Two points of the plane define two families of circles crossing at right angles; the bilinear map (az+b)/(cz+d) is built from rotations, stretches, shifts and one inversion, and it sends circles and lines to circles and lines.
The second half of Examples XXI treats a complex function as a transformation: a rule that moves every point of one plane to a point of another, and so carries figures to figures. The simplest rational functions, az+b and (az+b)/(cz+d), turn out to be made of a few familiar moves, and they have a remarkable property: they never bend a circle into anything but a circle (or a straight line). These maps reappear throughout complex analysis, and the circle geometry built here is what Chapter X uses to understand logz.
The idea
Fix two points A=a and B=b. For a third point P=z, the complex number (z−b)/(z−a) carries two pieces of information:
its modulus λ=PB/PA, the ratio of the distances from P to B and to A;
its amplitude θ, the angle at P between the directions to A and to B, that is, the angle APB with a sign.
Ask where P can be if one of these is held fixed. By the inscribed angle theorem, fixing the angle APB puts P on an arc of a circle through A and B. Fixing the ratio PB/PA also gives a circle (the circle of Apollonius), which surrounds A or B. Hardy (Examples XXI, 19–20) proves both, and observes that every circle of the first family crosses every circle of the second at right angles.
ExploreTwo families of circles
Try this.Drag P. Keep it on the ochre arc and watch θ; keep it on the green circle and watch λ. Move P across the line AB and watch the sign of θ. Then switch on both whole families and drag A and B.
xy
A = a, B = b (drag)
P = z (drag)
am{(z − b)/(z − a)} = θ (solid arc)
|z − b| / |z − a| = λ
At P: λ=z−az−b=0.66, θ=amz−az−b=68∘. The green circle has its centre on the line AB. The two curves cross at P at 90∘.
What to notice
On the solid ochre arc θ is constant; on the dashed part of the same circle it differs by π (the angle at P there is the supplement). On the green circle λ is constant, and λ=1 gives the perpendicular bisector. The two curves always cross at 90∘. Small green circles hug A when λ is large and B when λ is small: A and B are the limiting points.
Now the transformations (Examples XXI, 21). Each of the following moves every point of the plane:
w=z+β is a translation (shift by the displacement β);
w=ρz with ρ>0 is a magnification about 0;
w=(cosϕ+isinϕ)z is a rotation about 0 through ϕ;
w=1/z sends a point of modulus r and amplitude θ to modulus 1/r and amplitude −θ: an inversion in the unit circle followed by a reflection in the real axis.
The general linear map w=az+b is a rotation through ama, a magnification by ∣a∣ and a translation by b, one after another. And the bilinear map w=(az+b)/(cz+d) with c=0 can be rewritten as
w=ca+cbc−ad⋅cz+d1,
that is, a linear map, then 1/(), then another linear map. Linear maps obviously send circles to circles and lines to lines. So everything hangs on the middle step.
ExploreA bilinear map in three steps
Try this.With w=1/z, drag the blue circle so that it passes through the red pole 0, then away from it. Do the same with the ochre line. Then choose another map and step through 1, 2, 3.
z-plane (drag the centre, the rim point and the two line points)
xy
image after step 3: w=z1
uv
a circle and its image
a straight line and its image
pole −d/c (left); a/c, where far-away points go (right)
Show the image after step
w=z1,ad−bc=−1=0. Pole: z=−d/c=0. Image of the circle: a circle, centre 0.698−0.465i, radius 0.349. Image of the line: a circle, centre −0.638−0.517i, radius 0.821.
What to notice
A circle not through the pole goes to a circle; a circle through the pole goes to a straight line; a line through the pole stays a line; a line missing the pole becomes a circle (through the image of “far away”, the hollow red point a/c). Step 1 only moves and stretches; all the bending happens in step 2.
Precise statements
TheoremArcs of constant angleHardy: Examples XXI, 19
For 0<θ<π, the points z with am{(z−b)/(z−a)}=θ form an arc of a circle through a and b (without its ends). The same circle’s other arc is where the amplitude is θ−π. As θ varies, these circles make up all the circles through a and b: the coaxal system of the common point kind.
TheoremCircles of ApolloniusHardy: Examples XXI, 20
For λ>0, λ=1, the points z with ∣z−b∣=λ∣z−a∣ form a circle whose centre K=(b−λ2a)/(1−λ2) lies on the line AB, and whose radius is λ∣a−b∣/∣1−λ2∣. For λ=1 it is the perpendicular bisector of AB. These circles form the coaxal system of the limiting point kind, with limiting points a and b, and each cuts every circle through a and b at right angles.
it is a composite of translations, rotations, magnifications and the map z↦1/z;
it is one-to-one: z=(−dw+b)/(cw−a);
it maps every circle and every straight line to a circle or a straight line. A circle or line through the pole −d/c goes to a straight line; the others go to circles.
Dividing by 1−λ2=0: zzˉ−zKˉ−zˉK+1−λ2∣b∣2−λ2∣a∣2=0, with K=1−λ2b−λ2a.
Complete the square
Since (z−K)(zˉ−Kˉ)=zzˉ−zKˉ−zˉK+∣K∣2, the equation reads ∣z−K∣2=∣K∣2−1−λ2∣b∣2−λ2∣a∣2. A short computation shows the right side equals (1−λ2)2λ2∣a−b∣2>0. So the locus is the circle with centre K and that radius.
The centre is on AB
K=1−λ21b−1−λ2λ2a has real weights adding to 1, so it lies on the line through a and b (Examples XXI, 12).
∎
For the bilinear map, the clever step is to describe circles and lines by one kind of equation. A circle ∣z−K∣2=ρ2 expands to zzˉ−Kˉz−Kzˉ+∣K∣2−ρ2=0, and a line, say R(Bˉz)=const, can be written Bˉz+Bzˉ+C=0. Both are of the form
Azzˉ+Bˉz+Bzˉ+C=0,A,C real,∣B∣2>AC,(∗)
with A=0 for a circle and A=0 for a line. (Conversely, every equation (∗) is a circle or a line.)
ProofBilinear maps send circles and lines to circles and lines
Strategy
Decompose the map into simple steps and show that each step turns an equation of the form (∗) into another one of the same form.
The decomposition
If c=0 the map is linear, w=(a/d)z+b/d. If c=0, check by combining fractions: ca+c(cz+d)bc−ad=c(cz+d)a(cz+d)+bc−ad=cz+daz+b. So the map is z1=cz+d, then z2=1/z1, then w=a/c+kz2 with k=(bc−ad)/c=0.
Linear steps
A linear map w=pz+q (p=0) is a rotation, a magnification and a translation; each moves a circle to a congruent-up-to-scale circle and a line to a line. (Algebraically: substitute z=(w−q)/p in (∗) and expand; the form is preserved with A replaced by A/∣p∣2.)
The step w=1/z
Substitute z=1/w in (∗) and multiply by wwˉ: A+Bˉwˉ+Bw+Cwwˉ=0. This is (∗) again, with the roles of A and C exchanged and B replaced by Bˉ.
Which become lines
The image is a line exactly when its new A, namely the old C, is 0, that is, when z=0 satisfied the original equation: the figure passed through 0, the pole of 1/z. Tracing back through the first step, the figure passes through the pole −d/c of the whole map.
∎
ProofFour points on a circle
Strategy
The amplitude of the cross ratio is the difference of two angles, both subtended by the chord z3z4; compare with the arcs of constant angle above.
Split the amplitude
(z1z2,z3z4)=z1−z4z1−z3/z2−z4z2−z3, so its amplitude is θ1−θ2 (up to 2π), where θk=am{(zk−z3)/(zk−z4)} is the signed angle at zk subtended by z3,z4.
Real means the angles agree up to π
The cross ratio is real iff θ1−θ2 is a multiple of π, i.e. θ1=θ2 or θ1=θ2±π.
Use the arcs of constant angle
By the first theorem, the points where the angle is θ2 or θ2±π form exactly the circle through z3, z4 and z2 (or the line through them, if θ2 is 0 or π). So the cross ratio is real iff z1 is on that circle or line.
∎
Worked examples and exercises
ExerciseWhat 1/z and (1+z)/(1−z) do to a polar grid●●○Hardy: Examples XXI, 23
Find the images, under w=1/z and under w=(1+z)/(1−z), of the circles ∣z∣=ρ and of the straight lines through 0.
Hint
For the second map, solve for z: z=(w−1)/(w+1). Then ∣z∣=ρ and “amz is fixed” become conditions on (w−1)/(w+1), which you met above.
Solution
w=1/z.∣w∣=1/∣z∣ and amw=−amz. So the circle ∣z∣=ρ goes to the circle ∣w∣=1/ρ, and the line through 0 at angle ϕ goes to the line through 0 at angle −ϕ. (Both pass through the pole 0, so lines go to lines.)
w=(1+z)/(1−z). Here z=(w−1)/(w+1). The circle ∣z∣=ρ becomes ∣w−1∣=ρ∣w+1∣: for ρ=1 a circle of Apollonius with limiting points 1 and −1, for ρ=1 the imaginary axis. A line through 0, where amz=ϕ or ϕ±π, becomes the set where am{(w−1)/(w+1)} is ϕ or ϕ±π: a whole circle through 1 and −1 (the real axis when ϕ=0). So the map turns the polar grid into the two coaxal systems with points ±1, which explains why they are orthogonal: circles and rays about 0 are.
Exercise●●○Hardy: Examples XXI, 24
Show that z=(aZ+b)/(cZ+d) (with ad−bc=0) makes the circle ∣z∣=1 correspond to a straight line in the Z-plane if and only if ∣a∣=∣c∣.
Hint
∣z∣=1 means ∣aZ+b∣=∣cZ+d∣.
Solution
The points Z that go to ∣z∣=1 satisfy ∣aZ+b∣=∣cZ+d∣.
If ∣a∣=∣c∣ (both nonzero, since otherwise ad−bc=0), divide by ∣a∣=∣c∣: ∣Z+b/a∣=∣Z+d/c∣. The points −b/a and −d/c are distinct (else ad=bc), so this is their perpendicular bisector, a straight line.
If ∣a∣=∣c∣: when a=0 the condition is ∣Z+d/c∣=(∣a∣/∣c∣)∣Z+b/a∣ (if c=0), a circle of Apollonius since the ratio is not 1; when c=0 it is ∣aZ+b∣=∣d∣, a circle; when a=0 it is ∣cZ+d∣=∣b∣, a circle. So the image is a line exactly when ∣a∣=∣c∣. (Equivalently: the pole −d/c maps to ∞, and the circle ∣z∣=1 “passes through ∞” exactly when a/c, the image of Z=∞, lies on it, i.e. ∣a/c∣=1.)
ExerciseComplex functions of a real variable●●○Hardy: Examples XXI, 31 (i)–(iii)
As the real number t varies, what curves are traced by z=a+(b−a)t, by z=c+ρ1−ti1+ti (ρ>0), and by z=(a+bt)/(c+dt) (with bc−ad=0)? For the first, which values of t give the segment from a to b?
Hint
For the second, compute ∣(1+ti)/(1−ti)∣. For the third, think of t as a point moving along the real axis of a plane, and apply the bilinear theorem.
Solution
(i)z−a=t(b−a) is a real multiple of b−a: the line through a and b. The segment between them is 0≤t≤1; t<0 gives the part beyond a, and t>1 the part beyond b.
(ii)∣1+ti∣=∣1−ti∣, so ∣z−c∣=ρ: the circle of centre c and radius ρ. Every point except c−ρ is reached exactly once (that point corresponds to t=±∞), so z goes round the circle once as t runs over all reals.
(iii) The map Z↦(a+bZ)/(c+dZ) is bilinear because bc−ad=0. As t runs along the real axis (a straight line) its image runs along a circle or a straight line, by the bilinear theorem. (It is a straight line exactly when the pole −c/d is real or d=0.)
If (z1z2,z3z4)=λ, show that swapping z1 and z2 gives 1/λ and swapping z2 and z3 gives 1−λ.
Hint
For the second, put u=z1−z3 and so on, and use (z1−z3)(z2−z4)−(z1−z4)(z2−z3)=(z1−z2)(z3−z4), which you can verify by expanding.
Solution
Swapping z1,z2 exchanges numerator (z1−z3)(z2−z4) with denominator (z1−z4)(z2−z3), giving 1/λ.
Expanding, (z1−z3)(z2−z4)−(z1−z4)(z2−z3)=−z1z4−z3z2+z1z3+z4z2=(z1−z2)(z3−z4). Dividing by (z1−z4)(z2−z3): λ−1=(z1−z4)(z2−z3)(z1−z2)(z3−z4). Now swapping z2,z3 in the definition gives (z1z3,z2z4)=(z1−z4)(z3−z2)(z1−z2)(z3−z4)=−(λ−1)=1−λ. Combining these two moves generates the six values λ, 1−λ, 1/λ, 1/(1−λ), (λ−1)/λ, λ/(λ−1) that Hardy lists.
The circle and line images can also be seen on a whole grid at once. The shared map widget below shows Hardy’s Examples XXI, 23 map applied to a polar grid; compare it with the exercise above. (It also reports a winding number, which belongs to Chapter X; ignore it for now.)
ExploreA polar grid under (1+z)/(1−z)
Try this.Look at where the circles ∣z∣=r (blue) and the rays from 0 (ochre) go. Shrink the purple circle to a small circle round z=1.
w=1−z1+z
z-plane
xy
w-plane, w = f(z)
uv
circles |z| = r and their images
rays arg z = θ and their images
circle (drag centre and rim) and its image
w = 0
Grid
Circle: centre 0, radius 0.6 (it encloses z = 0). Its image winds 0 times round w=0. Maps the unit disc onto the right half-plane.
What to notice
Circles about 0 become circles of Apollonius about the limiting points ±1, rays become arcs of circles through ±1, and right angles are kept. A small circle round the pole z=1 becomes a huge circle: points near the pole are thrown far out.
Check yourself
Check yourselfEnter a number
Compute the cross ratio (z1z2,z3z4) for z1=1, z2=−1, z3=i, z4=−i.
Why
(1+i)(−1−i)(1−i)(−1+i)=−(1+i)2−(1−i)2=2i−2i=−1. The value is real, as it must be for four points on the unit circle, and equal to −1: the pairs ±1 and ±i are harmonic.
Check yourselfTrue or false?
The map w=1/z sends every straight line to a straight line.
Why
False. Only lines through the pole 0 go to lines. A line missing 0 goes to a circle through 0; for instance the line Rz=1 goes to the circle ∣w−21∣=21.
Check yourselfChoose one
What is the set of z with ∣z−1∣=∣z+1∣?
Why
It is the set of points equidistant from 1 and −1: the perpendicular bisector of the segment between them, which is the imaginary axis. This is the circle of Apollonius with λ=1, degenerated into a line.
Check yourselfChoose one
Under w=(z−i)/(z+i), what is the image of the real axis?
Why
For real z, ∣z−i∣=∣z+i∣ (the real axis is the perpendicular bisector of ±i), so ∣w∣=1. A bilinear map can only produce circles or lines, and here we get the unit circle; the point w=1 would need z=∞. This is the map in the widget’s third preset.
Hardy’s language
Summary
Summary
For fixed a,b: constant am{(z−b)/(z−a)} gives arcs of circles through a,b; constant ∣(z−b)/(z−a)∣ gives circles of Apollonius. The two families are orthogonal.
w=az+b is rotate, stretch, shift; w=1/z is inversion followed by reflection; w=(az+b)/(cz+d) is linear, then 1/(), then linear.
Circles and lines are the solutions of Azzˉ+Bˉz+Bzˉ+C=0; every step of a bilinear map preserves this form, so circles and lines go to circles and lines.
A figure through the pole −d/c goes to a line. Four points are concyclic (or collinear) iff their cross ratio is real.