Many limits come in a form where substituting the limiting value gives the meaningless "0/0": (sinx)/x, (1−cosx)/x2, (tanx−x)/(x−sinx) as x→0. Every derivative is such a limit, and so are many questions about how fast one quantity vanishes compared with another. §154 gives two systematic methods. The first compares Taylor expansions; the second, which students know as L’Hôpital’s rule, replaces the ratio of the functions by the ratio of their derivatives. Knowing why each works tells you when it fails.
The idea
If f(0)=ϕ(0)=0, the ratio f(x)/ϕ(x) near 0 depends on how fast each function vanishes. Taylor’s theorem answers that: near 0, each function behaves like the first non-zero term of its expansion. For example
tanx−x=31x3+152x5+⋯,x−sinx=61x3−1201x5+⋯,
so both vanish like x3, and their ratio tends to 31/61=2. Everything else in the expansions is small compared with the x3 terms and disappears in the limit.
ExploreLeading terms decide
Try this.For each example, read the two expansions, predict the limit from their leading terms, and check it against the graph and the table as the probe x shrinks. Then choose the last example, x2sin(1/x)/x, and switch on f′/ϕ′.
x−sinxtanx−x as x→0 (Examples LVIII, 3)
x
f(x)/φ(x)
the limit
x
f/φ
0.3
2.08408
0.03
2.00081
0.003
2.000008
Numerator: tanx−x=31x3+152x5+⋯; denominator: x−sinx=61x3−1201x5+⋯. The ratio of the leading terms gives the limit 2.
What to notice
Away from 0 the ratio can be far from its limit; only the leading terms matter as x→0. In the last example the ratio itself squeezes to 0, but the ratio of derivatives, 2xsin(1/x)−cos(1/x), swings between about −1 and 1 for ever. The derivative method can fail when the limit exists.
Precise statements
Hardy’s two methods differ in what they assume. The first needs derivatives at0; the second needs derivatives near0, and a limit for their ratio.
TheoremMethod (a): the first non-vanishing derivativesHardy: §154 (a)
Suppose that f and ϕ have n derivatives at x=0, that f(k)(0)=ϕ(k)(0)=0 for 0≤k≤n−1, and that ϕ(n)(0)=0. Then
x→0limϕ(x)f(x)=ϕ(n)(0)f(n)(0).
TheoremMethod (b): the ratio of derivatives (L'Hôpital's rule)Hardy: §154 (b), (3)
Suppose that f and ϕ are continuous on [0,h] and differentiable on (0,h], that f(0)=ϕ(0)=0, and that f′(x)/ϕ′(x)→l as x→+0. Then f(x)/ϕ(x)→l as x→+0. The same holds for x→−0 and x→0, with l=±∞, and, by repeating the argument, with f(n)/ϕ(n) in place of f′/ϕ′ when the earlier derivatives vanish at 0.
The condition in (b) is sufficient, not necessary: f/ϕ may tend to a limit when f′/ϕ′ does not (the last example above). Other variants (limits as x→a or x→∞, or the form ∞/∞) are usually reduced to these by a substitution such as x=a+t or x=1/t.
In practice method (a) is usually applied through known expansions: if f(x)=anxn+o(xn) and ϕ(x)=bnxn+o(xn) with bn=0, then f/ϕ→an/bn. The expansions of sin, cos, tan and so on can be multiplied and divided like polynomials, keeping track of the order of the error. Because an expansion with remainder o(xn) is unique (exercise below), any correct way of finding it gives the coefficients f(k)(0)/k!.
Proofs
ProofMethod (a)
Strategy
Expand both functions by the theorem of §151. Only the terms in xn survive, and the common factor xn/n! cancels.
Expand
Since the first n−1 derivatives (and the values) vanish at 0,
f(x)=n!xn{f(n)(0)+η},ϕ(x)=n!xn{ϕ(n)(0)+η′},
where η,η′→0 as x→0.
Divide
For small x=0 we have ∣η′∣<∣ϕ(n)(0)∣, so ϕ(x)=0 and
ϕ(x)f(x)=ϕ(n)(0)+η′f(n)(0)+η→ϕ(n)(0)f(n)(0).
∎
Method (a) applied literally to (tanx−x)/(x−sinx) needs three derivatives of each function: f′′′(0)=2 and ϕ′′′(0)=1. Method (b) needs only one step, because f′/ϕ′ simplifies at once: 1−cosxsec2x−1=cos2x(1−cosx)sin2x=cos2x1+cosx→2. That is Hardy’s reason for preferring (b) when the derivatives are simpler than the functions.
ProofMethod (b)
Strategy
Cauchy’s mean value theorem (§128) turns the ratio f(x)/ϕ(x)=(f(x)−f(0))/(ϕ(x)−ϕ(0)) into a ratio of derivatives at some point ξ between 0 and x. As x→0, ξ is squeezed to 0 too. The only work is to make sure ϕ does not vanish, so that the ratios make sense.
ϕ′ has no zeros in some interval (0,k)
The hypothesis that f′(x)/ϕ′(x) tends to a limit presupposes that this ratio is defined for all small positive x; so there is a k with ϕ′(x)=0 for 0<x<k.
Hence ϕ(x)=0 for 0<x<k
A derivative that never vanishes on an interval keeps one sign there (Darboux’s theorem, §129). So ϕ is strictly monotone on (0,k) and, since ϕ(0)=0 and ϕ is continuous, ϕ(x)=0 for 0<x<k.
Apply Cauchy's mean value theorem on [0,x]
For 0<x<k there is a ξ with 0<ξ<x and
ϕ(x)f(x)=ϕ(x)−ϕ(0)f(x)−f(0)=ϕ′(ξ)f′(ξ).
Let x→+0
Given ε>0, choose δ≤k so that ∣f′(t)/ϕ′(t)−l∣<ε for 0<t<δ. If 0<x<δ then 0<ξ<δ, so ∣f(x)/ϕ(x)−l∣<ε. Hence f(x)/ϕ(x)→l.
∎
Notice where the converse breaks: the point ξ depends on x in an uncontrolled way, so knowing f′(ξ)/ϕ′(ξ) only at these special points says nothing about f′/ϕ′ at all points near 0.
Worked examples and exercises
ExerciseExpansions are unique●○○Hardy: Examples LVI, 1
Show that if a0+a1x+⋯+anxn+o(xn)=b0+b1x+⋯+bnxn+o(xn) as x→0, then ak=bk for every k. Deduce that if f has n derivatives at 0 and f(x)=c0+c1x+⋯+cnxn+o(xn), then ck=f(k)(0)/k!.
Hint
Let x→0. Then divide by x and let x→0 again.
Solution
Subtract: (a0−b0)+(a1−b1)x+⋯+(an−bn)xn=o(xn). Letting x→0 gives a0=b0. Cancel those terms, divide by x, and let x→0: a1=b1. Continue n times (each division by x keeps the right side o(xn−k)). For the second part, §151 gives one expansion with ck=f(k)(0)/k!; by uniqueness every other expansion has the same coefficients.
Exercise●●○Hardy: Examples LVIII, 2
Find the limit of (1−x)2x−(n+1)xn+1+nxn+2 as x→1.
Hint
Put x=1+t, or use method (a) at x=1: check the values and first derivatives of numerator and denominator there.
Hint
Alternatively: the expression equals x+2x2+⋯+nxn for x=1.
Solution
Let g(x)=x−(n+1)xn+1+nxn+2. Then g(1)=1−(n+1)+n=0; g′(x)=1−(n+1)2xn+n(n+2)xn+1, so g′(1)=1−(n+1)2+n(n+2)=0; and g′′(x)=−n(n+1)2xn−1+n(n+1)(n+2)xn, so g′′(1)=n(n+1). The denominator (1−x)2 has value and first derivative 0 at 1 and second derivative 2. By method (a) (at x=1) the limit is n(n+1)/2.
This agrees with the second hint: the expression is ∑k=1nkxk, whose value at x=1 is 1+2+⋯+n=21n(n+1).
Exercise●○○Hardy: Examples LVIII, 5
Find the limit of x{x2+a2−x} as x→∞.
Hint
Hardy suggests x=1/y, which turns it into a 0/0 form as y→+0.
Solution
With x=1/y (y>0) the expression is y21+a2y2−1. Now 1+u=1+21u+o(u), so the numerator is 21a2y2+o(y2) and the limit is 21a2. (Without calculus: multiply and divide by x2+a2+x to get a2x/(x2+a2+x)→a2/2.)
Exercise●●○Hardy: Examples LVIII, 7
Find the limits as x→0 of x31(cosecx−x1−6x) and x31(cotx−x1+3x).
Hint
Write cosecx=x1(1−6x2+120x4−⋯)−1 and expand (1−u)−1=1+u+u2+⋯ with u=x2/6−x4/120+O(x6).
Solution
With u=6x2−120x4+O(x6) we get u+u2=6x2+x4(361−1201)+O(x6)=6x2+3607x4+O(x6). So cosecx=x1+6x+3607x3+O(x5), and the first limit is 3607.
For the cotangent, cotx=cosxcosecx=(1−2x2+24x4)(x1+6x+3607x3)+O(x5)=x1+x(61−21)+x3(3607−121+241)+O(x5). The coefficients are −31 and 3607−30+15=−3608=−451. So the second limit is −451.
Exercise●●○Hardy: Examples LVIII, 8
Show that (sinxarcsinx−x2)/x6→181 and (tanxarctanx−x2)/x6→92 as x→0.
Hint
You need arcsinx=x+6x3+403x5+O(x7), tanx=x+3x3+152x5+O(x7) and arctanx=x−3x3+5x5+O(x7). Multiply, keeping terms up to x6.
Solution
sinxarcsinx=(x−6x3+120x5)(x+6x3+403x5)+O(x8). The x4 coefficient is 61−61=0, and the x6 coefficient is 403−361+1201=36027−10+3=181. So the first limit is 181.
tanxarctanx=(x+3x3+152x5)(x−3x3+5x5)+O(x8). The x4 coefficient is again 0, and the x6 coefficient is 51−91+152=459−5+6=92.
Method (b) would need six differentiations of each product: expansions are far quicker.
Check yourself
Check yourselfTrue or false?
If f(0)=ϕ(0)=0 and f(x)/ϕ(x)→l as x→0, then f′(x)/ϕ′(x)→l as well.
Why
False. Hardy’s Examples LVIII, 1: f=x2sin(1/x), ϕ=x. Then f/ϕ=xsin(1/x)→0, but f′/ϕ′=2xsin(1/x)−cos(1/x) oscillates. The rule only goes one way.
Check yourselfEnter a number
What is x→0limx3x−sinx?
Why
x−sinx=6x3−120x5+⋯, so the limit is 61. (Method (a): the third derivatives at 0 are 1 and 6.)
Check yourselfChoose one
A student computes x→0limx+1sinx “by L’Hôpital” as lim1cosx=1. What is wrong?
Why
The proof of method (b) uses f(0)=ϕ(0)=0 to write f/ϕ as a ratio of increments. Here ϕ(0)=1, and the function is simply continuous at 0, with value 0. Always check the form before differentiating.
Check yourselfEnter a number
What is x→0lim3sinx−sin3xtan3x−3tanx? (Examples LVIII, 3.)
Why
tan3x−3tanx=327−3x3+⋯=8x3+⋯ and 3sinx−sin3x=627−3x3+⋯=4x3+⋯, so the limit is 2. The same happens with any n=0,±1 in place of 3: the ratio is (n3−n)/6(n3−n)/3=2.
Hardy’s language
Summary
Summary
For 0/0 at x=0: if the first non-vanishing derivatives of f and ϕ have the same order n, the limit is f(n)(0)/ϕ(n)(0); in practice, compare the leading terms of the expansions.
L’Hôpital: if f(0)=ϕ(0)=0 and f′/ϕ′→l, then f/ϕ→l; proved by Cauchy’s mean value theorem.
The rule is sufficient, not necessary: x2sin(1/x)/x→0 while f′/ϕ′ oscillates. And it applies only to genuine 0/0 (or ∞/∞) forms.
Expansions with remainder o(xn) are unique, so they can be found by any correct algebra on known series.