+q an integer. With 1+x=xts we get x=1/(ts−1) and 1+x=ts/(ts−1). Then
(1+x)pxq=(ts−1)ptsp⋅(ts−1)q1=tsp(ts−1)−(p+q),
rational because sp and p+q are integers; and dx/dt=−sts−1/(ts−1)2 is rational. (The third case, p an integer, is like the first with x=us, s the denominator of q.) A theorem of Chebyshev says these three cases are the only ones in which the integral is elementary.
ExerciseA curve rationalised by y=tx●●○Hardy: Examples XLIX, 14
Let y be defined by y2(x−y)=x2. Show that ∫R(x,y)dx can be reduced to the integral of a rational function.
Hint
Put y=tx in the equation and solve for x.
Solution
With y=tx: t2x2(x−tx)=x2, so (for x=0) t2(1−t)x=1, that is
x=t2(1−t)1,y=tx=t(1−t)1.
Both are rational in t, and so is dx/dt. By the rationalisation theorem, ∫R(x,y)dx=∫R{x(t),y(t)}x′(t)dt is the integral of a rational function. The trick works because every line y=tx through the origin meets the curve in only one other point: the origin counts as a double point of this cubic.
Check yourself
Check yourselfChoose one
After the substitution x=t2 (t>0), the integral ∫sinxdx becomes
Why
Replace x by t2anddx by f′(t)dt=2tdt. Forgetting the factor f′(t) (first option) is the most common error; dividing by it (third) is the second most common. The fourth substitutes in the wrong direction.
Check yourselfEnter a number
Using ∫dx/(1+x)=2x−2log(1+x), compute the increase of this integral from x=0 to x=4 (to three decimals). Compare with the totals in the widget.
Why
At x=4: 4−2log3; at x=0: 0. The difference is 4−2log3≈1.8028, the value the widget shows for both pictures with t1=2.
Check yourselfChoose one
Which substitution rationalises ∫1+4xxdx?
Why
With x=t4 the root becomes t, dx=4t3dt, and the integrand 4t7/(1+t) is rational. With x=t2 a square root t remains. The others leave a fourth root of something.
Check yourselfTrue or false?
Any differentiable f may be used in the substitution x=f(t), and the answer can always be converted back to x.
Why
False. The first half of the theorem holds for any differentiable f, but to express the result in x we need t as a function of x, which requires f to be invertible (for example strictly monotonic) on the interval used. With x=sint we restrict t to [−21π,21π] for exactly this reason.
Hardy’s language
Summary
Summary
Substitution: if x=f(t), then ∫ψ(x)dx=∫ψ{f(t)}f′(t)dt. The factor f′(t) is the stretch of the axis; never forget it.
To return to x, the substitution must be invertible on the interval in question.
Rationalisation: find t with x and y both rational in t; then ∫R(x,y)dx is the integral of a rational function.
Roots of a linear function, nax+b, are rationalised by ax+b=tn; several roots of x by a common power x=tk.
Whether a rationalising t exists depends on the curve y(x); for y2 quadratic it always does (next lesson).