You are on page 1of 9

Notes on Chapter 5

Learning goals §5.1:


• Idea of projection on transparent screen (or drawing canvas)
• Costruzione legittima
• Equidistant points are no longer equidistant after projection
• Projections of parallel lines meet on horizon, unless they are parallel to the canvas
• Maps of the form f(y) = (ay + b)/(cy + d) (fractional linear maps) begin to appear

5.1.1 The line from (-1,0) to (n, 0) has slope Δy/Δx = -1/(n + 1), hence its equation is
y = c – x/(n + 1). Since it passes through (n, 0) we have c = n/(n + 1), as claimed.

Extra Exercise A: In 5.1.1 a perspective drawing of a line is made on a 1-dimensional


vertical canvas. Compare this to the 2-dimensional case of figure 5.5. Compute the
projection map in this 2-dimensional case and compare your formulas with the formula y
= n / (n + 1) from 5.1.1. On each row horizontal distances are shrunk by a certain factor.
Compute this factor!

Solution Extra Exercise:


In figure 5.5, the horizontal width on the bottom row of the tiles equals 1 cm, whereas the
distance between the vanishing points V and W equals 3 cm. The distance between the
bottom line and the horizon equals 1.2 cm, which we lazily round off to 1 cm. So we take
the horizon at y = y0 and take W = (-3, 1) and V = (0, 1) as the two indicated vanishing
points on the horizon.
The line from (k, 0) on the bottom line to V has equation y = (1 – x/k). The line from (k +
j, 0) to W has equation y = (k + j – x)/(k + j + 3). Intersecting these lines we find (k + j +
3)(k – x) = k(k + j – x), i.e. x = 3k/(j + 3) and y = j/(j + 3).
This is very similar to the formula y = n/(n + 1), in 5.1.1. The expression x = 3k/(j + 3)
means that on the j-th row the horizontal distance Δk is multiplied by a factor 3/(j + 3).
For the front row (j = 0) this factor is 1 (as it should!); for the second (j = 1) it is 3/4 and
indeed the horizontal tile width equals ¾ × 1cm = 0.75 cm; on the third row the width
equals 3/5 × 1cm = 0.6 cm, and so on. Check this by measuring the distances! In the
limit j → ∞ the horizontal distances shrink to 0.

Remark: Note that distances are distorted and equidistant points do not remain
equidistant. Horizontally, however, distances are shrunk by a fixed factor (depending on
the row), so ratios of horizontal distances are preserved. If for example P, Q, R are points
on the same row, then the same ratio |PQ| : |QR| holds in reality and in the picture.
This is not true for vertical distances: for example, all rows have the same “depth” in
reality, but not in the figure. The same is seen in exercise 5.1.1: the equidistant points (n,
0) are mapped to non-equidistant points n/(n + 1). We will see in §5.7 that a certain “ratio
of ratios” is conserved (the so-called “cross ratio”).

1
5.1.2 2 – n/(n + 1) = (n + 2) /(n + 1), hence f(n/(n + 1)) = (n + 1) /(n + 2). So the
translation x → x + 1 on the x-axis corresponds via the projection to y → 1/(2 – y) on the
y-axis. In section 5.6 and 7.3 we shall meet these linear fractional functions again.
5.1.3 f(y) = y ⇔ y = 1. This is the image of the fix point of the corresponding
transformation x → x + 1 on the x-axis, i.e. of x = ∞. This image is the horizon. The
corresponding line of sight is the line through the eye parallel to the “floor”.

Learning goal §5.2:


• Perspective image of tiled floor if neither rows nor columns are parallel to canvas.

5.2.2

Learning goals §5.3:


• The definition of the real projective plane (denoted by RP2 or P2(R) or P(R3) or
simply by P2) as the set of all lines in R3 passing through the origin O.
• The definition of “points” and “lines” in this projective plane.
• Distinct “lines” in P2 always intersect in exactly one “point”.
• Two distinct “points” determine exactly one “line”.
• If V is a plane in R3 not passing through the origin O, then the points of V
correspond naturally to “points” of P2, so we can view V as a subset of P2. The
complement of this subset in P2 corresponds to “infinite points” of V.
• Similarly, the real projective line P1 is the set of lines through O in R2. If L is any
line in R2 not passing through O, then the points of L can be identified with
“points” of P1, hence L can be viewed as a subset of P1. Its complement consists
of exactly one point, “the infinite point of L”.

5.3.1 x=y

2
5.3.2 x = 0, y = 0, z = 0, x + y + z = 0: any two of these R3-planes = P2-lines intersect
along an R3-line = P2-point; any three of these R3-planes = P2-lines intersect in the origin
alone, which corresponds to the empty set in P2.
5.3.3 If B ≠ E then AB and BC have at least two distinct P2-points in common. By the
first axiom, this implies that the P2-lines AB and BC coincide. Hence A, B and C are on a
P2-line, contradicting the assumption.
5.3.4 By 5.3.3 AB ∩ BC = {B}. Analogously, BC ∩ CD = {C}. Since B ≠ C the three
lines AB, BC and CD have no common point.

Learning goals §5.4:


• Homogeneous coordinates are as important to projective geometry as ordinary
coordinates are to Euclidean geometry. They can reduce geometric problems to
(hopefully simple) calculations. We will pay more attention to them than Stillwell
does.
• Extra: Duality

5.4.1 Solve the linear system: a + 2b + 3c = 0 and a + b + c = 0. Solutions generated by


(a, b, c) = (1, -2, 1). Hence R3-plane (= P2-line) is given by the equation x – 2y + z = 0.
5.4.2 R3-line through origin generated by (x, y, z) = (1, -2, 1). This is the P2-point with
homogeneous coordinates (1:-2:1).
5.4.3 It doesn’t matter whether the unknowns are called a, b, c or x, y, z, the equations
are the same! The triple (a, b, c) corresponds to a P2-line; the triple (x, y, z) corresponds
to a P2-point. The P2-point lies on the P2-line if and only if ax + by + cz = 0. This
symmetry between P2-points and P2-lines is called duality. It is characteristic for
projective geometry.

INTERMEZZO: Duality

Dualizing a theorem about P2 means: replace the word “point” by “line” and conversely.
“Incidence” remains “incidence”. By dualizing the proof of the theorem you get a proof
of the dual theorem. Hence the dual of a true statement is also true!

As an example we dualize Pappos’ theorem. We omit the quotation marks around “line”
and “point”, so by “line” we mean a P2-line and by “point” we mean a P2-point.

Pappos Dual of Pappos


Given 2 distinct lines l and m, Given 2 distinct points L and M
3 distinct points A, B, C on l, 3 distinct lines a, b, c through L,
and 3 distinct points P, Q, R on m. and 3 distinct lines p, q, r through M.
Assume that the 6 points A, …, R Assume that the 6 lines a, …, r
are distinct from the intersection point are distinct from the line o = LM
O := l ∩ m. joining L and M.
Let AQ ∩ BP denote the intersection point Let l(aq , bp) denote the line joining the
of the lines AQ and BP, etc. points a ∩ q and b ∩ p, etc.
Then the 3 points AQ ∩ BP, AR ∩ CP and Then the 3 lines l(aq , bp), l(ar , cp) and
BR ∩ CQ are collinear, l(br , cq) are concurrent,

3
i.e. lie on a common line n. i.e. pass through a common point N.

Draw a figure of Pappos’ theorem. (For its dual, see next page.)
So using duality you get one theorem for free! Usually, this theorem is really new.
If you are unlucky, the dual theorem is identical to the original theorem and you get
nothing new. Sometimes the dual theorem is the converse of the original theorem.

Extra exercise: Dualize Desargues’ theorem: if two triangles are perspective from a line,
then they are perspective from a point. Is the dual theorem the converse of the original
theorem?

Figure of dual of Pappos’ theorem:

4
Learning goals §5.5 and §5.6:
• Given two projective lines l1 and l2 in P2 and a point P not on those lines, one can
project l1 onto l2 from P. This defines a bijection (linear projective map) between
the two lines.
• One can iterate the construction above to get bijections l1 ↔ l2 ↔ … ↔ lN. If we
take l1 = lN = l, then we get projective linear bijections of the line l to itself. These
are called projective automorphisms of the line l. We can look at projective
automorphisms of projective lines from different angles, depending on how we
look at projective lines:
a) P1 = R ∪ {∞}
b) P1 is the set of points with homogeneous coordinates (x : y)
c) P1 = P(R2) is the set of lines in R2 passing through (0,0).
(c) is just the mathematical definition of the projective line and corresponds to abstract
linear algebra; (b) is the same as (c) expressed in coordinates; (a) is just a “translation” of
(b) using the inhomogeneous coordinate x/y.
Projective automorphisms of the projective line:
We look at projective automorphisms of the projective line from the angles (a), (b) and
(c). We will see that the different descriptions are compatible with each other.
a) If we view R ∪ {∞} as a (“the”) projective line, then projective automorphisms
are exactly the fractional linear maps that map
x → (ax + b) / (cx + d),
where a,b,c,d are constants satisfying ad – bc ≠ 0.
b) If we choose homogeneous coordinates (x : y) on a projective line l, then the
projective automorphisms are exactly the maps that map
(x : y) → (ax + by : cx + dy),
where a,b,c,d are constants satisfying ad – bc ≠ 0.
o Note that the finite point (x:1), corresponding the real number x, is mapped
to (ax + b : cx + d) = (x′ : 1), where x′ := (ax + b) / (cx + d). So we see
that this description is compatible with the previous one.
o Note also that the infinite point (1:0), corresponding to ∞, is mapped onto
a/c. This is of course exactly what we want, since the limit of (ax + b) /
(cx + d) as x → ∞ is a/c.
c) If we return to the definition of P1 = P(R2) as the set of lines in R2 passing through
(0,0), then the projective automorphisms of P1 correspond to 2× 2-matrices
a b 
A=  of real numbers satisfying det(A) = ad – bc ≠ 0, i.e. they correspond
c d 
exactly to the invertible matrices defining invertible linear maps of R2. Indeed, if
p is an element of P1, i.e. p is a of lines in R2 passing through (0,0), then Ap is also
a line in R2 passing through (0,0), hence Ap is also an element of P1. If p has
homogeneous coordinates (x:y), i.e. if p is the line in R2 generated by P(x, y), then
Ap is the line generated by A(P) = (ax + by, cx + dy). Hence the action of A on P1
expressed in homogeneous coordinates is just
(x : y) → (ax + by : cx + dy),
as before. This shows that descriptions (a), (b) and (c) are compatible.

5
5.5.1 f1 (f2(x)) = a1(a2 x + b2) + b1= A x + B, where A = a1a2 and B = a1b2 + b1. Note that
f1 (f2(x)) is in general different from f2 (f1(x)).
5.5.2 More generally, the composition of any number of functions of the form f(x) = ax
+ b is again of this form, by 5.5.1.
5.5.3 Project (x, 1) on y = 1 onto (x′ , 0) the parallel line y = 0 from P(p,q). Assume that
q ≠ 0, 1. (What happens if q = 0? What happens if q = 1?) Then (x′ – p) / q = (x – p) / (q
– 1), hence x′ = ax + b with a = q / (q – 1) and b = p / (1 – q).

5.6.1 The algebra is easy:


y = (ax + b) / (cx + d)
y ⋅ (cx + d) = ax + b
cxy + dy = ax + b
cxy – ax = b – dy
(cy – a) ⋅ x = b – dy
x = (b – dy) / (cy – a).
(Note that this corresponds to inverting a 2 × 2 – matrix.)
It is more tricky to determine what goes wrong if ad – bc = 0, especially since we allow
the symbol “∞” using the rules
∞ ± c = ∞ for all real numbers c
c⋅ ∞ = ∞ for all real numbers c ≠ 0
∞⋅ ∞ = ∞
c / ∞ = 0 for all real numbers c
c / 0 = ∞ for all real numbers c ≠ 0
0 / 0, ∞ / ∞ and 0⋅ ∞ remain undefined.
(a ⋅ ∞ + b) / (c ⋅ ∞ + d) = a / c. This is defined unless a = c = 0.
Note that in this system -∞ = ∞! (Compare to figure 5.11 on p.97).
Instead of doing this formally (which is a bit messy because of all the case distinctions
between finite, infinite, zero and non-zero numbers), we consider a typical example,
which clearly shows what is going on. After the example we will return to the definition
of P1-points to show that everything is fine.
Here is the example: consider the expression (6x + 2) / (3x + 1). For all x ≠ -1/3
(including x =∞) this equals 2. So what goes wrong if we try to solve the impossible
equation 10 = (6x + 2) / (3x + 1)? The above yields 30x + 10 = 6x + 2, hence x = -1/3. For
this value of x the denominator of (6x + 2) / (3x + 1) vanishes. If the numerator was non-
zero, this would be no problem, since we know that c / 0 = ∞ for all real numbers c ≠ 0.
However, the numerator 6x + 2 does vanish for -1/3, so the expression returns the
undefined quotient “0/0” for x = -1/3. In general, if ad – bc = 0, then the expression (ax +
b) / (cx + d) is not well-defined x = -d/c = -b/a in R ∪ {∞}, since it returns the undefined
quotient 0:0. For all other values it is the constant a/b = c/d in R ∪ {∞}. In some sense
one could say that the map (6x + 2) / (3x + 1) takes the value 2 for all x ≠ -1/3, whereas
at x = -1/3 it takes all values of R ∪ {∞} simultaneously.
This discussion becomes much cleaner if we return to the definition of projective
points: they are lines through the origin in R2. Hence they are given by a line ax + by = 0.
This line is determined by its coefficients (a, b). We can have a = 0 (y-axis) or b = 0 (x-
axis), but not a = b = 0. Conversely, a pair (a, b) ≠ (0,0) determines a line and two pairs

6
determine the same line if and only if one is a non-zero multiple of the other. In other
words, the line is determined by the ratio (a : b). These are called the homogeneous
coordinates of the point on P1. We can identify R ∪ {∞} with P1 with having x∈ R
correspond to (x : 1) and ∞ to (1:0). Geometrically, x∈ R corresponds to the line through
the origin intersecting the line y = 1 in (x, 1) and ∞ corresponds to the horizontal through
the origin.
The 2 × 2 – matrix with rows (a, b) and (c, d) maps the line represented by (x : y)
to the line represented by (ax + b : cx + d). This map is bijective with inverse
determined by the 2 × 2 – matrix (d, -b; -c, a) if and only if its determinant ad – bc is
non-zero. If the determinant ad – bc vanishes, then the map is not invertible. This means
that there is a line (namely the one generated by (b, -a) ≡ (d, -c)) which is mapped to
(0,0) in R2 . Hence the corresponding element (b: -a) = (d: -c) of P1 has no well-defined
image in P1.
The corresponding description of the map on R ∪ {∞} is given in exercise 5.6.1.
In that exercise we had to pay special attention to the special elements ∞ and 0. The
advantage of the matrix/vector/homogeneous coordinate approach is that the special
elements (0:1) and (1:0) need no separate treatment: they are ordinary lines through the
origin in R2 just and the matrix sends them to other ordinary lines.

5.6.2 One should find A = a1 a2 + b1 c2, B = a1 b2 + b1 d2, C = c1 a2 + d1 c2 and D = c1 b2 +


d1 d2 (see exercise 5.6.3).

5.6.3 This follows from the formulas found in 5.6.2. Alternatively, we can use the
definition of the projective line as the set of lines in R2 passing through (0,0). Indeed,
x in exercise 5.6.2 corresponds to (x : 1), which is the point of the projective line
corresponding to the line generated by (x, 1) in R2. Now consider the linear
transformation f of R2 defined by the 2 × 2 – matrix
a b
c d
 
It maps the column vector (x, 1) to (ax+b, cx+d). If cx+d ≠ 0, then the line generated by
(ax+b, cx+d) is the line generated by (f(x), 1), where f(x) = (ax + b)/(cx + d). Hence the
map f(x) = (ax + b)/(cx + d) of P1 = R ∪{∞} corresponds to the linear transformation of
R2 defined by the 2 x 2 – matrix above. From this point of view, 5.6.3 is trivial. 5.6.2
follows immediately from it without any computation.

5.6.4 The determinant of the product of 2 matrices equals the product of the determinants.

7
Learning goals §5.7 and §5.8:
• In section §5.1 we have seen that distances are not preserved by projections, not
even ratios of distances.
• The cross ratio, however, is a ratio of ratios which is conserved.
• Four distinct points ABCD on a line l have the same cross ratio as EFGH on a line
m if and only if there is a projective transformation from l to m which maps
ABCD to EFGH respectively.
• Given three distinct points ABC on l an EFG on m, then there always is exactly
one projective transformation from l onto m mapping ABC to EFG, respectively.
• Now add a fourth point D, then the unique transformation of the previous item
maps D to the unique point H having the correct cross ratio, i.e. [EF; GH] = [AB;
CD]. Given ABCD and EFG, this condition determines H uniquely.
• The 24 permutations of ABCD give rise to 6 different values of the cross ratio. So
the 24 permutations fall into 6 subsets consisting of 4 permutations each having
the same cross ratio.
• The most important subset of four permutations with the same cross ratio is:
[AB; CD] = [BA; DC] = [CD; AB] = [DC; BA].
• Interchanging AB or CD (but not both) corresponds to λ → 1/λ . This implies
that [AB; CD] = -1 ↔ [AB; CD] = [BA; CD] ↔ [AB; CD] = [AB; DC]. In this case
we say that the pairs AB and CD form a harmonic set.
• Complete quadrangular sets give rise to harmonic sets.

5.7.1 The equation [p,q; r,s] = c, with given p, q, r and c is equivalent to a linear equation
in s. What happens if the coefficient of s vanishes?

5.7.2 [p,q; r,s] = (r – p)(s – q) / (r – q)(s – p) = 3 (s – 2) / s = 4/3, i.e. 4s = 9s – 18, i.e. s =


18/5 = 3.6.

5.7.3 In reality, the rows have the same width. Hence the cross ratio of 4 successive rows
equals 4/3. Since the cross ratio is constant under projective transformations, the same
holds for their perspective images. Hence
[0, 1; 1 + e, 1 + e + e2] = (1 + e) (e + e2) / e (1 + e + e2)
= (1 + e)(1 + e) / ( 1 + e + e2)
= (1 + 2e + e2) / ( 1 + e + e2)
equals 4/3, i.e. 3 + 6e + 3e2 = 4 + 4e + 4e2 which is equivalent to (e – 1)2 = 0. Hence we
find e = 1. Explain!

5.8.1 – 5.8.3: Easy


5.8.4 Let A denote the transformation that maps y to 1/y:
A(y) = 1/y
and let B be the transformation that maps y to 1 – y:
B(y) = 1 – y.
Note that A(A(y)) = 1 / (1/y) = y and B(B(y)) = 1 – (1 – y) = y. Hence the transformations
A2 and B2 are just the identity. We write
A2 = B2 = I.
Next, we compute the transformations AB and BA: AB maps y to

8
A(B(y)) = 1 / (1 – y)
and BA maps y to
B(A(y)) = 1 – 1/y.
Next, we consider products of three factors. We only have to compute ABA and BAB,
since other products contain A2 or B2, which are just I. For example AAB = IB = B; so
these products give nothing new. So here we go:
A(B(A(y))) = 1 / B(A(y)) = 1 / (1 – 1/y) = y / (y – 1)
and
B(A(B(y))) = 1 – A(B(y)) = 1 – 1/(1 – y) = y / (y – 1).
We conclude that
ABA = BAB.
This implies that there are no new products with four or more factors. Indeed, if
we multiply ABA = BAB by A we get a factor AA which cancels and if we multiply by B
we get a factor BB which also cancels. Hence the 6 values y, A(y) = 1/y, B(y) = 1 – y,
A(B(y)) = 1/(1 – y), B(A(y)) = 1 – 1/y and A(B(A(y))) = y / (y – 1) is all there is.

If you know a little group theory: A and B generate the group {I, A, B, AB, BA, ABA} and
the group is completely described by A2 = B2 = I and ABA = BAB.
5.8.5 We use the standard notation for permutations. For example (142) is the
permutation that maps 1 to 4, 4 to 2, 2 to 1 and 3 to itself.
5.8.6 By 5.8.1 the permutation (34) corresponds to the transformation A on the cross
ratio. By 5.8.2 the permutation (12) also corresponds to A.
By 5.8.3, the permutation (23) corresponds to B.
Since these 3 permutations generate all permutations, the corresponding transformations
on the cross ratio are generated by A and B.
For example:
[p, q; r, s]  [q, p; r, s]  [q, r; p, s]  [q, r; s, p]
y 1/y  1 – 1/y  1 / (1 – 1/y) = y / (y – 1).
I A  BA  ABA
and
[p, q; r, s]  [p, r; q, s]  [r, p; q, s]  [r, p; s, q]  [r, s; p, q]
y 1–y  1 / (1 – y)  1 – y y
I B  AB  AAB = B  BB = I.
In the language of group theory: we have a group homomorphism from the group S4 of
permutations of 4 elements onto the group {I, A, B, AB, BA, ABA}. This homomorphism
is surjective and its kernel is {(1), (12)(34), (13)(24), (14)(23)}.
This means
[p, q; r, s] = [q, p; s, r] = [r, s; p, q] = [s, r; q, p] = y.
The group S4 consists of 6 cosets of the kernel, each consisting of 4 permutations having
the same cross ratio:
[p, q; r, s] = [q, p; s, r] = [r, s; p, q] = [s, r; q, p] = y
[p, r; q, s] = [r, p; s, q] = [q, s; p, r] = [s, q; r, p] = 1 – y
[p, q; s, r] = [q, p; r, s] = [s, r; p, q] = [r, s; q, p] = 1 / y
[q, r; p, s] = [r, q; s, p] = [p, s; q, r] = [s, p; r, q] = 1 – 1/y
[q, r; s, p] = [r, q; p, s] = [p, s; r, q] = [s, p; q, r] = y / (y – 1)
[r, p; q, s] = [p, r; s, q] = [q, s; r, p] = [s, q; p, r] = 1 / (1 – y).

You might also like