(1 + z/n)ⁿ → exp z for complex z, from the logarithmic series; and the derivatives of log(1 + tz), exp φ(t) and (1 + tz)ᵐ needed for the binomial theorem.
For real x Hardy proved (1+x/n)n→ex (§215), the limit behind compound interest. §246 extends it to complex z, with a proof that runs entirely through the logarithmic series of the last lesson. Along the way it collects three differentiation rules for complex functions of a real variable t, which are exactly what the general binomial theorem (§247) needs.
The idea
(1+z/n)n is the product of n equal factors 1+z/n. Multiplying by 1+z/n turns a point through the small angle am(1+z/n)≈Imz/n and stretches it by ∣1+z/n∣≈1+Rez/n. So the points 1,(1+z/n),(1+z/n)2,…,(1+z/n)n form a polygon that turns and grows in n equal small steps, through a total angle of about Imz and a total stretch of about eRez. That is what expz does: it turns 1 through Imz and stretches it by eRez.
Explore(1 + z/n)ⁿ as a polygon
Try this.With z=πi, increase n from 1 to 200 and watch the end of the polygon. Then drag z elsewhere, for example to 1+3i and to −2+2i.
(1+nz)n→expz
ReIm
z (drag it)
the points (1 + z/n)ᵏ, k = 0 … n
the curve exp(tz), 0 ≤ t ≤ 1, ending at exp z
z=3.14i. Each step multiplies by 1+z/n=1+0.7854i: a turn through 0.666 radians and a stretch by 1.272. (1+z/n)n=−2.3206+1.2037i, expz=−1; distance 1.79.
What to notice
For z=πi and n=1 the single step lands at 1+πi, far from the answer; as n grows the polygon hugs the dashed half-circle more and more closely and its end approaches exp(πi)=−1. Each step is a slightly-more-than-unit stretch: ∣1+πi/n∣n=(1+π2/n2)n/2, which tends to 1. The error shrinks roughly like 1/n: doubling n halves the distance.
Precise statements
Throughout, h and t are real, and logarithms and powers have their principal values.
TheoremThe basic limitHardy: §246 (1)
For every complex z, h→0limhlog(1+hz)=z.
TheoremThe exponential limitHardy: §246 (2)
For every complex z,
n→∞lim(1+nz)n=expz.
TheoremThree derivativesHardy: §246 (3), (4)
Let z and m be complex, and t real with ∣tz∣<1. Then
dtdlog(1+tz)=1+tzz,dtd(1+tz)m=mz(1+tz)m−1.
More generally, if ϕ(t)=ψ(t)+iχ(t) with ψ, χ differentiable, then dtdexpϕ(t)=ϕ′(t)expϕ(t).
Proofs
ProofThe basic limit
Strategy
Expand log(1+hz) by the logarithmic series (valid once ∣hz∣<1), divide by h, and bound everything after the first term by a geometric series.
Expand
For ∣hz∣<1, log(1+hz)=hz−21(hz)2+31(hz)3−⋯ (§244). So hlog(1+hz)=z+ϕ(h,z) with ϕ(h,z)=−21hz2+31h2z3−⋯.
Bound the rest
∣ϕ(h,z)∣≤∣hz2∣{1+∣hz∣+∣hz∣2+⋯}=1−∣hz∣∣hz2∣, using k1≤1 for each coefficient.
Let h→0
The bound tends to 0, so log(1+hz)/h→z.
∎
ProofThe exponential limit
Strategy
Take h=1/n in the basic limit: nlog(1+z/n)→z. Then apply exp, which is continuous, and note that exp{nlog(1+z/n)} is exactly (1+z/n)n.
The exponent
By the basic limit with h=1/n: wn=nlog(1+z/n)→z.
Rewrite the power
expwn={explog(1+z/n)}n=(1+z/n)n, by the addition theorem (expnw=(expw)n) and because exp of any logarithm of a number is that number.
Continuity of exp
∣expwn−expz∣=∣expz∣∣exp(wn−z)−1∣≤∣expz∣∣wn−z∣e∣wn−z∣, by the inequality ∣expu−1∣≤∣u∣e∣u∣ (lesson on the exponential series). Since wn−z→0, the right side tends to 0.
∎
ProofThree derivatives
Strategy
For log(1+tz): a difference quotient reduces to the basic limit. For expϕ: write expϕ=eψ(cosχ+isinχ) and differentiate the real and imaginary parts. For the power: combine the two with (1+tz)m=exp{mlog(1+tz)}.
Derivative of log(1+tz)
For small real h, log(1+tz+hz)−log(1+tz)=log(1+1+tzhz). This uses the product rule for principal logarithms, which holds here because 1+tz and 1+hz/(1+tz) both lie in the right half-plane (when ∣tz∣<1 and h is small), so their principal amplitudes add to something in (−π,π). Dividing by h and applying the basic limit with z/(1+tz) in place of z gives z/(1+tz).
With ϕ(t)=mlog(1+tz) and the first step, ϕ′(t)=mz/(1+tz), so dtd(1+tz)m=1+tzmzexp{mlog(1+tz)}=mzexp{(m−1)log(1+tz)}=mz(1+tz)m−1, all powers principal. (The middle equality uses exp{mlogw}/w=exp{(m−1)logw}, since w=explogw.)
∎
Worked examples and exercises
ExerciseEuler's limit directly●●○Hardy: §246
Without using the theorem, show that (1+πi/n)n→−1 by computing the modulus and amplitude of 1+πi/n.
Hint
∣1+πi/n∣n=(1+π2/n2)n/2, and am(1+πi/n)=arctan(π/n).
Solution
By De Moivre, (1+πi/n)n has modulus (1+π2/n2)n/2 and amplitude narctan(π/n). For the modulus: log of it is 2nlog(1+π2/n2), which lies between 0 and 2n⋅π2/n2=π2/(2n), so it tends to 0 and the modulus tends to 1. For the amplitude: arctanu/u→1 as u→0, so narctan(π/n)=π⋅π/narctan(π/n)→π. Hence the limit is cosπ+isinπ=−1.
ExerciseHow fast?●●○Hardy: §246
Show that nlog(1+z/n)−z=−2nz2+O(1/n2), and deduce that (1+z/n)n=expz{1−2nz2+O(1/n2)}. What does this predict for z=πi and n=100?
Hint
Use the series for log(1+z/n) and expu=1+u+O(u2).
Solution
nlog(1+z/n)=z−2nz2+3n2z3−⋯, and the tail after −z2/(2n) is at most n2∣z∣31−∣z∣/n1 in modulus, so it is O(1/n2). Then (1+z/n)n=expz⋅exp(u) with u=−z2/(2n)+O(1/n2), and expu=1+u+O(u2)=1−z2/(2n)+O(1/n2).
For z=πi: z2=−π2, so (1+πi/n)n≈−1⋅(1+π2/(2n)). At n=100 the error is about π2/200≈0.049, along the real axis (beyond −1). The explorer’s read-out agrees: the distance at n=100 is about 0.05.
ExerciseA derivative to check●○○Hardy: §246 (4)
For z=i and m=21, compute dtd(1+it)1/2 at t=0 in two ways: from formula (4), and by differentiating (1+it)1/2=(1+t2)1/4{cos(21arctant)+isin(21arctant)}.
Hint
At t=0 the principal value of (1+it)−1/2 is 1.
Solution
Formula (4): mz(1+tz)m−1=21i⋅1=21i at t=0.
Directly: the modulus (1+t2)1/4 has derivative 21t(1+t2)−3/4=0 at t=0; the amplitude 21arctant has derivative 21 at t=0. At t=0 the point is 1, moving with velocity (radial 0)+i⋅1⋅21=21i. ✓
Exerciseexp is the limit along any route●●○Hardy: §246
Show that (1+z/x)x→expz as x→∞ through real values, where the power has its principal value.
Hint
Put h=1/x in the basic limit; for real x the power is exp{xlog(1+z/x)} by definition.
Solution
By definition (§236) the principal value of (1+z/x)x is exp{xlog(1+z/x)}. With h=1/x, the basic limit gives xlog(1+z/x)→z as x→∞, and continuity of exp (as in the proof) gives the result. Only the principal value is claimed: for non-integer x the other values of the power are exp{xlog(1+z/x)+2kπix}, which do not converge.
Check yourself
Check yourselfChoose one
In the polygon 1,(1+z/n),(1+z/n)2,…, what does each step do?
Why
Consecutive points are successive powers, so each is the previous one times 1+z/n. Adding z/n would produce a straight line from 1 to 1+z. Adding the terms zk/k! is the different polygon of the exponential series (lesson on §242); both polygons end near expz, but by different routes.
Check yourselfEnter a number
What is limn→∞∣1+πi/n∣n?
Why
∣1+πi/n∣n=(1+π2/n2)n/2, whose logarithm is at most π2/(2n)→0. So the limit is 1=∣expπi∣.
Check yourselfTrue or false?
If ϕ(t) is a differentiable complex function of the real variable t, then dtdexpϕ(t)=ϕ′(t)expϕ(t), exactly as for real ϕ.
Why
True; the proof writes expϕ=eψ(cosχ+isinχ) and differentiates. The chain rule for exp has the same form for complex-valued functions of a real variable.
Hardy’s language
Summary
Summary
log(1+hz)/h→z as real h→0, from the logarithmic series with remainder at most ∣hz2∣/(1−∣hz∣).
(1+z/n)n=exp{nlog(1+z/n)}→expz for every complex z; the error is about ∣z2expz∣/(2n).
Geometrically: n equal small turns and stretches add up to a turn through Imz and a stretch by eRez.
dtdexpϕ=ϕ′expϕ and dtd(1+tz)m=mz(1+tz)m−1 (principal values, ∣tz∣<1).