The Quillen-Suslin Theorem

Swayam Chube

December 1, 2024

§1 F INITE F REE R ESOLUTIONS

D EFINITION 1.1. A module E is said to be stably free if there exists a finite free module F of
such that E ⊕ F is finite free.
E is said to have a finite free resolution if there is a resolution

0 → En → · · · → E0 → E → 0

such that each Ei is a finite free module.

P ROPOSITION 1.2. Let M be projective. Then M is stably free if and only if M admits a
finite free resolution.

Proof. Suppose first that M is stably free. Then, there is a finite free F such that E = M ⊕ F
is finite free. Thus, 0 → F → E → M → 0 is a finite free resolution of M.
On the other hand, suppose M admits a finite free resolution,

0 → En → · · · → E0 → M → 0,

where n is the smallest such. We shall induct on this n. The base case with n = 0 is trivial
since M is free. Let M1 = ker ( E0 → M ). Then, M1 has a finite free resolution

0 → En → · · · → E1 → M1 → 0

of length n − 1 whence the induction hypothesis applies and there is a finite free F such
that M1 ⊕ F is finite free. Using the fact that M is projective, we have

M ⊕ ( M1 ⊕ F ) ∼
= ( M ⊕ M1 ) ⊕ F ∼
= E ⊕ F,

and hence, M is stably free. ■

D EFINITION 1.3. A resolution

0 → En → · · · → E0 → M → 0

is said to be stably free if each Ei is stably free for 0 ⩽ i ⩽ n.

1
P ROPOSITION 1.4. M has a finite free resolution of length n ⩾ 1 if and only if it has a
stably free resolution of length n.
Proof. Obviously every finite free resolution is stably free. Suppose now that M has a
stably free resolution of length n:

0 → En → · · · → E0 → M → 0

Choose any index 0 ⩽ i < i + 1 ⩽ n. There are finite free modules Fi , Fi+1 corresponding
to Ei , Ei+1 respectively. Set F = Fi ⊕ Fi+1 . Then, we have a stably free resolution:

0 → En → · · · → Ei+1 ⊕ F → Ei ⊕ F → Ei−1 → · · · → E0 → M → 0,

with the modified map being ( Ei+1 → Ei , id F ).

Applying the above construction successively to pairs ( E0 , E1 ), ( E1 , E2 ) and so on, we
end up with a finite free resolution of M. ■
D EFINITION 1.5. M1 and M2 are said to be stably isomorphic if there exist finite free modules
F1 and F2 such that M1 ⊕ F1 ∼
= M2 ⊕ F2 .
L EMMA 1.6 (S CHANUEL ). Let 0 → K → P → M → 0 and 0 → K ′ → P′ → M → 0 be
exact sequences where P and P′ are projective. Then K ⊕ P′ ∼
= K ′ ⊕ P.
Proof. Treat K and K ′ as submodules of P and P′ respectively. The projectivity of P and P′
gives a commutative diagram

0 / K / P / M / 0
u w id
 
0 / K′ / P′ / M / 0
f g
where u is the restriction of w to K. Consider the sequence 0 → K − → P ⊕ K′ −
→ P′ → 0
where
f ( x ) = ( x, u( x )) and g(y, z) = w(y) − z.
We contend that this is exact.
• Exactness at K is trivial.
• It is easy to see that g ◦ f = 0. Suppose (y, z) ∈ ker g, that is, w(y) = z. Since z ∈ K ′ ,
we must have that y ∈ K whence u(y) = z, which proves exactness at P ⊕ K ′ .
• Choose some x ′ ∈ P′ . We can choose an x ∈ P such that the images of x and x ′ in M
are the same. Thus, x ′ − w( x ) ∈ K ′ whence exactness at P′ follows.
Finally, since P′ is projective, the sequence splits, giving us the desired conclusion. ■
L EMMA 1.7. Suppose M1 and M2 are stably isomorphic. Let

0 → N1 → E1 → M1 → 0 and 0 → N2 → E2 → M2 → 0

be exact sequences where E1 and E2 are stably free. Then N1 is stably isomorphic to N2 .

2
Proof. There are finite free modules F1 , F2 such that M1 ⊕ F1 ∼
= M2 ⊕ F2 . We may modify
the above short exact sequences to obtain

0 → N1 → E1 ⊕ F1 → M1 ⊕ F1 → 0 and 0 → N2 → E2 ⊕ F2 → M2 ⊕ F2 → 0.

Invoking Lemma 1.6,

N1 ⊕ E2 ⊕ F2 ∼
= N2 ⊕ E1 ⊕ F1 .
Since both E1 , E2 are stably free, there is a finite free module F such that both E1 ⊕ F and
E2 ⊕ F are finite free. Thus,

N1 ⊕ ( E2 ⊕ F ⊕ F2 ) ∼
= N2 ⊕ ( E1 ⊕ F ⊕ F1 )
and the conclusion follows. ■
D EFINITION 1.8. The minimal length of a stably free resolution of a module is called its
stably free dimension.
T HEOREM 1.9. Let M be a module admitting a stably free resolution

0 → En → · · · → E0 → M → 0

of length n. Let
Fm → · · · → F0 → M
be an exact sequence with Fi stably free for 0 ⩽ i ⩽ m.
(a) If m < n − 1, then there exists a stably free module Fm+1 such that the above sequence
can be continued exactly to

Fm+1 → Fm → · · · → F0 → M

(b) If m = n − 1 and Fn = ker ( Fn−1 → Fn−2 ). Then Fn is stably free.

Proof. For 0 ⩽ i ⩽ n, define Ki = ker( Ei → Ei−1 ) with the convention that E−1 = M.
Similarly, define Ki′ = ker( Fi → Fi−1 ). Using Lemma 1.7, repeatedly along with the exact
sequences

0 → Ki → Ei → Ei−1 → 0 and 0 → Ki′ → Fi → Ki′−1 → 0,

we conclude that Km and Km ′ are stably isomorphic. Thus, there exist finite free modules

F, F ′ such that Km ⊕ F ∼ ′ ⊕ F′ .
= Km

(a) m < n − 1 : We have

Em+1 ⊕ F ↠ Km ⊕ F ∼ ′
= Km ⊕ F ′ → Km
′
→ 0.

Set Fm+1 = Em+1 ⊕ F which is easily seen to be stably free.

(b) m = n − 1 : We can choose Km = En . Then, En ⊕ F is stably free, whence so is

′ ⊕ F ′ , in particular, so is K ′ . This completes the proof.
Km ■
m

3
C OROLLARY. If 0 → M1 → E → M → 0 is exact, M has stably free dimension ⩽ n, and E
is stably free, then M1 has stably free dimension ⩽ n − 1.

Proof. Let 0 → En → · · · → E0 → M → 0. We have an incomplete stably free resolution

E → M → 0. We may now invoke Theorem 1.9 with F0 = E to obtain a resolution

0 → Fn → · · · → F0 = E → M → 0.

But note that M1 = ker( E → M) and hence, there is a stably free resolution

0 → Fn → · · · → F1 → M1 → 0,

and the conclusion follows. ■

R EMARK 1.10. Let 0 → M′ → M → M′′ → 0 be a short exact sequence of finitely
generated modules. Then, there are surjections φ : Rm → M′ and ψ : Rn → M′′ , where
R is the base ring. There is also the canonical injection ι : Rm → Rm ⊕ Rn and the
canonical surjection π : Rm ⊕ Rn → Rn . Define the map Φ : Rm ⊕ Rn → M given by
Φ( x, y) = f ( φ( x )) + ψ
e(y), where ψe : Rn → M is a lift of the map ψ : Rn → M′′ .
We contend that Φ is surjective. Indeed, let m ∈ M then there is a y ∈ Rn such that
ψ(y) = g(m). It is easy to see that m − ψ e(y) ∈ ker g = im f and hence, there is an x ∈ Rm
such that f ◦ φ( x ) = m − ψ e(y). It follows that Φ( x, y) = m. Finally, the Snake Lemma
gives a nice exact diagram.

0 0 0

  
0 / M1′ / M1 / M1′′ /0

  
0 / Rm
ι /
Rm ⊕ Rn
π / Rn /0
ψ
e
φ Φ ψ
  y 
0 / M′ / M / M′′ /0
f g

  
0 0 0

L EMMA 1.11. Let M′′ be finitely presented and M finitely generated. If M′ is the kernel of
a surjection M ↠ M′′ , then M′ is finitely generated.

Proof. We first prove this when M is finite free. Since M′′ is finitely presented, there is an
exact sequence 0 → K → F → M′′ → 0, where F is a finite free module and K is finitely
generated. Due to Lemma 1.6, M′ ⊕ F ∼ = K ⊕ M, whence M′ is finitely generated.
Now, suppose M is just finitely generated. It can be written as the quotient of a free

4
module F ↠ M. This gives a commutative diagram

0 / K / F

  !
0 / M′ / M / M′′ / 0

 
0 0
whence M′ is finitely generated. ■
T HEOREM 1.12. Let 0 → M′ → M → M′′ → 0 be an exact sequence. If any two of these
modules have a finite free resolution, then so does the third.
Proof. There are three possible cases. We shall tacitly use Proposition 1.4 throughout this
proof.
M′ and M: We induct on the stable free dimension of M. For the base case with the stable
free dimension 0, M is stable free and the conclusion follows since M′ too has a
finite free resolution. Next, suppose the stable free dimension of M is n ⩾ 1. Due
to Remark 1.10 and Corollary 1, the stable free dimension of M1 is at most n − 1
whence the induction hypothesis applies and M1′′ has a finite free resolution and the
conclusion follows.
M′ and M′′ : Induct on the maximum of the stable free dimension of M′ and M′′ . The base
case occurs when both M′ and M′′ have stably free dimension 0, that is, both are
stably free, consequently, projective. It follows that M ∼
= M′ ⊕ M′′ is stably free.
Next, for the induction step, using Remark 1.10 and Corollary 1 we see that the
maximum of the stably free dimension of M1′ and M1′′ is at most n − 1, whereby the
induction hypothesis applies and the conclusion follows.
M and M′′ : We induct on the stably free dimension of M′′ . In the base case, M′′ is stably
free, in particular, projective, and hence, M′ ⊕ M′′ ∼
= M, whence M′ is also stably
free.
As for the inductive step, again use Remark 1.10 and Corollary 1 to conclude. ■

§2 S ERRE ’ S T HEOREM
T HEOREM 2.1. Let R be a Noetherian ring. If every finite R-module has a finite free
resolution, then every finite R[ X ]-module has a finite free resolution.
Proof. Let M be a finite R[ X ]-module. There is a filtration

M = M0 ⊋ M1 ⊋ · · · ⊋ Mn = 0,

where Mi /Mi+1 ∼= R[ X ]/Pi for some prime Pi . In light of Theorem 1.12, it suffices to
prove the theorem in the case M = R[ X ]/P for some prime P.

5
 Suppose the theorem is false. Let Σ be the collection of all primes P such that R[ X ]/P
does not admit a finite free resolution. Choose P in Σ that maximizes p = P ∩ R.
Let R0 = R/p, K0 its quotient field, P0 = P/pR[ X ] and M = R[ X ]/P. We may view
M as an R0 [ X ]-module, equal to R0 [ X ]/P0 . Let f 1 , . . . , f n be a finite set of generators for
P0 , and let f be a polynomial of minimal degree in P0 .
We c an write f i = qi f + ri for 1 ⩽ i ⩽ n with qi , ri ∈ K0 [ X ] and deg ri < deg f or ri = 0.
Let d0 be a common denominator for all coefficients of all qi , ri . Then, d0 ̸= 0 and

d0 f i = qi′ f + ri′ ,

where qi′ = d0 qi , ri′ = d0 ri ∈ R0 [ X ] Since deg f is minimal in P0 , it follows that ri′ = 0 for
all i, so d0 P0 ⊆ ( f ).
Let N0 = P0 /( f ), so that N0 is a module over R0 [ X ], and hence, N0 can also be viewed
as an R[ X ]-module. When so viewed, we denote N0 by N. Let d ∈ R be any element
reducing to d0 mod p. Since d0 ̸= 0, d ∈ / p.
The module N0 has a filtration such that each successive quotient is isomorphic to
R0 [ X ]/Q0 where Q0 is an associated prime of N0 . Let Q be the pullback of Q0 to R[ X ]. It
is easy to argue that these prime ideals Q are precisely the associated primes of N in R[ X ].
Since d0 kills N0 , d must kill N and hence, d lies in every associated prime of N.
Note that each associated prime Q of N contains P and due to the preceding paragraph,
Q ∩ R ⊋ P ∩ R. Due to the maximality involved in the choice of P, every successive
quotient in the filtration of N has a finite free resolution, whence N has a finite free
resolution.
By assumption, p has a finite free resolution as an R-module, say

0 → En → · · · → E0 → p → 0.

Then
0 → En [ X ] → · · · → E0 [ X ] → p[ X ] → 0
is a finite free resolution of p[ X ] ⊆ R[ X ] as an R[ X ]-module. From the exact sequence

0 → p[ X ] → R[ X ] → R0 [ X ] → 0,

it follows that R0 [ X ] has a finite free resolution as an R[ X ]-module.

There is a surjective R[ X ]-linear map µ f : R0 [ X ] → ( f ), which is just multiplication by
f . The kernel of this map is trivial since R0 [ X ] is an integral domain. It follows that ( f ) too
has a finite free resolution as an R[ X ]-module.
From the exact sequence of R[ X ]-modules

0 → ( f ) → P0 → N → 0,

we conclude that P0 has a finite free resolution as an R[ X ]-module. Next, from another
exact sequence of R[ X ]-modules

0 → pR[ X ] → P → P0 → 0,

6
it follows that P has a finite free resolution as an R[ X ]-module. Finally from

0 → P → R[ X ] → R[ X ]/P → 0,

we conclude that R[ X ]/P admits a finite free resolution as an R[ X ]-module, a contradiction.

This completes the proof. ■
T HEOREM 2.2 (S ERRE ). Let k be a field. Every finite projective module over k [ X1 , . . . , Xn ]
admits a finite free resolution. Equivalently, is stably free.

§3 U NIMODULAR P OLYNOMIAL V ECTORS

D EFINITION 3.1. Let A be a commutative ring. An n-tuple ( f 1 , . . . , f n ) ∈ An is said to be
unimodular if they generate the unit ideal in A. A unimodular vector is said to have the
unimodular extension property if there exists a matrix in GLn ( A) with ( f 1 , . . . , f n )⊤ as the
first column.
R EMARK 3.2. Note that a unimodular column vector ( f 1 , . . . , f n )⊤ has the unimodular
extension property if and only if some column vector obtained after a series of row and
column operations has that property.
T HEOREM 3.3 (H ORROCKS ). Let (o, m, k ) be a local ring and A = o[ X ]. Let f be a
unimodular column vector in A(n) such that some component in f has leading coefficient
1. Then f has the unimodular extension property.
Proof. If n = 1, then there is nothing to prove. Next, if n = 2, then ( f 1 , f 2 ) = (1) and hence,
there are g1 , g2 ∈ A such that f 1 g1 + f 2 g2 = 1, whence
 
f 1 − g2
det = 1.
f 2 g1

Now, assume n ⩾ 3 and induct on the smallest degree d of a component of f with

leading coefficient 1. The base case with d = 0 is trivial. Suppose now that d ⩾ 1.
Using row operations, we may suppose that deg f i < d for i ̸= 1. Since there is a linear
combination ∑in=1 gi f i = 1, not all coefficients of f 2 , . . . , f n can lie in m, for if they did, then
g1 f 1 ≡ 1 (mod m)[ X ], which is absurd, since f 1 is not a unit modulo m[ X ].
Without loss of generality, suppose that some coefficient of f 2 does not lie in m. Write

f 1 ( X ) = X d + a d −1 X d −1 + · · · + a 0 a i ∈ o
f 2 ( X ) = bs X s + · · · + b0 bi ∈ o, s ⩽ d − 1

such that some bi is a unit. Lt a be the ideal generated by all leading coefficients of
polynomials of the form g1 f 1 + g2 f 2 of degree ⩽ d − 1. We claim that a contains all the bi .
This can be seen inductively. First, bs lies in a because of X d−s f 2 ( X ). Next, bs−1 is realised
as X d−s f 2 ( X ) − bs f 1 ( X ) has leading coefficient bs−1 − bs ad−1 . But since a already contains
bs , it must also contain bs−1 . Continue this way. Recall that one of the bi ’s is a unit and
hence, a is the unit ideal.

7
 Thus, there is a linear combination h = g1 f 1 + g2 f 2 having degree ⩽ d − 1 and leading
coefficient 1. If deg f 3 < deg h, then h + f 3 has leading coefficient 1 and degree ⩽ d − 1.
Now suppose deg f 3 = deg h. If the leading coefficient of f 3 is a unit, then multiply by its
inverse to make the leading coefficient 1. If, on the other hand, it is not a unit, then the
leading coefficient of h + f 3 is a unit and hence, can be made 1 after multiplying by its
inverse. Now, the induction hypothesis applies, thereby completing the proof. ■

D EFINITION 3.4. Let A be a commutative ring. For two column vectors f , g ∈ A(n) , we
write f ∼ g to mean that there exists M ∈ GLn ( A) such that f = Mg, and we say that f is
equivalent to g over A.

P ROPOSITION 3.5. Let (o, m, k ) be a local ring. Let f be a unimodular vector in o[ X ](n)
such that some component has leading coefficient 1. Then f ∼ f (0) over o[ X ].

Proof. Note that f (0) ∈ o(n) has at least one component which is a unit, for if not, then
the constant term of any linear combination would always lie in m. Hence, it follows that
f (0) ∼ e1 . On the the other hand, due to Theorem 3.3, f ∼ e1 , thereby completing the
proof. ■
L EMMA 3.6. Let R be an integral domain, and S ⊆ R a multiplicatively closed subset
containing 1. Let X and Y be independent variables. If f ( X ) ∼ f (0) over S−1 R[ X ], then
there is a c ∈ S such that f ( X + cY ) ∼ f ( X ) over R[ X, Y ].

Proof. Let M ∈ GLn (S−1 R[ X ]) be such that f ( X ) = M( X ) f (0). That is, M( X )−1 f ( X ) =
f (0). The right hand side is independent of X and hence, M( X + Y )−1 f ( X + Y ) = f (0)
when viewed over S−1 R[ X, Y ]. Set G ( X, Y ) = M ( X ) M( X + Y )−1 ∈ S−1 R[ X, Y ], then
G ( X, Y ) f ( X + Y ) = f ( X ).
By construction, we have G ( X, 0) = I, the identity matrix and hence, we can write
G ( X, Y ) = I + YH ( X, Y ) for some matrix H ( X, Y ) with entries in S−1 R[ X, Y ]. There is
some c ∈ S such that cH has entries in R[ X, Y ]. Then, G ( X, cY ) has entries in R[ X, Y ]. Now,
since deg M( X ) is invertible in S−1 R[ X ], it must be an element of S−1 R. Further, since
deg M( X + cY ) = det M( X ), we have det G ( X, cY ) = 1, thereby completing the proof. ■

T HEOREM 3.7. Let R be an integral domain, and let f be a unimodular vector in R[ X ](n) ,
such that one component has leading coefficient 1. Then f ( X ) ∼ f (0) over R[ X ].

Proof. Let J be the set of elements c ∈ R such that f ( X + cY ) is equivalent to f ( X ) over

R[ X, Y ]. We claim that J is an ideal.

• Let c ∈ J and a ∈ R. Then, f ( X + caY ) = f ( X + c( aY )) is equivalent to f ( X ) over

R[ X, aY ], which is a subring of R[ X, Y ], whence the equivalence holds over the latter
too.

• Let c, c′ ∈ J. Then f ( X + (c − c′ )Y ) is equivalent to f ( X ) over R[ X, (c − c′ )Y ], which

is again a subring of R[ X, Y ], whence the equivalence holds over the latter too.

8
We next contend that J is the unit ideal. Suppose not, then we can choose a maximal ideal m
containing J. Due to Proposition 3.5, f ( X ) is equivalent to f (0) over Rm [ X ], consequently,
using Lemma 3.6, there is a c ∈ R \ m such that f ( X + cY ) is equivalent to f ( X ) over
R[ X, Y ], a contradiction to the fact that J ⊆ m. Thus, J is the unit ideal in R and there exists
an M( X, Y ) ∈ GLn ( R[ X, Y ]) such that f ( X + Y ) = M( X, Y ) f ( X ). Substituting X = 0, we
get f (Y ) = M(0, Y ) f ( X ) where M(0, Y ) is also invertible and the conclusion follows. ■

T HEOREM 3.8. Let k be a field and f a unimodular vector in k [ X1 , . . . , Xn ](n) . Then f has
the unimodular extension property.

Proof. The proof of this is quite similar to that of Noether Normalization. We induct on r.
N
Suppose first that r ⩾ 2. Let Yr = Xr and Xi = Yi + Yr i for some suitable choice of Ni ’s
such that at least one component of g(Y1 , . . . , Yr ) = f ( X1 , . . . , Xr ) has leading coefficient
equal to 1.
Due to Theorem 3.7, using the fact that k [y1 , . . . , Yr−1 ] is an integral domain, we have
that
g(Y1 , . . . , Yr ) = M(Y1 , . . . , Yr ) g(Y1 , . . . , Yr−1 , 0)
where M ∈ GLn (k [Y1 , . . . , Yr ]). Note that g(Y1 , . . . , Yr−1 , 0) is unimodular over k [Y1 , . . . , Yr−1 ]
and hence, has the unimodular extension property, whence so does g(Y1 , . . . , Yr ). This
completes the induction step.
Finally, we must handle the base case of k [ X ], which is a PID. This is straightforward
for if f = ( f 1 , . . . , f n )⊤ , then making repeated use of the Euclidean algorithm, we can make
one of the components a unit, since gcd( f 1 , . . . , f n ) = 1. This completes the proof. ■
D EFINITION 3.9. A (commutative) ring A is said to have the unimodular extension property
if for every n ⩾ 1, every unimodular vector f ∈ A(n) has the property.the property.

L EMMA 3.10. Let A have the unimodular extension property. If E is a stably free A-module,
then E is free.

Proof. Let F be a finite free module of rank m such that E ⊕ F is finite free. We first show
that E is free when m = 1. That is, E ⊕ A ∼ = A(n) for some positive integer n. We may
treat both E and A as submodules of A(n) and let u1 = ( a11 , . . . , an1 )⊤ be a basis for the
submodule A as an A-module. Consider the canonical projection A(n) → A. This map
sends u1 to 1 and being A-linear, it is of the form

( x1 , . . . , x n ) ⊤ 7 → α1 x1 + · · · + α n x n

for some α1 , . . . , αn ∈ A. Thus, u1 is unimodular.

The unimodular extension property furnishes an
 
a11 · · · a1n
M = (u1 , . . . , un ) =  ... .. ..  ∈ GL ( A).

. .  n
an1 · · · ann

9
Using column operations, on M, one can make sure that u2 , . . . , un ∈ E while maintaining
M ∈ GLn ( A). Since u1 , . . . , un form a basis for A(n) , we see that u2 , . . . , un must span E,
whence E is free.
Finally, if F has rank m ⩾ 2, then write F = F ′ ⊕ A and use the first half of the proof to
induct downwards. ■
T HEOREM 3.11 (Q UILLEN -S USLIN ). Let k be a field. Every finite projective module over
k [ X1 , . . . , Xn ] is free.

Proof. Let P be a projective module over k [ X1 , . . . , Xn ]. Due to Theorem 2.2, P is stably free.
Next, due to Theorem 3.8 and Lemma 3.10, P must be free. ■

10