Appendices · Appendices

The infinite in analysis and geometry

Appendix IV 30 min

Appendix IV: in analysis ∞\infty is only part of a phrase; in homogeneous geometry the points and the line at infinity are as real as any others.

Why this matters

Students who learn analysis and projective geometry side by side hear “infinity” used in two incompatible ways. In analysis we are told that ∞\infty is not a number, and that "ϕ(n)→∞\phi(n) \to \infty" is shorthand for a statement about large values. In geometry we are told that parallel lines meet “at infinity”, and that there is a whole “line at infinity”. Appendix IV, Hardy’s shortest, explains why there is no contradiction: the geometer’s infinite elements are genuine objects, defined without any limits, and the familiar story of a point “running off to infinity” is only an illustration of them.

(The appendix has no section numbers. Hardy first defines “ϕ(n)→∞\phi(n) \to \infty” as a whole phrase in §60.)

The idea

The infinite of analysis. Throughout the book, ∞\infty never stands alone. "ϕ(n)→∞\phi(n) \to \infty" is defined as one unit (§60): for every Δ\Delta there is an n0n_0 with ϕ(n)>Δ\phi(n) > \Delta for n≥n0n \ge n_0. ”∑un\sum u_n to infinity” means a limit of finite sums. No object called ∞\infty is ever produced; the symbol has meaning only inside these phrases. Hardy calls it an incomplete symbol, and the infinite of analysis a limiting infinite.

The infinite of geometry. Now change what a “point” is. Instead of a pair of numbers (x,y)(x, y), let a point be a class of triads (x,y,z)(x, y, z), not all zero, where two triads count as the same point when they are proportional. Write the class as (x:y:z)(x : y : z). Most of these classes have z≠0z \ne 0 and can be scaled to (x/z:y/z:1)(x/z : y/z : 1), so they correspond to ordinary points (x/z,y/z)(x/z, y/z). But the classes with z=0z = 0 are there too, on the same footing: they are not limits of anything, they are simply classes of triads. These are the points at infinity, and together they form the line at infinity z=0z = 0.

The idea is clearest one dimension down, on a line.

Explore A point of the projective line is a line through the origin

Try this. Drag the purple point round the dashed semicircle (or use the slider). The purple line through OO consists of all multiples of one pair (x,z)(x, z): this whole line is one point of the projective line. Follow where it meets the ochre line z=1z = 1. Then turn the purple line towards the horizontal, from the left and from the right.

xz
  • a point of the projective line: all multiples of (x, z) (drag)
  • the ordinary line z = 1
  • its correlate x/z
The point is the class of pairs (x:z)=(0.54:0.841)(x : z) = (0.54 : 0.841), that is, every pair (λx,λz)(\lambda x, \lambda z) with λ≠0\lambda \ne 0. Its correlate on the ordinary line is x/z=0.642x/z = 0.642. Turn towards the horizontal from either side and the correlate runs off to ±∞\pm\infty, while the class itself just moves continuously to (1:0)(1 : 0).
What to notice

Each purple line meets z=1z = 1 at the single number x/zx/z, its correlate on the ordinary line. As the purple line turns towards the horizontal, the correlate runs off to +∞+\infty or −∞-\infty (that is the limiting infinite of analysis), but the purple line itself never does anything strange: it simply becomes the horizontal line, the class (1:0)(1 : 0). That class is an honest point, the same one whichever way you approach, and it has no correlate, because the horizontal line never meets z=1z = 1.

Precise statements

Definition Common Cartesian geometry Hardy: Appendix IV

A point is a pair of real numbers (x,y)(x, y). A line is the set of points satisfying a relation ax+by+c=0ax + by + c = 0, where aa and bb are not both zero.

In this geometry two distinct lines meet in one point or, if they are parallel (a:b=a′:b′a : b = a' : b'), in none.

Definition Real homogeneous geometry Hardy: Appendix IV

A point is a class of triads of real numbers (x,y,z)(x, y, z), not all zero, two triads being in the same class when they are proportional; we write it (x:y:z)(x : y : z). A line is the set of points satisfying a relation ax+by+cz=0ax + by + cz = 0, where aa, bb, cc are not all zero. In homogeneous Cartesian geometry the points with z=0z = 0 are called points at infinity, and the line z=0z = 0 is the line at infinity.

The relation ax+by+cz=0ax + by + cz = 0 either holds for every triad in a class or for none, since it is unchanged when (x,y,z)(x, y, z) is multiplied by λ≠0\lambda \ne 0. So “the point lies on the line” makes sense.

Theorem Two lines always meet

In real homogeneous geometry, two distinct lines ax+by+cz=0ax + by + cz = 0 and a′x+b′y+c′z=0a'x + b'y + c'z = 0 have exactly one common point, namely the class of

(bc′−cb′,  ca′−ac′,  ab′−ba′).(bc' - cb',\; ca' - ac',\; ab' - ba').
Theorem The correlation with common geometry Hardy: Appendix IV

The map (x,y)↦(x:y:1)(x, y) \mapsto (x : y : 1) is a one-to-one correspondence between the points of common Cartesian geometry and the points (x:y:z)(x : y : z) of homogeneous geometry with z≠0z \ne 0. It carries the line ax+by+c=0ax + by + c = 0 onto the line ax+by+cz=0ax + by + cz = 0 with its single point at infinity, (b:−a:0)(b : -a : 0), removed.

So a line of common geometry is a line of homogeneous geometry with one point missing, and two parallel lines are two homogeneous lines whose only common point is that missing point.

Explore Parallel lines meet at a point of the line at infinity

Try this. Move the angles and offsets of the two lines and watch the common point in the read-out, both as a class (x:y:z)(x : y : z) and in ordinary coordinates. Bring the two angles close together, then press “Make them parallel”.

xy
  • first line
  • second line
  • common point (x/z, y/z)
Homogeneous equations −0.342x+0.94y−z=0-0.342x + 0.94y - z = 0 and −0.766x+0.643y+z=0-0.766x + 0.643y + z = 0. Common point: the class (x:y:z)=(1:0.7:0.316)(x : y : z) = (1 : 0.7 : 0.316). In ordinary coordinates it is (3.16,2.22)(3.16, 2.22). Bring the angles together and zz shrinks smoothly to 0 while the ordinary point runs away.
The common point is computed as the cross product of the two coefficient triads; it is never the zero triad unless the lines coincide.
What to notice

As the lines approach parallel, the ordinary intersection point races away, but the homogeneous triad changes smoothly: its zz shrinks continuously to 00. When the lines are parallel the common point is still there, (cos⁡ψ:sin⁡ψ:0)(\cos\psi : \sin\psi : 0), recording the direction ψ\psi the lines share. Parallel lines with the same direction meet at the same point at infinity, whatever their offsets.

Proofs

Proof Two lines always meet
Strategy

A common point is a nonzero solution (x,y,z)(x, y, z) of two linear equations in three unknowns. The given triad is a solution, it is nonzero when the lines are distinct, and every solution is a multiple of it.

The triad satisfies both equations

With P=(bc′−cb′, ca′−ac′, ab′−ba′)P = (bc' - cb',\ ca' - ac',\ ab' - ba'): a(bc′−cb′)+b(ca′−ac′)+c(ab′−ba′)=0a(bc' - cb') + b(ca' - ac') + c(ab' - ba') = 0 (all six terms cancel in pairs), and the same computation with a′,b′,c′a', b', c' in front gives 00. (PP is the cross product of the two coefficient triads, which is perpendicular to both.)

The triad is not zero

If all three components vanish, then (a,b,c)(a, b, c) and (a′,b′,c′)(a', b', c') are proportional, so the two equations define the same line. For distinct lines, P≠(0,0,0)P \ne (0, 0, 0), so PP defines a point.

There is no other common point

Since the coefficient triads are not proportional, the two equations are independent, and their solutions form a set of dimension 3−2=13 - 2 = 1: the multiples of one nonzero triad, here PP. All of these are the same point.

Proof The correlation with common geometry
Strategy

Points with z≠0z \ne 0 have exactly one representative with z=1z = 1; read off the ordinary coordinates from it. Then compare the equations of the lines.

One-to-one on points

If z≠0z \ne 0, the class (x:y:z)(x : y : z) contains exactly one triad with last entry 11, namely (x/z,y/z,1)(x/z, y/z, 1). So (x:y:z)↦(x/z,y/z)(x : y : z) \mapsto (x/z, y/z) inverts (x,y)↦(x:y:1)(x, y) \mapsto (x : y : 1).

Lines go to lines

(x,y)(x, y) satisfies ax+by+c=0ax + by + c = 0 exactly when (x,y,1)(x, y, 1) satisfies ax+by+cz=0ax + by + cz = 0. So the ordinary line corresponds to the points of the homogeneous line with z≠0z \ne 0.

Exactly one point is missing

The points of ax+by+cz=0ax + by + cz = 0 with z=0z = 0 satisfy ax+by=0ax + by = 0 with (x,y)≠(0,0)(x, y) \ne (0, 0); since a,ba, b are not both zero, these form the single class (b:−a:0)(b : -a : 0). This is the point at infinity of the line, and it depends only on the direction a:ba : b, not on cc: parallel lines share it.

The illustration and the reality. Move a point along the homogeneous line ax+by+cz=0ax + by + cz = 0 towards its special point (b:−a:0)(b : -a : 0). The correlated ordinary point moves along ax+by+c=0ax + by + c = 0 and its distance from the origin tends to infinity. This is where the vocabulary (“point at infinity”) comes from, and it is a useful picture. But it is only a picture. The special point was defined above as a class of triads, with no limit in sight, and it would exist even if nobody ever drew the correlation. Hardy’s complaint is that elementary textbooks present the picture as if it were the definition, which leaves students thinking that the geometer’s “point at infinity” is the analyst’s "→∞\to \infty" in disguise. It is not: one is an actual object, the other a manner of speaking.

Exercises

Our own exercises on the ideas of the appendix.

Exercise Where parallel lines meet ●○○

Find the common point of the lines x+2y−3=0x + 2y - 3 = 0 and 2x+4y+1=02x + 4y + 1 = 0 in homogeneous coordinates. Check that it lies on both lines and on the line at infinity.

Solution

Homogeneously the lines are x+2y−3z=0x + 2y - 3z = 0 and 2x+4y+z=02x + 4y + z = 0, with triads (1,2,−3)(1, 2, -3) and (2,4,1)(2, 4, 1). The formula gives (2⋅1−(−3)⋅4, (−3)⋅2−1⋅1, 1⋅4−2⋅2)=(14,−7,0)(2 \cdot 1 - (-3)\cdot 4,\ (-3)\cdot 2 - 1 \cdot 1,\ 1 \cdot 4 - 2 \cdot 2) = (14, -7, 0), the class (2:−1:0)(2 : -1 : 0). Check: 2+2(−1)−0=02 + 2(-1) - 0 = 0 and 4+4(−1)+0=04 + 4(-1) + 0 = 0, and z=0z = 0. It is the point at infinity in the direction (2,−1)(2, -1), which is the direction of both lines.

Exercise Every ordinary line meets the line at infinity once ●○○

Show that the line at infinity z=0z = 0 meets every other line in exactly one point, and that two ordinary lines have the same point at infinity if and only if they are parallel.

Solution

A line other than z=0z = 0 has an equation ax+by+cz=0ax + by + cz = 0 with (a,b)≠(0,0)(a, b) \ne (0, 0) (if a=b=0a = b = 0 it would be cz=0cz = 0, the line at infinity). Its points with z=0z = 0 satisfy ax+by=0ax + by = 0, which gives the single class (b:−a:0)(b : -a : 0). Two lines ax+by+c=0ax + by + c = 0 and a′x+b′y+c′=0a'x + b'y + c' = 0 have the same point at infinity when (b:−a)=(b′:−a′)(b : -a) = (b' : -a'), that is a:b=a′:b′a : b = a' : b', which is the condition for them to be parallel (or equal).

Exercise Two points determine one line ●●○

Show that two distinct points (x1:y1:z1)(x_1 : y_1 : z_1) and (x2:y2:z2)(x_2 : y_2 : z_2) lie on exactly one line. Which line joins (1:0:0)(1 : 0 : 0) and (0:1:0)(0 : 1 : 0)?

Hint
The roles of points and lines are symmetric: a line is also given by a class of triads (a:b:c)(a : b : c).

Solution

A line (a:b:c)(a : b : c) contains both points when ax1+by1+cz1=0ax_1 + by_1 + cz_1 = 0 and ax2+by2+cz2=0ax_2 + by_2 + cz_2 = 0. This is the same algebra as in the theorem with the roles exchanged: since the point triads are not proportional, the solutions (a,b,c)(a, b, c) are the multiples of the cross product (y1z2−z1y2, z1x2−x1z2, x1y2−y1x2)≠0(y_1z_2 - z_1y_2,\ z_1x_2 - x_1z_2,\ x_1y_2 - y_1x_2) \ne 0, which is one line. For (1,0,0)(1, 0, 0) and (0,1,0)(0, 1, 0) the cross product is (0,0,1)(0, 0, 1): the line z=0z = 0. Two different points at infinity are joined by the line at infinity, as they must be.

Exercise The circular points at infinity ●●○

Hardy mentions “the circular points at infinity”. In complex homogeneous geometry (triads of complex numbers), show that every circle

x2+y2+2gxz+2fyz+cz2=0x^2 + y^2 + 2gxz + 2fyz + cz^2 = 0

(the homogeneous form of x2+y2+2gx+2fy+c=0x^2 + y^2 + 2gx + 2fy + c = 0) passes through the two points (1:i:0)(1 : i : 0) and (1:−i:0)(1 : -i : 0).

Solution

At (1,±i,0)(1, \pm i, 0): x2+y2=1+(±i)2=1−1=0x^2 + y^2 = 1 + (\pm i)^2 = 1 - 1 = 0, and every other term contains z=0z = 0. So both points lie on every circle, whatever gg, ff, cc. These two points are not real, which is why they never appear in drawings; they are as actual as (1:0:0)(1 : 0 : 0), but in the complex plane of triads.

Exercise Dividing by zero on the projective line ●●○

On the real projective line (classes (x:z)(x : z) of pairs, not both zero), define J(x:z)=(z:x)J(x : z) = (z : x). Show that JJ is a well-defined one-to-one map of the projective line onto itself, that it corresponds to t↦1/tt \mapsto 1/t for ordinary numbers t=x/z≠0t = x/z \ne 0, and that it exchanges 0=(0:1)0 = (0 : 1) with the special point (1:0)(1 : 0). Contrast this with the statement ”1/t→∞1/t \to \infty as t→0t \to 0” in analysis.

Solution

If (x,z)(x, z) and (λx,λz)(\lambda x, \lambda z) are in the same class, so are (z,x)(z, x) and (λz,λx)(\lambda z, \lambda x); and (z,x)≠(0,0)(z, x) \ne (0, 0). So JJ is well defined, and J(J(p))=pJ(J(p)) = p shows it is one-to-one and onto. For z≠0z \ne 0, x≠0x \ne 0: the point t=x/zt = x/z goes to (z:x)(z : x), which corresponds to z/x=1/tz/x = 1/t. Also J(0:1)=(1:0)J(0 : 1) = (1 : 0) and J(1:0)=(0:1)J(1 : 0) = (0 : 1).

In analysis, 1/t1/t tends to +∞+\infty as t→0t \to 0 from the right and to −∞-\infty from the left, and "1/01/0" is still meaningless. On the projective line JJ is simply defined at 00, with value the special point, and approaching from either side gives the same point: the projective line closes up like a circle (as the semicircle in the first widget suggests). The two statements are about different objects.

Exercise Taking ∞\infty out of a sentence ●○○

Rewrite each statement without the symbol ∞\infty: (a) f(x)→∞f(x) \to \infty as x→∞x \to \infty; (b) ∑n=1∞2−n=1\sum_{n=1}^\infty 2^{-n} = 1; (c) “the series ∑1/n\sum 1/n diverges to ∞\infty”.

Solution

(a) For every Δ\Delta there is an x0x_0 such that f(x)>Δf(x) > \Delta whenever x>x0x > x_0 (§90). (b) For every δ>0\delta > 0 there is an n0n_0 such that ∣∑k=1n2−k−1∣<δ\left|\sum_{k=1}^{n} 2^{-k} - 1\right| < \delta for all n≥n0n \ge n_0; that is, the partial sums tend to 11. (c) For every Δ\Delta there is an n0n_0 such that 1+12+⋯+1n>Δ1 + \tfrac12 + \dots + \tfrac1n > \Delta for all n≥n0n \ge n_0. In each case the whole phrase containing ∞\infty is replaced by a statement about real numbers. That is what Hardy means by calling ∞\infty an incomplete symbol.

Check yourself

Check yourself True or false?

In common Cartesian geometry, any two distinct lines have exactly one common point.

Why

False: parallel lines, such as y=xy = x and y=x+1y = x + 1, have none. Homogeneous geometry adds points with z=0z = 0, and then any two distinct lines meet in exactly one point.

Check yourself Choose one

What is the point at infinity of the line y=3x+1y = 3x + 1?

Why

Homogeneously the line is 3x−y+z=03x - y + z = 0. Points with z=0z = 0 satisfy 3x−y=03x - y = 0, the class (1:3:0)(1 : 3 : 0): the direction of the line, slope 33. (3:−1:0)(3 : -1 : 0) uses the coefficients instead of the direction; it does not satisfy 3x−y=03x - y = 0. (0:0:1)(0 : 0 : 1) is the origin and (1:3:1)(1 : 3 : 1) is the ordinary point (1,3)(1, 3), neither on the line nor at infinity.

Check yourself True or false?

In Hardy’s analysis, ∞\infty denotes a number larger than every real number.

Why

False. There is no such real number, and Hardy never introduces one. ∞\infty appears only inside phrases such as "ϕ(n)→∞\phi(n) \to \infty" or "∑1∞\sum_1^\infty", each defined as a whole by a statement about real numbers.

Check yourself Choose one

According to Hardy, what is the status of the picture “the point at infinity is where a point ends up when it moves off along the line to infinity”?

Why

Hardy calls the correlation “historically important” and “often useful for purposes of illustration”, but insists that no rational account of the geometrical infinite can be based on it. Confusing the two is exactly the mistake the appendix is written to prevent.

Hardy’s language

The infinite of analysis is a ‘limiting’ and not an ‘actual’ infinite.

— G. H. Hardy, Appendix IV

Summary

Summary
  • In analysis, ∞\infty is an incomplete symbol: phrases such as "→∞\to \infty" are defined as wholes, and no object ∞\infty exists.
  • In homogeneous geometry a point is a class of proportional triads (x:y:z)(x : y : z); the points with z=0z = 0 form the line at infinity, and they are points like any others.
  • Any two distinct lines meet in exactly one point, the class of the cross product of their coefficient triads; parallel lines meet at infinity, at the point recording their direction.
  • The correlation (x,y)↔(x:y:1)(x, y) \leftrightarrow (x : y : 1) explains the name: moving towards a point at infinity sends the ordinary correlate off to infinity. But that is an illustration, not the definition.