Chapter 4

G-projectivity
The central notion in this chapter is t h a t of 'Gorenstein projective modules';
it was introduced by Enochs and J e n d a in [25]1. We first present a different
view (from t h a t taken in chapter 1) on the G-class, then we move on to define
Gorenstein projective modules and prove t h a t the finite ones among them are
exactly the modules in the G-class. In the last two sections we focus on Cohen-
Macaulay local rings with dualizing modules. Over such rings the Gorenstein
projective modules can be identified as special modules in the Auslander class;
this view - - also due to Enochs et al. - - proves to be very fruitful, and a neat
theory for Gorenstein projective dimension becomes available.

4.1 The G-class Revisited

From Auslander's original definition - - see (1.1.2) - - it is not obvious how to
define "non-finite modules in the G-class", let alone how to dualize the notion.
In this section we show how to characterize modules in the G-class in terms of
complete resolutions by finite free modules, and this will be the starting point
for our future generalizations and dualizations.

Let M be some R-module, and let

O-~ K n - ~ Pn-1 -~ P~-2 -~ " " ~ P~ ~ Po ~ C - ~ O

be an exact sequence where the Pt-s are projective modules. It is easy to prove
t h a t E x t ~ ( K n , M ) = E x t ~ + n ( C , M) for m > 0; it is done by breaking the long
exact sequence into short ones, and using the fact that E x t ~ ( P t , M ) = 0 for all
m > 0 because the Pe-s are projective.

1Strictly speaking, only finite Gorenstein projective modules were defined in this paper.
But in [32], and other later papers, the same authors have tacitly understood the definition to
encompass also non-finite modules; of course, we do the same.
92 4. G-PROJECTIVITY

For the application of this classical technique - - sometimes called "dimension

shift" - - it is, of course, not vital that the P~-s are projective, only that the mo-
dules Ext~(P~, M) vanish for the module M in question. We already used the
technique in Observation (1.2.5), and we will resort to it frequently in this and
the next two chapters. To avoid redundancy we will apply the technique, once
and for all, in a very general setting. This gives us a lemma - - (4.1.1) and
two parallels: (4.1.6) and (4.1.7) - - to which we can then refer. To justify this
approach, let us point out that it is tempting in Lemma (4.1.1)(a) to think of
X as a complex of injective modules; in most applications, however, X will be a
complex of projective modules!

(4.1.1) L e m m a . Let X be an R - c o m p l e x and let M be an R-module. The

following hold:
(a) If Ext'~(M, X t ) = 0 for all m > 0 and g _> s u p X , then

E x t , ( M , C x ) = Ext~+n(M, C~,~)

for all m, n > 0 and e > sup X.

(b) If X is homologically trivial, then

nomR(C , M)
for all ~ 6 Z.
(c) If X is homological]y trivial, and E x t ~ ( X t , M) = 0 for all m > 0 and
6 Z, then

E x t ~ ( c X , M) = Ext R
m+n ( C xl _ n , M )

for all m, n > 0 and g 6 Z. Furthermore, the following are equivalent:

(i) Homn(X, M ) is homologically trivial.
(ii) Ext~(C x , M) = 0 for all e 6 Z.
(iii) E x t ~ ( C x, M ) = 0 for all m > 0 and ~ 6 Z.

Proof. (a): For each / _> s u p X we have a short exact sequence

(t) O-+C t+1

x -~ Xl -+ C x -~ O,

cf. (A.1.7.2). Since Ext'~(M,Xt) = 0 for m > O, the associated long exact
sequence,

• .. --+ E x t , ( M , X t ) --+ E x t , ( M , C x ) -~
E x t ~ +1 (M, cX+l) -+ E x t ~ +1 (M, X t ) - - + . . . ,

yields identities

E x t , ( M , C x ) = E x t ~ +1 (M, cX+l)
4.1. THE G-CLASS REVISITED 93

for m > 0. Piecing these together we get the desired identity.

(b): Applying the left-exact functor H o m R ( - , M) to the right-exact sequence
Xt+l -+ Xt -+ C~Y -~ 0, we get a left-exact sequence:

0 -+ Homn(Cex, M) --+ Homn(Xt, M) ~ Homn(Xt+l, M).

Evidently, the kernel ZH_~mR(X'M), i.e., Ker(HomR(Xt, M) -~ HomR(X~+I, M))

is isomorphic to HomR(C X, M) as wanted.
(c): The equivalence of (i) and (ii) is the fact that the complex HomR(X, M)
is homologically trivial if and only if the functor H o m R ( - , M ) leaves all the
short exact sequences 0 -+ Z~~( --+ Xt -+ C x -+ 0 exact. As in (a) the iden-
tity of Ext modules follows from (t); and now that we have E x t ~ ( C X , M ) --
Ext~(CX+m_l, M) for all m > 0 and ~ E Z, we see that also (ii) and (iii) are
equivalent. []

(4.1.2) D e f i n i t i o n . Let L E CL(R) be homologically trivial. We say that L

is a complete resolution by finite free modules if and only if the dual complex
L* = Homn(L, R) is homologically trivial.

We can now apply the "general dimension shift lemma", (4.1.1), to show that
modules in the G-class and complete resolutions by finite free modules are close
kin.

(4.1.3) P r o p o s i t i o n . Let L E cL(R) be homologically trivial. The following are

equivalent:
( i) L is a complete resolution by finite free modules.
(ii) A11 the cokernels C L, e E Z, belong to G(R).
(iii) HomR(L, T) is homologically trivial for every module T E jro(R).

Proof. It is clear by the Definition that (iii) is stronger than (i).

(i) ~ (ii): Fix an n E Z and set C = cL; we want to prove that C E G(R).
Since both L and L* are homologically trivial, we have E x t , ( C , R) = 0 for
m > 0 by Lemma (4.1.1)(c), and by (A.1.7.3) and (b) in the same Lemma it
follows that the dualized complex L* has

(t) CL__n+ 1 ~' Z L- - n -----~"( c L ) * ---- C * .

----

Dualizing once more yields a complex L** which is isomorphic to L; in particular,

it is homologically trivial and, as above, it follows that E x t , ( C * , R) = 0 for m >
0. The isomorphism between L and L** is the canonical one, (fL R, which in degree
is just the biduality map 5Le : Lt -+ Lt**. By (A.1.7.3), Lemma (4.1.1)(b), and
(t) the complex L** has

C L. = . i nL** ~ (C L
- - 1. -- . n+l) ~
= C**,
94 4. G-PROJECTIVITY

so we have an exact ladder

• "" ) Ln+l ) Ln ) C ) 0

• "" > Ln+I** > L.** > C** >0

and it follows by the five lemma that the biduality map 5c is an isomorphism.
(ii) ~ (iii): To prove that HomR(L,T) is homologically trivial, it is by
Lemma (4.1.1)(c) sufficient to see that E x t ~ ( C L , T ) = 0 for all ~ E Z. For T in
~'o(R) this is immediate by Corollary (2.4.2) as G-dimR C~ _< 0. []

The main result of this section describes modules in the G-class as infinite
syzygies of finite free modules.

(4.1.4) T h e o r e m . A finite R-module M belongs to G(R) if and only if there

exists a complete resolution by finite free modules L with CLo ~-- M.
Proof. The "if" part follows by the Proposition.
To prove "only if" we assume that M E G(R), and set out to construct a
complete resolution by finite free modules L E CL(R) with C L -~ M: When M
belongs to the G-class, then so does the dual module M*, cf. Observation (1.1.7).
Take a resolution L' of M* by finite free modules, then we have an exact sequence

GQLI1 L' 0tL'l L' x' M*

(t) "'" + > L~ - ~ L'l--1 - )''" - - ~ L~ ---+ --~ O.

Also the dualized sequence,

0 - ~ M** x'* >L~* (o~')*> . . . (or-l)

r' *>L~_I . (op')*> L~* ( 0r't + l*) > . . . ,

is exact, because its homology modules are E x t , ( M * , R), cf. (A.4.3). Also take
a resolution of M by finite free modules:
Ltt L tt LI' L II ~tt
• .. °~+b L~' o, > L~'_I o~_1> . . . ~ L~' ,~ M -~ O.

Let L be the complex in CL(R) obtained by pasting L" and L'*. That is, L has
modules

Lt=L}' for g k O , and

I *
Lt = (L'*)t+l = (L_(t+l)) for e < O ;

and differentials

0 ~ = 0 L'' for ~ > 0 ,

= = on' * for and
OoL = A'*SMA".
4.1. THE G-CLASS REVISITED 95

In degrees 0 , - 1 , and - 2 the complex L looks as follows:

... o~L" > L~ ~'*~M~" >L~* (oi)>L~.

L' * (o#') > ....

To see that L is homologically trivial, we note that

L It L It
BL = Be = Ze = Zf for ~ > 0 ;
B L = B oL" = Ker A" = KerA'*6MA" = ZOL;
BL1 = Im~'*(~M)~" : ImA'* = K e r ( 0 L ' ) * = zL1; and
BL =
B /+1
L'* = Zt+l
L'* = Z~ for g < --1.

Now that L _'~ O, we have CoL ~ BL1 = ImA'* ="~ M** ='~ M as wanted,
cf. (A.1.7.3).
It is equally straightforward to see that L* is homologically trivial. One can,
namely, consider it as the splice of the sequences

L' ** (o~')** (~tL- lI ) **

( oLl!) * * ~'**
(o~+~) > L t'** ) L 't - 1 ** > ... > L o'** > (M*)** -+ 0

and

0~ M* ~''* > L g * (at")* > - . . (o,~'0" > L t,,_ 1 . (oy")*> L~. (o,~'~)*> - " .
The first one is isomorphic to (t) and, in particular, exact. The second is exact
because its homology modules are E x t , ( M , R). This concludes the proof. []

(4.1.5) E x a m p l e . Let R be a local ring, and assume that x and y are elements
in the maximal ideal with

(t) AnnR(x) = (y) and AnnR(y) = (x).

The complex

L .... ~R~R~R~R~R~...
is then homologically trivial, and HomR(L, R) ~ ~ I L , so L is a complete reso-
lution by finite free modules. The modules R/(x) and R/(y) are not projective,
but all the cokernels in L have this form, so it follows by the Theorem that
R/(x) and R/(y) belong to G(R).
The immediate concrete example of such a ring is the Gorenstein ring
R = k ~ X , Y ] / ( X Y ) , where k is a field. More generally we can set R =
R'[[X, Y]]/(XY), where R' is any local ring; then the residue classes x and y
of, respectively, X and Y have the property (t). It follows by [49, Theorem 23.5]
and [12, Proposition 3.1.19(b)] that R is Gorenstein if and only if R' is so,
and by [12, Theorems 2.1.2 and 2.1.9] the same holds for the Cohen-Macaulay
property. In particular we now have examples of non-projective modules in the
G-class of non-Gorenstein rings.
96 4. G-PROJECTIVITY

The last two lemmas are parallel to Lemma (4.1.1); they will come in handy at
a later point.

(4.1.6) L e m m a . Let X be an R - c o m p l e x and let M be an R-module. The

following hold:
(a) If Ext~(Xt, M) = 0 for all m > 0 and e <_ infX, then
Ext~*(Z~¥, M) = Ext~+n (Z~X'_n,M)
for all m, n > 0 and g <_ inf X.
(b) I f X is homologically trivial, then
zHomR(M,X) ~a HomR(M, Zx)
l
for all g • Z.
(c) I f X is honmlogically trivial, and Ext~(M, Xt) = 0 for all m > 0 and
g • Z, then
E x t , ( M , Zx) = Ext~+n(M, zX+n)
for all m, n > 0 and g • Z. Furthermore, the following are equivalent:
(i) Homn(M, X ) is homologically trivial.
(ii) Ext]~(M, Zx) = 0 for all g • Z.
(iii) E x t , ( M , Zx) = 0 f o r a l l m > 0 a n d g • Z.

Proof. Similar to the proof of Lemma (4.1.1). []

(4.1.7) L e m m a . Let X be an R - c o m p l e x and let M be an R-module. The

following hold:
(a) I f Tornm(Xe, M) = 0 for all m > 0 and ~ <_infX, then
Tornm(Z x , M) = Torm+n(Zt_n,
n x M)
for all m, n > 0 and g _< inf X.
(b) If X is homologically trivial, then
C M®nx ~- M ®n C x
for all g 6 Z.
(c) I f X is homologically trivial, and Tornm(M, X t ) = 0 for all m > 0 and
g 6 Z, then
Tor~(M, C x) -- Tor~+,(M, cX_n)
for all m, n > 0 and g 6 Z. Furthermore, the following are equivalent:
(i) M ®n X is homologically trivia/.
(ii) WOrld(M,Ctx) = 0 for all t • Z.
(iii) Tornm(M, C x) = 0 for all m > 0 and ~ • Z.

Proof. Similar to the proof of Lemma (4.1.1). []

4.2. GORENSTEIN PROJECTIVE MODULES 97

Notes
The view we have taken on the G-class in this section is quite different from
that taken in chapter 1, but it is still part of Auslander's original work: the hard
part of Theorem (4.1.4) is covered by [1, Proposition 8, p. 67].

4.2 Gorenstein Projective M o d u l e s

We introduce Gorenstein projective modules - - a notion that includes the usual
projective modules - - and we prove that a finite module belongs to the G-class
if and only if it is Gorenstein projective.

(4.2.1) D e f i n i t i o n s . Let P E CP(R) be homologically trivial. We say that P

is a complete projective resolution if and only if the complex Homn(P, Q) is
homologically trivial for every projective R-module Q.
A module M is said to be Gorenstein projective if and only if there exists a
complete projective resolution P with C P ~ M.

(4.2.2) O b s e r v a t i o n . Let P ' be a projective module, then the complex P --

0 --~ P ' - ~ P ' --+ 0, concentrated in degrees 0 and - 1 , is a complete pro-
jective resolution with C p ~ P ' . Thus, every projective module is Gorenstein
projective.
By Theorem (4.1.4) and Proposition (4.1.3) it follows that all modules in
the G-class are Gorenstein projective. The converse also holds, that is, finite
Gorenstein projective modules belong to the G-class. This is the contents of
Theorem (4.2.6).

(4.2.3) R e m a r k . If M is a Gorenstein projective R-module and p is a prime

ideal in R, then it is not obvious from the definition that Mp is a Gorenstein
projective Rp-module. It is, however, so (at least) if R is a Cohen-Macaulay
local ring with a dualizing module; we prove this in Proposition (4.4.14).

(4.2.4) L e m m a . Let M be an R - m o d u l e and assume that E x t ~ ( M , Q ) = 0 for

all m > 0 and all projective modules Q. I f T is a module of finite fiat dimension,
then E x t ~ ( M , T ) = 0 for m > 0.

Proof. If T E ~-0(R), then T has finite projective dimension; this follows by

Jensen's [45, Proposition 6], see also Theorem (3.4.14). Let

Q = O-~Q~,-+...~QI-~Qo~O

be a projective resolution of T, then sup Q = 0, C0Q ~ T, and C Q = Qu. For

m > 0 we then have

E x t ~ (M, T) -- Ext m+u

R (M, Qu)

by Lemma (4.1.1)(a), so E x t ~ ( M , T ) = 0 for m > 0 as wanted. []

98 4. G-PROJECTIVITY

(4.2.5) P r o p o s i t i o n . If P E CP(R) is homologically trivial, then the following

are equivalent:
( i) P is a complete projective resolution.
( ii) All the cokernels C P, ~ E Z, are Gorenstein projective modules.
( iii) HomR(P,T) is homologically trivial for every module T E J:o(R).
In particular: if M is Gorenstein projective and T E Y:o(R), then E x t , ( M , T) =
0 f o r m > O.

Proof. It is clear from the definitions in (4.2.1) that (i) ~ (ii), and that (iii) is
stronger than (i). If all the cokernels in P are Gorenstein projective, then, by
(4.2.1) and Lemma (4.1.1)(c), we have E x t ~ ( C P, Q) = 0 for all m > 0, all ~ E Z,
and all projective modules Q. For every ~ E Z and T E ~'0(R) it then follows by
Lemma (4.2.4) that E x t ~ ( C P, T) = 0 for m > 0. This proves the last assertion,
and it follows by Lemma (4.1.1)(c) that Homn(P, T) is homologically trivial, so
( ii) implies ( iii). []

The last assertion in (4.2.5) can be interpreted as saying that, as far as modules
of finite flat dimension are concerned, Gorenstein projective modules behave as
projectives.

(4.2.6) T h e o r e m . A finite R - m o d u l e is Gorenstein projective ff and only ff it

belongs to the G-class. That is,

M is finite and Gorenstein projective ¢=~ M E G(R).

Proof. The "if" part is, as observed in (4.2.2), immediate by Theorem (4.1.4)
and Proposition (4.1.3); the converse, however, requires a little more work. Let
M be a finite Gorenstein projective R-module, we want to construct a complete
resolution L by finite free R-modules such that CoL -~ M. We get the left half
of a complex L E CL(R) by taking a resolution of M by finite free modules:

...--~ L e - - ~ . . . - ~ L1--~ Lo--~ M--~ O.

It is now sufficient to prove that M fits in a short exact sequence

(t) O --+ M -+ L _ I "-~ C-1 -"~ O,

where L-1 is a finite free module and C-1 is a finite Gorenstein projective
module. The right half of L can then be constructed recursively: the n-th
step supplies a finite free module L - n (and an obvious differential) and a finite
Gorenstein projective module C_u. A complex L constructed this way is ho-
mologically trivial and has C L -~ i . For ~ < 0 we have n x t ~ ( C L, R) = 0 by
Proposition (4.2.5), because the cokernel C L = Ct is Gorenstein projective; and
for g _> 0 it follows by Lemma (4.1.1)(c) that

Ext~(C L, R) = Ext~+e(M, R) = 0,
4.3. G-PROJECTIVES OVER COHEN-MACAULAY RINGS 99

so L is a complete resolution by finite free modules. T h a t is, the proof is complete

when the short exact sequence (t) is established.
Since M is Gorenstein projective there exists a complete projective resolution
P with zP1 ~ C P ~ M, cf. (A.1.7.3). T h a t is, there is a short exact sequence

0 --+ M ~ P-1 -+ cP1 --+ O,

where P-1 is projective and cP1 is Gorenstein projective, cf. Proposition (4.2.5).
For a suitable projective module Q the sum P-1 • Q is free, and adding to P
the homologically trivial complex 0 --~ Q - ~ Q -4 0 (concentrated in degrees - 1
and - 2 ) , we get a new complete projective resolution P ' with C P' ~ M and a
free module in degree - 1 . Thus, we can assume that P-1 is free. Since M is
finite, the image of M in P-1 is contained in a finite free submodule L-1 of P-1.
We now have a short exact ladder
0 ~ M ~ L-1 ~ C-1 ~0

1= I I
0 ~M ) P-1 ~ c_P1 ~0

To see that C-1 is Gorenstein projective, it is sufficient to prove that

E x t ' ( C _ 1 , Q) = 0 for every projective R-module Q. This follows by a result 2
similar to [25, Theorem 2.13] (see the remarks on p. 626 ibid.). But this is easy:
E x t ~ ( c P i , Q) = 0 and Ext~t(L_l , Q) = 0 for every projective module Q, so we
have a commutative diagram

0 -4 Homn(C_P1, Q) --~ H o m n ( P - 1 , Q) -4 H o m n ( M , Q) --~ 0

i I i--
0 ~ H o m n ( C - 1 , Q) ~ HomR(L-1,Q) -~ H o m n ( M , Q ) ~ E x t ~ ( C - 1 , Q ) --+ 0

and we can immediately see that the map H o m R ( L - 1 , Q ) ~ H o m R ( M , Q ) is

surjective and, therefore, E x t , ( C _ 1 , Q) = 0 as desired. []

Notes
The proof of Theorem (4.2.6) is due to Avramov et al.; it will appear in [6].
In (5.1.11) we will use the same technique to prove that finite Gorenstein flat
modules belong to the G-class.

4.3 G-projectives over Cohen-Macaulay Rings

The purpose of this section is to characterize Gorenstein projective modules
over Cohen-Macaulay rings as distinguished modules in the Auslander class.
This view is due to Enochs, Jenda, and Xu [32].
2It is spelled out in Corollary (4.3.5) and proved for modules over a Cohen-Macaulay local
ring with a dualizing module.
100 4. G-PROJECTIVITY

(4.3.1) S e t u p . In this section R is a C o h e n - M a c a u l a y local ring with a

d u a l i z i n g m o d u l e D.

Enochs' notion of flat preenvelopes plays a key role in the proof of the main
theorem, so we start by recalling the definition. The extra assumptions on R
are irrelevant for (4.3.2) and (4.3.3) but, needless to say, crucial for (4.3.4).

(4.3.2) Flat Preenvelopes. Let M be an R-module. A homomorphism

¢: M -4 F, where F is a flat R-module, is said to be a fiat preenvelope of M if
and only if the sequence

Homn(F,F') HomR(¢,F'))H o m R ( M , F ' ) ~0

is exact for every flat R - m o d u l e F ' . T h a t is, if F' is fiat and v: M --~ F' is a
homomorphism, then there exists a v' E HomR(F, F ' ) such t h a t v = v'¢.
¢
M ,F

F'
Every module over a Noetherian ring has a fiat preenvelope, cf. [21, Proposi-
tion 5.1].

(4.3.3) L e m m a . Let M be an R-module. If there exists an injective homo-

morphism from M into a fiat R-module, then every fiat preenvelope of M is
injective.

Proof. Let v: M --4 F ' be an injective homomorphism from M into a flat mo-
dule F ' , and let ¢: M --4 F be a flat preenvelope of M. There exists then a
h o m o m o r p h i s m # : F --4 F ' such t h a t v = v'¢, and since v is injective so is
¢. []

(4.3.4) T h e o r e m . For an R-module M the next three conditions are equivalent.

( i) M is Gorenstein projective.
(ii) M E Ao(R) and E x t , ( M , Q) = 0 for all m > 0 and all projective
modules Q.
(iii) M E Ao(R) and E x t ~ ( M , T ) -- 0 for all m > 0 and all T E 3Co(R).

Proof. T h e third condition is stronger than the second; this leaves us two impli-
cations to prove.
(i) ~ (iii): It was proved in Proposition (4.2.5) that E x t ~ ( M , T ) -- 0 for all
m > 0 and T E 3c0(R); now we prove that M meets conditions (1)-(3) of The-
orem (3.4.6). Let E be a faithfully injective R-module, then T = H o m R ( D , E)
4.3. G-PROJECTIVES OVER COHEN-MACAULAY RINGS 101

belongs to ~'0(R). Let P be a complete projective resolution with C P ~ M; by

commutativity and adjointness (A.2.8) we have

HomR(D ®R P, E) ~ HomR(P, HomR(D, E)) = HomR(P, T),

and the latter complex is homologically trivial by Proposition (4.2.5). By faith-

fulness of H o m R ( - , E) it then follows that D ®R P is homologically trivial; in
particular, TorRm(D, M) = Torm(D, R P ) = 0 for m > 0, cf. Lemma (4.1.7)(c), so
CO
M meets the first condition in Theorem (3.4.6). Furthermore, we have

z D ? RP :"" CoD®aP "~

= D @R CoP "~
=D®RM

by (A.1.7.3) and Lemma (4.1.7)(b). Also the complex H o m R ( D , D ®R P ) is

homologically trivial; this follows because it is isomorphic to the complete pro-
jective resolution P: the isomorphism is the natural one, ~/D, where the t-th
component ('),D)t = ")'De is invertible as Pt E .Ao(R). For the same reason, for
each g E Z we have E x t , ( D , D®RPe) = 0 for m > 0, and D®RPe = (D®RP)~,
so by Lemma (4.1.6)(c) it follows that E x t , ( D , ZD®RP) = 0 for all g E Z and
m > 0. In particular, E x t , ( D , D ®R M) = 0 for m > 0, cf. ($), so M satisfies
also the second condition in (3.4.6). In view of ($) it follows by Lemma (4.1.6)(b)
that

zH1
mR(D'D®RP)_ '~ HomR(D, D ®R M);

and zP1 ~ M, cf. (A.1.7.3), so we have an exact ladder

0 -~ M -~ P-I -~ P-2 -~ "'"

0 ~ HomR(D, D ®R M) -~ HomR(D, D ®R P - i ) ~ HomR(D, D ®R P - 2 ) --~ "'"

and the five lemma applies to show that the canonical map 3' 0 is an isomor-
phism. Hereby, also the third condition in Theorem (3.4.6) is met, and it follows
that M E Ao(R).
(ii) ~ (i): We assume that M belongs to the Auslander class and has
E x t , ( M , Q) = 0 for all integers m > 0 and all projective modules Q. Our target
is construction of a complete projective resolution P with C P ~ M. First, note
that we get the left half of a complex P E CP(R) for free by taking a projective
resolution of M:

. . . ~ - - . ~ ~ M ~ O .

Next, note that to establish the right half of P it is sufficient to prove the
existence of a short exact sequence

(*) 0 ~ M ~ P-1 -'~ C-1 ~ O,

102 4. G-PROJECTIVITY

where P_I is projective and C-1 is a module with the same properties as M.
Then the right half can be constructed recursively: the n-th step supplies a
projective module P - n (and an obvious differential) and a module C_,~ E A o ( R )
with Ext~(C_,~, Q) = 0 for m > 0 and Q projective. A complex P established
this way is homologically trivial with C~' -~ M. Let Q be a projective R-module;
for g > 0 we have E x t ~ ( c P , v ) = Ext~+l(M,Q) = 0 by Lemma (4.1.1)(c) and
the assumptions on M, and for g < 0 we have Ext]~(C P, Q) = 0 because C P = Ce
is a module with the same properties as M. Thus, P will be a complete projective
resolution, and the Theorem is, therefore, proved when we have established the
short exact sequence (,).
First, choose an injective module I such that D ®R M can be embedded in
I, and apply HomR(D, - ) to the sequence 0 -+ D ®R M --+ I. This yields an
exact sequence

(*) O ~ M --~ T,

where we have used that H o m R ( D , D ®R M ) ~ M as M E A o ( R ) , and we

have set T = H o m R ( D , I ) . Next, choose a fiat module F ' such that T is a
homomorphic image of F ' , and consider the short exact sequence

(tt) O ~ K - + F ' - ~ T - + O.

Applying H o m R ( M , - ) to (tt) we get an exact sequence

HomR(M, F ' ) Homn(M,!o))HomR(M, T) --+ Ext]~(M, K).

Since F ' is flat and T E ~0(R), by Foxby equivalence (3.4.11), also K E ~'0(R)
and, therefore, E x t l ( M , K ) = 0 by Lemma (4.2.4) and the assumptions on M.
The composition map H o m n ( M , ~) is, consequently, surjective, so there exists a
homomorphism u E HomR(M, F ' ) such that # = ~v, and since # is injective so
is u. Now take a flat preenvelope ¢: M ~ F , cf. (4.3.2). Since F ' is flat and u
is injective, also ¢ is injective, cf. Lemma (4.3.3), so we have an exact sequence

($:~) 0 -~ M ~¢ F.

Choose a projective module P-1 such that F is a homomorphic image of P - l ,

that is,
(**) 0---~ Z---~ P-1 --~ F---~ O

is exact. Arguing on ($:~) and (**) as we did above on (*) and (tt), we prove
the existence of an injective homomorphism 0: M --~ P-1 such that ¢ = lr0, and
setting C-1 -- Coker0, we have a short exact sequence

(**) O ---~ M -~ P_ I --+C-1 ---~0.

W h a t now remains to be proved is that C-1 has the same properties as M.

The projective module P-1 belongs to the Auslander class, and by assumption so
4.3. G-PROJECTIVES OVER COHEN-MACAULAY RINGS 103

does M; by Corollary (3.4.7)(a) it then follows from (**) that also C - i E Ao(R).
Let Q be projective; for m > 0 we have E x t ~ ( M , Q ) -- 0 = E x t ~ ( P _ I , Q ) , so
it follows from the long exact sequence of Ext modules associated to (**) that
E x t , ( C _ 1 , Q) -- 0 for m > 1. To prove that E x t , ( C _ 1 , Q) = 0, we consider the
right-exact sequence

HomR(P_I,Q) nomR(O,Q)~HomR(M,Q) -+ E x t , ( C - i , Q) ~ 0.

Since Q is flat and ¢: M---} F is a flat preenvelope, there exists, for each
~? E HomR(M,Q), a homomorphism y': F -+ Q such that ~1 = Y'¢; that is,
= ffTr0 = nomR(0, Q)(ff~r).
P-1

M , F

Q
Thus, the induced map I-IomR(O, Q) is surjective and, therefore, E x t , ( C _ 1 , Q) =
0. This concludes the proof. []

The next result holds over Noetherian rings in general (it is the dual of [25,
Theorem 2.13]), but the general version has a different proof.

(4.3.5) C o r o l l a r y . Let 0 -+ M' ~ M --4 M" -~ 0 be a short exact sequence of

R-modules. The following hold:
(a) If M" is Gorenstein projective, then M is Gorenstein projective if and
only if M' is so.
(b) If M' and M are Gorenstein projective, then M" is Gorenstein projective
if and only if E x t ~ ( M " , Q) = 0 for every projective module Q.
(c) If the sequence splits, then M is Gorenstein projective if and only if both
M' and M" are so.
Proof. (a): Assume that M " is Gorenstein projective, then, in particular, M "
belongs to the Auslander class, and it follows by Corollary (3.4.7)(a) that M E
Ao(R) if and only if M' E Ao(R). Let Q be a projective R-module; inspection
of the long exact sequence
• .---+ E x t ~ ( M " , Q) -+ E x t , ( M , Q) -+
(t) n x t ~ (M', Q) ~ Ex~R.
,,n+l ~vl~,,,Q) ~ . . -
shows that E x t ~ ( M , Q ) = E x t ~ ( M ' , Q ) for m > 0, as E x t ~ ( M " , Q ) -- 0 for
m > 0. It now follows by the Theorem that M is Gorenstein projective if and
only if M ' is so.
104 4. G-PROJECTIVITY

(b): It follows by Corollary (3.4.7)(a) that M " belongs to the Auslander class
because M' and M do so. Let Q be a projective module, then E x t , ( M , Q) =
0 = E x t ~ ( M ' , Q) for m > 0, so from (t) it follows that E x t ~ ( M " , Q) = 0 for
m > 1. The assertion is now immediate by the Theorem.
(c): If the sequence 0 ~ M' --+ M -4 M " --+ 0 splits, we have isomorphisms

E x t , ( M , Q) ~ E x t ~ ( M ' , Q) @ Ext mn (M I! , Q)

for all integers m > 0 and all projective modules Q. The assertion is then evident
by the Theorem and Corollary (3.4.7)(b). []

(4.3.6) D e f i n i t i o n . We use the notation cOP(R) for the full subcategory (of
C(R)) of complexes of Gorenstein projective modules, and we use it with sub-
scripts [] and D (defined as usual cf. (2.3.1)).

The last results of this section are auxiliaries needed for the proof of the main
theorem in section 4.4.

(4.3.7) L e m m a . If A • c~P(R) is homologically trivial and F • CF(R), then

also the complex Homa(A, F) is homologically trivial.
Proof. If F = 0 the assertion is trivial, so we assume that F is non-zero. Fur-
thermore, we can, without loss of generality, assume that At = 0 and Ft = 0 for
< 0. Set u = sup {~ • Z I Fe ~ 0}; we proceed by induction on u.
I f u = 0 then F is a flat module, and E x t , ( A t , F) = 0 for all m > 0 and ~ • Z,
cf. Theorem (4.3.4). Note that C A = 0 for ~ <_ 0; it follows by Lemma (4.1.1)(c)
that

Ext~(C A, F) = Ext~+t(C A, F) = 0

for ~ _> 0, so HomR(A, F) is homologically trivial, again by (4.1.1)(c).

Let u > 0 and assume that Homn(A, F) is homologically trivial for all com-
plexes F • CF(R) concentrated in at most u - 1 degrees. The short exact sequen-
ce of complexes 0 --4 [-,,-1F --~ F --4 E"F~ --~ 0 is degree-wise split, cf. (A.l.17),
so it stays exact after application of Homa(A, - ) . The complexes HomR(A, Fu)
and Homn(A, r u _ t F ) are homologically trivial by, respectively, the induction
base and hypothesis, so it follows that also HomR(A, F) is homologically trivial.
[]

(4.3.8) P r o p o s i t i o n . If X is equivalent to A • C~P(R) and U ~- F • CF(R),

then RHomR(X, U) is represented by HomR(A, F).
Proof. Take a projective resolution P • CP(R) of X, then R H o m n ( X , U) is
represented by the complex Homn(P, F). Since P _ X ~_ A there is by (A.3.6)
a quasi-isomorphism a : P = ~ A, and hence a morphism

HomR(a, F) : HomR(A, F) ~ HomR(P, F).

4.4. GORENSTEIN PROJECTIVE DIMENSION 105

The mapping cone M ( a ) is homologically trivial, and it follows by Corol-

lary (4.3.5)(c) that it belongs to c~P(R). By (A.2.1.4) we have
M(HomR(a, F)) ~ Z1SOmR(Jkd(a), F),
so it follows from the Lemma that the mapping cone Ad(HOmR(a, F)) is homo-
logically trivial, and HOmR(a, F) is, therefore, a quasi-isomorphism. In partic-
ular, the two complexes HomR(A, F) and HomR(P, F) are equivalent, so also
HomR(A, F) represents RHomR(X, U). []

(4.3.9) L e m m a . Let F be a fiat R-module. If X E C(D)(R) is equivalent to

A • C-~p (R) and n _> sup X , then

Extnm (Cn,
A F) = H_(,~+n)(RHomn(X, F))

for m > O. In particular, there is an inequality:

inf (RHomR(C A, F)) _> inf (RHomR(X, F)) + n.

Proof. Since n _> s u p X = supA we have An-1 ~ EncA, cf. (A.1.14.3), and
since F is flat it follows by the Proposition znat RHomR(C~, F) is represented
by HomR(E-n(An-~),F). For m > 0 the isomorphism class Ext R A F) is
m (Cn,
then represented by
H_m(HOmR(E-n(An-1), F) ) = H_m(EnHomR( An3, F) )
= H_(m+n)(Homa(An"q , F))
= H-(m+n)( E-nHOmR(A, F))
= H_(m+n)(HOmR(A, F)),
el. (A.2.1.3), (A.1.3.1), and (A.1.20.2). It also follows from the Proposition
that the complex HomR(A,F) represents R H o m R ( X , F ) , so E x t , ( C A , F ) =
H-(m+n)(RHomR(X, F)) as wanted, and the inequality of infima follows. []

Notes
The proof of Theorem (4.3.4) is based on an idea due to Enochs and Xu; it was
communicated to the author by Foxby.
The Auslander class is defined for every local ring with a dualizing complex,
but for non-Cohen-Macaulay rings the relation to Gorenstein projective modules
is yet to be uncovered.

4.4 Gorenstein Projective D i m e n s i o n

Since every projective module is Gorenstein projective, cf. Observation (4.2.2),
the definition of Gorenstein projective dimension - - (4.4.2) below - - makes sense
over any Noetherian ring. However, the only successful approach (that we know
of) to a nice functorial description goes via the Auslander class, and to make it
work it is (so far) necessary to take the base ring Cohen-Macaulay.
106 4. G-PROJECTIVITY

(4.4.1) S e t u p . In this section R is a C o h e n - M a c a u l a y local ring w i t h a

dualizing m o d u l e D.

(4.4.2) Definition. The Gorenstein projective dimension, Gpd R X, of a com-

plex X E C(3)(R) is defined as

G p d n X = inf {sup {~ E Z IAe # 0 } I X _~ A E c~P(R)}.

Note that the set over which infimum is taken is non-empty: any complex X E
C(~)(R) has a projective resolution X ~~- P E CP(R), and CP(R) C C~P(R).

(4.4.3) O b s e r v a t i o n . We note the following facts about the Gorenstein projec-

tive dimension of X E C(~)(R):

Gpd R X E {-oo} U Z U {co};

p d n X >_GpdRX_> supX; and
Gpd n X = - c o ~ X-~0.

While the Definition and the Observation above make perfect sense over any
Noetherian ring, the proof (at least) of the next theorem relies heavily on the
fact that the base ring is local Cohen-Macaulay and has a dualizing module.

(4.4.4) G P D T h e o r e m . Let X E C(~) (R) and n E Z. The following are equiv-

alent:
(i) X is equivalent to a complex A E caP(R) concentrated in degrees at
most n; and A can be chosen with At = 0 tbr /? < inf X.
(ii) Gpd R X < n.
(iii) X E A ( R ) and n >_ infU - inf (RHomR(X, U)) for all U 7k 0 in J:(R).
(iv) X E A ( R ) , n >_ supX, and n >_ - i n f (RHomR(X,Q)) for all projective
modules Q.
(v) n >_ sup X and the module C A is Gorenstein projective whenever
A E C-~P(R) is equivalent to X .

Proof. It is immediate by Definition (4.4.2) that (i) implies (ii).

(ii) ~ (iii): Choose a complex A in ¢~P(R) concentrated in degrees at most
n and equivalent to X. It follows by Proposition (3.1.14) that A, and thereby X,
belongs to the Auslander class. Let U E 5r(R) be homologically non-trivial, set
i = inf U, and choose by (A.5.5) a complex F _~ U in CF(R) with Fe = 0 for e < i.
By Proposition (4.3.8) the complex HomR(A, F) represents RHomR(X, U), in
particular, inf (RHomR(X, U)) = inf (Homn(A, F)). For g < i - n and p E Z
either p > n or p + e < n + ~ < i, so the module

HomR(A,F)e = H HomR(Ap, Fp+t)

pEZ
4.4. GORENSTEIN PROJECTIVE DIMENSION 107

vanishes. In particular, the homology modules Ht(HomR(A,F)) vanish for

t < i - n, so inf (RHomR(X, U)) > i - n = inf U - n as desired.
(iii) ~ (iv): Let E be a faithfully injective R-module, then HomR(D,E) E
3r0(R), and by Lemma (3.4.3)(c), (A.4.10), and adjointness (A.4.21) we have

sup X = sup (D ®L X)
= - inf (RHomR(X ®L D, E))
= - inf (RHomn(X, RHomR(D, E)))
= - inf (RHomn(X, HomR(D, E)))
_~n.

(iv) ~ (v): Choose a complex A e C~P(R) equivalent to X, and consider

the short exact sequence of complexes 0 ~ [ - n - l A -~ c n A ~ ~ n C A
n -+ O. By
Proposition (3.1.14) the complex En-1A belongs to A(R), and since n _> sup X
we have c n A -~ A ~ X E A(R); it, therefore, follows by Lemma (3.1.13)
that C A e Ao(R). For projective modules Q we have - i n f (RHomR(C A, Q)) _<
- inf (RHomR(X, Q)) - n _< 0 by Lemma (4.3.9), so it follows by Theorem (4.3.4)
that C A is Gorenstein projective.
(v) =~ (i): Choose by (A.3.2) a projective resolution A e CP(R) C_c-~P(R) of
X with Ae = 0 for e < infX. Since n _> supX = supA it follows by (A.1.14.2)
that X "~ c n A , and c n A E CaP(R) as C A is Gorenstein projective. []

(4.4.5) G P D Corollary. For a complex X E C(3)(R) the next three conditions

are equivalent.
(i) X e ~(R).
(ii) Gpd R X < o0.
(iii) X e C([])(R) and GpdRX < supX + dimR.
Furthermore, if X E A(R), then

Gpd R X = sup {inf U - inf (RHomR(X, U ) ) I U E ~'(R) A U ~ 0}

= sup { - i n f (RHomR(X, Q ) ) I Q E CoP(R)}.

Proof. It follows by the Theorem that (ii) implies (i), and (iii) is clearly stronger
than (ii). For X E .A(R) and Q projective it follows by Lemma (3.4.13)(a) that

- inf (RHomR(X, Q)) < sup X + dim R,

so by the equivalence of (ii) and (iv) in the Theorem we have Gpd R X < s u p X +
dim R as wanted. This proves the equivalence of the three conditions.
For X E A(R) the equalities now follow by the equivalence of (ii), (iii), and
(iv) in the Theorem. []

As one would expect by now, the Gorenstein projective dimension agrees with
the G-dimension for complexes with finite homology.
108 4. G-PROJECTIVITY

(4.4.6) C o r o l l a r y ( G D - G P D E q u a l i t y ) . For every X E c((f~(R) there is an

equality:

G - d i m n X = Gpd R X.

Proof. It follows by GD Corollary (2.3.8), Theorem (3.1.10), and GPD Corol-

lary (4.4.5) that the two dimensions are simultaneously finite, namely when X
belongs to R ( R ) = A(f)(R). The equality is now immediate by (EF) in Theo-
rem (2.4.7) and the equalities in (4.4.5). []

The next proposition shows that Gorenstein projective dimension is a refinement

of projective dimension.

(4.4.7) P r o p o s i t i o n ( G P D - P D I n e q u a l i t y ) . For every X E C(3)(R) there is

an inequality:

Gpd n X _< pd n X,

and equality holds if pd n X < c~.

Proof. The inequality is, as we have already observed, immediate because pro-
jective modules are Gorenstein projective. Furthermore, equality holds if X ~ 0,
so we assume that pd n X = p E Z and choose, by (A.5.4.1), an R-module T
such that p = - i n f ( R H o m n ( X , T)). Also choose a projective module Q such
that T is a homomorphic image of Q. The short exact sequence of modules
0 ~ K --+ Q ~ T --+ 0 induces, cf. (A.4.7), a long exact sequence of homology
modules:

• .. ~ H _ p ( R H o m n ( X , Q)) ~ H _ v ( R H o m n ( X , T ) ) --+
H _ ( p + l ) ( R H o m n ( X , K ) ) -~ . . . .

Since, by (A.5.4.1), H _ ( p + l ) ( R H o m n ( X , K ) ) = 0 while H _ v ( R H o m R ( X , T))

0, we conclude that also H _ p ( R H o m n ( X , Q)) is non-zero. This proves, in view
of GPD Corollary (4.4.5), that Gpd n X _> p, and hence equality holds. []

By GPD Corollary (4.4.5) the next theorem is just a rewrite of the A version
(3.1.12).

(4.4.8) Gorenstein Theorem, G P D V e r s i o n . Let R be a Cohen-Macaulay

local ring with residue field k. If R admits a dualizing module, then the following
are equivalent:
( i) R is Gorenstein.
(ii) Gpd n k < cx~.
(iii) Gpd n M < oo for all finite R-modules M.
(iv) Gpd n M < c~ for all R-modules M.
(v) G p d n X < cx~ for all complexes X E C(D)(R). []
4.4. GORENSTEIN PROJECTIVE DIMENSION 109

In (4.4.9)-(4.4.13) we treat Gorenstein projective dimension for modules: we

rewrite (4.4.4) and (4.4.5) in classical terms of resolutions and Ext modules.

(4.4.9) D e f i n i t i o n . A Gorenstein projective resolution of a module M is defined

the usual way, cf. (1.2.1). All modules have a projective resolution and, hence,
a Gorenstein projective one.

(4.4.10) L e m m a . L e t M be an R - m o d u l e . I f M is equivalent to A E C-~P(R),

then the truncated c o m p l e x

AoD . . . . -+ A t -~ . . . --4 A2 -+ A I ---) Z A --+ 0

is a Gorenstein projective resolution o f M .

Proof. Suppose M is equivalent to A E c~P(R), then i n f A = 0, so AoD --~ A _~

M by (A.1.14.4), and we have an exact sequence of modules:

(t) -.. --+ At -~ ... ~ A2 -~ A1 -+ ZoA ~ M ~ 0.

Set v = inf {t E Z [ A t ~ 0}, then also the sequence

0 --4 ZAo ~ Ao -~ . . . ~ Av+l --+ Av ~ 0

is exact. All the modules A o , . . . , Av are Gorenstein projective, so it follows by

repeated applications of Corollary (4.3.5)(a) t h a t Z A is Gorenstein projective,
and therefore AoD is a Gorenstein projective resolution of M , cf. (t)- []

(4.4.11) R e m a r k . It follows by the L e m m a and Definition (4.4.2) that an R - m o -

dule M is Gorenstein projective if and only if G p d n M _< 0. T h a t is,

M is Gorenstein projective ¢=~ G p d n M = 0 V M = 0.

(4.4.12) G P D T h e o r e m f o r M o d u l e s . L e t M be an R - m o d u l e and n E No.

T h e following are equivalent:
( i) M has a Gorenstein projective resolution o f length at most n. /.e., there
is an exact sequence o f m o d u l e s 0 --~ An -+ "'" --4 A1 "-4 Ao ~ M -+ O,
where Ao, A 1 , . . . , A n are Gorenstein projective.
(ii) Gpd R M <_ n.
(iii) M E A o ( R ) and E x t ~ ( M , T ) = 0 for all m > n and all T E .To(R).
(iv) M E A o ( R ) and E x t ~ ( M , Q ) = 0 for all m > n and all projective mo-
dules Q.
(v) In any Gorenstein projective resolution o f M ,

• .. --+ A t -~ A t - 1 -4 " " -+ Ao -~ M --4 O,

the kernel a Kn = Ker(An_l --+ An-2) is a Gorenstein projective module.

aAppropriately interpreted for small n as Ko = M and K1 = Ker(Ao --+ M).
110 4. G-PROJECTIVITY

Proof. If the sequence .-. ~ A t --+ A t - 1 ~ " " ~ Ao ~ M ~ 0 is exact, then M

is equivalent to A . . . . -~ A t -~ Ae-1 --+ .. • -~ Ao ~ O. The complex A belongs
to c-~P(R), and it has C A ~ M, C A ~ Ker(Ao -+ M), and C A ~ ZA_x =
Ker(Ae_l -~ A t - 2 ) for ~ _> 2. In view of the Lemma the equivalence of the five
conditions now follows from Theorem (4.4.4). []

(4.4.13) G P D Corollary for M o d u l e s . For an R - m o d u l e M the next three

conditions are equivalent.
(i) M e A o ( R ) .
(ii) Gpd R M < o~.
(iii) G p d R M < dimR.
Furthermore, if M E A o ( R ) , then

Gpd R M = sup (m • No I S T • ~'0(R) : E x t , ( M , T) ~ 0}

= s u p ( m • No I 3 Q • CP(R): E x t ~ ( M , Q ) ~ 0}.

Proof. Immediate from Corollary (4.4.5). []

The next proposition shows that the Gorenstein projective dimension cannot
grow under localization. In particular, it follows that Mp is Gorenstein projective
over Rp if M is Gorenstein projective over R and, as we remarked in (4.2.3),
this is not obvious from the definition.

(4.4.14) P r o p o s i t i o n . Let X G C(~)(R). For every p G SpecR there is an

inequality:

GpdR0 X 0 < Gpd R X-

Proof. Let p be a prime ideal. If X is equivalent to A E Cg~P(R), then Xp is

equivalent to Ap. It is therefore sufficient to prove that a localized module Mp
is Gorenstein projective over Rp if M is Gorenstein projective over R.
Let M be a Gorenstein projective R-module and set d = dim Rv. It follows
from the definitions in (4.2.1) that there is an exact sequence

(t ) O--+ M - - r P_] -+ P - 2 - + . " ~ P - d - + C - + O,

where the modules P - 1 , . . . , P - d are projective. Since M and the projective

modules all belong to the Auslander class, it follows by repeated applications
of Corollary (3.4.7)(a) that also C E A0(R). Localizing at p we get an exact
sequence

(:~) 0 -+ g p ~ ( P - 1 ) p ~ ( P - 2 ) p ~ "'" --~ ( P - d ) p ~ Cp --'~0,

where the modules (Pt)p are projective over Rp, and Mp and Cp belong to
A(Rp), cf. Observation (3.1.7). From GPD Corollary (4.4.13) it follows that
GpdRp Cp _< d, and since (:~) is exact it follows by GPD Theorem (4.4.12) that
Mp is Gorenstein projective. []
4.4. GORENSTEIN PROJECTIVE DIMENSION 111

Finally we will now use Foxby equivalence to establish a series of test expressions
for the Gorenstein projective dimension.

(4.4.15) L e m m a . If X • .A(R) and U • Z(R), then

inf (RHomR(X, U)) = inf (RItomR(X, RHomR(D, U))).
Proof. In the calculation below the first equality follows as X • A(R), the second
follows by Horn evaluation (A.4.24) as U • Z(R), the third by Lemma (3.4.3)(a),
the fourth by commutativity (A.4.19) and the last one by adjointness (A.4.21).
inf (RHomR(X, V)) = inf (RHomR(RHomR(D, D ®~ X), V))
= inf ( n ®~ RHomR(D ®L X, U))
= inf (RHomR(D ®L X, U))
= inf (RHomR(X ®~ D, U))
= inf (RHomR(X, RHomR(D,U))). []

(4.4.16) T h e o r e m . If X is a complex of finite Gorenstein projective dimension,

i.e., X E .4(R), then the next five numbers axe equal.
(D) Gpd R X,
(EF) sup {inf U - i n f (RHomR(X, U ) ) I U E Y(R) A V 7~ 0},
(El) sup {inf U - inf (RHomR(X, U)) [ U • Z(R) A V ;~ 0},
(too) s u p { - i n f ( R H o m R ( X , T ) ) [ T • Z0(R)}, and
(EQ) sup {-- inf (RHomR(X, Q)) I Q • CP(R)} •
Proof. The numbers (D), (EF), and (EQ) are equal by GPD Corollary (4.4.5),
and it is obvious that (too) _< (m). This leaves us two inequalities to prove.
"(EQ) _< (El,,)": Let Q be a projective R-module, then, by Foxby equiv-
alence (3.4.11), the module T = D ®R Q has finite injective dimension, and
Q - HomR(D,T) represents RHomR(D,T), cf. Theorems (3.4.6) and (3.4.9).
It now follows by the Lemma that
- inf (RHomR(X, T)) = - inf (RHomR(X, Q)),
and hence the inequality follows.
"(El) ___~(EF)": If U • Z(R), then
inf (RHomR(X, U)) = inf (RHomR(X, RHomR(D, U)))
by the Lemma, and RHomR(D,U) • ~'(R) by (d) in the Foxby equiva-
lence Theorem (3.3.2). Furthermore, we have inf U = inf (RHomR(D, U)) by
Lemma (3.4.3)(d), so

inf U - inf (RHomR(X, U))

= inf (RHomR(D, U)) - inf (RHomR(X, RHomR(D, U))).
This proves the desired inequality, and with that the five numbers are equal. []
112 4. G-PROJECTIVITY

(4.4.17) Corollary. If M is a module of finite Gorenstein projective dimension,

i.e., M E Ao(R), then the next four numbers are equal.

(D) Gpd R M,
(EFo) sup{m E NO ] 3 T E ~ o ( R ) : E x t ' ~ ( M , T ) ¢ O } ,
(too) sup{mENo] 3TEZo(R):Ext'~(M,T)~O}, and
(EQ) sup {m E No ] 3 Q E CoP(R): E x t , ( M , Q) ~ 0). []

(4.4.18) O b s e r v a t i o n . The test expression (EQ) in Corollary (4.4.17) can be

traced back to the definition of Gorenstein projective modules, and one might,
therefore, expect it to hold over Noetherian rings in general. The test expression
(m,,), on the other hand, was established through Foxby equivalence, so it is
possible that it will only hold over Cohen-Macaulay rings. Both test expressions
are valid for G-dimension of finite modules over general Noetherian rings, cf. (R)
and (EF,,) in Corollary (2.4.8). But the crux of the matter is that the two test
expressions can only agree for non-finite modules if the ring is Cohen-Macaulay:
suppose R is local and not Cohen-Macaulay, by [10, Proposition 5.4] there exists
an R-module M with p d R M = dimR, but idRT < d i m R - 1 for all T E Z0(R)
(see page 13). Thus,

sup{m E No I 3 T E 2:0(R): E x t ~ ( M , T ) ~ 0} < d i m R - 1,

but, proceeding as in the proof of Proposition (4.4.7), it is easy to see that

sup{m E No I 3 Q e CP(R) : E x t ~ ( M , Q ) ~ 0} = dimR.

Here we tacitly assume that the Gorenstein projective dimension is defined as

in (4.4.2), so that Gpd R M _< pd R M -- dim R < oc.

Notes
The GPD Theorem (4.4.12) - - and (4.4.4) - - is modeled on Cartan and
Eilenberg's characterization of projective dimension [13, Proposition VI.2.1].
The proof of (4.4.4) follows the pattern from Foxby's notes [33], and this in-
cludes the auxiliary results (4.3.7), (4.3.8), and (4.3.9).
The proof of Proposition (4.4.14) is due to Foxby; it appeared in [39], and so
did the equalities in GPD Corollary (4.4.13).