Thesis
Thesis
Thesis
Trieste, Italy
Title
Syzygies
Author
Yairon Cid Ruiz
Supervisor
Prof. Lothar Göttsche
- August, 2016 -
To mom and dad,
of course.
ii
Algebra is the offer made by the devil to the mathematician.
The devil says: “I will give you this powerful machine, it will
answer any question you like. All you need to do is give me
your soul: give up geometry and you will have this marvellous
machine”.
Michael Atiyah
iii
Acknowledgements
I am very grateful to my supervisor Lothar Göttsche for his invaluable support, guidance
and encouragement throughout the preparation of this project and during all this year in
ICTP. Very special thanks go to Tarig Abdelgadir for his advices, the long and helpful
conversations, and all the interesting things I was able to learn in his reading courses. I
feel deep gratitude in my heart to all the professors of the Mathematics Section of ICTP
for showing me a whole new world and for helping me to better appreciate the beauty of
mathematics.
iv
Contents
Introduction 1
1 Basic concepts 3
1.1 Modules and Resolutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.2 Two important types of rings . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.3 Homology and derived functors . . . . . . . . . . . . . . . . . . . . . . . . 8
1.4 Regular sequences and the Koszul complex . . . . . . . . . . . . . . . . . . 13
1.5 Dimension and depth . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
3 Applying syzygies 53
3.1 The µ-basis of a rational surface . . . . . . . . . . . . . . . . . . . . . . . . 53
3.2 The freeness of the syzygies . . . . . . . . . . . . . . . . . . . . . . . . . . 58
3.3 The outline of our proof . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
3.4 Bounds for the regularity and the Betti numbers . . . . . . . . . . . . . . . 64
3.5 The case with projective dimension one . . . . . . . . . . . . . . . . . . . . 72
3.6 The case with projective dimension two . . . . . . . . . . . . . . . . . . . . 74
3.7 Appendix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
Bibliography 82
v
Introduction
The word “syzygy” comes to mathematics from a rather interesting route. It comes from
the Greek “syzygos” and meaning yoked together. It was widely used in astronomy in the
18th century to describe three planets in a line (and thus yoked by some mysterious force).
Cayley and Sylvester were the first to use it in mathematics.
The current usage inside mathematics is as follows. If M is a module and F a free module
with a surjective map ϕ : F → M , then the kernel K = Ker(ϕ) is called the 0th syzygy
of M . The three modules K, F and M lie in a “line”
ϕ
0→K→F −
→ M → 0,
In general we can continue this process of finding the “syzygies” and construct a “free
resolution” for a module. For example, let R = K[x, y] and M = R/m where m = (x, y).
A generator for M is given by 1 and we find
π
0→m→R→
− M → 0.
The module of syzygies is m, which is generated by x and y. These elements are not
independent, since they satisfy the non-trivial relation (−y)x + xy = 0. Resolving again
for m we get
−y
x x y
0 → R −−−−−−→ R2 −−−−−−→ m → 0.
Finally gluing together these two exact sequences we find a free resolution for M
−y
x x y π
0 → R −−−−−−→ R2 −−−−−−→ R →
− M → 0.
1
2
It turns out that the finiteness of the process in the previous example is not casual
and every finitely generated module M (not necessarily graded) over a polynomial ring
R = K[x1 , . . . , xn ] , where K is a field, has a free resolution of length at most n.
As in many places of mathematics Hilbert was the first person looking at this “type” of
problems; and in his famous “Syzygy Theorem”, published in 1890, he states that every
L
finitely generated graded module M = Mi over the polynomial ring C[x1 , . . . , xn ] has a
free resolution of length at most n.
In this report we shall study several results and theorems related with syzygies, just like
previous one. The outline is as follows
• In Chapter 1, we summarize the basic concepts and tools coming from Commutative
Algebra and Homological Algebra that we will need throughout this work.
– The Hilbert’s Syzygy Theorem which could be seen as the starting point of this
now rich theory.
One of the main goals of this report is giving a “fair” introduction to the “theory of syzygies”
and make it an accessible read for any graduate student without previous knowledge on
the subject (like myself).
Chapter 1
Basic concepts
In this chapter we will introduce some basic facts with the hope of making a self-contained
work, but we will only prove a statement, if proving it is as easy as quoting it. The main
reference for the topics related with Commutative Algebra will be the classic book by
Atiyah and MacDonald [10] and for the use of some tools coming from Homological Algebra
we will follow the book by Rotman [19].
Throughout all this exposition the word “ring” shall mean a commutative ring with
an identity element, hence we will differ mainly in this aspect from the text [19], e.g.
for us will not exist “left R-modules” or “right R-modules”, for us only matters the term
“R-modules”.
Definition 1.1. Let R be a ring (commutative with identity element, as always) then an
R-module is an abelian group M on which R acts linearly by a mapping R × M → M that
satisfies the axioms
a(x + y) = ax + ay
(a + b)x = ax + bx
(ab)x = a(bx)
1x = x
3
1.1. Modules and Resolutions 4
One of the most natural ways of constructing an R-module is taking several copies of R,
more formally by defining the free module which is an R-module of the form i∈I Mi ,
L
where each Mi ∼
= R. It turns out that we can always give a good description of an arbitrary
module using these free modules.
Construction 1.2. Let M be an arbitrary module and choose a set of generators {xi }i∈I
L
of this module, then we define the free module F0 = i∈I Rxi ; hence we can obtain an
d0
exact sequence 0 → K0 → F0 − → M → 0. We may repeat this operation for K0 to
obtain the exact sequence 0 → K1 → F1 → K0 → 0, and iterating successively we get
0 → Kn → Fn → Kn−1 → 0 where each Fn is a free module.
Gluing all these short exact sequences we obtain the long exact sequence
dn+1 d n d2 1 d0 d
. . . −−−→ Fn −→ ... −
→ F1 −
→ F0 −
→ M →0
After this construction several questions come to our minds because certainly this con-
struction is not unique neither the syzygies of M . To solve this problem of uniqueness
we will use some techniques from Homological Algebra, the path that we have chosen to
achieve this goal is different from the one taken in the book by Eisenbud [3]. We have
decided to work in the general settlement of a commutative ring with a unit element
without assuming extra properties, although later on we will use this marvelous text [3]
for stronger results that can only be achieved making additional assumptions. The main
ingredient for this generalization will be that of a projective module.
Definition 1.3. A projective module P is a module that given any surjective R-linear
map β : B → C and any R-linear map α : P → C, there exists an R-linear map φ : P → B
such that α = β ◦ φ, i.e makes the following diagram
P
φ
α
B C 0
β
commute.
1.2. Two important types of rings 5
This type of module will give us the fundamental tool for describing modules in general.
dn+1 nd 2 d 1 d 0 d
. . . −−−→ Pn −→ ... −
→ P1 −
→ P0 −
→ M →0
The study of projective modules generalizes that of free modules because every free module
is a projective module, and we will see that by doing this we have identified the key
property that will abstract and simplify our work.
Then we define the term syzygy for any projective resolution of a module and we will see
how this theory is independent of the chosen projective resolution.
dn+1 n d 2 d 1 d0 d
Definition 1.6. Let . . . −−−→ Pn −→ ... −
→ P1 −
→ P0 −
→ M → 0 be a projective resolution
of M . For n ≥ 0 we denote Kn = Ker(dn ) as the nth syzygy of M.
In this short section we shall introduce the basic definitions of modules over a local ring
and over a graded polynomial ring. One of the biggest advantages of working over
these two types of rings is the possibility of applying Nakayama’s lemma in a pleasant
way.
Definition 1.7. A ring R with exactly one maximal ideal m is called a local ring and
will be denoted by (R, m) (or (R, m, K) if we want to stress the residue field K ∼
= R/m).
Definition 1.8. For the polynomial ring R = K[x1 , . . . , xn ] we will use the standard
grading with deg(xc11 xc22 . . . xcnn ) = c1 + c2 + . . . + cn . The K-vector space generated by the
monomials of degree i is denoted by Ri . The graded polynomial ring R has a direct sum
decomposition R = i∈N Ri as k-vector spaces with Ri Rj ⊆ Ri+j for all i, j ∈ N.
L
1.2. Two important types of rings 6
L
We say that M is a graded R-module, if it has a direct sum decomposition M = i∈Z Mi
as K-vector spaces and Ri Mj ⊆ Mi+j for all i, j ∈ Z. Of particular interest for us is the
free graded module R(−p) defined by R(−p)i = Ri−p , i.e., for p ≥ 0 the module R(−p) is
shifted p degrees with R(−p)p = R0 = K.
Lemma 1.9. (Nakayama’s lemma for local rings) Let (R, m, K) be a local ring and M be
a finitely generated R-module.
(i) If mM ∼
= M , then M = 0.
Proof. (i) Suppose that {x1 , . . . , xr } is a minimal system of generators for M . Using the
hypothesis mM ∼ = M we have x1 = α1 x1 + . . . + αr xr for some α1 , . . . , αr ∈ m. Since
1 − α1 is invertible we get the contradiction x1 = (1 − α1 )−1 (α2 x2 + . . . + αr xr ).
(i) If IM = M , then M = 0.
Proof. (i) Since M is finitely generated there exists a smallest integer j ∈ Z with Mj 6= 0.
Because I is a proper ideal all homogeneous elements in I has positive degree and this
implies the contradiction Mj 6⊂ IM .
With the use of Nakayama’s lemma we can define the concept of minimal number of
generators for finitely generated modules over a local ring or over a graded polynomial
ring, i.e., all minimal systems of generators have the same number of elements.
Lemma 1.11. Let (R, m, K) be a local ring (R = K[x1 , . . . , xn ] be the graded polynomial
ring with m = (x1 , . . . , xn ) the irrelevant ideal). Let M be a finitely generated R-module,
1.2. Two important types of rings 7
Proof. We shall prove that {v1 , . . . , vr } is a minimal system of generators for M if and
only if the set of residue classes {v1 , . . . , vr } is a basis for the K-vector space M ⊗R K.
Proof. From the initial exact sequence we can get the exact sequence
ψ ϕ
F/mF −
→ G/mG −
→ H/mH → 0.
The map ϕ is an isomorphism if and only if ϕ maps the elementary vectors of G/mG into
a basis of the K-vector space H/mH (∼= H ⊗R K), by Nakayama’s lemma this happens if
and only if ϕ maps the elementary vectors of G into a minimal system of generators of H.
Definition 1.13. A complex (or chain complex) C is a sequence of modules and maps
dn+1 nd
C : . . . → An+1 −−−→ An −→ An−1 → . . . , n ∈ Z
dn+1 dn d 2 d1
. . . −−−→ Pn −→ ... −
→ P1 −
→ P0 → 0
and by doing this we do not loose any information since M = Coker(d1 ) = P0 /Im(d1 ) =
P0 /Ker(d0 ). Then we will apply certain functors to this complex and by computing the
homology of the resulting complex we will obtain the derived functors T or and Ext. That
will be our two fundamental tools.
dn+1
Definition 1.15. Given a module M with its respective deleted complex C : . . . −−−→
dn d2 d1
Pn −→ ... −→ P1 −→ P0 → 0 coming from a projective resolution. For any module N we
apply the functor ⊗R N
dn+1 ⊗R 1 n R d ⊗ 1 d ⊗ 1 d ⊗ 1
D : . . . −−−−−→ Pn ⊗R N −−−−→ . . . −−2−−
→ P1 ⊗R N −−1−−
R R
→ P0 ⊗R N → 0
dn+1
Definition 1.16. Given a module M with its respective deleted complex C : . . . −−−→
dn d2 d1
Pn −→ ... −
→ P1 −
→ P0 → 0 coming from a projective resolution. For any module N we
1.3. Homology and derived functors 9
1 d∗ 2 nd∗ d∗ d∗n+1
E : 0 → HomR (P0 , N ) −
→ HomR (P1 , N ) −
→ . . . −→ HomR (Pn , N ) −−−→ . . .
Now surprisingly enough we have that both functors are independent of the projective
resolution chosen which is a result coming from the property of projective modules (see
[19], Theorem 6.9 (Comparison Theorem)).
(3) T ornR (M, N ) does not depend on the projective resolution chosen.
(2) ExtnR (M, N ) does not depend on the projective resolution chosen.
A fundamental result that we will use several times is the following “long exact sequence”
argument for the functors Tor and Ext.
Proof. (See [19], Theorem 6.21, Theorem 6.26 and Theorem 6.27 ).
Using the concept of split exact sequence (for a good explanation on this, see [13] page
131 ) there is a complete characterization of projective modules.
Theorem 1.20. A module M is projective if and only if every short exact sequence
0 → A → B → M → 0 splits.
M
φ
id
0 A B M 0,
πM
F0 M
i
φ
ω α
β
B C 0,
then we see that M satisfy diagram of Definition 1.3 with φ = ω ◦ i, and finally M is a
projective module.
Corollary 1.21. A module M is projective if and only if M is a direct summand of a free
R-module. More specifically, M is projective if and only if F0 ∼
= M ⊕ K0 .
i
Proof. Look at the previous proof and see that all is around the exact sequence 0 → K0 →
−
d0
F0 −
→ M → 0.
Example 1.22. We have Z/15Z = Z/5Z ⊕ Z/3Z, then both Z/5Z and Z/3Z are projective
Z/15Z-modules, but neither of them are free because they have less than 15 elements.
And we end this section with the following theorem that is a stone where much of our
work rests. We have tried to give our own proof by using explicitly the syzygies of the
construction 1.2, which seems reasonable given the title of this work.
Theorem 1.23. Let M be a module if Ext1 (M, N ) = 0 for every module N , then M is a
projective module.
i0 d
0 → K0 →
− F0 −
→ M → 0, (1.1)
dn+1 dn d3 d2 d1 d0
... Fn ... F2 F1 F0 M 0
d1 id id
i d0
0 K0 F0 M 0
1.3. Homology and derived functors 12
This map is well-defined from the R-linearity of ϕ(−) and [−], is an R-linear map because
Φ(α ∗ϕ+β ∗ψ)(x) = (α ∗ϕ+β ∗ψ)([x]) = α ∗ϕ([x])+β ∗ψ([x]) = α ∗Φ(ϕ)(x)+β ∗Φ(ψ)(x).
It is bijective because any f ∈ Hom(F1 , K0 ) that vanishes on K1 is constant on every
coset of F1 /K1 .
d1
F1 F0
d1 id
i
K0 F0
d∗1
Hom(F1 , K0 ) Hom(F0 , K0 )
∗
d1 id
i∗
Hom(K0 , K0 ) Hom(F0 , K0 )
.
1
A good way of imagine this isomorphism is like when we compute the coordinate ring of an affine
variety.
1.4. Regular sequences and the Koszul complex 13
0 Hom(K0 , K0 ) Hom(F0 , K0 )
ϕ id
i∗
Hom(K0 , K0 ) Hom(F0 , K0 ),
∗
where ϕ is an isomorphism, because the fact that d1 is surjective implies that d1 is
injective. From this follows that necessarily the map i∗ : Hom(F0 , K0 ) → Hom(K0 , K0 ) is
surjective and there exists g ∈ Hom(F0 , K0 ) such that i∗ (g) = idK0 , that is g ◦ i = idK0 .
Therefore we say that the exact sequence 1.1 splits and following the proof of the previous
Theorem 1.20 we get that M is projective.
Remark 1.24. For a short proof of the previous theorem, see [19] page 199, simply apply
i∗
Theorem 1.19 to the short exact sequence 1.1 and obtain Hom(F0 , K0 ) −
→ Hom(K0 , K0 ) →
Ext1 (M, K0 ) = 0.
In this section we will discuss the concept of regular sequence and its tight relation with the
Koszul complex. Our exposition will be “mixed” between the books [9], [11] and [6]. From
[9] we will use Section 4 in Chapter 21, from [11] the Section 16 and from [6] the Section 1.6.
(i) M/(x1 , . . . , xn )M 6= 0
The sequence (x1 , . . . , xn ) is called a weak M-sequence if only the condition (ii) is re-
quired to be satisfied.
1.4. Regular sequences and the Koszul complex 14
Theorem 1.26. If x1 , . . . , xn is an M -sequence then then so is xk11 , . . . , xknn for any positive
integers k1 , . . . , kn .
We start with a small discussion on the simplest type of Koszul complex. For any x ∈ R
we can construct
x
K(x) : 0 → R → − R → 0,
and here we see how special are the homologies, with H1 (K(x)) = Ann(x) and H0 (K(x)) =
R/xR.
K0 (x) = R;
K1 (x) = free module E with basis {e1 , . . . , en };
..
.
^p
Kp (x) = free module E with basis ei1 ∧ . . . ∧ eip , i1 < . . . < ip ;
..
.
^n
Kn (x) = free module E of rank 1 with basis e1 ∧ . . . ∧ en .
by
p
(−1)j+1 xij ei1 ∧ . . . ∧ ecij ∧ . . . ∧ eip .
X
d(ei1 ∧ . . . ∧ eip ) =
j=1
A simple verification shows that d2 = 0, then using the previous modules and the boundary
map, we define K(x) as
In the previous definition we see that the zero homology is H0 (K(x)) = R/I, where
1.4. Regular sequences and the Koszul complex 15
The next Lemma gives the “invariant property” that we would like for the Koszul complex.
Suppose that x = (x1 , . . . , xn ) and y = (y1 , . . . , yn ) are n-tuples of elements in R. If the
0
ideal I = (y1 , . . . , yn ) is contained in the ideal I = (x1 , . . . , xn ), then we can make
n
X
yi = aij xj with aij ∈ R.
j=1
0 0
Let {e1 , . . . , en } be a basis for K1 (y) then we define the R-linear map f : K1 (y) → K1 (x)
given by
n
0 X
f (ei ) = aij ej ,
j=1
f ∧p : Kp (y) → Kp (x),
0
0 Kn (y) ... Kp (y) ... K1 (y) R R/I 0
det(A) f ∧p f id can
Therefore, if two n-tuples x = (x1 , . . . , xn ) and y = (y1 , . . . , yn ) generate the same ideal
and the determinant of the linear transformation is a unit, then the two Koszul complexes
are isomorphic. An important and special case is when (y) is a permutation of (x).
1.4. Regular sequences and the Koszul complex 16
ei+1 ie ei−1 1 e
K: . . . −−→ Ki −
→ Ki−1 −−→ . . . −
→ K0 → 0
and
fi+1 if fi−1 1f
L: . . . −−→ Li −
→ Li−1 −−→ . . . −
→ L0 → 0,
di+1 M d
i
M dk−1 d1
K ⊗L : . . . −−→ (Kj ⊗R Lk ) −
→ (Kj ⊗R Lk ) −−→ . . . −
→ (K0 ⊗R L0 ) → 0,
j+k=i j+k=i−1
K(x1 , . . . , xn ) ∼
= K(x1 ) ⊗ . . . ⊗ K(xn ).
Proof. Follows from the definitions of tensor product of complexes and Koszul complex.
Now we can extend the notion of Koszul complex for any R-module.
Given an element x ∈ R we study what happens what when we tensor an arbitrary complex
dp+1 dp d2 d1
C : . . . −−→ Cp −→ . . . −
→ C1 −
→ C0 → 0 with the simple Koszul complex K(x). We can
1.4. Regular sequences and the Koszul complex 17
0 → C → C ⊗ K(x) → (C ⊗ K(x))/C → 0.
x
The Koszul complex K(x) is simply 0 → R → − R → 0, then (C ⊗ K(x))p = (Cp ⊗R R) ⊕
∼
(Cp−1 ⊗R R) = Cp ⊕ Cp−1 . Making this identification we can make explicit the previous
exact sequence of complexes as
i πC p
0 Cp+1 Cp+1 ⊕ Cp Cp 0
dp+1 ep+1 dp
i πCp−1
0 Cp Cp ⊕ Cp−1 Cp−1 0
dp ep dp−1
i πCp−2
0 Cp−1 Cp−1 ⊕ Cp−2 Cp−2 0
∂
. . . → Hp (C) → Hp (C ⊗ K(x)) → Hp ((C ⊗ K(x))/C) →
− Hp−1 (C) → . . . ,
by the previous discussions we have Hp ((C ⊗ K(x))/C) ∼ = Hp−1 (C) and from a simple
diagram chasing we get ∂ = (−1)p−1 x. Therefore we get the exact sequence
(−1)p−1 x
. . . → Hp (C) → Hp (C ⊗ K(x)) → Hp−1 (C) −−−−−→ Hp−1 (C) → . . . ,
(−1)p x
. . . → Hp (C) −−−−→ Hp (C) → Hp (C ⊗ K(x)) → Hp−1 (C) → . . . .
1.4. Regular sequences and the Koszul complex 18
For p ≥ 1, if u+v ∈ Cp ⊕Cp−1 is a cycle (i.e. ep (u+v) = 0) with u ∈ Cp and v ∈ Cp−1 , then
we get (−1)p xv = dp (u) and dp−1 (v) = 0. Thus ep+1 ((−1)p u) = (−1)2p xu + (−1)p dp (u) =
x(u + v), and we get at once xHp (C ⊗ K(x)) = 0.
We want to apply the previous considerations of the tensor product with C = K(x1 , . . . , xn−1 ; M )
and x = xn . To abbreviate a little bit, we will use the notation Hp K(x; M ) instead of
Hp (K(x; M )), and we are going to call it the Koszul homology.
(−1)p x
. . . → Hp K(x1 , . . . , xn−1 ; M ) −−−−→Hp K(x1 , . . . , xn−1 ; M ) → Hp K(x1 , . . . , xn ; M ) →
(−1)p−1 x
→ Hp−1 K(x1 , . . . , xn−1 ; M ) −−−−−→Hp−1 K(x1 , . . . , xn−1 ; M ) → Hp−1 K(x1 , . . . , xn ; M ) → . . . ;
Proof. These are consequences from Lemma 1.30 and Theorem 1.28.
(i) Let x = (x1 , . . . , xn ) be a regular sequence for M . Then Hp K(x; M ) = 0 for p > 0
(of course, H0 K(x; M ) = M/IM ), i.e., the augmented Koszul complex is exact.
(ii) Conversely, suppose R is a local ring (or a graded polynomial ring), and x =
(x1 , . . . , xn ) inside the maximal ideal (or x = (x1 , . . . , xn ) inside the irrelevant ideal
of the polynomial ring). Suppose M is finitely generated over R, and H1 K(x; M ) = 0.
Then (x1 , . . . , xn ) is an M -regular sequence.
Proof. (i) We proceed making induction on n. For n = 1 is clear, because as previously seen
H1 K(x; M ) = {m ∈ M | xm = 0} = 0 when x ∈ R is regular on M . So we assume n > 1.
The case p > 1 comes directly from the piece of exact sequence Hp K(x1 , . . . , xn−1 ; M ) →
Hp K(x1 , . . . , xn ; M ) → Hp−1 K(x1 , . . . , xn−1 ; M ), which from the inductive hypothesis
turns into 0 → Hp K(x1 , . . . , xn ; M ) → 0. For p = 1, we take the tail of the sequence of
Theorem 1.31 and the isomorphism H0 K(x1 , . . . , xn−1 ; M ) ∼ = M/(x1 , . . . , xn−1 )M to get
the exact sequence
nx
H1 K(x1 , . . . , xn−1 ; M ) → H1 K(x1 , . . . , xn ; M ) → M/(x1 , . . . , xn−1 )M −→ M/(x1 , . . . , xn−1 )M.
−xj
H1 K(x1 , . . . , xj−1 ; M ) −−→ H1 K(x1 , . . . , xj−1 ; M ) → H1 K(x1 , . . . , xj ; M ),
xj
0 = H1 K(x1 , . . . , xj ; M ) → M/(x1 , . . . , xj−1 )M −
→ M/(x1 , . . . , xj−1 )M
case of R is a local ring or a graded polynomial ring, and can be achieved with only
checking the simple condition H1 K(x; M ) = 0. Also we can get important results for
regular sequences by using the Koszul complex, for example from Lemma 1.27 we know
that any permutation of a regular sequence is also regular in the case of local rings or
graded polynomial rings.
An important case where we want to apply the previous theorem is when M = R, i.e.
x = (x1 , . . . , xn ) is an R-regular sequence. In this case we know that the augmented Koszul
complex is
0 → Kn (x) → . . . → K1 (x) → K0 (x) → R/I → 0
Therefore we can regard the augmented Koszul complex as a free resolution of R/I, when
the defining sequence of I is regular.
When the ring R is a Noetherian any ideal has a maximal M -regular sequence. Surprisingly
enough, the length of a maximal M -regular sequence is well-determined by means of the
Koszul complex. First we have to prove that the Koszul complex is an exact functor.
of homology modules.
Proof. The modules in the Koszul complex are free, hence flat R-modules.
1.4. Regular sequences and the Koszul complex 21
In general the Koszul complex cannot tell you if a sequence is exact or not (in the local
and the graded case we saw it is true). But when the ring R is Noetherian, it can give you
something even more important.
q = sup{i | Hi K(y1 , . . . , yn ; M ) 6= 0}
1 x
0 → M −→ M → M1 → 0.
x 1 1 x
. . . → Hp K(y; M ) −→ Hp K(y; M ) → Hp K(y; M1 ) → Hp−1 K(y; M ) −→ Hp−1 K(y; M ) → . . . ,
for every p. Thus Hq+1 K(y; M1 ) 6= 0 and Hp K(y; M1 ) = 0 for p > q + 1. But x2 , . . . , xm
is a maximal M1 -sequence in I and by the inductive hypothesis m − 1 = n − (q + 1).
Therefore m = n − q.
This previous theorem has the interesting interpretation, that regular sequences have
always length smaller or equal than the number of elements in any generating set of an
ideal and that all maximal regular sequences have the same length.
There is a dual version of the Koszul complex from which we can obtain all the same
1.4. Regular sequences and the Koszul complex 22
results. In fact, in the book of Eisenbud [3] is taken this approach to the Koszul cohomology.
1 n d∗ d∗
HomR (K(x), M ) : 0 → HomR (K0 (x), M ) −
→ . . . −→ HomR (Kn (x), M ) → 0.
The Koszul cohomology is given by H p (HomR (K(x), M )) and we are going to denote
it by H p K(x; M ).
It turns out that the Koszul complex is self-dual. With the isomorphism HomR (K(x), M ) ∼
=
∗ ∗
K(x) ⊗R M , we can reduce the problem to prove that K(x) and K(x) are isomorphic.
For the dual Koszul complex the basis of each module Kp (x)∗ is given by
where (ei1 ∧ . . . ∧ eip )∗ is the function that takes 1 on ei1 ∧ . . . ∧ eip and 0 on all the other
basis elements.
We can consider the following commutative diagram, that in fact induces an isomorphism
between the complexes K(x) and K(x)∗ (See [6], pages 47 and 48 ).
dn dp+1 dp d1
K(x) : 0 Kn (x) ... Kp (x) ... K0 (x) 0
ωn ωp ω0
d∗1 d∗n−p d∗n−p+1 d∗n
K(x)∗ : 0 K0 (x)∗ ... ∗
Kn−p ... Kn (x)∗ 0
(i) The complexes K(x) and K(x)∗ are isomorphic (the previous diagram is an isomor-
phism between them; we say that K(x) is self-dual).
(ii) More generally, for every R-module M the complexes K(x; M ) = K(x) ⊗R M and
1.5. Dimension and depth 23
(iii) Hp K(x; M ) ∼
= H n−p K(x; M ) for p = 0, . . . , n.
One very basic notion in mathematics is that of dimension and we shall deal with the
concept in this section. Here we will call the classic Krull dimension, define the projective
dimension and give a meaning to the length of “maximal regular sequences”.
Let R be a ring. The supremum of the lengths r, taken over all strictly decreasing chains
p0 ) p1 ) . . . ) pr of prime ideals of R, is called the Krull dimension, or simply the
dimension of R, and denoted by
dim R.
The codimension of a prime ideal p is defined as codim p = dim Rp and the dimension
as dim p = dim R/p. For an arbitrary ideal I we define the codimension as
Now comes the numerical invariant with biggest importance for us.
nd 2 d 1 d 0 d
0 → Pn −→ ... −
→ P1 −
→ P0 −
→ M → 0.
1.5. Dimension and depth 24
Example 1.41. Let V be a module over the field K (vector space) then pdK (V ) = 0, since
we can always find a basis.
Proposition 1.42. Let {Kn } be the syzygies of M coming from a projective resolution of
M , then Extn+1 (M, N ) ∼
= Ext1 (Kn−1 , N ) for every module N and every n ≥ 1.
n d 2 d
1 d
0 d
Proof. Suppose that . . . → Pn −→ ... −
→ P1 −
→ P0 −
→ M → 0 is the projective resolution
that defines the syzygies {Kn }. Then using Kn−1 = Ker(dn−1 ) = Im(dn ) we get the
following projective resolution of Kn−1
and from this follows Ext1 (Kn−1 , N ) = Ker(d∗n+2 )/Im(d∗n+1 ) = Extn+1 (M, N ).
Theorem 1.43. The following three conditions are equivalent for a module M
(1) pd(M ) ≤ n;
Proof. (1) ⇒ (2) If pd(M ) ≤ n then there is a projective resolution where Pk = 0 for all
k ≥ n + 1. Therefore Hom(Pk , N ) = 0 for any module N and also
(3) ⇒ (1) We proceed with the construction 1.2 of finding a free resolution (which is also
projective) until we obtain an exact sequence
dn−12 d
1 0d d
0 → Kn−1 → Fn−1 −−−→ . . . −
→ F1 −
→ F0 −
→ M →0
where Kn−1 is the (n − 1)th syzygy. If we obtain before a zero syzygy then the result
follows trivially. Otherwise, by hypothesis and Proposition 1.42 we have Ext1 (Kn−1 , N ) =
Extn+1 (M, N ) = 0 for all N . Then Theorem 1.23 implies that Kn−1 is projective and
pd(M ) ≤ n.
Corollary 1.44. Let M be an R-module with pd(M ) ≤ n, then for any projective resolution
of M the corresponding syzygies Kj (j ≥ n − 1) are projective modules.
Proof. For j ≥ n − 1 and any module N we have Ext1 (Kj , N ) = Extj+2 (M, N ) = 0 by
Proposition 1.42 and the previous Theorem 1.43. Thus Kj is projective from Theorem
1.23.
Corollary 1.45. Suppose M is a module with pd(M ) = n < ∞ then the sequence
dn−12 d
1 0d d
0 → Kn−1 → Fn−1 −−−→ . . . −
→ F1 −
→ F0 −
→ M →0
This previous corollary says that any module M with n = pdR (M ) < ∞ has a projective
resolution of length n that is “almost free” and that our initial and rather naive construc-
tion 1.2 will always give a projective resolution exactly at the step n = pd(M ), i.e., a
projective resolution of minimal length. Later we will see that for a finitely generated
module over K[x1 , . . . , xn ] we can get a free resolution of length n = pd(M ). For a very
nice treatment on Free Resolutions one can see [15] in Chapter 6.
Proof. Using the previous characterization of projective dimension and Theorem 1.18 (3),
the result follows.
Example 1.48. Let R = K[x]/(x2 ) and M the R-module defined by M = (x), then
pdR (M ) = ∞.2
Proof. First let’s compute who are F0 and K0 , since M is generated by one element
x
we get F0 = R, then K0 = Ker(R → − M ) = (x) = M . Inductively for any n we get
x
the exact sequence 0 → Kn → R → − Kn−1 (= M ) → 0, then all syzygies are equal to
x
Ker(R →− M) = M.
i x
For any y ∈ R \ M we have xy ∈ / R \ M , then the exact sequence 0 → K0 → − R→ − M →0
does not split because i(K0 ) = K0 (= M ) cannot be a direct summand of R. From Theorem
1.20 we get M is not projective. Therefore all syzygies are not projective and by Corollary
1.45 we get pdR (M ) = ∞.
After defining the projective dimension of any R-module then we can define the global
dimension of the ring R.
gDim(R) = sup{pdR (M ) | M ∈ MR },
From Theorem 1.35 in the previous section, we saw that for a Noetherian ring R and a
finite R-module M in the case IM 6= M we can define the numeric invariant of “length of
maximal sequences”.
Definition 1.52. Let R be a Noetherian ring, M a finite R-module, and I an ideal such
that IM 6= M . The common length of all the maximal M -sequences in I is called the
depth of I on M, denoted by depth(I, M). When M = R is simply called the depth
of I and denoted by depth(I). If IM = M we adopt the convention depth(I, M ) = ∞.
2
Taken from [15], Exercise 11, page 258.
1.5. Dimension and depth 27
(ii) Conversely, if R is Noetherian, and M , N are finite and HomR (N, M ) = 0, implies
that I contains an M -regular element.
x
0 → Extn−1 n
R (N, M/x1 M ) → ExtR (N, M ) −
→1
ExtnR (N, M ).
But the multiplication by x1 annihilates N , therefore we have the isomorphism ExtnR (N, M ) ∼
=
Extn−1
R (N, M/x1 M ). Again, the inductive hypothesis and the fact that (x2 , . . . , xn ) is
regular in M/x1 M , gives the expected result ExtnR (N, M ) ∼ = HomR (N, M/xM ).
Theorem 1.55. (Rees). Let R be a Noetherian ring, M a finite R-module, and I an
ideal such that IM 6= M . Then all maximal M -sequences in I have the same length n
given by
depth(I, M ) = min{i | ExtiR (R/I, M ) 6= 0}.
by Proposition 1.53
Exti−1 ∼
R (R/I, M ) = HomR (R/I, M/(x1 , . . . , xi−1 )M ) = 0.
ExtnR (R/I) ∼
= HomR (R/I, M/xM ) 6= 0.
A very special case is when (R, m, K) is a Noetherian local ring, and M finite R-module.
We call the depth of M as
At this point we should stress that the condition IM 6= M is not superfluous and the
concept of depth is not well defined when IM = M .
Example 1.56. Let K be a field and R = K[[x]][y]. Then we have that {x, y} and {xy −1}
are both maximal R-sequences.3
Taking the quotient K[[x]][y]/(xy−1) (the “Rabinowski’s trick”), is like adjoining the inverse
of x, i.e., K[[x]][y]/(xy − 1) ∼
= K[[x]][x−1 ]. In general we know that ∞ k
k=0 ak x ∈ K[[x]] is
P
In the following proposition we collect some important results. (From now on, V (I) denotes
the set of prime ideals containing I.)
3
Exercise 1.2.20, [6].
1.5. Dimension and depth 29
In the case of Noetherian local rings there is an important relation between the depth and
the dimension.
Proposition 1.58. Let (R, m) be a Noetherian local ring and M 6= 0 a finite R-module.
Then depth(M ) ≤ dim(M ).
Proof. See Proposition 1.2.12, [6]. Also one could see Theorem 6.5, [11].
Proof. We have depth(I) = inf {depth(Rp ) | p ∈ V (I)} and codim(I) = inf {dim(Rp ) |
p ∈ V (I)}, then the previous Proposition 1.58 implies the inequality.
Finally, as one could expect there is a relation between the characterization of depth by
means of the Koszul complex and by means of the Ext functor.
Hn−m K(x, M ) ∼
= HomR (R/I, M/yM ) ∼
= Extm
R (R/I, M ).
In this chapter as the title says we will state and prove some interesting results in the
theory of syzygies. To begin with, we will choose the Hilbert’s syzygy theorem that was
the starting point of this theory in his famous paper on Invariant Theory [7]. Secondly
we shall prove a beautiful result conjectured by Serre: “every finitely generated projective
module over K[x1 , . . . , xn ] is free”. Lastly we shall deal with the theory of our syzygies in
the case of a local ring.
31
2.1. Towards Hilbert’s syzygy theorem and more 32
. . . → 0 → Extn+2 (C, N ) → 0 → . . .
which may be regarded as an R[x]-module taking into account the grading of the xk ’s.
Also we have that if P is R-projective then P ⊗R R[x] is R[x]-projective, suppose we have
the following diagram as in Definition 1.3, where B and C are R[x]-modules
P ⊗R R[x]
φ
α
B C 0.
β
Given the graded structure of P ⊗R R[x], it is enough to induce a function φ that accom-
plish α(P ⊗R xk ) = (β ◦ φ)(P ⊗R xk ) as an R-linear map and then we extend it to all of
P ⊗R R[x] with the grading of the xk ’s. The isomorphism P ⊗R xk ∼ = P , the projectivity
of P and looking at B and C as R-modules induces a map φ : P ⊗R R[x] → B, where
each restriction φ |P ⊗R xk is an R-linear map that makes commute the previous diagram.
nd 2 d 1 0 d d
0 → Pn −→ ... −
→ P1 −
→ P0 −
→ M → 0.
The ring of polynomials R[x] is a flat R-module (it is even free), then we can tensor
2.1. Towards Hilbert’s syzygy theorem and more 33
d ⊗ 1
n R d ⊗ 1 d ⊗ 1 d ⊗ 1
0 → Pn ⊗R R[x] −−−−→ . . . −−2−−
→ P1 ⊗R R[x] −−1−−
R
→ P0 ⊗R R[x] −−0−−
R R
→ M [x] → 0.
If M is an R[x]-module certainly we can see it as an R-module and we can also make the
construction M [x] = M ⊗R R[x] that is an R[x]-module, where technically the action of x
is given by x(m ⊗R xk ) = m ⊗R xk+1 . Although xm is perfectly well defined, the operation
x(m ⊗R xk ) = xm ⊗R xk is not correct for our construction 2.1.
α β
0 → M [x] −
→ M [x] →
− M → 0.
so m0 = m1 = . . . = mn = 0 and α is injective.
By the given definitions we have trivially β ◦ α = 0 then Im(α) ⊂ Ker(β). But let
w = nk=0 mk ⊗R xk ∈ Ker(β), hence β(w) = 0 and nk=0 xk mk = 0, which gives
P P
un−1 = mn
un−2 = mn−1 + xmn
un−3 = mn−2 + xmn−1 + x2 mn
..
.
u0 = m1 + xm2 + . . . + xn−1 mn
2.1. Towards Hilbert’s syzygy theorem and more 34
n−1 n−1
k
(uk−1 − xuk ) ⊗R xk + un−1 ⊗R xn
X X
α( uk ⊗R x ) = −xu0 ⊗R 1 +
k=0 k=1
n
mk ⊗R xk .
X
=
k=0
At this point we are finally ready to state and prove the main result of this section which
relates the projective dimension over R[x] and over R.
pdR[x] (M ) ≤ 1 + pdR (M ).
α β
0 → M [x] −
→ M [x] →
− M → 0,
with Lemma 2.1 we find the inequality pdR[x] (M ) ≤ 1 + pdR[x] (M [x]) and Lemma 2.2 gives
pdR[x] (M [x]) ≤ pdR (M ), which combined prove our assertion.
Corollary 2.5. Let M be an R[x1 , x2 , . . . , xn ]-module then pdR[x1 ,x2 ,...,xn ] (M ) ≤ n +
pdR (M ).
Proof. Make induction by successively taking R[x1 , . . . , xk−1 , xk ] = R[x1 , . . . , xk−1 ][xk ].
Corollary 2.6. Let K be a field and M be a K[x1 , x2 , . . . , xn ]-module then pdK[x1 ,x2 ,...,xn ] (M ) ≤
n.
Remark 2.8. We have achieved our goal of exposing a version of Hilbert’s syzygy theorem
in the most general settlement inside Commutative Algebra, i.e for commutative ring R;
for a different approach see [3] in Chapter 19 and [5] in Chapter 3. Now unfortunately we
will have to put additional properties to our ring R in order to get more interesting results.
In the future we will stress which type of ring we are working and if it is not mentioned by
2.1. Towards Hilbert’s syzygy theorem and more 35
Proposition 2.9. Let R be a Noetherian ring and M a finitely generated R-module with
pdR (M ) = n < ∞, then it has a projective resolution of length n which is composed of
finitely generated modules.
Proof. Again our prays will be answered by the construction 1.2. Take a projective
resolution like in Corollary 1.45
dn−1 2 d 1 d0 d
0 → Kn−1 → Fn−1 −−−→ . . . −
→ F1 −
→ F0 −
→ M → 0.
Since M is finitely generated then can we can take the free module F0 also as a finitely
generated module. The module F0 is Noetherian because R is Noetherian and F0 is finitely
generated (see [3] Proposition 1.4 ), therefore we have that K0 = Ker(d0 ) ⊂ F0 is finitely
generated because is a submodule of a Noetherian module.
Proof. We can continue infinitely the process in the previous Proposition 2.9 to obtain
dn+1 dn dn−1 2 d 1 d0 d
. . . −−−→ Fn −→ Fn−1 −−−→ . . . −
→ F1 −
→ F0 −
→ M → 0,
Proof. From Corollary 2.6 we have pd(M ) ≤ n. The ring K[x1 , x2 , . . . , xn ] is Noetherian
because of Hilbert Basis theorem (see [3] Theorem 1.2 ), then the previous Proposition
2.9 implies M has a projective resolution of length at most n that is composed of finitely
2.2. On Serre’s conjecture 36
generated modules. Finally by the Quillen-Suslin Theorem 2.31 the modules in the
projective resolution are actually free.
We finish this section by remarking that in Corollary 2.7 we have actually an equality
gDim(R[x1 , x2 , . . . , xn ]) = n + gDim(R),
but proving it lies outside the interests of this work. The way of proving it is by making a
change of rings (see [19] page 245). Here we have tried to give an example to illustrate
this construction, where is present the equality pdR[x] (R) = 1 + pdR (R).
Example 2.12. Let R be a ring and R[x] its polynomial ring, then pdR[x] (R) = 1.
Proof. Certainly we have to elaborate more the enunciate: how can we see R as an R[x]-
module? Using the projection map π : R[x] → R[x]/xR[x] ∼ = R, we define for p ∈ R[x]
and r ∈ R the scalar product p ∗ r = π(p)r, i.e., if p(x) = an xn + . . . + a0 then p ∗ r = a0 r
(for a better explanation on this type of construction, see [10] page 30 ). Then we can get
a projective resolution of R as an R[x]-module by
x π
0 → R[x] →
− R[x] →
− R→0
where the first map is multiplication by x, hence we get pdR[x] (R) ≤ 1. But xR[x] cannot
be a direct summand of R[x] and then the previous exact sequence does not split. Finally,
by Theorem 1.20 we get R is not a projective R[x]-module and pdR[x] (R) = 1.
It the famous article “Faisceaux algébriques cohérents” (FAC, 1955), Serre wrote: “On
ignore s’il existe des A-modules projectifs de type fini qui ne soient pas libres” (A =
K[x1 , . . . , xn ], K a field); shortly thereafter, the freeness of finitely generated projective
modules over K[x1 , . . . , xn ] became known to the mathematical world as "Serre’s Conjec-
ture". After twenty years, in 1976 Daniel Quillen and Andrei Suslin independently proved
that finitely generated projective modules over K[x1 , . . . , xn ] are, indeed, free.1
1
Quoted from [8], where there is a complete development both historical and mathematical of the
Serre’s Conjecture.
2.2. On Serre’s conjecture 37
During this section we will prove this strong assertion, that in part gave the Fields Medal
to Quillen in 1978. In [8] there are several proofs, but we have chosen to follow one that is
very close to our work. Using our developed machinery to deal with syzygies we will prove
the Serre’s theorem: every finitely generated projective module over K[x1 , . . . , xn ] is stably
free. Finally we will follow the “Suslin’s elementary proof” to show that every stably free
module over K[x1 , . . . , xn ] is free.
From now on, we will use the term f.g. every time we want to say finitely generated. We
will follow the proof from [19] (pages 251 − 256) and we will fill some details that are left
to the reader.
Lemma 2.15. A projective module P has a finite free resolution if and only if is stably
free.2
Proof. (⇐) If P is stably free then F = P ⊕ G with F and G f.g. free. Therefore
0 → G → F → P → 0 is a finite free resolution of P .
0 → Fn → . . . → F1 → K0 → 0.
By the inductive hypothesis K0 is stably free and there exists f.g. free modules F and G
with F = G ⊕ K0 . Therefore we have G ⊕ F0 ∼ = G ⊕ K0 ⊕ P ∼ = F ⊕ P and P is stably free.
Lemma 2.16. Given two exact sequences, where the P ’s and Q’s are projective,
dn+1n d 2dn−1 1 0 d d d
0 → A −−−→ Pn −→ Pn−1 −−−→ . . . −
→ P1 −
→ P0 −
→ M → 0, (2.2)
en+1 en en−1 e2 e1 e0
0 → B −−→ Qn −→ Qn−1 −−→ . . . −
→ Q1 −
→ Q0 −
→ M → 0, (2.3)
then A ⊕ Qn ⊕ Pn−1 ⊕ . . . ∼
= B ⊕ Pn ⊕ Qn−1 ⊕ . . . .3
Proof. For the exact sequences 2.2 and 2.3 we denote their syzygies as Kj = Ker(dj ) and
Lj = Ker(ej ). For their corresponding 0th syzygies we have K0 and L0
0 d
0 → K 0 → P0 −
→ M →0
0 e
0 → L0 → Q0 −
→ M → 0.
Using that P0 and Q0 are projective modules, by Schanuel’s Lemma (Theorem 3.62, [19])
we get the isomorphism K0 ⊕ Q0 ∼ = L0 ⊕ P0 .
dn+1
n d 2dn−1 1 d d
0 → A −−−→ Pn −→ Pn−1 −−−→ . . . −
→ P1 −
→ K0 → 0,
en+1
n e 2 en−1
1 e e
0 → B −−→ Qn −→ Qn−1 −−→ . . . −
→ Q1 −
→ L0 → 0,
where d1 and e1 are the restrictions of d1 and e1 to their images. Then we can construct
the new exact sequences
dn+1 n 3d 1 2d i ◦d 0
d1 ⊕idQ
0 → A −−−→ Pn −→ ... −
→ P2 −− −→ P1 ⊕ Q0 −−−−−→ K0 ⊕ Q0 → 0 (2.4)
en+1n 3 e 1 2 e j ◦e
0
e1 ⊕idP
0 → B −−→ Qn −→ ... −
→ Q2 −−−→ Q1 ⊕ P0 −−−−→ L 0 ⊕ P0 → 0 (2.5)
where i1 (x) = (x, 0) ∀x ∈ P1 , j1 (y) = (y, 0) ∀y ∈ Q1 and idP0 , idQ0 are identity maps.
From Ker(d1 ⊕ idQ0 ) ∼ = Ker(d1 ) = K1 , Ker(e1 ⊕ idP0 ) ∼
= Ker(e1 ) = L1 , Ker(i1 ◦ d2 ) =
3
Exercise 3.37 from [19].
2.2. On Serre’s conjecture 39
A ⊕ Qn ⊕ Pn−1 ⊕ . . . ∼
= B ⊕ Pn ⊕ Qn−1 ⊕ . . . .
Lemma 2.17. Let R be a Noetherian ring and M an R-module with a finite free resolution
of length ≤ n. Then every free resolution(not necessarily finite) of M composed of f.g.
modules, has a stably free and f.g n-th syzygy.4
n d 2 1d 0 d d
Proof. Let 0 → Fn −→ ... −
→ F1 −→ F0 − → M → 0 be the given finite free resolution of
M with length n (if it had length smaller than n we fill it with some Fj = 0). All syzygies
are f.g. because over a Noetherian ring submodules of f.g. modules are f.g. (just like in
Proposition 2.9).
en+1 ne en−1 2 1 e 0 e e
Let . . . −−→ Gn −→ Gn−1 −−→ . . . −→ G1 −→ G0 −→ M → 0 be a free resolution of M
with all Gj f.g. free. Then we want to prove that Qn = Ker(en ) = Im(en+1 ) is stably free
and f.g., the finiteness of Qn follows in the same way by the previous Noetherian argument.
n d2 1 d0 d d
0 → 0 →Fn −→ ... −
→ F1 −
→ F0 −
→ M →0
i n e 2 1 e 0 e e
0 → Qn →
− Gn −→ ... −
→ G1 −
→ G0 −
→ M → 0,
n d 3 2d 1 d
0 d d
0 → Pn −→ ... −
→ P2 −
→ P1 −
→ P0 −
→ M → 0.
nd 3 d1 2 i ◦d
1 F d ⊕id
0
d0 ◦πP
0 → Pn −→ ... −
→ P2 −− −→ P1 ⊕ F −− −−→ P0 ⊕ F −−−−→ M → 0,
where i1 is an inclusion map (as before), idF the identity map and πM0 a projection map.
With the 0-th syzygy K0 = Ker(d0 ) we can get
n d 3 1 d2 i ◦d
1 F d ⊕id
0 → Pn −→ ... −
→ P2 −− −→ P1 ⊕ F −− −−→ K0 ⊕ F → 0
0
d0 ◦πP
0 → K0 ⊕ F → P0 ⊕ F −−−−→ M → 0.
Then K0 ⊕ F has a projective resolution of length n − 1 with each member stably free,
hence by the inductive hypothesis there exists a finite free resolution 0 → Em → . . . →
E0 → K0 ⊕ F → 0 and gluing it with the second equation we get the finite free resolution
0 → Em → . . . → E0 → P0 ⊕ F → M → 0
Proof. The two modules that have finite free resolution are f.g.; if L and M , then N is f.g.
because the second map is surjective; if L and N , then M ∼ = L + N is f.g.; if M and N ,
then L is f.g. because is a submodule of M and the Noetherian hypothesis. Therefore we
assume that the free modules are f.g.
.. .. ..
. . .
0 F1 G1 H1 0
0 F0 G0 H0 0
0 L M N 0
0 0 0
Assuming two of {L, M, N } have a finite free resolution, then by Lemma 2.17 two of
{Im , Jm , Km } are stably free, where m is the maximal length between the two given finite
free resolutions. If one of these is Km then the sequence 0 → Im → Jm → Km → 0
splits (Jm ∼= Im ⊕ Km ) and the third one is necessarily stably free, thus we get the
resolutions composed of stably free modules: 0 → Im → Fm → . . . → F0 → L → 0,
0 → Jm → Gm → . . . → G0 → M → 0 and 0 → Km → Hm → . . . → H0 → N → 0.
The remaining case is when Im and Jm are stably free, then by a simple diagram chasing
0 → Im → Jm → Hm → . . . → H0 → N → 0
Therefore in every cases by Lemma 2.18 the three modules L, M and N have finite free
resolution.
Let M be an R-module then its annihilator is the ideal Ann(M ) = {r ∈ R | rM = 0}, and
for an element x ∈ M its annihilator is the ideal Ann(x) = {r ∈ R | rx = 0}. Sometimes
we will write AnnR (M ) when we want to stress the ring over which we are taking the
annihilator.
2.2. On Serre’s conjecture 42
Proof. The collection of ideals (M ) is not empty since M 6= 0, then (M ) has a maximal
P P
element I = Ann(x) for some x ∈ M \ {0}, because R is Noetherian (see Chapter 6, [10]).
Suppose by contradiction a 6∈ I, b 6∈ I and ab ∈ I; then I + (a) ) I, bx 6= 0 and abx = 0.
Combining this we get Ann(bx) ⊃ I + (a) ) I, that contradicts the maximality of I in
P
(M ).
Proposition 2.21. Let R be an Noetherian ring and M 6= 0 be a f.g. R-module. Then
there is a chain
0 = M0 ( M1 ( M2 ( . . . ( Mn = M
where Mi /Mi−1 ∼
= R/pi where each pi is a prime ideal, and for each we have pi ⊃ Ann(M ).
P
Proof. By Lemma 2.20, let p1 = Ann(x1 ) be the prime ideal that is maximal in (M ),
M1 we get M1 ∼
x1
we define M1 = (x1 ) and with the surjective homomorphism R −→ = R/p1 .
P
In the same way choose p2 = Ann(x2 + M1 ) as the maximal element in (M/M1 ),
x2 +M1
define M2 = (x1 , x2 ) and with the surjective homomorphism R −− −−→ M2 /M1 follows the
∼
isomorphism M2 /M1 = R/p2 .
Proof. By contradiction suppose there is a f.g. R[x]-module that has no finite free resolution.
Then the family of ideals of R
X
= {AnnR (M ) | M is a f.g. R[x]-module and has no finite free resolution}
0 → M0 (= 0) → M1 → R[x]/p1 → 0
0 → M1 → M2 → R[x]/p2 → 0
..
.
0 → Mn−1 → Mn (= M ∗ ) → R[x]/pn → 0.
Thus if we show that each R[x]/pj has a finite free resolution, applying Theorem 2.19
iteratively we get that M ∗ has a finite free resolution, which is the contradiction we are
looking for.
Therefore we only have to deal with the case pk ∩ R = I, and I will be a prime ideal in R
because by Proposition 2.21 the ideal pk is prime in R[x]. Also in this case R0 ∼
= R/I is
an integral domain. Here using R[x]I ⊂ pk we get the isomorphism
R[x]/(R[x]I) ∼
R0 [x]/qk ∼
= = R[x]/pk (2.6)
pk /(R[x]I)
where R0 [x] = R[x]/(R[x]I) and qk = pk /(R[x]I). We may see R0 [x] and qk as R0 [x]-
modules and as R[x]-modules.
We have AnnR0 (qk /(f )) = J/I for some ideal J in R, then AnnR0 (qk /(f )) 6= 0 implies
J ) I. But then we have J(qk /(f )) = 0 when we look it as an R[x]-module via the map
R[x] → R[x]/(R[x]I). Thus by the maximality of I we get that qk /(f ) has a finite free
resolution as an R[x]-module.
By the hypothesis of the theorem we have that I has a finite free resolution
0 → Fr → . . . → F1 → F0 → I → 0,
2.2. On Serre’s conjecture 44
then R[x]I ∼
= I ⊗R R[x] has a finite free resolution.
we have that (f ) ∼
= R0 [x] because R0 [x] is an integral domain. Therefore applying Theorem
2.19 we finally get that qk and R0 [x]/qk ∼ = R[x]/pk have finite free resolution as R[x]-
modules.
Theorem 2.23. (Serre) If K is a field, then every f.g. projective K[x1 , . . . , xn ]-module
is stably free.
Proof. The proof follows by induction on n. If n = 1 then K[x] is a principal ideal domain
and clearly every f.g. module (ideal) has a finite free resolution (are actually free). If
n > 1 then by the induction hypothesis every f.g. module over K[x1 , . . . , xn−1 ] has a finite
free resolution. From Hilbert Basis Theorem K[x1 , . . . , xn−1 ] is a Noetherian ring, then by
Theorem 2.22 we get that every f.g. module over K[x1 , . . . , xn−1 , xn ] = K[x1 , . . . , xn−1 ][xn ]
has a finite free resolution.
In particular, any f.g. projective module over K[x1 , . . . , xn ] has a finite free resolution,
therefore by Lemma 2.15 is a stably free module.
In this subsection we shall prove that any stably free module over R = K[x1 , . . . , xn ] is a
free module. Our exposition will be based on Chapter 3, Section 2 from [8] and Chapter
21, Section 3 from [9]. We shall show a simplification made by Vaserstein to the proof of
Suslin.
2.2. On Serre’s conjecture 45
The ideas introduced by Suslin are very important because they established the calculus
of unimodular rows as a new theme for research in commutative ring theory, and made
the Serre’s conjecture accessible to any graduate student.
Definition 2.24. Let R be a ring and (f1 , . . . , fr ) ∈ R1×r be a row of elements. We say
that (f1 , . . . , fr ) is a unimodular row if f1 R + . . . + fr R = R.
For preference of notation we shall work using columns and throughout this subsection f
will always denote the column vector f = (f1 , . . . , fr )t ∈ Rr×1 . We say that f has the uni-
modular extension property if there exists a matrix GLr (R) with f as the first column.
Theorem 2.25. (Horrocks) Let (R, m) be a local ring and let R[x] be the polynomial ring
in one variable over R. Let f be a unimodular vector in R[x]r such that some component
has leading coefficient 1. Then f has the unimodular extension property in R[x].
Proof. If r = 1 the statement is trivial and if r = 2 we can always construct such matrix
because 1 ∈ (f1 , f2 ). Thus we assume that r ≥ 3 and we will proceed by induction on the
smallest degree d of a component of f with leading coefficient 1. We will make several
row operations that will take f into the elementary vector e1 = (1, 0, . . . , 0)t , since this
is equivalent to the existence of elementary matrices E1 , . . . , Em with Em . . . E1 f = e1 ,
we will get that f is the first column of the matrix E1−1 . . . Em
−1
∈ GLr (R[x]). Also these
elementary transformations does not change the unimodularity of a vector.
We assume that f1 has leading coefficient 1 and degree d. By the possibility of applying the
Euclidean division algorithm when a polynomial has leading coefficient 1, we may assume
that deg(fi ) < d for all 2 ≤ i ≤ r. Since f is unimodular, there exist polynomials gi with
Pr
i=1 fi gi = 1 and from this we may conclude that not all coefficients of f2 , . . . , fr belongs
to the maximal ideal m. By contradiction we suppose that all coefficients of f2 , . . . , fr
belong to m, then for the polynomial f1 g1 we have (f1 g1 )0 ∈ 1 − m and (f1 g1 )j ∈ m for
j ≥ 1. We have that f1 = xd + ad−1 xd−1 + . . . + a0 and g1 = cm xm + cm−1 xm−1 + . . . + c0 ,
the coefficient (f1 g1 )m+d = cm implies that cm ∈ m, and following inductively with the
coefficients (f1 g1 )k = ck−d + ad−1 ck−d+1 + . . . ∈ m until k = d we get that all ci ∈ m. So
we get a0 c0 ∈ 1 − m and a0 c0 ∈ m, that implies the contradiction 1 ∈ m.
Therefore we assume that some coefficient of f2 does not lie in m and so is a unit since R
2.2. On Serre’s conjecture 46
is local. We write
so that some bi is a unit. Let I be the ideal generated by all the leading coefficients
of polynomials h1 f1 + h2 f2 of degree ≤ d − 1. Then I contains all the coefficients
bi . This can be proved by induction, we have trivially that bs ∈ I, with the linear
combination ps−1 (x) = xd−s f2 (x) − bs f1 (x) = (bs−1 + bs ad−1 )xd−1 + . . . follows bs−1 ∈ I,
with ps−2 (x) = xps−1 (x) − (bs−1 + bs ad−1 )f1 (x) we get bs−2 ∈ I, and proceeding in this
way we can obtain that bi ∈ I for 0 ≤ i ≤ s.
Over a ring R, for two column vectors f and g we write f ∼GLr (R) g (or just f ∼ g) if
there exists M ∈ GLr (R) such that
f = M g,
and we say that f is equivalent to g over R. The previous theorem implies that over
a local ring (R, m), a unimodular vector f ∈ R[x]r with one component having leading
coefficient 1 is R[x]-equivalent to the elementary vector e1 = (1, 0, . . . , 0)t .
Corollary 2.26. Let (R, m) be a local ring. Let f be a unimodular vector in R[x]r with
some component having leading coefficient 1. Then f (x) ∼ f (0) over R[x].
Proof. We have that necessarily f (0) ∈ Rr has some component not inside m and so a
unit. Therefore, f (x) ∼ e1 ∼ f (0) over R[x].
Lemma 2.27. Let R be an integral domain, and let S be a multiplicative subset in R. Let
x, y be independent variables. If f (x) ∼ f (0) over RS [x], then there exists c ∈ S such that
f (x + cy) ∼ f (x) over R[x, y].
Proof. Let M ∈ GLr (RS [x]) be such that f (x) = M (x)f (0). Then M (x)−1 f (x) = f (0) is
2.2. On Serre’s conjecture 47
with H(x, y) ∈ RS [x, y]. We can clear denominators by choosing some c ∈ S such
that cH ∈ R[x, y], then G(x, cy) has coefficients in R. Since det(M (x)) is a con-
stant in RS then det(M (x + cy)) is equal to this same constant, and det(G(x, cy)) =
det(M (x))det(M (x + cy))−1 = 1. Therefore G(x, cy) ∈ GLr (R[x, y]) and G(x, cy)f (x +
cy) = f (x).
Theorem 2.28. Let R be an entire ring, and let f be a unimodular row in R[x]r , such
that one component has leading coefficient 1. Then f (x) ∼ f (0) over R[x].
Proof. Let J denote the set of elements c ∈ R such that f (x + cy) ∼ f (x) over R[x, y].
If c1 , c2 ∈ J and a ∈ R, then f (x + cay) ∼ f (x) over R[x, ay] (so also over R[x, y]) and
f (x + (c1 + c2 )y) ∼ f (x + c1 y) ∼ f (x) over R[x, y]. Therefore J is an ideal.
Let p be an arbitrary prime ideal in R. By Corollary 2.26 we have that f (x) ∼ f (0)
over Rp [x], and by Lemma 2.27 there exists some c ∈ R \ p such that f (x + cy) ∼ f (x)
over R[x, y]. Hence J is not contained in any maximal ideal, and we get that J = (1).
Therefore there exists an invertible matrix M (x, y) over R[x, y] such that
Since det(M (x, y)) is a unit in R then it does not depend on x neither y, and the
substitution x = 0 maintains the matrix invertible. So we get
and we rename y by x.
In one of the steps for proving Noether Normalization Theorem we normally show that for
2.2. On Serre’s conjecture 48
y n = xn , yi = xi − xri i , (2.7)
for some suitable ri ’s, we can get a new polynomial q(y1 , . . . , yn ) ∈ K[y1 , . . . , yn ] such that
p(x1 , . . . , xn ) = q(y1 , . . . , yn ) and the leading coefficient with respect to yn belongs to the
field K.
Theorem 2.29. Let K be a field and let f be a unimodular vector in K[x1 , . . . , xn ]r . Then
f has the unimodular extension property.
Therefore we assume that n ≥ 2. Now we see f as a polynomial vector in K[x1 , . . . , xn−1 ][xn ]
and making a substitution like in 2.7 we can get a polynomial vector g(y1 , . . . , yn ) such
that
f (x1 , . . . , xn ) = g(y1 , . . . , yn )
and has one component with leading coefficient 1 (actually a unit, but that’s the same to
our purposes).
From Theorem 2.28 there exist an invertible matrix N (y1 , . . . , yn ) ∈ K[y1 , . . . , yn ] such
that
g(y1 , . . . , yn ) = N (y1 , . . . , yn )g(y1 , . . . , yn−1 , 0)
and g(y1 , . . . , yn−1 , 0) is unimodular. Therefore from the inductive hypothesis we have
that
g(y1 , . . . , yn ) = M (y1 , . . . , yn )e1
with M invertible, and a substitution back to the variables xi ’s gives the result
r
f (x1 , . . . , xn ) = M (x1 − xr11 , . . . , xn−1 − xn−1
n−1
, xn )e1 .
Theorem 2.30. Let M be a stably free module over R = K[x1 , . . . , xn ], then M is free.
π : Rr → R
be the projection map induced from the direct sum M ⊕ R = Rr . We choose u1 as the gen-
erator of R inside M ⊕ R = Rr with π(u1 ) = 1, then we have that u1 = (a11 , a21 , . . . , an1 )t
is unimodular.
From Theorem 2.29 we find an invertible matrix F ∈ Rr×r with u1 as first column. Let
uj = F ej for j = 1 . . . , n,
where ej is the j-th unit column vector. Since F induces an automorphism over Rr , we
have that {u1 , u2 , . . . , ur } is a basis for Rr .
For i ≥ 2 we define
wi = ui − π(ui )u1 ,
Proof. By Theorem 2.23 f.g. projective K[x1 , . . . , xn ]-module is stably free. Then from
Theorem 2.30 every stably free K[x1 , . . . , xn ]-module is free.
This section discusses some interesting results for Noetherian local rings and throughout
this section R will always denote (R, m, K), i.e., a local ring with maximal ideal m and
residue field K.
In the case of local rings (and graded, as we will see later) we can take advantage of a
very special type of resolutions: “minimal free resolutions”. As the name suggests, now in
2.3. The Auslander-Buchsbaum formula 50
the Construction 1.2 rather than taking an arbitrary set of generators, we instead take in
every step a minimal set of generators. Making this process we obtain a “minimal free
resolution”
dn+1 dn d2 d1
F : . . . −−−→ Fn −→ ... −
→ F1 − → F0 .
From Theorem 1.12 we know that taking a minimal system of generators is equivalent to
the condition Im(di ) ⊂ mFi−1 for i ≥ 1. Therefore we can make the following definition
Definition 2.32. Let (R, m, K) be a local ring and M be an R-module. A free resolution
dn+1 nd 2 d 1 d 0 d
F : . . . −−−→ Fn −→ ... −
→ F1 −
→ F0 −
→ M →0
Theorem 2.33. Let (R, m, K) be a local ring and M be a finitely generated nonzero
R-module, then pdR (M ) is the length of every minimal free resolution of M . Furthermore,
pdR (M ) is the biggest integer i for which T oriR (K, M ) = T oriR (M, K) 6= 0.
R
T orj+1 (M, K) = 0 for j ≥ pdR (M ). (2.8)
nd 2 d 1 d0 d
F : 0 → Fn −→ ... −
→ F1 −
→ F0 −
→ M →0
is a projective resolution of length n. If i is the biggest integer for which T oriR (M, K) 6= 0,
then from 2.8 we have n ≥ pdR (M ) ≥ i. But if F is a minimal free resolution then the
differentials in the complex F ⊗R K are 0,
0 0 0
F ⊗R K : 0 → Fn →
− ... →
− F1 →
− F0 .
With this previous theorem we get a type of “Quillen-Suslin Theorem” 2.31 in the case of
local rings.
2.3. The Auslander-Buchsbaum formula 51
Corollary 2.34. Let (R, m, K) be a local ring and M be a finitely generated projective
module, then M is free.
Proof. We have pdR (M ) = 0, then with the previous theorem we have a minimal free
resolution of the form 0 → F → M → 0.
or just
pdR (M ) + depth(M ) = depth(R).
If pdR (M ) > 0, the we make the first step of taking a minimal free resolution
ϕ
F : 0→N −
→ F → M → 0,
We shall explode the characterization of depth by the Koszul complex (Theorem 1.35). Let
x = (x1 , . . . , xn ) be a set of generators of the maximal ideal m. We apply the Proposition
1.34 to obtain the long exact sequence
and Hj K(x, M ) = 0.
Hn−d+1 K(x, M ) 6= 0.
Since depth(N ) = d we know that Hn−d K(x, N ) 6= 0, and from the exact sequence
Φ
Hn−d+1 K(x, M ) → Hn−d K(x, N ) −
→ Hn−d K(x, F )
will be more that enough to prove that the induced map Φ from ϕ, is the zero map.
If pdR (N ) > 0, by the induction hypothesis we have d < depth(R), so that in fact
Hn−d K(x, F ) = 0. Otherwise pdR (N ) = 0, so that N is free, and we have
In this case the induced map Φ can be identified as ϕ ⊗ id. We know that Im(ϕ) ⊂ mF
and from Theorem 1.31 (ii) we have that m annihilates Hn−d K(x), thus the tensor product
ϕ ⊗ id is the zero map.
Chapter 3
Applying syzygies
In the article [1] is proved that any rational surface has a µ-basis. That strong result
obtained by Falai Chen, David Cox, and Yang Liu, geometrically means that any rational
surface is the intersection of three moving planes without extraneous factors. There, is
stated: “This is an unexpected result, for after ten years of exploration, researchers in the
geometric modelling community generally believed that this was not true”.
In [1] there are left several questions of interest for further research and better understanding.
During this chapter we shall address the question:
We will try to make a more or less detailed exposition of the bound we found for the degree
of the polynomials in a minimal µ-basis for any rational surface. This chapter could be seen
as our small result in this rich theory of syzygies and also it will serve the purpose of giving
small introduction to the study of the graded polynomial ring R = K[x1 , . . . , xn ].
In this section we will discuss the existence of a µ-basis for any rational surface, because
it turns out that a principal ingredient is dealing with our syzygies!. We give a different
proof for the result about syzygies stated there, which we also generalize.
Our goal here is to roughly give the principal definitions to prove the Theorem 3.1 in [1]
and use our different approach with syzygies.
53
3.1. The µ-basis of a rational surface 54
P (s, t) = (a(s, t), b(s, t), c(s, t), d(s, t)) (3.1)
where a, b, c, d ∈ R[s, t] are bi-degree (m, n) polynomials and gcd(a, b, c, d) = 1. Where the
rational surface is properly parametrized, i.e., the map
!
a(s, t) b(s, t) c(s, t)
(s, t) = , ,
d(s, t) d(s, t) d(s, t)
is birational.
Definition 3.2. A moving plane is a quadruple (A(s, t), B(s, t), C(s, t), D(s, t)) ∈ R[s, t]4
such that
a(s, t) b(s, t) c(s, t)
A(s, t) + B(s, t) + C(s, t) + D(s, t) = 0 (3.2)
d(s, t) d(s, t) d(s, t)
for all s, t ∈ R.
From the previous definition we see that a moving plane is a collection of planes parametrized
by polynomials, that for each s, t ∈ R the point a(s,t) , b(s,t) c(s,t)
,
d(s,t) d(s,t) d(s,t)
of the rational surface
is contained in the corresponding plane to (s, t).
We denote the collection of moving planes of the rational surface P (s, t) as Ls,t ⊂ R[s, t]4 .
The equation 3.2 is equivalent to
for all s, t ∈ R. Thus the polynomial above is the zero polynomial and the collection Ls,t
is equal to the 0-th syzygy K0 of the R[s, t]-module(ideal) generated by the polynomials
{a(s, t), b(s, t), c(s, t), d(s, t)}, that is I = (a, b, c, d) ⊂ R[s, t]. Thus we have the following
exact sequence
[a,b,c,d]
0 → K0 (= Ls,t ) → R[s, t]4 −−−−→ I → 0. (3.3)
In [1] is used the notation Syz(a, b, c, d) to denote the 0-th syzygy K0 , which seems appro-
priate now to use because we want to stress that the previous exact sequence depends on
the generators {a, b, c, d}, and is not a general construction like 1.2 where we pointed out
that was not important the generators chosen.
for some constant α ∈ R. Then p, q, r are said to be a µ − basis of the rational surface
3.1. Where [p, q, r] is defined as the outer product
p2 p3 p4 p1 p3 p4 p1 p2 p4 p1 p1 p3
[p, q, r] = q2 q3 q4 , − ,
q q3 q4 q1 q2 q4 , − q q2 q3 .
1 1
r2 r3 r4 r1 r3 r4 r1 r2 r4 r1 r2 r3
The previous condition means that the rational surface P (s, t) can be represented as the
intersection of three moving planes p, q and r without extraneous factors.
Now we state the following result about syzygies (Proposition 2.1 in [1]).
Proposition 3.4. Let a, b, c, d ∈ R[s, t] be four relatively prime polynomials. Then the
syzygy module Syz(a, b, c, d) is a free module of rank 3.
Proof. The proof of being free will be given in the next section and the proof of having
rank 3 can be found in [1].
In the following theorem is proved the existence of a µ-basis for any rational surface and
we will see that the previous result on syzygies is the fundamental base that reduces all to
linear algebra.
Theorem 3.5. For any rational surface defined as in 3.1, there always exist three moving
planes p, q, r such that 3.4 holds.1
Proof. Using that a, b, c, d are coprime then by Proposition 3.4 the R[s, t]-module Syz(a, b, c, d)
is a free module of rank 3, and we choose a basis {p, q, r}. Then we shall prove there is a
linear dependence between [p, q, r] and P (s, t) (as elements in a module), i.e., there exist
polynomials f, g ∈ R[s, t] such that f [p, q, r] = gP (s, t), and we also may assume that
1
See [1] for a different argument in proving the linear dependence between [p, q, r] and P (s, t).
3.1. The µ-basis of a rational surface 56
and if we prove that every 2 × 2-minor vanishes then we get the expected result. For
example if we take the first minor we get
b −a 0 0
p2 p3 p4 p1 p3 p4
p1 p2 p3 p4
b q2 q3 q4 + a q1 q3 q4 = = 0.
q1 q2 q3 q4
r2 r3 r4 r1 r3 r4
r1 r3 r3 r4
The last equality follows from the fact (b, −a, 0, 0) ∈ Syz(a, b, c, d) and p, q, r form a basis
for Syz(a, b, c, d). In the same way we can compute that all minors vanish.
Since gcd(f, g) = 1 and gcd(a, b, c, d) = 1, we may assume f = 1. Using that (−b, a, 0, 0),
(−c, 0, a, 0) and (−d, 0, 0, a) belongs to Syz(a, b, c, d), there exist polynomials hi,j ∈ R[s, t]
such that
Forming the outer product of the above three vector polynomials, one has
Thus we have g|a2 , and similarly we can get g|b2 , g|c2 and g|d2 . Hence g|gcd(a, b, c, d) = 1,
g is a constant and we get the equality
[p, q, r] = αP (s, t)
for some α ∈ R.
The term “without extraneous factors” means that we can achieve the [p, q, r] = αP (s, t)
with α ∈ R, because we can always get some vectors p, q, r that achieve the equality
3.1. The µ-basis of a rational surface 57
“with extraneous factors”, that is [p, q, r] = gP (s, t) for some g ∈ R[s, t]. For instance, if
P = (a, b, c, d), then we make
p = (−d, 0, 0, a)
q = (0, −d, 0, b)
r = (0, 0, −d, c)
Surprisingly enough, there is a 1 − 1 relationship between a µ-basis and a basis for the
syzygy module Syz(a, b, c, d) given in the following form:
• (Corollary 3.1, [1]) p, q and r form an µ-basis if and only if p, q and r are a basis
of Syz(a, b, c, d).
Definition 3.6. p, q and r are said to form a minimal µ-basis of the rational surface
3.1 if
(i) among all the triples p1 , q1 and r1 satisfying 3.4, degt (p) + degt (q) + degt (r) is the
smallest, and
(ii) among all triples p1 , q1 and r1 satisfying 3.4 and item (i), degs (p)+degs (q)+degs (r)
is the smallest.
Here, degt (p) = max1≤i≤4 (degt (pi )) when p = (p1 , p2 , p3 , p4 ), and degt (q), degt (r), degs (p),
degs (q), degs (r) are defined similarly.
will be to prove the existence of a basis for the syzygy module Syz(a, b, c, d), where the poly-
nomials in this basis are bounded in terms of the maximal degree max(deg(a), deg(b), deg(c),
deg(d)) (deg denotes the maximal total degree in the variables s and t).
3.2. The freeness of the syzygies 58
Proving the freeness of Syz(a, b, c, d) is an important step in [1] to show the existence
of a µ-basis. In this section we give a different proof for that statement, which we also
generalize because we will need the case of three variables after homogenizing the ideal
I = (a, b, c, d).
Statement in the Appendix of [1]: Let K be a field and R = K[s, t]. For any
f1 , . . . , fk ∈ R, the syzygy module Syz(f1 , . . . , fk ) = {(h1 , . . . , hk ) ∈ Rk | h1 f1 + . . . +
hk fk ≡ 0} is a free module.
We claim that for any n ≥ 2 and any ideal in K[x1 , . . . , xn ] the (n-2)-th syzygy Kn−2
is free (for us the first syzygy is the 0-th syzygy, although there are other sources that
start counting in 1). For n = 1 is not needed because any nonzero ideal in k[x] is
principal, therefore is isomorphic to k[x] and clearly free. For n = 2 is just the previous
statement, because by Hilbert Basis Theorem any ideal is finitely generated, so for any
ideal I = (f1 , . . . , fk ) we have that K0 = Syz(f1 , . . . , fk ) is free. For n = 3 we have that
K1 is free, i.e., the syzygy of Syz(f1 , . . . , fk ) is free.
We shall see that the proof will be rather short by quoting from previous sections. In some
sense the next theorem is like the Hilbert’s Syzygy Theorem in a sharped way for modules
inside the polynomial ring, i.e., ideals.
0 → I → R → R/I → 0.
Using Theorem 1.19 for any R-module N we get the exact sequence
By Corollary 2.6 we know that pdR (R/I) ≤ n, then applying Theorem 1.43 we get
Extn+1 n
R (R/I, N ) = 0. We always have ExtR (R, N ) = 0 (for n ≥ 1), thus we transform the
previous exact sequence into
. . . → 0 → ExtnR (I, N ) → 0 → . . . ,
3.2. The freeness of the syzygies 59
Now we are ready to prove our promised generalization of Proposition 3.4 in the following
theorem.
d
n dn−1 dn−2 dn−3
1 d 0 d
Proof. Let . . . −→ Pn−1 −−−→ Pn−2 −−−→ Pn−3 −−−→ . . . −→ P0 −
→ I → 0 be a projective
resolution of I where each Pi is a f.g. projective module. From Quillen-Suslin Theorem
2.31 we know that in this case the modules Pi are actually free.
With the obtained bound pdR (I) ≤ n − 1 and Corollary 1.44, we get that the (n-2)-th
syzygy Kn−2 = Ker(dn−2 ) is a projective module. Using Pn−2 is f.g. and R Noetherian
we get that Pn−2 is a Noetherian module, then Kn−2 ⊂ Pn−2 is a f.g. module. Finally, by
the Quillen-Suslin Theorem 2.31 the (n-2)-th syzygy Kn−2 is a free K[x1 , . . . , xn ]-module.
Remark 3.9. In the previous proof we only need that Pn−2 is f.g. to get that Kn−2 is
free. Here we see that for the special case n = 2 we can take a projective resolution for
I = (f1 , . . . , fk ) as
[f1 ,...,fk ]
. . . → R[s, t]k −−−−−→ I → 0,
i [f1 ,...,fk ]
− R[s, t]k −−−−−→ I → 0
0 → K0 = Syz(f1 , . . . , fk ) →
Remark 3.10. We have given a specific reference for the Hilbert Syzygy Theorem as
Corollary 2.6, because the standard proof using the Koszul complex only works in the
graded case. For instance, in the excellent book [3] by Eisenbud the proof of Hilbert Syzygy
Theorem for the non graded case says:
• (Corollary 19.8 in [3]) “Every finitely generated module over K[x1 , . . . , xn ] has a
finite free resolution”.
3.2. The freeness of the syzygies 60
Actually if we go through the proof of this Corollary, we see that can only be concluded the
existence of a resolution of length ≤ n + 1, and this result is obtained after a process of
“homogenizing” the initial module.
Using a computer algebra system like Macaulay2 ([2], [4]) we can check the veracity of
all kinds of results we expect to be true or that we think we have proven. To test our
previous result, we programmed the following simple code to see if we could find an ideal
with projective dimension equal to numVar, i.e., numVar is n in the polynomial ring
K[x0 , . . . , xn−1 ].
numVar = 3
R = QQ[vars (0..(numVar-1))]
constructRandomIdeal = (numGen, maxDeg) -> (
I := {};
for i from 1 to numGen do (
polynom := random(maxDeg,R);
I = append(I, polynom);
);
ideal I
)
checkIneq = (numIter, numGen, maxDeg) -> (
J := 0;
for i from 1 to numIter do (
I := constructRandomIdeal(numGen, maxDeg);
n := pdim module I;
if n == numVar then (
J = I;
break;
);
);
J
)
The implemented function checkIneq tries to find an ideal with projective dimension
equal to numVar and generated by numGen polynomials of degree at most maxDeg;
and the function tries to do this process numIter times. We have run the previous
function checkIneq with several configurations and never has been able to find an ideal
I ∈ K[x0 , . . . , xn−1 ] with pd(I) = n. Therefore computations made in Macaulay2 agree
with our given proof.
We could play a little more with Macaulay2 and using its functionalities we can implement
a very short algorithm to compute the projective dimension of a module. From Corollary
1.45 we know that the “obvious” method of computing one kernel (syzygy) after another
will always give the projective dimension of a module. The following simple algorithm does
this iterative process and with it computes the projective dimension of a module.
3.3. The outline of our proof 61
In this section we describe how we are going to prove the existence of a basis with a certain
bound in terms of the maximal degree in the polynomials a, b, c and d. For notational
purposes from now on, we assume that a1 , a2 , a3 , a4 ∈ R[s, t] are the polynomials defining
the rational surface 3.1 and that d = max1≤i≤4 (deg(ai )). Therefore we will try to solve
the following problem for the rest of this document.
For us is not clear at all how to define a bound in the “degree of the polynomials” if we
are working with modules over the polynomial ring R[s, t]. Instead we will try to reduce
the problem to a graded case where we can exploit the structure of the grading.
Given the ideal I = (a1 , a2 , a3 , a4 ) ⊂ R[s, t], then we define the homogeneous ideal
Iˆ = (b1 , b2 , b3 , b4 ) with
s t
bi = ud ai ( , ) ∈ R[s, t, u]. (3.5)
u u
For the ideal Iˆ ⊂ R[s, t, u] we can find a graded free resolution that starts as
we can continue this process of finding a graded free resolution until the step
and from Theorem 3.8 we know that the first syzygy K1 is a free R[s, t, u]-module. Therefore
we can assure that the ideal Iˆ has a graded free resolution of the form
Intuitively we could expect that if we make the substitution u = 1 in the previous resolution
we find a free resolution (perhaps no longer graded) of the ideal I = (a1 , a2 , a3 , a4 ). This
hunch turns out to be true and the way of state it rigorously is to say that tensoring with
⊗R[s,t,u] R[s, t, u]/(u − 1) is an exact functor, for a proof of this fact one can see Corollary
19.8 in [3] or Proposition 1.1.5 in [6].
Finding a resolution like in 3.6 with certain “regularity” or bound for the degree of the
polynomials involved in the differential maps of 3.6 will give us a resolution of I in the
form
[a1 ,a2 ,a3 ,a4 ]
0 → R[s, t]b → R[s, t]a → R[s, t]4 −−−−−−−→ I → 0,
d ⊗ R d ⊗ R
0 → Rb −−−−→ Ra −−
2 S 1 S
−−→ Syz(a1 , a2 , a3 , a4 ) → 0.
We know that Syz(a1 , a2 , a3 , a4 ) is a free module of rank 3, thus the previous exact
sequence splits and we can conclude a = b + 3. Unfortunately we will have to work
a lot more to give a bound in this case and we will dedicate a complete section for
this case.
We can easily compute various examples for the two previous cases.
ˆ
This section will be devoted to find a bound for the regularity and the Betti numbers of I.
We will try to be more specific and rigorous than in the previous section and our main
reference will be the chapter “Graded Free Resolutions” of [17]. For convenience we will
make an exposition for S = R[s, t, u], although all the definitions given here work for the
general case S = K[x1 , . . . , xn ].
di+1 d
i d
2 1d 0 d
. . . −−→ Fi −
→ ... −
→ F1 −
→ F0 −
→ M → 0,
d0 : F0 → K−1
fj 7→ mj for 1 ≤ j ≤ r,
Step i + 1: Set Ki = Ker(di ) (from Proposition 2.9 in [17], we know that Ker(di ) is a
graded module). Choose homogeneous generators l1 , . . . , ls of Ki . Let c1 , . . . , cs be their
3.4. Bounds for the regularity and the Betti numbers 65
di+1 : Fi+1 → Ki ⊂ Fi
gj 7→ lj for 1 ≤ j ≤ s,
The constructed complex is exact and is just like in 1.2 but only taking some extra care for
the homogeneous elements.
Example 3.16. Let R = K[x, y] and J = (x3 , xy, y 5 ), then for the module R/J we have
the following graded free resolution
y 0
−x −y 4
2
0 x x3 xy y 5
0 → R(−4)⊕R(−6) −−−−−−−−−−−→ R(−3)⊕R(−2)⊕R(−5) −−−−−−−−−−−→ R → R/J → 0.
As a consequence of the Nakayama’s lemma (Lemma 1.10 for the graded case) we can
define the concept of minimal system of homogeneous generators for any finitely
generated graded S-module (Lemma 1.11). Using this fact we can give a “minimal
resolution” for any module M by choosing a minimal system of generators in each step of
the previous construction. Luckily, from Theorem 1.12 we can come up with an equivalent
condition which is more easy definition to handle.
Definition 3.17. A graded free resolution of a graded finitely generated module S-module
is said to be minimal if
This means, that no invertible elements (non-zero constants) appear in the differential
matrices.
The resolution given in Example 3.16 is minimal, but the ones given in Example 3.11
and Example 3.12 are not minimal. Unfortunately we are interested in resolutions that
are not always minimal, because we want extract information about the syzygy module
Syz(b1 , b2 , b3 , b4 ) and in several cases {b1 , b2 , b3 , b4 } is not a minimal system of homogeneous
3.4. Bounds for the regularity and the Betti numbers 66
Theorem 3.18. (i) A graded free resolution is minimal if and only if at each step we
choose a minimal system of homogeneous generators.
Using this convenient uniqueness we can define the Betti numbers for a graded finitely
generated S-module M . For the rest of this section Fi will denote the modules in a minimal
free resolution of M .
βi (M ) = dimR (Fi ).
Theorem 3.20.
βi (M ) = dimR (T oriS (M, R)).
Proof. If we tensor the minimal free resolution F of M with ⊗S R we can obtain the
complex
0 0 0 0
F ⊗S R : − Rβi (M ) →
... → − Rβ1 (M ) →
− ... → − Rβ0 (M ) .
Thus we have
T oriS (M, R) ∼
= Hi (F ⊗S R) = Rβi (M ) .
Definition 3.21. We define the graded Betti numbers of M by
In the same way we can compute the graded Betti numbers by means of the T or functor.
Theorem 3.22.
βi,p (M ) = dimR (T oriS (M, R)p ).
3.4. Bounds for the regularity and the Betti numbers 67
Just as in the case for local rings, for graded polynomial rings the projective dimension is
determined by the length of a minimal free resolution.
Proof. Almost verbatim to Theorem 2.33, with m = (s, t, u) the irrelevant ideal.
Corollary 3.24. Let M be a graded finitely generated projective S-module, then M is free.
We would like to know how the graded Betti numbers behave and thus we study the
concept of regularity.
Computing the depth of the ideal Iˆ will proof that the condition gcd(a1 , a2 , a3 , a4 ) = 1 is
an essential assumption to obtain good bounds for the regularity and the Betti numbers.
Proposition 3.26. ˆ ≥ 2.
depth(I)
Proof. Step 1: From the hypothesis gcd(a1 , a2 , a3 , a4 ) = 1 we can conclude that gcd(b1 , b2 ,
b3 , b4 ) = 1. By contradiction suppose that g ∈ R[s, t, u] is a non-constant polynomial with
g | bi , then we may assume that g is a polynomial that depends only on the variable
u because g(s, t, 1) is a common divisor for the polynomials ai . From the construction
3.5 we know that one of the polynomials bi has a term that is free of u, without loss of
generality we assume that b1 (s, t, u) = λsα td−α + up(s, t, u) with p ∈ R[s, t, u]. Then for
some h ∈ R[s, t, u] we have b1 (s, t, u) = g(u)h(s, t, u) and thus h has a term of the form
3.4. Bounds for the regularity and the Betti numbers 68
βsα td−α that multiplied with the highest power of u in g gives a term of degree bigger
than d which cannot be reduced, and this contradicts the fact that b1 is homogeneous of
degree d.
Step 2: There exists two elements p and q in Iˆ = (b1 , b2 , b3 , b4 ), that are relatively prime,
i.e., gcd(p, q) = 1. Disappointingly, we were able to prove this statement only by making a
more complicated reformulation.
For n = 1 we assume that the only polynomial in this case is equal to 1, so the ideal
is whole polynomial ring and we can find such infinite sequence. In the case n > 1, we
0 0
compute g = gcd(f1 , . . . , fn−1 ) and the new polynomials f1 = f1 /g, . . . , fn−1 = fn−1 /g.
0 0
Then using gcd(f1 , . . . , fn−1 ) = 1 and the inductive hypothesis we can obtain an infinite
0 0 0 0 0
sequence {hi }∞
i=1 ⊂ (f1 , . . . , fn−1 ) with gcd(hi , hj ) = 1 for i 6= j.
0 0 0 0
For each hi we have that gcd(fn , fn + ghi ) = gcd(fn , fn + ghi − fn ) = gcd(fn , ghi ). From
gcd(f1 , . . . , fn ) = 1 we conclude that gcd(fn , g) = 1 and for some j ∈ N we should have
0 0
gcd(fn , ghj ) = 1, because all the hi ’s have different prime factors but fn can have only a
finite amount of prime factors.
Now we use another induction argument where the inductive step is similar to the previous
paragraph. Suppose we have computed a sequence of polynomials h1 , . . . , hk and a
polynomial gk , with the properties gcd(hi , hj ) = 1 for 1 ≤ i < j ≤ k and gcd(hi , gk ) = 1
for 1 ≤ i ≤ k. Again, for each hi 0 we have
0 0 0 0
gcd(h1 . . . hk , h1 . . . hk + gk hi ) = gcd(h1 . . . hk , gk hi ) = gcd(h1 . . . hk + gk hi , gk hi ), (3.7)
0
and should exist some j ∈ N with gcd(h1 h2 . . . hk , gk hj ) = 1. Thus we define the next
elements in the inductive step as
0 0
hk+1 = h1 h2 . . . hk + gk hj and gk+1 = gk hj .
From the equality 3.7 we have that gcd(h1 h2 . . . hk , hk+1 ) = gcd(h1 h2 . . . hk , gk+1 ) =
gcd(hk+1 , gk+1 ) = 1, which implies gcd(hi , hj ) = 1 for 1 ≤ i < j ≤ k +1 and gcd(hi , gk+1 ) =
1 for 1 ≤ i ≤ k + 1. Starting with h1 = fn , g1 = g and following this iteratively process we
can construct the required sequence {hi }∞ i=1 ⊂ (f1 , . . . , fn ) with gcd(hi , hj ) = 1 for i 6= j.
3.4. Bounds for the regularity and the Betti numbers 69
Step 3: We choose the two relatively prime elements p and q from the previous step. Then
p is regular on R[s, t, u], and q is regular on R[s, t, u]/p because gcd(p, q) = 1. Therefore
{p, q} is a regular sequence and depth(I) ˆ ≥ 2.
In [20] is proven that for a general homogeneous ideal J ⊂ K[x1 , . . . , xn ] (in any character-
istic) generated by homogeneous elements with degree ≤ d, the regularity of J is bounded
by
n−2
reg(J) ≤ (2d)2 . (3.8)
Using similar ideas, in [16] is proven a bound for the regularity of a quotient ring, with
if dim(K[x1 , . . . , xn ]/J) ≤ 1.
Proof. If we prove dim(S/I)ˆ ≤ 1, then we could cite Theorem 19.4 from [16] and obtain
ˆ = reg(S/I)
the linear bound for the regularity reg(I) ˆ + 1 ≤ 3d − 2, that in some sense is
the best result we could expect.
From the previous Proposition 3.26 we obtain a lower bound for the codimension (or
height) of Iˆ by means of codim(I)
ˆ ≥ depth(I)
ˆ ≥ 2 (see Proposition 1.59).
ˆ + dim(S/I)
codim(I) ˆ ≤ dim(S) = 3,
ˆ ≤ 1.
that implies our desired inequality dim(S/I)
Remark 3.28. From the fact that all the elements in the differential matrices are homoge-
ˆ = 0 for p < i + d (see Proposition
neous of degree at least one, we can conclude that βi,p (I)
12.3, [17]).
With the computation of the regularity will be enough for our problem in the case
ˆ = 1. Therefore for the rest of this section we will assume that pdS (I)
pdS (I) ˆ = 2 and we
ˆ R)) in this case.
shall try to find a bound for the Betti number β2 = dimR (T or2S (I,
Our main tool for this computation will be the Koszul complex. We know that m = {s, t, u}
is a regular sequence in S = R[s, t, u], then the Koszul complex K(m) has zero homology
3.4. Bounds for the regularity and the Betti numbers 70
Hi K(m) = 0 for i ≥ 1 (see Theorem 1.32). So we can regard the Koszul complex as a
graded free resolution of R = S/(s, t, u), that is
^3 d ^2 d ^1 d ^0
K(m) : 0→ S(−3)3 −
→3
S(−2)3 −
→2
S(−1)3 −
→1
S 3 → R → 0.
then we find the following relationship between the Tor functor and the Koszul homol-
ogy
ˆ ∼
Hi K(m; I) ˆ ∼
= T oriS (R, I) ˆ R).
= T oriS (I,
At this moment we have an explicit formula for computing the Betti number
ˆ
β2 = dimR (H2 K(m; I)),
and yet we don’t know how to compute a bound for the R-dimension of the S-module
ˆ Instead we are going to use the fact that each map id ⊗S di has degree 0, and
H2 K(m; I).
so we can form the complex
^3 (id⊗S d3 )p−3 ^2 (id⊗S d2 )p−2
0 → Iˆp−3 ⊗S S 3 −−−−−−−→ Iˆp−2 ⊗S S 3 −−−−−−−→
^1 (id⊗S d1 )p−1 ^0
Iˆp−1 ⊗S S 3 −−−−−−−→ Iˆp ⊗S S 3 → 0,
For the moment we assume that we know the Hilbert function of the ideal Iˆ and our bound
will be related with it. We make this “fair” assumption because for completely general
ideals we cannot give an interesting result and later we will try to give some bounds for
ˆ
the Hilbert function of I.
ˆ p = Ker(id ⊗S d2 )p = Ker((id ⊗S d2 )p ) .
H2 K(m; I)
Im(id ⊗S d3 )p Im((id ⊗S d3 )p−1 )
Then using the fact that Ker((id ⊗S d2 )p ) and Im((id ⊗S d3 )p−1 ) are R-vector spaces, we
can compute the graded Betti number β2,p with
ˆ = 2 we have H3 K(m; I)
Since pdS (I) ˆ ∼ ˆ R) = 0 and so Ker(id ⊗S d3 ) = 0. From
= T or3S (I,
this we conclude that (id ⊗S d3 )p−1 is an injective map and dimR (Im((id ⊗S d3 )p−1 )) =
dimR (Iˆp−1 ⊗S 3 S 3 ) = HIˆ(p − 1).
V
th12 = −uh13 ,
sh12 = uh23 ,
sh13 = −th23 .
Therefore one of the terms can completely define the other two, e.g., by knowing h12 we
can determine h13 and h23 . This simple fact implies the inequality
Theorem 3.30. β2 ≤ HIˆ(3d).
3.5. The case with projective dimension one 72
Here we give a simple combinatorial bound for the Hilbert function of the ideal Iˆ =
(b1 , b2 , b3 , b4 ) ⊆ R[s, t, u] generated by four elements. For any p ≥ d we have that the vector
space Iˆp is generated by elements of the form mbi (1 ≤ i ≤ 4) where m is a monomial
of degree p − d. We know that the number of monomials of degree p − d in R[s, t, u] is
p−d+2
2
. Therefore we have have that the Hilbert function of Iˆ is bounded by
!
p−d+2
HIˆ(p) ≤ 4 ,
2
Lemma 3.31. The grading of the previous resolution 3.9 can be bounded just as if it were
a minimal resolution, i.e, α = max(α1 , α2 , α3 ) ≤ 3d − 1.
3.5. The case with projective dimension one 73
ˆ R)p = 0,
Ker((d1 ⊗S R)p ) = T or1S (I,
hence we may conclude that the map (d1 ⊗S R)p is an injective map, and (S(−α1 ) ⊕
S(−α2 ) ⊕ S(−α3 ))p = 0 because (S(−d)4 )q = 0 for q > d. Therefore we have the inequality
α ≤ 3d − 1, which is the same result we can obtain for the minimal free resolution (i.e.
(3d − 2) + 1).
From the inequalities α1 , α2 , α3 ≤ 3d − 1 and the fact that in the previous resolution the
graded free module S(−d)4 has all its grading exactly in d, we have that the degree of the
polynomials p̂i , qˆi , rˆi is bounded by 2d − 1 = 3d − 1 − d. Therefore we have obtained the
following result.
Remark 3.32. Here we have found a family of surfaces for which a minimal µ-basis has
an optimal bound for the degree of its polynomials, i.e., linear with bound ≤ 2d − 1. One
could say that we already had this type of bound for the case in which a1 , a2 , a3 , a3 were
n−2
homogeneous polynomials since the beginning, because we could quote the bound (2d)2
ˆ = 1 is much broader than the
equal to 2d for n = 2. But the family of surfaces with pdS (I)
one defined by homogeneous polynomials. For instance the surface
P (s, t) = (st, st + t, t + 1, 2t − 1)
is not defined by homogeneous polynomials, but the homogeneous ideal Iˆ = (st, st + tu, tu +
ˆ = 1 and we can find the following minimal free resolution
u2 , 2tu − u2 ) has pdS (I)
t 0
−u s
2
0 −u u tu st
0 → S(−3)2 −−−−−−−−−−→ S(−2)3 −−−−−−−−−−−→ Iˆ → 0.
3.6. The case with projective dimension two 74
ˆ = 2 is of the
As we have already computed, we remember that the case with pdS (I)
form
dˆ2 dˆ1 [b1 ,b2 ,b3 ,b4 ]
0 → Sa − → S 4 −−−−−−→ Iˆ → 0,
→ S a+3 − (3.10)
dˆ dˆ
0 → Sa −
→2
S a+3 −
→1
Syz(b1 , b2 , b3 , b4 ) → 0
Making a proof with the same spirit as in Theorem 3.20, we have that
ˆ R)) = β2 .
a = dimR (T or2S (I,
Using a similar argument as in Lemma 3.31 we have that the grading in the free module
S β2 +3 is bounded by 3d − 1, and using the fact that Im(dˆ2 ) ⊆ mS a+3 then the grading in
the free module S β2 is the same as the one for the minimal free resolution. Therefore, the
polynomials in the entries of the matrix d1 have degree bounded by 3d − 1 − d = 2d − 1, and
the polynomials in the entries of the matrix d2 are bounded by (3d − 2) + 2 − d = 2d.
Now finally, we can return to our original problem in R = R[s, t]. We apply the tensor
product with ⊗S S/(u − 1) to obtain the exact sequence
where d1 = dˆ1 ⊗S S/(u − 1) and d2 = dˆ2 ⊗S S/(u − 1) are matrices with entries in R
bounded in degrees by 2d − 1 and 2d, respectively.
3.6. The case with projective dimension two 75
For the rest of this document we shall work with the exact sequence
d d
0 → R β2 −
→2
Rβ2 +3 −
→1
Syz(a1 , a2 , a3 , a4 ) (⊂ R4 ) → 0, (3.11)
which is a split exact sequence because we know that Syz(a1 , a2 , a3 , a4 ) is a free module
of rank 3.
Here we are in a position where the only thing that we “know” is the split exact sequence
3.11 (for the only thing we have bounds is for the matrices d1 and d2 ) and we want to
find a bound for a basis of Syz(a1 , a2 , a3 , a4 ). For this task we will follow the paper [21],
where there is an algorithmic proof based on the original Suslin’s proof for the Serre’s
conjecture.
The “calculus of unimodular rows” presented in the previous chapter is an elegant theory
that deals with the problem
We define the degree of a matrix M = (aij ) ∈ Rr×s as the maximum degree of the
polynomial entries of M , i.e., deg(M ) = max(deg(aij )). For an “effective” solution of
completing an unimodular matrix we are going to use the following result from [21].
Theorem 3.33. Let R = R[s, t] and assume that F ∈ Rr×s (r < s) is unimodular. Then
there exists a square matrix M ∈ Rs×s such that
(i) M is unimodular,
3.6. The case with projective dimension two 76
(iii) deg(M ) ≤ 2D(1 + 2D)(1 + D4 )(1 + D)4 , where D = r(1 + deg(F )).
This previous result is given for completing rows (i.e., r < s), but we want to complete
columns (i.e., r > s). By simply taking the transpose of (ii) in the previous theorem we
get the following corollary.
Corollary 3.34. Let R = R[s, t] and assume that F ∈ Rr×s (r > s) is unimodular. Then
there exists a square matrix M ∈ Rr×r such that
(i) M is unimodular,
Is
(ii) M F = ∈ Rr×s ,
0
(iii) deg(M ) ≤ 2D(1 + 2D)(1 + D4 )(1 + D)4 , where D = s(1 + deg(F )).
• r = β2 + 3 and s = β2 ,
F G
0 → Rs −
→ Rr −
→ Syz(a1 , a2 , a3 , a4 ) (⊂ R4 ) → 0. (3.12)
Since this sequence splits, there exists a matrix H ∈ Rr×s with HF = Is and so the matrix
F is unimodular. We can apply the previous Corollary 3.34 and from now on M ∈ Rr×r
will denote a matrix that satisfies (i), (ii), (iii) from 3.34.
Rr = Im(F ) ⊕ V. (3.13)
3.6. The case with projective dimension two 77
because Ker(G) = Im(F ). Finally we are ready to find a basis for Syz(a1 , a2 , a3 , a4 ).
Theorem 3.35. There exist a basis for Syz(a1 , a2 , a3 , a4 ) formed by the three vectors
p, q, r ∈ R4 , with the polynomial entries bounded in degree by
p = GN es+1 ,
q = GN es+2 ,
r = GN es+3 .
In contrast to the proof of Theorem 2.30, here we don’t need to fully identify the module V ∼
2
=
Syz(a1 , a2 , a3 , a4 ) as a direct summand inside Rr .
3.6. The case with projective dimension two 78
Finally, using the previous bounds deg(G) ≤ 2d − 1 and 3.14, we obtain the result
2(2d − 1)(β2 + 2)β2 (1 + 2d)(1 + 2β2 (1 + 2d))(1 + (β2 (1 + 2d))4 )(1 + β2 (1 + 2d))4 (3.15)
and substituting the previous result β2 ≤ 4 2d+22
∈ O(d2 ) we can find a bound for the
degree of a minimal µ-basis that completely depends on d. We can compute that the result
in 3.15 is in the order of
O(d33 ).
In the end of this section we put some of our thoughts about the results obtained throughout
this chapter.
• We have given an explicit bound for the degree of a minimal µ-basis for any rational
surface as in 3.1. The result of the bound is in the order of O(d33 ).
• This big and not so attractive bound comes mostly from the solution of the Quillen-
Suslin Theorem. Therefore could be interesting to find better bounds for an “effective”
solution of the Quillen-Suslin Theorem.
ˆ ≤ 3d − 2, so it is a
• The result that we found for the regularity is linear with reg(I)
pretty good bound and optimal from a complexity point of view.
• The formula 3.15 is given also in terms of β2 in case that we already know who is β2 .
The bound for β2 is in the order of O(d2 ), but having a small knowledge of the ideal
Iˆ and its Hilbert function this result could be improved drastically.
• Perhaps, for “aesthetic purposes” it could be better to follow the “philosophy” of [21]
and just say: “we have proven that the degree of a minimal µ-basis for any rational
surface is bounded by a single exponential bound dO(1) ”.3
3
Clearly we have that O(2) = O(1)
3.7. Appendix 79
3.7 Appendix
The main result we shall follow from [21] is the following theorem.
Theorem 3.36. (Theorem 3.1, [21]) Let R = K[x1 , . . . , xn ] and assume that F ∈
Rr×s (r < s) is unimodular. Then there exists a square matrix M ∈ Rs×s such that
(i) M is unimodular,
With the pioneering work of Mayr and Meyer [12], we know that the general problem
of solving linear equation systems in polynomial rings involves necessarily a doubly
exponential bound for the degree of the polynomials appearing in the solution. Also from
[12], we know that the best possible bound for the regularity of an ideal in the general case
is double exponential as in 3.8. Therefore the previous result has important theoretical
consequences because it proofs that a solution for the Serre’s conjecture exists with a
single exponential bound for the degree.
In our case n = 2 and we will have to make some small “adjustments” to find an actual
constant instead of the complexity class O(2). We will follow exactly the same proof as
in [21], and in certain steps we will substitute phrases like n + 3n ∈ O(n) by the exact
computation n + 3n = 4n.
Proposition 3.37. (Proposition 4.1, [21]) Assume that F ∈ Rr×s (R = K[x1 , . . . , xn ], r <
s) is unimodular. Then there exists a square matrix M ∈ Rs×s such that:
(i) M is unimodular,
(ii) F M = [fij (x1 , . . . , xn−1 , 0)] (i.e., F M is equal to the r × s matrix obtained by
specializing the indeterminate xn to zero in the matrix F ),
(iii) deg(M ) ≤ D(1 + 2D)(1 + D2n )(1 + D)2n , with D = r(1 + deg(F )).
Proof. We denote F (t) as the matrix F (t) = [fij (x1 , . . . , xn−1 , t)] and inside this proposition
d = 1 + deg(F ).
Claim 1. (Procedure 4.6, Step 1 and Step 2, [21]) There exists elements c1 , . . . , cN ∈
K[x1 , . . . , xn−1 ] with N ≤ (1 + rd)2n such that xn ∈ (c1 , . . . , cN ). Also we can find elements
3.7. Appendix 80
x n = a1 c 1 + . . . + aN c N
and with
max {deg(ak ck )} ≤ 1 + (rd)2n .
1≤k≤N
k
X
bk = ah ch ,
h=1
Proof. (of Theorem 3.36) For a matrix F = [fij (x1 , . . . , xn )] the substitution of a variable
xi for 0 does not increase the degree of the matrix, and keeps the unimodularity because
the condition of being unimodular is equivalent to the fact that the (r × r)-minors have
no common zeros. Therefore, applying the previous proposition n times we can find a
0
unimodular square Rs×s matrix M with
0
F M = [fij (0, 0, . . . , 0)]
0
deg(M ) ≤ nD(1 + 2D)(1 + D2n )(1 + D)2n .
Since the matrix [fij (0, 0, . . . , 0)] has rank r, we can find an elementary matrix E ∈ Ks×s
0
with [fij (0, 0, . . . , 0)]E = [Ir , 0]. Making M = M E, we get F M = [Ir , 0] and the same
3.7. Appendix 81
inequality
deg(M ) ≤ nD(1 + 2D)(1 + D2n )(1 + D)2n .
Bibliography
[1] Falai Chen, David Cox, and Yang Liu, The µ-basis and implicitization of a rational parametric surface,
Journal of Symbolic Computation 39 (2005), no. 6, 689–706.
[2] David Eisenbud, Daniel R. Grayson, Mike Stillman, Bernd Sturmfels (Eds.), Computations in algebraic
geometry with macaulay 2, 1st ed., Vol. 8, Springer, 2002.
[3] David Eisenbud, Commutative algebra with a view towards algebraic geometry, Graduate Texts in
Mathematics, 150, Springer-Verlag, 1995.
[4] Daniel R. Grayson and Michael E. Stillman, Macaulay2, a software system for research in algebraic
geometry. Available at http://www.math.uiuc.edu/Macaulay2/.
[5] E. Graham Evans, Phillip Griffith, Syzygies, London Mathematical Society Lecture Note Series, 106,
Cambridge University Press, 1985.
[6] Winfried Bruns, H. Jürgen Herzog, Cohen-macaulay rings, 2nd ed., Cambridge Studies in Advanced
Mathematics, Cambridge University Press, 1998.
[7] David Hilbert, Über die theorie der algebraischen formen, Math. Ann. 36 (1890), 473–534.
[8] T. Y. Lam, Serre’s problem on projective modules, Springer Monographs in Mathematics, Springer
Verlag, 2006.
[9] Serge Lang, Algebra, 3rd ed., Graduate Texts in Mathematics 211, Springer-Verlag New York, 2002.
[11] Hideyuki Matsumura, Commutative ring theory, 1st ed., Cambridge Studies in Advanced Mathematics
volume 8, Cambridge University Press, 1989.
[12] Ernst W. Mayr, Albert R. Meyer, The complexity of the word problems for commutative semigroups
and polynomial ideals, Advances in mathematics 46 (1982), no. 3, 305–329.
[15] David A. Cox, John Little, Donal O’Shea, Using algebraic geometry, 2nd ed., Graduate texts in
mathematics, 185, Springer Verlag, 2005.
[16] Irena Peeva, Syzygies and hilbert functions, CRC Press, 2007.
82
Bibliography 83
[17] , Graded syzygies, Vol. 14, Springer Science & Business Media, 2010.
[18] D. C. Youla, P. F. Pickel, The quillen-suslin theorem and the structure of n-dimensional elementary
polynomial matrices, IEEE transactions on circuits and systems 31 (1984), no. 6, 513–518.
[20] Giulio Caviglia, Enrico Sbarra, Characteristic-free bounds for the castelnuovo-mumford regularity,
Compos. Math 141 (2005), no. 6, 1365–1373.
[21] Leandro Caniglia, Guillermo Cortiñas, Silvia Danón, Joos Heintz, Teresa Krick, Pablo Solernó,
Algorithmic aspects of suslin’s proof of serre’s conjecture, computational complexity 3 (1993), no. 1,
31–55.