You are on page 1of 23

a

r
X
i
v
:
0
9
0
7
.
4
1
5
1
v
2


[
m
a
t
h
.
A
G
]


3
0

J
u
l

2
0
0
9
GLOBAL ASPECTS OF THE GEOMETRY OF SURFACES
BRIAN HARBOURNE
Abstract. Several open problems related to the behavior of the monoid of effective divisors and the nef cone
for smooth projective surfaces over an algebraically closed field are discussed, motivating and putting into
historical context concepts such as Mori dream spaces, Seshadri constants and the resurgence of homogeneous
ideals in polynomial rings. Some recent work on these topics is discussed along with the problem of which
ordinary powers of homogeneous ideals contain given symbolic powers of those ideals. Exercises, with
solutions, are included.
I. Lecture: Bounded Negativity
I.1. Introduction. A surface here will always mean a smooth irreducible closed 2 dimensional subscheme
of projective space, over an algebraically closed field k. A prime divisor on X is a reduced irreducible curve.
Notation I.1.1. Let X be a surface. The divisor class group Cl(X) is the free abelian group G on the prime
divisors, modulo linear equivalence. By intersection theory we have a bilinear form on G which descends to
Cl(X). Two divisors which induce the same intersections on curves (which in our situation are themselves
divisors, since X is a surface) are said to be numerically equivalent. We will denote numerical equivalence
on divisors by ∼; thus D
1
∼ D
2
means that D
1
· C = D
2
· C for all curves C. We denote Cl(X) modulo
numerical equivalence by Num(X). Also, EFF(X) denotes the submonoid of Cl(X) consisting of the classes
of effective divisors on X, and NEF(X) denotes the submonoid of Cl(X) of all classes F such that F · C ≥ 0
for all C ∈ EFF(X).
A basic fact is that Num(X) is a free abelian group of finite rank. If X is rational, then Num(X) = Cl(X).
If X is obtained by blowing up points p
1
, . . . , p
r
∈ P
2
, for example, then Cl(X) is the free abelian group on
the class L of the pullback of a line and on the classes E
i
of the blowings up of the points p
i
, hence Cl(X)
has rank r +1. The intersection form on Cl(X) in this case is defined by −L
2
= E
2
i
with L· E
i
= E
j
· E
i
= 0
for all i and all j = i, and the canonical class is K
X
= −3L +E
1
+· · · +E
r
.
We now recall the Adjunction Theorem and Riemann-Roch for surfaces:
Theorem I.1.2 (Adjunction). Let C be a prime divisor on a surface X and let K
X
be the canonical class
on X. Then there is a non-negative integer p
C
such that C
2
= 2p
C
−2 −C · K
X
.
Theorem I.1.3 (Riemann-Roch). Given any divisor D on a surface X, let χ(O
X
(D)) denote h
0
(X, O
X
(D))−
h
1
(X, O
X
(D)) +h
2
(X, O
X
(D)). Then
χ(O
X
(D)) =
D
2
−K
X
· D
2
+ χ(O
X
).
Riemann-Roch becomes especially useful when taken together with Serre duality, which for a surface X
says that h
i
(X, O
X
(D)) = h
2−i
(X, O
X
(K
X
−D)). Castelnuovo’s criterion for rationality is also useful:
Theorem I.1.4 (Castelnuovo). A surface X is rational if and only if h
0
(X, O
X
(2K
X
)) = h
1
(X, O
X
) = 0.
Next, we recall the Hodge Index Theorem:
Date: July 30, 2009.
Acknowledgments: These notes were prepared for lectures given at the summer school of the SFB/TR 45 Bonn-Essen-Mainz,
financed by the Deutsche Forschungsgemeinschaft, that took place March 23-27, 2009 at the Pedagogical University of Cracow,
with the goal of improving the training of PhD students and postdocs in the area, in particular of the members of the SFB/TR
45. I thank the organizers of the school, Stefan M¨ uller-Stach and Tomasz Szemberg, for their invitation to give these lectures
(and Tomasz for his careful reading of these notes), and I thank Joaquim Ro´e, fellow speaker at the Summer School in Krakow
and my host in Barcelona the week before the School when some of the work on these notes was carried out. I also thank Zach
Teitler for sharing his notes [Te] on multiplier ideals with the participants, and I thank Burt Totaro and J´ anos Koll´ ar for their
comments regarding the Bounded Negativity Conjecture.
1
2 B. HARBOURNE
Theorem I.1.5 (HIT). Given a surface X and D ∈ Num(X) with D
2
> 0, then the intersection form on
the space D

⊆ Num(X) of classes F with F · D = 0 is negative definite.
Finally, we recall the semicontinuity principle. We say points p
1
, . . . , p
r
are essentially distinct points of a
surface X, and that X
r+1
is the blow up of X at p
1
, . . . , p
r
, if p
1
∈ X = X
1
, π
1
: X
2
→ X
1
is the blow up of
X
1
at p
1
, and for 1 < i ≤ r we have p
i
∈ X
i
, and π
i
: X
i+1
→ X
i
is the blow up of X
i
at p
i
. By identifying
X
i+1
with X
i
away from p
i
, we can regard p
i+1
as being in X
i
when π
i
(p
i+1
) = p
i
. In this way distinct
points p
1
, . . . , p
r
∈ X can be regarded as being essentially distinct. Let π
j,i
: X
j
→ X
i
be the morphism
π
j−1
◦ · · · ◦ π
i
whenever j > i, and let E
i
be the divisor (or divisor class, depending on context) given by
π
−1
r,i
(p
i
). When X = P
2
, let L be the pullback to X
r+1
of a general line on X. Otherwise we will assume L
is the pullback to X
r+1
of some ample divisor on X.
Theorem I.1.6 (Semicontinuity Principle). Let a, a
1
, . . . , a
r
be integers, let p
1
, . . . , p
r
be general points of
X = P
2
and denote by X
r+1
the blow up of X at p
1
, . . . , p
r
, with L, E
1
, . . . , E
r
being the usual associated
classes. Also, given essentially distinct points p

1
, . . . , p

r
of X, let X

r+1
denote the blow up of X at p

1
, . . . , p

r
,
and let L

, E

1
, . . . , E

r
denote the associated classes.
(a) If aL −

i
a
i
E
i
∈ EFF(X
r+1
), then aL

i
a
i
E

i
∈ EFF(X

r+1
) for every choice of essentially
distinct points p

1
, . . . , p

r
of X.
(b) If aL

i
a
i
E

i
∈ NEF(X

r+1
) for some choice of essentially distinct points p

1
, . . . , p

r
of X and if
(aL

i
a
i
E

i
)
2
> 0, then aL −

i
a
i
E
i
∈ NEF(X
r+1
).
Proof. Following [K], parameterize essentially distinct points of X by schemes W
i
where we set b
0
: W
1
→ W
0
to be X → Spec (k) and recursively we define W
i+1
→ W
i
×
Wi−1
W
i
to be the blow up of the diagonal in
W
i
×
Wi−1
W
i
, setting b
i
: W
i+1
→ W
i
to be the composition of W
i+1
→ W
i
×
Wi−1
W
i
with the projection
π
1i
: W
i
×
Wi−1
W
i
→ W
i
to the first factor. The morphisms b
i
are smooth ([EGA, 17.3, 19.4]). [Here is a
proof. Note b
0
is smooth (since X is). Assuming b
i−1
is smooth, we see the projection W
i
×
Wi−1
W
i
→ W
i
is
smooth ([Hr, Proposition III.10.1(d)]), and the exceptional locus B
i+1
⊂ W
i+1
for b
i
is smooth ([Hr, Theorem
II.8.24(b)]) and locally isomorphic to W
i
× P
1
. We now see that b
i
is smooth by checking surjectivity of
the induced maps on Zariski tangent spaces ([Hr, Proposition III.10.4(iii)]) at points x ∈ W
i+1
. Away from
B
i+1
, W
i+1
→ W
i
×
Wi−1
W
i
is an isomorphism and W
i
×
Wi−1
W
i
→ W
i
is smooth, hence surjectivity follows
for points x ∈ B
i+1
. At points x ∈ B
i+1
, the composition B
i+1
⊂ W
i+1
→ W
i
is smooth and thus the
map on tangent spaces induced by B
i+1
→ W
i
is already surjective at x, hence so is the one induced by
W
i+1
→ W
i
. Thus b
i
is smooth.]
Consider the pullbacks B

i
to W
r+1
of the divisors B
i
. For any ample divisor L on X, let L
′′
be the pullback
to W
r+1
via the blow ups b
i
and the projections π
2i
on the second factors. Let F = O
Wr+1
(aL
′′

i
a
i
B

i
).
Then for any essentially distinct points p

1
, . . . , p

r
of X we have a uniquely determined point w ∈ W
r
, the
fiber (W
r+1
)
w
of W
r+1
over w is X

r+1
, and the restriction F
w
of F to (W
r+1
)
w
is O
X

r+1
(aL

i
a
i
E

i
). By
the semicontinuity theorem ([Hr, Theorem III.12.8]), h
0
((W
r+1
)
w
, F
w
) is an upper semicontinuous function
of w. This implies (a).
Now consider (b). If (aL −

i
a
i
E
i
)
2
= (aL

i
a
i
E

i
)
2
> 0, we have s(aL −

i
a
i
E
i
) ∈ EFF(X
r+1
)
for some s ≫ 0. Pick some effective divisor C whose class is s(aL −

i
a
i
E
i
). For each prime divisor
component D of C, there is an open set of points p
i
for which D remains prime, since being effective is
a closed condition by (a), and since for only finitely many classes D

= a

L −

i
a

i
E
i
could D

and the
class of D − D

(or even s(aL −

i
a
i
E
i
) − D

in place of D − D

) both conceivably be classes of effective
divisors. Thus the decomposition of C as a sum of prime divisors is well-defined for general points, and
each component specializes to an effective divisor on X

r+1
which thus meets aL

i
a
i
E

i
, and hence
aL −

i
a
i
E
i
, non-negatively, so aL −

i
a
i
E
i
∈ NEF(X
r+1
), proving (b).
Here is a version of the same result stated for generic points, where X now is any surface and L comes
via pullback from some ample divisor on X:
Theorem I.1.7 (Semicontinuity Principle 2). Let a, a
1
, . . . , a
r
be integers, let p
1
, . . . , p
r
be generic points of
a surface X and denote by X
r+1
the blow up of X at p
1
, . . . , p
r
, with E
1
, . . . , E
r
being the usual associated
classes and L the pullback to X
r+1
from X of some ample divisor on X. Also, given essentially distinct
points p

1
, . . . , p

r
of X, let X

r+1
denote the blow up of X at p

1
, . . . , p

r
, and let L

, E

1
, . . . , E

r
denote the
associated classes.
GLOBAL ASPECTS OF THE GEOMETRY OF SURFACES 3
(a) If aL −

i
a
i
E
i
∈ EFF(X
r+1
), then aL

i
a
i
E

i
∈ EFF(X

r+1
) for every choice of essentially
distinct points p

1
, . . . , p

r
of X.
(b) If aL

i
a
i
E

i
∈ NEF(X

r+1
) for some choice of essentially distinct points p

1
, . . . , p

r
of X, then
aL −

i
a
i
E
i
∈ NEF(X
r+1
).
Proof. The proof of (a) is the same as for Theorem I.1.6(a). The proof for (b) is even simpler than before
since now we are not claiming that having a specific divisor F = aL −

i
a
i
E
i
be nef is an open condition
on the points p
i
. Instead, if F were not nef, then F · C < 0 for some C = c
0
L−

i
c
i
E
i
∈ EFF(X
r+1
), hence
C

= c
0
L

i
c
i
E

i
∈ EFF(X
r+1
) by (a), so F

· C

= F · C < 0 for F

= aL

i
a
i
E

i
, contradicting our
assumption that F

∈ NEF(X

r+1
).
Remark I.1.8. It is not hard to show that F ∈ NEF(X) implies F
2
≥ 0 (this is Exercise I.4.1(a)). It is
certainly possible, however, to have H ∈ EFF(X) with H
2
< 0. The question of the extent to which this
can happen is the main motivation for these notes. If in fact there is no H ∈ EFF(X) with H
2
< 0, then it
is easy to see that EFF(X) ⊆ NEF(X). It can also happen that NEF(X) ⊆ EFF(X), but in general neither
containment holds. For example, for n > 0, the base curve C on the Hirzebruch surface H
n
is effective but
has C
2
= −n so is not nef. For an example of a nef divisor which is not effective, see Exercise I.4.1(b).
However, in Exercise I.4.1(b), the class F is in fact ample (see Exercise III.2.4), thus some multiple of F is
effective (in fact 2F ∈ EFF(X) by Riemann-Roch), but divisors can be nef without being ample and without
any multiple being effective. For example, suppose X is given by blowing up r = s
2
generic points p
i
∈ P
2
.
Nagata [N2] proved that h
0
(X, O
X
(mF)) = 0 for all m > 0 when F = sL − E
1
− · · · − E
r
and s > 3. But
by specializing the points p
i
to general points of a smooth curve of degree s, we see that sL−E
1
−· · · −E
r
is nef after specializing, and hence nef to begin with by Theorem I.1.7. Thus for r = s
2
generic points p
i
,
sL −E
1
−· · · −E
r
is nef but not ample (since F
2
= 0), and, for each m > 0, m(sL −E
1
− · · · −E
r
) is not
the class of an effective divisor.
I.2. A Motivational Folklore Conjecture. There is a long-standing open conjecture involving bounded-
ness of negativity on surfaces. Let us say that a surface X has bounded negativity if there is an integer n
X
such that C
2
≥ n
X
for each prime divisor C ⊂ X.
Conjecture I.2.1 (Folklore: The Bounded Negativity Conjecture). Every surface X in characteristic 0 has
bounded negativity.
Remark I.2.2. Conjecture I.2.1 is false in positive characteristic. I thank Burt Totaro for bringing to my
attention the following example pointed out by J´anos Koll´ ar at a talk by Richard Harris at MSRI in Jan-
uary, 2009 (http://www.msri.org/communications/vmath/VMathVideos/VideoInfo/4111/show video).
Let X = C ×C, where C is a curve of genus g
C
≥ 2 defined over a finite field of characteristic p > 0. Let Γ
q
be the graph in X of the Frobenius morphism defined by taking qth powers, where q is a sufficiently large
power of p. Then Γ
q
is a curve on X with X
2
= q(2 −2g
C
) [Hr, Exercise V.1.10]. Since q can be arbitrarily
large, X does not have bounded negativity. However, it is as far as I know still an open problem even in
positive characteristic to determine which surfaces fail to have bounded negativity.
Some surfaces are known to have bounded negativity.
Corollary I.2.3. A surface X has bounded negativity if −mK
X
∈ EFF(X) for some positive integer m.
Proof. Since −mK
X
∈ EFF(X), there are only finitely many prime divisors C such that −mK
X
· C < 0. So,
apart from finitely many prime divisors C, we have −mK
X
· C ≥ 0, in which case C
2
= 2p
C
−2 −C · K
X

−2.
Example I.2.4. In particular, bounded negativity holds for K3 surfaces, Enriques surfaces, abelian surfaces,
and relatively minimal rational surfaces. But it is not always clear when it holds if one blows up points on
those surfaces.
Let EFF(X)/∼ denote the image of EFF(X) in Num(X). In preparation for giving a criterion for bounded
negativity to hold on X, we have the following proposition (taken from [Ro]):
Proposition I.2.5. If EFF(X)/∼ is finitely generated, then there are only finitely many prime divisors C
with C
2
< 0.
4 B. HARBOURNE
Proof. Let C
1
, . . . , C
r
be prime divisors whose classes generate EFF(X)/∼. Since each C
i
is the class of an
effective divisor, there are only finitely many prime divisors D such that D· C
i
< 0 for some i. Now let C be
a prime divisor with C
2
< 0; we have C ∼

i
m
i
C
i
for some m
i
≥ 0 and so 0 > C
2
=

i
m
i
C · C
i
, hence
C · C
j
< 0 for some j, and so C = C
j
.
This then gives a criterion for bounded negativity to hold.
Corollary I.2.6. If X is a surface such that EFF(X)/∼ is finitely generated, then bounded negativity holds
for X.
Remark I.2.7. It is difficult in general to determine whether EFF(X)/∼ is finitely generated, even for
rational surfaces. Here are some cases where it is known. If X is a rational surface with K
2
X
> 0, then
EFF(X) is finitely generated (see Exercise I.4.5 if −K
X
is nef and K
2
X
> 1; see [Ro] for the case that
K
2
X
> 0 and X is obtained by blowing up at most 8 points of P
2
; or see [LH] or [TVV] for K
2
X
> 0 in
general). We also have EFF(X) finitely generated if X is obtained by blowing up points on a line or conic
in P
2
(see Exercises I.4.2 and I.4.4), or, more generally, if X is rational and −K
X
is big [TVV].
When X be obtained by blowing up r < 9 generic points of P
2
, then K
2
X
> 0 and so EFF(X) is finitely
generated as mentioned above, but in fact −K
X
is ample, which with adjunction implies C
2
≥ −1 for any
prime divisor C. For r = 9, EFF(X) is not finitely generated (see Exercise I.4.7, for example), but it is
still true that C
2
≥ −1 for any prime divisor C, although the proof is somewhat technical. Here now is
a conjecture for a case where EFF(X) is definitely not finitely generated (see Exercise I.4.7) but where
Conjecture I.2.1 is not yet known:
Conjecture I.2.8 ([Ha3]). Let X be obtained by blowing up r > 9 generic points of P
2
. Then C
2
≥ −1 for
every prime divisor C, with equality if and only if C is a smooth rational curve with K
X
· C = −1.
I.3. An Asymptotic Approach to Bounded Negativity. While no general lower bound for C
2
for prime
divisors C on a surface X is known, given a nef divisor F, we can instead ask for a lower bound on C
2
/(F ·C)
2
for all prime divisors C with F · C > 0. As motivation for introducing multipoint Seshadri constants, we
now study this question in the case that X is obtained by blowing up r > 0 points p
1
, . . . , p
r
∈ P
2
, taking
F to be L. Since C is prime and we assume L · C > 0, we see that C = dL −

i
m
i
E
i
for some m
i
≥ 0.
Let
λ
L
(X) = inf
_
C
2
(C · L)
2
: C ∈ EFF(X), C · L > 0, C · E
i
≥ 0 for all i, C ·

i
E
i
> 0
_
(I.3.1)
= inf
_
C
2
(C · L)
2
: C is a prime divisor on X and C · L > 0
_
(The second equality is Exercise I.4.8.) It is clear that the infimum exists: Let C = dL −

i
m
i
E
i
. Since
L −E
i
∈ NEF(X), we see d ≥ m
i
for all i. Hence C
2
/(C · L)
2
= (d
2

i
m
2
i
)/d
2
≥ 1 −r.
Problem I.3.2. Compute λ
L
(X), or at least give good estimates for it.
We now recall a quantity ε introduced by G. V. Chudnovsky [Ch] (for any r points in any projective space)
and Demailly [D] (for a single point, i.e., r = 1, but on any smooth variety) now known as a multipoint
Seshadri constant; see also [Lz]. (Chudnovsky’s version, denoted
´

0
(p
1
, . . . , p
r
), is actually equal to rε.) Let
X be obtained by blowing up distinct points p
1
, . . . , p
r
∈ P
2
. Then
ε(P
2
; p
1
, . . . , p
r
) = inf
_
d

i
m
i
: dL −

i
m
i
E
i
∈ EFF(X), m
i
≥ 0,

i
m
i
> 0
_
As alternative definitions (see Exercise I.4.9) we have:
ε(P
2
; p
1
, . . . , p
r
) = inf
_
C · L

i
C · E
i
: C is prime and

i
C · E
i
> 0
_
(I.3.3)
= sup
_
m
d
: dL −m

i
E
i
∈ NEF(X), m > 0
_
.
GLOBAL ASPECTS OF THE GEOMETRY OF SURFACES 5
Remark I.3.4. In general, ε(P
2
; p
1
, . . . , p
r
) is itself hard to compute. However, by Exercise I.4.10, if
F · C = 0 for some F = dL −m

i
E
i
∈ NEF(X) and C = aL−

i
m
i
E
i
∈ EFF(X) with d > 0 and a > 0,
then ε(P
2
; p
1
, . . . , p
r
) = m/d.
For our asymptotic application of Seshadri constants to bounded negativity, we will use the following
elementary inequality:
Lemma I.3.5. Given integers d > 0 and d ≥ m
i
≥ 0 for all i, we have

i
m
2
i
d
2

i
m
i
d
.
Proof. Just note that d ≥ m
i
≥ 0 implies d

i
m
i

i
m
2
i
; dividing by d
2
gives the result.
Since L−E
i
∈ NEF(X) for each i, if dL−

i
m
i
E
i
∈ EFF(X) then d ≥ m
i
for all i, hence
d
P
i
mi
≥ 1/r.
In particular, ε(P
2
; p
1
, . . . , p
r
) ≥ 1/r > 0 so 1/ε(P
2
; p
1
, . . . , p
r
) makes sense. Applying the lemma now gives:
Corollary I.3.6. Let X be obtained by blowing up distinct points p
1
, . . . , p
r
∈ P
2
. Then
λ
L
(X) ≥ 1 −
1
ε(P
2
; p
1
, . . . , p
r
)
.
Proof.
λ
L
(X) = inf
_
C
2
(C · L)
2
: C ∈ EFF(X), C · L > 0, C · E
i
≥ 0 for all i, C ·

i
E
i
> 0
_
= inf
_
1 −

i
m
2
i
d
2
: C = dL −

i
m
i
E
i
∈ EFF(X), d > 0, m
i
≥ 0,

i
m
i
> 0
_
= inf
_
1 −

i
m
2
i
d
2
: dL −

i
m
i
E
i
∈ EFF(X), m
i
≥ 0,

i
m
i
> 0
_
≥ inf
_
1 −

i
m
i
d
: dL −

i
m
i
E
i
∈ EFF(X), m
i
≥ 0,

i
m
i
> 0
_
= 1 −sup
_

i
m
i
d
: dL −

i
m
i
E
i
∈ EFF(X), m
i
≥ 0,

i
m
i
> 0
_
= 1 −
1
ε(P
2
; p
1
, . . . , p
r
)

Remark I.3.7. Sometimes equality holds, but usually not. See Exercises I.4.11 and I.4.12.
We close this lecture with some remarks about Mori dream spaces. Mori dream spaces give interesting
examples of surfaces with bounded negativity. Let X be obtained by blowing up points p
1
, . . . , p
r
∈ P
2
. Let
L be the total transform of a line and let E
i
be the blow up of the point p
i
. Define the Cox ring Cox(X) of
X to be the ring whose additive structure is given by

(a0,...,ar)∈Z
r+1
H
0
(X, O
X
(a
0
L −a
1
E
1
−· · · −a
e
E
r
)),
and where multiplication is given by the natural maps H
0
(X, O
X
(F
1
)) ⊗H
0
(X, O
X
(F
2
)) → H
0
(X, O
X
(F
1
+
F
2
)). If Cox(X) is finitely generated we say that X is a Mori dream space [HK]. If X is a Mori dream space,
then EFF(X) must be finitely generated, and hence X has bounded negativity by Proposition I.2.5.
Remark I.3.8. If X is obtained by blowing up at most 8 points of P
2
, then X is a Mori dream space.
(Proof: By the Hodge Index Theorem if F is a nontrivial nef divisor, then −K
X
· F > 0, hence the result
follows over the complex numbers from [GM, Corollary 1] by [Ro, Theorem 2] and [Ha2, Theorem III.1].) In
fact, if X is any rational surface with K
2
X
> 0, then X is a Mori dream space. (The same proof applies, but
without the assumption of the complex numbers, using [LH, Proposition 4.3(a)] in place of [Ro]; alternatively,
see [TVV].) If K
2
X
= 0 but −K
X
is not nef, we can again conclude that X is a Mori dream space. (By
[LH, Proposition 4.3(c)], EFF(X) is finitely generated, and by Exercise I.4.6(b) and [Ha2, Theorem III.1],
nef divisors are semi-ample (i.e., have a positive multiple which is effective and base point free). Now apply
6 B. HARBOURNE
[GM, Corollary 1].) In each of these cases, −K
X
is big (see Exercise I.4.13), hence these (in addition to the
examples of Exercises I.4.3 and I.4.4 of blow ups of points on a line or conic) are all subsumed by the result
of [TVV] that a rational surface with big −K
X
is a Mori dream space. However, not all rational surfaces
which are Mori dream spaces have big −K
X
. For example, let C be an irreducible cubic curve, and blow up
the curve r > 9 times, each time at successive infinitely near points of the cubic, starting with a flex point
of the cubic. By Exercise I.4.14, EFF(X) is finitely generated and any nef class F has F · (−K
X
) ≥ 0. By
[Ha6, Theorem 3.1 and Corollary 3.4], every nef class is semi-ample, and by [GM], X is a Mori dream space
since EFF(X) is finitely generated and any nef class is semi-ample, but −K
X
is not big since it is a prime
divisor of negative self-intersection.
Here is a question I do not know the answer to:
Question I.3.9. If X is a rational surface with EFF(X) finitely generated, is X a Mori dream space?
I.4. Exercises.
Exercise I.4.1. Let X be a surface.
(a) Show F ∈ NEF(X) implies F
2
≥ 0.
(b) Assume X is obtained by blowing up r = 21 general points p
i
∈ P
2
. Then Cl(X) has basis
L, E
1
, . . . , E
21
, where L is the pullback of the class of a line and E
i
is the class of the blow up of p
i
.
Let F = 5L −

i
E
i
; show that F ∈ NEF(X) \ EFF(X).
Solution (Exercise I.4.1). (a) Let A be ample, F nef and F
2
< 0. We will show that there are positive
integers s, a and f such that aA+fF is ample and saA+sfF is effective, but such that F · (aA+fF) < 0,
which is impossible if F is nef. To show aA+fF is ample it is enough by the Nakai-Moisezon criterion [Hr]
to show that (aA + fF) · C > 0 for every curve C, and that (aA + fF)
2
> 0. But A · C > 0 since A is
ample and F · C ≥ 0 since F is nef, so (aA+fF) · C > 0. Since A is ample, aA ∈ EFF(X) for a ≫ 0, hence
A· F ≥ 0. Thus, taking t = f/a, we have (aA+fF)· F = a(A+tF)· F < 0 for t >
A·F
−F
2
, but a(A+tF)· F = 0
and (aA + fF)
2
= a
2
(A
2
+ 2tA · F + t
2
F
2
) = a
2
(A
2
+ tA · F) > 0 for t =
A·F
−F
2
, so by choosing a and f
such that t is slightly larger than
A·F
−F
2
we will still have (aA + fF)
2
> 0 (and hence aA + fF is ample so
saA+ sfF is effective for s ≫ 0) while also having (saA+sfF) · F < 0, contradicting F being nef. Hence
we must have F
2
≥ 0.
(b) Consider points p

i
which lie on a smooth quintic. Let Q be the proper transform of that quintic. Then
Q is nef but Q is linearly equivalent to F

= 5L

− E

1
− · · · − E

21
. Now by the semicontinuity principle,
Theorem I.1.6, F = 5L − E
1
− · · · − E
21
∈ NEF(X) when the points p
i
are general. But the points are
general so impose 21 independent conditions on the 21 dimensional space of all quintics (since we can always
choose each successive point not to be a base point of the linear system of quintics through the previous
points); i.e., h
0
(X, O
X
(F)) = 0 hence F ∈ EFF(X).
Exercise I.4.2. Find an explicit finite set of generators for EFF(X) and NEF(X) in case X is obtained by
blowing up r ≥ 1 distinct points on a line in P
2
.
Solution (Exercise I.4.2). This solution is based on [Ha3, Proposition I.5.2]. Let the points be p
1
, . . . , p
r
.
Then Cl(X) has basis L, E
1
, . . . , E
r
, where L is the pullback of the class of a line and E
i
is the class of the
blow up of p
i
. Let Λ = L − E
1
− · · · − E
r
and let L
i
= L −E
i
. Clearly Λ ∈ EFF(X) and E
i
and L
i
are in
EFF(X) for each i. Since L
2
i
= 0 and L
i
is the class of a prime divisor, we see L
i
∈ NEF(X). To prove that
Λ, E
1
, . . . , E
r
generate EFF(X), it is enough to prove that every effective, reduced, irreducible divisor can be
written as a non-negative integer combination of Λ and E
1
, . . . , E
r
. So let C = aΛ+

b
i
E
i
be the class of an
effective, reduced and irreducible divisor. If C is Λ or L
j
, then the claim is true as L
j
= Λ+

i=j
E
i
, so we
may assume that C is not one of these divisors. But then the intersection with them must be non-negative
and hence b
j
= C · L
j
≥ 0. Putting this into 0 ≤ C · Λ = a −

b
i
implies the non-negativity of a. Moreover,
GLOBAL ASPECTS OF THE GEOMETRY OF SURFACES 7
if F is nef then F meets each E
i
and Λ non-negatively, and the argument we just used on C shows that any
such class can be written as F = (a −

i
b
i
)L +

i
b
i
L
i
for non-negative integers a, b
1
, . . . , b
r
, and hence
L, L
1
, . . . , L
r
generate NEF(X).
Exercise I.4.3. Let X be obtained by blowing up points p
1
, . . . , p
r
∈ P
2
. If the points p
i
are collinear,
show that X is a Mori dream space.
Solution (Exercise I.4.3). The generators are given by taking a basis for H
0
(X, O
X
(G)) for each G among
E
1
, . . . , E
r
, Λ and L
1
, . . . , L
r
. This is because if D is an effective divisor, then by the solution to Exercise
I.4.2, D = N + M, where N is fixed and consists of a sum of non-negative multiples of the E
i
and Λ, and
M ∈ NEF(X) and hence M =

i≥0
m
i
L
i
for some non-negative m
i
(where we take L
0
= L). Thus it is
enough to show that

i
H
0
(X, O
X
(L
i
))
⊗mi
→ H
0
(X, O
X
(M)) is surjective. Do this inductively by showing
that H
0
(X, O
X
(F)) ⊗H
0
(X, O
X
(L
i
)) → H
0
(X, O
X
(F +L
i
)) is surjective for each F ∈ NEF(X) and hence
in fact that H
0
(X, O
X
(F
1
)) ⊗H
0
(X, O
X
(F
2
)) → H
0
(X, O
X
(F
1
+F
2
)) is surjective whenever F
1
and F
2
are
nef (see [Ha1, Theorem 2.8]). Alternatively, see [Ot].
Exercise I.4.4. Let X be obtained by blowing up points p
1
, . . . , p
r
on a smooth conic in P
2
with r ≥ 3. (If
r < 3, the points are collinear and the result is given by Exercise I.4.2. Also, the conic does not need to be
smooth here but smoothness simplifies the argument a bit.)
(a) Show EFF(X) is finitely generated.
(b) Cite the literature to show that X is a Mori dream space.
Solution (Exercise I.4.4). (a) Let L
ij
, i = j, be the class of the proper transform of the line through p
i
and
p
j
, let L be the class of the total transform of a line, let D be the class of the proper transform of the conic
and let E
i
be the class of the blow up of p
i
for each i > 0.
Let C be the class of a prime divisor. Note that −K
X
= D + L. Hence if C
2
< 0 but C = D,
then adjunction forces L · C ≤ 1, and hence C is either L
ij
or E
i
for some i and j. If C
2
≥ 0, write
C = a
0
L − a
1
E
1
− · · · − a
r
E
r
. Since C · E
i
≥ 0, we have a
i
≥ 0 for all i. By reindexing we may
assume that a
1
≥ a
2
≥ · · · ≥ a
r
≥ 0. Let D
i
= 2L − E
1
− · · · − E
i
= D + E
i+1
+ · · · + E
r
. Thus
C = (a
0
− a
1
− a
3
)L + (a
1
− a
2
)(L − E
1
) + (a
2
− a
3
)L
12
+ (a
3
− a
2
)D
3
+ · · · + (a
r−1
− a
r
)D
r−1
+ a
r
D
r
,
where a
0
−a
1
−a
3
≥ 0 since C · L
12
≥ 0 implies a
0
≥ a
1
+a
2
≥ a
1
+a
3
. Note that L = L
12
+E
1
+E
2
and
that L−E
1
= L
12
+E
2
. In particular, the class of every prime divisor is a sum of non-negative multiples of
classes of the form L
ij
, E
i
and D.
(b) Since −K
X
= D+L, if F is nef with −K
X
· F = 0, then F · L = 0, hence F = 0 by the Hodge Index
Theorem. The fact that X is a Mori dream space now follows by [GM, Corollary 1] and [Ha2, Theorem
III.1], or directly by [GM, Corollary 3], or by [TVV].
The basic idea of part (a) of the next exercise is taken from [Ro].
Exercise I.4.5. Let X be a rational surface such that −K
X
is nef.
(a) If C is a prime divisor on X such that p
C
> 0, show that C +K
X
∈ EFF(X).
(b) For each integer n, show that there are only finitely many classes C of prime divisors with C
2
≤ n
if K
2
X
> 0.
(c) If E is a class such that E
2
= E · K
X
= −1, show that E ∈ EFF(X).
(d) If K
2
X
> 1, show that −K
X
−E ∈ EFF(X) for any class E such that E
2
= E · K
X
= −1.
(e) Conclude that EFF(X) is finitely generated if K
2
X
> 1.
Solution (Exercise I.4.5). First note that K
2
X
≥ 0 since −K
X
is nef. Next note that h
2
(X, O
X
(−K
X
)) =
h
0
(X, O
X
(2K
X
)) is 0 since X is rational. Hence h
0
(X, O
X
(−K
X
)) ≥ K
2
X
+ 1 > 0 by Riemann-Roch, so
−K
X
∈ EFF(X).
(a) Take cohomology of 0 → O
X
(K
X
) → O
X
(C + K
X
) → O
C
(C + K
X
) → 0. Since X is rational,
h
1
(X, O
X
(K
X
)) = h
1
(X, O
X
) = 0, and h
0
(X, O
X
(K
X
)) = 0 since −K
X
∈ EFF(X) is nontrivial. Thus
0 < p
C
= h
0
(C, O
C
(C +K
X
)) = h
0
(X, O
X
(C +K
X
)).
(b) By adjunction and the fact that −K
X
is nef we have C
2
= 2p
C
− 2 − K
X
· C ≥ −2, so for each n it
is enough to show that there are only finitely many C with C
2
= n. So say C
2
= n, hence 0 ≤ −K
X
· C ≤
C
2
+ 2 = n + 2 by adjunction and the fact that −K
X
is nef. Now let N = K
2
X
C − (K
X
· C)K
X
, so
C = ((−K
X
· C)(−K
X
) +N)/(K
2
X
). Thus to show there are only finitely many such C, it is enough to show
8 B. HARBOURNE
that −K
X
· C is bounded (but we already saw that 0 ≤ −K
X
· C ≤ n + 2) and that there are only finitely
many possibilities for N. To see the latter, note that N · K
X
= 0, so (K
2
X
)
2
C
2
= (K
X
· C)
2
K
2
X
+N
2
, hence
(K
2
X
)(K
2
X
n−(n+2)
2
) ≤ (K
2
X
)
2
C
2
−(K
X
· C)
2
K
2
X
= N
2
. Thus N
2
is bounded below, but N ∈ K

X
and K

X
is negative definite by the Hodge Index Theorem, so intuitively there are only finitely many lattice elements
N in K

X
of length at most
_
−(K
2
X
)(K
2
X
n −(n + 2)
2
). More rigorously, since Cl(X) is free abelian of finite
rank, there are only finitely many elements of Cl(X) orthogonal to K
X
with self-intersection no less than
(K
2
X
)(K
2
X
n −(n + 2)
2
), and hence there are only finitely many possibilities for N.
(c) Since −K
X
is nef but −K
X
· (K
X
− E) < 0, we see K
X
− E ∈ EFF(X). Thus h
2
(X, O
X
(E)) = 0,
hence h
0
(X, O
X
(E)) ≥ 1 by Riemann-Roch, so E ∈ EFF(X).
(d) Since −K
X
· (2K
X
+ E) < 0, we see that 0 = h
0
(X, O
X
(2K
X
+ E)) = h
2
(X, O
X
(−K
X
− E)). Now
h
0
(X, O
X
(−K
X
−E)) ≥ K
2
X
−1 by Riemann-Roch, so −K
X
−E ∈ EFF(X).
(e) Let E be any class such that E
2
= K
X
· E = −1. Then E ∈ EFF(X). Let C be the class of a prime
divisor that is a component of E. Then p
C
= 0, since otherwise C + K
X
and hence E +K
X
is in EFF(X)
by (a). But −K
X
· (E + K
X
) < 0, so this is impossible. Thus E is a sum of prime divisors C with p
C
= 0.
Likewise, −K
X
−E is a sum of classes of prime divisors with p
C
= 0, since otherwise −K
X
−E+K
X
= −E
is in EFF(X). So for some C
i
with p
Ci
= 0 we have −K
X
=

i
C
i
and this sum involves at least two
summands.
By part (b), there are only finitely many classes D of prime divisors with D
2
≤ 0 and p
D
= 0. We will now
see that these classes D, together with the C
i
, generate EFF(X). Given any prime divisor C, it is enough to
show that either C−D ∈ EFF(X) for some such D or that C−C
i
∈ EFF(X) for some i. This is clear by (a)
if p
C
> 0, so assume p
C
= 0. It is again clear if C
2
≤ 0, so assume C
2
> 0. We may assume C · C
1
≤ C · C
i
for all i. Note that h
2
(X, O
X
(C−C
1
)) = h
0
(X, O
X
(K
X
−C+C
1
)) = h
0
(X, O
X
(−

i>1
C
i
−C)) = 0. Thus
h
0
(X, O
X
(C −C
1
)) ≥ (C
2
−2C· C
1
−K
X
· C +(C
2
1
+K
X
· C
1
))/2 +1 = (C
2
−2C · C
1
−K
X
· C −2)/2 +1 =
(C
2
+

i>1
C · C
i
−C · C
1
)/2 ≥ C
2
/2 > 0, so C −C
1
∈ EFF(X).
Exercise I.4.6. Let X be a rational surface with K
2
X
= 0.
(a) Show that −K
X
∈ EFF(X).
(b) Assume in addition that −K
X
is not nef. If F is nef with −K
X
· F = 0, show that F = 0.
Solution (Exercise I.4.6). (a) Apply Riemann-Roch, using h
2
(X, O
X
(−K
X
)) = h
0
(X, O
X
(2K
X
)) = 0.
(b) This follows by the Hodge Index Theorem. Suppose F = 0. Since F is nef, we have F
2
≥ 0 (by
Exercise I.4.1). If F
2
> 0, then F

is negative definite, hence −K
X
· F = 0 and K
2
X
= 0 imply that
−K
X
= 0, but this contradicts the fact that −K
X
= 0. Thus F
2
= 0. Since F
2
= 0 and K
X
· F = 0, we see
for any elements v and w in the span of −K
X
and F in Cl(X) that v · w = 0. But for any ample divisor A we
have i = A· (−mK
X
+F) > 0 for m ≫ 0, since j = −K
X
· A > 0. Let v = j(−mK
X
+F) and let w = −iK
X
.
Then v − w ∈ A

, but (v − w)
2
= 0, so v = w, hence j(−mK
X
+ F) = −iK
X
so jF = (mj − i)K
X
. Thus
mj − i < 0 (since −3(mj − i) = (mj − i)K
X
· L = jF · L ≥ 0 but F · L = 0 implies F = 0) so −K
X
is nef
(being a positive rational multiple of a nef class), contrary to hypothesis.
Exercise I.4.7. Let X be obtained by blowing up 9 points p
1
, . . . , p
9
∈ P
2
on a smooth plane cubic D

.
Let D be the proper transform of D

. Let L be the pullback of the class of a line and let E
i
be the class of
the blow up of p
i
.
(a) Show that N ∈ K

implies N
2
is even.
(b) Let N be any class in K

∩ E

9
. Show that E = N +E
9
+ (N
2
/2)K
X
satisfies E
2
= K
X
· E = −1.
Conclude that E ∈ EFF(X).
(c) If the points are sufficiently general and the ground field is the complex numbers, show that each
such E is the class of a prime divisor. Conclude that EFF(X) is not finitely generated.
Solution (Exercise I.4.7). (a) By Riemann-Roch, N
2
/2 = (N
2
−K
X
· N)/2 is an integer.
(b) That E = N + E
9
+ (N
2
/2)K
X
satisfies E
2
= K
X
· E = −1 is easy. Since −K
X
= D is nef,
and since −K
X
· (K
X
− E) < 0, we see that h
2
(X, O
X
(E)) = 0. Now apply Riemann-Roch to see that
h
0
(X, O
X
(E)) ≥ 1, so E ∈ EFF(X).
(c) Suppose that some E = N +E
9
+ (N
2
/2)K
X
is not the class of a prime divisor. Since −K
X
· E = 1
and −K
X
is nef, if E has two or more components, then one of them must be disjoint from D, hence in
the kernel of the mapping Cl(X) → Cl(D). But the kernel here is the same as the kernel of K

X
→ Cl
0
(D),
where Cl
0
(D) is the subgroup of divisor classes of degree 0, which is a torus which can be identified with
GLOBAL ASPECTS OF THE GEOMETRY OF SURFACES 9
D. Since the complex numbers have infinite dimension over the rationals, it’s easy to choose points p
i
∈ D,
such that the map K

X
→ Cl
0
(D) is injective, hence no prime divisor on X is disjoint from D, so E must be
prime. Conclude by applying Proposition I.2.5.
Exercise I.4.8. Let X be obtained by blowing up r > 0 points p
1
, . . . , p
r
∈ P
2
, and let L and E
i
be as
usual. Show that
inf
_
C
2
(C · L)
2
: C is a prime divisor on X and C · L > 0
_
=inf
_
C
2
(C · L)
2
: C ∈ EFF(X), C · L > 0, C · E
i
≥ 0 for all i, C ·

i
E
i
> 0
_
.
Solution (Exercise I.4.8). Let inf
1
be the first infimum in the statement above and let inf
2
be the second.
Note that inf
1
is equal to inf
3
= inf
_
C
2
(C·L)
2
: C is a prime divisor on X, C· L > 0 and C·

i
E
i
> 0
_
, since
any prime C with C · L > 0 but C ·

i
E
i
= 0 is a positive multiple of L, in which case C

= C −E
1
is the
class of a prime divisor with
C
′2
(C

·L)
2
<
C
2
(C·L)
2
.
Since inf
2
is an infimum over a bigger set than is inf
3
, we see that inf
2
≤ inf
3
= inf
1
. Thus, to see
inf
2
= inf
1
, it is enough to see for any D that is effective with D· L > 0, D· E
i
≥ 0 for all i and D·

i
E
i
> 0
that there is a prime C with C · L > 0 and
C
2
(C·L)
2

D
2
(D·L)
2
.
Suppose D satisfies the given conditions. Write D as dL −

i
m
i
E
i
. We can also write D as

j
C
j
for
some prime divisors C
j
. Let F be obtained by deleting every summand C
j
(if any) for which C
j
= E
i
for
some i. Writing F = dL −

i
m

i
E
i
we see m

i
≥ m
i
for all i, hence
F
2
(F·L)
2

D
2
(D·L)
2
. Thus we may assume
that C
j
= E
i
for all i and j and hence that C
j
· E
i
≥ 0 for all i and j and that C
j
· L ≥ 0 for all j. If for
some j we have C
j
·

i
E
i
= 0, then C
j
is a positive multiple of L, so we can replace C
j
by C
j
− E
1
; the
latter is still the class of a prime divisor, but this change reduces
F
2
(F·L)
2
. Thus with these changes we may
assume each summand C
j
of F satisfies the conditions imposed on D.
By induction it is clearly enough to check that if D
1
and D
2
satisfy the conditions on D, then
min
i
_
D
2
i
(D
i
· L)
2
_

(D
1
+D
2
)
2
((D
1
+D
2
) · L)
2
.
If we write D
1
= aL−

i
a
i
E
i
and D
2
= bL−

i
b
i
E
i
, and assume that the minimum occurs for i = 1, this
is just (a
2

i
a
2
i
)/a
2
≤ ((a +b)
2

i
(a
i
+b
i
)
2
)/(a +b)
2
, or 1 −

i
(a
i
/a)
2
≤ 1 −

i
((a
i
+b
i
)/(a +b))
2
.
I.e., it is enough to show that

i
(a
i
/a)
2

i
((a
i
+ b
i
)/(a + b))
2
if

i
(a
i
/a)
2

i
(b
i
/b)
2
. I.e., given
vectors v and w in Euclidean space with non-negative entries and given positive reals a and b, we must show
(v/a)
2
≥ ((v +w)/(a + b))
2
if (v/a)
2
≥ (w/b)
2
.
But b
2
v
2
≥ a
2
w
2
by hypothesis, so b|v| ≥ a|w|, hence bv
2
= b|v|
2
≥ a|w||v| ≥ av·w, so 2abv
2
≥ 2a
2
v·w and
thus (a+b)
2
v
2
= a
2
v
2
+b
2
v
2
+2abv
2
≥ a
2
v
2
+a
2
w
2
+2a
2
v·w = a
2
(v+w)
2
, whence (v/a)
2
≥ ((v+w)/(a+b))
2
.
Exercise I.4.9. Let X be obtained by blowing up points p
1
, . . . , p
r
∈ P
2
.
(a) Show that ε(P
2
; p
1
, . . . , p
r
) = inf
_
C·L
P
i
C·Ei
: C is prime and C · L > 0
_
.
(b) Show that ε

(P
2
; p
1
, . . . , p
r
) = ε(P
2
; p
1
, . . . , p
r
), where ε

(P
2
; p
1
, . . . , p
r
) = sup
_
m
d
: dL−m

i
E
i

NEF(X), d > 0
_
.
Solution (Exercise I.4.9). (a) This just amounts to the easy fact that
d
1
+d
2
m
1
+m
2
≥ min
_
d
1
m
1
,
d
2
m
2
_
.
By definition
ε(P
2
; p
1
, . . . , p
r
) = inf
_
d

i
m
i
: dL −

i
m
i
E
i
∈ EFF(X), m
i
≥ 0,

i
m
i
> 0
_
.
Suppose D = dL −

i
m
i
E
i
∈ EFF(X) satisfies the conditions of the definition of ε(P
2
; p
1
, . . . , p
r
). Write
D =

i
C
i
as a sum of classes of prime divisors C
i
. Deleting all C
i
of the form E
j
reduces
d
P
i
mi
, so we may
10 B. HARBOURNE
assume C
i
= E
j
for all i and j. If for some i we have C
i
· E
j
= 0 for all j, then deleting that C
i
from the sum
also reduces
d
P
i
mi
. Hence we may assume that D = dL −

i
m
i
E
i
=

i
C
i
, where each C
i
is prime and
satisfies the conditions in the definition of ε(P
2
; p
1
, . . . , p
r
). Write C
i
= d
i
L −

j
m
ij
E
j
. Let µ
i
=

j
m
ij
.
Then it suffices to show that (

i
d
i
)/(

i
µ
i
) ≥ min
i
{d
i

i
}, which follows by repeated application of the
easy fact above.
(b) If aL −b

i
E
i
∈ NEF(X), then ad ≥ b

i
m
i
whenever dL −

i
m
i
E
i
∈ EFF(X). Thus
ε(P
2
; p
1
, . . . , p
r
) ≥ ε

(P
2
; p
1
, . . . , p
r
).
Conversely, for any positive integers a and b such that b/a < ε(P
2
; p
1
, . . . , p
r
), we have (aL − b

i
E
i
) ·
(dL −

i
m
i
E
i
) ≥ 0 for all dL −

i
m
i
E
i
∈ EFF(X), and hence aL − b

i
E
i
∈ NEF(X) so b/a ≤
ε

(P
2
; p
1
, . . . , p
r
). Since we can choose positive integers a and b such that b/a is less than (but arbitrarily
close to) ε(P
2
; p
1
, . . . , p
r
), the result follows.
Exercise I.4.10. Let X be the blow up of P
2
at r distinct points p
1
, . . . , p
r
. Suppose F · C = 0 for some
F = dL − m

i
E
i
∈ NEF(X) and C = aL −

i
m
i
E
i
∈ EFF(X) with d > 0 and a > 0. Show that
ε(P
2
; p
1
, . . . , p
r
) = m/d.
Solution (Exercise I.4.10). Since F ∈ NEF(X), clearly ε(P
2
; p
1
, . . . , p
r
) ≥ m/d. But ε(P
2
; p
1
, . . . , p
r
) >
m/d would imply that F

= d

L −m

i
E
i
∈ NEF(X) for some m

/d

> m/d, but in that case F

· C < 0,
contradicting F

∈ NEF(X).
Exercise I.4.11. Let X be obtained by blowing up collinear points p
1
, . . . , p
r
∈ P
2
. Show that
ε(P
2
; p
1
, . . . , p
r
) = 1/r
and that equality holds in Corollary I.3.6.
Solution (Exercise I.4.11). Since C = L −E
1
−· · · −E
r
is the class of a prime divisor and F = rL −E
1

· · · − E
r
= (r − 1)L + C is a sum of prime divisors each of which F meets non-negatively, we see that F is
nef. Clearly λ
L
(X) ≤ C
2
/(C · L)
2
= 1 −r. But F · C = 0, so ε(P
2
; p
1
, . . . , p
r
) = 1/r by Exercise I.4.10, and
we have 1 −r = 1 −1/ε(P
2
; p
1
, . . . , p
r
).
Remark (on Exercise I.4.11). Exercise I.4.11 shows that equality holds in Corollary I.3.6 when the points
are collinear, but the converse is not true. Here is an example where equality holds but the points are not
collinear. Suppose we consider 9 points on a smooth cubic, three of which are collinear. Let X be obtained
by blowing up the nine points. The proper transform of the cubic is clearly effective and (being prime of
non-negative self-intersection) it also is nef so ε(P
2
; p
1
, . . . , p
r
) = 1/3 by Exercise I.4.10, and the proper
transform of the line through the three collinear points has self-intersection −2, so −2 ≥ λ
L
(X), hence
−2 ≥ λ
L
(X) ≥ 1 −1/ε(P
2
; p
1
, . . . , p
r
) = −2.
There is another way to look at what Exercise I.4.11 tells us, however. The solution to Exercise I.4.11 shows
that λ
L
(X) = 1−r and ε(P
2
; p
1
, . . . , p
r
) = 1/r if the points are collinear. Conversely, if either λ
L
(X) = 1−r
or ε(P
2
; p
1
, . . . , p
r
) = 1/r, then the points p
1
, . . . , p
r
∈ P
2
are collinear. For suppose ε(P
2
; p
1
, . . . , p
r
) = 1/r.
In any case, F = (r −1)L −E
1
−· · · −E
r
= (L −E
1
−E
2
) + (L −E
3
) +· · · + (L −E
r
) is effective and the
classes of the prime components of L−E
1
−E
2
consist of classes E
j
and L−E
1
−E
2
−E
j1
−· · · −E
js
, where
p
1
, p
2
, p
j1
, . . . , p
js
are all of the points which lie on the line through p
1
and p
2
. If the points were not all
collinear, then there would be at most r −1 such points, so F would meet each of its prime components non-
negatively. Thus F would be nef and we would have the contradiction that 1/r = ε(P
2
; p
1
, . . . , p
r
) ≥ 1/(r−1).
Finally, suppose λ
L
(X) = 1−r. Since F = rL−E
1
−· · · −E
r
= (L−E
1
)+(L−E
2
)+(L−E
3
)+· · · +(L−E
r
)
is always nef, we see that ε(P
2
; p
1
, . . . , p
r
) ≥ 1/r always holds. But this means we have 1 − r = λ
L
(X) ≥
1 −1/ε(P
2
; p
1
, . . . , p
r
) ≥ 1 −1/(1/r) = 1 −r, hence ε(P
2
; p
1
, . . . , p
r
) = 1/r which we saw above implies the
points are collinear.
Exercise I.4.12. Find a set of points p
1
, . . . , p
r
∈ P
2
such that the inequality in Corollary I.3.6 is strict.
Solution (Exercise I.4.12). Consider ten points p
1
, . . . , p
10
on a smooth conic. From the solution to Exercise
I.4.4, the only prime divisors C of negative self-intersection come from the points, from the lines through
pairs of points and from the conic itself. The infimum defining λ
L
(X) must come from prime divisors C
of negative self-intersection. By just checking the possibilities we see λ
L
(X) = −6/4 comes from C =
2L −E
1
−· · · −E
10
∈ EFF(X) and we also see F = 5L − E
1
−· · · −E
10
= C + 3L is nef. Since F · C = 0,
by Exercise I.4.10 we see ε(P
2
; p
1
, . . . , p
r
) = 1/5 and hence λ
L
(X) = −6/4 ≥ 1 −1/ε(P
2
; p
1
, . . . , p
r
) = −4.
GLOBAL ASPECTS OF THE GEOMETRY OF SURFACES 11
Exercise I.4.13. Let X be a rational surface such that K
2
X
= 0 but −K
X
∈ NEF(X). Show that −K
X
is
big (i.e., some positive multiple −mK
X
is effective and can be written as −mK
X
= M + N where M and
N are effective and M
2
> 0).
Solution (Exercise I.4.13). By Exercise I.4.6, −K
X
∈ EFF(X). Since −K
X
is not nef, there is a prime
divisor C such that −K
X
· C < 0. Thus −K
X
− C is effective hence so is −mK
X
− C for m ≥ 1, and
(−mK
X
−C)
2
= (2m−1)K
X
· C + (K
X
· C +C
2
) = (2m−1)K
X
· C + 2p
C
−2, so (−mK
X
−C)
2
> 0 for
m ≥ 2. Thus −mK
X
= (−mK
X
−C) +C is big for m ≥ 2.
Exercise I.4.14. Let X = X
r+1
be the rational surface such that X
1
= P
2
, and for each i ≥ 1, X
i+1
→ X
i
is the blow up of p
i
, where p
1
∈ X
1
is a flex of an irreducible plane cubic C, and then for each i ≥ 1, p
i+1
is
the point of the proper transform of C on X
i+1
infinitely near to p
i
. (Thus p
1
, . . . , p
r
are essentially distinct
points.) Assume r ≥ 3. Show that the class of any prime divisor D with D
2
< 0 is either E
r
, or E
i
− E
i+1
for 1 ≤ i < r, or L − E
1
− E
2
− E
3
or D = −K
X
(if r > 9), and show that a divisor class F is nef if
and only if −K
X
· F ≥ 0 and F is a non-negative integer linear combination of L, L − E
1
, 2L − E
1
− E
2
,
3L−E
1
−E
2
−E
3
, . . ., 3L−E
1
−· · · −E
r
= −K
X
. Conclude that EFF(X) is generated by E
r
, E
i
−E
i+1
for 1 ≤ i < r, L −E
1
−E
2
−E
3
and D = −K
X
.
Solution (Exercise I.4.14). It is easy to see that each of the classes listed is the class of a prime divisor D with
D
2
< 0; for example, L−E
1
−E
2
−E
3
is the class of the proper transformof the line tangent to C at p
1
(i.e., the
flex line), while the the class of the proper transform of C to X is 3L−E
1
−· · ·−E
r
= −K
X
, which has negative
self-intersection exactly when r > 9. Suppose D = aL−a
1
E
1
−· · · −a
r
E
r
is the class of a prime divisor with
D
2
< 0 which is not in the given list. Then D meets each of the listed classes non-negatively; i.e., D· E
r
≥ 0
(so a
r
≥ 0), D · (E
i
− E
i+1
) ≥ 0 (so a
i
≥ a
i+1
for each i = 1, . . . , r − 1, hence a
1
≥ a
2
≥ · · · ≥ a
r
≥ 0) and
D·(L−E
1
−E
2
−E
3
) ≥ 0 (so a ≥ a
1
+a
2
+a
3
). It is not hard to see the non-negative integer linear combinations
of L, L−E
1
, 2L−E
1
−E
2
, 3L−E
1
−E
2
−E
3
, . . ., 3L−E
1
−· · ·−E
r
= −K
X
are precisely the classes which meet
E
r
, E
i
−E
i+1
for i > 0 and L−E
1
−E
2
−E
3
non-negatively (see [Ha6]). But each of L, L−E
1
, 2L−E
1
−E
2
,
3L−E
1
−E
2
−E
3
, . . ., 3L−E
1
−· · ·−E
r
= −K
X
is a sum of the listed classes of negative self-intersection (for
example, L = (L−E
1
−E
2
−E
3
)+((E
1
−E
2
)+· · ·+(E
r
−E
r−1
)+E
r
)+((E
2
−E
3
)+· · ·+(E
r
−E
r−1
)+E
r
)+
((E
3
−E
4
) +· · · +(E
r
−E
r−1
) +E
r
) and 3L−E
1
−· · · −E
r−1
= −K
X
+E
r
; moreover, if r = 9, then −K
r
=
3(L−E
1
−E
2
−E
3
)+2(E
1
−E
2
)+4(E
2
−E
3
)+6(E
3
−E
4
)+5(E
4
−E
5
)+4(E
5
−E
6
)+· · ·+(E
8
−E
9
), if r = 8, then
−K
r
= 3(L−E
1
−E
2
−E
3
)+2(E
1
−E
2
)+4(E
2
−E
3
)+6(E
3
−E
4
)+5(E
4
−E
5
)+4(E
5
−E
6
)+· · ·+2(E
7
−E
8
)+E
8
,
etc.). Thus D· (−K
X
) ≥ 0 implies D is a sum of classes of negative self-intersection, each of which it meets
non-negatively, so D
2
≥ 0. Thus our list of classes of prime divisors of negative self-intersection is complete.
Because F meets E
r
, E
i
−E
i+1
for i > 0 and L−E
1
−E
2
−E
3
non-negatively if F is nef, this also shows that
any nef class F is a non-negative integer linear combination of L, L−E
1
, 2L−E
1
−E
2
, 3L−E
1
−E
2
−E
3
,
. . ., 3L −E
1
−· · · −E
r
= −K
X
.
Since each of L, L − E
1
, 2L − E
1
− E
2
, 3L − E
1
− E
2
− E
3
, . . ., 3L − E
1
− · · · − E
r
= −K
X
is a non-
negative integer linear combination of the listed classes of negative self-intersection, we see the latter generate
EFF(X), and we also see that a class F is nef if and only if it is a non-negative integer linear combination
F of L, L −E
1
, 2L −E
1
−E
2
, 3L −E
1
−E
2
−E
3
, . . ., 3L −E
1
−· · · −E
r
= −K
X
with −K
X
· F ≥ 0.
12 B. HARBOURNE
II. Lecture: Abnormality
II.1. Abnormal Curves. One of the difficulties in studying Conjecture I.2.1 and Problem I.3.2 is the
possibility of there being infinitely many prime divisors C with C
2
< 0, possibly (for all anyone knows) even
with C
2
arbitrarily negative. As an intermediate step, it might be worthwhile to define and study a class of
effective divisors C with C
2
< 0 which are so bad as to form a finite set. Doing so turns out to have useful
applications to computing ε(P
2
; p
1
, . . . , p
r
).
Definition II.1.1. Consider a surface X obtained by blowing up a finite set of points p
1
, . . . , p
r
∈ P
2
. Let
C = dL−

i
m
i
E
i
∈ EFF(X) and assume m
i
≥ 0 for all i with m
i
> 0 for some i. Working formally (i.e., in
Cl(X) ⊗
Z
Q), let C = dL − m

i
E
i
, where m = (

i
m
i
)/r. Following Nagata [N1], we say C is abnormal
if C
2
< 0. This is equivalent to d/

i
m
i
< 1/

r, and also to d/(rm) < 1/

r.
We note that not every curve C with C
2
< 0 is abnormal (see Exercise II.3.1); in fact, X has at most
finitely many prime divisors which are abnormal curves (see Exercise II.3.2), but X can have infinitely many
prime C with C
2
< 0 (see Exercise I.4.7).
One application of the concept of abnormality is to computing ε(P
2
; p
1
, . . . , p
r
):
Theorem II.1.2. Let X be a surface obtained by blowing up a finite set of points p
1
, . . . , p
r
∈ P
2
. Then
ε(P
2
; p
1
, . . . , p
r
) < 1/

r if and only if X has an abnormal prime divisor.
Proof. If X has an abnormal prime divisor, then ε(P
2
; p
1
, . . . , p
r
) < 1/

r follows by definition of ε. Con-
versely, assume ε(P
2
; p
1
, . . . , p
r
) < 1/

r. Then there is a class C = dL −

i
m
i
E
i
∈ EFF(X) with
d/

i
m
i
< 1/

r and hence C
2
< 0. Write C =

i
C
i
as a sum of prime divisors C
i
. We may assume no
summand is of the form E
j
, since after removing all such summands we still have an abnormal curve. Thus
every summand C
j
is of the form d
j
L −

i
m
ij
E
i
with m
ij
≥ 0. Now C =

i
C
i
, so C
2
=

ij
C
i
· C
j
< 0
hence C
i
· C
j
< 0 for some i and j. But if neither C
i
nor C
j
were abnormal, then it is easy to see that
C
i
· C
j
≥ 0.
Corollary II.1.3. Let X be a surface obtained by blowing up a finite set of points p
1
, . . . , p
r
∈ P
2
. If
there are no prime divisors on X which are abnormal, then ε(P
2
; p
1
, . . . , p
r
) = 1/

r. If there are abnormal
prime divisors on X, then ε(P
2
; p
1
, . . . , p
r
) = d/

i
m
i
< 1/

r for some abnormal prime divisor C =
dL −

i
m
i
E
i
.
Proof. The first statement follows from Theorem II.1.2. The second follows from the fact that there are only
finitely many abnormal prime divisors and hence the infimum in the definition of ε(P
2
; p
1
, . . . , p
r
) as given
in (I.3.3) is actually a minimum (see Exercises I.4.9(a) and II.3.2). The fact that d/

i
m
i
< 1/

r is just
the definition of abnormality.
The values of ε(P
2
; p
1
, . . . , p
r
) are known when X is obtained by blowing up r generic points of P
2
if either
r ≤ 9 or r is a square. It is an open problem to compute ε(P
2
; p
1
, . . . , p
r
) when r > 9 is not a square. There
is a long-standing conjecture, however, which implies (and in fact is equivalent to) ε(P
2
; p
1
, . . . , p
r
) = 1/

r
for r > 9:
Conjecture II.1.4 (Nagata [N2]). If X is obtained by blowing up r > 9 generic points of P
2
, then X has
no abnormal curves.
Nagata proved this when r is a square [N2]. The conjecture is still open, although it is known in various
special cases. For example, the conjecture is equivalent to:
Conjecture II.1.5. If dL − m(E
1
+ · · · +E
r
) ∈ EFF(X) when X is obtained by blowing up r > 9 generic
points of P
2
, then d > m

r.
By [HR3, Corollary 4.1], this is true when m ≤ t(t −3)/2, where t = ⌊

r⌋. In addition, Dumnicki shows
Conjecture II.1.5 is true when m ≤ 42.
II.2. A Dual Problem. Let X be obtained by blowing up r points of P
2
. Recall that ε(P
2
; p
1
, . . . , p
r
) is
the supremum of 1/t over all t such that tL −

i
E
i
∈ NEF(X). There is a dual notion which Chudnovsky
[Ch] attributes to Waldschmidt [W].
GLOBAL ASPECTS OF THE GEOMETRY OF SURFACES 13
Definition II.2.1.
γ(P
2
; p
1
, . . . , p
r
) = inf
_
d
m
: dL −m

i
E
i
∈ EFF(X)
_
Clearly (as Chudnovsky [Ch] remarks), we have rε(P
2
; p
1
, . . . , p
r
) ≤ γ(P
2
; p
1
, . . . , p
r
). Also, since d/m >

r implies (dL − m

i
E
i
)
2
> 0, we see in that case for D = dL − m

i
E
i
that sD ∈ EFF(X) for s ≫ 0,
hence γ(P
2
; p
1
, . . . , p
r
) ≤ d/m for all d/m >

r; i.e., γ(P
2
; p
1
, . . . , p
r
) ≤

r. Thus we have:
Corollary II.2.2.
rε(P
2
; p
1
, . . . , p
r
) ≤ γ(P
2
; p
1
, . . . , p
r
) ≤

r
As Chudnovsky [Ch] points out, although in general rε(P
2
; p
1
, . . . , p
r
) < γ(P
2
; p
1
, . . . , p
r
) (see Exercise
II.3.3), if the points p
1
, . . . , p
r
are generic we have rε(P
2
; p
1
, . . . , p
r
) = γ(P
2
; p
1
, . . . , p
r
) (Exercise II.3.4).
Thus Nagata’s conjecture (Conjecture II.1.4) is also equivalent to γ(P
2
; p
1
, . . . , p
r
) =

r for r > 9 generic
points of P
2
. Chudnovsky [Ch] also remarks that γ(P
2
; p
1
, . . . , p
r
) is actually a limit:
Proposition II.2.3. Let X be obtained by blowing up r distinct points p
i
∈ P
2
. Then
γ(P
2
; p
1
, . . . , p
r
) = lim
m→∞
d
m
m
where d
m
is the least t such that tL −m

i
E
i
∈ EFF(X). Moreover, for each n ≥ 1, we have
γ(P
2
; p
1
, . . . , p
r
) ≤
d
n
n
.
Proof. Clearly, d
rm
≤ rd
m
, so d
rm
/(rm) ≤ d
m
/m. Therefore, d
m!
/(m!) ≤ d
n
/n for every n|m!. Thus
d
m!
/(m!) is a non-increasing sequence, so lim
m→∞
d
m!
m!
exists; call it l. Given any δ > 0, we check for all
n ≫ 0 that l ≤ d
n
/n ≤ l + δ. Pick m large enough that l ≤
d
m!
m!
≤ l + δ/2. Say n ≥ m! and write
n = a(m!) +c, where c is an integer with 0 ≤ c < m!. Then d
n
≤ d
(a+1)(m!)
≤ (a + 1)d
m!
so
l ≤
d
n!
n!

d
n
n

(a + 1)d
m!
a(m!) + c
=
ad
m!
a(m!) +c
+
d
m!
a(m!) + c

d
m!
m!
+
d
m!
a(m!)
≤ l +δ/2 +d
m!
/(a(m!)),
and for n ≫ 0 we will have a large enough such that d
m!
/(a(m!)) ≤ δ/2. We also now see γ(P
2
; p
1
, . . . , p
r
) ≤
d
n!
n!

dn
n
.
II.3. Exercises.
Exercise II.3.1. Let X be obtained by blowing up r points p
i
∈ P
2
.
(a) Show that C
2
≤ C
2
for any divisor C on X.
(b) If C
1
= a
0
L−

i
a
i
and C
2
= b
0
L−

i
b
i
, where a
1
≥ a
2
≥ · · · ≥ a
r
≥ 0 and b
1
≥ b
2
≥ · · · ≥ b
r
≥ 0,
show that C
1
· C
2
≤ C
1
· C
2
.
(c) If C
1
and C
2
are abnormal, show that C
1
· C
2
< 0.
(d) Give an example of a curve C with C
2
< 0 but such that C is not abnormal.
Solution (Exercise II.3.1). (a) Let C = dL−

r
i=1
m
i
E
i
. Thus we need to show that d
2

i
m
2
i
≤ d
2
−rm
2
;
i.e., that

i
m
2
i
≥ rm
2
. Let v = (m
1
, . . . , m
r
) and let v = (m, . . . , m). Then we need to show, with respect
to the Euclidean dot product, that 0 ≤ v
2
−v
2
, but v
2
= v · v, so 0 ≤ (v −v)
2
= v
2
+v
2
−2v · v = v
2
−v
2
,
as required.
(b) If a =

i
a
i
/r and b =

i
b
i
/r, it suffices to show that

i
a
i
b
i
≥ rab. But rab = a

i
b
i
, so we
need only show

i
a
i
b
i
≥ a

i
b
i
. This is equivalent to showing

i
(ra
i
)b
i
≥ (ra)

i
b
i
, where ra =

i
a
i
;
i.e., we can reduce to the case that a is an integer. If a
1
, · · · , a
r
are not all equal, we can pick some j such
that a
j
> a and some l such that a > a
l
. Let a

j
= a
j
− 1 and a

l
= a
l
+ 1, and a

i
= a
i
for i = j, l. Then

i
(a
i
− a)
2
>

i
(a

i
− a)
2
and

i
a
i
b
i
= (b
j
− b
l
) +

i
a

i
b
i

i
a

i
b
i
. By repeating this procedure we
eventually obtain a sequence a

i
, 1 ≤ i ≤ r, such that

i
(a

i
−a)
2
= 0 and hence a = a

i
for all 1 ≤ i ≤ r and
so

i
a
i
b
i

i
a

i
b
i
= rab.
14 B. HARBOURNE
(c) Let C
1
= aL − b

i
E
i
and let C
2
= cL − d

i
E
i
. Then a/(rb) < 1/

r and c/(rd) < 1/

r, so
ac/(rbd) < 1 so C
1
· C
2
< 0.
(d) An easy example is given by L − E
1
− E
1
− 0E
3
− 0E
4
. For a more interesting example, choose an
irreducible quartic plane curve C

with a triple point. Blow up the triple point and eight additional points
on C

. The proper transform of C

is C = 4L − 3E
1
− E
2
− · · · − E
9
. Then C
2
> 0 but C
2
= −1. More
generally, if you blow up 9 or more general enough points of P
2
, then there are infinitely many exceptional
curves (i.e., the prime divisors E with E
2
= E · K
X
= −1) by Exercise I.4.7(c), but by Exercise II.3.2 at
most finitely many of them are abnormal.
Exercise II.3.2. Let X be obtained by blowing up r points p
i
∈ P
2
. Then there are at most finitely many
prime divisors C which are abnormal. In fact, there are at most r + 1 of them [Sz].
Solution (Exercise II.3.2). Suppose there were an infinite set S of them. We get a mapping φ : S → S
r
by
choosing, for each C ∈ S, a permutation π such that if C = dL−

i
m
i
E
i
, then m
π(1)
≥ m
π(2)
≥ · · · ≥ m
π(r)
.
Thus there must be two prime divisors C
1
= C
2
with φ(C
1
) = φ(C
2
) if S is infinite. Hence by Exercise
II.3.1(b, c) we have C
1
· C
2
< 0, but C
1
= C
2
implies 0 ≤ C
1
· C
2
, which is impossible.
To see that there are at most r + 1, suppose there were more, say C
i
, for 1 ≤ i ≤ t for t > r + 1. Since
Cl(X) has rank r + 1, there is a relation

i
m
i
C
i
= 0 where the C
i
are distinct. Let P =

i,mi>0
m
i
C
i
and let N = −

i,mi<0
m
i
C
i
. Then P − N = 0 hence P = N. Now, P is abnormal, hence P
2
< 0, but
P
2
= P · N ≥ 0 which is a contradiction.
Exercise II.3.3. Give an example such that rε(P
2
; p
1
, . . . , p
r
) < γ(P
2
; p
1
, . . . , p
r
).
Solution (Exercise II.3.3). Consider four points p
1
, . . . , p
4
, exactly three of which (say p
1
, p
2
, p
3
) are
collinear. Let C = L − E
1
− E
2
− E
3
and let L
ij
= L − E
i
− E
j
. Then F = 3L − E
1
− E
2
− E
3
− E
4
=
C+L+(L−E
4
) and H = 3L−E
1
−E
2
−E
3
−2E
4
= C+2(L−E
4
) are nef and C = 5L−3E
1
−3E
2
−3E
3
−3E
4
=
2C + L
14
+L
24
+ L
34
∈ EFF(X). Since H · C = 0, we see γ(P
2
; p
1
, . . . , p
r
) = 5/3, and since F · C = 0, we
see by Exercise I.4.10 that ε(P
2
; p
1
, . . . , p
r
) = 1/3.
Exercise II.3.4. Let X be obtained by blowing up r generic points p
i
∈ P
2
.
(a) Compute ε(P
2
; p
1
, . . . , p
r
) for each r ≤ 9 and each r which is a perfect square.
(b) Show that rε(P
2
; p
1
, . . . , p
r
) = γ(P
2
; p
1
, . . . , p
r
).
Solution (Exercise II.3.4). (a) Suppose r = d
2
is a perfect square. Let X

be obtained by blowing up
r = d
2
points p

i
∈ P
2
on a smooth plane curve C of degree d, hence the class C

= dL

i
E

i
of the
proper transform of C is nef. Let X be obtained by blowing up r = d
2
generic points p
i
∈ P
2
. Since by
Theorem I.1.7 for any divisor tL

i
E

i
∈ NEF(X

) we also have tL −

i
E
i
∈ NEF(X) we see that
ε(P
2
; p

1
, . . . , p

r
) ≤ ε(P
2
; p
1
, . . . , p
r
). But C

∈ NEF(X

) ∩ EFF(X

) together with (C

)
2
= 0 implies that
ε(P
2
; p

1
, . . . , p

r
) = 1/d by Remark I.3.4. On the other hand, ε(P
2
; p
1
, . . . , p
r
) ≤ 1/

r by Corollary II.1.3.
Thus ε(P
2
; p
1
, . . . , p
r
) = 1/

r when r is a perfect square.
For the case of r ≤ 9 generic points, if C = tL−

i
m
i
E
i
is the class of an abnormal prime divisor, then so is
C

= tL−

i
m

i
E
i
, where the m

i
are obtained by a permutation of the m
i
such that m

1
≥ m

2
≥ · · · ≥ m

r
.
Given two abnormal prime divisors, C and D, we thus see that the permuted divisors C

and D

have
C

· D

< 0, and hence C

= D

. I.e., up to permutations, C and D are the same, so if X has any abnormal
curve, that curve gives the value of ε(P
2
; p
1
, . . . , p
r
).
Thus, since C = L − E
1
− E
2
is abnormal for r = 2 or 3, we see ε(P
2
; p
1
, p
2
) = ε(P
2
; p
1
, p
2
, p
3
) = 1/2.
For r = 5 or 6, take C = 2L − E
1
− · · · − E
5
to see ε(P
2
; p
1
, . . . , p
5
) = 2/5. For r = 7, take C =
3L−2E
1
−E
2
−· · ·−E
7
to see that ε(P
2
; p
1
, . . . , p
7
) = 3/8, and for r = 8, take C = 6L−3E
1
−2E
2
−· · ·−2E
8
to
see that ε(P
2
; p
1
, . . . , p
7
) = 6/17. (For the fact that 3L−2E
1
−E
2
−· · · −E
7
and 6L−3E
1
−2E
2
−· · · −2E
8
are classes of prime divisors, use Exercise I.4.7 over the complex numbers. More generally, one can use
quadratic transforms to see that 3L − 2E
1
− E
2
− · · · − E
7
and 6L − 3E
1
− 2E
2
− · · · − 2E
8
are smooth
rational curves.)
(b) Since the points are generic, if C = tL −

i
m
i
E
i
is the class of an effective divisor, then so is
C

= tL −

i
m

i
E
i
, where the m

i
are obtained by any permutation of the m
i
. Thus rC ∈ EFF(X)
for any C = aL − a
1
E
1
− · · · − a
r
E
r
∈ EFF(X). But rC = raL − (a
1
+ · · · + a
r
)(E
1
+ · · · + E
r
) and
ar/(r(a
1
+ · · · + a
r
)) = a/(a
1
+ · · · + a
r
), so rε(P
2
; p
1
, . . . , p
r
) ≥ γ(P
2
; p
1
, . . . , p
r
). This together with
Corollary II.2.2 gives rε(P
2
; p
1
, . . . , p
r
) = γ(P
2
; p
1
, . . . , p
r
).
GLOBAL ASPECTS OF THE GEOMETRY OF SURFACES 15
III. Lecture: Computation of Seshadri Constants
III.1. Estimating Seshadri Constants. Given distinct points p
i
∈ P
2
we now consider the problem of
estimating ε(P
2
; p
1
, . . . , p
r
). Getting an upper bound less than 1/

r is, by Corollary II.1.3, equivalent to
showing the existence of abnormal curves, and this is often quite hard. Thus much of the focus has been on
getting increasingly better lower bounds.
There have been two main methods used for this. Both methods can be adapted to studying Seshadri
constants on surfaces in general. For purposes of exposition we will continue to focus on the case of P
2
.
The first method is to explicitly construct nef divisors. For example, if one shows some divisor F =
dL − m

i
E
i
is nef, then we know m/d ≤ ε(P
2
; p
1
, . . . , p
r
). This is the method used by [Bi], [Ha5] and
[Ha4]. Both authors first construct a nef divisor F

= d

L −

i
m
i
E
i
, and then use an averaging process to
get a nef divisor of the form F = dL −m

i
E
i
.
The second main method is to rule out the possible occurrence of abnormal curves. This method has
been applied by [X], [T], [ST], [SS], [HR1] and [HR2]. Fundamentally it depends on the fact that if F =
tL −m

i
E
i
has F
2
> 0, then, as we show below, there are only finitely many classes C = dL −

i
m
i
E
i
that could possibly be the class of a prime divisor with F · C < 0 [HR1, Lemma 2.1.3]. If one can show that
none of these finitely many classes is the class of a prime divisor, then F is nef and m/t ≤ ε(P
2
; p
1
, . . . , p
r
).
Proposition III.1.1. Let X be obtained by blowing up distinct points p
1
, . . . , p
r
∈ P
2
, with L and E
i
as
usual. Assume that F = tL − m

i
E
i
has F
2
> 0 and t > 0. Then there is an explicitly computable finite
set S
F
of classes which contains the class of every prime divisor C with C · F < 0 (if any).
Proof. Since F
2
> 0 and F · L > 0, we can find an explicit s such that sF ∈ EFF(X) (but the smaller s is
the smaller S
F
will be).
Let E =

i
E
i
, and choose nef divisors H
i
that span Cl(X). For example, H
0
= L, and H
i
= L −E
i
for
i > 0, or choose h
i
> 0 large enough such that H
0
= h
0
L−E and H
i
= h
i
L−E−E
i
are in EFF(X)∩NEF(X).
(It is clear that h
0
= r and h
i
= r +1 will suffice, but the smaller one can choose the h
i
the smaller S
F
will
be. Being able to choose smaller values of the h
i
will depend on having some knowledge of how the points
p
i
are arranged, since if the points are collinear, then h
0
= r and h
i
= r + 1 are best possible.)
If C is the class of a prime divisor with F · C < 0, then sF −C and C are both in EFF(X), hence both
meet every H
i
non-negatively so 0 ≤ C · H
i
≤ sF · H
i
for each i.
Since the classes H
i
generate Cl(X), if for two classes C
1
and C
2
we have C
1
· H
i
= C
2
· H
i
for all i, then
C
1
= C
2
. Thus there are only finitely many possible classes C with 0 ≤ C · H
i
≤ sF · H
i
for all i.
When the r points p
i
are general, one can narrow down the set S
F
even more [Sz].
Lemma III.1.2. Let X be obtained by blowing up general points p
1
, . . . , p
r
∈ P
2
, with L and E
i
as usual.
Assume that C = dL −

i
m
i
E
i
is abnormal; then all but at most one of the coefficients m
i
are equal.
Proof. By Exercise II.3.2 there are at most r + 1 prime divisors C = dL −

i
m
i
E
i
which are abnormal,
but since the points are general any permutation of the m
i
is again an abnormal prime divisor. We may
assume that m
1
≥ · · · ≥ m
r
≥ 0. Suppose that there is an index i such that m
1
> m
i
> m
r
. Then there
are i − 2 permutations ω
j
which are transpositions of m
r
with m
j
, where 1 < j < i. There are r − i − 1
more transpositions α
j
of m
1
with m
j
, where i < j < r. In addition, there are six permutations in which
we permute m
1
, m
i
and m
r
with each other only. This gives (i − 2) + (r − i − 1) + 6 = r + 3 distinct
permutations, contradicting there being at most r +1 abnormal prime divisors. Thus at most two values can
occur among the m
i
. The only other possibility to be ruled out is if the two values each occur at least twice.
So assume that m
1
= · · · = m
j
> m
j+1
= · · · = m
r
, where r ≥ 4 and 2 ≤ j ≤ r −2. The number of distinct
arrangements of the m
i
is
_
r
j
_
. Looking at Pascal’s triangle it is clear that
_
r
j
_
> r + 1. (Since the entries in
the triangle we’re interested in are on the row beginning 1 r · · · , but more than two spots from either end,
we see
_
r
j
_
is the sum of two entries on the row above it, each entry being at least r − 1, so
_
r
j
_
≥ 2r − 2,
hence 2r −2 > r + 1, since r ≥ 4.)
The restrictions on possible abnormal prime divisors can be made even more stringent; see [HR1] and
[HR2].
Example III.1.3. Suppose we blow up six general points p
1
, . . . , p
6
. We will use the method of ruling
out abnormal curves to check that F = 5L − 2

i
E
i
is nef, and hence that ε(P
2
; p
1
, . . . , p
6
) ≥ 2/5. Since
C = 2L −E
1
−· · · −E
5
∈ EFF(X) has F · C = 0, this shows ε(P
2
; p
1
, . . . , p
6
) = 2/5 by Exercise I.4.10.
16 B. HARBOURNE
First, note that H
0
= 3L −

i
E
i
= (2L −E
1
−· · · −E
4
) + (L −E
5
−E
6
) is nef, since each summand is
the class of a prime divisor which H
0
meets non-negatively. Also, H
i
= H
0
− E
i
∈ NEF(X). For example,
H
5
= (2L − E
1
− · · · −E
5
) + (L −E
5
−E
6
), but H
5
meets each summand non-negatively, each of which is
the class of a prime divisor.
Suppose C is the class of a prime divisor such that 0 > C · F. Then C is abnormal and by Lemma III.1.2
we may (after reindexing, if need be) assume that C = dL−m

i
E
i
−kE
1
for some k. First suppose k = 0.
Then we have 0 ≤ C · H
0
= 3d −6m ≤ F · H
0
= 3, 0 ≤ C · H
1
= 3d −7m ≤ F · H
1
= 1.
Thus 7m ≤ 3d ≤ 6m + 3, so m ≤ 3. For m = 1 we get d = 3 (which fails 5d − 12m = F · C < 0), for
m = 2 we get d = 5 (which also fails 5d − 12m < 0), and for m = 3 we get d = 7. But since C is supposed
to be a prime divisor it should satisfy adjunction and thus must have −2 ≤ C
2
+C · K
X
, but for d = 7 with
m = 3 we find C
2
+C · K
X
= −8.
So suppose C = dL − m

i
E
i
− kE
1
for some k > 0 so 5d − 12m − 2k = C · F < 0. We have
0 ≤ C · H
0
= 3d − 6m − k ≤ F · H
0
= 3 and 0 ≤ C · H
1
= 3d − 7m − 2k ≤ F · H
1
= 1. Thus
7m + 2k ≤ 3d ≤ 6m + k + 3, so m + k ≤ 3. Thus (d, m, k) is either (1, 0, 1), (2, 0, 3) or (3, 1, 1), giving
C = L−E
1
(which fails F ·C < 0), C = 2L−3E
1
(which is not in EFF(X)), and C = 3L−2E
1
−E
2
−· · · −E
6
(which also fails F · C < 0).
Finally, assume C = dL−m

i
E
i
−kE
1
for some k < 0, so 5d−12m−2k = C· F < 0. Since C should be
the class of a prime divisor with C· L > 0, we have C· E
i
≥ 0 for all i, hence −k ≤ m. We have 0 ≤ C· H
0
=
3d−6m−k ≤ F · H
0
= 3 and 0 ≤ C· H
1
= 3d−7m−2k ≤ F · H
1
= 1. Thus 7m+2k ≤ 3d ≤ 6m+k +3 and
hence also 5m ≤ 7m+2k ≤ 3d. Since F = (2 −E
1
−· · · −E
5
) +(2 −E
2
−· · · −E
6
) +(L−E
1
−E
6
), we see
F ∈ EFF(X), hence F −C should also be effective, so d ≤ 5, whence 5m ≤ 3d ≤ 15 implies 1 ≤ −k ≤ m ≤ 3.
The simultaneous solutions to 7m + 2k ≤ 3d ≤ 6m + k + 3, 1 ≤ −k ≤ m ≤ 3 and 5d − 12m− 2k < 0 are
(d, m, k) ∈ {(5, 3, −3), (6, 3, −2), (4, 2, −1)}. None of these are effective. For example, E = 2L−E
2
−· · · −E
6
is a prime divisor, but C = 5L−3(E
2
+· · · +E
6
) for (d, m, k) = (5, 3, −3); since E· C < 0, C−E is effective
if C is, and likewise so are C − 2E and C −3E, but C −3E = −L is not effective, hence neither is C. The
same argument handles the other two cases.
Thus F is nef, as claimed.
We now give an example of the alternative approach using the method of [Ha5] and [Ha4], based on the
idea of unloading [R].
Proposition III.1.4. Let d, r, n be positive integers such that r < d

n and r ≤ n. Then for n general
points p
i
, we have
ε(P
2
; p
1
, . . . , p
n
) ≥
r
nd
.
Proof. It is enough to show that ndL−r(E
1
+· · · +E
n
) ∈ NEF(X), where X is the blow up of P
2
at general
points p
1
, . . . , p
n
. By Theorem I.1.6(b), it is enough to find essentially distinct points p

i
of X
1
= P
2
such
that ndL

− r(E

1
+ · · · + E

n
) ∈ NEF(X

n+1
). Choose any smooth plane curve C
1
of degree d. Let p

1
∈ C
1
.
Recursively, let X

i+1
be the blow up of X

i
at p

i
, let C
i+1
be the proper transform of C
i
, and let p

i+1
be the
point of C
i+1
infinitely near to p

i
. This defines p

1
, . . . , p

r
. If n > r, for r < i ≤ n, choose p

i
to be infinitely
near to p

i−1
but choose p

r+1
not to be on C
r+1
.
Thus dL

−E

1
−· · · −E

r
is the class of C
r+1
, i.e., the proper transform of C
1
, hence the class of a prime
divisor, as are E

i
− E

i+1
for each 1 ≤ i < n and E

n
. In particular E

i
− E

j
∈ EFF(X

n+1
) for every j > i,
and hence so is E

1
+· · · +E

r
−rE

i
for every i > r. Since ndL

−n(E

1
+· · · +E

r
) ∈ EFF(X

n+1
), we see
F = ndL

−r(E

1
+· · · +E

n
) = (ndL

−n(E

1
+· · · +E

r
)) +

r<i≤n
(E

1
+· · · +E

r
−rE

i
) ∈ EFF(X

n+1
)
(this is the unloading step). The irreducible components of this sum are C
r+1
and E

i
− E

i+1
for various i,
but F meets each one non-negatively (this is clear for E

i
−E

i+1
, and F · C
r+1
≥ 0 since nd
2
−r
2
> 0). So
F ∈ NEF(X

n+1
), as required.
As another variation we have:
Proposition III.1.5. Let d, r, n be positive integers such that n ≥ r > d

n. Then for n general points
p
i
∈ P
2
, we have
ε(P
2
; p
1
, . . . , p
n
) ≥
d
r
.
GLOBAL ASPECTS OF THE GEOMETRY OF SURFACES 17
Proof. See Exercise III.2.3.
Example III.1.6. Again suppose we blow up six general points p
1
, . . . , p
6
; let X be the surface we obtain.
Then 5 > 2

6, so by Proposition III.1.5, we see that ε(P
2
; p
1
, . . . , p
6
) ≥
2
5
and hence that F = 5L−2

i
E
i

NEF(X). Since C = 2L − E
1
− · · · − E
5
∈ EFF(X) and F · C = 0, we see by Remark I.3.4 that in fact
ε(P
2
; p
1
, . . . , p
6
) =
2
5
.
III.2. Exercises.
Exercise III.2.1. Compute ε(P
2
; p
1
, . . . , p
r
) and γ(P
2
; p
1
, . . . , p
r
) for every choice of r < 9 distinct points
of P
2
.
Solution (Exercise III.2.1). This can be done using the various possibilities (worked out in [GHM]) for
EFF(X) where X is the blow up of P
2
at the r points. Some of these cases are discussed in [Ch].
Exercise III.2.2. Let X be the blow up of 12 general points. Study whether F = 7L −2(E
1
+· · · + E
12
)
is nef, using the method of Example III.1.3.
Solution (Exercise III.2.2). It is nef, using Proposition III.1.5 with r = 7 and d = 2. However, using the
method of Example III.1.3 one is left with showing that in none of the following cases is C = dL − (m +
k)E
1
−m(E
2
+· · · +E
12
) an abnormal prime divisor:
d= 7 m= 2 k= 1
d= 10 m= 3 k= 0
d= 3 m= 1 k= -1
This is clear for the last case, since 3L − E
1
− · · · − E
11
is not effective. The other two cases are harder
to eliminate, but it is known that except for a few exceptional cases which do not occur here that general
points of small multiplicity impose independent conditions on curves of degree d, if there are curves of degree
d passing through the points with the specified multiplicities. (How big “small” is keeps increasing as more
research is done, but certainly multiplicity at most 3 is covered by the results; see [Du].)
Exercise III.2.3. Let d, r, n be positive integers such that n ≥ r > d

n. Then for n general points p
i
∈ P
2
,
we have
ε(P
2
; p
1
, . . . , p
n
) ≥
d
r
.
Solution (Exercise III.2.3). Mimic the proof of Proposition III.1.4. It is enough by the semicontinuity
principle to find essentially distinct points p

i
of X
1
= P
2
such that rdL

−d
2
(E

1
+· · · +E

n
) ∈ NEF(X

n+1
).
Choose any smooth plane curve C
1
of degree d. Let p

1
∈ C
1
. Recursively, let X

i+1
be the blow up of X

i
at p

i
, let C
i+1
be the proper transform of C
i
, and let p

i+1
be the point of C
i+1
infinitely near to p

i
. This
defines p

1
, . . . , p

r
. If n > r, for r < i ≤ n, choose p

i
to be infinitely near to p

i−1
but choose p

r+1
not to be
on C
r+1
.
Thus dL

−E

1
−· · · −E

r
is the class of C
r+1
, i.e., the proper transform of C
1
, hence the class of a prime
divisor, as are E

i
−E

i+1
for each 1 ≤ i < n and E

n
. In particular E

i
−E

j
∈ EFF(X) for every j > i. Since
rdL

−r(E

1
+· · · +E

r
) ∈ EFF(X

n+1
) and since r
2
> nd
2
, by adding to rdL

−r(E

1
+· · · +E

r
) the classes
m
n
E

n
and m
ij
(E

i
−E

j
) with i ≤ r and j > r for appropriate choices of m
ij
≥ 0 (this is the unloading step),
we obtain F = rdL

−d
2
(E

1
+· · · +E

n
) with F
2
> 0. But F · C
r+1
= 0, F · (E

i
−E

j
) = 0 and F · E

r
> 0,
so F ∈ NEF(X

n+1
), as required.
Exercise III.2.4. Show F = 5L − E
1
− · · · − E
21
is ample, when the E
i
are obtained by blowing up 21
general points p
i
of P
2
.
Solution (Exercise III.2.4). By Proposition III.1.4, using r = 9 and d = 2, we see that
ε(P
2
; p
1
, . . . , p
21
) ≥
9
42
.
Thus D = 42L − 9(E
1
+ · · · + E
21
) ∈ NEF(X), so clearly 45L − 9(E
1
+ · · · + E
21
) = 9F (and even
43L−9(E
1
+· · · +E
21
), for that matter) is ample by the Nakai-Moiseson criterion [Hr] since F
2
> 0 and F
meets every curve positively (any prime divisor orthogonal to D must meet D+L positively, since the only
prime divisors orthogonal to L are the E
i
, which meet D positively).
18 B. HARBOURNE
IV. Lecture: The Containment Problem (an application to Commutative Algebra)
IV.1. Background. The notions we’ve discussed above can be applied to questions of commutative algebra,
especially problems involving ideals of fat points. Let p
1
, . . . , p
s
∈ P
n
be distinct points. Let R = k[P
n
] =
k[x
0
, . . . , x
n
] be the homogeneous coordinate ring of P
n
. Let I(p
i
) ⊂ R be the ideal generated by all forms
vanishing at p
i
. Given a 0-cycle Z = m
1
p
1
+ · · · + m
s
p
s
(i.e., an element in the free abelian group on the
points p
i
) with m
i
≥ 0 for all i, let I(Z) be the homogeneous ideal ∩
i
I(p
i
)
mi
. This is a saturated ideal which
defines a 0-dimensional subscheme of P
n
. We will abuse notation and use the 0-cycle Z = m
1
p
1
+· · · +m
s
p
s
to denote this subscheme, which we refer to as a fat point subscheme. We will denote the sheaf of ideals
corresponding to I(Z) by I
Z
, hence I(Z) = ⊕
t≥0
H
0
(P
n
, I
Z
(t)), where I
Z
(t) = I
Z

O
P
n
O
P
n(t). In fact,
more is true:
Proposition IV.1.1. Given distinct points p
i
∈ P
n
and integers m
i
. Let Z be the fat point scheme

mi≥0
m
i
p
i
, let π : X → P
n
be the morphism obtained by blowing up the points p
i
, let H be the pullback
to X of a general hyperplane and let E
i
be the blow up of p
i
. Then there is a natural isomorphism I
Z
(t)

=
π

(O
X
(tH−

i
m
i
E
i
)) such that H
0
(P
n
, I
Z
(t))

= H
0
(X, O
X
(tH−

i
a
i
E
i
)) and so I(Z) can be identified
with ⊕
t≥0
H
0
(X, O
X
(tH −

i
m
i
E
i
)). Moreover, if m
i
≥ 0 for all i, then H
q
(P
n
, I
Z
(t))

= H
q
(X, O
X
(tH−

i
a
i
E
i
)) holds for all q ≥ 0.
Proof. First, I
Z
= Π
mi≥0
I
mi
pi
. If m ≥ 0 and π is the blow up of a single point p ∈ P
n
where we set E =
π
−1
(p), then we have a natural morphism I
m
p
→ π


−1
I
m
p
) which induces a morphism I
m
p
→ π


−1
I
m
p
·
O
X
) = π

O
X
(−mE) and thus
I
Z
= Π
mi≥0
I
mi
pi
→ Π
mi≥0
π

O
X
(−m
i
E) = π

O
X
(

mi≥0
−m
i
E) ֒→ π

O
X
(

i
−m
i
E).
By the projection formula ([Hr, Exercise II.5.1(d)]), we have a natural isomorphism
π

(O
X
(tH −

i
m
i
E
i
))

= O
P
n(t) ⊗π

(O
X
(−

i
m
i
E
i
)),
so I
Z
(t)

= π

(O
X
(tH −

i
m
i
E
i
)) follows if we show that I
Z

= π

(O
X
(−

i
m
i
E
i
)).
This is trivial if n = 1, since then blowing up has no effect. So assume n > 1. For convenience we write
L for O
X
(−

i
m
i
E
i
), notationally suppressing its dependence on the m
i
.
We start by noting that π

O
X
= O
P
n. (See the argument of [Hr, Corollary III.11.4]: since π is projective
by [Hr, Proposition II.7.16(c)], π

O
X
is coherent. Thus π

O
X
is locally a sheaf of finitely generated O
P
n-
modules. Since π is birational, on any affine open of P
n
, the ring B given by π

O
X
and the ring A given
by O
P
n both have the same function field, with A being integrally closed since P
n
is smooth, hence normal
and B being module finite over A since π is projective and O
X
and hence π

O
X
are coherent [Hr, Corollary
II.5.20]; i.e., B is an integral extension of the integrally closed ring A, with the same function field, so A = B
and thus π

O
X
= O
P
n.)
Now we show that π

O
X
(−m
i
E
i
) is either O
P
n (if m
i
≤ 0) or I
mi
pi
(if m
i
> 0). If m
i
≤ 0, then we have
a morphism O
X
→ O
X
(−m
i
E
i
), hence O
P
2 = π

O
X
→ π

O
X
(−m
i
E
i
). This is clearly an isomorphism
except possibly at the point p
i
. Let p
i
∈ U be an affine open neighborhood. Consider the commutative
diagram
O
P
n(U)

=
→ O
X

−1
(U)) ֒→ O
X
(−m
i
E
i
)(π
−1
(U))
↓ ↓
O
P
n(U \ {p
i
}) → O
X
(−m
i
E
i
)(π
−1
(U) \ E
i
)
The left vertical arrow is an equality by [Hr, Exercise I.3.20] (see also [Hr, Proposition II.6.3A]) or by [Hr,
Exercise III.3.5] and the right vertical arrow is injective since X is integral. The bottom arrow is also an
isomorphism (since U \ {p
i
}

= π
−1
(U) \ {π
−1
(p
i
)} = π
−1
(U) \ E
i
), hence the other arrows are isomorphisms
too, whence π

O
X
(−m
i
E
i
)

= O
P
n.
If m
i
> 0, consider the canonical morphism I
mi
pi
→ π


−1
I
mi
pi
· O
X
) = π

O
X
(−m
i
E
i
). Now, π


−1
I
mi
pi
)
is the sheaf associated to the presheaf U → I
mi
pi
(U), hence π


−1
I
mi
pi
· O
X
) is the sheaf associated to the
presheaf U → I
mi
pi
(U) · O
X

−1
(U)) = I
mi
pi
(U) · O
P
2 (U) = I
mi
pi
(U). I.e, the canonical sheaf morphism
I
mi
pi
→ π

O
X
(−m
i
E
i
) comes from an isomorphism of presheaves, hence is an isomorphism itself.
GLOBAL ASPECTS OF THE GEOMETRY OF SURFACES 19
Thus I
Z
→ π

L is locally an isomorphism hence it is an isomorphism, so
H
0
(P
n
, I
Z
(t))

= H
0
(P
n
, π

L(t)) = H
0
(X, O
X
(tH −

i
m
i
E
i
)).
Now assume that m
i
≥ 0 for all i and let L denote O
X
(tH −

i
m
i
E
i
) for an arbitrary integer t. We
conclude by applying [Hr, Exercise III.8.1], showing that H
l
(X, L) = H
l
(P
n
, π

L) for all l > 0. For this we
need to show that R
l
π

L = 0 for all l > 0, and to do this it is enough to check that the stalks vanish. This is
clear at points away from each point p
i
since π is an isomorphism then. Thus R
l
π

L has support at most at
the points p
i
, hence at p
i
it is equal to the inverse limit of H
l
(jE
i
, O
jEi
(m
i
)) over j by [Hr, Theorem III.11.1]
(as in the proof of [Hr, Corollary V.3.4]), where O
jEi
(m
i
) denotes O
jEi

OX
O
X
(−m
i
E
i
). Thus it suffices
to show that H
l
(jE
i
, O
jEi
(m
i
)) = 0. Look at the exact sequence 0 → O
X
(−jE
i
) → O
X
→ O
jEi
→ 0 and
tensor through by O
X
((m
i
+j)H −m
i
E
i
) to get 0 → O
X
((m
i
+j)(H −E
i
)) → O
X
((m
i
+j)H −m
i
E
i
) →
O
jEi
(m
i
) → 0. The result will follow by showing that h
l
(X, O
X
(aH −bE
i
)) = 0 for all l > 0 if a ≥ b.
Let Y be a prime divisor whose class is H − E
i
if b > 0 or just H if b = 0. Consider 0 → O
X
(−Y ) →
O
X
→ O
Y
→ 0 and tensor through by O
X
(aH − bH
i
) to get 0 → O
X
(a

H − b

E
i
) → O
X
(aH − bE
i
) →
O
Y
(aH

− bE

i
) → 0, where H

= H ∩ Y , E

i
= E
i
∩ Y , a

= a − 1 and b

is the maximum of b − 1 and
0. Taking cohomology of this exact sequence shows that h
l
= 0 for the ends for all l > 0 then h
l
= 0 for
the middle term for all l > 0. Since O
Y
(aH

− bE

i
) is of the same form as what we wish to prove, but in
dimension one less, and since the result is true in dimension 1 (i.e., when Y = P
1
), we may assume the
rightmost term has h
l
= 0 for all l > 0 by induction. Showing the same for the leftmost term eventually
reduces to showing h
l
(X, O
X
) = 0 for all l > 0. For this mimic the argument of [Hr, Proposition V.3.4].
IV.2. Symbolic Powers. Let P be a prime ideal in a polynomial ring R = k[x
0
, . . . , x
n
] over an algebraically
closed field k. By the Nullstellensatz, we know that P = ∩
P⊆M maximal
M. The symbolic power P
(m)
of P
can be defined as P
(m)
= ∩
P
m
⊆M maximal
M
m
(see [Es, Theorem 3.14]). This generalizes nicely to the case
of an ideal I(Z) = ∩I(p
i
) of points p
1
, . . . , p
s
∈ P
n
, where we define the m-th symbolic power I(Z)
(m)
to be
I(Z)
(m)
= ∩
i
(I(p
i
)
m
).
I.e., I(Z)
(m)
= I(mZ), where mZ is the fat point scheme mp
1
+ · · · + mp
s
. (This is consistent with the
definition of symbolic powers used in [HH], in terms of primary decompositions.)
We will for simplicity focus here on the case of symbolic powers of ideals of points in projective space.
See [PSC] for greater generality.
IV.3. The Containment Problem. Let Z = p
1
+· · · +p
s
⊂ P
n
and let I = I(Z). Clearly, I
m
⊆ I
(m)
. In
fact, we have:
Lemma IV.3.1. Let Z = p
1
+· · · +p
s
⊂ P
n
and let I = I(Z) ⊆ R = k[P
n
]. Then I
r
⊆ I
(m)
if and only if
r ≥ m.
Proof. See Exercise IV.6.1.
Understanding the reverse containment is a much harder largely open problem:
Problem IV.3.2. Let Z = p
1
+· · · +p
s
⊂ P
n
and let I = I(Z) ⊆ R = k[P
n
]. Determine all r and m such
that I
(m)
⊆ I
r
.
Since I
(m)
⊆ I
r
implies I
m
⊆ I
(m)
⊆ I
r
⊆ I
(r)
, by Lemma IV.3.1 we see m ≥ r. Also, I
(1)
= I
1
, and
clearly, m

≥ m implies I
(m

)
⊆ I
(m)
, so we can restate Problem IV.3.2 as:
Problem IV.3.3. Let Z = p
1
+· · · + p
s
⊂ P
n
and let I = I(Z) ⊆ R = k[P
n
]. Given r ≥ 2, determine the
least m ≥ r such that I
(m)
⊆ I
r
.
As an asymptotic first step, this suggests the following definition and problem:
Definition IV.3.4 ([BH]). Let Z = p
1
+ · · · + p
s
⊂ P
n
and let I = I(Z) ⊆ R = k[P
n
]. Then define the
resurgence of I to be
ρ(I) = sup
_
m
r
: I
(m)
⊆ I
r
_
.
Problem IV.3.5 ([BH]). Let Z = p
1
+ · · · +p
s
⊂ P
n
and let I = I(Z) ⊆ R = k[P
n
]. Compute or at least
give bounds on ρ(I).
20 B. HARBOURNE
It is not clear a priori that ρ(I) is even finite. Results of Swanson [Sw] showed in many cases that it is
and inspired the results of [ELS] and [HH]. We state a simplified version of the result of [HH]:
Theorem IV.3.6. Let Z = p
1
+ · · · + p
s
⊂ P
n
and let I = I(Z) ⊂ R = k[P
n
]. Then I
(nr)
⊆ I
r
for each
r ≥ 1, hence ρ(I) ≤ n.
Both for [ELS] and for [HH], the proof essentially involves finding an ideal J such that one can check both
that I
(nr)
⊆ J and that J ⊆ I
r
; [ELS] uses asymptotic multiplier ideals for J (see [Te] for an exposition of
this approach), while the proof of [HH] uses Frobenius powers for J (with a dash of tight closure to get the
general result). Example IV.3.10 exhibits the role of Frobenius powers; it is actually a special case of the
Hochster-Huneke proof of Theorem IV.3.6. For the example we will need some results on Frobenius powers:
Definition IV.3.7. Let I ⊆ R = k[P
n
] be an ideal. Assume char(k) = p > 0 and let q be a power of p.
Define the q-th Frobenius power I
[q]
of I to be the ideal generated by I
q
.
Proposition IV.3.8. Let I, J ⊆ R = k[P
n
] be ideals, where char(k) = p > 0 and q is a power of p. Then
(I ∩ J)
[q]
= I
[q]
∩ J
[q]
.
Proof. See [HS, Lemma 13.1.3] or [PSC, Example 8.4.4].
To apply Lemma IV.3.8, we will also want to note:
Lemma IV.3.9. Let I ⊆ R = k[P
n
] be an ideal generated by s elements, and assume char(k) = p > 0 and
q is a power of p. Then I
sq
⊆ I
[q]
.
Proof. See Exercise IV.6.2.
Example IV.3.10. Consider Z = p
1
+· · ·+p
s
⊂ P
n
and let I = I(Z) ⊆ R = k[P
n
]. Assume char(k) = p > 0
and that q is a power of p. Then I(p
i
)
qn
⊆ I(p
i
)
[q]
by Lemma IV.3.9 since the ideal of a point in P
n
is
generated by n linear forms, so I
(qn)
= ∩
i
I(p
i
)
qn
⊆ ∩
i
I(p
i
)
[q]
⊆ (∩
i
I(p
i
))
[q]
⊆ I
q
by Proposition IV.3.8 and
the obvious fact that I
[q]
⊆ I
q
.
IV.4. Estimating the Resurgence. In this section we show how to use γ from Definition II.2.1 and the
regularity of an ideal to give bounds on ρ(I). First we show how to interpret γ in this context. Given points
p
1
, . . . , p
s
∈ P
n
, let I = I(Z) for Z = p
1
+· · · +p
s
. Define γ(I) to be the infimum of d
m
/m where d
m
is the
least degree t such that I
(m)
contains a nonzero form of degree t. (By Proposition IV.1.1, this is consistent
with Definition II.2.1; i.e., γ(I) = γ(P
2
; p
1
, . . . , p
s
).) As in Proposition II.2.3, this is actually a limit which
is decreasing on multiplicative subsequences; i.e., d
ms
/(ms) ≤ d
m
/m for all s > 0. More generally, given a
homogeneous ideal 0 = J ⊆ k[P
n
], we will denote the least degree t such that J contains a nonzero form of
degree t by α(J). Thus α(J) is the degree in which the ideal starts (hence the use of the first letter, α, of
the Greek alphabet to denote this concept). One can also regard α(J) as the M-order of J, where M is the
ideal generated by the variables (i.e., α(J) is the greatest power of M containing J).
As noted by the remark after Exercise I.4.11, ε(P
2
; p
1
, . . . , p
s
) ≥ 1/s, hence by Corollary II.2.2 we have
γ(I) ≥ 1. By a similar argument, this remains true for P
n
. In particular, γ(I) > 0, so it makes sense to
divide by γ(I).
Given a homogeneous ideal J ⊆ R = k[P
n
], for any t ≥ 0 let J
t
be the k-vector space span of the forms of
degree t in k (called the homogeneous component of J of degree t). Note that R/J is also graded; we define
(R/J)
t
to be R
t
/J
t
. We recall that the regularity reg(I) of I is the least degree t ≥ 0 such that (R/I)
t
and
(R/I)
t−1
have the same vector space dimension. We have the following theorem:
Theorem IV.4.1 ([BH]). Let Z = p
1
+· · · +p
s
⊂ P
n
, let I = I(Z) and let r and m be positive integers.
(a) If α(I
(m)
) < rα(I), then I
(m)
⊆ I
r
.
(b) If rreg(I) ≤ α(I
(m)
), then I
(m)
⊆ I
r
.
(c)
α(I)
γ(I)
≤ ρ(I) ≤
reg(I)
γ(I)
(d) If α(I) = reg(I), then I
(m)
⊆ I
r
if and only if α(I
(m)
) ≥ rα(I).
GLOBAL ASPECTS OF THE GEOMETRY OF SURFACES 21
Proof. (a) This is clear, since α(I
r
) = rα(I) and so in this case I
(m)
has a nonzero element of degree less
than any nonzero element of I
r
.
(b) First we check that rreg(I) ≤ α(I
(m)
) implies that r ≤ m. Since I
m
⊆ I
(m)
, we see that α(I
(m)
) ≤
α(I
m
) = mα(I). But α(I) ≤ reg(I) since for all 0 ≤ t < α(I) we have dim
k
(R/I)
t
> dim
k
(R/I)
t−1
. Thus
rreg(I) ≤ α(I
(m)
) ≤ mα(I) ≤ mreg(I). But for any nonzero ideal I properly contained in (x
0
, . . . , x
n
) we
have α(I) > 0, so reg(I) > 0 and we see m ≥ r and hence I
(m)
⊆ I
(r)
.
Now we use the facts that reg(I
r
) ≤ r reg(I) and I
r
t
= I
(r)
t
for all t ≥ reg(I
r
) [GGP]; see also [AV]. Thus
for t < rreg(I) ≤ α(I
(m)
) we have 0 = I
(m)
t
⊆ I
r
t
, while for t ≥ rreg(I) ≥ reg(I
r
), we have I
(m)
t
⊆ I
(r)
t
= I
r
t
,
so I
(m)
t
⊆ I
r
t
holds for all t and we have I
(m)
⊆ I
r
.
(c) For any 0 < m/r <
α(I)
γ(I)
, since
α(I)
γ(I)
= lim
s→∞
msα(I)/α(I
(ms)
) by Proposition II.2.3, for s ≫ 0 we
have m/r < msα(I)/α(I
(ms)
), and hence α(I
(ms)
) < rsα(I), so I
(ms)
⊆ I
rs
for s ≫ 0 by (a), hence m/r =
ms/(rs) ≤ ρ(I); i.e.,
α(I)
γ(I)
≤ ρ(I). And for any m/r ≥ reg(I)/γ(I), we have rreg(I) ≤ mγ(I) ≤ α(I
(m)
) so
I
(m)
⊆ I
r
by (b) and hence ρ(I) ≤
reg(I)
γ(I)
.
(d) If α(I
(m)
) < rα(I), then containment fails by (a), while if α(I
(m)
) ≥ rα(I), then containment holds
by (b).
Example IV.4.2. Let I be the ideal of p
1
, . . . , p
s
∈ P
n
for s =
_
d+n−1
n
_
general points. Then α(I) =
reg(I) = d, hence I
(m)
⊆ I
r
if and only if α(I
(m)
) ≥ rα(I). Unfortunately, α(I
(m)
) is not in general known.
See however Exercise IV.6.3.
IV.5. A Question and a Conjecture. The paper [BH] gives examples of reduced schemes Z
i
⊂ P
n
of
finite sets of points such that lim
i
ρ(I(Z
i
)) = n. This shows that the bound given in Theorem IV.3.6 is in
some sense sharp. However, one can hope to do better. In fact, Huneke has raised the following question:
Question IV.5.1 (Huneke). Let I ⊂ k[P
2
] be the ideal I = I(Z) where Z = p
1
+· · · +p
s
⊂ P
2
for a finite
set of distinct points p
i
. Must it be true that I
(3)
⊆ I
2
?
In the case of the ideal I of any s generic points of P
2
, [BH] showed that the answer is yes. This and
additional examples, both in dimension 2 and in higher dimensions, suggested the following conjecture (this
is a simplified version of [PSC, Conjecture 8.4.2]):
Conjecture IV.5.2 (Harbourne). Let I ⊂ k[P
n
] be the ideal I = I(Z) where Z = p
1
+· · · +p
s
⊂ P
n
for a
finite set of distinct points p
i
. Then I
(m)
⊆ I
r
if m ≥ rn −(n −1).
Example IV.5.3. Let I be the ideal of distinct points p
1
, . . . , p
s
∈ P
n
. Mimicking the argument of Example
IV.3.10 shows in fact that I
(rn−(n−1))
⊆ I
r
holds if char(k) = p > 0 and r is a power of p. See Exercise
IV.6.4.
We thus obtain an observation of Huneke:
Corollary IV.5.4. Question IV.5.1 has an affirmative answer when char(k) = 2.
Taking r = 2 in Conjecture IV.5.2 suggests in light of Theorem IV.4.1(a) the following possibly easier
question:
Question IV.5.5. Let 0 = I ⊂ k[P
n
] be any homogeneous ideal. Must it be true that α(I
(n+1)
) ≥ 2α(I)?
What is known is that α(I
(n+1)
) ≥
n+1
n
α(I). (For example, if I is the ideal of a set of points, this follows
from [Ch, Theorem 1]; alternatively, we have I
(rn)
⊆ I
r
by Theorem IV.3.6, and hence α(I
(rn)
) ≥ rα(I),
or α(I
(rn)
)/(rn) ≥ α(I)/n. But Proposition II.2.3 and its proof holds also for P
n
. Taking the limit as
r → ∞ gives γ(I) ≥ α(I)/n and, since α(I
(m)
)/m ≥ γ(I) for every m ≥ 0 as in Proposition II.2.3, we have
α(I
(n+1)
) ≥
n+1
n
α(I).) In fact, examples suggest that α(I
(rn−n+1)
) ≥ rα(I) + n − 1 may hold for the ideal
of any finite set of points in P
n
(and perhaps for any nontrivial homogeneous ideal in k[P
n
]).
22 B. HARBOURNE
IV.6. Exercises.
Exercise IV.6.1. Prove Lemma IV.3.1: Let Z = p
1
+· · · +p
r
⊂ P
n
and let I = I(Z) ⊆ R = k[P
n
]. Then
I
r
⊆ I
(m)
if and only if r ≥ m.
Solution (Exercise IV.6.1). If r ≥ m, then I
r
⊆ I
m
⊆ I
(m)
. Conversely, assume I
r
⊆ I
(m)
. Localize at p
1
and contract to R
i
, where p
1
∈ U
i
= Spec(R
i
)

= A
n
is a standard affine open neighborhood (the complement
of x
i
= 0, where R
i
= k[x
0
/x
i
, · · · , x
n
/x
i
] for some x
i
not vanishing at p
1
) to get J(p
1
)
r
⊆ J(p
1
)
m
, and
hence r ≥ m, where J(p
1
) ⊂ R
i
is the ideal of p
1
in R
i
.
Exercise IV.6.2. Prove Lemma IV.3.9: Let I ⊆ R = k[P
n
] be an ideal generated by s elements, and
assume char(k) = p > 0 and q is a power of p. Then I
sq
⊆ I
[q]
.
Solution (Exercise IV.6.2). Let f
1
, . . . , f
s
generate I. Then monomials in the f
i
of degree sq generate I
sq
,
and for each such monomial there must be an i such that f
q
i
is a factor. Thus each monomial is in I
[q]
.
Exercise IV.6.3. Let I ⊂ R = k[P
2
] be the ideal of non-collinear points p
1
, p
2
, p
3
∈ P
2
. Then I
(m)
⊆ I
r
if
and only if m ≥
4r−1
3
.
Solution (Exercise IV.6.3). Let X be the blow up of P
2
at the three points. Note that B = 3L − 2(E
1
+
E
2
+E
3
) = (L−E
1
−E
2
) +(L−E
1
−E
3
) +(L−E
2
−E
3
) ∈ EFF(X) and C = 2L−E
1
−E
2
−E
3
∈ NEF(X)
(since C is the class of a prime divisor of positive self-intersection). Now consider m = 2s +i for s ≥ 0 and
0 ≤ i ≤ 1. Then α(I
(m)
) = 3s + 2i, since sB +iC = (3s + 2i)L−(2s +i)(E
1
+E
2
+E
3
) ∈ EFF(X) is clear
but ((3s +2i −1)L−(2s +i)(E
1
+E
2
+E
3
)) · (2L−E
1
−E
2
−E
3
) < 0. In particular, α(I) = 2. Since R
2
/I
2
and R
1
/I
1
both have dimension 3, we also have reg(I) = 2.
Thus I
(m)
⊆ I
r
holds by Theorem IV.4.1(d) for m = 2s + i exactly when 3s + 2i ≥ 2r. First check the
case that m is even: containment holds exactly when 3m/2 ≥ 2r which is equivalent to (3m+ 1)/2 ≥ 2r, or
m ≥ (4r − 1)/3. Now say m is odd: containment holds exactly when (3m + 1)/2 = (3(m− 1) + 4)/2 ≥ 2r
which is again equivalent to m ≥ (4r −1)/3.
Exercise IV.6.4. Justify Example IV.5.3: Let J be the ideal of distinct points p
1
, . . . , p
s
∈ P
n
. Mimicking
the argument of Example IV.3.10 shows in fact that J
(rn−(n−1))
⊆ J
r
holds if char(k) = p > 0 and r is a
power of p.
Solution (Exercise IV.6.4). The same argument as given in Example IV.3.10 works except that we need to
refine the statement of Lemma IV.3.9 so that I
qs−(s−1)
⊆ I
[q]
. Let f
1
, . . . , f
s
generate I. Then monomials in
the f
i
of degree sq −(s −1) generate I
sq−(s−1)
, but for each such monomial there must in fact be an i such
that f
q
i
is a factor (if not, the monomial has degree at most s(q −1) in the f
i
, which is less than sq −(s−1)).
Thus each monomial is in I
[q]
.
References
[AV] A. Arsie and J. E. Vatne. A Note on Symbolic and Ordinary Powers of Homogeneous Ideals, Ann. Univ. Ferrara - Sez.
VII - Sc. Mat. Vol. IL, 19–30 (2003). (http://www.uib.no/People/nmajv/03.pdf).
[Bi] P. Biran. Constructing new ample divisors out of old ones, Duke Math. J. 98 (1999), no. 1, 113–135.
[BH] C. Bocci and B. Harbourne. Comparing Powers and Symbolic Powers of Ideals, to appear, Journal of Algebraic Geom-
etry.
[Ch] G. V. Chudnovsky. Singular points on complex hypersurfaces and multidimensional Schwarz Lemma, S´eminaire de
Th´eorie des Nombres, Paris 1979–80, S´eminaire Delange-Pisot-Poitou, Progress in Math vol. 12, M-J Bertin, editor,
Birkh¨ auser, Boston-Basel-Stutgart (1981).
[D] J.-P. Demailly. Singular Hermitian metrics on positive line bundles, Complex algebraic varieties (Bayreuth, 1990), Lect.
Notes Math. 1507, Springer-Verlag, 1992, pp. 87–104.
[Du] M. Dumnicki. Reduction method for linear systems of plane curves with base fat points, preprint 2006
(math.AG/0606716).
[ELS] L. Ein, R. Lazarsfeld and K. Smith. Uniform bounds and symbolic powers on smooth varieties, Invent. Math. 144
(2001), p. 241-252.
[Es] D. Eisenbud. Commutative Algebra With A View Toward Algebraic Geometry, New York : Springer-Verlag, 1995, pp.
xvi+797.
GLOBAL ASPECTS OF THE GEOMETRY OF SURFACES 23
[EGA] A. Grothendieck and J. Dieudonn´e.
´
Elements de G´eom´etrie Alg´ebrique IV, Publ. Math. 32, Inst. Hautes
´
Etudes Sci.,
1967.
[GM] C. Galindo and F. Monserrat. The total coordinate ring of a smooth projective surface, Journal of Algebra 284 (2005)
91–101.
[GHM] A. Geramita, B. Harbourne and J. Migliore. Classifying Hilbert functions of fat point subschemes in P
2
, Collect. Math.
60, 2 (2009), 159–192.
[GGP] A. V. Geramita, A. Gimigliano and Y. Pitteloud. Graded Betti numbers of some embedded rational n-folds, Math.
Annalen 301 (1995), 363-380.
[Ha1] B. Harbourne. Free Resolutions of Fat Point Ideals on P
2
, J. Pure Appl. Alg. 125, 213–234 (1998).
[Ha2] B. Harbourne. Anticanonical rational surfaces, Trans. Amer. Math. Soc. 349, 1191–1208 (1997).
[Ha3] B. Harbourne. The geometry of rational surfaces and Hilbert functions of points in the plane, Canadian Mathematical
Society Conference Proceedings 6 , 95–111 (1986).
[Ha4] B. Harbourne. Seshadri constants and very ample divisors on algebraic surfaces, J. Reine Angew. Math. 559 (2003)
115–122.
[Ha5] B. Harbourne. On Nagata’s Conjecture, J. Alg. 236 (2001), 692–702.
[Ha6] B. Harbourne. Complete linear systems on rational surfaces, Trans. Amer. Math. Soc. 289, 213–226 (1985).
[HR1] B. Harbourne and J. Ro´e. Discrete behavior of Seshadri constants on surfaces, Journal of Pure and Applied Algebra
212 (2008), 616–627.
[HR2] B. Harbourne and J. Ro´e. Computing multi-point Seshadri constants on P
2
, to appear, Bulletin of the Belgian Math-
ematical Society - Simon Stevin.
[HR3] B. Harbourne and J. Ro´e. Linear systems with multiple base points in P
2
, Adv. Geom. 4 (2004), 41–59.
[Hr] R. Hartshorne. Algebraic Geometry, Graduate texts in mathematics (52), New York, Springer-Verlag, 496 pp., 1977.
[HH] M. Hochster and C. Huneke. Comparison of symbolic and ordinary powers of ideals, Invent. Math. 147 (2002), no. 2,
349–369.
[HK] Y. Hu and S. Keel. Mori dream spaces and GIT, Michigan Math. J. 48 (2000) 331–348.
[HS] C. Huneke and I. Swanson. Integral Closure of Ideals, Rings, and Modules. London Math. Soc. Lecture Note Series 336
(2006), Cambridge University Press, pp. 421.
[K] S. L. Kleiman. Multiple-point formulas. I: Iteration, Acta Math., 147:2 (1981), 13–49.
[LH] M. Lahyane and B. Harbourne. Irreducibility of (−1)-classes of anticanonical rational surfaces, Pac. J. Math., 218 No.
1 (2005), 101–114.
[Lz] R. Lazarsfeld. Positivity in Algebraic Geometry I, Springer-Verlag, 2004.
[M] H. Matsumura. Commutative Algebra. W. A. Benjamin, New York, (1970), pp. 212 + xii
[N1] M. Nagata. On rational surfaces, II, Mem. Coll. Sci. Univ. Kyoto, Ser. A Math. 33 (1960), 271–293.
[N2] M. Nagata. On the 14-th problem of Hilbert, Amer. J. Math. 81 (1959), 766–772.
[Ot] J. Ottem. On the Cox ring of P
2
blown up in several points on a line, arXiv:0901.4277v3, preprint 2009.
[PSC] T. Bauer, S. Di Rocco, B. Harbourne, M. Kapustka, A. Knutsen, W. Syzdek, and T. Szemberg. A primer on Seshadri
constants, to appear in the AMS Contemporary Mathematics series volume “Interactions of Classical and Numerical
Algebraic Geometry,” Proceedings of a conference in honor of A. J. Sommese, held at Notre Dame, May 22–24 2008.
[R] J. Ro´e. On the existence of plane curves with imposed multiple points, J. Pure Appl. Alg. 156 (2001), 115–126.
[Ro] J. Rosoff. Effective divisor dlasses and blowings-up of P
2
, Pac. J. Math. 89, No. 2 (1980), 419–429.
[Sw] I. Swanson. Linear equivalence of topologies, Math. Z. 234 (2000), 755–775.
[Sz] T. Szemberg. Global and local positivity of line bundles, Habilitation, 2001.
[SS] B. Strycharz-Szemberg and T. Szemberg. Remarks on the Nagata Conjecture, Serdica Math. J. 30 (2004), 405–430.
[ST] T. Szemberg and H. Tutaj-Gasi´ nska. General blow ups of the projective plane, Proc. Amer. Math. Soc. 130 (2002),
no. 9, 2515–2524.
[Te] Z. Teitler. Bounding symbolic powers via asymptotic multiplier ideals, preprint, 2009 (arXiv:0907.4045).
[TVV] D. Testa, A. V´ arilly-Alvarado and M. Velasco. Big rational surfaces, arXiv:0901.1094, preprint 2009.
[T] H. Tutaj-Gasi´ nska. A bound for Seshadri constants on P
2
, Math. Nachr. 257 (2003), no. 1, 108–116.
[W] M. Waldschmidt. Properties arithmetiques de fonctions de pleusieurs variables, II, S´eminaire Lelong: Analyse, 16
e
anee, 1975–76, no. 10, 108–135. Berlin, Springer-Verlag, LNM 578, 1977.
[X] G. Xu. Curves in P
2
and symplectic packings, Math. Ann. 299 (1994), 609–613.
Department of Mathematics, University of Nebraska, Lincoln, NE 68588-0130 USA
E-mail address: bharbour@math.unl.edu

2

B. HARBOURNE

Theorem I.1.5 (HIT). Given a surface X and D ∈ Num(X) with D2 > 0, then the intersection form on the space D⊥ ⊆ Num(X) of classes F with F · D = 0 is negative definite. Finally, we recall the semicontinuity principle. We say points p1 , . . . , pr are essentially distinct points of a surface X, and that Xr+1 is the blow up of X at p1 , . . . , pr , if p1 ∈ X = X1 , π1 : X2 → X1 is the blow up of X1 at p1 , and for 1 < i ≤ r we have pi ∈ Xi , and πi : Xi+1 → Xi is the blow up of Xi at pi . By identifying Xi+1 with Xi away from pi , we can regard pi+1 as being in Xi when πi (pi+1 ) = pi . In this way distinct points p1 , . . . , pr ∈ X can be regarded as being essentially distinct. Let πj,i : Xj → Xi be the morphism πj−1 ◦ · · · ◦ πi whenever j > i, and let Ei be the divisor (or divisor class, depending on context) given by −1 πr,i (pi ). When X = P2 , let L be the pullback to Xr+1 of a general line on X. Otherwise we will assume L is the pullback to Xr+1 of some ample divisor on X. Theorem I.1.6 (Semicontinuity Principle). Let a, a1 , . . . , ar be integers, let p1 , . . . , pr be general points of X = P2 and denote by Xr+1 the blow up of X at p1 , . . . , pr , with L, E1 , . . . , Er being the usual associated ′ classes. Also, given essentially distinct points p′ , . . . , p′ of X, let Xr+1 denote the blow up of X at p′ , . . . , p′ , 1 r 1 r ′ ′ ′ and let L , E1 , . . . , Er denote the associated classes. ′ ′ (a) If aL − i ai Ei ∈ EFF(Xr+1 ), then aL′ − i ai Ei ∈ EFF(Xr+1 ) for every choice of essentially ′ ′ distinct points p1 , . . . , pr of X. ′ ′ (b) If aL′ − i ai Ei ∈ NEF(Xr+1 ) for some choice of essentially distinct points p′ , . . . , p′ of X and if 1 r ′ 2 ′ (aL − i ai Ei ) > 0, then aL − i ai Ei ∈ NEF(Xr+1 ). Proof. Following [K], parameterize essentially distinct points of X by schemes Wi where we set b0 : W1 → W0 to be X → Spec (k) and recursively we define Wi+1 → Wi ×Wi−1 Wi to be the blow up of the diagonal in Wi ×Wi−1 Wi , setting bi : Wi+1 → Wi to be the composition of Wi+1 → Wi ×Wi−1 Wi with the projection π1i : Wi ×Wi−1 Wi → Wi to the first factor. The morphisms bi are smooth ([EGA, 17.3, 19.4]). [Here is a proof. Note b0 is smooth (since X is). Assuming bi−1 is smooth, we see the projection Wi ×Wi−1 Wi → Wi is smooth ([Hr, Proposition III.10.1(d)]), and the exceptional locus Bi+1 ⊂ Wi+1 for bi is smooth ([Hr, Theorem II.8.24(b)]) and locally isomorphic to Wi × P1 . We now see that bi is smooth by checking surjectivity of the induced maps on Zariski tangent spaces ([Hr, Proposition III.10.4(iii)]) at points x ∈ Wi+1 . Away from Bi+1 , Wi+1 → Wi ×Wi−1 Wi is an isomorphism and Wi ×Wi−1 Wi → Wi is smooth, hence surjectivity follows for points x ∈ Bi+1 . At points x ∈ Bi+1 , the composition Bi+1 ⊂ Wi+1 → Wi is smooth and thus the map on tangent spaces induced by Bi+1 → Wi is already surjective at x, hence so is the one induced by Wi+1 → Wi . Thus bi is smooth.] ′ Consider the pullbacks Bi to Wr+1 of the divisors Bi . For any ample divisor L on X, let L′′ be the pullback ′ to Wr+1 via the blow ups bi and the projections π2i on the second factors. Let F = OWr+1 (aL′′ − i ai Bi ). ′ ′ Then for any essentially distinct points p1 , . . . , pr of X we have a uniquely determined point w ∈ Wr , the ′ ′ ′ fiber (Wr+1 )w of Wr+1 over w is Xr+1 , and the restriction Fw of F to (Wr+1 )w is OXr+1 (aL′ − i ai Ei ). By the semicontinuity theorem ([Hr, Theorem III.12.8]), h0 ((Wr+1 )w , Fw ) is an upper semicontinuous function of w. This implies (a). ′ Now consider (b). If (aL − i ai Ei )2 = (aL′ − i ai Ei )2 > 0, we have s(aL − i ai Ei ) ∈ EFF(Xr+1 ) for some s ≫ 0. Pick some effective divisor C whose class is s(aL − i ai Ei ). For each prime divisor component D of C, there is an open set of points pi for which D remains prime, since being effective is a closed condition by (a), and since for only finitely many classes D′ = a′ L − i a′ Ei could D′ and the i class of D − D′ (or even s(aL − i ai Ei ) − D′ in place of D − D′ ) both conceivably be classes of effective divisors. Thus the decomposition of C as a sum of prime divisors is well-defined for general points, and ′ ′ each component specializes to an effective divisor on Xr+1 which thus meets aL′ − i ai Ei , and hence aL − i ai Ei , non-negatively, so aL − i ai Ei ∈ NEF(Xr+1 ), proving (b). Here is a version of the same result stated for generic points, where X now is any surface and L comes via pullback from some ample divisor on X: Theorem I.1.7 (Semicontinuity Principle 2). Let a, a1 , . . . , ar be integers, let p1 , . . . , pr be generic points of a surface X and denote by Xr+1 the blow up of X at p1 , . . . , pr , with E1 , . . . , Er being the usual associated classes and L the pullback to Xr+1 from X of some ample divisor on X. Also, given essentially distinct ′ ′ ′ points p′ , . . . , p′ of X, let Xr+1 denote the blow up of X at p′ , . . . , p′ , and let L′ , E1 , . . . , Er denote the 1 r 1 r associated classes.

3. . Corollary I. Since −mKX ∈ EFF(X). X does not have bounded negativity.1 is false in positive characteristic.2. So. then r 1 aL − i ai Ei ∈ NEF(Xr+1 ).1. If EFF(X)/∼ is finitely generated.4.2. .2.1(b). m(sL − E1 − · · · − Er ) is not the class of an effective divisor. The proof of (a) is the same as for Theorem I.2. sL − E1 − · · · − Er is nef but not ample (since F 2 = 0). the class F is in fact ample (see Exercise III. see Exercise I. But it is not always clear when it holds if one blows up points on those surfaces. in Exercise I.1. . suppose X is given by blowing up r = s2 generic points pi ∈ P2 . where q is a sufficiently large power of p.7. thus some multiple of F is effective (in fact 2F ∈ EFF(X) by Riemann-Roch). I. Let us say that a surface X has bounded negativity if there is an integer nX such that C 2 ≥ nX for each prime divisor C ⊂ X. the base curve C on the Hirzebruch surface Hn is effective but has C 2 = −n so is not nef.4. The proof for (b) is even simpler than before since now we are not claiming that having a specific divisor F = aL − i ai Ei be nef is an open condition on the points pi . The question of the extent to which this can happen is the main motivation for these notes. and relatively minimal rational surfaces. Remark I. there are only finitely many prime divisors C such that −mKX · C < 0. A surface X has bounded negativity if −mKX ∈ EFF(X) for some positive integer m. Instead.org/communications/vmath/VMathVideos/VideoInfo/4111/show video). Let X = C × C. OX (mF )) = 0 for all m > 0 when F = sL − E1 − · · · − Er and s > 3. Proof. It is certainly possible. Let EFF(X)/ denote the image of EFF(X) in Num(X).2.4). But by specializing the points pi to general points of a smooth curve of degree s. In preparation for giving a criterion for bounded ∼ negativity to hold on X. Remark I. Since q can be arbitrarily large. but divisors can be nef without being ample and without any multiple being effective.6(a). hence ′ ′ C ′ = c0 L′ − i ci Ei ∈ EFF(Xr+1 ) by (a). we have the following proposition (taken from [Ro]): Proposition I. apart from finitely many prime divisors C. 1 r ′ ′ (b) If aL′ − i ai Ei ∈ NEF(Xr+1 ) for some choice of essentially distinct points p′ . . . Conjecture I. then aL′ − i ai Ei ∈ EFF(Xr+1 ) for every choice of essentially distinct points p′ .5. I thank Burt Totaro for bringing to my attention the following example pointed out by J´nos Koll´r at a talk by Richard Harris at MSRI in Jana a uary. In particular.msri. and. . Every surface X in characteristic 0 has bounded negativity. For an example of a nef divisor which is not effective. However. Some surfaces are known to have bounded negativity. but in general neither containment holds. However. . If in fact there is no H ∈ EFF(X) with H 2 < 0. if F were not nef. For example.2. 2009 (http://www.10]. then there are only finitely many prime divisors C with C 2 < 0. Proof.GLOBAL ASPECTS OF THE GEOMETRY OF SURFACES 3 ′ ′ (a) If aL − i ai Ei ∈ EFF(Xr+1 ).1(b). . p′ of X. It is not hard to show that F ∈ NEF(X) implies F 2 ≥ 0 (this is Exercise I. Example I. Let Γq be the graph in X of the Frobenius morphism defined by taking qth powers.1.2. Then Γq is a curve on X with X 2 = q(2 − 2gC ) [Hr.1 (Folklore: The Bounded Negativity Conjecture). in which case C 2 = 2pC − 2 − C · KX ≥ −2. and hence nef to begin with by Theorem I. however. For example. abelian surfaces. for each m > 0.8.4.4.1. p′ of X. contradicting our ′ ′ assumption that F ∈ NEF(Xr+1 ). for n > 0. where C is a curve of genus gC ≥ 2 defined over a finite field of characteristic p > 0. then it is easy to see that EFF(X) ⊆ NEF(X).2. Thus for r = s2 generic points pi . so F ′ · C ′ = F · C < 0 for F ′ = aL′ − i ai Ei . Enriques surfaces. bounded negativity holds for K3 surfaces. it is as far as I know still an open problem even in positive characteristic to determine which surfaces fail to have bounded negativity.2. There is a long-standing open conjecture involving boundedness of negativity on surfaces. Exercise V. to have H ∈ EFF(X) with H 2 < 0. It can also happen that NEF(X) ⊆ EFF(X). . A Motivational Folklore Conjecture. Nagata [N2] proved that h0 (X. Conjecture I.1(a)). then F · C < 0 for some C = c0 L − i ci Ei ∈ EFF(Xr+1 ). we see that sL − E1 − · · · − Er is nef after specializing. we have −mKX · C ≥ 0.

pr ).2. pr ) = inf C ·L : C is prime and i C · Ei m : dL − m d C · Ei > 0 i = sup i Ei ∈ NEF(X).) It is clear that the infimum exists: Let C = dL − L − Ei ∈ NEF(X). there are only finitely many prime divisors D such that D · Ci < 0 for some i. or. Since We now recall a quantity ε introduced by G. . see also [Lz]. we see that C = dL − i mi Ei for some mi ≥ 0. . Since C is prime and we assume L · C > 0. or see [LH] or [TVV] for KX > 0 in general). i mi Ei . but it is still true that C 2 ≥ −1 for any prime divisor C. pr ) = inf mi Ei ∈ EFF(X). which with adjunction implies C 2 ≥ −1 for any prime divisor C. pr ∈ P2 .4.3. more generally.1) Ei > 0 : C ∈ EFF(X). Corollary I. C · λL (X) = inf (C · L)2 i = inf C2 : C is a prime divisor on X and C · L > 0 (C · L)2 (The second equality is Exercise I. C · L > 0.4. for example). we have C ∼ i mi Ci for some mi ≥ 0 and so 0 > C 2 = i mi C · Ci . . i Problem I. then 2 EFF(X) is finitely generated (see Exercise I. . As motivation for introducing multipoint Seshadri constants.2. (Chudnovsky’s version. . . . given a nef divisor F .5 if −KX is nef and KX > 1. pr ∈ P2 . .2. we can instead ask for a lower bound on C 2 /(F ·C)2 for all prime divisors C with F · C > 0. For r = 9.1 is not yet known: Conjecture I. i. denoted Ω0 (p1 . If X is a rational surface with KX > 0. . then KX > 0 and so EFF(X) is finitely generated as mentioned above. EFF(X) is not finitely generated (see Exercise I.3. .9) we have: (I. .4. We also have EFF(X) finitely generated if X is obtained by blowing up points on a line or conic in P2 (see Exercises I. Now let C be a prime divisor with C 2 < 0. then bounded negativity holds ∼ for X. or at least give good estimates for it. Let C1 .2 and I. . but in fact −KX is ample. .2. .3) ε(P2 . . Here now is a conjecture for a case where EFF(X) is definitely not finitely generated (see Exercise I. Since each Ci is the class of an effective divisor. V. . An Asymptotic Approach to Bounded Negativity. we now study this question in the case that X is obtained by blowing up r > 0 points p1 . p1 .4.. taking F to be L. with equality if and only if C is a smooth rational curve with KX · C = −1. mi ≥ 0. HARBOURNE Proof. Chudnovsky [Ch] (for any r points in any projective space) and Demailly [D] (for a single point.4. This then gives a criterion for bounded negativity to hold. Then C 2 ≥ −1 for every prime divisor C. Hence C 2 /(C · L)2 = (d2 − i m2 )/d2 ≥ 1 − r. Compute λL (X). is actually equal to rε. : dL − i mi i i As alternative definitions (see Exercise I. Here are some cases where it is known. I. we see d ≥ mi for all i.4). . 2 When X be obtained by blowing up r < 9 generic points of P2 .8 ([Ha3]). see [Ro] for the case that 2 2 KX > 0 and X is obtained by blowing up at most 8 points of P2 . It is difficult in general to determine whether EFF(X)/∼ is finitely generated.4. Cr be prime divisors whose classes generate EFF(X)/∼. .7. Let C2 (I. . but on any smooth variety) now known as a multipoint Seshadri constant.7) but where Conjecture I. Then d mi > 0 ε(P2 .3.8. r = 1.e. . and so C = Cj .2. .4. although the proof is somewhat technical. Let X be obtained by blowing up r > 9 generic points of P2 . m > 0 . hence C · Cj < 0 for some j. Remark I.7. if X is rational and −KX is big [TVV]. C · Ei ≥ 0 for all i.6.3. . p1 . While no general lower bound for C 2 for prime divisors C on a surface X is known.4 B. .) Let X be obtained by blowing up distinct points p1 . even for 2 rational surfaces. . . If X is a surface such that EFF(X)/ is finitely generated.

2. mi Ei ∈ EFF(X). if F · C = 0 for some F = dL − m i Ei ∈ NEF(X) and C = aL − i mi Ei ∈ EFF(X) with d > 0 and a > 0. .. if dL − i mi Ei ∈ EFF(X) then d ≥ mi for all i. . Applying the lemma now gives: Ei > 0 i = inf 1 − = inf 1 − ≥ inf 1 − = 1 − sup =1− m2 i d2 i : C = dL − : dL − i mi Ei ∈ EFF(X).4. OX (F1 + F2 )). . Proposition 4. i Corollary I. mi ≥ 0. mi ≥ 0. . hence P dmi ≥ 1/r. However. See Exercises I.6. If X is obtained by blowing up at most 8 points of P2 .12. . Let X be obtained by blowing up distinct points p1 . p1 . Sometimes equality holds. hence the result follows over the complex numbers from [GM.1]. OX (F1 )) ⊗ H 0 (X. . Theorem III. i In particular. d > 0. p1 . For our asymptotic application of Seshadri constants to bounded negativity. then EFF(X) must be finitely generated.3. . Theorem III. . have a positive multiple which is effective and base point free). . ε(P2 . Let L be the total transform of a line and let Ei be the blow up of the point pi . 2 see [TVV]. (Proof: By the Hodge Index Theorem if F is a nontrivial nef divisor. if X is any rational surface with KX > 0.ar )∈Zr+1 H 0 (X. . We close this lecture with some remarks about Mori dream spaces. . pr ) = m/d.4. λL (X) ≥ 1 − ε(P2 . dividing by d2 gives the result. . Theorem 2] and [Ha2. . mi ≥ 0.. . we can again conclude that X is a Mori dream space. .11 and I. Then 1 . pr ) ≥ 1/r > 0 so 1/ε(P2 . nef divisors are semi-ample (i.. and hence X has bounded negativity by Proposition I.3(a)] in place of [Ro]. (The same proof applies. we will use the following elementary inequality: Lemma I. EFF(X) is finitely generated. . and by Exercise I. C · (C · L)2 i Since L − Ei ∈ NEF(X) for each i. p1 . Remark I.8.1]. then ε(P2 . .3. (By [LH.5.3(c)]. . C · L > 0. pr ∈ P2 . pr ) Remark I. . but usually not. d mi ≥ i m2 . pr ) makes sense. OX (F2 )) → H 0 (X. mi : dL − d i mi > 0 i i mi : dL − d mi > 0 i i 1 ε(P2 . . we have Proof. .. and where multiplication is given by the natural maps H 0 (X. Corollary 1] by [Ro. mi ≥ 0. In general.7.e. . Given integers d > 0 and d ≥ mi ≥ 0 for all i. .3. . . . Define the Cox ring Cox(X) of X to be the ring whose additive structure is given by (a0 .) If KX = 0 but −KX is not nef.4.4.6(b) and [Ha2.3. . p1 . If X is a Mori dream space.GLOBAL ASPECTS OF THE GEOMETRY OF SURFACES 5 Remark I. alternatively. using [LH. C · Ei ≥ 0 for all i. λL (X) = inf C2 : C ∈ EFF(X). but without the assumption of the complex numbers. . then X is a Mori dream space. then −KX · F > 0. . mi Ei ∈ EFF(X). Let X be obtained by blowing up points p1 .4.3. Now apply . then X is a Mori dream space. pr ∈ P2 . ε(P2 . . .. pr ) is itself hard to compute.10. p1 . Just note that d ≥ mi ≥ 0 implies d i i m2 i d2 ≤ i mi . by Exercise I.) In 2 fact. pr ) Proof. Mori dream spaces give interesting examples of surfaces with bounded negativity. p1 . Proposition 4. OX (a0 L − a1 E1 − · · · − ae Er )). mi > 0 i mi > 0 i m2 i d2 i i mi Ei ∈ EFF(X). .5. If Cox(X) is finitely generated we say that X is a Mori dream space [HK].

. contradicting F being nef.4.4. and blow up the curve r > 9 times. (b) Consider points p′ which lie on a smooth quintic. Hence we must have F 2 ≥ 0. EFF(X) is finitely generated and any nef class F has F · (−KX ) ≥ 0. F = 5L − E1 − · · · − E21 ∈ NEF(X) when the points pi are general. so (aA + f F ) · C > 0.4. . . . . . . Here is a question I do not know the answer to: Question I. reduced. Then i ′ ′ Q is nef but Q is linearly equivalent to F ′ = 5L′ − E1 − · · · − E21 . Theorem 3. (a) Let A be ample. . Let F = 5L − i Ei . To show aA + f F is ample it is enough by the Nakai-Moisezon criterion [Hr] to show that (aA + f F ) · C > 0 for every curve C. Er .4.1 and Corollary 3. we have (aA+ f F )·F = a(A+ tF )·F < 0 for t > −F 2 . but −KX is not big since it is a prime divisor of negative self-intersection.2. But A · C > 0 since A is ample and F · C ≥ 0 since F is nef. E1 .1. a and f such that aA + f F is ample and saA + sf F is effective. Let Λ = L − E1 − · · · − Er and let Li = L − Ei .4 of blow ups of points on a line or conic) are all subsumed by the result of [TVV] that a rational surface with big −KX is a Mori dream space. . pr .2]. By [Ha6. but such that F · (aA + f F ) < 0. OX (F )) = 0 hence F ∈ EFF(X). Let the points be p1 .9. .. Then Cl(X) has basis L.13). irreducible divisor can be written as a non-negative integer combination of Λ and E1 . So let C = aΛ + bi Ei be the class of an effective. . .14. Solution (Exercise I. Since L2 = 0 and Li is the class of a prime divisor.4. . .e. However. .4.3 and I. X is a Mori dream space since EFF(X) is finitely generated and any nef class is semi-ample. Moreover. show that F ∈ NEF(X) \ EFF(X). i. so by choosing a and f A·F such that t is slightly larger than −F 2 we will still have (aA + f F )2 > 0 (and hence aA + f F is ample so saA + sf F is effective for s ≫ 0) while also having (saA + sf F ) · F < 0. Exercise I. (a) Show F ∈ NEF(X) implies F 2 ≥ 0. Corollary 1]. so we may assume that C is not one of these divisors. Theorem I. Solution (Exercise I. This solution is based on [Ha3.4. where L is the pullback of the class of a line and Ei is the class of the blow up of pi . −KX is big (see Exercise I. Putting this into 0 ≤ C · Λ = a − bi implies the non-negativity of a. Let X be a surface. Exercises. then the claim is true as Lj = Λ + i=j Ei . Now by the semicontinuity principle. h0 (X. If X is a rational surface with EFF(X) finitely generated. If C is Λ or Lj .5. By Exercise I.4]. Find an explicit finite set of generators for EFF(X) and NEF(X) in case X is obtained by blowing up r ≥ 1 distinct points on a line in P2 .1). it is enough to prove that every effective. Er .1. every nef class is semi-ample. and by [GM]. Then Cl(X) has basis L.6 B.4. not all rational surfaces which are Mori dream spaces have big −KX . aA ∈ EFF(X) for a ≫ 0. We will show that there are positive integers s. . Thus. hence A·F A·F ≥ 0. . Er generate EFF(X). E21 . E1 . Proposition I. reduced and irreducible divisor. . Let Q be the proper transform of that quintic. taking t = f /a. and that (aA + f F )2 > 0. HARBOURNE [GM. Exercise I.2). . let C be an irreducible cubic curve.) In each of these cases.6. E1 .3. hence these (in addition to the examples of Exercises I.4. we see Li ∈ NEF(X). but a(A+ tF )·F = 0 A·F 2 2 2 2 2 2 2 and (aA + f F ) = a (A + 2tA · F + t F ) = a (A + tA · F ) > 0 for t = −F 2 . F nef and F 2 < 0. . (b) Assume X is obtained by blowing up r = 21 general points pi ∈ P2 . Since A is ample. which is impossible if F is nef. is X a Mori dream space? I. . each time at successive infinitely near points of the cubic. But the points are general so impose 21 independent conditions on the 21 dimensional space of all quintics (since we can always choose each successive point not to be a base point of the linear system of quintics through the previous points). Clearly Λ ∈ EFF(X) and Ei and Li are in EFF(X) for each i. For example. starting with a flex point of the cubic. To prove that i Λ. But then the intersection with them must be non-negative and hence bj = C · Lj ≥ 0. where L is the pullback of the class of a line and Ei is the class of the blow up of pi .

4). Since X is rational.4. Solution (Exercise I.5.4. OX (F )) ⊗ H 0 (X.3. h1 (X. it is enough to show .4. the conic does not need to be smooth here but smoothness simplifies the argument a bit. Let X be obtained by blowing up points p1 .GLOBAL ASPECTS OF THE GEOMETRY OF SURFACES 7 if F is nef then F meets each Ei and Λ non-negatively. br . OX (KX )) = h1 (X. OX (M )) is surjective. hence 0 ≤ −KX · C ≤ 2 C 2 + 2 = n + 2 by adjunction and the fact that −KX is nef. . show that E ∈ EFF(X). In particular.) (a) Show EFF(X) is finitely generated. OX ) = 0. then by the solution to Exercise I. OX (F1 )) ⊗ H 0 (X. Thus it is enough to show that i H 0 (X.2. .4. so 2 C = ((−KX · C)(−KX ) + N )/(KX ). Theorem III. and M ∈ NEF(X) and hence M = i≥0 mi Li for some non-negative mi (where we take L0 = L). so for each n it is enough to show that there are only finitely many C with C 2 = n. (b) Cite the literature to show that X is a Mori dream space. the class of every prime divisor is a sum of non-negative multiples of classes of the form Lij . and hence C is either Lij or Ei for some i and j. L1 . pr on a smooth conic in P2 with r ≥ 3. . then F · L = 0. This is because if D is an effective divisor.4. . . if F is nef with −KX · F = 0. OX (KX )) = 0 since −KX ∈ EFF(X) is nontrivial. see [Ot]. 2 (d) If KX > 1. . Theorem 2. i = j. and the argument we just used on C shows that any such class can be written as F = (a − i bi )L + i bi Li for non-negative integers a. so −KX ∈ EFF(X). Corollary 3]. Er . be the class of the proper transform of the line through pi and pj . Lr . where N is fixed and consists of a sum of non-negative multiples of the Ei and Λ. . The generators are given by taking a basis for H 0 (X. Do this inductively by showing that H 0 (X. . . Corollary 1] and [Ha2.4. hence F = 0 by the Hodge Index Theorem. (b) For each integer n. show that C + KX ∈ EFF(X). (a) If C is a prime divisor on X such that pC > 0. Thus to show there are only finitely many such C.5).2. OC (C + KX )) = h0 (X. or by [TVV]. OX (−KX )) ≥ KX + 1 > 0 by Riemann-Roch. OX (Li )) → H 0 (X. Exercise I. and hence L. . Alternatively. where a0 − a1 − a3 ≥ 0 since C · L12 ≥ 0 implies a0 ≥ a1 + a2 ≥ a1 + a3 . Let X be a rational surface such that −KX is nef. Thus 0 < pC = h0 (C. . Hence if C 2 < 0 but C = D. Since C · Ei ≥ 0. 2 (e) Conclude that EFF(X) is finitely generated if KX > 1. OX (G)) for each G among E1 . and h0 (X. . If the points pi are collinear. If C 2 ≥ 0. OX (−KX )) = 0 2 h (X.3). Also. Next note that h2 (X. . b1 . . Ei and D. Let Di = 2L − E1 − · · · − Ei = D + Ei+1 + · · · + Er . let D be the class of the proper transform of the conic and let Ei be the class of the blow up of pi for each i > 0. Hence h0 (X. OX (C + KX )). show that −KX − E ∈ EFF(X) for any class E such that E 2 = E · KX = −1. show that X is a Mori dream space. (b) Since −KX = D + L. . (c) If E is a class such that E 2 = E · KX = −1. we have ai ≥ 0 for all i. Thus C = (a0 − a1 − a3 )L + (a1 − a2 )(L − E1 ) + (a2 − a3 )L12 + (a3 − a2 )D3 + · · · + (ar−1 − ar )Dr−1 + ar Dr . First note that KX ≥ 0 since −KX is nef. OX (F1 + F2 )) is surjective whenever F1 and F2 are nef (see [Ha1. . By reindexing we may assume that a1 ≥ a2 ≥ · · · ≥ ar ≥ 0. or directly by [GM.1].8]). . 2 Solution (Exercise I. let L be the class of the total transform of a line. the points are collinear and the result is given by Exercise I. (a) Let Lij . The fact that X is a Mori dream space now follows by [GM. (a) Take cohomology of 0 → OX (KX ) → OX (C + KX ) → OC (C + KX ) → 0. The basic idea of part (a) of the next exercise is taken from [Ro].4. . show that there are only finitely many classes C of prime divisors with C 2 ≤ n 2 if KX > 0. Λ and L1 . . . OX (2KX )) is 0 since X is rational. (b) By adjunction and the fact that −KX is nef we have C 2 = 2pC − 2 − KX · C ≥ −2. Exercise I. Now let N = KX C − (KX · C)KX . . Exercise I.4. Note that L = L12 + E1 + E2 and that L − E1 = L12 + E2 . (If r < 3. So say C 2 = n. OX (Li ))⊗mi → H 0 (X. Note that −KX = D + L. pr ∈ P2 . Let X be obtained by blowing up points p1 . D = N + M . . then adjunction forces L · C ≤ 1. OX (F + Li )) is surjective for each F ∈ NEF(X) and hence in fact that H 0 (X. OX (F2 )) → H 0 (X. Lr generate NEF(X).4. . write C = a0 L − a1 E1 − · · · − ar Er . . Solution (Exercise I. Let C be the class of a prime divisor.

there are only finitely many elements of Cl(X) orthogonal to KX with self-intersection no less than 2 2 (KX )(KX n − (n + 2)2 ). Thus 2 h0 (X. Thus mj − i < 0 (since −3(mj − i) = (mj − i)KX · L = jF · L ≥ 0 but F · L = 0 implies F = 0) so −KX is nef (being a positive rational multiple of a nef class).4. we see that 0 = h0 (X. and since −KX · (KX − E) < 0. p9 ∈ P2 on a smooth plane cubic D′ . We may assume C · C1 ≤ C · Ci for all i. Since −KX · E = 1 and −KX is nef. hence h0 (X. OX (E)) = 0. we see KX − E ∈ EFF(X). Let v = j(−mKX + F ) and let w = −iKX . so −KX − E ∈ EFF(X). it is enough to show that either C − D ∈ EFF(X) for some such D or that C − Ci ∈ EFF(X) for some i. so E ∈ EFF(X). so assume C 2 > 0. ⊥ (b) Let N be any class in K ⊥ ∩ E9 .7. Now apply Riemann-Roch to see that h0 (X. hence j(−mKX + F ) = −iKX so jF = (mj − i)KX . (a) By Riemann-Roch. (d) Since −KX · (2KX + E) < 0. Likewise. OX (E)) ≥ 1 by Riemann-Roch. (a) Show that −KX ∈ EFF(X). OX (− i>1 Ci − C)) = 0. . Since −KX = D is nef. OX (2KX )) = 0. Since F = 0 and KX · F = 0. OX (KX − C + C1 )) = h0 (X. so C − C1 ∈ EFF(X). but this contradicts the fact that −KX = 0. More rigorously. but (v − w)2 = 0. We will now see that these classes D. so intuitively there are only finitely many lattice elements ⊥ 2 2 N in KX of length at most −(KX )(KX n − (n + 2)2 ). so this is impossible. . Thus F = 0. there are only finitely many classes D of prime divisors with D2 ≤ 0 and pD = 0. since Cl(X) is free abelian of finite rank. 2 Exercise I. (b) That E = N + E9 + (N 2 /2)KX satisfies E 2 = KX · E = −1 is easy. hence −KX · F = 0 and KX = 0 imply that 2 2 −KX = 0. Thus h2 (X. To see the latter. Conclude that E ∈ EFF(X).1). contrary to hypothesis. show that each such E is the class of a prime divisor. Show that E = N + E9 + (N 2 /2)KX satisfies E 2 = KX · E = −1. so assume pC = 0. then F ⊥ is negative definite. OX (C − C1 )) ≥ (C 2 − 2C · C1 − KX · C + (C1 + KX · C1 ))/2 + 1 = (C 2 − 2C · C1 − KX · C − 2)/2 + 1 = 2 2 (C + i>1 C · Ci − C · C1 )/2 ≥ C /2 > 0. so v = w. By part (b). hence 2 2 2 2 2 2 2 2 2 2 ⊥ ⊥ (KX )(KX n − (n + 2) ) ≤ (KX ) C − (KX · C) KX = N . (b) Assume in addition that −KX is not nef. OX (−KX − E)). But for any ample divisor A we have i = A · (−mKX + F ) > 0 for m ≫ 0. Then E ∈ EFF(X).6). OX (−KX )) = h0 (X. since otherwise C + KX and hence E + KX is in EFF(X) by (a). we see for any elements v and w in the span of −KX and F in Cl(X) that v · w = 0. Then pC = 0. OX (C − C1 )) = h0 (X.7). since otherwise −KX − E + KX = −E is in EFF(X). (a) Apply Riemann-Roch. N 2 /2 = (N 2 − KX · N )/2 is an integer. If F is nef with −KX · F = 0. Suppose F = 0. hence in ⊥ the kernel of the mapping Cl(X) → Cl(D). Let X be obtained by blowing up 9 points p1 . Exercise I. Let D be the proper transform of D′ . generate EFF(X). (e) Let E be any class such that E 2 = KX · E = −1. (c) Suppose that some E = N + E9 + (N 2 /2)KX is not the class of a prime divisor. So for some Ci with pCi = 0 we have −KX = i Ci and this sum involves at least two summands. Solution (Exercise I. and hence there are only finitely many possibilities for N .4. together with the Ci . but N ∈ KX and KX is negative definite by the Hodge Index Theorem. (b) This follows by the Hodge Index Theorem. Then v − w ∈ A⊥ . Let C be the class of a prime divisor that is a component of E. using h2 (X. But the kernel here is the same as the kernel of KX → Cl0 (D).4. But −KX · (E + KX ) < 0. This is clear by (a) if pC > 0. Thus E is a sum of prime divisors C with pC = 0. since j = −KX · A > 0. we see that h2 (X. show that F = 0. if E has two or more components. . (c) If the points are sufficiently general and the ground field is the complex numbers. OX (−KX − E)) ≥ KX − 1 by Riemann-Roch. Now 2 h0 (X. Thus N is bounded below. Given any prime divisor C. Solution (Exercise I. Let L be the pullback of the class of a line and let Ei be the class of the blow up of pi . which is a torus which can be identified with . 0 where Cl (D) is the subgroup of divisor classes of degree 0. then one of them must be disjoint from D. we have F 2 ≥ 0 (by 2 Exercise I. Note that h2 (X.4. OX (E)) ≥ 1. . HARBOURNE that −KX · C is bounded (but we already saw that 0 ≤ −KX · C ≤ n + 2) and that there are only finitely 2 2 many possibilities for N . (a) Show that N ∈ K ⊥ implies N 2 is even.6. note that N · KX = 0. Let X be a rational surface with KX = 0. −KX − E is a sum of classes of prime divisors with pC = 0. It is again clear if C 2 ≤ 0. so E ∈ EFF(X). so (KX )2 C 2 = (KX · C)2 KX + N 2 . OX (2KX + E)) = h2 (X.8 B.4. If F 2 > 0. Since F is nef. Conclude that EFF(X) is not finitely generated. (c) Since −KX is nef but −KX · (KX − E) < 0. OX (E)) = 0.

so we may i .e. to see inf 2 = inf 1 . Note that inf 1 is equal to inf 3 = inf C2 (C·L)2 any prime C with C · L > 0 but C · i Ei = 0 is a positive multiple of L. . whence (v/a)2 ≥ ((v+w)/(a+b))2 . p1 . then Cj is a positive multiple of L. m1 + m2 m1 m2 By definition ε(P2 . ((D1 + D2 ) · L)2 : C is prime and C · L > 0 . . Suppose D satisfies the given conditions. . Since the complex numbers have infinite dimension over the rationals.e. it is enough to show that i (ai /a)2 ≥ i ((ai + bi )/(a + b))2 if i (ai /a)2 ≥ i (bi /b)2 . pr ) = inf d : dL − i mi mi Ei ∈ EFF(X). and let L and Ei be as usual.2. . i i Suppose D = dL − i mi Ei ∈ EFF(X) satisfies the conditions of the definition of ε(P2 .5. . . so 2abv 2 ≥ 2a2 v·w and thus (a+b)2 v 2 = a2 v 2 +b2 v 2 +2abv 2 ≥ a2 v 2 +a2 w2 +2a2 v·w = a2 (v+w)2 . then min i 2 Di (Di · L)2 : C is a prime divisor on X. it’s easy to choose points pi ∈ D. mi ≥ 0. ⊥ such that the map KX → Cl0 (D) is injective. . p1 . Let F be obtained by deleting every summand Cj (if any) for which Cj = Ei for 2 D2 some i. pr ) = ε(P2 . d > 0 . .9.GLOBAL ASPECTS OF THE GEOMETRY OF SURFACES 9 D. in which case C = C − E1 is the ′2 C2 class of a prime divisor with (CC·L)2 < (C·L)2 . Deleting all Ci of the form Ej reduces P dmi . Exercise I. pr ) = sup NEF(X). . But b2 v 2 ≥ a2 w2 by hypothesis. pr ). . pr ∈ P2 . . we see that inf 2 ≤ inf 3 = inf 1 . . but this change reduces (FF 2 . Show that C2 : C is a prime divisor on X and C · L > 0 inf (C · L)2 C2 Ei > 0 . Solution (Exercise I. Let X be obtained by blowing up points p1 . . .. p1 . . . since ≤ If we write D1 = aL − i ai Ei and D2 = bL − i bi Ei . ≥ min . Let X be obtained by blowing up r > 0 points p1 . so E must be prime. If for some j we have Cj · i Ei = 0. : C ∈ EFF(X). Thus with these changes we may ·L) assume each summand Cj of F satisfies the conditions imposed on D. . . Write D = i Ci as a sum of classes of prime divisors Ci . . I. . p1 .4. Write D as dL − i mi Ei . C · L > 0. . we must show (v/a)2 ≥ ((v + w)/(a + b))2 if (v/a)2 ≥ (w/b)2 . ′ Since inf 2 is an infimum over a bigger set than is inf 3 .4.8). .8. so b|v| ≥ a|w|. Let inf 1 be the first infimum in the statement above and let inf 2 be the second. p1 .4. . Thus we may assume i i ·L) that Cj = Ei for all i and j and hence that Cj · Ei ≥ 0 for all i and j and that Cj · L ≥ 0 for all j.4. the 2 latter is still the class of a prime divisor. or 1 − i (ai /a)2 ≤ 1 − i ((ai + bi )/(a + b))2 . (a) This just amounts to the easy fact that d1 d2 d1 + d2 . p1 . hence bv 2 = b|v|2 ≥ a|w||v| ≥ av·w. pr ) = inf PC·L i C·Ei (D1 + D2 )2 . C · Ei ≥ 0 for all i. By induction it is clearly enough to check that if D1 and D2 satisfy the conditions on D. so we can replace Cj by Cj − E1 . Exercise I. .. pr ). this is just (a2 − i a2 )/a2 ≤ ((a + b)2 − i (ai + bi )2 )/(a + b)2 . Writing F = dL − i m′ Ei we see m′ ≥ mi for all i. : dL − m i Ei ∈ mi > 0 . C · L > 0 and C · i ′ Ei > 0 . . . where ε′ (P2 . hence no prime divisor on X is disjoint from D. Conclude by applying Proposition I. C · = inf (C · L)2 i Solution (Exercise I. . . it is enough to see for any D that is effective with D · L > 0. . . and assume that the minimum occurs for i = 1. m d (b) Show that ε′ (P2 . .9). pr ∈ P2 . D · Ei ≥ 0 for all i and D · i Ei > 0 C2 D2 that there is a prime C with C · L > 0 and (C·L)2 ≤ (D·L)2 . hence (FF 2 ≤ (D·L)2 . (a) Show that ε(P2 . given vectors v and w in Euclidean space with non-negative entries and given positive reals a and b. We can also write D as j Cj for some prime divisors Cj . . Thus. i I.

p1 . where p1 .4. Then it suffices to show that ( i di )/( i µi ) ≥ mini {di /µi }. . hence −2 ≥ λL (X) ≥ 1 − 1/ε(P2 . . pr ) ≥ 1 − 1/(1/r) = 1 − r. . . . .4. . pr ) = 1/r by Exercise I. the only prime divisors C of negative self-intersection come from the points. p1 . Solution (Exercise I. . pr ) = −2. . . . . . pr ). . p1 . . p1 . But F · C = 0.11 tells us. .12. . Exercise I. . . so ε(P2 . p1 . pr ) ≥ 1/r always holds. .10. pr ) = 1/r which we saw above implies the points are collinear. . we have (aL − b i Ei ) · (dL − i mi Ei ) ≥ 0 for all dL − i mi Ei ∈ EFF(X). . Thus Conversely.4. Hence we may assume that D = dL − i mi Ei = i Ci . . pr ). .4. . . The solution to Exercise I. p1 . pr ) = 1/r. The infimum defining λL (X) must come from prime divisors C of negative self-intersection. . . Since F · C = 0. From the solution to Exercise I. . F = (r − 1)L − E1 − · · · − Er = (L − E1 − E2 ) + (L − E3 ) + · · · + (L − Er ) is effective and the classes of the prime components of L − E1 − E2 consist of classes Ej and L − E1 − E2 − Ej1 − · · · − Ejs . pr ) = 1/3 by Exercise I. . pr ). . then deleting that Ci from the sum also reduces P dmi .3.10. . . .10). Write Ci = di L − j mij Ej . . But this means we have 1 − r = λL (X) ≥ 1 − 1/ε(P2 . Since F ∈ NEF(X). pr ) > m/d would imply that F ′ = d′ L − m′ i Ei ∈ NEF(X) for some m′ /d′ > m/d. for any positive integers a and b such that b/a < ε(P2 . . .4. however. pr ∈ P2 are collinear. Since C = L − E1 − · · · − Er is the class of a prime divisor and F = rL − E1 − · · · − Er = (r − 1)L + C is a sum of prime divisors each of which F meets non-negatively.10 B. . But ε(P2 . p2 . . .11. p1 . Exercise I. . pj1 . Let X be the blow up of P2 at r distinct points p1 .4. Suppose F · C = 0 for some F = dL − m i Ei ∈ NEF(X) and C = aL − i mi Ei ∈ EFF(X) with d > 0 and a > 0. There is another way to look at what Exercise I. pr ∈ P2 . If for some i we have Ci · Ej = 0 for all j. p1 . pr ) ≥ ε′ (P2 . . p1 .4. . Exercise I. p1 . ε(P2 .11 shows that equality holds in Corollary I. . . . . pr ) = 1/5 and hence λL (X) = −6/4 ≥ 1 − 1/ε(P2 . Clearly λL (X) ≤ C 2 /(C · L)2 = 1 − r. Since F = rL−E1 −· · ·−Er = (L−E1 )+(L−E2 )+(L−E3 )+· · ·+(L−Er ) is always nef.4. pr ) = m/d. . . . (b) If aL − b i Ei ∈ NEF(X). For suppose ε(P2 .3. pjs are all of the points which lie on the line through p1 and p2 . .12). .4. HARBOURNE assume Ci = Ej for all i and j. p1 . . p1 . p1 . . . p1 . pr ). . . pr ∈ P2 such that the inequality in Corollary I. . Remark (on Exercise I.4. p1 . .4. p1 .3. . .6 when the points are collinear. . pr ). where each Ci is prime and i satisfies the conditions in the definition of ε(P2 . . clearly ε(P2 . . . p1 . and hence aL − b i Ei ∈ NEF(X) so b/a ≤ ε′ (P2 . pr ) = −4. . which follows by repeated application of the easy fact above. then ad ≥ b i mi whenever dL − i mi Ei ∈ EFF(X). . but the converse is not true. . pr ) = 1/r. . . . contradicting F ′ ∈ NEF(X). Solution (Exercise I. . Let X be obtained by blowing up the nine points. from the lines through pairs of points and from the conic itself. we see that F is nef. . . p1 .11 shows that λL (X) = 1 − r and ε(P2 . Since we can choose positive integers a and b such that b/a is less than (but arbitrarily close to) ε(P2 .10 we see ε(P2 . . Exercise I. p1 . . Let µi = j mij . . Here is an example where equality holds but the points are not collinear. Suppose we consider 9 points on a smooth cubic. . Thus F would be nef and we would have the contradiction that 1/r = ε(P2 . The proper transform of the cubic is clearly effective and (being prime of non-negative self-intersection) it also is nef so ε(P2 . Find a set of points p1 .11). we see that ε(P2 . suppose λL (X) = 1−r. . p1 . then there would be at most r − 1 such points. . . hence ε(P2 . .6 is strict. . . pr . . and the proper transform of the line through the three collinear points has self-intersection −2. . .4.4. by Exercise I. By just checking the possibilities we see λL (X) = −6/4 comes from C = 2L − E1 − · · · − E10 ∈ EFF(X) and we also see F = 5L − E1 − · · · − E10 = C + 3L is nef. . . if either λL (X) = 1 − r or ε(P2 . so −2 ≥ λL (X). . . . . Let X be obtained by blowing up collinear points p1 . . pr ) = 1/r if the points are collinear. . p10 on a smooth conic. pr ) = 1/r ε(P2 . and we have 1 − r = 1 − 1/ε(P2 . three of which are collinear. the result follows. p1 . . . . . . In any case.4. . . If the points were not all collinear. . pr ).10. so F would meet each of its prime components nonnegatively. . . p1 . pr ) ≥ m/d. Solution (Exercise I. . Show that and that equality holds in Corollary I. . p1 . . .4. . . pr ) ≥ 1/(r−1). . Finally. but in that case F ′ · C < 0. . . Conversely. . . Show that ε(P2 . . Consider ten points p1 . then the points p1 .11).6.

13.4. Thus D · (−KX ) ≥ 0 implies D is a sum of classes of negative self-intersection. . 2L − E1 − E2 . Show that the class of any prime divisor D with D2 < 0 is either Er . 3L−E1 −E2 −E3 . D · (Ei − Ei+1 ) ≥ 0 (so ai ≥ ai+1 for each i = 1. 3L − E1 − E2 − E3 . . . Ei − Ei+1 for i > 0 and L − E1 − E2 − E3 non-negatively if F is nef. and for each i ≥ 1. L−E1 . .4. Thus our list of classes of prime divisors of negative self-intersection is complete. 3L − E1 − E2 − E3 . L − E1 − E2 − E3 and D = −KX . .. L = (L−E1 −E2 −E3 )+((E1 −E2 )+· · ·+(Er −Er−1 )+Er )+((E2 −E3 )+· · ·+(Er −Er−1 )+Er )+ ((E3 − E4 ) + · · · + (Er − Er−1 ) + Er ) and 3L − E1 − · · · − Er−1 = −KX + Er . . . where p1 ∈ X1 is a flex of an irreducible plane cubic C. some positive multiple −mKX is effective and can be written as −mKX = M + N where M and N are effective and M 2 > 0). if r = 9. 3L − E1 − · · · − Er = −KX with −KX · F ≥ 0. and then for each i ≥ 1.e. It is easy to see that each of the classes listed is the class of a prime divisor D with D2 < 0.. so (−mKX − C)2 > 0 for m ≥ 2.4. 2L − E1 − E2 . . L − E1 . if r = 8. Ei − Ei+1 for i > 0 and L − E1 − E2 − E3 non-negatively (see [Ha6]). 3L−E1 −· · ·−Er = −KX are precisely the classes which meet Er . . Suppose D = aL − a1 E1 − · · · − ar Er is the class of a prime divisor with D2 < 0 which is not in the given list. . we see the latter generate EFF(X). 3L − E1 − · · · − Er = −KX . 3L − E1 − · · · − Er = −KX . pr are essentially distinct points. and we also see that a class F is nef if and only if it is a non-negative integer linear combination F of L. 2L − E1 − E2 .14).). .e. . or Ei − Ei+1 for 1 ≤ i < r. Let X be a rational surface such that KX = 0 but −KX ∈ NEF(X).. 3L−E1 −· · ·−Er = −KX is a sum of the listed classes of negative self-intersection (for example. then −Kr = 3(L−E1 −E2 −E3 )+2(E1 −E2 )+4(E2 −E3 )+6(E3 −E4 )+5(E4 −E5 )+4(E5 −E6 )+· · ·+2(E7 −E8 )+E8 . By Exercise I. . L−E1 −E2 −E3 is the class of the proper transform of the line tangent to C at p1 (i.. Because F meets Er . 2L − E1 − E2 . It is not hard to see the non-negative integer linear combinations of L. or L − E1 − E2 − E3 or D = −KX (if r > 9). . Xi+1 → Xi is the blow up of pi .13). .) Assume r ≥ 3. etc. . for example.. Then D meets each of the listed classes non-negatively. 3L − E1 − E2 − E3 . . . 2L − E1 − E2 . pi+1 is the point of the proper transform of C on Xi+1 infinitely near to pi . . Ei − Ei+1 for 1 ≤ i < r. . while the the class of the proper transform of C to X is 3L−E1−· · ·−Er = −KX . −KX ∈ EFF(X). . L − E1 . there is a prime divisor C such that −KX · C < 0. 2L−E1 −E2 ..e. moreover. 3L−E1 −E2 −E3 . which has negative self-intersection exactly when r > 9. and (−mKX − C)2 = (2m − 1)KX · C + (KX · C + C 2 ) = (2m − 1)KX · C + 2pC − 2.4. . hence a1 ≥ a2 ≥ · · · ≥ ar ≥ 0) and D·(L−E1 −E2 −E3 ) ≥ 0 (so a ≥ a1 +a2 +a3 ). Solution (Exercise I.GLOBAL ASPECTS OF THE GEOMETRY OF SURFACES 11 2 Exercise I.. Show that −KX is big (i.. . Solution (Exercise I. . each of which it meets non-negatively.6. and show that a divisor class F is nef if and only if −KX · F ≥ 0 and F is a non-negative integer linear combination of L. D · Er ≥ 0 (so ar ≥ 0). Thus −mKX = (−mKX − C) + C is big for m ≥ 2. Since each of L.4. Since −KX is not nef. Conclude that EFF(X) is generated by Er . so D2 ≥ 0.14. then −Kr = 3(L−E1 −E2 −E3 )+2(E1 −E2 )+4(E2 −E3 )+6(E3 −E4 )+5(E4 −E5 )+4(E5 −E6 )+· · ·+(E8 −E9 ). (Thus p1 . . Let X = Xr+1 be the rational surface such that X1 = P2 . L − E1 . r − 1. L − E1 . 3L − E1 − · · · − Er = −KX is a nonnegative integer linear combination of the listed classes of negative self-intersection. But each of L. L − E1 . . the flex line).. this also shows that any nef class F is a non-negative integer linear combination of L. Exercise I. 3L − E1 − E2 − E3 . i. Thus −KX − C is effective hence so is −mKX − C for m ≥ 1. .

. Con√ 2 versely. Abnormal Curves. Definition II.3) is actually a minimum (see Exercises I. . it might be worthwhile to define and study a class of effective divisors C with C 2 < 0 which are so bad as to form a finite set. Let X be a surface obtained by blowing up a finite set of points p1 . pr ) are known when X is obtained by blowing up r generic points of P2 if either r ≤ 9 or r is a square. . One application of the concept of abnormality is to computing ε(P2 .1. . pr ) when r > 9 is not a square. HARBOURNE II. . .2. . pr ) < 1/ r. .1.3. pr ∈ P2 . Let X be obtained by blowing up r points of P2 .4. p1 . X has at most finitely many prime divisors which are abnormal curves (see Exercise II.1.2). the conjecture is equivalent to: Conjecture II. The fact that d/ i mi < 1/ r is just the definition of abnormality. . p1 .1. Then √ ε(P2 . . As an intermediate step. II. where t = ⌊ r⌋. . Nagata proved this when r is a square [N2]. pr ∈ P2 . . . . . possibly (for all anyone knows) even with C 2 arbitrarily negative. since after removing all such summands we still have an abnormal curve. . which implies (and in fact is equivalent to) ε(P2 . The first statement follows from Theorem II. Working formally (i.1 and Problem I. Write C = i Ci as a sum of prime divisors Ci . . then X has no abnormal curves. The conjecture is still open. . . If X is obtained by blowing up r > 9 generic points of P2 . then ε(P2 . Doing so turns out to have useful applications to computing ε(P2 . . √ By [HR3.7). p1 . Lecture: Abnormality II. . If X has an abnormal prime divisor. We may assume no summand is of the form Ej . pr ) = 1/ r. Let C = dL − i mi Ei ∈ EFF(X) and assume mi ≥ 0 for all i with mi > 0 for some i. p1 . . .5 is true when m ≤ 42. For example. so C = ij C i · C j < 0 hence C i · C j < 0 for some i and j.1. Corollary II.9(a) and II. . The second follows from the fact that there are only finitely many abnormal prime divisors and hence the infimum in the definition of ε(P2 . pr ) = d/ i mi < 1/ r for some abnormal prime divisor C = dL − i mi Ei . . we say C is abnormal √ √ 2 if C < 0.2). . The values of ε(P2 . . pr ). p1 . p1 . If there are abnormal . Let X be a surface obtained by blowing up a finite set of points p1 . . If dL − m(E1 + · · · + Er ) ∈ EFF(X) when X is obtained by blowing up r > 9 generic √ points of P2 . .2. Then there is a class C = dL − i mi Ei ∈ EFF(X) with √ 2 d/ i mi < 1/ r and hence C < 0.4 (Nagata [N2]). .1). . One of the difficulties in studying Conjecture I. p1 . . . . and also to d/(rm) < 1/ r.3. prime divisors on X. in Cl(X) ⊗Z Q). Thus 2 every summand Cj is of the form dj L − i mij Ei with mij ≥ 0. . It is an open problem to compute ε(P2 . . . Corollary 4. . but X can have infinitely many prime C with C 2 < 0 (see Exercise I.3. . . This is equivalent to d/ i mi < 1/ r. But if neither Ci nor Cj were abnormal. Following Nagata [N1].2 is the possibility of there being infinitely many prime divisors C with C 2 < 0. . Dumnicki shows Conjecture II. p1 . pr ) < 1/ r if and only if X has an abnormal prime divisor. . p1 . .1. . pr√as given ) in (I. pr ) is the supremum of 1/t over all t such that tL − i Ei ∈ NEF(X). then ε(P√ p1 . this is true when m ≤ t(t − 3)/2. . pr ): Theorem II. .4. p1 . . There √ is a long-standing conjecture. although it is known in various special cases. If √ 2 there are no prime divisors on X which are abnormal. . p1 . √ Proof. . then it is easy to see that C i · C j ≥ 0.3. where m = ( i mi )/r. Proof. . assume ε(P .3. . . . let C = dL − m i Ei . however. pr ) < 1/ r follows by definition of ε. .1. pr ) = 1/ r for r > 9: Conjecture II. Recall that ε(P2 . Now C = i C i .2. . in fact. .. .1.e.12 B.2. then ε(P2 . In addition. Consider a surface X obtained by blowing up a finite set of points p1 .1]. then d > m r. pr ∈ P2 . .1. . . We note that not every curve C with C 2 < 0 is abnormal (see Exercise II.5. A Dual Problem. There is a dual notion which Chudnovsky [Ch] attributes to Waldschmidt [W].3. .

. p1 . p1 . . that 0 ≤ v 2 − v 2 .3. . ar are not all equal. Then dn ≤ d(a+1)(m!) ≤ (a + 1)dm! so l≤ dn (a + 1)dm! adm! dm! dm! dm! dn! ≤ ≤ = + ≤ + ≤ l + δ/2 + dm! /(a(m!)).4) is also equivalent to γ(P2 . . If a1 . Then γ(P2 .1. .3. . . . if the points p1 . but v 2 = v · v. . p1 . . show that C 1 · C 2 < 0. (b) If a = i ai /r and b = i bi /r. Let X be obtained by blowing up r points pi ∈ P2 .e.3. . where ra = i ai . . But rab = a i bi . . (a) Let C = dL− r mi Ei .3. . m). .2. . p1 . . .e. pr ) = inf d : dL − m m i Ei ∈ EFF(X) [Ch] remarks).2. pr ) = r for r > 9 generic points of P2 . (d) Give an example of a curve C with C 2 < 0 but such that C is not abnormal. . so 0 ≤ (v − v)2 = v 2 + v 2 − 2v · v = v 2 − v 2 . . Moreover. such that i (a′ − a)2 = 0 and hence a = a′ for all 1 ≤ i ≤ r and i i i so i ai bi ≥ i a′ bi = rab.3). rε(P2 . p1 . Say n ≥ m! and write m! n = a(m!) + c. pr ) = γ(P2 . . γ(P2 . we check for all m! n ≫ 0 that l ≤ dn /n ≤ l + δ. . . although in general rε(P2 . . Solution (Exercise II. pr ) ≤ r. Exercises. .e. . . p1 . p1 . . (b) If C1 = a0 L − i ai and C2 = b0 L − i bi . . Clearly. drm ≤ rdm . . . where a1 ≥ a2 ≥ · · · ≥ ar ≥ 0 and b1 ≥ b2 ≥ · · · ≥ br ≥ 0. p1 . i. . p1 . . pr ) (see Exercise II. . Thus we have: Corollary II. .3. 1 ≤ i ≤ r. . Chudnovsky [Ch] also remarks that γ(P2 . Pick m large enough that l ≤ dm! ≤ l + δ/2. . . . with respect i to the Euclidean dot product. Given any δ > 0. mr ) and let v = (m. . . . . . . By repeating this procedure we i i i i eventually obtain a sequence a′ . i i=1 i. . .. . This is equivalent to showing i (rai )bi ≥ (ra) i bi . we can reduce to the case that a is an integer. i. (c) If C1 and C2 are abnormal. pr ) ≤ γ(P2 . pr ) ≤ √ r As Chudnovsky [Ch] points out. Exercise II. pr ) ≤ dn dn! n! ≤ n . pr ) = lim where dm is the least t such that tL − m m→∞ dm m Ei ∈ EFF(X). n! n a(m!) + c a(m!) + c a(m!) + c m! a(m!) and for n ≫ 0 we will have a large enough such that dm! /(a(m!)) ≤ δ/2. Let X be obtained by blowing up r distinct points pi ∈ P2 . II. 2 (a) Show that C 2 ≤ C for any divisor C on X. . . . . . p1 . . . pr ) ≤ Proof. so limm→∞ dm! exists. Also. we have rε(P2 . . i . √ hence γ(P2 . pr ) ≤ d/m for all d/m > r. it suffices to show that i ai bi ≥ rab. Thus dm! /(m!) is a non-increasing sequence. p1 . pr ) is actually a limit: Proposition II.4). . p1 . so drm /(rm) ≤ dm /m. . call it l.3. pr ) (Exercise II. we can pick some j such that aj > a and some l such that a > al . Let v = (m1 . . √ Thus Nagata’s conjecture (Conjecture II.GLOBAL ASPECTS OF THE GEOMETRY OF SURFACES 13 Definition II. . l. where c is an integer with 0 ≤ c < m!. . . Thus we need to show that d2 − i m2 ≤ d2 −rm2 . p1 . . · · · .. Let a′ = aj − 1 and a′ = al + 1.. p1 . we have i dn . . n γ(P2 . dm! /(m!) ≤ dn /n for every n|m!. pr ) < γ(P2 . Then j i l (ai − a)2 > i (a′ − a)2 and i ai bi = (bj − bl ) + i a′ bi ≥ i a′ bi . for each n ≥ 1. γ(P2 . p1 . pr are generic we have rε(P2 . Then we need to show. . . . p1 . . . . pr ).1. that i m2 ≥ rm2 . .1). . so we need only show i ai bi ≥ a i bi . .2. we see in that case for D = dL − m i√ i that sD ∈ EFF(X) for s ≫ 0. since d/m > √ Clearly (as Chudnovsky 2 E r implies (dL − m i Ei ) > 0.1. and a′ = ai for i = j. show that C1 · C2 ≤ C 1 · C 2 . .2. as required. We also now see γ(P2 . . . Therefore. pr ) ≤ γ(P2 .

1 2 r i i Given two abnormal prime divisors. r 1 √ Thus ε(P2 . p1 . p1 . suppose there were more.3. Thus. pr ) = 5/3. take C = 3L−2E1 −E2 −· · ·−E7 to see that ε(P2 . . there are at most r + 1 of them [Sz].. but by Exercise II. . p1 . take C = 6L−3E1 −2E2 −· · ·−2E8 to see that ε(P2 . . pr ) = γ(P2 . where the m′ are obtained by a permutation of the mi such that m′ ≥ m′ ≥ · · · ≥ m′ . p1 . Then C > 0 but C 2 = −1. . pr ) ≤ ε(P .4. More generally. P is abnormal. Thus there must be two prime divisors C1 = C2 with φ(C1 ) = φ(C2 ) if S is infinite. . . pr ) for each r ≤ 9 and each r which is a perfect square. pr ). . For a more interesting example.3. p1 . . p1 . p2 . I.3. Since by ′ Theorem I. p1 .1. . . For the case of r ≤ 9 generic points. p3 ) = 1/2. Then F = 3L − E1 − E2 − E3 − E4 = C+L+(L−E4 ) and H = 3L−E1 −E2 −E3 −2E4 = C+2(L−E4 ) are nef and C = 5L−3E1 −3E2 −3E3 −3E4 = 2C + L14 + L24 + L34 ∈ EFF(X). . and for r = 8. since C = L − E1 − E2 is abnormal for r = 2 or 3. To see that there are at most r + 1.4. p2 . Exercise II. Thus rC ∈ EFF(X) i i for any C = aL − a1 E1 − · · · − ar Er ∈ EFF(X). . . . the prime divisors E with E 2 = E · KX = −1) by Exercise I. p′ ) = 1/d by Remark I. .3. p1 . .7 for any divisor tL′ − i Ei ∈ NEF(X ′ ) we also have tL − i Ei ∈ NEF(X) we see that 2 ′ ′ 2 ε(P .e. Let X be obtained by blowing up r points pi ∈ P2 . Since H · C = 0. Suppose there were an infinite set S of them.3. In fact. a permutation π such that if C = dL− i mi Ei .3). . . for 1 ≤ i ≤ t for t > r + 1. we see γ(P2 . we thus see that the permuted divisors C ′ and D′ have C ′ · D′ < 0. . . exactly three of which (say p1 . p4 . . . there is a relation i mi Ci = 0 where the Ci are distinct. . Then there are at most finitely many prime divisors C which are abnormal. Solution (Exercise II. Solution (Exercise II. .1(b.2 gives rε(P2 . p1 . . . take C = 2L − E1 − · · · − E5 to see ε(P2 . . p1 .2. Let X be obtained by blowing up r = d generic points pi ∈ P . HARBOURNE √ √ (c) Let C 1 = aL − b i Ei and let C 2 = cL − d i Ei . then there are infinitely many exceptional curves (i. For r = 7. c) we have C1 · C2 < 0. but P 2 = P · N ≥ 0 which is a contradiction. . so if X has any abnormal curve. For r = 5 or 6. .10 that ε(P2 .mi >0 mi Ci and let N = − i. Let C = L − E1 − E2 − E3 and let Lij = L − Ei − Ej . we see ε(P2 . p′ .3. . pr ). . one can use quadratic transforms to see that 3L − 2E1 − E2 − · · · − E7 and 6L − 3E1 − 2E2 − · · · − 2E8 are smooth rational curves. The proper transform of C ′ is C = 4L − 3E1 − E2 − · · · − E9 .mi <0 mi Ci . . say Ci . . up to permutations. .14 B. . . if you blow up 9 or more general enough points of P2 . pr ). . . . and since F · C = 0. . . . p1 . we see by Exercise I. .3. that curve gives the value of ε(P2 . . . if C = tL − i mi Ei is the class of an effective divisor. . p1 . On the other hand. ε(P2 . Consider four points p1 . . pr ) = 1/3. for each C ∈ S. Give an example such that rε(P2 . pr ) = 1/ r when r is a perfect square. p7 ) = 6/17. But rC = raL − (a1 + · · · + ar )(E1 + · · · + Er ) and ar/(r(a1 + · · · + ar )) = a/(a1 + · · · + ar ). .2.1. . p1 . . hence P 2 < 0. . .4. . . then mπ(1) ≥ mπ(2) ≥ · · · ≥ mπ(r) . then so is C ′ = tL − i m′ Ei . . .4. Solution (Exercise II. p1 . . then so is C ′ = tL − i m′ Ei . hence the class C ′ = dL′ − i Ei of the i 2 2 proper transform of C is nef. . We get a mapping φ : S → Sr by choosing. . (For the fact that 3L − 2E1 − E2 − · · · − E7 and 6L − 3E1 − 2E2 − · · · − 2E8 are classes of prime divisors. pr ). .3. Now. . p3 ) are collinear. . Let P = i. Let X be obtained by blowing up r generic points pi ∈ P2 . C and D. . pr ) ≤ 1/ r by Corollary II.2). (a) Suppose r = d2 is a perfect square. C and D are the same. pr ) = γ(P2 . . . Exercise II. pr ) ≥ γ(P2 . Then a/(rb) < 1/ r and c/(rd) < 1/ r. Since Cl(X) has rank r + 1.3. . choose an irreducible quartic plane curve C ′ with a triple point. . use Exercise I. . . . so ac/(rbd) < 1 so C 1 · C 2 < 0. and hence C ′ = D′ . . pr ).e. p2 ) = ε(P2 . .3.2 at most finitely many of them are abnormal. This together with Corollary II. pr ). Then P − N = 0 hence P = N . . .3. . p7 ) = 3/8. . . . (a) Compute ε(P2 . p5 ) = 2/5. Blow up the triple point and eight additional points 2 on C ′ . p1 . p1 .4).4. where the m′ are obtained by any permutation of the mi . p1 .7 over the complex numbers. Let X ′ be obtained by blowing up ′ r = d2 points p′ ∈ P2 on a smooth plane curve C of degree d. . p1 . . so rε(P2 . but C1 = C2 implies 0 ≤ C1 · C2 . .7(c). . Exercise II.) (b) Since the points are generic. if C = tL− i mi Ei is the class of an abnormal prime divisor. p1 . which is impossible.. Hence by Exercise II. . p1 . p1 . (b) Show that rε(P2 . . But C ′ ∈ NEF(X ′ ) ∩ EFF(X ′ ) together with √ ′ )2 = 0 implies that (C ε(P2 . More generally. . pr ) < γ(P2 . (d) An easy example is given by L − E1 − E1 − 0E3 − 0E4 .

then h0 = r and hi = r + 1 are best possible.3. If one can show that none of these finitely many classes is the class of a prime divisor. as we show below.1. Fundamentally it depends on the fact that if F = tL − m i Ei has F 2 > 0. equivalent to showing the existence of abnormal curves. [ST]. . . Proof.3]. contradicting there being at most r + 1 abnormal prime divisors. Then there is an explicitly computable finite set SF of classes which contains the class of every prime divisor C with C · F < 0 (if any). . Let X be obtained by blowing up distinct points p1 . . Since the classes Hi generate Cl(X). so r ≥ 2r − 2. . For purposes of exposition we will continue to focus on the case of P2 . and hence that ε(P2 .2 there are at most r + 1 prime divisors C = dL − i mi Ei which are abnormal. .1. Getting an upper bound less than 1/ r is. (It is clear that h0 = r and hi = r + 1 will suffice. Proposition III. . This is the method used by [Bi]. When the r points pi are general. . Both authors first construct a nef divisor F ′ = d′ L − i mi Ei . Let E = i Ei . but since the points are general any permutation of the mi is again an abnormal prime divisor. and choose nef divisors Hi that span Cl(X). p1 .4. Since C = 2L − E1 − · · · − E5 ∈ EFF(X) has F · C = 0. Lemma III. . . Thus much of the focus has been on getting increasingly better lower bounds. p1 . with L and Ei as usual.1. Given distinct points p√∈ P2 we now consider the problem of i estimating ε(P2 . with L and Ei as usual. then sF − C and C are both in EFF(X). Thus there are only finitely many possible classes C with 0 ≤ C · Hi ≤ sF · Hi for all i. see [HR1] and [HR2].1. Proof. By Exercise II. hence both meet every Hi non-negatively so 0 ≤ C · Hi ≤ sF · Hi for each i. where 1 < j < i. .) The restrictions on possible abnormal prime divisors can be made even more stringent. Example III. Being able to choose smaller values of the hi will depend on having some knowledge of how the points pi are arranged. one can narrow down the set SF even more [Sz]. pr ). then we know m/d ≤ ε(P2 . [SS]. p6 ) ≥ 2/5. if for two classes C1 and C2 we have C1 · Hi = C2 · Hi for all i. Since F 2 > 0 and F · L > 0. The only other possibility to be ruled out is if the two values each occur at least twice. Assume that C = dL − i mi Ei is abnormal. [T]. pr ∈ P2 . [Ha5] and [Ha4]. . There are r − i − 1 more transpositions αj of m1 with mj . The first method is to explicitly construct nef divisors. Thus at most two values can occur among the mi . Lemma 2. then C1 = C2 . . . since r ≥ 4. This method has been applied by [X]. . then. Suppose we blow up six general points p1 . The number of distinct arrangements of the mi is r . Both methods can be adapted to studying Seshadri constants on surfaces in general. j j hence 2r − 2 > r + 1. p1 . or choose hi > 0 large enough such that H0 = h0 L−E and Hi = hi L−E−Ei are in EFF(X)∩NEF(X). but more than two spots from either end. For example. H0 = L. Estimating Seshadri Constants. pr ). . we can find an explicit s such that sF ∈ EFF(X) (but the smaller s is the smaller SF will be). then all but at most one of the coefficients mi are equal. then F is nef and m/t ≤ ε(P2 . Suppose that there is an index i such that m1 > mi > mr . by Corollary II. . We may assume that m1 ≥ · · · ≥ mr ≥ 0.3. Looking at Pascal’s triangle it is clear that r > r + 1. (Since the entries in j j the triangle we’re interested in are on the row beginning 1 r · · · . . . . and Hi = L − Ei for i > 0. where i < j < r. there are six permutations in which we permute m1 . where r ≥ 4 and 2 ≤ j ≤ r − 2. p6 . The second main method is to rule out the possible occurrence of abnormal curves.3. . and then use an averaging process to get a nef divisor of the form F = dL − m i Ei . .10. Lecture: Computation of Seshadri Constants III. this shows ε(P2 . There have been two main methods used for this. . We will use the method of ruling out abnormal curves to check that F = 5L − 2 i Ei is nef. . if one shows some divisor F = dL − m i Ei is nef. Let X be obtained by blowing up general points p1 .) If C is the class of a prime divisor with F · C < 0. p6 ) = 2/5 by Exercise I. there are only finitely many classes C = dL − i mi Ei that could possibly be the class of a prime divisor with F · C < 0 [HR1.1.1. . pr ∈ P2 .GLOBAL ASPECTS OF THE GEOMETRY OF SURFACES 15 III.2. So assume that m1 = · · · = mj > mj+1 = · · · = mr . In addition. we see r is the sum of two entries on the row above it. p1 .1. . . For example. each entry being at least r − 1. but the smaller one can choose the hi the smaller SF will be. pr ). . . This gives (i − 2) + (r − i − 1) + 6 = r + 3 distinct permutations. Assume that F = tL − m i Ei has F 2 > 0 and t > 0. [HR1] and [HR2]. Then there are i − 2 permutations ωj which are transpositions of mr with mj . . mi and mr with each other only. and this is often quite hard. . since if the points are collinear. p1 . . .

. . So ′ F ∈ NEF(Xn+1 ).4. The irreducible components of this sum are Cr+1 and Ei − Ei+1 for various i. for m = 2 we get d = 5 (which also fails 5d − 12m < 0). Thus F is nef. k) = (5. since each summand is the class of a prime divisor which H0 meets non-negatively. choose p′ to be infinitely 1 r i i near to p′ but choose p′ r+1 not to be on Cr+1 . 3) or (3. the proper transform of C1 . so 5d − 12m − 2k = C · F < 0. 3. . . it is enough to find essentially distinct points p′ of X1 = P2 such i ′ ′ ′ that ndL′ − r(E1 + · · · + En ) ∈ NEF(Xn+1 ). but C = 5L − 3(E2 + · · · + E6 ) for (d. . For m = 1 we get d = 3 (which fails 5d − 12m = F · C < 0).1. C − E is effective if C is. pn . n be positive integers such that r < d n and r ≤ n.2 we may (after reindexing. None of these are effective. 3. and likewise so are C − 2E and C − 3E. This defines p′ . . 1. E = 2L − E2 − · · · − E6 is a prime divisor. Thus 7m + 2k ≤ 3d ≤ 6m + k + 3.. based on the idea of unloading [R]. 1). p1 . −3). For example.1. and for m = 3 we get d = 7. We have 0 ≤ C · H0 = 3d − 6m − k ≤ F · H0 = 3 and 0 ≤ C · H1 = 3d − 7m − 2k ≤ F · H1 = 1. C = 2L−3E1 (which is not in EFF(X)). if need be) assume that C = dL − m i Ei − kE1 for some k. nd Proof. we have r ε(P2 .1.6(b). hence the class of a prime ′ ′ ′ ′ ′ ′ divisor. hence F − C should also be effective. 0 ≤ C · H1 = 3d − 7m ≤ F · H1 = 1. Let d. So suppose C = dL − m i Ei − kE1 for some k > 0 so 5d − 12m − 2k = C · F < 0. 1 ≤ −k ≤ m ≤ 3 and 5d − 12m − 2k < 0 are (d. Then we have 0 ≤ C · H0 = 3d − 6m ≤ F · H0 = 3. ′ ′ ′ ′ ′ ′ ′ and hence so is E1 + · · · + Er − rEi for every i > r. But since C is supposed to be a prime divisor it should satisfy adjunction and thus must have −2 ≤ C 2 + C · KX . and F · Cr+1 ≥ 0 since nd2 − r2 > 0). for r < i ≤ n. If n > r. . The same argument handles the other two cases. as claimed. as required. Hi = H0 − Ei ∈ NEF(X). r As another variation we have: . so d ≤ 5. Suppose C is the class of a prime divisor such that 0 > C · F . √ Proposition III. but H5 meets each summand non-negatively. and let p′ be the i i+1 point of Ci+1 infinitely near to p′ . In particular Ei − Ej ∈ EFF(Xn+1 ) for every j > i. Thus 7m ≤ 3d ≤ 6m + 3.1. we have d ε(P2 . each of which is the class of a prime divisor. r. For example. . HARBOURNE First. so m + k ≤ 3. n be positive integers such that n ≥ r > d n. 3. −2). Let p′ ∈ C1 . .e. k) is either (1. but for d = 7 with m = 3 we find C 2 + C · KX = −8. It is enough to show that ndL − r(E1 + · · · + En ) ∈ NEF(X). By Theorem I. since E · C < 0. m. (2. . pn ) ≥ . −3). . . (6. and C = 3L−2E1 −E2 −· · ·−E6 (which also fails F · C < 0). note that H0 = 3L − i Ei = (2L − E1 − · · · − E4 ) + (L − E5 − E6 ) is nef. but C − 3E = −L is not effective. m. −1)}. we see F ∈ EFF(X). giving C = L−E1 (which fails F ·C < 0). let Ci+1 be the proper transform of Ci . hence −k ≤ m. Choose any smooth plane curve C1 of degree d. Thus (d. 0. whence 5m ≤ 3d ≤ 15 implies 1 ≤ −k ≤ m ≤ 3. where X is the blow up of P2 at general points p1 . Since ndL − n(E1 + · · · + Er ) ∈ EFF(Xn+1 ).5. 2. Then for n general points pi ∈ P2 . assume C = dL − m i Ei − kE1 for some k < 0. Finally. r. Then for n general points pi . First suppose k = 0. H5 = (2L − E1 − · · · − E5 ) + (L − E5 − E6 ). p′ . . . p1 . . i−1 ′ ′ Thus dL′ − E1 − · · · − Er is the class of Cr+1 . 1). 1 ′ Recursively. k) ∈ {(5. ′ ′ but F meets each one non-negatively (this is clear for Ei − Ei+1 . hence neither is C. let Xi+1 be the blow up of Xi′ at p′ . Then C is abnormal and by Lemma III. we have C · Ei ≥ 0 for all i. (4. Also. pn ) ≥ . so m ≤ 3. Let d. We have 0 ≤ C · H0 = 3d − 6m − k ≤ F · H0 = 3 and 0 ≤ C · H1 = 3d − 7m − 2k ≤ F · H1 = 1. . Since F = (2 − E1 − · · · − E5 ) + (2 − E2 − · · · − E6 ) + (L − E1 − E6 ). we see ′ ′ ′ ′ F = ndL′ − r(E1 + · · · + En ) = (ndL′ − n(E1 + · · · + Er )) + r<i≤n ′ ′ ′ ′ (E1 + · · · + Er − rEi ) ∈ EFF(Xn+1 ) ′ ′ (this is the unloading step). i. m.16 B. 0. The simultaneous solutions to 7m + 2k ≤ 3d ≤ 6m + k + 3. We now give an example of the alternative approach using the method of [Ha5] and [Ha4]. Thus 7m + 2k ≤ 3d ≤ 6m + k + 3 and hence also 5m ≤ 7m + 2k ≤ 3d. as are Ei − Ei+1 for each 1 ≤ i < n and En . Since C should be the class of a prime divisor with C · L > 0. √ Proposition III.

It is enough by the semicontinuity ′ ′ ′ principle to find essentially distinct points p′ of X1 = P2 such that rdL′ − d2 (E1 + · · · + En ) ∈ NEF(Xn+1 ). √ Then 5 > 2 6. pr ) for every choice of r < 9 distinct points of P2 . Compute ε(P2 .3. as are Ei − Ei+1 for each 1 ≤ i < n and En . . which meet D positively). Example III.2.3. i. p1 . This can be done using the various possibilities (worked out in [GHM]) for EFF(X) where X is the blow up of P2 at the r points. . But F · Cr+1 = 0.4. by adding to rdL′ − r(E1 + · · · + Er ) the classes ′ ′ ′ mn En and mij (Ei − Ej ) with i ≤ r and j > r for appropriate choices of mij ≥ 0 (this is the unloading step). ε(P2 . so clearly 45L − 9(E1 + · · · + E21 ) = 9F (and even 43L − 9(E1 + · · · + E21 ). let Ci+1 be the proper transform of Ci . . Recursively. .1. Show F = 5L − E1 − · · · − E21 is ample. pr ) and γ(P2 . . for r < i ≤ n. but it is known that except for a few exceptional cases which do not occur here that general points of small multiplicity impose independent conditions on curves of degree d.1. so by Proposition III. . This i i+1 i defines p′ .2.1.e. (How big “small” is keeps increasing as more research is done. . By Proposition III. we have d ε(P2 .5 with r = 7 and d = 2.) √ Exercise III. . It is nef. for that matter) is ample by the Nakai-Moiseson criterion [Hr] since F 2 > 0 and F meets every curve positively (any prime divisor orthogonal to D must meet D + L positively. Exercise III. However. .4 that in fact ε(P2 . using Proposition III.1). . 5 III. Some of these cases are discussed in [Ch].. Exercise III. Since ′ ′ ′ ′ 2 2 ′ ′ rdL − r(E1 + · · · + Er ) ∈ EFF(Xn+1 ) and since r > nd . Solution (Exercise III. . see [Du]. r. . but certainly multiplicity at most 3 is covered by the results. p′ .4).2. using r = 9 and d = 2. p6 . Let d.2. p6 ) = 2 . we see that 9 .3). p1 . . if there are curves of degree d passing through the points with the specified multiplicities. p1 .1. we see that ε(P2 . as required. . ′ so F ∈ NEF(Xn+1 ). p1 . using the method of Example III. .1. Solution (Exercise III. Solution (Exercise III. .2.6. since the only prime divisors orthogonal to L are the Ei .2).1. See Exercise III. .4. Then for n general points pi ∈ P2 . pn ) ≥ .3 one is left with showing that in none of the following cases is C = dL − (m + k)E1 − m(E2 + · · · + E12 ) an abnormal prime divisor: d= 7 m= 2 k= 1 d= 10 m= 3 k= 0 d= 3 m= 1 k= -1 This is clear for the last case. Let X be the blow up of 12 general points. . Exercise III. . F · (Ei − Ej ) = 0 and F · Er > 0. choose p′ to be infinitely near to p′ but choose p′ 1 r r+1 not to be i i−1 on Cr+1 . r Solution (Exercise III. .1. . p6 ) ≥ 2 and hence that F = 5L−2 i Ei ∈ 5 NEF(X).3. . .GLOBAL ASPECTS OF THE GEOMETRY OF SURFACES 17 Proof.2.2.2.3.2. and let p′ be the point of Ci+1 infinitely near to p′ . . . hence the class of a prime ′ ′ ′ ′ ′ divisor. p21 ) ≥ 42 Thus D = 42L − 9(E1 + · · · + E21 ) ∈ NEF(X). Study whether F = 7L − 2(E1 + · · · + E12 ) is nef. . Since C = 2L − E1 − · · · − E5 ∈ EFF(X) and F · C = 0. . Exercises.5. ′ ′ Thus dL′ − E1 − · · · − Er is the class of Cr+1 . Mimic the proof of Proposition III. we see by Remark I. when the Ei are obtained by blowing up 21 general points pi of P2 . using the method of Example III. The other two cases are harder to eliminate. p1 . Let p1 ∈ C1 . Again suppose we blow up six general points p1 .2. . i ′ ′ Choose any smooth plane curve C1 of degree d.4. If n > r.2. ′ ′ ′ ′ ′ we obtain F = rdL′ − d2 (E1 + · · · + En ) with F 2 > 0. p1 . . n be positive integers such that n ≥ r > d n. . . the proper transform of C1 .1. . . let X be the surface we obtain. let Xi+1 be the blow up of Xi′ at p′ . In particular Ei − Ej ∈ EFF(X) for every j > i. since 3L − E1 − · · · − E11 is not effective.

OX (−mi Ei )(π −1 (U )) ↓ OX (−mi Ei )(π −1 (U ) \ Ei ) . Let I(pi ) ⊂ R be the ideal generated by all forms vanishing at pi . First. .e. hence OP2 = π∗ OX → π∗ OX (−mi Ei ). Now. IZ (t)) ∼ H 0 (X. hence is an isomorphism itself. OX (tH − i ai Ei )) and so I(Z) can be identified = with ⊕t≥0 H 0 (X. let H be the pullback to X of a general hyperplane and let Ei be the blow up of pi .1. so A = B and thus π∗ OX = OPn . The bottom arrow is also an isomorphism (since U \ {pi } ∼ π −1 (U ) \ {π −1 (pi )} = π −1 (U ) \ Ei ). i i so IZ (t) ∼ π∗ (OX (tH − i mi Ei )) follows if we show that IZ ∼ π∗ (OX (− i mi Ei )). We will denote the sheaf of ideals corresponding to I(Z) by IZ .e. OX (tH − = ai Ei )) holds for all q ≥ 0.e.) m Now we show that π∗ OX (−mi Ei ) is either OPn (if mi ≤ 0) or Ipi i (if mi > 0).5. If mi ≤ 0. hence the other arrows are isomorphisms = too. then we have a morphism OX → OX (−mi Ei ). we have a natural isomorphism π∗ (OX (tH − mi Ei )) ∼ OPn (t) ⊗ π∗ (OX (− = mi Ei )).16(c)]. xn ] be the homogeneous coordinate ring of Pn . since then blowing up has no effect. π∗ (π −1 Ipi i ) mi −1 mi is the sheaf associated to the presheaf U → Ipi (U ). hence normal and B being module finite over A since π is projective and OX and hence π∗ OX are coherent [Hr. By the projection formula ([Hr.20] (see also [Hr. We start by noting that π∗ OX = OPn . Proposition II. Background. I. Exercise III. Exercise I. notationally suppressing its dependence on the mi . Let p1 . The notions we’ve discussed above can be applied to questions of commutative algebra. So assume n > 1. hence I(Z) = ⊕t≥0 H 0 (Pn . let I(Z) be the homogeneous ideal ∩i I(pi )mi . This is a saturated ideal which defines a 0-dimensional subscheme of Pn .3. Let Z be the fat point scheme n mi ≥0 mi pi . where IZ (t) = IZ ⊗OPn OPn (t). Given distinct points pi ∈ Pn and integers mi . then H q (Pn . . Let pi ∈ U be an affine open neighborhood.7. i. on any affine open of Pn . Proposition II. This is clearly an isomorphism except possibly at the point pi . the canonical sheaf morphism mi Ipi → π∗ OX (−mi Ei ) comes from an isomorphism of presheaves.4]: since π is projective by [Hr. Then there is a natural isomorphism IZ (t) ∼ = π∗ (OX (tH − i mi Ei )) such that H 0 (Pn . . Let R = k[Pn ] = k[x0 ..1. If m ≥ 0 and π is the blow up of a single point p ∈ Pn where we set E = −1 m m m m π (p).3A]) or by [Hr. π∗ OX is coherent. i m Proof. consider the canonical morphism Ipi i → π∗ (π −1 Ipi i · OX ) = π∗ OX (−mi Ei ). HARBOURNE IV. then we have a natural morphism Ip → π∗ (π −1 Ip ) which induces a morphism Ip → π∗ (π −1 Ip · OX ) = π∗ OX (−mE) and thus m IZ = Πmi ≥0 Ipi i → Πmi ≥0 π∗ OX (−mi E) = π∗ OX ( mi ≥0 −mi E) ֒→ π∗ OX ( i −mi E). = m m m If mi > 0.20]. IZ = Πmi ≥0 Ipi i .5. which we refer to as a fat point subscheme. whence π∗ OX (−mi Ei ) ∼ OPn . We will abuse notation and use the 0-cycle Z = m1 p1 + · · · + ms ps to denote this subscheme. Exercise II. Lecture: The Containment Problem (an application to Commutative Algebra) IV.1(d)]). Given a 0-cycle Z = m1 p1 + · · · + ms ps (i. . .6. (See the argument of [Hr. with A being integrally closed since Pn is smooth. if mi ≥ 0 for all i.5] and the right vertical arrow is injective since X is integral. an element in the free abelian group on the points pi ) with mi ≥ 0 for all i. For convenience we write L for OX (− i mi Ei ).3. B is an integral extension of the integrally closed ring A. let π : X → P be the morphism obtained by blowing up the points pi . Thus π∗ OX is locally a sheaf of finitely generated OPn modules. Moreover. Since π is birational. IZ (t)). = = This is trivial if n = 1. the ring B given by π∗ OX and the ring A given by OPn both have the same function field.18 B. ps ∈ Pn be distinct points.. . In fact. . hence π∗ (π Ipi · OX ) is the sheaf associated to the m m m presheaf U → Ipi i (U ) · OX (π −1 (U )) = Ipi i (U ) · OP2 (U ) = Ipi i (U ). Corollary II. more is true: Proposition IV. with the same function field.1. IZ (t)) ∼ H q (X. especially problems involving ideals of fat points. . Consider the commutative diagram OPn (U ) → OX (π −1 (U )) ↓ OPn (U \ {pi }) → ∼ = ֒→ The left vertical arrow is an equality by [Hr. OX (tH − i mi Ei )). Corollary III.11.

8. π∗ L) for all l > 0. Then I r ⊆ I (m) if and only if r ≥ m.1.GLOBAL ASPECTS OF THE GEOMETRY OF SURFACES 19 Thus IZ → π∗ L is locally an isomorphism hence it is an isomorphism.3. Proposition V. hence at pi it is equal to the inverse limit of H l (jEi . OjEi (mi )) over j by [Hr. . Consider 0 → OX (−Y ) → OX → OY → 0 and tensor through by OX (aH − bHi ) to get 0 → OX (a′ H − b′ Ei ) → OX (aH − bEi ) → ′ ′ OY (aH ′ − bEi ) → 0. For this we need to show that Rl π∗ L = 0 for all l > 0. Let Z = p1 + · · · + ps ⊂ Pn and let I = I(Z) ⊆ R = k[Pn ].2 as: . Exercise III. and since the result is true in dimension 1 (i. L) = H l (Pn .3. Symbolic Powers. where mZ is the fat point scheme mp1 + · · · + mps . Let P be a prime ideal in a polynomial ring R = k[x0 . i Now assume that mi ≥ 0 for all i and let L denote OX (tH − i mi Ei ) for an arbitrary integer t. Corollary V.1 we see m ≥ r. m′ ≥ m implies I (m ) ⊆ I (m) .2. OX (aH − bEi )) = 0 for all l > 0 if a ≥ b. .11. Showing the same for the leftmost term eventually reduces to showing hl (X. The symbolic power P (m) of P can be defined as P (m) = ∩P m ⊆M maximal M m (see [Es. π∗ L(t)) = H 0 (X.3. Let Z = p1 + · · · + ps ⊂ Pn and let I = I(Z). The result will follow by showing that hl (X. Also. Thus it suffices to show that H l (jEi . Ei = Ei ∩ Y . I (1) = I 1 .. a′ = a − 1 and b′ is the maximum of b − 1 and 0.1. determine the least m ≥ r such that I (m) ⊆ I r . . I m ⊆ I (m) . . where OjEi (mi ) denotes OjEi ⊗OX OX (−mi Ei ). IZ (t)) ∼ H 0 (Pn . . Thus Rl π∗ L has support at most at the points pi . Proof.1] (as in the proof of [Hr. where we define the m-th symbolic power I(Z)(m) to be I. Theorem III.3. so H 0 (Pn .) We will for simplicity focus here on the case of symbolic powers of ideals of points in projective space. See [PSC] for greater generality. Then define the resurgence of I to be m (m) :I ⊆ Ir . . ps ∈ Pn . I(Z)(m) = ∩i (I(pi )m ).3. Let Z = p1 + · · · + ps ⊂ Pn and let I = I(Z) ⊆ R = k[Pn ]. Understanding the reverse containment is a much harder largely open problem: Problem IV.. Taking cohomology of this exact sequence shows that hl = 0 for the ends for all l > 0 then hl = 0 for ′ the middle term for all l > 0.1]. See Exercise IV. I(Z)(m) = I(mZ). Theorem 3. so we can restate Problem IV. Let Y be a prime divisor whose class is H − Ei if b > 0 or just H if b = 0. and to do this it is enough to check that the stalks vanish. Problem IV. . Let Z = p1 + · · · + ps ⊂ Pn and let I = I(Z) ⊆ R = k[Pn ]. By the Nullstellensatz. We conclude by applying [Hr.4 ([BH]). in terms of primary decompositions.3. where H ′ = H ∩ Y . Look at the exact sequence 0 → OX (−jEi ) → OX → OjEi → 0 and tensor through by OX ((mi + j)H − mi Ei ) to get 0 → OX ((mi + j)(H − Ei )) → OX ((mi + j)H − mi Ei ) → OjEi (mi ) → 0. This generalizes nicely to the case of an ideal I(Z) = ∩I(pi ) of points p1 . Let Z = p1 + · · · + ps ⊂ Pn and let I = I(Z) ⊆ R = k[Pn ]. .4]. (This is consistent with the definition of symbolic powers used in [HH]. For this mimic the argument of [Hr. OjEi (mi )) = 0.2. This is clear at points away from each point pi since π is an isomorphism then. Let Z = p1 + · · · + ps ⊂ Pn and let I = I(Z) ⊆ R = k[Pn ]. but in dimension one less. In fact.3. by Lemma IV. Compute or at least give bounds on ρ(I). Given r ≥ 2. xn ] over an algebraically closed field k. showing that H l (X. Since OY (aH ′ − bEi ) is of the same form as what we wish to prove. we may assume the rightmost term has hl = 0 for all l > 0 by induction.e. Clearly.3.4]). when Y = P1 ).14]). and ′ clearly. ρ(I) = sup r Problem IV. this suggests the following definition and problem: Definition IV.5 ([BH]). we know that P = ∩P ⊆M maximal M .6. IV. As an asymptotic first step.3. The Containment Problem.e. OX (tH − = mi Ei )). we have: Lemma IV.3.3. IV. Determine all r and m such that I (m) ⊆ I r . Since I (m) ⊆ I r implies I m ⊆ I (m) ⊆ I r ⊆ I (r) . OX ) = 0 for all l > 0.

Let Z = p1 + · · · + ps ⊂ Pn . First we show how to interpret γ in this context.1. let I = I(Z) and let r and m be positive integers. IV.e. To apply Lemma IV.2. Estimating the Resurgence. We recall that the regularity reg(I) of I is the least degree t ≥ 0 such that (R/I)t and (R/I)t−1 have the same vector space dimension. . so I (qn) = ∩i I(pi )qn ⊆ ∩i I(pi )[q] ⊆ (∩i I(pi ))[q] ⊆ I q by Proposition IV. (a) If α(I (m) ) < rα(I). then I (m) ⊆ I r if and only if α(I (m) ) ≥ rα(I). hence by Corollary II. [ELS] uses asymptotic multiplier ideals for J (see [Te] for an exposition of this approach).4. Assume char(k) = p > 0 and let q be a power of p.4. Let I ⊆ R = k[Pn ] be an ideal generated by s elements. α(J) is the greatest power of M containing J). (By Proposition IV. Define the q-th Frobenius power I [q] of I to be the ideal generated by I q . J ⊆ R = k[Pn ] be ideals. (c) reg(I) α(I) ≤ ρ(I) ≤ γ(I) γ(I) (d) If α(I) = reg(I).e. Then I (nr) ⊆ I r for each r ≥ 1. We have the following theorem: Theorem IV. Thus α(J) is the degree in which the ideal starts (hence the use of the first letter.9. Lemma 13..2 we have γ(I) ≥ 1.3. Define γ(I) to be the infimum of dm /m where dm is the least degree t such that I (m) contains a nonzero form of degree t. ε(P2 . Then I(pi )qn ⊆ I(pi )[q] by Lemma IV. dms /(ms) ≤ dm /m for all s > 0. .1. i. Proof.3.2.2.10. In particular. Proof. Example IV.3. We state a simplified version of the result of [HH]: Theorem IV.10 exhibits the role of Frobenius powers.. γ(I) > 0. it is actually a special case of the Hochster-Huneke proof of Theorem IV.6.20 B. where M is the ideal generated by the variables (i. for any t ≥ 0 let Jt be the k-vector space span of the forms of degree t in k (called the homogeneous component of J of degree t). In this section we show how to use γ from Definition II.2. . α. ps ).3.4. . we will denote the least degree t such that J contains a nonzero form of degree t by α(J).3. then I (m) ⊆ I r .4]. Let I. Example 8. See Exercise IV. . One can also regard α(J) as the M -order of J.3. and assume char(k) = p > 0 and q is a power of p. Let Z = p1 + · · · + ps ⊂ Pn and let I = I(Z) ⊂ R = k[Pn ]. while the proof of [HH] uses Frobenius powers for J (with a dash of tight closure to get the general result).3. p1 . . let I = I(Z) for Z = p1 + · · · + ps .. then I (m) ⊆ I r . where char(k) = p > 0 and q is a power of p. i. (b) If rreg(I) ≤ α(I (m) ). we will also want to note: Lemma IV. . γ(I) = γ(P2 .2. p1 . we define (R/J)t to be Rt /Jt . More generally. HARBOURNE It is not clear a priori that ρ(I) is even finite. .3. the proof essentially involves finding an ideal J such that one can check both that I (nr) ⊆ J and that J ⊆ I r . given a homogeneous ideal 0 = J ⊆ k[Pn ]. hence ρ(I) ≤ n. this remains true for Pn . Let I ⊆ R = k[Pn ] be an ideal.8. so it makes sense to divide by γ(I).6. Note that R/J is also graded. See [HS.6.1. Then (I ∩ J)[q] = I [q] ∩ J [q] .1 and the regularity of an ideal to give bounds on ρ(I).3.) As in Proposition II.e.9 since the ideal of a point in Pn is generated by n linear forms. of the Greek alphabet to denote this concept).3] or [PSC.3. Consider Z = p1 +· · ·+ps ⊂ Pn and let I = I(Z) ⊆ R = k[Pn ]. Given points p1 . .4.8.1 ([BH]). Assume char(k) = p > 0 and that q is a power of p.1. this is actually a limit which is decreasing on multiplicative subsequences. . Then I sq ⊆ I [q] . ps ) ≥ 1/s. ps ∈ Pn .11. . .8 and the obvious fact that I [q] ⊆ I q . Proposition IV.7. By a similar argument. . Example IV. For the example we will need some results on Frobenius powers: Definition IV. Results of Swanson [Sw] showed in many cases that it is and inspired the results of [ELS] and [HH]. this is consistent with Definition II. Given a homogeneous ideal J ⊆ R = k[Pn ].3. As noted by the remark after Exercise I. Both for [ELS] and for [HH].

2]): Conjecture IV. Thus (m) (m) (r) r r for t < rreg(I) ≤ α(I (m) ) we have 0 = It ⊆ It .) In fact. Then I (m) ⊆ I r if m ≥ rn − (n − 1). (For example.5. Taking the limit as r → ∞ gives γ(I) ≥ α(I)/n and. But for any nonzero ideal I properly contained in (x0 .1 (Huneke). (b) First we check that rreg(I) ≤ α(I (m) ) implies that r ≤ m. we have I (rn) ⊆ I r by Theorem IV.. if I is the ideal of a set of points.5. (m) r (m) r so It ⊆ It holds for all t and we have I ⊆I . or α(I (rn) )/(rn) ≥ α(I)/n. Let I ⊂ k[Pn ] be the ideal I = I(Z) where Z = p1 + · · · + ps ⊂ Pn for a finite set of distinct points pi .6. hence m/r = ms/(rs) ≤ ρ(I).6.5. Unfortunately. . suggested the following conjecture (this is a simplified version of [PSC. this follows n from [Ch. Must it be true that I (3) ⊆ I 2 ? In the case of the ideal I of any s generic points of P2 . . Proof.1 has an affirmative answer when char(k) = 2. Since I m ⊆ I (m) . then containment fails by (a). Conjecture 8. we have rreg(I) ≤ mγ(I) ≤ α(I (m) ) so γ(I) (d) If α(I (m) ) < rα(I).5. . and hence α(I (ms) ) < rsα(I). Mimicking the argument of Example IV.e.3 and its proof holds also for Pn . while if α(I (m) ) ≥ rα(I).4.6. and hence α(I (rn) ) ≥ rα(I).GLOBAL ASPECTS OF THE GEOMETRY OF SURFACES 21 have m/r < msα(I)/α(I (ms) ). .2. ps ∈ Pn for s = d+n−1 general points. one can hope to do better. . Theorem 1].10 shows in fact that I (rn−(n−1)) ⊆ I r holds if char(k) = p > 0 and r is a power of p. hence I (m) ⊆ I r if and only if α(I (m) ) ≥ rα(I). we have It ⊆ It = It . See Exercise IV. for s ≫ 0 we γ(I) γ(I) Example IV.2 (Harbourne). . We thus obtain an observation of Huneke: Corollary IV. This and additional examples. since α(I r ) = rα(I) and so in this case I (m) has a nonzero element of degree less than any nonzero element of I r .2. Example IV. (c) For any 0 < m/r < α(I) . alternatively. . Thus rreg(I) ≤ α(I (m) ) ≤ mα(I) ≤ mreg(I).3. (a) This is clear.4. . we see that α(I (m) ) ≤ α(I m ) = mα(I). α(I) ≤ ρ(I). A Question and a Conjecture. see also [AV].3. See however Exercise IV.2 suggests in light of Theorem IV.1(a) the following possibly easier question: Question IV. we have α(I (n+1) ) ≥ n+1 α(I). while for t ≥ rreg(I) ≥ reg(I r ). Let I be the ideal of p1 . However. both in dimension 2 and in higher dimensions. . Let 0 = I ⊂ k[Pn ] be any homogeneous ideal. (r) r Now we use the facts that reg(I r ) ≤ r reg(I) and It = It for all t ≥ reg(I r ) [GGP]. [BH] showed that the answer is yes. Taking r = 2 in Conjecture IV. i. IV. Let I be the ideal of distinct points p1 . α(I (m) ) is not in general known.5.5.4.3. ps ∈ Pn .4.5.2. .2.3.5. Question IV. xn ) we have α(I) > 0. .3. so reg(I) > 0 and we see m ≥ r and hence I (m) ⊆ I (r) . then containment holds by (b). In fact. And for any m/r ≥ reg(I)/γ(I). I (m) ⊆ I r by (b) and hence ρ(I) ≤ reg(I) γ(I) . since α(I) = lims→∞ msα(I)/α(I (ms) ) by Proposition II.3. But Proposition II. This shows that the bound given in Theorem IV. since α(I (m) )/m ≥ γ(I) for every m ≥ 0 as in Proposition II. But α(I) ≤ reg(I) since for all 0 ≤ t < α(I) we have dimk (R/I)t > dimk (R/I)t−1 . Must it be true that α(I (n+1) ) ≥ 2α(I)? What is known is that α(I (n+1) ) ≥ n+1 α(I). . so I (ms) ⊆ I rs for s ≫ 0 by (a). examples suggest that α(I (rn−n+1) ) ≥ rα(I) + n − 1 may hold for the ideal n of any finite set of points in Pn (and perhaps for any nontrivial homogeneous ideal in k[Pn ]). Then α(I) = n reg(I) = d. The paper [BH] gives examples of reduced schemes Zi ⊂ Pn of finite sets of points such that limi ρ(I(Zi )) = n.4.6 is in some sense sharp.5. Let I ⊂ k[P2 ] be the ideal I = I(Z) where Z = p1 + · · · + ps ⊂ P2 for a finite set of distinct points pi . Huneke has raised the following question: Question IV. .3.

22 B. Uniform bounds and symbolic powers on smooth varieties. 1992.1. and assume char(k) = p > 0 and q is a power of p. p3 ∈ P2 . p2 .2). 87–104. Thus I (m) ⊆ I r holds by Theorem IV.4. Bocci and B. then I r ⊆ I m ⊆ I (m) . 3 Solution (Exercise IV. Commutative Algebra With A View Toward Algebraic Geometry. Solution (Exercise IV.3. p. to appear.6. Invent.9 so that I qs−(s−1) ⊆ I [q] . [D] [Du] [ELS] [Es] . Then monomials in the fi of degree sq − (s − 1) generate I sq−(s−1) . . Duke Math. . the monomial has degree at most s(q − 1) in the fi . Smith. 98 (1999). Justify Example IV.6. Singular Hermitian metrics on positive line bundles. New York : Springer-Verlag. Math.9: Let I ⊆ R = k[Pn ] be an ideal generated by s elements. α(I) = 2. Then I sq ⊆ I [q] . which is less than sq − (s − 1)).2. Boston-Basel-Stutgart (1981). Complex algebraic varieties (Bayreuth. Thus each monomial is in I [q] . Eisenbud. 1995. Conversely. Thus each monomial is in I [q] . S´minaire Delange-Pisot-Poitou. 12. In particular. Reduction method for linear systems of plane curves with base fat points. ps ∈ Pn . Then I r ⊆ I (m) if and only if r ≥ m. since sB + iC = (3s + 2i)L − (2s + i)(E1 + E2 + E3 ) ∈ EFF(X) is clear but ((3s + 2i − 1)L − (2s + i)(E1 + E2 + E3 )) · (2L − E1 − E2 − E3 ) < 0. Note that B = 3L − 2(E1 + E2 + E3 ) = (L − E1 − E2 ) + (L − E1 − E3 ) + (L − E2 − E3 ) ∈ EFF(X) and C = 2L − E1 − E2 − E3 ∈ NEF(X) (since C is the class of a prime divisor of positive self-intersection). Since R2 /I2 and R1 /I1 both have dimension 3. Dumnicki. S´minaire de e Th´orie des Nombres. we also have reg(I) = 2. but for each such monomial there must in fact be an i such that fiq is a factor (if not. a J. Let f1 . Then I (m) ⊆ I r if and only if m ≥ 4r−1 . Demailly.-P. Univ. R. 241-252. Biran. . Ein. .uib. Then α(I (m) ) = 3s + 2i. Singular points on complex hypersurfaces and multidimensional Schwarz Lemma. Now say m is odd: containment holds exactly when (3m + 1)/2 = (3(m − 1) + 4)/2 ≥ 2r which is again equivalent to m ≥ (4r − 1)/3.1: Let Z = p1 + · · · + pr ⊂ Pn and let I = I(Z) ⊆ R = k[Pn ]. 19–30 (2003). Springer-Verlag.Sez. VII .no/People/nmajv/03. e e Birkh¨user. Let f1 . References [AV] [Bi] [BH] [Ch] A.6. Mat. First check the case that m is even: containment holds exactly when 3m/2 ≥ 2r which is equivalent to (3m + 1)/2 ≥ 2r.3).1(d) for m = 2s + i exactly when 3s + 2i ≥ 2r.3.10 works except that we need to refine the statement of Lemma IV. J. Then monomials in the fi of degree sq generate I sq . Exercise IV. assume I r ⊆ I (m) . Lazarsfeld and K. E. or m ≥ (4r − 1)/3. and for each such monomial there must be an i such that fiq is a factor.3: Let J be the ideal of distinct points p1 . where J(p1 ) ⊂ Ri is the ideal of p1 in Ri .10 shows in fact that J (rn−(n−1)) ⊆ J r holds if char(k) = p > 0 and r is a power of p. . . (http://www.3. L. Mimicking the argument of Example IV. HARBOURNE IV.3. Ferrara . Let I ⊂ R = k[P2 ] be the ideal of non-collinear points p1 . IL. Comparing Powers and Symbolic Powers of Ideals. fs generate I.6. pp. C.5. 1507. Constructing new ample divisors out of old ones. 1990).6. no. Journal of Algebraic Geometry. A Note on Symbolic and Ordinary Powers of Homogeneous Ideals. Let X be the blow up of P2 at the three points. Exercises. Exercise IV. where p1 ∈ Ui = Spec(Ri ) ∼ An is a standard affine open neighborhood (the complement = of xi = 0.3. V. where Ri = k[x0 /xi . xn /xi ] for some xi not vanishing at p1 ) to get J(p1 )r ⊆ J(p1 )m . . Solution (Exercise IV. xvi+797. and hence r ≥ m. Ann. Localize at p1 and contract to Ri . 144 (2001).6. Arsie and J.4). Notes Math. fs generate I. editor. Paris 1979–80. . G. Vatne. 113–135. .4. preprint 2006 (math. P. If r ≥ m.6. . Exercise IV. Chudnovsky.pdf). Exercise IV. Solution (Exercise IV. Harbourne. .AG/0606716).6. D. Lect. M.1). Prove Lemma IV. . 1. Now consider m = 2s + i for s ≥ 0 and 0 ≤ i ≤ 1.Sc. · · · . Prove Lemma IV. The same argument as given in Example IV.6.3. Vol. M-J Bertin. pp. Progress in Math vol.

2. Math. S.edu . Gimigliano and Y. Michigan Math. Geramita. A Math. Positivity in Algebraic Geometry I. Collect. Galindo and F. Grothendieck and J. L. 692–702. 101–114. Huneke. Amer.. Ro´. Ro´. 16e e anee. Xu. Adv. [GM] C. 2515–2524. V´rilly-Alvarado and M. Math. Mem. Springer-Verlag. [Ha1] B. Annalen 301 (1995). [K] S.. 147:2 (1981). Keel. Lecture Note Series 336 (2006). Free Resolutions of Fat Point Ideals on P2 . A. [HH] M. Integral Closure of Ideals. Pac. preprint. to appear. Swanson. Computing multi-point Seshadri constants on P2 . 1 (2005). Bauer. Trans. W. 2009 (arXiv:0907. Univ. A. 349–369.4045). 48 (2000) 331–348. 766–772.4277v3. to appear in the AMS Contemporary Mathematics series volume “Interactions of Classical and Numerical Algebraic Geometry. Pure Appl. Matsumura. Testa. [Sz] T. Algebraic Geometry. Szemberg. 1. e [Ro] J. Harbourne. Soc. A primer on Seshadri constants. 108–116. Nachr. New York. [GHM] A. Ottem. 32. No. On Nagata’s Conjecture. 108–135. 616–627. Lazarsfeld. and Modules. On the Cox ring of P2 blown up in several points on a line. (1970). Invent. Trans. Velasco. Math. Migliore. General blow ups of the projective plane. A. Ro´. [Te] Z. n no. 213–226 (1985). Monserrat. Bounding symbolic powers via asymptotic multiplier ideals.. Strycharz-Szemberg and T. 95–111 (1986). J. Commutative Algebra. LNM 578. 609–613. London Math. [Ha6] B. 349. Amer. 419–429. e [Hr] R. B. On rational surfaces. Springer-Verlag. Kleiman. Di Rocco. Math. [ST] T. [M] H. Harbourne. B. 89. [TVV] D. Z. preprint 2009. 234 (2000). 2001. Anticanonical rational surfaces. May 22–24 2008. [R] J. 2 (1980). J. Math. 81 (1959). NE 68588-0130 USA E-mail address: bharbour@math. 1977.Simon Stevin. Harbourne. Tutaj-Gasi´ska. Nagata. Math. Pitteloud. Geom. pp. 496 pp. J. Kyoto. Sci. Nagata. Properties arithmetiques de fonctions de pleusieurs variables. Harbourne. 236 (2001). University of Nebraska. [HK] Y. On the 14-th problem of Hilbert. V. 271–293. Harbourne. 10. J. J. held at Notre Dame. Proc. Math. 9. Tutaj-Gasi´ska. Hartshorne. Sommese. Reine Angew. pp. arXiv:0901. Discrete behavior of Seshadri constants on surfaces. [Ha3] B. a [T] H. Remarks on the Nagata Conjecture. [HS] C. J. [Sw] I. 213–234 (1998). Bulletin of the Belgian Mathe ematical Society . Elements de G´om´trie Alg´brique IV. [N2] M. [SS] B. Math. 1975–76. Math. 2 (2009). Journal of Algebra 284 (2005) 91–101. 218 No. 159–192. 33 (1960). 421. Math. Soc. 257 (2003). 559 (2003) 115–122. Lahyane and B. Pure Appl. Harbourne. 299 (1994). Big rational surfaces. Dieudonn´. Graduate texts in mathematics (52). 130 (2002). Hautes Etudes Sci. [HR1] B. Classifying Hilbert functions of fat point subschemes in P2 . Berlin. Inst. Ro´. Ser. On the existence of plane curves with imposed multiple points. Soc. Global and local positivity of line bundles. [HR3] B. Teitler.. Huneke and I. Math. 1977. Curves in P2 and symplectic packings. 115–126. Seshadri constants and very ample divisors on algebraic surfaces. J. Szemberg and H. [GGP] A. Harbourne and J. Cambridge University Press. 41–59. Multiple-point formulas. e ´ e e e 1967. II. Canadian Mathematical Society Conference Proceedings 6 . Serdica Math. Springer-Verlag. [Ha4] B. M. Habilitation. 156 (2001). Harbourne and J. S´minaire Lelong: Analyse.unl. 405–430. and T. Syzdek. [Ha5] B. Lincoln. 755–775. Math. The total coordinate ring of a smooth projective surface. Irreducibility of (−1)-classes of anticanonical rational surfaces. Effective divisor dlasses and blowings-up of P2 . 2004. A bound for Seshadri constants on P2 . n [W] M. Harbourne and J. Waldschmidt. Mori dream spaces and GIT. Hochster and C. Knutsen. 30 (2004). Harbourne and J. Math. 125. Alg. 147 (2002). The geometry of rational surfaces and Hilbert functions of points in the plane. Kapustka. [Ot] J. Rosoff. W. Szemberg. [PSC] T. Geramita. Amer. Pac. Swanson. Department of Mathematics. 4 (2004). J. 363-380. preprint 2009. J. 60.” Proceedings of a conference in honor of A. Szemberg. 1191–1208 (1997). A. Journal of Pure and Applied Algebra e 212 (2008). Amer. Publ. 13–49. Linear equivalence of topologies. Math.1094. Linear systems with multiple base points in P2 . [LH] M. [X] G. Hu and S. Complete linear systems on rational surfaces. [Ha2] B. Comparison of symbolic and ordinary powers of ideals. Harbourne. [Lz] R. Rings. Alg. arXiv:0901.GLOBAL ASPECTS OF THE GEOMETRY OF SURFACES 23 ´ [EGA] A. Benjamin. Alg. 289. no. Harbourne. Graded Betti numbers of some embedded rational n-folds. no. Soc. I: Iteration. II. Acta Math. [HR2] B. J. New York. no. Ann. Coll. 212 + xii [N1] M.