Warning 1 These are sketchy and incomplete draft lecture notes written for

myself for my 256A course.

Course home page [Link]
Text: Hartshorne, Algebraic Geometry. We cover same topics, but often
different viewpoint. Lectures do not necessarily contain complete proofs.

Example 1 Solve x2 + y 2 = z 2 in integers. 32 + 42 = 52 .

Algebraic solution: Can assume x, y, z coprime, so z is odd, and x (say) is
odd.
Then x2 = (z − y)(z + y) is a product of coprime factors, so z − y = r2 ,
z + y = s2 , with r and s odd and positive, so z = (r2 + s2 )/2, y = (s2 − r2 )/2,
x = rs. Examples with (r, s) = (1, 3), (1, 5), (3, 5).
Geometric solution: X = x/z, Y = y/z, so X 2 + Y 2 = 1. Put t = y/(x + 1)
DIAGRAM
t ∈ Q corresponds to solutions other than (−1, 0)
Y = t(X + 1) so t2 (X + 1)2 + X 2 = 1, so (X + 1)((t2 + 1)X + t2 − 1) = 0
(Must have root at X = −1), so factors.)
2
so X = 1−t 2t
1+t2 , Y = 1+t2
Example: t = 1/2, X = 3/5, Y = 4/5.
This is a BIRATIONAL map from the circle to the line: an isomorphism
except in codimension at least 1. (X 6= −1, t2 6= −1). No analogue for smooth
manifolds.
Circle forms GROUP of rotations: product of (x1 , y1 )(x2 , y2 ) = (x1 x2 −
y1 y2 , x1 y2 + x2 y1 ).
Example of ALGEBRAIC GROUP G taking a commutative ring R to a
group G(R).
What is G(C)? Answer: For (x, y) ∈ G(C) put z = x + iy (note that x
and y are COMPLEX with x2 + y 2 = 1). Then x − iy = z −1 so x = z+1/z 2 ,
y = z−1/z2i , so G(C) = C ∗
. Example of a torus splitting over C.

Example 2 Solve y 2 = x3 + x2
DIAGRAM There is a NODE at (0, 0), simplest example of a SINGULAR-
ITY.
Put y = tx. 3 intersection points,2 at (0,0), so 3rd must be RATIONAL.
t2 x2 = x3 + x2 , so x = t2 − 1, y = t3 − t. Example: (3,6).
t → (t2 − 1, t3 − 1) maps A1 to curve, 2:1 at 0, 1:1 elsewhere. It is a
RESOLUTION of the singularity, and an example of BLOWING UP. (Hironaka
resolved all sings in char 0; char ¿0 still open.)

Example 3 Solve xn +y n = z n in rationals . Algebraic geometry over Q is HARD .

Example 4 Solve x3 +y 3 = 9 in rationals. (Dudeney, Canterbury puzzles #20;

Fermat)

1

3
3
Solution: 415280564497
348671682660 + 676702467503
348671682660 = 9. Dudeney found this by hand!
How?
No double points, so cannot use previous method. Chord-tangent process
leads to new points starting from 2 known points (possibly the same). This
almost defines a group (a + b + c = 0 ↔ a, b, c collinear. Identity: add “point
at infinity”. Example of a PROJECTIVE variety that is a group: ABELIAN
variety.
This is NOT birational to A1 : it has genus 1, not 0 as a Riemann surface.
Projective coordinates: (x : y : z) mod λ. x3 + y 3 = 9z 3 . z 6= 1: as above.
Extra point (1 : −1 : 0).
Associativity hard to prove directly but follows from a1 + . . . an = b1 + . . . +
bn ↔there is a rational function with zeros at a’s, poles at b’s.
2 2
Tangent to (a, b) is y = − ab2 (x − a) + b. So x3 + (− ab2 (x − a) + b)3 = 9.
Cubic in x with 2 roots a, and product of roots = -constant term/leading term
, so third root is −((9/b2 )3 − 9)/a2 (1 − a6 /b6 ) = (9b6 − 93 )/a2 (b6 − a6 ). For
(a, b) = (1, 2) this gives (−17/7, 20/7).
Cubic curves are examples of abelian varieties: connected projective alge-
braic groups. Abelian varieties are abelian (proof for complex numbers: any
map from projective variety to C is constant by maximum principle, so adjoint
map to automorphism group of tangent space of origin is 0.) Warning: abelian
linear groups are NOT abelian: this is a old name for the sympectic group, an
affine algebraic group.
Singular cubics such as y 2 = x2 +x3 also have a group law on singular points,
though this is affine not projective.

Example 5 Bezout’s theorem. Essentially first proved by Newton??? (check

this)

Example 6 Pappus’s theorem. One of the 2 interesting theorems about lines.

DIAGRAM of Pappus’s theorem
Special case of:

Example 7 Pascal’s theorem.

DIAGRAM
`1 ∩ `4 , `2 ∩ `5 , `3 ∩ `6 collinear (Pascal line)
Proof via algebraic geom: Pick P on conic. Choose λ with p1 p3 p5 −λp2 p4 p6 =
0 on P . (pi = 0 is `i )
Then this is a cubic intersecting the conic in 7 > 2 × 3 points, so must have
a component in common by Bezout, which must be the conic. So cubic factors
as the conic times a line, which is the Pascal line.

Example 8 27 lines on a cubic surface.

Example: Fermat surface w3 + x3 + y 3 + z 3 = 0 in P 3 . Typical line: (1 : −1 :
t : −t). Acted on by group of permutation, multiplying coordinates by cube
root of 1, order 33 4!. 27 images under this group.

2
Example 9 Dini’s proof of finite field Kakeya conjecture: size of Kakeya set in
F n is at least cn |F |n . Kakaya set: any set containing a line in every direction
(unit line for Euclidean space, full line for finite fields).
Besicovich showed that a needle can be turned inside a set of arbitrarily small
area; these sorts of questions are important in higher dimensional harmonic
analysis, and the finite field case was an analogue suggested by Wolfe.
Proof first show that a Kakeya set cannot lie in hypersurface of degree less
than |F |. If f is the polynomial of degree d < |F | defining the algebraic set and
fd its highest degree component, then for all v we can find x such that f (x + vt)
vanishes for all t ∈ F , so the coefficient fd (v) of td vanishes. As this is true for
any v and fd has degree less than |F |, we have fd = 0, so f = 0. (A polynomial
of degree less than |F | cannot vanish at all points of F , though a polynomial of
degree |F | such as x|F | − x can.)
Next
 observe  that polynomials of degree 6 d in n variables have dimen-
n+d
sion so we can find a hypersurface of degree at most d containing
n
any
 set with less  than this many elements. So any Kakaya set has at least
n + |F | − 1
> |F |n /n! elements. 
n

Affine scheme: something like zeros of polynomials. Scheme: covered by

affine schemes (example:P 3 ). Compare with definition of differentiable manifold.
Vector bundles in differential geometry: functions, differential forms, tangent
fields, etc. For schemes use SHEAVES (cotangent bundle of something with a
singularity does not exist, but its cotangent sheaf does.

1 Varieties
1.1 Affine varieties
k=field. Affine space =k n = An .
Algebraic set Y = zeros of a set T of polynomials: Y = Z(T )
Closed under finite union, arbitrary intersection. So form closed sets of
ZARISKI TOPOLOGY on k n .

Example 10 A1 Closed sets=finite subsets, whole space. NOT HAUSDORFF!

Example 11 A2 Closed sets: points, curves, whole space. (DIAGRAM) Zariski

topology is NOT product topology (DIAGRAM)

Example 12 Determinantal varieties. Amn = m × n matrices = linear maps

from k n → k n . Subset of matrices of rank 6 N is an affine subvariety, given by
matrices such that all minors of rank N + 1 vanish. In particular the subset of
maps k m → k n that are onto is open.

3
 Recall k[x1 , . . . , xn ] is Noetherian. Recall 3 conditions for Noetherian ring:
every ideal f.g., every ascending chain stabilizes, every set of ideals has a maxi-
mal element.
Proof R Noetherian implies R[x] Noetherian. Look at ideal I0 ⊆ I1 ⊆ . . . ⊆
In ⊆ . . . of leading coefficients of polynomials in I of degree at most n. Stabilizes
at N say, so I generated by its polynomials of degree at most N . 

Exercise 1 Show that if R is Noetherian then so is R[[x]].

Topological space called Noetherian if closed sets satisfy descending chain

condition.
So An is Noetherian because k[x1 , . . . , xn ] is Noetherian.
Noetherian equivalent to every open set is compact. (or quasicompact in
Bourbaki’s terminology). Noetherian+Hausdorff implies finite: (complement of
a point is open, so compact, so closed, so points are open).
A set is called IRREDUCIBLE if it is nonempty and not the union of proper
closed subsets. In a Noetherian space, every closed set is a finite union of irre-
ducible closed subsets (proof by Noetherian induction: look at minimal closed
counterexample). This is a refinement of decomposition into connected compo-
nents: for example xy = 0 is connected but reducible.

Example 13 x2 + y 2 + z 2 = 0, xyz = 0. Union of 6 lines x = 0, y = ±iz, etc.

Example 14 xy = 1 has only 1 irreducible component over R. Irreducible

components are connected in Zariski topology, but need not be connected in
usual topology! (It is connected over C.)

Example 15 Families of mostly irreducible algebraic sets can have reducible

members. For example, look at intersection of xy = z with z = c: this is
irreducible for c 6= 0, but a union of 2 lines for c = 0. (DIAGRAM)

Definition 1 (provisional) An affine variety is an irreducible closed subset of

affine space.

This definition is in fact a bit misleading: it turns out that the non-zero
points of the affine line also form an affine variety, as they are isomorphic to an
affine variety xy = 1, so we will later modify it.
Nullstellensatz: What is relation between
(1) Subsets Y of An
(2) Ideals I of k[x1 , . . . , xn ]?
Y → I(Y ) = polynomials vanishing on Y
Z(a) ← a zeros of polynomials in ideal a.
Z(I(Y )) is closure of Y by definition.
Is I(Z(a)) = a? √ NO! a = (x2 ), I(Z(a)) = (x). More generally, if f n ∈ a
then f ∈ I(Z(a))√so a ⊆ I(Z(a)).
Is I(Z(a)) = a? No! Over R, take a = (x2 + y 2 + 1) so Z(a) = ∅.

4
 First look at easier case of maximal ideals.
What ideals do points correspond to? (a1, . . .) → (x1 − a1 , . . . , xn − an )
(max ideal).Do all maximal ideals come from points? NO! (x2 + 1) is maximal
in R[x].
Problems in last 2 examples caused because R is not algebraically closed.

Theorem 1 (Weak Hilbert nullstellensatz: zeros position theorem). If k is al-

gebraic closed, then any max ideal I of k[x1 , . . . , xn ] is of the form (x1 −
a1 , . . . , xn − an ).
Proof We know that K = k[x1 , . . . , xn ]/I is a field. Renumber so that x1 , . . . xi
are algebraic independent, and remaining ones are algebraic over them. We have
k ⊆ F = k(x1 , . . . xi ) ⊆ K
where K is a finite module over F and F is a finitely generated field extension
over k. We want to show that F is finitely generated as a k algebra (it is trivially
finitely generated as a k field!) Note difference between being finitely generated
as a field or ring or module: these are all quite different!
Pick yP1 , . . . , ym as basis for the F -module K. Then
xa = P ta,b yb
ya yb = ta,b,c yc
for some t’s. Let T be k-algebra generated by the t’s, so
k⊆T ⊆F ⊆K
The first extension is a finitely generated algebra extension, the second a
finitely generated field extension, and the third a finitely generated module.
Then T is Noetherian as the number of t’s is finite. Moreover K is generated
by the y’s as a T -module, as this module contains the x’s and is closed under
multiplication. So K is a finitely generated module over the Noetherian ring T ,
so its submodule F is also a finitely generated module over T , so F is a finitely
generated k-algebra (this is much stronger than saying it is a finitely generated
field extension).
But if F is finitely generated by the elements fj /gj and i > 0 choose some
irreducible polynomial P in k[x1 , . . . xi ] not dividing any gi , using the fact that
there are infinitely many primes (Euclid). Then 1/P is not in k[f1 /g1 , . . .]. This
is a contradiction so i = 0 so k = F . Therefore K is a finite extension of the
field k.
So far we have not used the fact that k is algebraically closed. Finally we
use fact the k is algebraically closed to deduce that K = k. So each xj = aj for
some aj ∈ k. 

√
Theorem 2 Strong Nullstellensatz. If k is algebraic closed then a = I(Z(a)).
Proof √
It is trivial that a ⊂ I(Z(a)). Rabinowitsch trick: add extra variable
x0 to deduce strong from weak nullstellensatz. Suppose a is generated by

5
f1 , . . . , fm . Suppose f ∈ I(Z(a) so that f vanishes whenever f1 , . . . , fm van-
ish. Then f1 , . . . , fm , 1 − x0 f have no zeros in An+1 , so are not in any max ideal
by weak nullstellensatz, so its ideal contains 1. So

1 = g0 (1 − x0 f ) + g1 f1 + . . . + gm fm

for some g’s. Now put x0 = 1/f . We get 1 = g1 f1 + . . . + gm fm in the field of

rational functions. Clear denominators of the g’s by multiplying by some power
of f to get
f N = h1 f1 + . . . + hm fm
√
where N is the max power of x0 in the g’s. But this just says that f ∈ a. 
So we get a 1:1 correspondence √
Closed sets of An ↔Radical ideals a of k[x1 , . . . , xn ] a = a.

Example 16 Intersection of line y = 0 with parabola y = x2 (DIAGRAM).

p
Ideal generated by 2 ideals is (y, y − x2 ) = (y, x2 ) NOT radical! (y, x2 ) =
(y, x). Ideal generated by 2 radical ideals need not be radical.

Example 17 Look at ideal generated by a condition that an n × n matrix is

nilpotent (given by n2 polynomials of degree n in n2 matrices). This ideal is
NOT radical: for example by the Nullstellensatz its radical contains the trace,
as if a matrix is nilpotent it has trace 0. For 2 by 2 matrices the smallest power
of the trace in the ideal is the cube (not the square as one might guess). It can
be really hard to find the radical of an ideal, or even tell if an ideal is radical;
this seems to be an open problem for the ideal saying that 2 matrices commute.
(The “moduli space” of pairs of commuting matrices is a notoriously hard space
to understand!)
Refinement for schemes: Lasker-Noether theorem says that any ideal of a
Noetherian ring is intersection of primary ideals. Lasker’s definition of primary
√
was that ab ∈ p → a ∈ p ∨ b ∈ p. It turns out to be more convenient to
focus on the MODULE R/I rather than the IDEAL I. Recall that an associ-
ated prime of a module is a prime ideal that is the annihilator of some nonzero
element. (Any maximal element of set of annihilators of elements is prime, as
if I is the annihilator of m and xy ∈ I then either ym = 0 so y ∈ I or x is
in the annihilator of ym 6= 0 so x ∈ I as I is maximal. In general maximal
elements of sets of ideals have a strong tendency to be prime.) For N rings
one can show that an ideal is primary in Lasker’s sense if and only if the mod-
ule R/I has exactly one associated prime, and it is convenient to adopt this
property as a definition: a module is called COPRIMARY if it has exactly 1
associated prime. (A submodule is called primary if the quotient is coprimary,
but this depends on 2 modules rather than 1.) Better viewed as a theorem about
finitely generated modules over Noetherian rings: 0 is intersection of primary
submodules. (proved by world chess champion Lasker for ideals in polynomial
and power series rings, generalized by Noether and Grothendieck to coherent
sheaves over Noetherian schemes.) Special case: finite abelian groups. Lasker’s

6
original paper proving the special case of ideals in polynomial rings was about
100 pages long, involving a complicated induction on dimension. The more gen-
eral theorem about modules over Noetherian rings can be proved in a few lines
as follows. (This shows the advantage of using the “correct” definitions: they
almost force you to write down the simple proof.)
Proof: 1. in any Noetherian module, 0 is intersection of finite number of ir-
reducible submodules (irreducible = not intersection of 2 larger submodules), as
by Noetherian induction if it is not there is a max submodule that is not, which
gives a contradiction. 2. Any irreducible submodule is a primary submodule.
Equivalently if 0 is irreducible in M then M is coprimary. This follows because
if p, q are associated primes then there are submodules R/p, R/q which have
zero intersection, as annihilator of any nonzero element of these submodules is
p 6= q.

Exercise 2 Check that, for Noetherian rings, if R/q is coprimary then q is primary
in Lasker’s sense.
Hints: (1) First reduce to the case q = [Link] we have to show that if p is the only
associated prime of R then it is nilpotent, or equivalently every element is nilpotent.
(2) Ifa ∈ p is not nilpotent then Ra 6= 0, so pick an associated prime of Ra . Show that
its inverse image in R is prime, and is the annihilator of an b for some n. Deduce that
is an associated prime of R but does not contain a, and obtain a contradiction.

Example 18 Look at the ideal generated by (xy, y 2 ). The algebraic set is just
y = 0 but this does not give a complete picture of the ideal: informally there is
a little bit sticking out at the origin. A primary decomposition of this ideal is
the intersection of the ideals (y) and (x, y 2 ). This primary decomposition is not
unique, even if we remove redundant elements and have only one primary ideal
for each prime: it can also be given as the intersection of (y) and (x+y, y 2 ). So a
minimal decomposition into irreducible algebraic sets is unique, but we do NOT
get uniqueness for the more general case of ideals or subschemes. The “point
with a bit sticking out” is called an embedded component, meaning that its
underlying algebraic set is contained in the algebraic set of a larger irreducible
component. In general the primary ideals of the “maximal”irreducible algebraic
sets are unique, while the primary ideals of the embedded components need not
be (though their radicals are uniquely determined).

Example 19 Primary decomposition is a generalization of the structure theo-

rem for finitely generated abelian groups: if M is the sum of Z n0 and p-groups
for various p then 0 is the intersection of the primary submodules whose quo-
tients are Z n0 and the p groups. Note that this decomposition is unique for
FINITE groups but not unique in general: for example Z + Z/2Z has more
than 1 such decomposition. As a special case we “generalize” the fundamental
theorem of arithmetic: primary decomposition of Z/nZ closely related to the
factorization of n into primes.
Warning: primary is not the same as power of a prime. For example, in
k[x, y] any ideal containing some power of (x, y) is primary, but need not be a

7
power of (x, y).
So affine varieties correspond to finitely generated algebras over k with no
nilpotents. Application: can we take quotient V /G of affine variety V by a
group G? Taking quotient of points is unclear: how do we make this an affine
variety? Idea: coordinate ring of V /G should be invariant elements of coordinate
ring of V . This is obviously an algebra over k with no nilpotents. Is it finitely
generated? A: Sometimes (Hilbert) but not always (Nagata).

Example 20 Consider affine space An acted on in the obvious way by the sym-
metric group Sn permuting coordinates. The quotient has coordinate ring the
symmetric polynomials, which is a polynomial ring in the elementary symmetric
polynomials. So the quotient is affine space again. (It is unusual for a quotient
to be so well behaved: this happens for (complex) reflection groups.) The quo-
tient my not be what you expect: for example if we take the quotient of the real
affine line by z → −z the quotient has coordinate ring R[x2 ] so the quotient
is again the affine line, which seems wrong as this is not the topological space
quotient. The reason is that the quotient is really pairs of points {z, −z} fixed
by complex conjugation, which includes things like {i, −i}.

Example 21 Orthogonal group has ring of invariants generated by (x, x), a

polynomial ring in 1 variable, corresponding to the fact that the quotient (set
of spheres centered at the origin) is the affine line. The special linear group of
V acting on V n has no nontrivial invariants for n < dim(V ) as in this case the
group acts transitively on a dense open set (n linearly independent vectors).
But for n = dim(V ) the determinant is an invariant.

Example 22 Classical invariant theory: G = SL2 (C) acting on an xn +an−1 xn−1 y+

. . .+a0 y n , A = C[a0 , . . . , an ]. AG is the ring of invariants of binary forms, shown
to be finitely generated by Gordan. More complicated examples in more vari-
ables shown to be finitely generated by Hilbert. Example of an invariant: the
discriminant b2 − 4ac of ax2 + bxy + cy2 .
(Gordan is supposed to have said about Hilbert’s finiteness proof “this is
not math; this is theology” as Hilbert’s proof was not constructive. It is not al
all clear if he really said this, and in any case it may have been a compliment
rather than a complaint; Gordan thought highly of Hilbert’s work.)
Proof We do the case when G is finite. A is graded by degree. Let I be ideal
generated by positive degree elements of AG . Then I is a finitely generated ideal
by Hilbert basis theorem, with generators i1 , . . . , ik which we can assume are
fixed by G. We want to show that these generate AG as an ALGEBRA, which
is much stronger than saying they generate IDEAL I. (Example: subring of
k[x, y] generated by xy ∗ is NOT finitely generated, even though corresponding
ideal is. We need to use some special property of subrings fixed by a finite
group.)
Need Reynolds operator ρ given by taking average under action of G (needs
char=0). Key properties: ρ(ab) = aρ(b) if a fixed by G, ρ(1) = 1. Not true that

8
ρ(ab) = ρ(a)ρ(b) in general. ρ is a projection of AG modules from A to AG but
is not a ring homomorphism.
We show by induction on degree of x that if x ∈ AG then it is in algebra
generated by i’s.
We know
x = a1 i1 + . . . + ak ik
for some a’s in A as x is in I. Apply Reynolds operator:

x = ρ(x) = ρ(a1 )i1 + . . . + ρ(ak )ik

By induction ρ(aj ) is in AG as it has degree less than that of x, so x ∈ AG . 

Compact groups: similar as can still integrate over the group:
Noncompact groups such as SLn (C): Use Weyl’s unitarian trick: invariant
vectors (for finite dimensional complex reps of the complex group) same as for
compact subgroup SUn , so still get Reynolds operator. Works for all semisim-
ple or reductive algebraic groups (key point: reps are completely reducible), but
NOT for some unipotent groups (Nagata counterexample to Hilbert conjecture:
take k acting unipotently on k 2 , and copy this 16 times to get k 16 acting on
k 32 . Then the ring of invariants of a “generic” 13-dimensional subspace of k 16
is not finitely generated.). Char p harder as groups need not be completely re-
ducible; e.g. Z/ pZ acting on 2-dim space over F p . Haboush proved Mumford’s
conjecture giving a sort of nonlinear analogue of Reynolds operator, which can
be used to prove finitely generated of invariants for reductive groups as in char
0. Summary: quotient of affine variety by reductive group is affine variety.
Quotients by groups are used in constructing moduli spaces: for example
(moduli space of elliptic curves) is roughly (Hilbert scheme of cubic curves in
P 2 )/(action of group of automorphisms of P 2 )

Example 23 Hyperelliptic curves y 2 = an xn + . . . + a0 . Rewrite this as

y 2 z n−2 = an xn z0 + . . . + a0 x0 z n . The group SL2 (k) acts on the space of degree
n forms in x and z. We want to find the quotient space, which will be closely
related to the moduli space of hyperelliptic curves. The coordinate ring is the
space of all invariant polynomials in a0 , . . . , an ; in other words we want to know
the invariants of binary quantics. (These are finitely generated by Gordan’s
theorem.)

Example 24 Cyclic quotient singularities. Take quotient of a cyclic group

acting on a vector space. Example: cyclic group of order n acting on A2 by
(x, y) → (ζx, ζy) where ζ is a primitive nth root of 1. The invariant polynomials
are generated by the n + 1 elements zi = xi y n−i . There are many relations
between them as zi zj = zk zl whenever i + j = k + l.

Example 25 (Suggested by Petya on mathoverflow from Kontsevich) A moduli

space as a quotient. Look at configuration of cyclohexane: chemist Hermann
Sachse in 1890 discovered that there are 2 forms of this. Take a cycle of 6

9
unit lines in R3 such that each meets the next at some fixed angle, say right
angles. (Cyclohexane has angle about 109 degrees.) The configuration space
is affine variety given by intersection of 12 quadrics in 18 dimensional space.
Moduli space of configurations is quotient of this by 6-dimensional group of
Euclidean motions, so we might incorrectly guess it has dimension 0. There are
two components, one (“chair”) of dimension 0 invariant under inversion through
origin (omit opposite vertices of a cube) and one of dimension 1 invariant under
rotation by half a revolution (omit adjacent vertices of a cube.) This is flexible
because one can fix 3 consecutive edges then rotate this by 1/2 revolution to
make endpoints meet. There are similar statements for almost any edges
a, b, c, a, b, c and angles α, β, γ, α, β, γ, though there can sometimes be extra
configurations invariant under a reflection exchanging 2 opposite vertices.
Dimension. Intuitively obvious but hard to define and work with.
Hausdorff spaces have following examples:
Cantor showed that there is a bijective map from R → R2 .
Peano curve is a continuous surjective map from R → R2 .
Lebesgue covering dimension: any open cover has a refinement with no point
in more than n + 1 sets means dimension is at most n. Example: dimension
2 (DIAGRAM) no point in more than 3 sets. Not trivial to prove that n-dim
space has dimension n.
Dimension for non-Hausdorff spaces is TOTALLY different. Dimension de-
fined as sup n such that Z0 ⊂ Z1 ⊂ . . . ⊂ Zn are distinct and irreducible.
DIAGRAM pt ⊂ curve ⊂ A2
Warning: Any Hausdorff space has dim 0 with this definition, as only irre-
ducible sets are points. On the other hand, A1 has infinite Lebesgue covering
dimension, as any finite number of non-empty open sets intersect. So NO rela-
tion between two definitions of dimension.
Problem: What are irreducible subsets of (say) A4 ? Hard to describe, and
one needs nontrivial commutative algebra to calculate dimension.
Dimension of a ring: Dimension defined as sup n such that P0 ⊂ P1 ⊂ . . . ⊂
Pn are distinct prime ideals. Dim R = sup dim Rm , m maximal, so can be
reduced to dimension of local rings.
Algebraic definition: main idea is that higher dimensional spaces have more
functions on them.

Example 26 dim B/k = tr. deg. ring of quotient field of B over k.

Example 27 Best definition uses Hilbert polynomial of a local ring. Look

at dim(A/mk ): this is a polynomial in k for large k of degree d, and d is the
dimension of the local ring. It is also the same as the dimension of the completion
as these have the same Hilbert polynomials.
For example, k[x1 , . . . , xn ] has dimension n. k[x, y]/(y 2 − x3 − ax − b) has
dimension 1.

10
Warning 2 Noetherian local rings always have finite dimension, but Noethe-
rian rings can have infinite dimension. Non-Noetherian rings can have dimen-
sion 0 (example:k[x1 , . . . , xn , . . .](x2i )) so Noetherian is not the same as finite
dimensional.

Example 28 Dimension of Hilbert scheme of n points in Am (ideals in k[x1 , . . . , xm ]

with codimension n). Obvious guess for dimension is mn as Hilbert scheme
seems to be a sort of symmetric product of Am . This is correct for m = 1, 2
but fails horribly for larger m. Example: take m = 3 and look at ideals I
with mk ⊇ I ⊇ mk+1 with m = (x1 , x2 , x3 ). Any subspace of the vector space
mk /mk+1 will give such an ideal. The ideal mk has codimension a degree 3 poly-
nomial in k so mk /mk+1 has dimension given by a degree 2 polynomial. So the
Grassmannian of subspaces of about half this dimension has dimension given
by a degree 4 polynomial Q in k, and corresponds to ideals whose codimension
is a degree 3 polynomial P in k. As Q(k) is eventually larger than 3P (k) there
are components of dimension Q(k) greater than 3P (k) = mn when n (number
of points) is large enough. In particular there are 0-dimensional subschemes of
degree n that cannot be obtained as limits of n points.

1.2 Projective varieties

Definition 2 Projective space P n = 1-dim subspaces of k n+1 = nonzero points
(x0 : . . . : xn ) modulo scalar.
Projective space contains affine space (1 : x1 , . . . , xn ) together with the
points at infinity forming a copy of projective space of lower dimension. It is a
sort of compactification of affine space (at least over C). It is covered by n + 1
copies of affine space.
Historical background. Projective geometry was the study of properties
invariant under projection. For example, projection of a railway track onto a
picture of a railway track shows that “parallel” lines might meet after projection.
(DIAGRAM).
Analytic geometry (using coordinates) versus synthetic geometry (uses ax-
ioms about points lines, incidence relations,etc.)
Axioms for synthetic projective geometry: points+lines+incidence relation.
1. Any 2 distinct points meet a unique line.
2. Any 2 lines “in same plane” meet in a point. “In same plane” means they
meet 2 intersecting lines in 4 points. (DIAGRAM)

3. Non-degeneracy: any line meets at least 3 points. (Just to eliminate reducible

unions.)

• Dim 0: 1 point, no lines., boring.

• Dim 1: just 1 lines, with all points on it. Also boring.

11
 • Dim 2: Any 2 lines meet in a point, at least 2 lines. Projective plane.
Example: Fano plane (DIAGRAM).
• Dim > 3 There exist 2 lines that do not meet.

Examples: points, lines = 1, 2 dimensional subspaces of a vector space over a

division ring.

Theorem 3 Desargues. 2 triangles abc, ABC with aA, bB, cC meeting at E.

Suppose x lies on bc, BC; y lies on ac, AC, and z lines on ab, AB. Then
x, y, zare collinear
Proof In R3 . xyz all lie in plane abc and in plane ABC, so they lie in the
intersection which is a line if the 2 triangles are not coplanar. General case:
project from non-coplanar triangles. This proof FAILS in 2-dimensions! 

Remark 1 Another theorem whose proof uses the same trick of looking in 3
dimensions says that if we draw pairs of lines from three circles their intersection
points lie on a line.

Theorem 4

• Any projective space of dimension at least 3 satisfies Desargues theorem.

• Any projective space of dimension at least 2 satisfying Desargues theorem
is a projective space over a division ring.

• And the division ring is a field if and only if Pappus’s theorem holds.

So projective space are points and lines (boring), non-desarguesian planes

and projective spaces over division rings. Non-Desarguesian planes seem to
be mostly junk. Examples are the projective plane over the octonions, the
projectivization of the Moulton plane (double slopes of lines as the cross the
y-axis), finite planes of order 9, .... Hard problem: is there a projective plane
of order n for n=10 (no), 12 (open)...
Summary: synthetic projective geometry + Pappus leads to the definition
of projective space used here.
Remark: it is common in math for the case of dim at least 3 to be the
general case, the case of dimension 1 is trivial in some sense, and dim 2 is a
mess. Examples: Tits buildings classified in terms of algebraic groups for rank
at least 3. The notorious quasi-thin case in classification of finite simple groups
was a “rank 2” problem.
Affine space An ⇔ k[x1 , . . . , xn ]
Affine algebraic set ⇔radical ideal
Projective space P n ⇔ GRADED ring k[x0 , x1 , . . . , xn ]
Affine algebraic set ⇔GRADED radical ideal other than (x0 , . . . xn ).
Cone on projective variety=affine variety in An+1 .

12
Example 29 Twisted cubic T = points(t, t2 , t3 ) in A3 , given by ideal (y −
x2 , z − x3 ). In P 3 it is given by the points (s3 : s2 t : st2 : t3 ). Homogenizing
the generators gives the graded ideal (wy −x2 , w2 z − x3 ) which is NOT the full
ideal (wy −x2 , y 2 − xz, wz − xy) of the projective curve. In particular wz − xy is
not in the ideal generated by wy −x2 , w2 z − x3 . In fact the latter ideal is not
even reduced. Analyze it on the 4 affine spaces covering projective space:
• w = 1: we get the affine twisted cubic of degree 3.
• x = 1: we get wy = 1, w2 z = 1 so again we just get a single curve (affine
line - point).
• y = 1: we get w = x2 , w2 z = x3 . Eliminating w gives x4 z = x3 which
splits as xz = 1 and x3 = 0. So we get a sort of triple copy of the line
w = x = 0 as well as the twisted cubic.
• z = 1: We get wy = x2 , w2 = x3 . Eliminating y gives w2 = x3 . However
we can only eliminate y like this when w is nonzero, and we pick up an
extra line. Eliminating y gives x4 = x3 so this line has multiplicity 3
again.
The decomposition of the variety as a union of 2 irreducible curves corre-
sponds to a primary decomposition of its ideal. (wy −x2 , y 2 − xz, wz − xy) ∩
(wy −x2 , w2 , wx)The ideal is not radical: for example, (y 2 − xz)x is not in the
ideal but its cube is.

Example 30 How many point in projective space over a finite field? Method
1: projective space = affine space + projective space at infinity. Method 2:
Projective space = affine space -point/ multiplicative group. Zeta function is
exp( (points over F qn )tn /n) = 1/(1 − t)(1 − qt)(1 − q 2 t) . . . (1 − q dim t). Note
P
relation to complex cohomology: Betti numbers of complex projective space
can be read off from number of points over finite fields. Generalized by Weil
conjectures.

Example 31 Is the product of 2 projective varieties projective? Analogue for

affine varieties is easy: If Y and Z are given by ideals I and J in k[x1 , . . . xm ]
and k[y1 , . . . yn ] then the product is given by the ideal generated by I and J in
k[x1 , . . . xm , y1 , . . . yn ]. This uses the fact that Am × An = Am+n .
Analogue for projective space is FALSE: P m × P n 6= P m+n . The “obvious”
map between them taking (x0 : . . .) × (y0 : . . .) → (x0 : . . . : y0 : . . .) is simply
not well defined as multiplying the x’s by λ gives a different image.
In fact over the complex numbers, P 1 × P 1 and P 2 are not even homeomor-
phic as topological spaces: look at 2nd cohomology. Other differences: any two
curves in P 2 intersect (Bezout) but P 1 × P 1 has plenty of disjoint lines.
Instead, we can embed P m × P n as a projective variety inside P mn +m+n by
mapping (x0 : . . .) × (y0 : . . .) → (x0 y0 : . . . : xi yj : . . .) = (w00 : . . . : wij : . . .).
(Segre embedding) In terms of lines in vector space, this is given by taking the
tensor product of two lines in the tensor product of 2 vector spaces. What is

13
the ideal of its image? wij wkl = wik wjl . Check map is onto: Can assume that
w00 (say) is nonzero, so can assume it is 1. Then wkl = w0l wk0 = yl xk , so map
is onto.
For example, this identifies P 1 × P 1 with the nonsingular quadric wz = xy
in P 3 , which therefore has 2 rulings. (Any 2 nonsingular quadrics in P 3 are
isomorphic, and are ruled surfaces, sometimes used in architecture.)
Any nonsingular quadric in P 3 can be put in this form (pick norm 0 vectors);
for example the sphere x2 + y 2 + z 2 = 1 has 2 rulings by straight lines (over
complex numbers)! (x + iy)(x − iy) = (1 − z)(1 + z).

Example 32 Veronese surface is given by the points of the form (x2 : xy :

xz : y 2 : yz : z 2 ) in P 5 . Embedding of projective plane into P 5 . Equations:
wij wkl = wik wjl with wij = wji . Similar examples in higher dimensions.

Example 33 Example of a Grassmannian G(m, n) is m dimensional subspaces

of k m+n . G(m, n) = G(n, m) (duality of vector spaces) and G(1, n) is projective
space, so first nontrivial case is G(2, 2), the planes on k 4 or the lines in P 3 .
Can we find a projective variety whose points correspond naturally to this space
(and what does “naturally” mean?). Simplest nontrivial example of a HILBERT
SCHEME: parametrizes subschemes of projective space, this this case of degree
and dimension 1. (Note: lines in A3 do not form an affine or a projective
variety.)
We embed G(2, 2) in P 5 . Suppose a and b span a line in P 3 . Look at
 
a0 a1 a2 a3
b0 b1 b2 b3

Put sij = determinant of columns i and j. Then (sij ) is well defined in P 5

as changing 2 points of a line does not change rations of determinants. Is this
onto? No! lines in P 3 form 4-dim space, so there must be a nontrivial relation
between the sij . Relation is the Plucker relation

s01 s23 − s02 s13 + s03 s12 = 0

(proof: each term occurs twice with opposite signs). Check map is onto. Sup-
pose s01 (say) is 1. Then s23 is determined by other s. So point it the image
of  
1 0 s12 s13
0 1 s02 s03
So the set of lines in P 3 is isomorphic to a quadric in P 5 .
We can use  this to find cohomology
 of this quadric. The Grassmannian
1 0 ∗ ∗ 1 ∗ 0 ∗
is the union of etc, giving affine spaces of
0 1 ∗ ∗ 0 0 1 ∗
dimensions 0, 1, 2, 2, 3, 4, from which one can read off the cohomology, and the
number of points over finite fields. (Differs from cohomology of P 4 )

14
 Similar but more complicated argument shows that Grassmannians  G(m,n)
m+n
are intersections of quadrics in some projective space of dimension −
m
1. Details: Pick m vectors spanning subspace, which give a matrix of size
m+n
m × (m + n). Then the determinants pi1 ,...,im of all m × m minors
m
give projective space coordinates for a point of the Grassmannian. Find relations
between them:
X
0= (−1)λ pi1 ,...,im−1 ,jλ × pj1 ,...jλ−1 ,jλ+1 ,...,jm+1
λ

because each monomial occurs twice, with opposite signs. (Enough to check for
m × 2m matrix.)
Now need to check this map from Grassmannian to projective space is
ONTO. Image is covered by open affine subsets where some p is nonzero. Sup-
pose for example that p1,...,m = 1. Then we can find a point of the Grassmannian
with any given values of p1,...,r−1,r+1,...,m,s , in other words with at least m − 1
indices in 1, . . . , m, by choosing a matrix whose left columns form the iden-
tity matrix. But then the Plucker relations determine all the other p’s using
p1,...,m pi1 ,...,im =sum of terms with more indices in the set 1, . . . , m.
Grassmannians are particularly simple because they can be written as  dis- 
∗ ∗
joint unions of affine spaces. They can also be given as quotients: GLm+n / ,
0 ∗
which shows that their dimension is mn. In particular the quotient of two
AFFINE groups can be PROJECTIVE.
Used by Grothendieck in proof that components of Hilbert scheme are pro-
jective. Key idea: suppose a homogeneous ideal I = I0 + I1 + . . . of k[x1 , . . .] =
S0 + S1 + . . . has the property that dim(Ij ) = p(j) for j large and some poly-
nomial p (more or less the Hilbert polynomial: see later). Fix d sufficiently
large, so that in particular Id generates all larger I’s. Then the image of Id
in Sd is a point of the Grassmannian of p(d)-dimensional subspaces of a vector
space of dimension dim(Sd ). Relations defining the image of this subset: we
have the image of the map Id ⊗ Sj → Sd+j is contained in Id+j so has rank
at most p(d + j). Recall that linear maps of rank at most something form a
closed subset (a determinantal variety defined by lots of determinants being 0).
The condition that all these maps have rank at most something gives the ideal
defining the Hilbert scheme in the Grassmannian. (Actual construction more
sophisticated, because of the technical problem that all ideals generated by Id
of given dimension do not form a flat family. The restriction to ideals with
given Hilbert polynomial is needed because these form a better behaved (flat)
family.)
What does “natural” mean when we say that lines in P 3 is naturally iso-
morphic to this projective variety? Answer by Grothendieck: corresponding
functors from commutative rings to sets are isomorphic;
Ring R → lines in projective space over R

15
 Ring R → G(2, 2) over R.
Ring R → R-valued points of a projective variety in P 5
Isomorphic as functions: this means not only are F (x), G(x) isomorphic as
sets, but these isomorphisms commute with morphisms x → y.
This requires working over more general commutative rings. More generally,
any scheme determined up to isomorphism by its functor of points. Fundamental
question: given a functor, such as “Picard group” “Hilbert scheme” “isomor-
phism classes of abelian varieties”, is it represented by a scheme? Problem: it
is tricky to define the correct functors from rings.
Elementary examples of Hilbert schemes include hypersurfaces in projective
space, n points on a line, and Grassmannians. These examples are misleadingly
simple: in general Hilbert schemes seem to exhibit every imaginable sort of bad
behavior of projective schemes (they can be singular at all points of a component
for example).

Example 34 Hirzebruch [Link] quotient of (A2 − 0) × (A2 − 0) by Gm ×

Gm acting as (λ, µ)(s, t, x, y) = (λs, λt, µx, λ−a µy) for a an integer. There is
a map from F to P 1 taking (s, t, x, y) → (s : t). The fiber at any point is
isomorphic to P 1 so we get a fiber bundle, nontrivial unless a = 0. Fiber
bundle= “twisted product”; looks locally like a product, similar to a Moebius
band.

Example 35 Scrolls. F = F (a1 , . . . , an ) is the quotient of (A2 − 0) × (An − 0)

by Gm × Gm acting as (λ, µ)(s, t, x1 , . . . , xn ) = (λs, λt, λ−a1 µx1 , . . . , λ−an µxn ).
There is a map fromF to P 1 taking (s, t, x1 , . . . , xn ) → (s : t). The fiber at any
point is isomorphic to P n−1 so we get a fiber bundle, usually nontrivial. P
(ai + 1)-1
Assume all a’s
Ppositive. Embedding of F into projective space P i aj −i
by ratios of the (ai + 1) monomials (bihomogeneous polynomials) s t xj .
The image of each fiber is a linear subspace.
For n = 2 these are the Hirzebruch surfaces.
Quasiprojective varieties are covered by open affine varieties. It is unnatural
to assume all varieties should be embedded in projective space; this is rather
like demanding that all differentiable manifolds should be subsets of Euclidean
space. Weil defined abstract varieties as (roughly) things that can be cov-
ered by open affine varieties. Formal definition comes later using locally ringed
space as special case of schemes. Work slightly informally. Weil originally in-
troduced concept to construct abstract Jacobian varieties of curves over finite
fields, though these turned out to be projective.
Toric varieties. Any rational polyhedral cone C gives an affine variety,
with coordinate ring the group ring of the DUAL cone. (Taking duals means
maps of cones give maps of affine varieties.) Example: quadrant gives affine
space, 0 gives a torus, cone generated by (1, 1)(1, −1) gives singular variety
k[x, y, z]/(z 2 − xy). If we have a fan (collection of cones closed under taking
faces, intersections) we get a collection of affine varieties that can be glued

16
together using inclusions of cones. Example: projective space, P 1 × P 1 , weird
examples with an infinite number of cones (not of finite type, not quasicompact).
All complete abstract varieties that are 1-dim or 2-dim and nonsingular are
projective, but Hironaka gave example of a nonsingular complete 3-dim variety
that is not projective: see later. They don’t seem to be of much interest, and
Chow’s lemma says they are covered by projective varieties by a map that is
isomorphism over an open dense subset.

1.3 Morphisms
Background: recall definition of category. Examples: Sets, commutative rings,
abelian groups, differentiable manifolds. Key point: when defining mathemat-
ical objects, ask what the morphisms are. In algebraic geometry there are 2
sorts of morphism: regular maps and birational maps.
Regular functions on an affine variety = coordinate ring k[x1 , . . .]/I.
Regular function f on an open subset U of an affine variety V (quasiaffine
variety) are functions that are regular at all points p of U , meaning that f = g/h
in some neighborhood of p, where h is nonzero in this neighborhood.
We should check that this is compatible with the definition for affine vari-
eties! So suppose V = U1 ∪ U2 ∪ . . . and f is a function on V with f = gi /hi on
Ui with hi nonvanishing on Ui . Here the g’s and h’s are in the coordinate ring
of V , and we want to show that f is too. We have

1 = a1 h1 + . . . mod I

because the Ui cover V , so no max ideal contains all the h’s and I. This suggests
that
f = a1 h1 f + . . . mod I = a1 g1 + . . . mod I.
So we DEFINE f to be a1 g1 + . . . and check that hi f = a1 g1 hi + . . . = a1 gi h1 +
. . . mod I (because hi gj = hj gi mod I), which is gi mod I.
Similarly a function on a quasiprojective set (open subset of projective space)
is called regular if it is locally regular at all points, or equivalently regular on
all affine open subsets. This makes quasi projective varieties into RINGED
SPACES: regular functions form a SHEAF. (Define sheaves.)

Example 36 Regular functions on P 1 = A1 ∪A1 given by (f, g) with f ∈ k[x0 ],

g ∈ k[x1 ], that coincide on k[x, x−1 ]. So they are elements of k[x, x−1 ] that are
polynomials in both x and x−1 , so they are constants.

Definition 3 A morphism f : X → Y is a continuous map such that the

pullback of any regular function on any open U ⊂ Y is regular on f −1 (U ) ⊂ X.
(Same as morphism of ringed spaces)
Warning: for general schemes this definition is WRONG: should use mor-
phisms of LOCALLY ringed spaces, which for varieties happen to be same as
morphisms of ringed spaces.

17
 Examples of ringed spaces: manifolds with
top=C 0 , C 1 , C 2 , . . . C ∞ , C ω , algebraic structure
Morphisms in this → direction
Floppy Rigid
Local ring of a variety at a point p is lim→U 3p (regular functions on U ). ROUGHLY
functions defined near p. Check this is a local ring, with max ideal functions
vanishing at p. Suppose f 6= 0 at p. Then f = g/h with g 6= 0 at p, so g 6= 0 in
some neighborhood, so 1/f = h/g in this neighborhood is in the local ring.
So in fact a variety is a LOCALLY RINGED SPACE: this means stalks are
LOCAL rings. .
Moreover morphisms of varieties are automatically morphisms of locally
ringed spaces, meaning that pullback of something in a max ideal of a local
ring is in max ideal, in other words pullback of function vanishing at p vanishes
at f −1 (p). (Not always true for more general ringed spaces whose sections need
not be functions on a space.)

Example 37 There is a morphism from A1 to the cuspidal curve y 2 = x3 taking

t to (t2 , t3 ). Corresponding map off coordinate rings embeds k[t2 , t3 ] ⊂ k[t].
This is a homeomorphism of the underlying topological spaces but is NOT an
isomorphism of varieties.

Theorem 5 Suppose Y is AFFINE. (False if not affine). Then morphisms

from X to Y same as morphisms from O(Y ) to O(X) where O means regular
functions.
Proof Suppose ϕ ∈ Hom(X, Y ). Then ϕ∗ takes regular functions on Y to
regular functions on X so we get an element of Hom(O(Y ), O(X) (even if Y is
not affine). We need to construct an inverse Hom(O(Y ), O(X) ⇒ Hom(X, Y ).
Suppose h ∈ Hom(O(Y ), O(X) where O(Y ) = k[x1 , . . .]/I. Define ψ : X →
Y as follows. h(xi ) ∈ O(X), so if p ∈ X then h(xi )(p) ∈ k. Put
ψ(p) = (h(x1 )(p), . . .) ∈ k n
which defines a map ψ : X → k n . The image is in Y as h(I) = 0.
Check ψ is a morphism. xi ◦ ψ is regular on X for each coordinate function
xi on Y , so f (x1 , . . .) ◦ ψ = f (x1 ◦ ψ, . . .) is regular for any regular f . Easy to
check that map taking h to ψ is desired inverse map. 
Corollary: Category of affine algebraic sets over algebraically closed fields
k is equivalent to the opposite (define this) of category of finitely generated
reduced k-algebras. For example, (categorical) product of algebraic varieties
corresponds to coproduct=tensor product of the coordinate rings. Example:
A1 × A2 = A3 and k[x1 ] ⊗ k[x2 , x3 ] = k[x1 , x2 , x3 ]. Warning: this fails over
non-perfect fields, because the tensor product of two reduced finitely generated
algebras over a field need not be reduced. An example is given by taking a (non-
perfect) field k of characteristic p > 0 such as Fp (t) and forming the inseparable
extension K = k[t1/p ] where t has no pth root in k. Then K is a field so is
reduced, but K ⊗k K has nilpotent elements.

18
Example 38 Algebraic group G × G → G.
Ga : (x, y) → x + y This corresponds to a map k[x] ⊗ k[y] ← k[z] taking z to
x + y.
Gm : (x, y) → xy This corresponds to a map k[x, x−1 ] ⊗ k[y, y −1 ] ← k[z, z −1 ]
taking z to xy.
    
a1 b1 a2 b2 a1 a2 + b1 c2 a1 b2 + b1 d2
GL2 (k): = Coor-
c1 d1 c2 d2 c1 a2 + d1 c2 c1 b2+ d1 d2
dinate ring R is k[a, b, c, d]/(ad−bc−1). Corresponding map from ∆ : R → R⊗R
takes a to a1 a2 + b1 c2 , etc. Exercise: describe homomorphism from R to R cor-
responding to INVERSE of group.
(Coordinate rings of affine algebraic groups are commutative HOPF ALGE-
BRAS.)

Example 39 We check that the twisted cubic is isomorphic to P 1 . There is a

natural map from P 1 to the twisted cubic taking (s : t) to (s3 : s2 t : st2 : t3 )
and this is isomorphism of underlying top spaces, but this is NOT enough to
prove it is an isomorphism of varieties. We need to construct a morphism in
the other direction. The twisted cubic is covered by 2 affine subsets w 6= 0 and
z 6= 0 so we can define the inverse morphism on each of these two subsets and
check it is the same on their intersection.
w 6= 0 : (w : x : y : z) → (w : x)
z 6= 0 : (w : x : y : z) → (y : z)
As wz = xy these coincide on w 6= 0 ∧ z 6= 0.
Note that the corresponding graded rings k[s, t] and k[w, x, y, z]/(wz − xy, wx −x2 , xz −y 2 )
are NOT isomorphic!

Example 40 A1 − 0 is isomorphic to the affine variety xy = 1. However

the quasiaffine variety A2 −(0, 0) is not isomorphic to ANY affine variety. To see
this calculate its coordinate ring. Cover it by the 2 open affine subsets x 6= 0 and
y 6= 0. Then the Coordinate ring consists of pairs in k[x, x−1 , y] × k[x, y, y −1 ]
with same image in coordinate ring k[x, x−1 , y, y −1 ] of the intersection. This
is just k[x, y]. So maps from A2 − (0, 0) to affine varieties are “same as” maps
from A2 to affine varieties, so it cannot be affine as the natural map to A2 is
not an isomorphism.

Example 41 (Suggested by Arend Bayer on mathoverflow). The group GL2 (k)

acts in the obvious way on P 1 . If we fix a point of P 1 we get a morphism of
an affine variety onto a projective variety with affine fibers. In particular the
quotient of 2 affine groups need not be affine, and a fiber bundle with affine
fibers over a projective variety can be affine.

Example 42 Products of affine varieties. Recall product of objects of any

category. Product is unique up to unique isomorphism. (For example, there
are several ways to define the ordered pair in set theory, such as {a, {a, b}}
or { {a, 1}, {b, 2} } or {{a}, {a, b}}, but the all give canonically isomorphic
products of sets.) Product of affine varieties corresponds to coproduct=tensor

19
product of commutative rings. Notice that products and coproducts of objects
depend on the category. For example, the product of affine varieties does NOT
have the same topology as their product as topological spaces. Another random
example: the coproduct of 2 commutative rings is different in the categories of
commutative rings and non-commutative rings.

Example 43 Products of projective varieties. We want to check that the Segre

embedding gives the product P m × P n , which is unique up to canonical isomor-
phism. The proof is essentially trivial, but involves some bookkeeping and
unwinding definitions. We need to give maps from the Segre embedding to
P m and P n and then check it has the universal property. Recall the funda-
mental technique: cover by open affine subsets. Element of Segre embedding
is (z00 : . . . : zmn ) Define map to P m on (say) the open set with z00 6= 0 as
(z00 : z10 : . . . : zm0 ). On the open set with (say) z01 6= 0 the map is given by
(z01 : z11 : . . . : zm1 ). Check these two maps are same on the intersection: this
follows from the relations zij zkl = zil zjk . So we have a well defined morphism to
P m.
Now we have to check the universal property, so suppose we have a variety
V mapping to P m and P n . The first problem is that it is not that easy to
describe regular maps from V to projective space. We bypass this by covering
V with open affine sets such that each has image in one of the standard open
affine sets covering projective space, say z0 6= 0. So first assume V is affine.
Then a morphism to the open affine subset z0 6= 0 of P m is given by regular
functions (f0 = 1 : f1 : . . . : fm ) on V . Similarly a morphism to P n is given by
(g0 = 1 : g1 : . . . : gn ). Then we can define a morphism to the Segre embedding
by (f0 g0 : . . . : fm gn ) and check this has required properties (image is in Segre
embedding, etc.) For general V , the morphisms on sets of an open affine cover
are compatible on intersections by the uniqueness property of the product, so
they fit together to give a morphism to the image of the Segre embedding.
The Segre embedding is really a combination of 2 different operations: a
product of projective spaces as an abstract variety, together with an embedding
of this abstract variety into projective space. The embedding into projective
space is really irrelevant, and it is easier to define the abstract product of
varieties and forget about projective embeddings: this is what we will do for
products of schemes.

Example 44 Automorphisms of affine space. For A1 these correspond to au-

tomorphisms of k[x], which just map x → ax +b for nonzero a, so we get a
2-dimensional nonabelian group. For An with n > 1 the automorphism group
is MUCH larger. We can still map x → Ax +B for matrices A, B with A, B
invertible, but there are many other automorphisms not in this subgroup: for
example x → x, y → y + p(x) for any polynomial p, giving an infinite dimen-
sional abelian subgroup. Suppose we have a morphism of affine space given by
xi → fi (x1 , . . . , xn ). Look at the Jacobian ∂fi /∂xj . Then Jacobian of FG is
jacobian of F times Jacobian of G, so the jacobian of an automorphism is an

20
invertible polynomial, in other words a nonzero constant. Jacobian conjecture:
does the converse hold? (There have been many incorrect proposed solutions).

Example 45 Morphisms of the projective line P 1 to itself. Any morphism re-

stricts to a map from a (possibly empty) open set of A1 to A1 , which is either
empty or a rational function on A1 . Conversely rational functions on A1 extend
to morphisms of P 1 , so the morphisms of P 1 correspond to rational functions
together with ∞. The ones with inverses are just the ones of degree 1, corre-
sponding to the group PGL2 (k). The automorphisms of the affine line and the
projective line over the complex numbers are the same as the automorphisms of
the complex plane and the complex sphere in complex analysis, and morphisms
of the projective complex line to itself are the same as morphisms from the com-
plex sphere to itself. (However morphisms from the affine line to itself are not
the same as morphisms from the complex plane to itself.) This is a special case
of Serre’s GAGA: roughly speaking, in the projective setting analytic things
tend to be algebraic.

Example 46 Consider the image of the morphism from A2 → A2 taking

(x, y) → (x, xy). The image is plane −y axis + origin so is not affine or locally
closed.

Example 47 Ax-Grothendieck theorem. Suppose f is an injective morphism

from a variety to itself over an algebraically closed field, Then f is surjective.
(This is false over Q: consider x → x3 .) Proof:
1. Trivial over finite fields: any injective map on a finite set is surjective.
2. Trivial over algebraic extensions of finite fields (such as the algebraic closure):
take the finite field generated by the coefficients of the equations defining the
variety, the map f , and a point in the variety.
3. Now use fact that a 1st order statement is true for algebraically closed fields
of characteristic 0 if and only if it is true over algebraically closed fields of
large prime characteristic.
This is related to the Lefschetz principal. Suppose S is a statement in the first
order language of fields. Then the following are equivalent:
1. S is true in the complex numbers
2. S is true in some algebraically closed field of characteristic 0
3. S is true for some algebraically closed fields of arbitrarily large characteristic.

4. S is true for all algebraically closed fields of sufficiently large characteristic

The key point is that the theory of algebraically closed fields of given character-
istic is complete, which follows from the fact that it is categorical in uncountable
cardinals. So any two algebraically closed fields of the same characteristic are

21
indistinguishable using first order statements. Characteristic 0 is defined by the
statements 2 6= 0, 3 6= 0, . . . and any proof uses only a finite number of these, so
applies to all algebraically closed fields of sufficiently large characteristic.
This gives two powerful methods for proving things for all algebraically closed
fields in characteristic 0:
Use analysis, Riemannian geometry, hodge theory etc. Then apply Lefschetz
principle.
Prove for algebraic closures of finite fields, using counting arguments, Weil
conjectures, Frobenius endomorphism, etc. (Famous example: Mori’s bend and
break argument.)

1.4 Rational maps

Suppose Y is affine variety. As Y is irreducible we can take quotient field K(Y )
of integral domain O(Y ). Elements are called RATIONAL FUNCTIONS on Y .
No analogue for smooth manifolds: characteristic of algebraic geom. Analogous
to meromorphic functions on Riemann surfaces/complex manifolds.
Definition fails for projective varieties as O(Y ) is too small. So instead we
define the ring of rational functions to be
(regular functions on dense open set of Y )/(equivalent if equal on intersec-
tion).
(Note that any 2 dense open sets have dense intersection, and if Y is irre-
ducible all nonempty open sets are dense, and in this case the rational functions
form a field as the points where a nonzero function is nonzero is dense.)
So K(Y ) = lim→U dense open O(U ).
Similarly define a rational map from X to Y to be given by a morphism on
a dense open set, modulo obvious equivalence relation. So rational functions
are same as rational maps to A1 . Rational maps do NOT form morphisms of
a category: composition need not be defined, as the image of a rational map
may have dense open complement! To fix this define a rational map to be
DOMINANT if its image of some dense open set is dense. Two varieties are
called BIRATIONAL if there is an invertible rational map from one to the other.
This is a cruder equivalence relation than isomorphism. (For smooth metrizable
manifolds, any 2 of same dimension are “birational”.)

Example 48 The varieties A1 , P 1 , xy = 1, x3 = y 2 are all birational, but no

2 are isomorphic. Similarly P 1 × P 1 , P 2 , A2 , A2 − (0, 0)

Example 49 We show that that affine line is not birational to the elliptic curve
x3 + y 3 = 1. and more generally that there is no dominant map from A1 to this
curve. Algebraic proof: Suppose x(t)3 + y(t)3 = 1 for rational functions x,
y of t. Clearing denominators we get f (t)3 + g(t)3 = h(t)3 . Factoring we get
(f + g)(f + ωg)(f + ω 2 g) = h3 . As polynomials form a UFD and every unit is a
cube, the three terms on the left are all cubes, say f + ω k g = h3k . Eliminating f
and g from these 3 equations gives a linear relation between the h3k , which have

22
smaller degrees than f , g, h, so by induction on the degree there is no solution
in positive degree polynomials. (Same proof works for any exponent at least 3.
Note that over the rationals, zero degree is FAR harder than positive degree!)
Shorter proof using more algebraic geom: there is no non-constant map from a
curve to a curve of higher genus.

Example 50 We show the affine line is not birational to some elliptic curves
y 2 = 4x3 − g2 x − g3 using complex analysis/topology. Use Weierstrass ℘ to
construct an isomorphism from C/Λ to the curve. Recall that
X 1
℘(z) =
(z − λ)2
λ∈L

which is doubly periodic, except that this does not converge so we regularize it
by
1 X 1 1
℘(z) = 2 + − 2
z (z − λ)2 λ
λ∈L−0

This is not clearly doubly periodic, but its derivative is as the series then con-
verges. So ℘ is periodic up to “constants of integration”, and the fact that ℘ is
even implies that these constants all vanish.
Its Laurent expansion at 0 is z −2 +O(z 2 ) so ℘0 (z)3 = 4z −6 +(?)z −2 +(?)z 0 +
. . . so we get
℘0 (z)2 = 4℘(z)3 − g2 ℘(z) − g3
because both sides are doubly periodic with no poles and vanishing constant
term. So the map taking z to (℘(z), ℘0 (z)) maps C/Λ to the projective comple-
tion of the elliptic curve y 2 = 4x3 − g2 x − g3 .
Reason for name elliptic curve/function etc: we have
Z ℘
d℘
z= p
3
4℘(z) − g2 ℘(z) − g3
and the integral on the right is an elliptic integral related to finding the arc
length of an ellipse.
Then S 2 −finite number of points is never homeomorphic to C/Λ−finite number of points
so P 1 cannot be birational to y 2 = 4x3 − g2 x − g3 . The group law on the elliptic
curve is obvious in the analytic setting as it is a quotient of two abelian groups.
There is a similar construction for singly periodic functions:
X 1
= π cot(πz)
z−λ
λ∈Z

if the left hand side is regularized in a suitable way. This is the logarithmic
derivaative of Euler’s product formuls for the sine function, and there is an
analogue of this product formula for the Weierstrass function, leading to the
theory of theta functions.

23
Example 51 Cubic surfaces are usually rational. Informal argument: Take 6
points in general position in P 2 so that the space of cubics vanishing on all
of them is 10 − 6 = 4 dimensional, basis f1 , f2 , f3 , f4 . This gives a map from
P 2 − 6 points to P 3 ; image is some hypersurface. To find its degree, look at
number of intersection points of image with some line, say f1 = f2 = 0. There
are 3 × 3 = 9 points of intersection of these 2 cubics, consisting of the 6 points
we chose + 3 others. The images of these 3 other points are the 3 points of
intersection of the surface with the line, so degree=3. Now count dimensions:
space of cubic surfaces has dimension 20 − 1 = 19 (projective space). Dim of
space of 6 points =6 × 2 = 12. Subtract automorphisms of P 2 (dim 8) and add
dim aut(P 3 ) = 15 to get dim 12 − 8 + 15 = 19. So spaces have same dimension,
so that most cubic surfaces arise in this way.
27 lines on cubic surface come from: 6 blown up points + 15 lines through
2 points + 6 quadrics through 5 points.
Argument is rather sloppy (typical of old Italian style algebraic geom): for
example, what does “general position” mean for the 6 points? A. No 3 on line,
no 6 on conic, but this takes more work.

A variety is called unirational if it is finitely covered by a rational variety.

These are hard to distinguish from rational varieties. In 1 and 2 dimensions
unirational implies rational (Luroth, Castelnuovo) and for many years it was an
open problem to find any examples of unirational varieties that are not rational.
Clemens and Griffiths showed that a cubic three-fold is unirational but in general
not a rational variety

1.4.1 Blowing up
Blowup of An (0,0,..0) at a point means replace point by a copy of projective
space. Given by points (x1 , . . . , xn ) × (y1 : . . . : yn ) ∈ An × P n−1 with xi yj =
xj yi . Projection to An is an isomorphism except at the origin, where inverse
image is P n−1 . It is proper as it is a closed subset of An × P n−1 .
P n−1 is covered by affine spaces such as y1 6= 0. On this open subset we can
put y1 = 1 so xi = yi x1 : we get new coordinates x1 , y2 , . . . , yn . So blowing up
means roughly divide x2, . . . , xn by x1 , and same for other coordinates.
Example: y 2 = x3 . Blowup: put y = xt so get x2 t2 = x3 . Get x = 0,
x = t2 as 2 components. x = 0 is the exceptional curve. Can also put x = sy
so we get y 2 = y 3 s3 to get exceptional curve y = 0 and a curve 1 = ys. Note
result is NONSINGULAR (see next section...) so blowing up has resolved the
singularity.
Example: x2 + y 2 = z 2 Conical singularity DIAGRAM. Blowup: x = zs, y =
tz so we get z 2 (s2 + t2 − 1) = 0: Cylinder.
Example: y 8 = x5 Blowup x = yt gives y 3 = t5 . Blowup using y = st gives
s = t2 . Third blowup makes it nonsingular.
3

Example: xy 2 = z 2 (Pinch point or Whitney umbrella.) Blowing up along

the “worst” singularity at the origin just reproduces the same singularity. In-
stead blow up along line y = z = 0. (x, y, z) × (s : t) ∈ A3 × P 1 with yt = sz.

24
Take (say) s = 1 to get z = ty, so (x − t2 )y 2 = 0. This shows one cannot resolve
singularities by repeatedly blowing up along subvariety of “worst” singularities,
at least not using naive definitions.
Example: Blowing up real affine plane at a point gives real projective plane
minus a point, as a line bundle over projective line = circle, in other words a
Moebius band without its boundary. Note that blowing up can turn an ori-
entable manifold into a non-orientable one.
More generally can blow up along points or subvarieties or sheaves of ideals
or sheaves of graded algebras. (Example: blowing up a point along the sheaf
of graded algebras k[x0, . . .] gives projective space.) Meaning is roughly replace
each point by some projective variety, such as projective space of the normal
space at a point, or the projective variety of a graded algebra. Blowup pulls
apart the different normal vectors. Hironaka used repeated blowups to con-
struct a PROPER birational map from a nonsingular variety to a char 0 variety.
Idea is roughly that a singular variety is a projection of a nonsingular variety
in a higher dimensional space.
Blowup over ideal: map (x1 , . . . , xm ) → (x1 , . . . , xm ) × (g1 : . . . : gn ) ∈
Am × P n−1 where the g’s are some polynomials. Defined for points not in the
variety defined by the ideal (g1 , . . .). Take Zariski closure of points of X not in
this set. Up to isomorphism this only depends on the ideal generated by the g’s
and gives the blowup with center the ideal (or subscheme). Special case gi = xi
is just blowup at origin.
Example: Blow up affine plane along ideal (x, y 2 ) has coordinate rings
k[x, y, y 2 /x] and k[x, y, x/y 2 ]. The second is a polynomial ring, but the first
is singular y 2 = xz (conical singularity). Similarly blowing up along the ideal
(x2 , y 2 ) produces a pinch point. Blowing up can be used to resolve singularities,
but can also create new singularities. (This is why when blowing up along a
subvariety one tries to use nonsingular subvarieties.) Blowing up along nonre-
duced subscheme like this is not always bad: Hironaka’s theorem implies that
in char 0 one can resolve any singularity by blowing up along some possibly
nonreduced subscheme. Open problem: find a direct construction of such a
subscheme. Blowing up along nonreduced subschemes seems a powerful and
dangerous tool, but is not well understood.

1.4.2 The Atiyah flop

Flop: special sort of birational map used in classification of varieties of dim at
least 3. First example found by Atiyah in 1958 as follows. Look at xy = zt in
A4 , a 3-dim variety with a sing at origin. Blow up at 0. Exceptional variety is
XY = ZT in P 3 . This resolves singularity as along (say) X = 1 it looks like
Y = ZT in coordinates x, Y, Z, T . Exceptional variety is isomorphic to P 1 × P 1
which can be projected onto P 1 in 2 ways. So we can “half blow up” singularity
along line y = t = 0 by mapping it to A4 × P 1 as xy = zt, xZ = zX, tZ = yX.
If (say) X = 1 this becomes y = tZ, z = xZ, xy = zt, which reduces to affine
3-space so is nonsingular. We can also half blow it up to get xy = zt, xT = tX,
zT = yX. The rational map between these two spaces is the Atiyah flop: it is

25
a birational map between nonsingular 3-folds that changes a codimension 2 P 1
to a different P 1 and is otherwise an isomorphism. This also shows that in 3
dims there is in general no nonsingular “minimal” resolution (though there is
in dim ¡3). To find a minimal model in dimension 3, allow mild singularities,
called terminal singularities.

1.5 Nonsingular varieties

1.5.1 Tangent spaces
Tangent space of V with ideal generated by fi at (0, . . . , 0) given by vanishing
of LINEAR parts of fi . Origin is called a singular point if tangent space has
“wrong” dimension > dim(V). For other points change variable to make point
at origin. Dim of tangent space given by n − rank(∂fi /∂xj )
Example: f (x, y)nonsingular at 0 if some linear term nonzero. Converse true
if f REDUCED but not in general.
Example: For hypersurface, singular if f = ∂f /∂xi = 0. Example: y 2 =
x + x2 − x − 1 singular at (−1, 0).
3

Set of singular points is CLOSED. Proof: Condition for a matrix to have

rank < n − dim(V ) is a closed subset of all matrices. (Determinantal variety!)
Set of nonsingular points is NONEMPTY.
Apparent counter-example: x3 + y 3 = 1. nonsingular at all points except
in char 3 where ALL points are singular! This does not contradict result that
nonsingular points are dense, because in char 3 x3 + y 3 − 1 = (x + y − 1)3 .
Schemes CAN be singular at all points.
Proof: Reduce to case of hypersurface f = 0, as every variety is birational
to a hypersurface. If all points are singular then all derivatives of f vanish
whenever f vanishes so are divisible by f . As f is irreducible and derivatives
have lower degree, all derivatives vanish. This does NOT in general imply that
f is constant! But does imply that f is p’th power of some thing, so as f is
irreducible it must be constant, contradiction.
As nonsingular points are open and nonempty, they are DENSE.
At nonsingular points variety is smooth manifold over reals or complex num-
bers. (Warning: y 2 = x3 is a topological manifold even though it is singular at
(0, 0)).
Problem: Tangent space seems to depend on embedding of variety into affine
space. Need to find INTRINSIC definition of tangent space. A: It is given by the
Zariski tangent space (m/m2 )∗ where m is the max ideal of the local ring, so that
(m/m2 ) is a vector space over k. To see this it is enough to check it for the point
(0, . . . , 0) for a variety whose ideal is generated by the polynomials fi . We ob-
serve that (m/m2 ) is (x1 , . . . , xn )/(terms of degree > 2, fi ) which can be identi-
fied with the “cotangent space” (vector space with basis x1 , . . . , xn )/(linear partsof
fi ) whose dual is just the subspace of k n on which the linear parts of the fi
vanish, in other words the tangent space.
Another viewpoint: Zariski tangent space of local ring over a field k equal
to the residue field = homomorphisms to k[ε]/(ε2 ). Geometrically this is maps

26
from Spec(k[ε]/(ε2 )) which is a sort of point with an infinitesimal line coming
from it; an analogue of a “short smooth curve” in differential manifolds. (Note
use of non-reduced scheme.) This analogue fails for tangent space of Spec(p-adic
numbers).
Yet another viewpoint: consider tangent and cotangent vector fields on a
differentiable manifold. These are modules over the ring R of smooth functions,
and there is a linear derivation d from R to cotangent vector fields. Define
the module M of cotangent fields over a ring to be the image of the universal
derivation (generators df, relations d(f + g) = df + dg, d(f g) = f dg + gdf (Leib-
niz)). If we do this for the coordinate ring of an affine variety we get module
of cotangent fields. Example: for k[x1 , . . . xn ] this is a free module with basis
dx1 , . . . dxn . Module of tangent vector fields = dual Hom(M, R). Cotangent
space of point can be reconstructed by (localizing at that point and) taking
quotient by maximal ideal.
Direct construction of cotangent module: let I be kernel of R ⊗ R → R and
put M = I/I 2 (with R acting on left component), dr = 1 ⊗ r − r ⊗ 1. Check d is
a derivation: 1⊗ab−ba⊗1 = a(1⊗b−b⊗1)+b(1⊗a−a⊗1)+(1⊗a−a⊗1)(1⊗
b − b ⊗ 1). Check universal property: Suppose we have a derivation d to some
module PM . Define fP: R ⊗ R → M by f (aP ⊗ b) = [Link] this vanishes P on
I 2: P
if si ⊗ ti ∈ I,P uj ⊗ vj ∈ I ∈ I, Pthen P si t i = 0, P uj v j
P = 0 so f (( si ⊗
ti )( uj ⊗ vj )) = si uj d(ti vj ) = si dti uj vj + si ti uj dvj = 0 so f
vanishes on I 2 so gives a map from I/I 2 to M . This construction will later be
used for constructing the cotangent sheaf of a scheme. Geometric interpretation:
tangent space something like infinitesimal neighborhood of diagonal V in V ×V .
(Compare with tangent microbundle of a topological space.)
Definition: A local ring is called REGULAR if its dimension is dim(m/m2 )
and a variety is called NONSINGULAR at a point if the local ring is regular.

1.5.2 Example: resolution of E8 Du Val singularity x2 + y 3 + z 5 = 0

A du Val singularity, also called simple surface singularity, Kleinian singularity,
or rational double point or canonical singularity in dim 2, is a quotient of C 2
by action of finite group, such as binary icosahedral group of order 120 (inverse
image of A5 in unit quaternions) (Klein). Ring of invariant generated by 3
elements x, y, z satisfying the relation above. For example, for cyclic group of
order n acting as X → ζX, Y → ζ −1 Y , ring of invariant is generated by x =
X n , z = XY, y = Y n satisfying the relation z n = xy = (x + y)2 /4 − (x − y)2 /4,
so is of type An−1 .
cyclic An : x2 + y 2 + z n+1 = 0
dihedral Dn : x2 + zy 2 + z n−1 = 0
tetrahedral E6 : x2 + y 3 + z 4 = 0
octahedral E7 : x2 + y 3 + yz 3 = 0
icosahedral E8 : x2 + y 3 + z 5 = 0
Assume char=0. Only singularity is at (0,0,0).

27
 Blow up using (x1 : y1 : z1 ) as coordinates for P 2 . Cover P 2 by 3 copies of
2
A and check for singularities:
• x1 = 1: x2 + y13 x3 + z15 x5 = 0 → 1 + y13 x + z15 x3 = 0 Sing if also x = 0,
not possible.
• y1 = 1 : x21 y 2 + y 3 + z15 y 5 = 0 → x21 + y + z15 y 3 = 0 Sing if also 2x1 = 0,
5z14 y 3 = 0, 1 + 3z15 y 2 = 0, no possible.
• z1 = 1 : x21 z 2 +y13 z 3 +z 5 = 0 → x21 +y13 z +z 3 = 0. Singular if also 2x1 = 0,
3y12 = 0, y13 + 3z 2 = 0, so singular if all coordinates 0.
So we now have to resolve singularity of x21 + y13 z + z 3 = 0 at origin.
As before, blow up introducing coordinates (x2 : y2 : z2 ) with x2 z = z2 x1 ,
etc.
• x2 = 1: Nonsingular
• y2 = 1: x22 y12 + y14 z2 + y13 z23 = 0 → x22 + y12 z2 + y1 z23 = 0 Sing at (0,0,0)
• z2 = 1 : x22 z 2 + y23 z 4 + z 3 = 0 → x22 + y23 z 2 + z = 0 Nonsingular.
So we now have to resolve singularity of x22 + y12 z2 + y1 z23 = 0 at origin. Not
really clear what we have gained as this seems just as complicated as what we
started with! This shows that measures of the complexity of a singularity have
to detect quite subtle properties.
Introduce (x3 : y3 : z3 ).
• x3 = 1: Nonsingular
• y3 = 1: x23 y12 + y13 z3 + y14 z33 = 0 → x23 + y1 z3 + y12 z33 = 0 Sing if also
2x3 = 0, z3 + 2y1 z33 = 0, y1 + 3y12 z32 = 0 which forces all to be 0.
• z3 = 1 : x23 + y32 z2 + y3 z22 = 0 so also 2x3 = 0, 2y3 z2 + z22 =0, y32 + 2y3 z2 = 0
so sing is at (0, 0, 0)
So there are TWO singularities at (x3 : y3 : z3 ) = (0 : 1 : 0), (0 : 0 : 1).
First look at the singularity x23 +y1 z3 +y12 z33 = 0 at (x3 : y3 : z3 ) = (0 : 1 : 0).
Blowup introducing (x4 : y4 : z4 ).
• x4 = 1: 1 + y4 z4 + x33 y42 z43 = 0. Sing if also 3x23 y42 z43 = 0, y4 + 3x33 y42 z42 = 0,
z4 + 2x33 y4 z43 = 0 No solutions, so nonsingular.
• y4 = 1: x24 y1 + z4 + y14 z43 = 0, 2x4 y1 = 0, x24 + 4y13 z43 = 0, 1 + 3y14 z42 = 0.
No solutions.
• z4 = 1 : Similar to above case: no solutions.
So result of blowing up is NONSINGULAR. For later use, not that this applies
to anything of the form x2 + yz + yz(polynomial in y and z).
Now look at the other singularity, x23 + y32 z2 + y3 z22 = 0 at (x3 : y3 : z3 ) =
(0 : 0 : 1). Blowup introducing (x5 : y5 : z5 ).

28
 • x5 = 1: 1 + x3 y52 z5 + x3 y5 z52 = 0 Nonsingular
• y5 = 1: x25 + y3 z5 + y3 z52 = 0 Singular if also 2x5 = 0, z5 + z52 = 0,
y3 + 2y3 z5 = 0. Two singularities, at x5 = 0, y3 = 0, z5 = 0 or −1.
• z5 = 1 : Similar to previous case: Two singularities, at x5 = 0, z2 = 0,
y5 = 0 or −1.
So there are THREE singularities at (x5 : y5 : z5 ) = (0 : 0 : 1) or(0 : 1 :
0) or(0 : 1 : −1). Fortunately each of these 3 singularities looks like x2 + yz +
yz(monomial in y, z) so can be resolved with 1 further blowup.
Summary: altogether we needed 8 blowups, using a total of 27 variables,
as each blowup introduced 3 new variables. Problem with repeated blowups:
notation becomes a mess. Some authors reuse x, y, z each time, though this
becomes confusing. In general things can be a lot more complicated: for ex-
ample, there may be singular sets of dim > 0 which require blowing up along
more complicated varieties (see below), and even if there are non to start with,
blowing up a singular point may produced a singular set of dim > 0.
Remark: the intersection of v 2 + w2 + x2 + y 3 + z 5+6k = 0 with a small
sphere around the origin in C 5 is one of Milnor’s manifolds homeomorphic but
not diffeomorphic to S 7 .

Example 52 x4 + y 4 = z 2 . Used by Fermat to solve Fermat’s last theorem

for exponent 4. Only singularity is at origin: dimension 0. Blow it up using
(x1 : y1 : z1 ) as coordinates for projective space. Nonsingular along z1 = 1. But
along y1 = 1 we get x41 y 2 + y 2 = z12 which is singular along the LINE z1 = 0,
y = 0. So blowing up has increased the dimension of the singular set. (Blowing
up along this line resolves the singularity.)

Example 53 Blowups with a poor choice of center can even make a singularity
worse. x2 − yz = 0 blown up along line y = z = 0 gives x2 − y 2 z = 0. Here
the center of the blowup sticks out from the variety. Can even increase the
multiplicity of a singularity such as w2 x2 = y 2 z 2 .

Example 54 Why do people care about resolution of singularities? Many re-

sults use it for proofs. Example: analytic continuation of power |f |s of a poly-
nomial f . (Special case: analytic continuation of Γ.) Idea (Atiyah): resolve
singularity of polynomial so that inverse image has only normal crossing singu-
larities, so in local coordinates polynomial looks like xn1 1 xn2 2 . . . . which is easy
d s
to do as we can continue xs using dx x = sxs−1 (as in proof for Γ). Since the
map from the resolution is proper, we can push forward distributions, as the
pullback of a compactly supported function is still compactly supported.
Hironaka’s theorem is not really necessary: Bernstein found a more elemen-
tary proof using Bernstein polynomials that avoids it. Idea is to find a differ-
entiable operator D and a (Bernstein) polynomial b such that Df (x1 , . . .)s =
b(s)f (x1 , . . .)s−1 .

29
 The Malgrange Ehrenpreis theorem is an immediate consequence: By taking
Fourier transforms, finding a fundamental solution of a constant coefficient dif-
ferentiable operator is equivalent to finding a distributional inverse of a polyno-
mial f . But this is given by the constant term of the meromorphic distribution-
valued function f s of s at s = −1.
√
Example √ 55 Number fields. Z[ −3] is singular
√ at prime 2. Look at local
ring√Z2 [ 3]. Max ideal m generated
√ √by 2, −3 − 1. Square m2 generated by
2
4, 2 3 − 2. Maps onto Z/4Z[ 3]/(2 3 − 2) of order 8. So √ m/m has order
4 and tangent space has dimension 2. Integral closure is Z[( −3 + 1)/2]. This
times m/m2 has order 2.

1.5.3 Completions
Examples: Completion k[x] is k[[x]], completion of local ring Z(p) is Zp .
Completion R̂ of local ring R is inverse limit of R/mn = completion in Krull
topology.
Map from R to R̂ injective (∩mn = 0, Krull topology Hausdorff) if R Noethe-
rian but not in general even for natural examples: example local ring of germs
of SMOOTH functions near 0, or even worse k[[x1/n ]], where m = m2 .
S
Key property is Hensel’s lemma, stating that solutions in k can often be lifted
to solutions in R̂, by lifting to R/m, R/m2 , R/m3 , . . .. Many variations. Typical
example: If f0 (Z) = g0 (Z)h0 (Z) and g0 , h0 ∈ k[Z] are COPRIME then this lifts
to a factorization f = gh in R̂[Z]. Proof: suppose we have a factorization mod
mn . Want to lift this to a factorization mod mn+1 . Need to find a, b ∈ mn /mn+1
with g0 a + h0 b = something, which has a solution as g0 , h0 are coprime.
Application: y 2 = x3 + x2 . Take R to be local ring at 0. Then Z 2 = x + 1
has roots Z = ±1 in k, which lift to solutions in R̂ if char 6= 2, so y 2 − x2 (x + 1)
factorizes as (y − x + . . .)(y + x + . . .). Note that this FAILS in char 2; g0 , h0
are no longer coprime. So curve looks analytically like xy = 0 except in char 2
where it has a cusp.
Nasty properties of completions: completion of an integral local ring need
not be integral (see above); completion of a reduced local ring need not be
reduced. Henselization much better.

1.5.4 Elimination theory

Abhyankar’ slogan: “Eliminate the eliminators of elimination theory”
Briefly mentioned in theorem 5.7A, but very important. Problem: eliminate
y from x3 y 4 − 7x2 + y 6 − xy 8 and 3x2 y 5 + 4y 2 + 7x6 + x4 y 7 . Intersection of 2
plane curves. Some degree 99 polynomial in x. We want to find an EXPLICIT
FORMULA for it!
General problem: Given polynomials f (x) = am xm + . . . and g(x) = bn xn +
. . ., what is condition for them to have a common root? If they do then
f (x)p(x) = g(x)q(x) where deg(p) < n, deg(q) < m for some p, q as we can take

30
p = g/(x − α), q = f /(x − α). This is a set of linear equations so condition for
a nontrivial solution is
am am−1 am−2 ...
0 am am−1

a0 0
0 a1 a0 = 0n rows for a0 s, m rows for b0 s
bn bn−1 0 0
0 bn
b0 0
0 b1 b0
Q
(Determinant of Sylvester matrix) Also equal to const × (α − β).
More precisely, this condition is equivalent to f and g either have a common
root or both have zero leading coefficients (“common root at infinity”). An-
other way of putting this is that HOMOGENEOUS polynomials have a common
nontrivial solution. This determinant is called the RESULTANT of f and g.
(Compare Bezoutiant, catalecticant, determinant, harmonizant, canonizant,...)
Example: What is condition for f to have a multiple zero (or leading co-
efficient 0)? A. f and f 0 have common root, so resultant of f and f 0 is zero.
Example for cubic x3 + bx +c
1 0 b c 0
0 1 0 b c
3 0 b 0 0 = 4b3 + 27c2
0 3 0 b 0
0 0 3 0 b
Geometric meaning: Consider f and g to be homogeneous polynomials with
coefficients in k[y1 , . . . , yk ] Then f and g define hypersurfaces Hf , Hg in Ak ×P 1 .
Resultant gives projection of Hf ∩Hg in Ak . In particular IMAGE OF CLOSED
SET Hf ∩ Hg IS CLOSED!
More generally we want to show that Y × Z → Y is closed for any projective
variety Z. (Analogue for Ak × A1 → A1 is false (xy = 1 projected on x does
not have closed image. Over reals it is not even true that a polynomial maps
from Rn → R has closed image: for example x2 + (xy − 1)2 .) This is the ana-
logue of compactness for projective varieties. (Usual definition of compactness
useless, as all affine varieties are compact.) Recall concept of proper maps for
topological spaces: X → Y is called proper if it is continuous and universally
closed. Equivalent to continuous+closed+fibers compact. For locally compact
Hausdorff spaces this is equivalent to continuous + inverse image of compact is
compact. For Hausdorff spaces, compact is equivalent to (projection to a point
is proper). So we define a map of varieties to be proper if it is universally closed
(for Zariski topology on products!)
Now check that P 1 → point is proper (analogue of saying P 1 is compact
in complex topology). Sufficient to show P 1 × Ak → Ak is closed. Sup-
pose that a closed set S in P 1 × Ak is given by the zeros of f1 , f2 , . . ., where

31
each is a polynomial in X, Y, Z1 , . . . , Zk , homogeneous in X, Y . Look at re-
sultant of t1 f1 (X, Y, Z1 , . . .) + t2 f2 (X, Y, Z1 , . . .) + . . . and s1 f1 (X, Y, Z1 , . . .) +
s2 f2 (X, Y, Z1 , . . .) + . . .considered as homogeneous polynomials in X and Y .
This is a polynomial in s1 , s2 , . . . t1 , t2 , . . . Z1 , . . . , Zk . Then the projection of S
is given by the vanishing of all coefficients of sα tβ . So the image of S is closed.
Now we want to show that P n → point is proper. This would be trivial by
induction on n if P n = P n−1 × P 1 , but this is not true. It is close to being
true: correct version is: blowup of P n at a point is a (nontrivial) P 1 -bundle
over P n−1 . To see this look at graph Z of correspondence from P n → P n−1 .
Z is the set of pairs ((x0 : . . . : xn ), (y1 : . . . : yn )) with xi yj = xj yi . The
map Z → P n is an isomorphism except that it maps a whole P n−1 to (0:..:0:1)
so is blowup of P n at this point. Moreover Z → P n−1 is locally a projection:
over the hyperplane yi = 1 we have xj = xi yj so if we map ((x0 : . . . : xn )
to (x0 : xi ) × (x1 , . . . , xi−1 , xi+1 , . . .) we see that over the hyperplane we get a
projection from P 1 × An−1 → An−1 .
Since the map from Z → P n−1 looks locally like P 1 × An−1 → An−1 it is
proper. So the map from P n to a point is proper by induction.
Note use of blowup to change a rational map into a regular map defined
everywhere: common technique.
Remark: For n = 2 the surface Z is an example of a ruled surface over P 1 ,
not isomorphic to either P 2 or P 1 × P 1 . (Hirzebruch surface.)
Alternative short proof using Determinantal varieties. Suppose f1 , f2 , . . .
are homogeneous polynomials in Z1 , . . .. We want to show that the condition
that they have a common zero is closed in their coefficients. But they have
no common zero if and only if their ideal contains (Z1 , . . .)d for some d by the
Nullstellensatz. For each fixed d the condition that linear combinations of the f
contain all degree d monomials in the Z’s just says that a certain linear map with
coefficients that are polynomials in the coefficients of the f ’s is onto. But by
the theory of determinantal varieties being onto is a (horrendously complicated)
open condition on the linear map. So the points where the f have no common
zero is a union over d of open subsets, so is open. (Nonconstructive as we take
a union over d; can be made constructive by estimating max necessary d.)
These proofs suggest that although the image of a closed set is closed, it can
be VERY complicated to describe explicitly as the dimension grows: the proofs
require more than exponentially large numbers of equations with more than
exponentially large numbers of nonzero coefficients.

Example 56 Newton’s theorem that smooth ovals cannot be algebraically in-

tegrated. Used by Newton to show that the position of a planet in a periodic
orbit for a smoothly varying central force cannot be given by an algebraic func-
tion of time (though its orbit can often be described algebraically of course).
Since area swept out is proportional to time, this reduces to the question: given
a smooth oval, show that the area cut off by a secant is not an algebraic func-
tion of the secant. (For non-smooth ovals there are lots of counterexamples:
triangles, y 2 = x2 − x4 . This caused a lot of confusion since Newton did not
explicitly state the condition that the oval most be smooth, which was finally

32
pointed out by Arnold.) Newton’s proof: consider the spiral given by (distance
from center)=area swept out. Suppose this is locally algebraic. Since it is also
smooth, there are no singular points so it is algebraic. But then it has an infinite
number of intersection points with a line through the origin, which contradicts
“Bezout’s theorem” (stated by Newton.) In the case of triangles etc, Newton’s
spiral is a union of a countable number of algebraic curves spliced together.

1.6 Nonsingular curves

Basic invariant (of complex curves) is the genus. Some examples:
Genus 0: Projective line.
Genus 1: Elliptic curves. Take C/ lattice{1, τ } and use ℘02 = 4℘3 + b℘ + c to
map it to a cubic curve y 2 = 4x3 +bx+c. This gives it as branched double cover of
plane with 4 branch points (don’t forget possible branch point at infinity). Any
elliptic curve can be put into the form y 2 = x(x − 1)(x − λ) and 2 are the same
if and only if there is an aut of P 1 taking the set {0, 1, ∞, λ1 } → {0, 1, ∞, λ2 }.
This is possible if λ2 is λ1 , 1−λ1 , 1/λ1 , 1−1/λ1 , λ1 /(λ1 −1), 1/(1−λ1 ), image of
λ1 under a group of order 6 given by permutations of {0, 1, ∞}. So moduli space
is affine line modulo this group of order 6. Need rational function of λ invariant
under this group: use elliptic modular function j = 256(λ2 − λ + 1)/λ2 (λ − 1)2 .
This is the j-invariant of an elliptic curve: 2 complex elliptic curves are isomor-
phic if and only if they have the same j-invariant. So (course) moduli space =
affine line. (This is misleading: elliptic curves y 2 = x3 − x, y 2 = x3 − 1 with aut
groups that are larger than normal really correspond to 1/2 or 1/3 of a point
in the moduli space, and this cannot be described in a completely satisfactory
way with varieties or even schemes. The definitive description uses the moduli
STACK of elliptic curves, which can handle automorphisms in a satisfactory
way. Similarly almost any nontrivial moduli problem involves stacks.) Descrip-
tion of j in terms of τ is quite complicated: j = q −1 + 744 + 196884q + . . .
where q = e2πiτ ; see any good book on modular forms for details. Weird fact:
coefficients of j − 744 are the dimensions of the pieces of the natural graded rep-
resentation of the monster simple group: e.g. 196884 = 1 + 196883 (McKay).
Genus 2: Hyperelliptic curves: a double cover or projective line with 2n
branch points has Euler characteristic is 2 × 2 − 2n so genus is n − 1, so for
6 branch points we get genus 2. Moduli space is given by sets of 6 distinct
points on P 1 modulo action of PSL2 (C). This is essentially the same as the
problem of finding invariants of a sextic binary form, which is quite hard. (An-
swer: moduli space looks like A3 modulo action of a cyclic group of order 5
taking (x, y, z) → (ζx, ζ 2 y, ζ 3 z) where ζ 5 = 1. Ring of invariants contains by
x5 , x3 y, xy 2 , y 5 , x2 z, xz 3 , z 5 , yz.) No such description of moduli space seems to
be known for higher genus.
Genus 3: Some are hyperelliptic, but many are not. Example: degree 4
nonsingular curve in plane. Work out genus of degree d nonsingular curve by
considering it as a d-fold cover of a projective line: in general it will have d(d−1)
branch points of order 2, so Euler characteristic is 2d − d(d − 1) = 2 − 2g, so
g = (d − 1)(d − 2)/2. We get a 6-dimensional family of such curves (space of

33
polynomials is dimension 15, subtract 1 as we want projective space, subtract
another 8 for automorphisms of projective plane.) Hyperelliptic curves of genus
3 form a family of dimension 8 − 3 = 5. Example (Trott): 144(x4 + y 4 ) −
225(x2 + y 2 ) + 350x2 y 2 + 81 = 0. Graph is 4 beans, so has 28 real bitangents. In
general nonsingular degree 4 curves have 28 bitangents touching at 56 special
points. (These are not Weierstrass points: more functions than expected with
poles just at this point).
Genus 4: Cannot be obtained as nonsingular plane curves. Given by inter-
section of cubic and quadric in P 3 . Genus 5 given by intersection of 3 quadrics
in P 4 . Representing curves of higher genus gets hard: can use embeddings into
projective space (e.g. canonical embedding), or embeddings into plane with sin-
gularities, or branched coverings of the plane, or quotients of upper half plane
by discrete groups
General genus: Pick lots of points x, x1 , . . . xn in projective line. Choose n
transpositions on d points with product 1, acting transitively on the d points.
Construct Riemann surface by choosing cuts from x to xn then joining d copies
of the cut projective line these up using the transpositions. Result has Euler
characteristic 2d − n.
Hurwitz curves: what is most symmetrical complex algebraic curve of given
genus? Genus 1 has infinite automorphism group, so look at genus ¿1. Hurwitz
bound: automorphism group has order at most 84(g − 1). Idea of proof: look at
orbifold quotient. This has orbifold Euler characteristic (2g − 2)/|G| < 0. On
the other hand, if the underlying (orientable) surface has genus h and conical
singularities of orders p1 , p2 , . . . then the orbifold Euler characteristic is
2 − 2h − (1 − 1/p1 ) − (1 − 1/p2 ) − . . . .
The maximum negative value of this is −1/42 with h = 0, p1 = 2, p2 = 3, p3 = 7.
To see this look at various cases. First, h must be 0 or the value is at most
−1/2. Second, if there are at most 2 conical points the Euler characteristic is
positive. Third, if there at at least 4 conical points and the Euler characteristic
is negative it must be at most −1/6 coming from conical points of orders 2, 2, 2,
3. So there are exactly 3 conical points. Fourth, if no conical point has order 2
then the possibilities are 3, 3, 3 or 3, 3, 4, and so on, so the Euler characteristic
is at most −1/12. So we can assume one conical point has order 2. If there are
2 conical points of order 3 then the Euler characteristic is positive, so the other
two conical points have orders at least 3. If neither has order 3, the smallest
possibilitities are 2, 4, 4 and 2, 4, 5 with Euler characteristics 0, 1/20. So we
can assume one point has order 2 and another has order 3 and the third has
order at lest 3. If the third has order 3, 4, 5, 6, 7, 8 ... the Euler characteristic
is 1/6, 1/12, 1/30, 0, −1/42, −1/24, ... So we get the Hurwitz bound as the
largest possible negative value of the Euler charcateristic is −1/42. In fact we
get more, since the second largest possible value is −1/24, so if a finite group of
automorphisms does not achieve the Hurwitz bound then it has order at most
48(g − 1).
So (2g − 2)/|G| 6 −1/42 or |G| 6 84(g − 1). Moreover in this case G must
be a quotient of the orbifold fundamental group of the orbifold above, so is

34
generated by elements of orders 2, 3, 7, and product 1. Conversely any such
finite group is the automorphism group of a Hurwitz surface.
Exercise: classify the 17 wallpaper groups by finding all orbifolds of Euler
characteristic 0. These can also have folds and angles as singularities, and the
underlying topological surface can be a sphere or disc or torus but also be a
non-orientable surface such as a Klein bottle, Moebius band, projective plane.

Example 57 There is no Hurwitz surface of genus 2. In this case the automor-

phism group G has order 84. But by Sylow, any group of order 84 has a normal
subgroup of order 7, and the quotient group of order 12 has a normal subgroup
of order 4 or 3. In the first case every element of order 2 or 7 is in the normal
subgroup of order 28, which has no elements of order 3. Similarly in the second
case every element of order 3 or 7 is in the normal subgroup of order 21, which
has no elements of order 2. So there is no Hurwitz group of order 84.
So the maximum possible order of the automorphism group of a genus 2
surface is 48(g − 1) = 48. There is in fact a surface with this order of auto-
morphism group, given by taking the double branched cover of the projective
line branched at 0, 1, −1, i, −i, ∞ (the corners of an octahedron). The subgroup
of P SL2 (C) fixing this set of 6 points has order 24, generated by z → iz and
z → (1 + z)/(1 − z), and is isomorphic to the symmetric group S4 on 4 points.
The group of order 48 is a central extension of this where the center exchanges
the two branches. (This is a non-split extension: for example elements of order
4 in S4 lift to elements of order 8 in the automorphism group. It has a pre-
sentation x2 = y 3 = z 8 = xyz = 1, z 4 in the center.) The hyperelliptic curve
cannot be embedded as a non-singular plane curve, but can be constructed as
an abstract variety by taking two copies of y 2 = x(x2 + 1)(x2 − 1), and gluing
the open subsets x 6= 0 using the involution x → 1/x, y → y/x3 .

Example 58 There were 5 cases where the Euler characteristic is 0 given by

orbifold points of orders (), (2,2,2,2), (3, 3, 3), (2,4,4), (2, 3, 6), . These corre-
spond to the 5 possible cyclic groups acting on elliptic curves, of orders 1, 2, 3,
4, 6.

Example 59 The cases where the Euler characteristic is positive correspond

to finite groups acting on a sphere together with some “bad” orbifolds not
corresponding to any manifold.

Example 60 Suggested by Francesco Polizzi on mathoverflow. A Hurwitz

curve of genus 3 is the Klein quartic: x3 y + y 3 z + z 3 x = 0. This is nonsin-
gular (check) so has genus (d − 1)(d − 2)/2 = 3. It is the only genus 3 Hurwitz
curve with automorphism group of order Hurwitz’s upper bound 84(g−1) = 168;
automorphism group is simple group PSL2 (F 7 ) = SL3 (F 2 ). Obvious automor-
phisms x → y → z → x and x → ζ 4 x, y → ζ 2 y, z → ζz with ζ 7 = 1 generate a
subgroup of order 21.
Proof that it has
 automorphism
 group
 of 
order 168: The group PSL2 (Z)
0 1 1 1
has elements S = , T = with S 2 = (ST)3 = 1 and in
−1 0 0 1

35
PSL2 (F 7 ) the image of T has order 7, so we get a Hurwitz group or order 168.
It has a subgroup of order 21 acting on 3-dim space as above. The Klein quartic
is the only quartic invariant under this subgroup. It is not that easy to write
down an explicit automorphism not in this subgroup.
Following are essentially equivalent:

• Nonsingular projective curves, up to isomorphism of curves

• Curves, up to birational isomorphism
• Finitely generated algebraic function fields over k of transcendence degree
1, up to isomorphism

• (Over C) Compact connected Riemann surfaces.

Nonsingular curves → curves →function fields easy, as is nonsingular curves →
Riemann surfaces.
Compact Riemann surfaces →algebraic function fields: take field of mero-
morphic functions. Hard problem: show that enough meromorphic functions
exist. Not easy to show there are any nonconstant ones. Example: (C −0)/(z =
2z): find a meromorphic function f on C − 0 with f (z) = f (2z). Same prob-
lem in higher dimensions has no solution: Hopf surface = (C 2 − 0)/((z1 , z2 ) =
(2z1 , 2z2 )) is a compact 2-dimensional complex manifold that is NOT projective!
Function fields →Curve: By field theory, the function field has a separating
transcendence base, which has just one element x. It is a finite separable exten-
sion of k(x) so is a simple extension of the form k[x, y] which as y is algebraic
over x are related by some nonzero polynomial f (x, y) = 0. This gives the
equation of some curve.
Curve → nonsingular projective curve. Making the curve projective is trivial,
so the problem is to show that we can resolve the singularities of a curve. More
than 20 ways to do this. Original method due to Newton using Newton polygons,
Puiseux expansions as follows

1.6.1 Newton’s method

Think of f (x, y) as a polynomial in y. Allow rational powers of x. Then roots can
be written as Puiseux series = Taylor series in x1/N for some integer N . (Puiseux
series and Taylor series both invented by Newton.) Example: y 2 = x3 +x4 . Look
at Newton polygon of f , and find leading edge containing some y N using method
of rotating ruler. Terms on this leading edge have lowest weighted degree, so
form homogeneous polynomial in x and y for suitable weights of x, y. Idea is
that we substitute to reduce N or keep N same and reduce slope of leading
edge. 2 cases:

• If not all roots equal, then substitute y → y − axr/s to reduce least power
of y in f . This case only occurs a finite number of times as N is finite and
a positive integer.

36
 • All roots equal. Substitute y → y − axr/s to reduce slope of leading edge.
Can occur an infinite number of times, but this converges to a Puiseux
series for y.
A slight extension of this shows that the field of Puiseux-Laurent series over
C is algebraically closed. Proof similar to proof that if f (x, y) = 0 then y
is a Puiseux series, but allow f to be Puiseux series in x rather than just a
polynomial. A rare example of an explicit algebraically closed field. Field of
Laurent series over C has a unique extension of each degree, like finite fields.
Such fields are sometimes called quasi-finite fields. (Check definition, relation
to theory of finite fields.)
What does this have to do with resolution of singularities? Idea: curve
looks analytically locally like a product of curves y = axr/s + . . . or y s =
bxr + higher powers of x. (Note that different analytic branches can be the same
algebraic branch: example y 2 = x3 + x2 .) Blowing up reduces r or s until one
of them is 1, in which case curve is nonsingular. (When blowing up a point on
an arbitrary curve it can be hard to see that the point has been simplified.)
This is not quite a complete proof of resolution as we have only shown how
to resolve each analytic branch. So we will introduce 2 invariants to measure
nastiness of a singularity and show that blowing up improves invariants until
curve is nonsingular. Invariants are:
1. The multiplicity of the singularity
2. The min value of the slope of the leading edge of the Newton polygon, taken
over ALL choices of local analytic coordinates with y N a min degree term
and NO terms y N −1 x∗ (change coordinates by (x, y + power series in x); this
needs char=0). This slope is 6 1.
Look at effect of blowing up on these invariants. This depends on the roots of
the polynomial of terms of lowest weighted degree, on the leading edge of the
Newton polygon.
1. If this polynomial has more than 1 root, then blowing up reduces the multi-
plicity of the singularity, essentially because it will separate the roots. This
can only happen a finite number of times.
2. If this polynomial has all roots the same, then slope of leading edge is less
than 1 as coefficient of y N −1 x is 0. Then blowing up will increase the slope
of the leading edge. If this new slope is greater than 1 then the multiplicity
is reduced.
Note that we need to eliminate terms y N −1 x otherwise we seem to end up going
in circles.

1.6.2 Other methods

Riemann’s method: construct Riemann surface of a complex algebraic curve or
function field. Hartshorne gives an algebraic version of this.

37
 One word method: normalization. Normalization of an integral domain=integral
closure in quotient field: normal ring. A Noetherian ring is normal if and only
if it is nonsingular in codimension 1 (Ap regular whenever p has height at most
1) and if p has height at least 2 then Ap has depth at least 2 (Serre), so this
resolves singularities of curves. No analogue known in higher codimension, and
normalization not really useful for resolution in dim at least 3. (In dim 2 a
common method is to alternate normalization with blowing up singular points,
but this does not generalize.) Used in number theory: ring of integers is integral
closure of Z in number field.

1.7 Intersections in projective space

1.7.1 Hilbert polynomials
Problem: Suppose we have a finitely generated graded module M = ⊕Mn over
a finitely generated graded algebra
P R over a field. How does dim(Mn ) grow?
Encode as power series fM (x) = xn dim(Mn ). Then this function is RATIO-
NAL. Proof by induction on number of generators of R. Suppose r is in a finite
set of generators and has degree n. Look at

0 → ker(r) → M → M (n) → M (n)/rM (n) → 0

where M (n) is M with grading shifted by n. This gives

fker(r) − fM + fM (n) − fM (n)/rM (n) = 0

But by induction the first and last terms are rational functions, and fM (n) =
xn fM , so fM is rational. This also shows the denominator can be taken as
(1 − xni ) where the ni are the degrees of the generators.
Q
Important special case: if all generators have degree 1 then denominator is
(1 − x)n whose coefficients are polynomials in m for large m, so for SUFFI-
CIENTLY LARGE m, dim(Mm ) is a POLYNOMIAL in m called the Hilbert
polynomial.
Polynomials that are integer-valued on the integers need not have integral
coefficients:
  x(x − 1)/2. Spanned by 1, x, x(x −1)/2,x(x − 1)(x − 2)/3!, ... ,
x x
= x(x − 1) . . . (x − n + 1)/n!, ... (Proof: is 0 for x = 0, . . . n − 1
n n
and 1 for x = n., etc.) So leading coefficient is dxn /n! for some integer d. Most
important invariants are n and d; lower coefficients tend to vary with choice of
grading of M so not so useful.
Useful variations: replace dimension of vector space over a field by any other
invariant that is additive on exact sequences, such as length of a module.

1.7.2 Dimension of local rings

If R is a Noetherian local ring then ⊕mn /mn+1 is a graded ring, finitely gener-
ated over the field R/m by degree 1 elements, so has Hilbert function.
For a Noetherian local ring the following are equal:

38
1. geometric dimension = max length of chain of prime ideals -1
2. 1+deg of Hilbert polynomial of ⊕mn /mn+1
3. Smallest number of elements of m not contained in any other prime ideal.

4. Smallest system of parameters (set of elements of m generating a cofinite

ideal).
Proofs are hard commutative algebra.
Problem: most of these almost impossible to calculate directly. But Hilbert
polynomial is easy to calculate.

1.7.3 Degree of a projective variety

If I ⊆ R = k[x0 , . . . , xn ] is the graded ideal of a projective variety then R/I is
a graded module over R so has Hilbert polynomial. Then degree r of Hilbert
polynomial = dimension of variety, and degree of variety is defined to be r! ×
leading coef.

n
Example 61 Projective  space
 P . dimensions are 1, n+1, (n+1)(n+2)/2!, . . . , (n+
n+k
1) . . . (n + k)/k! = = k n /n! + . . . so dimension is n and degree is 1.
n
 
n n+k
Example 62 Hypersurface of degree d in P . Hilbert polynomial is −
n
 
n+k−d
= dk n−1 /(n − 1)! so dimension is n − 1 and degree is d.
n

Example 63 Twisted cubic: wz = xy, x2 = wy, y 2 = xz. Basis of coordinate

ring: 1; w, x, y, z;
In general cant have x2 , y 2 , xy, so get wi z k−i , wi xz k−i−1 , wi yz k−i−1 which
is k + 1 + k + k = 3k + 1. So dim = 1 and degree=3.

For a projective variety the Euler characteristic χ is defined to be constant

term of Hilbert polynomial (coherent cohomology Euler characteristic of sheaf
of regular functions, so does not depend on embedding of variety into projective
space). For historical reasons (−1)dim (χ − 1) is called the arithmetic genus.

Example
 64
 Arithmetic
 genus
 of a plane curve of degree d. Hilbert polynomial
2+k 2+k−d
is − = 1 − (2 − d)(1 − d)/2 so arithmetic genus is
2 2
(d − 1)(d − 2)/2. For nonsingular complex curves this is topological genus.
Remark. Hilbert polynomial is Euler characteristic of O(n), and higher
cohomology groups all vanish for n large.

39
Remark 2 The Hilbert polynomial of a closed subvariety or subscheme is es-
sentially the only discrete invariant. Hartshorne showed that 2 points in Hilbert
scheme are in same connected component if and only if they have the same
Hilbert polynomial. However connected components of Hilbert scheme often
have many irreducible components (with horrendous singularities).

1.7.4 Hironaka’s example

Hironaka gave example of a 3-dim nonsingular abstract variety that is not
projective. Take 2 curves in a nonsingular 3-fold intersecting transversely at 2
points P , Q. On X − P blowup first one curve then the strict transform of the
other. On X − Q do in other order. Then glue together. TODO

1.7.5 Bezout’s theorem

Old style: Intersection of 2 GENERIC curves of degrees m, n in P 2 has mn
points. More generally, intersection of hypersurfaces of degrees n1 , . . . has
n1 × . . . points. Old style “proof”: vary coefficients of each hypersurface so
it degenerates into a union of hyperplanes, and hope that the number of inter-
section points does not change under this process.
“Generic” means: theorem often true, but there sure seem to be a lot of
exceptions which are left as an exercise for the reader to figure out.

Example 65 Things that can go wrong:

1. Intersections may occur at infinity, or at complex points. Example: two
circles seem to have at most 2 intersection points. Other are at (1 : ±i : 0).
Solution: use projective space.

2. Curves may have a component in common. Solution: exclude this. Not so

easy to deal with in higher dimensions: for example, 3 planes in P 3 might
meet in a line.
3. Curve may have a singularity with multiplicity ¿1. Can either exclude these
or count points with multiplicities.

4. Curves may intersect non-transversely. Can either exclude these or count

points with multiplicities.
Problem: how does one define multiplicity of intersection of 2 curves? Ex-
ample: y 3 − x4 and y 5 − x6 : not obvious what the intersection multiplicity is.
Can try to deform curves slightly to get intersection multiplicities 1. “What is
true up to the limit is true at the limit”.

Example 66 How many lines intersect 4 lines in P 3 ? Expect a finite number,

as this is intersection of 4 hypersurfaces in 4-dim Grassmannian G(2, 2). Guess
solution as follows: degenerate lines so they meet in 2 pairs. Then there are
2 lines meeting all 4: line joining 2 intersection points, and intersection of 2

40
planes. Problem: make this rigorous! (Schubert calculus = cohomology ring of
Grassmannian = Littlewood-Richardson coefficients.)

Theorem 6 2 distinct irreducible curves in P 2 of degrees m, n intersect in mn

points, counted with multiplicities.

Problem: define multiplicity. Solution: prove theorem, define multiplicity

to be what is needed to make proof work! It will turn out to be dim (local
ring/(f, g)). (Reality check: this is 0 if f or g does not vanish, and 1 if curves
of f , g intersect transversely without multiple points. Also depends only on
completion mod f , g, so really is a local invariant.)
To prove theorem, calculate degree of intersection in 2 different ways. Inter-
section is given by ideal, and is a closed subscheme rather than a subvariety.

2 Schemes
Schemes generalize algebraic sets: need not be over fields, and need not be
reduced. Affine algebraic sets correspond to finitely generated algebras over a
field with no nilpotents: affine schemes correspond to any commutative rings.

2.1 Sheaves
Invented by Leray around 1950. Grauert’s comment on the work of Cartan and
Serre: “We have bows and arrows, the French have tanks”. Introduced into
algebraic geom by Serre in [Link] for sheaves and cohomology: cleans
up algebraic geom and makes it rigorous. Typical example: old definition of
arithmetic genus for a surface in P 3 is
 
µ0 − 1
pa = − (µ0 − 4)ε0 + ε1 /2 − ε0 + 2t
2

where µ0 for example is the degree of the section of the surface by a hyperplane,
and the other invariants are defined similarly. Arithmetic genus and irregu-
larity are invariant of the surface, independent of embedding. This was very
mysterious, and it was unclear how to generalize this to higher dimensions. In
terms of sheaf theory, arithmetic genus is essentially the Euler characteristic,
and irregularity is dim of H 1 . Trivial to generalize in higher dimensions: Hodge
numbers hpq = dim H q (Ωp ).
Let X be a topological space. For each open U ⊆ X put F(U ) =continuous
functions on U . Then whenever V ⊆ U we have a restriction morphism ρUV :
F(U ) → F(V ). This satisfies
1. ρU U is the identity map
2. ρUW = ρVW ◦ ρUV

41
In other words, we have a contravariant functor from the category of open sets
to the category of groups, or a presheaf of abelian groups.
In addition suppose that Ui is a cover of U . then
1. If the restriction of s, t ∈ F(U ) to all Ui are same, then s = t. (This implies
F(∅)has at most 1 element).

2. If we are given si ∈ Ui whose restrictions to Ui ∩ Uj are the same, then there

is some s with restrictions si . (Unique by condition 1.)
Y Y
F(U ) → F(Ui ) ⇒ F(Ui ∩ Uj ) exact
A presheaf satisfying these conditions is called a sheaf. Morphisms of presheaves
or sheaves are just natural transformations of functors. Sheaves form full sub-
category of presheaves.
More generally we can define a presheaf with values in any cat over any cat as
a contravariant functor. Most important cases are sheaves of sets and sheaves of
abelian groups. If the base cat has a Grothendieck topology (describes covering
families) then we can define sheaves. Philosophy of sheaves: the sheaves over a
top space form a weak model of set theory, in the sense that any constructive
operations on sets can be done on sheaves. More precisely they form a topos.
Similarly sheaves of abelian groups behave very much like abelian groups (they
are the “abelian groups” of the category of sheaves). For example, sheaves of
abelian groups form an abelian category with tensor products.

Example 67 Sheaves of continuous/smooth/analytic/holomorphic functions,

or regular functions on a variety.

Example 68 Skyscraper sheaf: F(U ) = A if p ∈ U , 0 otherwise.

Example 69 If f : Y → X put F(U ) = continuous sections of f over U to get

a sheaf. For example this gives the sheaf of a vector bundle. In fact any sheaf
can be obtained like this from the etale space of F. Construction of etale space
of a presheaf is as follows: define stalk of F at p to be lim→p∈U F(U ). Etale
space is the union of the stalks with following topology. Base of open sets given
by taking any open U and any s ∈ F(U ) and taking image of s at all stalks
of U . This makes projection into an etale map: every point of etale space has
an open neighborhood mapped homeomorphically to its image. Put F + (U ) =
sections of etale space over U . Then if F is a sheaf, F = F + . More generally,
any map from F to a sheaf factors through the sheaf F + in a unique way: this
is the universal sheaf generated by the presheaf.

Example 70 If A is a discrete group, define constant presheaf and constant

sheaf. Note that these are DIFFERENT in general.
Suppose f : X → Y is a continuous map of top spaces. There are 2
fundamental operations: f∗ taking sheaves on X to sheaves on Y defined by

42
f∗ (F)(V ) = F(f −1 (V )), and f −1 in other direction given by pullback of etale
space. The functor f −1 is left adjoint to f∗ : maps f −1 G → F are “same as”
maps G → f∗ (F).(Left adjoints are in practice often the “free” object while
right adjoints often “forgetful”; for example free group on a set is left adjoint
to forgetful functor from groups to sets.)
Now suppose X and Y are ringed spaces (spaces with a sheaf of rings) and
suppose that f is a morphism of ringed spaces. Then if G is a sheaf of OY
modules, f −1 G is a sheaf of f −1 OY modules but need not be a sheaf of OX
modules, so define f ∗ G = f −1 G ⊗f −1 OY OX . Then f ∗ is left adjoint to f∗
considered as functors between cats of modules.
A fundamental problem is to determine when a map of sheaves is injective
or surjective or an isomorphim. For injective maps this is straightforward a
map of sheaves is injective if and only if the maps of stalks are injective, and
this is in turn equivalent to the maps of sections being injective. For surjective
maps things are more subtle. It is still true that a map is surjective if and only
if the maps of stalks are surjective. However if a map of sheaves is surjective,
this does NOT imple that the corresponding maps of sections are surjective.
The fundamental example is the fact from complex analysis that the logarithm
function cannot be defined everywhere on the nonzero complex numbers: in
terms of sheaves this means that although the map of sheaves

0 → 2πiZ → O → O∗ → 0

is exact, the corresponding map of sections is NOT surjective on the right: for
example there is no global function “log” mapping to x, even though such a
function can be defined locally everywhere. This lack of surjectivity is funda-
mental in algebraic geometry, and is the cause of cohomology theory, which is a
way of controlling the lack of exactness. More precisely for any exact sequence

0→A→B→C→0

of sheaves we get the corresponding long exact sequence of cohomology groups.

Sheaves are a sort of generalization of vector bundles, which are in turn a
twisted generalization of trivial vector bundles. One difference between a sheaf
and a vector bundle is that sheaves may have support on a closed subset: for
example, if V is the vector bundle whose sections are smooth functions on R,
and x is multiplication by x, then the map x from V to V is a surjection as a
map of vector bundles but not as a map of sheaves: quotient is a skyscraper
sheaf at 0 that cannot be detected by vector bundles. In more technical terms,
vector bundles form a category that is additive but not abelian, while abelian
sheaves form an abelian category.
Example: the map from the 1-dimensional vector bundle to itself over R
given by multiplication by x has cokernel a skyscraper sheaf at 0,not corre-
sponding to a vector bundle.

43
2.2 Schemes
Several experiments with generalizing algebraic sets in the 1950s by Cheval-
ley, Serre, Grothendieck produced schemes as the most powerful. (Named first
used by Chevalley.) There are also many suggestions for generalizing schemes:
algebraic spaces, locally ringed toposes, stacks (Janos Kollar “Their study
is strongly recommended to people who would have been flagellants in earlier
times”), noncommutative algebraic geometry.
Affine schemes are locally ringed spaces ↔commutative rings; scheme is
called Spectrum of the ring Spec(R). Scheme is a locally ringed space looking
locally like an affine scheme, just as a quasiprojective variety looks locally like
affine variety.
Reason for name “spectrum”: Spectrum of an operator A is set of eigenvalues
= homomorphisms from C[A] → C = maximal ideals of the ring C[A]. Same
if A is a set of commuting operators on a complex vector space. This is the
finite dimensional case. Now look at a compact Hausdorff space X and C(X).
How to reconstruct X from C(X)? A. X is the set of max ideals of C(X) (if an
ideal does not vanish at any point, we can find a function in it that is positive
everywhere by taking a sum of squares). Topology given by closed sets=max
ideals containing some ideal. Maximal spectrum of commutative C ∗ algebra
= set of max ideals with topology. Similarly max ideals of coordinate ring of
algebraic set over C = points of algebraic set. This suggests we look at maximal
spectrum = max ideals of any commutative ring. Problem: suppose f : R → S.
We would like this to induce a map from Specm S → Specm R. However inverse
image of a max ideal need not be a max ideal, as subring of a field need not be
a field. Most we can say about it is that it is an integral domain. Inverse of a
prime ideal is a prime ideal, as subring of int domain is int domain, so PRIME
SPECTRUM of a ring is a contravariant functor.
Topology on Spec(R): Closed sets = prime ideals containing given set (or
ideal). Base of open sets given by D(f ) = prime ideals not containing f . Key
point: work with open sets D(f ) rather than with arbitrary open sets.
Weird properties: not only is Spec(R) not Hausdorff, but it may have points
that are not closed! In fact, closed points are exactly the maximal ideals.
(Remark: maximal ideals of C(X) for X compact Hausdorff are just points
but the prime ideals can be rather weird: for example, pick a sequence, and
take functions vanishing on an element of some ultrafilter. Don’t try to mix alg
geom and analysis.)

Example 71 Spec (field)=point. Spec(Z)=primes with 0. (0) is not a closed

point! Spec(C[x]) similar: affine line + extra generic point. Spec(C[x, y]) has
points of plane + generic points of irreducible curves+generic point for whole
space. Spec of discrete valuation ring has 2 points, one closed. Spec k[x, y]
localized at (x, y) or at (f (x, y)). Spec k[x, y]/(f (x, y)).

Example 72 Spec(Z[x]) Fibered over Spec(Z) = (0), (2), (3), (5), . . . Fiber
over (p) is spectrum of Fp [x] which consists of 0 (closure = fiber) and irre-

44
ducible polynomials over Fp = orbits of alg closure under Galois group (Frobe-
nius) (closed points). Fiber over 0 is Spec(Q[x]) = 0 (generic: closure=whole
space) plus irreducible polynomials over Q=orbits of alg numbers under absolute
Galois group of Q = content free irreducible polynomials over Z with positive
leading coefficient (closure = union of reduction mod p for all p). So describ-
ing Spec(Z[x]) requires all of alg number theory. DIAGRAM: draw closures of
(x2 + 1), (x − 5) which intersect in 2 points. Spec(Z[x]) is 2-dimensional, and
behaves like an algebraic surface.

Example 73 Nagata counterexample to almost everything: an infinite dimen-

sional Noetherian affine scheme. In the ring k[x1 , . . .] consider the primes ideals
(x1 ), (x2 , x3 ), (x4 , x5 , x6 ), . . . and let R be the ring given by inverting all ele-
ments not in any of these ideals. Then ideals generated by these elements are
exactly the max ideals of R. Any prime ideal is generated by irreducibles so
by elements with variables in just one of these subsets. So all prime ideals are
finitely generated and ring is Noetherian (Cohen). We have a map from R to
Rm embedding Spec(Rm ) as subset of Spec R, and these intersect only in 0 and
have union Spec R. Describe Spec Rm . On the other hand ring is infinite dimen-
sional as it has arbitrarily long chains of prime ideals (xn ) ⊂ (xn , xn+1 ) ⊂ . . ..
Noetherian local rings all finite dimensional: dim at most number of generators
of max ideal.
Variation: a 1-dim integral domain with all closed points singular. Take
subring generated by x2i , x3i for all i, and invert anything not in one of the prime
ideals (x2i , x3i ). This forces these ideals to be the maximal ideals. All local rings
at these points are singular, given by local ring of a cusp over a field of infinite
transcendence degree.
Noetherian rings can be quite strange: what is “correct” class of rings of
geometrical rings that aviods examples like this? No generally agreed answer;
Grothendieck suggested excellent rings.
Basic properties of topological space Spec(R)
1. Every closed irreducible set is closure of a point (and conversely!) If S is
a closed set of prime ideals, then it is set of prime ideals containing their
intersection I. If I is not prime, pick a, b 6∈ I, ab ∈ I. Then S is union of
primes containing (I, a) and (I, b) so is not irreducible. So if S is irreducible
then S is closure of prime ideal I.
2. Space is ALWAYS compact (if a set of elements generates the unit ideal, then
some finite subset does).

Recall localization RS of R at a multiplicative set S = ring generated by adjoin-

ing inverses of S. Problem: what is kernel of natural map R → RS ? Answer:
elements of R killed by some element of S. Proof: construct RS explicitly as
equivalence classes of pairs (r, s) = r/s, where r1 /s1 = r2 /s2 if s(r1 s2 −r2 s1 ) = 0
for some s ∈ S. If S has no zero divisors can omit s and this is usual construction
of rationals.

45
 Now define locally ringed space structure. Recall D(f ) form base of open
sets. Put O(D(f )) = Rf .
Key technical lemma for simplifying arguments about sheaves:

Lemma 1 Can define sheaves by defining them on a base for topology and
checking sheaf axioms just for coverings by this base.
So we just need to check the sheaf property for covers of D(f ) by D(fi ). By
replacing R by Rf can assume f = 1 so have to check for a cover of Spec(R) by
sets D(fi ). This means fi generate unit ideal, so some linear combination of a
finite number is 1 = a1 f1 + . . . + an fn .
First do easy check that r is determined by its restrictions: if r = 0 when
restricted to D(fi ) then fini r = 0 for some ni . Can replace fi by any power, so
can assume ni = 1 so rfi = 0. But then r = r1 = ra1 f1 + . . . = 0.
Hard part is to show that compatible elements ri /fini on D(fi ) lift to
m n
some r ∈ R. Compatibility condition says fimi fj j (ri fj j − rj fini ) = 0 for some
m’s. As before we can replace fi by some high power to assume that all m’s
and n’s are 1. So we have elements satisfying

1 = a1 f1 + . . . an fn
0 = fi fj (ri fj − rj fi )

and we want to find r with r = ri /fi in D(fi ), which will follow if r satisifies

0 = fi (fi r − ri )

Now replace ri fi by si so that 0 = si fj2 − sj fi2 and we have to solve fi2 r =

si . Then put gi = fi2 so we have 1 = b1 g1 + . . . for some bi , si gj = sj gi and we
have to solve gi r = si . But now we can just put r(= b1 g1 r + . . .) = b1 s1 + . . .
and check that gi r = gi b1 s1 + . . . = g1 b1 si + . . . = si .
Stalk of sheaf at a prime p is direct limit of Rf for f 6∈ p which is local ring
Rp . (Hartshorne does things backwards: uses local rings to define sheaf, then
checks values on D(f ).)
Problem: what does O(U ) look like for other open sets U . Correct answer:
who cares?
Morphism of schemes = morphism of LOCALLY ringed space. (Recall that
this means we need the extra condition that inverse of a maximal ideal of a
local ring is in the maximal ideal: geometrically this coresponds to saying that
if a functioin vanishes at a point, then its pullback vanishes on the inverse
images of this point.) Example of a morphism of ringed spaces that is not a
morphism of locally ringed spaces: R=DVR with quotient field K, then map
Spec K → Spec R with image the CLOSED point of Spec R, with map R → K
given by the inclusion. This is not local, as it does not map the max ideal of R
into the max ideal of K.
Morphisms between affine schemes correspond to homomorphisms of rings:
affine schemes is opposite cat of commutative rings. (As we have just seen, this
is FALSE for morphisms of ringed spaces: we need to use morphisms of locally

46
ringed spaces.) More generally, if X is any locally ringed space, then morphisms
of locally ringed spaces from Spec(R) → X are same as ring homomorphisms
from O(X) → R. In other words the functor Spec is adjoint to the functor O:
Spec R is a sort of universal locally ringed space generated by the ring R.
Proj of a graded ring S. Underlying set = homogeneous prime ideals not
containing all positive degree elements. Topologize by closed sets = primes
containing given homogeneous ideal. For any positive degree element f de-
fine D(f )= elements of P roj(S) not containing f , (these form a base for the
topology) and define O(D(f ))= degree 0 elements of S[f −1 ]. The D(f ) cover
Spec(S). Homogeneous primes of S not containing f = primes of Sf0 .
Example: what is a point of projective line over a ring? Wrong answers:

1. Union of 2 affine lines. (Correct for fields, fails in general as maps from X
to Y ∪ Z need not have image in Y or Z if X not a point.)
2. Pair of points not both zero, up to units. (Correct for fields, fails in general:
zero divisors, etc.)

3. Pair of elements generating unit ideal, up to units. Still fails for rings with
nontrivial invertible modules, though OK for local rings. Point is that R ⊕ R
can have rank 1 submodules (“lines”) that are not free modules.
4. Correct definition: pair of elements of an invertible module, generating all
stalks, up to isomorphism.

What are points of Proj(S) with values in a ring R, in other words the morphisms
from Spec(R) to Proj(S)? First look at case when R is a field k. Then Spec(R)
has just one point, so image is certainly in some affine subset D(f ). So we get
a morphism O(D(f )) → R. But O(D(f )) = degree 0 elements of localization
Sf = S[1/f ] so morphisms O(D(f )) → R = morphisms from S0 ⊕ Sn ⊕ . . . → R
taking f to a unit, up to multiplying elements of Sn by a unit (n = deg(f )). So
points are represented by some homomorphism S0 ⊕ Sn ⊕ . . . → R that is not
zero on positive degree elements, up to multiplying by units and changing n to
some multiple. If S is generated by degree 1 elements, can just take n = 1.
So points of projective space with values in a field is as expected.
Points with values in a local ring similar: key point is that there is a closed
point of Spec(R) in closure of every other point, so any open set of Proj(S)
containing this point contains whole image of Spec(R). In other words Spec(R)
has image in one of the open sets D(f ) as in the case of fields. Argument for fields
gives similar answer: points of projective space represented by (a0 : . . . : an )
such that at least one ai is a unit, up to multiplication by units.
Example: R=formal power series in x and y. Then (x : y) is a point in
projective space of quotient field that is NOT represented by a point with values
in R. Same for R = Z[x, y]
Example: integer valued points of projective space are represented by (a0 :
. . . : an ) coprime. Image is NOT usually in one of the open sets D(f ) unless
some ai is a unit. To see this represents a point, cover Spec(Z) by open subsets

47
and note that this point is well defined on each open subset and represents the
same morphism on intersections.
Example: One might guess from previous examples that points of projective
space are represented by (a0 : . . . : an ) whose elements generate the unit ideal.
This is FALSE in general, and related to nonuniquness of factorization in alg
number fields: √ both related to existence of nontrivial √ invertible√ modules. Exam-
ple: R = Z[ −5] √ Not a UFD as 6 = 2 × 3 = (1 +√ −5)(1 − −5),
√ non-principal
ideal = (2, 1 + −5). Look at point (2 : 1 + −5) = (1 − −5 : 3). If we
cover Spec(R) by open sets Spec(R[1/2]), Spec(R[1/3]) then these 2 expressions
represent points on the 2 open sets, equal on their intersection. So they give a
point of projective space.
Key property of Proj(S): there is a special line bundle O(1) over it. Line
bundle = sheaf of modules over O that is locally isomorphic to O (locally free
of rank 1). Informal picture: associate a 1-dim vector space over residue field
at each point. Informal examples: Moebius band, highest exterior power of
(co)tangent bundle over smooth manifold. Projective space = set of lines in
a vector space, so we can associate this line to each point of projective space.
(Similarly Grassmannians have a special vector bundle.) More formal construc-
tion: Cover projective space by open affine sets D(xi ). On each of these the line
bundle will be trivial, so we can identify it with the ring of regular functions.
How do we change in going from D(xi ) to D(xj )? A. Multiply by fij = xi /xj (or
(xi /xj )n ) called transition functions. So a global section of line bundle is given
by elements si of O(D(xi )) such that sj = si fij on Ui ∩ Uj . (Easy to get this
wrong way round!) For example, for projective space, put si = xj /xi , a degree
0 element of k[x0 , . . . , xn , xi−1 ], so xj gives a section of this line bundle, and sim-
ilarly degree k polynomials are sections of m’th tensor power O(m). Transition
functions have to be invertible in O(Ui ∩ Uj ) and compatible on O(Ui ∩ Uj ∩ Uk ):
fij fjk fki = 1. (A Cech 1=cocyle) Warning: in diff geom all vector bundles are
isomorphic to their duals as we can find a metric on them, but in alg geom this
is completely false and O(1) is totally different from its dual O(−1). Warning:
For projective space all sections of O(m) are given by elements of Sm but this
is not true for general graded rings.

Remark 3 In alg topology, classifying space for complex line bundles over CW
complexes is infinite dimensional complex projective space. Chern class = image
of generator of 2nd cohomology of CP∞ ; classifies complex line bundles. In
algebraic geom line bundles harder to classify: get continuous families.

2.3 First properties of schemes

Contains about a gazillion definitions, all important.

Definition 4 Define integral, irreducible, connected, reduced.

Definition 5 Locally Noetherian = covered by Spec of Noetherian ring. Noethe-

rian = locally Noetherian + compact.

48
 Common problem: Does “locally X” mean “ has a base of affine X” or
“every open affine subset is X”. Usually equivalent in practice. Want to show:
(1) Localization of X ring is X and (2) If Spec R is covered by (finite number
of) X sets D(fi ) then R is X. Check these 2 conditions for X=Noetherian.
(1) Localization of Noetherian ring obviously Noetherian. (2) Suppose Rfi
Noetherian. If I is an ideal of R then we can find a finite set of generators
of Ifi ; we can assume these are in R, so we have a finite set {g1 , . . .} that
generates all Rfi . We show this generates I. We have rfini ∈ (g1 , . . .) so
r = r1n1 +n2 +... ∈ (g1 , . . .).

Definition 6 f : X → Y is Locally of finite type if X is covered by affines such

that each inverse image is covered by finitely generated affines. Finite type if
each inverse image is covered by a finite number of finitely generated affines.
Example: Varieties are finite type as they are covered by a finite number of
affine varieties, which are finitely generated algebras. In fact abstract varieties
are exactly the schemes that are reduced and integral and of finite type over a
field. Hilbert scheme is locally of finite type over a field but not of finite type:
infinitely many connected components! Finite type can be thought of vaguely
as finite dimensional fibers.

Definition 7 f : X → Y is quasifinite if inverse image of any point is finite,

and finite if it is covered by affines such that inverse image is Spec of a finite
algebra (=finite as a MODULE not an ALGEBRA).
Finite is much stronger than quasifinite: example: R → RS , or any open
immersion, is always quasifinite, as inverse image of a point has at most 1
element, but is rarely finite. Finite implies finite type (but quasifinite does
not). Examples of finite morphisms: Spec A → Spec Z for A integers of algebraic
number field. Projection from projective curve to P 1 .
Open immersions and finite morphisms are in some sense the only ways to
get quasifinite ones. Grothendieck’ generalization of Zariski’s main theorem: if
Y is a quasi-compact separated scheme and f : X → Y is a separated, quasi-
finite, finitely presented morphism then there is a factorization into X → Z → Y
where the first map is an open immersion and the second one is finite.
Stein factorization: proper morphisms of Noetherian schemes can be written
as composition of a morphism with connected fibers and a finite morphism. For
these morphisms it follows that proper + quasi finite implies finite. (Stein fac-
torization and Zariski’s main theorem both follow from Grothendieck’s theorem
on formal functions, covered in chapter on cohomology.)

Definition 8 Open/closed subschemes. Given by open/closed subsets of scheme

such that induced map of sheaves is identity/quotient.
Note that open subschemes are determined by the subset, but closed sub-
schemes are NOT.

49
 Example: A localization Spec(R[f − 1] ⊆ Spec(R) is an open immersion. The
reduced induced closed subscheme is a closed subscheme. k[x]/(xn ) is a closed
subscheme of k[x]. Closed subschemes of affine schemes correspond to ideals.
Define pullbacks for categories.

Theorem 7 Fibered products X ⊗S Y (= pullbacks= base extension) exists for

schemes.
Proof Requires a lot of tiresome bookkeeping to do properly.
Case 1: everything affine. In category of affine schemes, this is just given by
tensor product. Check that this is still a fibered product in cat of all schemes.
Case 2: S affine. Now cover any 2 schemes over an affine S by affines Xi Yj ,
and construct fibered product by gluing together Xi ⊗S Yj .
Case 3: general case. Now suppose S is covered by affines Si . Construct
X ⊗S Y by gluing together Xi ⊗Si Yi where Xi , Yi are inverse images of Si . (Can
also use Xi ⊗Si Yi = Xi ⊗S Y ) 

Warning 3 The scheme product of X and Y over S not only has more open
sets than the topological product, but usually has far more points. Example:
Spec k[x] ×Spec k Spec k[y] has all irreducible curves as extra points.

Example 74 (suggested by Ryan Reich, Pete L. Clark on [Link])

Fiber products of reduced schemes can be nonreduced. Take pullback of identity
map and p’th power map of algebraic group Gm = Spec k[x, x−1 ] where k has
characteristic p. Result is k[x]/(xp − 1) (a non-reduced group scheme represent-
ing p’th roots of 1). Or take K ⊗k K where K is an inseparable extension of
the field k such as k[x]/(xp − a). Then tensor product is k[x, y]/(xp − a, y p − a)
so (x − y)p = 0.

2.4 Separated and proper morphisms

Separated is analogue of Hausdorff, proper is analogue of compact (or rather
proper). Problem: find definition of these concepts for top spaces that work well
for schemes. A. A top space X is Hausdorff ↔ diagonal is closed in X × X. If
X is compact then X → 1 is universally closed: X × Y → Y is closed for all Y .
Key point: for schemes we use the scheme product, not the product topology.

Definition 9 A morphism is called SEPARATED if the image of the diagonal

is closed.
Example: Any morphism R → S of rings gives a separated morphism.
Diagonal map corresponds to R ⊗S R → R so diagonal is closed subset of
Spec(R ⊗S R) of prime ideals containing ker(R ⊗S R → R).
Example: Line with 2 origins (bug eyed line).

Definition 10 A morphism is called PROPER if it is separated of finite type

and universally closed.

50
2.4.1 Valuation rings
Recall that a valuation ring is a domain such that for any 2 elements one divides
the other. They are all local rings. Valuation group = nonzero elements of
quotient field mod units: ordered by divisibility. DISCRETE valuation rings:
valuation group = integers. Nondiscrete valuation rings can be strange: any
Noetherian valuation ring is a DVR or a field. Example of valuation rings:
1. p-adic numbers
2. Formal power series in 1 variable

3. 1-dim regular local ring. Example: local ring of irreducible curve in nonsin-
gular surface.
4. Puiseux power series; valuation group = rationals. Nondiscrete, height 1
5. Put v(f ) = multiplicity of 0 for f ∈ k[x, y]. Then elements of quotient field
with v > 0form a DVR.

6. As above, but give x, y different integral weights.

7. As above, but given x, y IRRATIONAL weights: nondiscrete DVR.
8. Take smooth surface. Repeatedly blow up points on exceptional curve. Union
of local rings is a valuation ring (not trivial: need to know this resolved sings
of a function.)
9. Ring of formal powers series with exponents the nonnegative elements of a
totally ordered group.
The following are equivalent:

1. R is a DVR
2. R is a Noetherian valuation ring other than a field

3. R is a 1-dim regular Noetherian local ring

4. R is a valuation ring and a PID

Spec of a DVR R has 1 closed point and one generic point. Something like a
short smooth curve. They tend to be associated to divisors not contained in
singular locus: local ring of a hypersurface is a DVR whose valuation measures
the order of the zero of a function along the divisor.
Geometric meaning of valuation ring. Zariski defined spaces that were pre-
cursors to Spec and Proj. Take extension of fields k ⊆ K where we think of
K as rational functions on some variety over k. In 1-dim, valuation rings in K
containing k = Riemann surface (more or less) = complete nonsingular model
of curve. Try to copy this in higher dims: define Zariski-Riemann space = all
valuation rings of K containing k = places of K over k. (More generally can

51
use local rings instead of valuation rings). Zariski topology: basis of open sets
given by valuation rings not containing given finite set.
Recall spectrum of a DVR has 2 point (0) (open) and m = (p) (closed).
Stalk at (0) = regular functions on (0) = field of quotients, and stalk at (p) =
regular functions on whole space = R.
A morphism from Spec(R) to a scheme S is given by following date:
1. 2 points a, b of S with a in closure of b; the images of (p), (0).
2. Maps from the stalks of a and b to R and K that commute with the restriction
maps (R → K) and map the max ideals into the max ideals.

Theorem 8 Suppose f : X → Y is a morphism of Noetherian schemes of finite

type. Then f is separated if and only if given

Spec K → X
↓ % ↓
Spec R → Y

for a DVR R with quotient field K there is at most one diagonal arrow. f is
proper if and only if there is always exactly one diagonal arrow.
Meaning: think of Spec R as a short smooth curve with a special point. Then
condition says that maps from rest of curve can be extended to point in at most
one or exactly one way.
Recall morphisms from a LOCAL ring to projective space P n (over integers)
are given by tuples (a0 : . . . : an ) such that at least one is a unit, up to multipli-
cation by units. In particular if R is a valuation ring, then points of projective
space with values in R are same as points with values in quotient field K. Key
point: if (a0 : . . . : an ) has elements in K not all 0, we can divide by the element
with smallest valuation to make it a unit, so all other elements are then in R.
Corollary: Projective space over the integers is complete.
Application: when is a toric variety complete, when is a map between them
proper.

2.5 Sheaves of modules

Define sheaf of modules over a sheaf of rings. (Philosophy: sheaves form weak
model of intuitionistic set theory, so constructions for rings should work for
sheaves of rings.) Sheaves of modules form an abelian category.
Module M over ring R gives a sheaf of modules over Spec R with sections
over D(f ) given by Mf . Proof similar to construction of Spec R: need to show
these have sheaf property for open sets of the form D(f ). Stalk of sheaf at prime
ideal p is localization Mp just as for R.
Are sheaves of modules over an affine scheme same as modules over the
ring? A. No! Sheaves of modules have special property: suppose A ⊆ B are
open affine subschemes. Then M (A) = M (B)⊗O(B) O(A), or M (A) = M (B)f if

52
O(A) = O(B)f . Sheaves of modules with this extra property are (unfortunately)
called quasicoherent. For quasicoherent sheaves, if we know values on some
open affine subscheme we know values on all smaller subschemes; in particular
quasicoherent sheaves over affine schemes are essentially the same as modules
over rings. Example of a sheaf of modules that is not quasicoherent: take a
quasicoherent sheaf over affine scheme and just make it 0 on all proper subsets.
Coherent modules: over Noetherian rings these are same as finitely generated
modules. Hartshorne uses coherent to mean finitely generated over all rings
but this is not standard terminology. Over general rings, M coherent means
finitely generated + kernel of any finitely generated free module mapping to M is
finitely generated. Automatic for Noetherian rings as any submodule of a finitely
generated free module is finitely generated. Reason for extra condition: in
general kernels of maps between finitely generated modules need not be finitely
generated, for example any non finitely generated ideal is kernel of map between
finitely generated modules. So finitely generated modules need not form abelian
category, but coherent modules do. A ring is called coherent if it is coherent
as module over itself, so all Noetherian rings are coherent. Ring of polynomials
in infinitely many variables is coherent but not Noetherian. Easiest to forget
about coherence and just use finitely generated modules over Noetherian rings.
Coherent sheaves: quasicoherent sheaves that come from coherent modules.
So for affine Noetherian schemes, quasicoherent sheaves are same as finitely gen-
erated modules. Example: sheaf of regular functions, sheaf O(1) over projective
space.
Graded module M over graded ring R gives a quasicoherent sheaf whose
value on D(f ) (f homogeneous of positive degree) is degree 0 elements of Mf .
Warning: different graded modules M often give same sheaf: in fact if M is
changed on a finite number of pieces this does not change the sheaf.
Fundamental difference between sheaves and modules: sheaf of regular func-
tions is not projective in general; in fact there are sometimes not enough pro-
jectives. Recall that global sections of O(m) over projective space = degree
m polynomials. These are also same as maps of sheaves O(0) → O(m) or
O(k) → O(m + k). We have a map On+1 → O(1) defined by coordinate func-
tions that is surjective on each D(xi ) and therefore surjective, but does not split
as there are no maps from O(1) → O. Global section functor of sheaves not
right exact: cause of cohomology.
Notice difference between vector bundles in alg and diff geom: in diff geom
all vector bundles isomorphic to their duals as we can put a metric on them,
but in alg geom O(1) is quite different from its dual O(−1) Also vector bundles
in diff geom are projective. Serre-Swan theorem: vector bundes over a smooth
manifold or affine variety = coherent projective sheaves; this completely fails
for projective varieties.
Discuss free/stably free/projective/flat/torsion free
Free and stably free: example of a stably free module that is not free is
the tangent space of a 2-sphere: not free by hairy sphere theorem, stably free
as adding normal bundle makes it free. Algebraically we work over the ring
R[x, y, z]/(x2 + y 2 + z 2 − 1) and the tangent bundle T is the submodule of R3

53
of elements (a, b, c) with xa + yb + zc = 0. An isomorphism from R3 → T ⊕ R
takes (A, B, C) → (A − xD, B − yD, C − zD), D where D = xA + yB + zC.
Compare stably free and projective. Stably free modules are obviously
projective as any direct summand of a free module is projective. Note we
cannot allow infinitely generated free modules in the definition of stably free
as for any projective P there is a free module F with P ⊕ F = F : take
F = P ⊕ Q ⊕ P ⊕ Q ⊕ . . . . where P ⊕ Q is free, and use the Eilenberg swin-
dle. Example of a projective that is not stably free: Moebius band over circle
(cant get rid of reveral of orientation by adding free bundles), many other
√ vector
bundles,
√ non-principal ideals over Dedekind domains such as (2, 1 + −5) over
Z[ −5]. It is usually hard to distinguish free and stably free modules: Serre’s
question asked whether every projective=stably free module over k[x1 , . . . xn ] is
free (this is easy to prove for topological vector bundels over real affine space).
Solved positively by Quillen and Suslin.
Although free modules are projective, it is NOT true that free sheaves are
projective! In fact most schemes have very few projective sheaves. Example:
for the variety P 1 , the map from O(0) ⊕ O(0) → O(1) is surjective, but there
are no maps from O(1) → O(0), so O(1) is not projective. Twisting by O(−1)
shows that the free module O(0) of rank 1 is not projective either. In general
there are not enough projective sheaves to form projective resolutions, though
there are enough injective sheaves.

Warning 4 Vector bundles over schemes or varieties behave differently from

topological vector bundles over smooth manifolds. For example, over smooth
manifolds, vector bundles are always self dual, because we can put a positive
definite bilinear form on them using a partition of unity, but over algebraic
projective space the dual of O(1) is O(−1). Over smooth manifolds the spaces
of smooth sections of vector bundles are flat modules over the ring of smooth
functions, but the analogue of this for algebraic vector bundles is false. Over
smooth vector bundles exact sequences of bundles split (use positive definite
inner product again) but this is completely false for bundles over projective
space.
Projective and flat modules: Projectives are flat because any direct sum-
mand of a flat module is flat. Examples of flat modules that are not projective:
the module Q over Z. Finitely generated flat modules M over a Noetherian local
ring R are projective and even free: choose a map 0 → K → Rn → M → 0
that is an isomorphism when reduced mod m. Then since Tor1 (M, R/m) = 0,
we have K = Km. Since K is finitely generated, Nakayama’s lemma implies
that K = 0. (Proof of Nakayama: We have g = Ag for some matrix A with
coefficients in m where g is a vector of generators of M , and det(1 − A) is a unit,
so 1 − A is invertible.) So over locally Noetherian schemes, finitely generated
flat sheaves correspond roughly to “vector bundles” as they are free over local
rings.

Example 75 Classify vector bundles over P 1 . (Grothendieck.) Over A1 all

54
vector bundles are free (coordinate ring is a PID). So cover P 1 by 2 copies of
A1 on each of which vector bundle is free of rank n. Transition functions are
given by an n × n matrix with coordinates in k[x, x−1 ]. We can multiply on
left by an invertible mattrix in k[x] and on right by invertible matrix in k[x−1 ].
Using these operations we can make transition matrix M diagonal with entries
powers of x as follows. Use operations to make 1 column zero except for one
element, necessarily of the form xr , and choose r as large as possible, and take
this element to be in top left corner. (To show that such a maximal r exists,
observe that it is bounded above by integers t such that for some nonzero vector
v with coefficients in k[x−1 ], M v has all coefficients divisible by xt , and this t
is invariant under the matrix operations on the left and right, and is bounded
by the highest power of x appearing in M .) Then by induction we can make
matrix diagonal except for top row. Then using maximality of r we can clear
top row, using column operations to clear out powers of x that are at most
r, and row operations to clear out poers that are at least s where xs is the
diagonal power in this column. So if there are any entries left in the top row we
must have r < s contradicting maximality of s. So vector bundle splits as the
sum of one dimensional bundles of the form O(m). Number of copies of O(k)
is uniquely determined by looking at dimensioin of space of sections of bundle
twisted by line bundles.  
1 x
Example: is the matrix M is then it cannot be turned into the
x2
   
1 x
diagonal matrix ; instead it gets transformed into . This
x2 x
shows we need to take r maximal in the above argument. This corresponds to
the fact that the exact sequence 0 → O(0) → O(1)⊕O(1) → O(2) → 0 does not
split. So although any vector bundle is a sum of line bundles, it can be tricky
to find such a decomposition: one cannot just pick sub line bundles at random.
This classification is misleadingly simple: in this case all vector bundles are
sums of irreducibles, all irreducibles are 1-dimensional, and irreducibles form a
discrete set, none of which are true in general. Over other curves vector bundles
are quite hard to classify, and classifying them over varieties of dimension ¿1 is
a very difficult open problem, even for projective spaces. Example: Horrocks-
Mumford rank 2 bundle on P 4 .
Flat and torsion free: Any flat module is torsion free (multiplication by non
zero divisor is injective) as follows by taking tensor product with 0 → R →a R.
Torsion free module that is not flat: look at submodule M of k[x, y] generated
k[x,y]
by x and y. Then Tor1 (M, k) = k so M is not flat, but is a submodule of a
torsion free module so is torsion free.

2.6 Divisors
Divisors on a compact Riemann surface. A divisor D is a finite linear com-
bination of points with integer coefficients: free abelian group with basis the
Riemann surface. Degree = sum of coefficients. A meromorphic function has

55
a divisor of zeros, of degree 0 because #zeros-#poles=change in argument. 2
divisors equivalent if difference is divisor of a meromorphic function.
Divisor classes -¿ line bundles
Riemann Roch. H 0 (X, O(D)) = space of functions whose only poles are
along D.
H 1 (X, O(D)) = space of obstructions to finding a function with given sin-
gular parts, except that we only don’t care about singularities of order less than
n at the point x if D = nx + . . . .
Then dim H 0 (X, O(D)) − dim H 1 (X, O(D)) = deg D + 1 − g where g =
dim H 1 (X, O) is the genus. Proof: trivial for D = 0. Adding a point to D
increases H 0 by 1 if there is a function with a singularity of the appropriate
order at this point. If there is not then H 1 is decreased by 1. So in either
case the left hand side increases by 1. (Nontrivial point: show H 1 has finite
dimension.)
Roch’s part: H 0 (X.O(−D) × H 1 (X, O(K − D)) → C given by sum of
residues is a perfect pairing. Generalized by Serre duality. K = canonical class
= zeros of a meromorphic 1-form; gives a well defined divisor class. Combining
these gives usual RR theorem:
dim H 0 (X, O(D)) − dim H 0 (X, O(K − D)) = deg D + 1 − g
Exercise: show deg K = 2g − 2.
Picard variety.
Weil divisor: analogue of divisor of a Riemann surface.
Total quotient ringK of a ring R: invert all non zero divisors. R ⊆ K.
Problem: when does this give a sheaf? Easy if R is integral domain, as then
any open subset has sections K. Hartshorne cheats slightly by simply taking
the sheaf associated to this presheaf, but this loses control of it. Vital principal:
keep track of values of a sheaf on (a basis of) open affine sets. We show that
associated sheaf has same values as presheaf on open affines for Noetherian
schemes. Suppose Spec(R) is covered by open set D(fi ) (1 = a1 f1 + . . .) and we
have sections ai /bi where bi is not a zero divisor in Rfi . Can assume ai bj = aj bi
by multiplying by powers of the f ’s. Then bi is not a zero divisor in Rfi so
anything killing bi is killed by a power of fi . Want to find A, B with Abi = Bai .
Let I be ideal of elements such that Iai ⊆ Rbi for all I, so that I contains all
b’s and therefore has no annihilator (as any annihilator is killed by a power of
any f and therefore killed by 1, as the f generate the unit ideal). We want to
show that I has a non zero divisor B. Now use the fact that R is Noetherian
so we have a good theory of associated primes. If every element of I is a zero
divisor then I is in the union of the associated primes of R, which are finite in
number. But any ideal contained in a finite union of primes is contained in one
of them (prime avoidance: if ai not in Pi but in the others then a1 + a2 a3 . . . .an
not in any) so I is contained in some associated prime of R and therefore kills
some nonzero element of R, contradiction.
Cartier divisor: sheaf theoretic. Sections of K ∗ /O∗ . Cartier divisors ⊆Weil
divisors. Same as Weil on integral separated Noetherian schemes such that local
rings are UFDs (e.g. regular: deep theorem). Nonregular local rings not UFDs
in general, example k[[x2 , x3 ]]. Example of a nonregular local ring that is a

56
UFD: R = k[x, y, z]/(x2 + y 3 + z 5 ) is a UFD, so its localization at 0 is as well.
Proof: The subring k[x/z 3 , y/z 2 ] is a UFD as it is a polynomial ring, and it
contains z −1 . So its localization at z −1 is a UFD and is the same as R[z −1 ].
Now check that z is a prime of R. Since z is a prime and R[z −1 ] is a UFD, R
is also a UFD.

2.7 Projective morphisms

2.8 Differentials
2.9 Formal schemes

57