An irreducible projective variety of dimension 2 is called a projective surface. We call a surface S n ruled if it is covered by a family of lines. Recall that the Grassmanian 𝔾(1,n) is a projective variety parametrising lines in n; there is a sub-variety I n × 𝔾(1,n) such that the fibre over any point of 𝔾(1,n) is a line in n. We thus define a ruled surface by saying that there is a curve C G(1,n) such that S is the image p1(p2-1(C) I).

We have already seen two examples in the previous lecture. The first (trivial) example is the cone. Let C be a curve in n-1 and consider n-1 as obtained from n by projecting from a point P in n. Let S be the closure of the inverse image of C under the morphism n -{P}→ n-1. Clearly, S is a ruled surface. A slightly more interesting example is the surface Q in 3 that is defined by WY - XZ = 0. As we saw in the previous lecture, this is isomorphic to 1 × 1. In fact, if A = (w : x : y : z) is any point on Q, then the tangent plane TA to Q at A is defined by yW - zX + wY - xZ = 0. One computes that the intersection TA Q is the union of the parametric lines (here we assume that w⁄=0)

t ↦→ (w : x : xt : wt) and t ↦→ (w : wt : zt : z)

It follows that Q is a ruled surface in two different ways.

Now one could argue that the study of ruled surfaces is identical to the study of curves in 𝔾(2,n) for various n, but the latter approach is additionally complicated by the slightly more complicated geometry of 𝔾(2,n) as compared with the geometry of n. Hence, we study ruled surfaces more directly.

Consider a point A on a ruled surface S n and the projection pA : n -{A}→ n-1. If L is any line in S that does not contain A, then the image of L is again a line in n-1. Thus the closure T of the image of S -{A} is again a ruled surface providing that the image is a surface; the latter condition only fails when S is the cone over a curve in n-1 as constructed above. When T is a surface it is not isomorphic to S; when A is a smooth point of S, the correspondence between T and S is called an elementary transformation from T to S.

Recall that the closure of the graph of pA is an n-dimensional sub-variety ℙ˜nA n × n-1. The morphism ℙ˜nA n-1 identifies n-1 as the space of lines in n that contain A or equivalently the space of tangent directions to n at A. As is obvious, any point B of n such that B⁄=A determines a unique such tangent direction (given by the line joining A and B) so that morphism ℙ˜n n is an isomorphism outside A. On the other hand the fibre of this morphism over A is all of n-1.

If X is any irreducible variety in n that properly contains A, then let X˜A ℙ˜nA denote the closure of the graph of the restricted morphism pA|X : X -{A}→ n-1. When X is smooth at A the fibre of  ˜
XA X over the point A can be realised as the space of tangent directions to X at A. When X is singular, this could be used to define what one can think of as the locus of lines tangent to X at A; but this is not the collection of all lines in the Zariski tangent space of X at A when X is singular at A. In order to avoid confusion, the fibre of  ˜
XA X was called the apparent locus of X at A in classical geometry. Its algebraic description is as the “proj” of the associated graded ring obtained from the local ring of X at A. The morphism X˜A X is called the blow-up of X at A.

Returning to the ruled surface S, let S˜A S be the blow-up morphism as above. Let us further assume that A is a smooth point of S. By the above description, the inverse image of A in ˜SA is a smooth curve EA~
=1 which sits in n-1 as the locus of all directions which are tangent to S. Now, if L is a line in S which contains A then L -{A} is collapsed to a point B ∈ EA under pA. In general, there is a unique such line (in fact Q is essentially the only case where this is not the case). Thus, the elementary transformation from S to T with centre A is obtained by replacing the line LA which contains A with a point B which lies on a line EA which takes the place of LA in the ruling of T.

Let us temporarily specialise to the case when the base of the ruling is isomorphic to 1.

Let Cd denote the rational normal curve of degree d in d. Fix an isomorphism fd : 1 Cd. The projective spaces d and e can be put as disjoint linear spaces in the projective space d+e+1; let us denote these by Ld and Me respectively. Each point A of d+e+1 which is not contained in Ld Me gives rise (uniquely) to two points B and C in these linear spaces such that A lies on the line joining B and C in d+e+1. The linear span of Me and A meets Ld in a unique point B; similarly for C. The intersection of the span of Me and A with the span of Ld and A is the line joining B and C. In particular, we see that the locus of lines joining fd(t) with fe(t) as t varies over 1 gives a ruled surface Sd,e in d+e-1. A general hyperplane meets each of these lines in one point. It follows that the intersection of a general hyperplane with Sd,e gives a rational normal curve of degree d + e in d+e. Consider a hyperplane H that contains Ld. Its intersection with Sd,e already contains Cd. Let A be a point of H Sd,e \ Cd. Let L be the line in Sd,e that contains A. Since L meets H and A as well as the endpoint of L that lies in Cd, we see that L lies in H. Thus (from degree considerations) it follows that H Sd,e in general consists of Cd and e lines L1, …, Le from the ruling. Similarly, a general hyperplane containing Me intersects Sd,e in Ce and d lines from the ruling on Sd,e. Now, e general lines in a projective space n will have a linear span of dimension 2e - 1 as long as n 2e - 1. Thus, if d > e, the span of any e lines in the ruling of Sd,e determines a linear subspace of dimension 2e - 1 in d+e+1; in particular, there are many hyperplanes that contain such an e-tuple of lines. The residual intersection of such a hyperplane with Sd,e gives a curve of degree d. One can easily show that this is another rational normal curve of degree d which too spans a linear subspace of d+e+1 which is disjoint from Me. On the other hand we cannot take the linear span of d general lines in this situation since their span will be all of d+e+1. Using this one can show that Ce is the curve of least degree in Sd,e. Similarly, one can show that the elementary transform of Sd,e from a point not on Ce results in Sd-1,e, whereas the elementary transform of Sd,e from a point on Ce results in Sd,e-1.

In order to explain the next statement we need to introduce the notion of self-intersection of a curve on a surface S. Let S n be a surface and C a curve on S. Let F be a general homogeneous polynomial of large degree that vanishes on C and let V (F) denote the hypersurface in n defined by the vanishing of F. One can show that V (F) S is of the form C + D for D an irreducible curve on S. We define the self-intersection (C.C) of C as deg(C)deg(F) - (C D); where (C D) denotes the intersection of C and D as defined earlier. As in the case of most such definitions the main work goes into proving that this number is independent of the choice of F; in fact it is also independent of the embedding of S in projective space. We note that if H = n-1 S is the intersection of S with a hyperplane in n then (H.H) = deg(H) = deg(S).

Returning to the ruled surfaces Se,d, let L be a line which is the ruling of Se,d. A general hyperplane that contains L will meet all other lines in the ruling transversely at one point. It follows that the intersection of such a hyperplane with S is of the form C + L where C meets L in exactly one point. Thus (L.L) = 0. Now some easy calculations show that (Ce.Ce) = e - d in Sd,e. In particular, when d > e this is a negative number. This is not un-correlated with what we showed above—viz. the curve Ce does not “move” in this case.

Let us move on to more general ruled surfaces. Let S n be a ruled surface. We wish to study curves H on S with the property that they meet each line in the ruling of S transversely; such curves are called “sections” of the ruling for obvious reasons. By a procedure similar to the one used to construct Sd,e, it is not difficult to construct surfaces which have sections H where (H.H) can become as small as one wants. In the interests of restricting one’s attention to “interesting” surfaces one can look at the case where (H.H) is bounded below. One such restriction that has proved very worthwhile is to require that (H.H) > 0 for every section H. Such ruled surfaces are called “stable”. A slight weakening of this is to allow equality as well; in that case the ruled surface is called “semi-stable”.

M. S. Narasimhan and C. S. Seshadri were able to prove a remarkable topological description of stable (and semi-stable) ruled surfaces. This description could be used to find a parametrisation of all such surfaces. The study of these parameter spaces (or moduli spaces) has been the subject of numerous papers by algebraic geometers at the TIFR such as S. Ramanan, A. Ramanathan, V. B. Mehta, N. Nitsure, Y. Holla and others.