In Chapter IX the logarithm was defined as an integral, logx=∫1xdt/t. To define the logarithm of a complex number ζ in the same way, we have to integrate 1/z from 1 to ζ. But between two points of the plane there is no single interval to integrate over: there are infinitely many curves. §229 says what an integral along a curve means. Whether its value depends on the curve is the question the rest of the chapter turns on.
The idea
An ordinary integral ∫abf(x)dx is the limit of sums ∑f(xj)(xj+1−xj): cut [a,b] into small pieces, multiply each length by a value of f, and add. Along a curve C from A to B in the Argand diagram we do the same thing with complex numbers. Mark points z0=A,z1,…,zn=B along C; each step zj+1−zj is now a small complex number (a displacement with a direction), and we form
j=0∑n−1f(zj)(zj+1−zj).
Every product f(zj)(zj+1−zj) is a complex number, so the sum can be drawn as a chain of arrows laid tip to tail. As the steps get shorter, the end of the chain settles down: that end point is the integral of f along C.
ExploreIntegrate along a path you can bend
Try this.With f(z)=z2, increase the number of chords n and watch the end of the chain approach the hollow dot. Then drag the purple control point P to bend the path, keeping A and B fixed. Repeat with f(z)=x−iy, and with f(z)=1/z, bending the path so that it passes on the other side of the origin.
∑f(zj)Δzj→∫Cf(z)dz, f(z)=z2
z-plane: the path C from A to B
xy
the products f(z_j)Δz_j, tip to tail
ReIm
path C (drag A and B)
bend it with P
chords and the chain of products
the integral along C
the integral along the segment AB
f(z)
Sum with n=6 chords: 0.6411−0.6495i. Integral along this path: 0.4587−0.144i. Along the straight segment from A to B: 0.4587−0.144i. The two integrals agree.
What to notice
For z2 the integral does not move when you bend the path: it is always 31(B3−A3). For x−iy it moves with every change of shape. For 1/z it stays put while you bend the path without crossing the origin, but jumps (by 2πi) when the path is pulled across 0. That last behaviour is the logarithm, the subject of the next two lessons.
The explorer computes the sums; a definition must avoid limits of sums over arbitrary subdivisions and instead reduce everything to ordinary integrals of real functions, which we already have. That is what Hardy does: he describes the curve by a parameter.
Let the arc C from A to B be given by x=ϕ(t), y=ψ(t) for t0≤t≤t1, where ϕ and ψ have continuous derivatives and the point moves along C from A to B as t increases. If g(x,y) and h(x,y) are continuous, then
where the integral of a complex function of the real variable t means the integral of its real part plus i times the integral of its imaginary part. A path made of several smooth pieces (a polygon, say) is handled by adding the integrals over the pieces.
Two properties are used again and again.
TheoremIntegrals of derivatives depend only on the end-points
Let F be a polynomial or rational function of z, with derivative F′, and let C be a path from A to B on which F has no pole. Then
∫CF′(z)dz=F(B)−F(A),
whatever the shape of C. In particular ∫Czndz=n+1Bn+1−An+1 for every integer n≥0, and for every integer n≤−2 if C avoids 0.
LemmaThe size of an integralHardy: cf. §170 (1)
If ∣f(z)∣≤M on C and C has length L=∫t0t1∣z′(t)∣dt, then ∫Cf(z)dz≤ML.
The theorem leaves out exactly one power: n=−1. There is no rational function F with F′(z)=1/z (exercise below), and that is why ∫dz/z can, and does, depend on the path.
Proofs
ProofIntegrals of derivatives depend only on the end-points
Strategy
Compose F with the path to get the complex function t↦F(z(t)) of the real variable t. Show that its derivative is F′(z(t))z′(t), which is exactly the integrand in the parametric form. Then the fundamental theorem of the integral calculus (§165), applied to the real and imaginary parts separately, does the rest.
Differentiate z(t)n
For a complex function w(t)=p(t)+iq(t) of a real variable, define w′(t)=p′(t)+iq′(t). The product rule (w1w2)′=w1′w2+w1w2′ holds, because it holds for the real and imaginary parts, which are sums of products of real functions. By induction, dtdz(t)n=nz(t)n−1z′(t) for n≥1.
Extend to quotients
In the same way (1/w)′=−w′/w2 where w=0 (differentiate w⋅(1/w)=1). Hence for any rational function, dtdF(z(t))=F′(z(t))z′(t), with F′ given by the usual formal rules.
Integrate
Write F(z(t))=P(t)+iQ(t). Then F′(z(t))z′(t)=P′(t)+iQ′(t), and so
∫CF′(z)dz=∫t0t1P′dt+i∫t0t1Q′dt={P(t1)−P(t0)}+i{Q(t1)−Q(t0)}=F(B)−F(A).
Nothing depended on the shape of C
Only the values of F at the two ends appear. For several smooth pieces, the contributions telescope: {F(z1)−F(A)}+{F(z2)−F(z1)}+⋯=F(B)−F(A).
∎
ProofThe size of an integral
Strategy
Reduce to the real inequality ∣∫g∣≤∫∣g∣ for a complex function g of a real variable, then bound ∣f(z(t))z′(t)∣.
∣∫g∣≤∫∣g∣
Let I=∫t0t1g(t)dt. If I=0 there is nothing to prove; otherwise choose c=Iˉ/∣I∣, of modulus 1, so that cI=∣I∣ is real. Then ∣I∣=∫Re{cg(t)}dt≤∫∣cg(t)∣dt=∫∣g(t)∣dt.
Bound the integrand
With g(t)=f(z(t))z′(t) we have ∣g(t)∣≤M∣z′(t)∣, so ∫Cfdz≤M∫t0t1∣z′(t)∣dt=ML.
∎
Worked examples and exercises
ExerciseOne path, computed by hand●○○Hardy: §229
Let C be the parabolic arc z=t+it2, 0≤t≤1, from 0 to 1+i. Compute ∫Czdz from the definition and check it against the theorem.
Hint
z′(t)=1+2it. Multiply out (t+it2)(1+2it) and separate real and imaginary parts.
Solution
(t+it2)(1+2it)=t+2it2+it2−2t3=(t−2t3)+3it2. So ∫Czdz=∫01(t−2t3)dt+i∫013t2dt=(21−21)+i=i. The theorem gives 21{(1+i)2−0}=21⋅2i=i. ✓
Exercisex−iy depends on the path●●○Hardy: §229
Compute ∫zˉdz from 1 to i (a) along the straight segment, (b) along the quarter of the unit circle. Then compute it once round the whole unit circle, counter-clockwise.
Hint
(a) z=1+t(i−1), 0≤t≤1. (b) z=cost+isint, 0≤t≤21π; on the circle zˉ=1/z.
Solution
(a) z′(t)=i−1 and zˉ=1−t−it. The product is (1−t−it)(i−1)=(1−t)(i−1)−it(i−1)=−1+t+t+i(1−t+t), i.e. (2t−1)+i. Integrating over [0,1] gives 0+i=i.
(b) On the circle zˉz′(t)=(cost−isint)(−sint+icost)=i, so the integral is ∫0π/2idt=21πi.
The two answers differ, so zˉ has no “antiderivative” in the sense of the theorem. Once round the circle, the same computation gives ∫02πidt=2πi. In general ∮zˉdz=2i× (enclosed area), which is why every bend in the explorer changed the value.
ExerciseOnce round the origin with 1/z●○○Hardy: §229
Let C be the circle z=R(cost+isint), 0≤t≤2π. Show that ∫Cdz/z=2πi for every R>0, and that ∫Czndz=0 for every other integer n.
Hint
z′(t)=iz(t).
Solution
Since z′=iz, ∫Czndz=∫02πizn+1dt=iRn+1∫02π{cos(n+1)t+isin(n+1)t}dt by De Moivre’s theorem. For n=−1 both integrals of the trigonometric functions over a full period vanish. For n=−1 the integrand is i and the integral is 2πi. This agrees with the theorem: for n=−1 the circle starts and ends at the same point, so F(B)−F(A)=0.
ExerciseNo rational antiderivative for 1/z●●●Hardy: §229
Show that there is no rational function F with F′(z)=1/z.
Hint
If there were, the theorem would give ∫Cdz/z=0 round any closed path avoiding the poles of F. Which closed path is available?
Solution
Suppose F′=1/z, with F rational. F has finitely many poles, so choose R>0 so small that no pole of F other than possibly 0 lies in 0<∣z∣≤R, and also so that no pole lies on the circle ∣z∣=R. The circle is a closed path on which F has no pole, so the theorem gives ∮dz/z=F(B)−F(A)=0. But the previous exercise gives 2πi. Contradiction. So the integral of 1/z is a genuinely new function, which is how the logarithm enters.
ExerciseAn estimate●●○Hardy: §229, with the lemma
Let C be the upper half of the circle ∣z∣=R from R to −R. Show that ∫Cz2+1dz≤R2−1πR when R>1, so that the integral tends to 0 as R→∞.
Hint
On C, ∣z2+1∣≥∣z∣2−1.
Solution
On C, ∣z2+1∣≥∣z2∣−1=R2−1>0, so ∣1/(z2+1)∣≤1/(R2−1)=M. The length is L=πR. By the lemma the integral has modulus at most πR/(R2−1), which tends to 0.
Check yourself
Check yourselfChoose one
For which f does ∫Cf(z)dz from A=1 to B=i have the same value for every path (avoiding poles)?
Why
z2+3z is the derivative of the polynomial 31z3+23z2, so the theorem applies. x−iy and ∣z∣ are not functions of z in Hardy’s sense and their integrals depend on the shape. For 1/z the value depends on which side of 0 the path passes; there is no rational antiderivative.
Check yourselfEnter a number
What is ∫Cz2dz round the unit circle, once counter-clockwise?
Why
z2 is the derivative of 31z3, and on a closed path the end-points coincide, so the integral is 31(1−1)=0. The exceptional power is z−1, whose integral round the circle is 2πi.
Check yourselfTrue or false?
If f(z) is real for every z on C, then ∫Cf(z)dz is real.
Why
False. The steps dz are complex. For example f=1 along the segment from 0 to i gives ∫Cdz=i. The integral is real for every real f only if the path runs parallel to the real axis (so that every dz is real).
Check yourselfChoose one
Reversing the direction of C (from B to A instead of A to B)
Why
Reparametrise by s=t0+t1−t. Then z′ changes sign and the limits are swapped back, so the integral changes sign, exactly as ∫ab=−∫ba for real integrals. The direction is part of the path; that is why Hardy insists the point moves “from A to B”.
Hardy’s language
Summary
Summary
∫Cf(z)dz is the limit of sums of products f(zj)Δzj; by definition it is ∫t0t1f(z(t))z′(t)dt, split into real and imaginary parts.
If f=F′ with F rational, the integral is F(B)−F(A) for every path: the shape does not matter.
∫zndz is path-independent for every integer n except n=−1; once round the origin, ∮dz/z=2πi.
∣∫Cfdz∣≤ML. The direction of the path is part of the data.