Chapter X · The general theory of the logarithmic, exponential and circular functions

The general power aζa^\zeta

§§235–236 40 min

a^ζ = exp(ζ Log a) for any complex a ≠ 0 and ζ; its values lie on an equiangular spiral, there are infinitely many unless ζ is real and rational, and i^i is a sequence of real numbers.

Why this matters

What could 2i2^i, (1+i)2(1 + i)^{\sqrt 2} or iii^i possibly mean? Elementary algebra defines aζa^\zeta for rational ζ\zeta as roots of powers, and Chapter IX extended this to real ζ\zeta and positive aa by aζ=eζlog⁡aa^\zeta = e^{\zeta\log a}. §235 gives one definition covering every non-zero aa and every complex ζ\zeta, checks that it agrees with all the old ones, and finds the price: the power is in general infinitely many-valued, because the logarithm in its definition is.

The idea

Copy the real formula aζ=eζlog⁡aa^\zeta = e^{\zeta\log a}, but with the general logarithm: aζ=exp⁡(ζLog⁡a)a^\zeta = \exp(\zeta\operatorname{Log} a). Because Log⁡a\operatorname{Log} a has the values log⁡σ+i(ψ+2mπ)\log\sigma + i(\psi + 2m\pi), one for each integer mm, the numbers ζLog⁡a\zeta\operatorname{Log} a are

ζlog⁡σ+iζψ+2mπi ζ,m=0,±1,±2,…,\zeta\log\sigma + i\zeta\psi + 2m\pi i\,\zeta, \qquad m = 0, \pm1, \pm2, \dots,

equally spaced points on a straight line, a step 2πiζ2\pi i\zeta apart. Each of them gives a value of aζa^\zeta when we apply exp⁡\exp. Two of them give the same value only if they differ by a multiple of 2πi2\pi i, and that is rare. So we expect infinitely many values, lying on the image under exp⁡\exp of a straight line. That image is an equiangular spiral.

Explore All the values of a^ζ

Try this. Start with 20.3+0.1i2^{0.3 + 0.1i}. On the left are the numbers ζLog⁡a\zeta\operatorname{Log} a for m=−M,…,Mm = -M, \dots, M; on the right their exponentials, the values of aζa^\zeta. Change Im⁡ζ\operatorname{Im}\zeta to 00, then try Re⁡ζ=0\operatorname{Re}\zeta = 0. Finally pick the examples iii^i, (−1)1/3(-1)^{1/3}, 222^{\sqrt 2} and (1+i)1+i(1+i)^{1+i} from the menu (for the last, consecutive moduli differ by a factor e2π≈535e^{2\pi} \approx 535, so use the zoom).

aζ=exp⁡(ζLog⁡a),Log⁡a=log⁡σ+i(ψ+2mπ)a^\zeta = \exp(\zeta\operatorname{Log} a), \quad \operatorname{Log} a = \log\sigma + i(\psi + 2m\pi)
the numbers ζ Log a (one for each m)
ReIm
their exponentials: the values of a^ζ
ReIm
  • principal value (m = 0)
  • other values, m = −M … M
  • the line and its image under exp
a=2, ζ=0.3+0.1ia = 2, \ \zeta = 0.3 + 0.1i. Principal value 1.228+0.08527i1.228 + 0.08527i; m = 1 gives −0.2457+0.6091i-0.2457 + 0.6091i, m = −1 gives −0.5594−2.239i-0.5594 - 2.239i. 5 of the 7 values shown fit in the window. The moduli form a geometric progression and the amplitudes an arithmetic one: an equiangular spiral.
What to notice

The left-hand points are always equally spaced on a line, and the right-hand points on its image: an equiangular spiral, with the moduli in geometric progression. When ζ\zeta is real the line is vertical and the spiral becomes a circle; for (−1)1/3(-1)^{1/3} the points on the circle repeat after three steps (the three cube roots), for 222^{\sqrt 2} they never repeat. When ζ\zeta is purely imaginary the spiral becomes a ray: every value of iii^i is a positive real number.

Precise statements

Throughout, a=σ(cos⁡ψ+isin⁡ψ)a = \sigma(\cos\psi + i\sin\psi) with σ=∣a∣>0\sigma = |a| > 0 and −π<ψ≤π-\pi < \psi \le \pi, and ζ=ξ+iη\zeta = \xi + i\eta.

Definition The general power Hardy: §235

For a≠0a \ne 0 and any complex ζ\zeta,

aζ=exp⁡(ζLog⁡a),a^\zeta = \exp(\zeta\operatorname{Log} a),

where Log⁡a\operatorname{Log} a is any value of the logarithm of aa. Each value of Log⁡a\operatorname{Log} a gives a value of aζa^\zeta. The principal value of aζa^\zeta is exp⁡(ζlog⁡a)\exp(\zeta\log a), given by the principal logarithm.

Theorem Consistency with the earlier definitions Hardy: §235

(1) If a>0a > 0 and ζ\zeta is real, the principal value of aζa^\zeta is the real power eζlog⁡ae^{\zeta\log a} of Chapter IX. (2) If ζ=p/q\zeta = p/q is rational, the values of aζa^\zeta are exactly the qq values of ap/qa^{p/q} found in §48.

Theorem The general value of a^ζ Hardy: §236

The values of aζa^\zeta are

aζ=eξlog⁡σ−η(ψ+2mπ)[cos⁡{ηlog⁡σ+ξ(ψ+2mπ)}+isin⁡{ηlog⁡σ+ξ(ψ+2mπ)}],a^\zeta = e^{\xi\log\sigma - \eta(\psi + 2m\pi)}\bigl[\cos\{\eta\log\sigma + \xi(\psi + 2m\pi)\} + i\sin\{\eta\log\sigma + \xi(\psi + 2m\pi)\}\bigr],

mm any integer; the principal value is the case m=0m = 0.

Theorem How many values? Hardy: §236

aζa^\zeta has infinitely many distinct values unless ζ\zeta is real and rational. If ζ=p/q\zeta = p/q in lowest terms (q>0q > 0), it has exactly qq values.

Corollary De Moivre generalised Hardy: §236

If ζ\zeta is real and −π<ψ≤π-\pi < \psi \le \pi, the principal value of (cos⁡ψ+isin⁡ψ)ζ(\cos\psi + i\sin\psi)^\zeta is cos⁡ζψ+isin⁡ζψ\cos\zeta\psi + i\sin\zeta\psi.

The corollary needs the restriction on ψ\psi: for ψ=3π/2\psi = 3\pi/2 (not principal), (cos⁡ψ+isin⁡ψ)1/2(\cos\psi + i\sin\psi)^{1/2} has principal value (−i)1/2=cos⁡14π−isin⁡14π(-i)^{1/2} = \cos\frac14\pi - i\sin\frac14\pi, not cos⁡34π+isin⁡34π\cos\frac34\pi + i\sin\frac34\pi.

Proofs

Proof The general value of a^ζ
Strategy

Multiply out ζLog⁡a\zeta\operatorname{Log} a into real and imaginary parts, then apply the formula exp⁡(L+iM)=eL(cos⁡M+isin⁡M)\exp(L + iM) = e^L(\cos M + i\sin M) of §233.

The logarithm of aa

By §231, Log⁡a=log⁡σ+i(ψ+2mπ)\operatorname{Log} a = \log\sigma + i(\psi + 2m\pi).

Multiply by ζ\zeta

(ξ+iη){log⁡σ+i(ψ+2mπ)}=L+iM(\xi + i\eta)\{\log\sigma + i(\psi + 2m\pi)\} = L + iM with L=ξlog⁡σ−η(ψ+2mπ)L = \xi\log\sigma - \eta(\psi + 2m\pi) and M=ηlog⁡σ+ξ(ψ+2mπ)M = \eta\log\sigma + \xi(\psi + 2m\pi).

Exponentiate

aζ=exp⁡(L+iM)=eL(cos⁡M+isin⁡M)a^\zeta = \exp(L + iM) = e^L(\cos M + i\sin M), which is the formula. Putting m=0m = 0 uses the principal logarithm.

Proof How many values?
Strategy

Compare two values, with integers m≠nm \neq n. If η≠0\eta \ne 0 their moduli differ. If η=0\eta = 0 the moduli agree, and the values coincide exactly when the amplitudes differ by a multiple of 2π2\pi, which is a question about whether ξ\xi is rational.

The case η≠0\eta \neq 0

The modulus eξlog⁡σ−η(ψ+2mπ)e^{\xi\log\sigma - \eta(\psi + 2m\pi)} is a strictly monotone function of mm, so different mm give different values: infinitely many.

The case η=0\eta = 0: when do two values agree?

Now every value has modulus σξ\sigma^\xi, and the values for mm and nn have amplitudes ξ(ψ+2mπ)\xi(\psi + 2m\pi) and ξ(ψ+2nπ)\xi(\psi + 2n\pi). They are equal if and only if 2πξ(m−n)=2πk2\pi\xi(m - n) = 2\pi k for some integer kk, that is, ξ(m−n)\xi(m - n) is an integer.

ξ\xi irrational

Then ξ(m−n)\xi(m - n) is never an integer for m≠nm \ne n: all values are distinct, infinitely many.

ξ=p/q\xi = p/q in lowest terms

p(m−n)/qp(m - n)/q is an integer exactly when qq divides m−nm - n (since pp and qq have no common factor). So the values repeat with period qq in mm, and m=0,1,…,q−1m = 0, 1, \dots, q - 1 give exactly qq distinct values. (For ξ=0\xi = 0, q=1q = 1: the single value 11.)

Proof Consistency (2): rational exponents
Strategy

Compare the values just found with the qq-th roots of apa^p from Chapter III.

The values for ζ=p/q\zeta = p/q

With η=0\eta = 0, ξ=p/q\xi = p/q: ap/q=σp/q{cos⁡pq(ψ+2mπ)+isin⁡pq(ψ+2mπ)}a^{p/q} = \sigma^{p/q}\{\cos\tfrac pq(\psi + 2m\pi) + i\sin\tfrac pq(\psi + 2m\pi)\}.

They are qq-th roots of apa^p

Raising to the power qq by De Moivre: modulus σp\sigma^p, amplitude pψ+2pmπp\psi + 2pm\pi, i.e. apa^p. There are qq distinct such numbers (previous proof), and apa^p has exactly qq distinct qq-th roots (§48), so the two sets coincide.

Worked examples and exercises

Exercise i to the i ●○○ Hardy: Examples XCV, 1

Find all values of iii^i, and its principal value.

Hint
Log⁡i=(2k+12)πi\operatorname{Log} i = (2k + \tfrac12)\pi i.

Solution

ii=exp⁡(iLog⁡i)=exp⁡{i⋅(2k+12)πi}=exp⁡{−(2k+12)π}=e−(2k+12)πi^i = \exp(i\operatorname{Log} i) = \exp\{i \cdot (2k + \tfrac12)\pi i\} = \exp\{-(2k + \tfrac12)\pi\} = e^{-(2k + \frac12)\pi}, kk any integer. Every value is real and positive. The principal value (k=0k = 0) is e−π/2≈0.2079e^{-\pi/2} \approx 0.2079. The other values are this number multiplied by e−2kπe^{-2k\pi}, ranging from enormous (kk very negative) to tiny.

Exercise The values lie on an equiangular spiral ●●○ Hardy: Examples XCV, 2

Writing a value of aζa^\zeta as r(cos⁡θ+isin⁡θ)r(\cos\theta + i\sin\theta) with θ=ηlog⁡σ+ξ(ψ+2mπ)\theta = \eta\log\sigma + \xi(\psi + 2m\pi) (and ξ≠0\xi \ne 0), show that all the values lie on the spiral r=σ(ξ2+η2)/ξ e−ηθ/ξr = \sigma^{(\xi^2 + \eta^2)/\xi}\, e^{-\eta\theta/\xi}, and that the angle between this spiral and its radius vector does not depend on aa.

Hint
Eliminate s=ψ+2mπs = \psi + 2m\pi between log⁡r=ξlog⁡σ−ηs\log r = \xi\log\sigma - \eta s and θ=ηlog⁡σ+ξs\theta = \eta\log\sigma + \xi s.

Solution

From θ=ηlog⁡σ+ξs\theta = \eta\log\sigma + \xi s we get s=(θ−ηlog⁡σ)/ξs = (\theta - \eta\log\sigma)/\xi. Then log⁡r=ξlog⁡σ−η(θ−ηlog⁡σ)/ξ=ξ2+η2ξlog⁡σ−ηξθ\log r = \xi\log\sigma - \eta(\theta - \eta\log\sigma)/\xi = \dfrac{\xi^2 + \eta^2}{\xi}\log\sigma - \dfrac{\eta}{\xi}\theta, which is the stated spiral. For a curve r=Ce−ηθ/ξr = Ce^{-\eta\theta/\xi}, 1rdrdθ=−η/ξ\dfrac{1}{r}\dfrac{dr}{d\theta} = -\eta/\xi, and this is the cotangent of the angle between the tangent and the radius vector. It depends only on ζ\zeta, not on aa. (Consecutive values differ by the fixed factor exp⁡(2πiζ)\exp(2\pi i\zeta), so they are the vertices of an equiangular polygon inscribed in the spiral.)

Exercise The many values of e^ζ ●●○ Hardy: Examples XCV, 3

Show that the general value of eζe^\zeta (with a=ea = e in the definition) is eξ−2mπη{cos⁡(η+2mπξ)+isin⁡(η+2mπξ)}e^{\xi - 2m\pi\eta}\{\cos(\eta + 2m\pi\xi) + i\sin(\eta + 2m\pi\xi)\}, and that its principal value is exp⁡ζ\exp\zeta. What are the values of e1/2e^{1/2}?

Hint
Log⁡e=1+2mπi\operatorname{Log} e = 1 + 2m\pi i, so σ=e\sigma = e, ψ=0\psi = 0.

Solution

ζLog⁡e=(ξ+iη)(1+2mπi)=(ξ−2mπη)+i(η+2mπξ)\zeta\operatorname{Log} e = (\xi + i\eta)(1 + 2m\pi i) = (\xi - 2m\pi\eta) + i(\eta + 2m\pi\xi); exponentiate. With m=0m = 0 this is eξ(cos⁡η+isin⁡η)=exp⁡ζe^\xi(\cos\eta + i\sin\eta) = \exp\zeta. For ζ=12\zeta = \tfrac12: e1/2{cos⁡mπ+isin⁡mπ}=±ee^{1/2}\{\cos m\pi + i\sin m\pi\} = \pm\sqrt e: the two square roots of ee, of which the principal one is the positive one. This is why Hardy keeps the symbol exp⁡ζ\exp\zeta for the one-valued function.

Exercise Adding exponents ●●○ Hardy: Examples XCV, 7

Show that every value of aζ+ζ′a^{\zeta + \zeta'} is a value of aζ×aζ′a^\zeta \times a^{\zeta'}, but not conversely in general; and that the equation is true for principal values.

Hint
On the left one logarithm is used; on the right two, possibly different, ones.

Solution

A value of the product aζaζ′a^\zeta a^{\zeta'} is exp⁡{ζ(log⁡a+2mπi)}exp⁡{ζ′(log⁡a+2nπi)}=exp⁡{(ζ+ζ′)log⁡a+2πi(mζ+nζ′)}\exp\{\zeta(\log a + 2m\pi i)\}\exp\{\zeta'(\log a + 2n\pi i)\} = \exp\{(\zeta + \zeta')\log a + 2\pi i(m\zeta + n\zeta')\}, with mm and nn chosen independently. A value of aζ+ζ′a^{\zeta + \zeta'} is exp⁡{(ζ+ζ′)(log⁡a+2kπi)}\exp\{(\zeta + \zeta')(\log a + 2k\pi i)\}, which is the case m=n=km = n = k of the product. So every value of aζ+ζ′a^{\zeta + \zeta'} is a value of aζaζ′a^\zeta a^{\zeta'}.

With m≠nm \ne n the product generally has new values. Take a=1a = 1, ζ=ζ′=12\zeta = \zeta' = \tfrac12: 11/2=±11^{1/2} = \pm1, so the product 11/2×11/21^{1/2} \times 1^{1/2} has the values 11 and −1-1, while 11=11^{1} = 1 has only the value 11. So −1-1 is a value of the product but not of aζ+ζ′a^{\zeta + \zeta'}: the equation is not completely true. For principal values take m=n=k=0m = n = k = 0: both sides are exp⁡{(ζ+ζ′)log⁡a}\exp\{(\zeta + \zeta')\log a\}.

Exercise Find the fallacy ●●○ Hardy: Examples XCV, 11

“Since e2mπi=e2nπi=1e^{2m\pi i} = e^{2n\pi i} = 1 for all integers mm, nn, raising to the power ii gives e−2mπ=e−2nπe^{-2m\pi} = e^{-2n\pi}.” What is wrong?

Hint
Which values of 1i1^i and of (e2mπi)i(e^{2m\pi i})^i are being compared?

Solution

The rule (ew)i=ewi(e^w)^i = e^{wi} silently picks one value of a many-valued power. Here 1i=exp⁡(iLog⁡1)=exp⁡(i⋅2kπi)=e−2kπ1^i = \exp(i\operatorname{Log} 1) = \exp(i \cdot 2k\pi i) = e^{-2k\pi}: the number 11 raised to the power ii has all the values e−2kπe^{-2k\pi}. Raising both sides of e2mπi=e2nπie^{2m\pi i} = e^{2n\pi i} to the power ii produces the same set of values on each side; the argument then wrongly equates one particular value from the left (e−2mπe^{-2m\pi}, using the logarithm 2mπi2m\pi i) with one particular value from the right (e−2nπe^{-2n\pi}, using 2nπi2n\pi i). An equation between many-valued expressions does not let you choose values independently on the two sides.

Check yourself

Check yourself Enter a number

What is the principal value of iii^i (to four decimals)?

Why

log⁡i=12πi\log i = \tfrac12\pi i, so the principal value is exp⁡(i⋅12πi)=e−π/2≈0.2079\exp(i \cdot \tfrac12\pi i) = e^{-\pi/2} \approx 0.2079.

Check yourself Choose one

What is the principal value of (−8)1/3(-8)^{1/3}, in Hardy’s sense?

Why

log⁡(−8)=log⁡8+πi\log(-8) = \log 8 + \pi i, so the principal value is exp⁡{13(log⁡8+πi)}=2(cos⁡13π+isin⁡13π)=1+i3\exp\{\tfrac13(\log 8 + \pi i)\} = 2(\cos\tfrac13\pi + i\sin\tfrac13\pi) = 1 + i\sqrt3. The real cube root −2-2 is one of the three values (take m=1m = 1), but not the principal one. School algebra’s choice of −2-2 is a different convention.

Check yourself Choose one

How many distinct values has aζa^{\zeta} when a=2a = 2 and ζ=64\zeta = \tfrac{6}{4}?

Why

Reduce first: 64=32\tfrac64 = \tfrac32 in lowest terms, so there are q=2q = 2 values, ±22\pm 2\sqrt2. Counting by the unreduced denominator 44 would be wrong: the values for mm and m+2m + 2 already coincide.

Check yourself True or false?

Every value of iii^i is a real number.

Why

True: they are e−(2k+12)πe^{-(2k + \frac12)\pi}. In the explorer, iii^i is the case Re⁡ζ=0\operatorname{Re}\zeta = 0, where the spiral degenerates to a ray, here the positive real axis.

Hardy’s language

Summary

Summary
  • aζ=exp⁡(ζLog⁡a)a^\zeta = \exp(\zeta\operatorname{Log} a) for a≠0a \ne 0; principal value exp⁡(ζlog⁡a)\exp(\zeta\log a).
  • The numbers ζLog⁡a\zeta\operatorname{Log} a are equally spaced on a line; the values of aζa^\zeta lie on its image, an equiangular spiral (a circle if ζ\zeta is real, a ray if ζ\zeta is imaginary).
  • Infinitely many values unless ζ\zeta is real and rational; exactly qq values for ζ=p/q\zeta = p/q in lowest terms.
  • ii=e−(2k+12)πi^i = e^{-(2k + \frac12)\pi}, all real; principal value e−π/2e^{-\pi/2}.
  • Identities between powers may hold completely, only for principal values, or neither: always check which.