Chapter 6

G-injectivity
The central notion in this chapter is 'Gorenstein injective modules' as intro-
duced by Enochs and Jenda in [25]1. The first two sections follow the familiar
pattern from chapters 4 and 5: first we introduce Gorenstein injective modules
(over general Noetherian rings), next we prove that (over local Cohen-Macaulay
rings) they are distinguished modules in an Auslander category, and then a neat
theory for Gorenstein injective dimension unfolds. In section 6.3 we study dual-
ity between G-flatness and G-injectivity, and in the final section 6.4 we collect
additional stability results, mostly in the form of exercises.

6.1 Gorenstein Injective M o d u l e s

We introduce Gorenstein injective modules - - a notion that includes the usual in-
jective modules - - and we characterize Gorenstein injective modules over Cohen-
Macaulay rings as distinguished modules in the Bass class. This view is due to
Enochs, Jenda, and Xu [32].

(6.1.1) D e f i n i t i o n s . Let I E CI(R) be homologically trivial. We say that I is a

complete injective resolution if and only if the complex H o m n ( J , I) is homologi-
cally trivial for every injective R-module J.
A module N is said to be Gorenstein injective if and only if there exists a
complete injective resolution I with Z0I ~ N.

(6.1.2) O b s e r v a t i o n . Every injective module is Gorenstein injective: let I' be

injective, then the complex I = 0 ~ I' -Y-+ I' ~ O, concentrated in degrees 1 and
0, is a complete injective resolution with ZoI ~ I'.

(6.1.3) R e m a r k . If N is a Gorenstein injective R-module and p is a prime ideal

in R, then it is not obvious from the definition that Np is a Gorenstein injective
1Gorenstein injective modules over Gorenstein rings were studied by the same authors in
an earlier paper [22].
136 6. G-INJECTIVITY

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 (6.2.13).

(6.1.4) L e m m a . Let N be an R - m o d u l e and assume that E x t , ( J , N) = 0 for all

m > 0 and all injective modules J. I f T is a module of finite injective dimension,
then E x t , ( T , N ) = 0 for m > O.

Proof. Let

J = O--'~go-'~g-l-+'"-~g-v---~O

be an injective resolution of T, then inf J = 0, Zg ~ T, and Z Jr- = J - v . For

m > 0 we then have

E x t , ( T , N) -- Ext~+v(J_v, N)

by Lemma (4.1.6)(a) and, therefore, E x t , ( T , N) -- 0 for m > 0. []

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

are equivalent:
( i) I is a complete injective resolution.
(ii) All the kernels Z~, g E Z, are Gorenstein injective modules.
( iii) HomR(T, I) is homologically trivial for every module T E Zo( R).
In particular: if N is Gorenstein injective and T E Zo (R), then E x t ~ (T, N) = 0
for m > O.

Proof. It is clear from the definitions in (6.1.1) that (i)=v(ii) and (iii)=~(i). If all
the kernels in I are Gorenstein injective, then, by (6.1.1) and Lemma (4.1.6)(c),
we have E x t , ( J , ZtI) = 0 for all m > 0, all g E Z, and all injective modules
J. For every g E Z and T E Zo(R) it now follows by Lemma (6.1.4) that
E x t , ( T , Z~) = 0 for m > 0. This proves the last assertion, and it follows, again
by Lemma (4.1.6)(c), that HomR(T, I) is homologically trivial, so (ii) implies
( iii ) . []

The last assertion in (6.1.5) can be interpreted as saying that, as far as modules
of finite injective dimension are concerned, Gorenstein injective modules behave
as injectives.
The key ingredient in the proof of the main result of the section is Enochs'
notion of injective precovers; we start by recalling the definition.

(6.1.6) I n j e e t i v e Precovers. Let N be an R-module. A homomorphism

~/: I -~ N, where I is an injective R-module, is said to be an injective precover
of N if and only if the sequence

HomR(I',I) HomR(l',~)) H o m R ( I ' , N ) >0

6.1. GORENSTEIN INJECTIVE MODULES 137

is exact for every injective R-module I'. T h a t is, if I t is injective and v : I ' -+ N
is a homomorphism, then there exists a v' HomR(F, I) such that v = T/v'.

I'

I ,N

Every module over a Noetherian ring has an injective precover, cf. [21, Proposi-
tion 2.2].

(6.1.7) T h e o r e m . Let R be a Cohen-Macaulay local ring with a dualizing mo-

dule. For an R-module N the next three conditions are then equivalent.
( i) N is Gorenstein injective.
(ii) N E Bo(R) and E x t , ( J , N) -- 0 for all m > 0 and all injective modules J.
(iii) N E Bo(R) and E x t , ( T , N) = 0 for all m > 0 and all T E Zo(R).
Proof. The third condition is stronger than the second; this leaves us two impli-
cations to prove.
(i) ~ (iii): It was proved in Proposition (6.1.5) that E x t , ( T , N) = 0 for
all m > 0 and all T E Zo(R). The dualizing module D has finite injective
dimension so, in particular, E x t ~ ( D , N ) -- 0 for m > 0. T h a t is, N meets t h e
first condition in Theorem (3.4.9), and we now prove that it also meets conditions
(2) and (3). Let I be a complete injective resolution with ZoI ~ N. It follows by
Proposition (6.1.5) that the complex HomR(D, I) is homologically trivial, and
by (A.1.7.3) and (b) in Lemma (4.1.6) we have

(~) "'l(~Hmn(D'I)='~~o~Hmn(D'l) ='~ HomR(D, Z0I) ='~ HomR (D, N).

Also the complex D R HomR (D, I) is homologically trivial; this follows because
it is isomorphic to the complete injective resolution I: the isomorphism is the
natural one, ~/D, where the ~-th component (~D)t = ~iD is invertible as It
Bo(R). For the same reason, for each ~ Z we have TorRm(D, HomR(D,It)) --
0 for m > 0, and HomR(D, It) = H o m R ( D , I ) t , so by Lemma (4.1.7)(c) it
follows that TorRm(D, "~t
(~nomR(D,~)j = 0 for all e Z and m > 0. In particular,
TOrRm(D,HomR(D,N)) = 0 for m > 0, cf. (~:), so N satisfies also the second
condition in (3.4.9). In view of ($) it follows by Lemma (4.1.7)(b) that

C DRHmR(D'I) ~- D R HomR(D, N),

and C11 ~ N, cf. (A.1.7.3), so we have an exact ladder

.. ~ D R HomR(D, I2) -~ D R HomR(D, I1) -~ D R HomR(D, N) -+ 0

"~ /2 -~ /1 --+ N -+0

138 6. G-INJECTIVITY

and the five lemma applies to show that the canonical map ~D is an isomor-
phism. With this, also the third condition in Theorem (3.4.9) is satisfied, and it
follows that N E Bo(R).
(ii) ::v (i): We assume that N belongs to the Bass class and has E x t , ( J , N) =
0 for all integers m > 0 and all injective modules J. We want to construct
a complete injective resolution I with Z01 -~ N. The right half of a complex
I E CI(R) we get for free by taking an injective resolution of N:

0 - ~ Io ~ I _ 1 - ~ ...--+ I~ ~ . . . .

To establish the left half of I, it is sufficient to prove the existence of a short

exact sequence

(*) 0 -~ Zl ~ / 1 -~ N -~ 0

where I1 is injective and Z1 is a module with the same properties as N. Then

the left half can be constructed recursively: the n-th step supplies an in-
jective module In (and an obvious differential) and a module Zn E Bo(R) with
E x t , ( J , Zn) -- 0 for m > 0 and J injective. A complex I established this way
is homologically trivial and has Z01 ~ N. Let J be an injective R-module; for
e _< 0 we have Ext~(J,Z~) -- E x t ~ - l ( J , N ) -- 0 by Lemma (4.1.6)(c) and the
assumptions on N, and for e > 0 we have E x t , ( J , Z~) -- 0 because Z / -- Ze is
a module with the same properties as N. Thus, I will be a complete injective
resolution, and the Theorem is, therefore, proved when we have established the
short exact sequence (,).
First, choose a projective module P such that HomR(D, N) is a homomorphic
image of P, and apply D R - to the sequence P --+ HomR(D, N) --+ 0. This
yields an exact sequence

(*) T --~ N --+ 0,

where we have used that D R H o m R ( D , N ) '~ N as N E Bo(R), and we have set

T = D R P- Next, choose an injective module F such that T can be embedded
in 1 I, and consider the short exact sequence

(ft) 0 ~ T 2-} I' ~ C --+ 0.

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

HomR(I', N) HomR(~,g)) HomR(T, N) --+ E x t , ( C , N).

Since I ' is injective and T E Zo(R), by Foxby equivalence (3.4.11), also C E

Z0(R) and, therefore, E x t ~ ( C , N ) = 0 by Lemma (6.1.4) and the assumptions
on N . Consequently, the composition map HomR(L, N) is surjective, so there
exists a homomorphism v E H o m R ( F , N ) such that ~/ = w , and since 3' is
surjective so is v. Now, take an injective precover 7/: I ~ N, cf. (6.1.6). Since F
is injective there is a homomorphism v ~ E HomR(F, I) such that v = ~?vI, and
6.1. GORENSTEIN INJECTIVE MODULES 139

since v is surjective also y must be surjective. Set Z 1 = Ker y, then we have a

short exact sequence

O-+ Z1--+ I1 --~ N--+ O.

What now remains to be proved is that Z1 has the same properties as N.

Both N and the injective module 11 belong to the Bass class, so by Corol-
lary (3.4.10)(a) it follows from (:~:~)that also Z1 E Bo(R). Let J be injective; for
m > 0 we have E x t ~ ( J , I1) = 0 = E x t ~ ( J , N ) , so it follows from the long exact
sequence of Ext modules associated to (:~:~) that E x t , ( J , Z1) = 0 for m > 1.
Now, consider the right-exact sequence

Homn(J, I1) HomR(J,~?)) HomR(J, N) --+ E x t , ( J , Z l ) --4 0.

The induced map HomR(J, r/) is surjective because/1 is an injective precover of

N, so also E x t , ( J , Z1) = 0. This concludes the proof. []

The Bass class is defined for every local ring with a dualizing complex, but for
non-Cohen-Macaulay rings the relation to Gorenstein injective modules is yet
to be uncovered.
The next result is [25, Theorem 2.13]. A straightforward proof, similar to
that of Corollary (4.3.5), applies when the base ring is local Cohen-Macaulay
with a dualizing module.

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

R-modules. The following hold:
(a) If N' is Gorenstein injective, then N is Gorenstein injective if and only if
N" is so.
(b) If N and N " are Gorenstein injective, then N' is Gorenstein injective if
and only if E x t , ( J , N') = 0 for all injective modules J.
(c) If the sequence splits, then N is Gorenstein injective if and only if both
N' and N " are so. []

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

In the rest of this section, that is, in (6.1.10)-(6.1.12) we assume that R is a

C o h e n - M a c a u l a y local r i n g w i t h a d u a l i z i n g m o d u l e . These last three
results are auxiliaries needed for the proof of the main theorem in section 6.2.

(6.1.10) L e m m a . If B E C~I(R) is homologically trivial and J E C~(R), then

also the complex HomR(J, B) is homologically trivial.
140 6. G-INJECTIVITY

Proof. If J = 0 the assertion is trivial, so we assume that J is non-zero. We can

also, without loss of generality, assume that Bt = 0 for g > 0 and Je = 0 for
g < 0. Set u = sup {g E Z [ Je ~ 0}; we proceed by induction on u.
If u = 0 then J is an injective module, and E x t , ( J , Be) = 0 for all m > 0
and all g E Z, cf. Theorem (6.1.7). Note that Zs = 0 for g _> 0; it follows by
Lemma (4.1.6)(c) that

Ext]~(J, ZB) = Ext]~-e (J, ZB) = 0

for t < 0, so HomR(J, B) is homologically trivial, again by (4.1.6)(c).

Let u > 0 and assume that H o m R ( J , B ) is homologically trivial for all
complexes J E C~(R) concentrated in at most u - 1 degrees. The short ex-
act sequence of complexes 0 ~ E u - 1 J -~ J -~ E~Ju -~ 0 is degree-wise split,
cf. (A.l.17), so it stays exact after application of H o m R ( - , B). As the complex-
es HomR(Ju, B) and HomR( E~-I J, B) are homologically trivial by, respectively,
the induction base an hypothesis, it follows that also HomR(J, B) is homologi-
cally trivial. []

(6.1.11) P r o p o s i t i o n . If Y is equivalent to B e C~I(R) and U ~_ J e C~(R),

then RHoma(U, Y ) is represented by HomR(J, B).

Proof. Take an injective resolution I e C~(R) of Y, then RHomR(U,Y) is rep-

resented by the complex HomR(J, I). Since B -~ Y --. I there is by (A.3.5) a
quasi-isomorphism/3: B ~- ~ I, and hence a morphism

HomR(J,13) : HomR(J,B) > HomR(J,I).

The mapping cone A4(~) is homologically trivial, and it follows by Corol-

lary (6.1.8)(c) that it belongs to C~I(R). By (A.2.1.2) we have

Ad(HomR(J, fl)) -- HomR(J, .~4(/3)),

so it follows from the Lemma that the mapping cone A4(HomR(J, fl)) is homo-
logically trivial, and HomR(J,/3) is, therefore, a quasi-isomorphism, cf. (A.l.19).
In particular, the two complexes HomR(J, B) and HomR(J, I) are equivalent, so
also HomR(J, B) represents RHomR(U, Y). []

(6.1.12) L e m m a . Let J be an injective R-module. If Y E C(n)(R) is equivalent

to B E C~I(R) and n >_ - infY, then

E x t . ( J , zB_n) ----H-(m+n)(RHoma(J, Y))

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

inf (RHomR(J, ZB_n)) > inf (RHomR(J, Y)) + n.

6.2. GORENSTEIN INJECTIVE DIMENSION 141

Proof. Since - n < infY = i n f B we have v'_nB ~_ z-nZBn,_ cf. (A.l.14.1), and
since J is injective it follows by the Proposition that RHomR(J, zB_n) is repre-
sented by HomR(J, E" [--nB). For m > 0 the isomorphism class E x t , ( J , Z_Bn)
is now represented by

H-m(Homn(J, Z" c_.B)) = H_m(E"HomR(J, r_.B))

= H_(m+n)(HomR(J, [--nB))
= H-(m+n)( C-nHomR(J, B))
= H-(m+n)(HomR(J, B)),

cf. (A.2.1.1), (A.1.3.1), and (A.1.20.1). It also follows by the Proposition

that the complex HomR(J,B) represents RHomR(J,Y), so E x t ~ ( j , zBn) ---
H-(m+n)(RHomR(J, Y)) as wanted, and the inequality of infima follows. []

6.2 Gorenstein Injective D i m e n s i o n

By Observation (6.1.2) every injective module is Gorenstein injective, and the
definition of Gorenstein injective dimension, (6.2.2) below, makes sense over any
Noetherian ring. However, as for Gorenstein projective and flat dimensions, we
only know how to get a nice functorial description if we work over a Cohen-
Macaulay local ring with a dualizing module.

(6.2.1) Setup. 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.

(6.2.2) Definition. The Gorenstein injective dimension, GidR Y, of a complex

Y C(E)(R ) is defined as

GidRY = inf {sup {~ Z IS_t # 0} I Y -~ B C~I(R)}.

Note that the set over which infimum is taken is non-empty: any complex Y
C(E)(R ) has an injective resolution Y ~-~ I C~(R), and C~(R) C_c I(n).
(6.2.3) Observation. We note the following facts about the Gorenstein injective
dimension of Y 6 C(7-)(R):

GidR Y {-co} U Z U {co};

idn Y _>Gidn Y _> - inf Y; and
GidnY=-co v Y ~ 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
assumption that the base ring is local Cohen-Macaulay and has a dualizing
module.
142 6. G-INJECTIVITY

(6.2.4) G I D T h e o r e m . Let Y E C(F)(R) and n E Z. The following axe equiva-

lent:
(i) Y is equivalent to a complex B E C~(R) concentrated in degrees at
least - n ; and B can be chosen with Bt = 0 for g > sup Y.
(ii) Gidn Y _< n.
(iii) Y E B(R) and n >_ - sup U - inf (RHomR(U, Y)) for all U ~ 0 in Z(R).
(iv) Y E B(R), n >_ - i n f Y , and n _> - i n f ( R H o m R ( J , Y ) ) for all injective
modules J.
(v) n > - i n f Y and the module Z B is Gorenstein injective whenever
B E C~I(R) is equivalent to Y.

Proof. It is immediate by Definition (6.2.2) that (i) implies (ii).

(ii) ~ (iii): Choose a complex B E C~(R) concentrated in degrees at least
- n and equivalent to Y. It follows by Proposition (3.2.13) that B, and thereby
Y, belongs to the Bass class. Let U E Z(R) be homologically non-trivial, set s --
supU, and choose by (A.5.1) a complex J _~ U in C~(R) with Jt = 0 for > s.
By Proposition (6.1.11) the complex Homn(J, B) represents RHomR(U, Y), in
particular, inf (RHomn(U, Y)) = inf (Homn(J, B)). For ~ < - s - n and p E Z
either p > s or p + g _< s + g < - n , so the module

HOmR(J,B)t = H Homn(Jp, Bv+t )

pEZ

vanishes. In particular, Ht(Homn(J, B)) = 0 for g < - s - n and, therefore,

inf (RHomn(U, Y)) > - s - n = - sup U - n, as desired.
(iii) ~ (iv): Since D E Zo(R) we have

- i n f Y = - inf ( R H o m n ( D , Y)) < n,

cf. Lemma (3.4.3)(d).

(iv) ~ (v): Choose a complex B E C~-I(R) equivalent to Y, and consider
the short exact sequence of complexes 0 --+ ~ - n z Bn_ --+ B - n D ~ B l - n ~ ~ O.
By Proposition (3.2.13) the complex Bl_n-7 belongs to B(R), and since
- n < i n f Y = i n f B we have B _ n D --~ B _~ Y E B(R), cf. (A.1.14.4).
By Lemma (3.2.12) it now follows that Z_Bn E Bo(R). For injective mo-
dules J we have - inf ( R H o m n ( J , Z_Bn)) _< - inf ( R H o m n ( J , Y)) - n _< 0 by
Lemma (6.1.12), so it follows by Theorem (6.1.7) that Z_Bn is Gorenstein injec-
tive.
(v)=~(i): Choose by (A.3.2) an injective resolution B E C~(R) C_ C~I(R) of Y
with Be = 0 for g > supY. Since - n < i n f Y = i n f B it follows from (A.1.14.4)
that Y ~- B - n D , and B - n D E C~(R) as Z B_n is Gorenstein injective. []
6.2. GORENSTEIN INJECTIVE DIMENSION 143

(6.2.5) G I D C o r o l l a r y . For a complex Y E C(w)(R) the next three conditions

axe equivalent.
(i) Y B(R).
(ii) GidR Y < oe.
(iii) Y C(D)(R ) and GidRY < - i n f Y + d i m R .
Fhrthermore, if Y /3(R), then

GidR Y = sup { - sup V - inf (RnomR(V, Y)) I V Z(R) ^ U ;~ 0}

= sup { - inf (RHomR(J, Y)) I J Co1(R))

Proof. It follows by the Theorem that (ii) implies (i), and (iii) is clearly stronger
than (ii). For Y /3(R) and J injective it follows by Lemma (3.4.13)(c) that

- inf (RHomR(J, Y)) _< - inf Y + dim R,

so by the equivalence of (ii) and (iv) in the Theorem we have GidR Y _< - inf Y +
dim R as wanted. This proves the equivalence of the three conditions.
For Y /3(R) the equalities now follow by the equivalence of (ii), (iii), and
(iv) in the Theorem. []

The next proposition shows that Gorenstein injective dimension is a refinement

of injective dimension.

(6.2.6) P r o p o s i t i o n ( G I D - I D I n e q u a l i t y ) . For every complex Y E C(E) (R)

there is an inequality:

GidR Y _< idn Y,

and equality holds if idn Y < oo.

Proof. The inequality is, as we have already observed, immediate because in-
jective modules are Gorenstein injective. ~ r t h e r m o r e , equality holds if Y is
homologically trivial, so we assume that idR Y = j E Z and choose, by (A.5.2.1),
an R-module T such that j -- - inf (RHomR(T, Y)). Also choose a injective mo-
dule J such that T can be embedded in J. The short exact sequence of modules
0 - T - J -+ C -+ 0 induces, cf. (A.4.8), a long exact sequence of homology
modules:

-" -+ H - j ( R H o m R ( J , Y)) --+ H_j(RHomR(T, Y)) -+

H-(j+I) (RHomR(C, Y)) -+ . . . .

Since, by (A.5.2.1), H_(j+I)(RHomR(C, Y)) = 0 while H_j (RHomR(T, Y)) ~ 0,

we conclude that also H_j(RHomR(J,Y)) is non-zero. This proves, in view of
GID Corollary (6.2.5), that GidR Y >_ j , and hence equality holds. []

By GID Corollary (6.2.5) the next theorem is just a rewrite of the /3 version
(3.2.10).
144 6. G-INJECTIVITY

(6.2.7) G o r e n s t e i n T h e o r e m , G I D V e r s i o n . L e t R be a C o h e n - M a c a u l a y
local ring with residue field k. I f R a d m i t s a dualizing module, then the fol-
lowing are equivalent:
( i) R is Gorenstein.
(ii) GidR k < oo.
(iii) GidR N < oo for all finite R - m o d u l e s N .
(iv) GidR N < oe for all R - m o d u l e s N .
(v) GidR Y < oo for all complexes Y E C(o ) (R). []

In (6.2.8)-(6.2.12) we consider Gorenstein injective dimension for modules: we

rewrite (6.2.4) and (6.2.5) in classical terms of resolutions and Ext modules.

(6.2.8) D e f i n i t i o n . Let N be an R-module. A Gorenstein injective resolution

of N is a complex of Gorenstein injective R-modules,

B = O-~Bo-~B-I-~...~Be~...,

with homology concentrated in degree zero and H0(B) -- Z0B ~ N. T h a t is,

there is a h o m o m o r p h i s m L: N --~/3o such t h a t the sequence

O-+ N - ~ Bo-~ B_I-~...--+ Bt-~...

is exact.
Every module has an injective resolution and hence a Gorenstein injective
one.

(6.2.9) L e m m a . L e t N be an R - m o d u l e . If N is equivalent to B E C~I(R), then

the truncated c o m p l e x

C o B = O -+ c B -~ B _ I --+ B _ 2 -~ . . . -+ B t -+ . . .

is a Gorenstein injective resolution of N .

Proof. Suppose N is equivalent to B E C~I(R), then s u p B = 0, so C o B - B "~

N by (A.1.14.2), and we have an exact sequence of modules:

(t ) 0 -~ N --+ C~ ~ B - 1 --+ B - 2 -~ " " ~ B t -+ " "

Set u = sup {e E Z I B t # 0}, then also the sequence

0 --+ B,, -~ Bu-1 ~ -.. ~ B0 --* CoB -~ 0

is exact. All the modules B u , . . . , t3o are Gorenstein injective, so it follows by

repeated applications of Corollary (6.1.8)(a) t h a t C B is Gorenstein injective, and
therefore C o B is a Gorenstein injective resolution of N , cf. (t). []
6.2. GORENSTEIN INJECTIVE DIMENSION 145

(6.2.10) R e m a r k . It follows by the Lemma and Definition (6.2.2) that an R-mo-

dule N is Gorenstein injective if and only if GidR N < 0. T h a t is,

N is Gorenstein injective ==~ GidR N = 0 V N = 0.

(6.2.11) G I D T h e o r e m for M o d u l e s . Let N be an R - m o d u l e and n E No.

The following are equivalent:
( i) N has a Gorenstein injective resolution of length at most n. That is, there
is an exact sequence of modules 0 -~ N -4 Bo -~ B - 1 -~ ... -4 B - n -4 O,
where Bo, B - I , . . . , B - n axe Gorenstein injective.
(ii) GidR N _< n.
(iii) N E Bo(R) and E x t , ( T , N ) = 0 for all m > n and all T E Zo(R).
(iv) N E Bo( R) and E x t , ( J , N ) = 0 for all m > n and all injective modules J.
(v) In any Gorenstein injective resolution of N ,

O-~ N - ~ Bo - 4 B - 1 ~ . . . - - + B e - 4 ...

the cokernel 2 W - n = Coker(B_n+2 -~ B-n+1) is a Gorenstein injective

module.

Proof. If the sequence 0 --+ N ~ B0 --+ B-1 -~ -.- --+ Be -+ --- is exact, then N
is equivalent to B = 0 -4 B0 -4 B_I -4 "-" -4 Be -4 ---. The complex B belongs
to C~-I(R), and it has Zs -~ N , ZB-I "~= Coker(N -+ B0), and Z_Be "= CB-e+I --
Coker(B_e+2 -~ B-e+1) for g _> 2. In view of the Lemma the equivalence of the
five conditions now follows from Theorem (6.2.4). []

(6.2.12) G I D C o r o l l a r y for M o d u l e s . For an R - m o d u l e N the next three

conditions axe equivalent.
(i) N E Bo(R).
(ii) GidR N < oo.
(iii) GidR N <_ dim R.
Furthermore, ff N E B0(R), then

G i d R N = s u p { m E No I 3 T E Zo(R) : E x t ~ ( T , N ) 0}
= sup {m e No [ 3 J E Col(R) : E x t , ( J , N) 0}.

Proof. Immediate from Corollary (6.2.5). []

The next proposition shows that the Gorenstein injective dimension cannot grow
under localization. In particular, it follows that Np is Gorenstein injective over
Rp if N is Gorenstein injective over R and, as we remarked in (6.1.3), this is not
immediate from the definition.
2Appropriately interpreted for small n as Wo = N and W-1 = Coker(N --4 Bo).
146 6. G-INJECTIVITY

(6.2.13) P r o p o s i t i o n . Let Y E C(r-)(R). For every p E SpecR there is an

inequality:

Gidnp Yp < GidR Y.

Proof. If Y is equivalent to B E C~-I(R), then Yp is equivalent to Bp. It is,

therefore, sufficient to prove that a localized module Np is Gorenstein injective
over Rp if N is Gorenstein injective over R.
Let N be a Gorenstein injective R-module, and set d = dim Rp. It follows
from the definitions in (6.1.1) that there is an exact sequence

(t) 0 ~ K ~ - . . ~ Q ~ N ~ 0 ,

where the modules Id,... ,I1 are injective. Since N and the injective mo-
dules all belong to the Bass class, it follows by repeated applications of Corol-
lary (3.4.10)(a) that also K E 13o(R). Localizing at p we get an exact sequence

($) 0 --, K p ~ (Id)p --~ "" ~ (I2)p ~ (I~)~ ~ Np ~ O,

where the modules (It)p are injective over Rp, while Np and Kp belong to B(Rp),
cf. Observation (3.2.7). From GID Corollary (6.2.12) it follows that GidRp Kp <
d, and since (~:) is exact it follows by GID Theorem (6.2.11) that Np is Gorenstein
injective. []

We will now use Foxby equivalence to prove a formula for Gorenstein injective
dimension like that of Bass' for injective dimension (see page 13).

(6.2.14) L e m m a . If Y E B(R) and U E P(f)(R), then

inf (RHomR(U, Y)) = inf (RHomR(D ~ U, Y)).

Proof. The first equality in the calculation below follows as Y E B(R); it also
uses commutativity (A.4.19). The second equality follows by tensor evalua-
tion (A.4.23) as U E P(f)(R), the third by Lemma (3.4.3)(a), the fourth by
adjointness (A.4.21), and the last one by commutativity.

inf (RHomn(U, Y)) = inf (RHomR(U, RHomn(D, Y) ~ D))

= inf (RHomn(U, RHomn(D, Y)) ~ D)
= inf (RHomR(U, RHomR(D, Y)))
= inf (RHomR(U ~ D, Y))
= inf (RHomR(D ~ U, Y)). []
6.2. GORENSTEIN INJECTIVE DIMENSION 147

(6.2.15) T h e o r e m (Bass F o r m u l a for G I D ) . If Y is a complex with finite

homology and finite Gorenstein injective dimension, i.e., Y E B (f) (R), then

GidR Y = depth R - inf Y.

In particular,

Gida N = depth R

for finite modules N # 0 of finite Gorenstein injective dimension.

Proof. By GID Corollary (6.2.5) we have

GidR Y _< dim R - inf Y = depth R - inf Y

as R is Cohen-Macaulay. To prove the opposite inequality, let x l , . . . , Xd be

a maximal R-sequence and set T = R / ( x l , . . . , X d ) , then T belongs to P f ( R )
and has pd R T = depth R. By Theorem (3.4.6) the module D R T represents
D L T, and D R T E Zo(R) by Foxby equivalence (3.4.11). Now the inequality
in demand follows by (A.7.8), Lemma (6.2.14), and GID Corollary (6.2.5):

depth R - inf Y = pd a T - inf Y

= - inf (RHomR(T, Y))
= - inf (RHomR(D R T, Y))
_< GidR Y. []

(6.2.16) R e m a r k . We do not know if the existence of a finite R-module N ~ 0

of finite Gorenstein injective dimension has any implications for the ring. By
the celebrated Bass conjecture a local ring must be Cohen-Macaulay to accom-
modate a non-trivial finite module of finite injective dimension, so the question
seems to be: If among the non-trivial finite modules there is one of finite Goren-
stein injective dimension, is there then also one of finite injective dimension?

Notes

The Bass formula for Gorenstein injective dimension, Theorem (6.2.15), was
proved over Gorenstein rings by Enochs and Jenda [26, Theorem 4.3], see also [30,
Corollary 4.11].

If Y E C(D) (R) has finite injective dimension, then

idR Y = sup (depth Rp - widthRp Yp I P E Spec R).

This formula was proved for modules by Chouinard [14, Corollary 3.1] and ex-
tended to complexes by Yassemi [63, Theorem 2.10]; it holds over Noetherian
rings in general.
148 6. G-INJECTIVITY

One must ask if a similar formula holds for the Gorenstein injective dimen-
sion. T h a t is, (to be modest) if R is a Cohen-Macaulay local ring with a dualizing
module, is then

Gidn Y = sup {depth Rp - widthRp Yp [ p E Spec R}

for all complexes Y in the Bass class? For complexes with finite homology the
answer is positive: by (A.6.3.2) and Cohen-Macaulayness of R we have

sup {depth Rp - widthnp Yp [ p e Spec R} = sup {dim Rp - inf Yp [ p Spec R}

= dim R - inf Y
= depth R - inf Y,

so the desired formula was established in Theorem (6.2.15). In general, however,

the answer is not yet known. Enochs and Jenda have in [30] established a couple
of special cases where Gidn N = depth R - widthn N for a non-finite module N;
we return briefly to this point in Observation (6.3.7).

6.3 G-injective versus G-flat Dimension

In this section we establish the "G-parallels" of Ishikawa's formulas for flat and
injective dimension.

(6.3.1) 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.

We start by rewriting Gorenstein Theorems (3.3.5) and (3.4.12) in terms of

finiteness of Gorenstein dimensions.

(6.3.2) G o r e n s t e i n T h e o r e m , G F D / G I D V e r s i o n . Let R be a Cohen-

Macaulay local ring. If D is a dualizing module for R, then the following are
equivalent:
(i) R is Gorenstein.
(ii) GidR R < oo.
(ii') G f d n D < co.
(iii) Gidn N < c~ and fdn N < oo for some R-module N of finite depth.
(iii') idn M < oe and Gfdu M < oo for some R-module M of finite depth.
(iv) A homologically bounded complex X has finite Gorenstein fiat dimen-
sion if and only if it has finite Gorenstein injective dimension; that is,
G f d n X < oo ~ G i d n X < oo.

Proof. In view of GID Corollary (6.2.5) and GFD Corollary (5.2.6) the theorem
is just a reformulation of the the special complexes version (3.3.5) and the special
modules version (3.4.12). []
6.3. G-INJECTIVE VERSUS (]-FLAT DIMENSION 149

The next theorem is a parallel to Ishikawa's [42, Theorem 1.4].

(6.3.3) T h e o r e m . Let E be an injective R-module. For every X E C(n)(R)

there is an inequality:
Gidn(Homn(X, E)) <_ Gfdn X,
and equality holds if E is faithfully injective.
Proof. The inequality is trivial if X is not of finite Gorenstein flat dimension, so
we assume that X E A(R). Then, by Lemma (3.2.9)(a), H o m R ( X , E ) belongs
to the Bass class, and for every injective module J we have
- inf ( R H o m n ( J , Homn(X, E))) = - inf ( R H o m n ( J L X, E))
(9) < sup ( J L X),

by adjointness (A.4.21) and (A.5.2.1). The desired equality now follows by

GID Corollary (6.2.5) and GFD Corollary (5.2.6).
If E is faithfully injective then, again by Lemma (3.2.9)(a), X belongs to
A(R) if and only if Homn(X, E) is in B(R); that is, the two dimensions are
simultaneously finite. Furthermore, equality holds in (9), cf. (A.4.10), so the
desired equality follows by Corollaries (6.2.5) and (5.2.6). []

As an immediate corollary we obtain a special case of Theorem (6.4.2):

(6.3.4) C o r o l l a r y . An R-module M is Gorenstein fiat if and only if Homn(M, E)

is Gorenstein injective for every injective R-module E. []

The next result is supposed to be the dual of Theorem (6.3.3) - - i.e., the G -
parallel of Ishikawa's [42, Theorem 1.5] - - but, alas, it is not quite so. In (6.3.8)
and (6.3.9) we work out a couple of special cases where the duality works as it
should or, rather, as one could hope that it would.

(6.3.5) P r o p o s i t i o n . Let E be an injective R-module. For every Y E C(z)(R)

there is an inequality:
Gfda(Homn(Y, E)) _< Gidn Y,
and i r E is faithfully injective, then the two dimensions are simultaneously finite;
that is,
G f d a ( H o m a ( Y , E ) ) < oo ~ G i d a Y < oo.
Proof. The inequality is trivial if Y is not of finite Gorenstein injeetive dimen-
sion, so we assume that Y E B(R). Then Homn(Y, E) belongs to the Auslander
class, cf. Lemma (3.2.9)(b), and for every finite R-module T we have
sup (T ~ Homn(Y, E)) = sup ( R H o m n ( R H o m n ( T , Y), E))
<_ - inf (RHomR(T, r))
150 6. G-INJECTIVITY

by Hom evaluation (A.4.24) and (A.4.6.1). The desired inequality now follows
by GID Corollary (6.2.5) and (TI;) in Theorem (5.4.6).
If E is faithfully injective, then, again by Lemma (3.2.9)(b), we have
HomR(Y,E) A(R) if and only if Y B(R); that is, the two dimensions
axe simultaneously finite. []

(6.3.6) Corollary. ff N is a Gorenstein injective R-module, then Homn(N, E)

is Gorenstein fiat for every injective R-module E. []

(6.3.7) Observation. Let E = ER(k) be the injective hull of the residue

field. For any R-module N it then follows by (A.6.4) that widthRN =
depthR(HomR(N, E)), this is [63, Lemma 2.2], so widthR N < dim R = depth R.
Now, if N ~ 0 is Gorenstein injective, then HomR(N, E) ~ 0 is Gorenstein flat
by the Corollary, so

0 -- GfdR(HOmR(N, E)) > depth R - depthR(HomR(N, E))

= depth R - widthR N

by the AB formula (5.4.9). Thus,

(6.3.7.1) widthR N = depth R

for every Gorenstein injective R-module N.

(6.3.8) P r o p o s i t i o n . Let E be a faithfully injective R-module. For every com-

plex Y E C ~ (R) there is an equality:

GfdR(HomR(Y, E)) = GidR Y.

Proof. By Proposition (6.3.5) it is sufficient to prove that GidRY <

Gfdn(Homn(Y,E)) for Y E B(f)(R). Let x l , . . . ,Xd be a maximal R-sequen-
ce and set T = R / ( x l , . . . , X d ) , then it follows by Theorem (5.4.6), (A.4.10),
(A.7.8), and Theorem (6.2.15) that

GfdR(HomR(Y, E)) > sup (R/(Xl,..., Xd) ~ HomR(Y, E))

= sup (RHomR(RHomR(R/(Xl,..., x~), Y), E))
- - inf (RHomR(R/(xI,..., Xd), Y))
= pd R R / ( x l , . . . , x d ) - infY
= depth R - inf Y
= GidR Y. []
6.3. G - I N J E C T I V E VERSUS G - F L A T DIMENSION 151

(6.3.9) Observation. Let X E C(-~)(R), and let E and E' be faithfully injective
R-modules. For T E 7~f0(R) we have
sup (T ~ X) = - inf (RHomn(T ~ X, E'))
-- - inf (RHomn(T, R H o m n ( Z , E')))
= sup (RHoma(RHomR(T, HomR(X, E')), E))
= sup (T ~ HomR(HOmR(X, E'), E))
by adjointness (A.4.21), (A.4.10), and tensor evaluation (A.4.23). By (T~) in
Theorem (5.4.6) and Lemma (3.2.9) it now follows that
(t) GfdR X = GfdR(HOmn(HomR(X, E'), E)).
Set Y = HomR(X,E'), then Y E C(r)(R), and by (t), Proposition (6.3.5), and
Theorem (6.3.3) we have
GfdR X = GfdR(HomR(Y, E)) < GidR Y < GfdR X.
That is, if Y E C(E)(R) is equivalent to a complex HomR(X,E'), where
X E .A(R) and E' is faithfully injective, then

GfdR(HomR(Y, E)) = Gidn Y

for every faithfully injective R-module E.

(6.3.10) T h e o r e m . If Y is a complex of finite Gorenstein injective dimension,

i.e., Y E B(R), then the following numbers are equal:

(D) Gidn Y,
(EI) sup{-supU-inf(RHomn(U,Y))[UEZ(R) A U~0}, and
(EE) sup {-- inf (RHomn(ER(R/p), Y)) [ p E Spec R}.

Furthermore, if Y ~_ Homn(X, E), where X E A(R) and E is a faithfully

injective R-module, then also the next three numbers are equal, and equal to
those above.

(m,0 sup { - inf (RHomn(T, Y)) [ T E Zf0(R)},

(EF) sup {-- sup U - inf (RHomn(U, Y)) [ U E ~-(R) A U ~ 0}, and
(E~) sup { - inf ( R H o m n ( R / ( x ) , Y)) [x = x l , . . . , xt is an R-sequence}.
Proof. It was shown in GID Corollary (6.2.5) that the numbers (D) and (m) axe
equal. Furthermore, every injective R-module is a direct sum of indecomposable
injectives, i.e., modules of the form En(R/p), so in view of the functorial iso-
morphism

H o m , ( H E p , - ) ~ H HOmR(Ep,-),
p p
152 6. G-INJECTIVITY

it follows, still by Corollary (6.2.5), that (D) ---- (EE).

Now, assume that Y ~_ Homn(X, E), where X A(R) and E is a faithfully
injective R-module. For U C(D)(R) we then have

- inf (RHomR(U, Y)) = - inf (RHomR(U, HomR(X, E)))

(t) = - inf (RHomR(U L X, E))
= sup (v L X)
by adjointness (A.4.21) and (A.4.10). Since (D) = GfdR X by Theorem (6.3.3), it
is sufficient to prove that the numbers (m~), (EF), and (Ex) are equal to GfdR X;
and in view of (t) this is immediate from Theorem (5.4.6). []

(6.3.11) Corollary. If N is a module of finite Gorenstein injective dimension,

i.e., N E Bo(R), then the foUowing numbers axe equal:
(D) GidR N,
(too) sup {m e No ] 3 T Z0(R) : E x t , ( T , N) # 0}, and
(EE) sup {m No I 3 p Spec R: Ext~(ER(R/p), N) # 0}.

Fhrthermore, if N ~- HomR(M,E), where M Ao(R) and E is a faithfully

injective R-module, then also the next three numbers are equal, and equal to
those above.

(Eg) sup {m C 3 T e zf0(R) : E x t , ( T , N) ~ 0},

(EFo) sup {m C 3 T e ~0(R) : E x t ~ ( T , N ) ~ 0}, and
(Ex) sup {m C No Ext~(R/(z), N) ~ 0 for some R-seq. z = Xl,..., xt}.

Notes

The equality (6.3.7.1 was proved by Enochs and Jenda [30, Lemma 4.1] under
slightly different conditions.

Theorem (6.3.2) - - the GFD/GID version of the Gorenstein Theorem - -

strengthens the PD/ID version (see page 6), and it is natural to ask if an even
stronger version exists: does the existence of an R-module (or complex) of finite
depth, finite Gorenstein flat dimension, and finite Gorenstein injective dimension
imply that R is Gorenstein? The answer is not known (to the author).

6.4 Exercises in S t a b i l i t y
In the previous section we used the functorial characterizations of Gorenstein
flat and injective dimensions to prove a couple of stability results. While this
approach is fast, it also has a serious drawback: it only works over certain Cohen-
Macaulay rings. Some of the results, however, hold over general Noetherian
rings; and in this section we show how to prove them by working with resolutions.
6.4. EXERCISES IN STABILITY 153

The main theorem of this section is (6.4.2): a module is Gorenstein fiat if and
only if the dual with respect to every injective module is Gorenstein injective; it
is the general version of Corollary (6.3.4).
While detailed proofs are provided for the first three results, the rest of the
section can be taken as a series of exercises; the proofs are, at any rate, reduced
to hints. This is particularly true for the final (6.4.13) which is only interesting
from a "derived category point of view" and should be proved by within this
framework.

(6.4.1) P r o p o s i t i o n . Let E be an injective R-module. If F E CF(R) is a com-

plete fiat resolution, then Homn(F, E) is a complete injective resolution; and
the converse holds if E is faithfully injective.

Proof. If F E CF(R) and E is injective, then HomR(F,E) is a complex

of injective modules. Furthermore, if F is homologically trivial, then so is
HomR(F, E); and the converse holds if E is faithfully injective. For every (in-
jective) module J we have

HomR(J R F, E) -~ HomR(J, HomR (F, E))

by adjointness (A.2.8), so if J R F is homologically trivial, then so is

HomR(J, HomR(F, E)); and, again, the converse holds if E is faithfully injec-
tive. []

(6.4.2) T h e o r e m . The following are equivalent for an R-module M:

(i) M is Gorenstein Nat.
(ii) HomR (M, E) is Gorenstein injective for some faithfully injective
R-module E.
(iii) HomR(M, E) is Gorenstein injective for every injective R-module E.

Proof. It is evident that (iii) implies (ii); this leaves us two implications to
prove.
(i) ~ (iii): Let F be a complete fiat resolution with C F - M, and let E be
injective. Then, by the Proposition, HomR(F, E) is a complete injective resolu-
tion, and ~Homa(F,E)
"~0 _~ HomR(M, E) by Lemma (4.1.1) (b) , so HomR(M, E) is
Gorenstein injective as wanted.
(ii) ~ (i): We assume that E is a faithfully injective R-module such that
HomR(M, E) is Gorenstein injective, and we set out to construct a complete fiat
resolution F with C F ~ M. If we can construct a short exact sequence

($) O --~ M --~ F_ I --~ C-1 -~0,

where F-1 is flat and C-1 is a module with the same property as M (that
is, H o m R ( C _ I , E ) is Gorenstein injective), then the right half of a complex
F E CF(R) can be constructed recursively. The left half of F we get for free
154 6. G-INJECTIV1TY

by taking a flat resolution of M, and a complex F established this way is ho-

mologically trivial with CoF ~ M. Consider the homologically trivial complex
HomR(F, E) of injective modules. By Lemma (4.1.1)(b) we have

(*) zHomR(F,E) ~.~HomR(CFt,E),

7Homa(F,E) is a Gorenstein injective module, because

so for g > 0 the kernel ~t
c F t = C-e is a module with the same property as M. Let J be an injective
module; for e > 0 we then have

E x t , ( J , Z~ mR(f'E)) = 0,

cf. Proposition (6.1.5), and for g < 0 we have

E x t , ( J , ZtHmR(F'E)) = E x t ~ - l ( j , ZoHOmR(F,E))
= Extl-e(J, UomR(M, E)) = 0

by Lemma (4.1.6)(c), (4.1.1)(b), and the assumption on M. Thus, it follows by

(4.1.6)(c) that HomR(J, HomR(F, E)) is homologically trivial for every injective
module J; that is, HomR(F, E) is a complete injective resolution and, therefore,
F is a complete flat resolution by Proposition (6.4.1). To prove the theorem it
is now sufficient to construct the short exact sequence ($).
The module M v = HomR(M, E) is Gorenstein injective by assumption, so
by definition we have a short exact sequence

O~ Z ~ I-~ Mv ~O,

where I is injective. Applying the exact functor _v = H o m R ( - , E ) , we get

another short exact sequence

0 -+ -h~/vv v) I v --). Z v --~ O.

The canonical map (iE: M ~ M vv is injective because E is faithfully injective,

so we have an injective map u = 0v5 E from M into the flat module I v. Let
: M --~ F-1 be a flat preenvelope of M, then is injective by Lemma (4.3.3),
so with C-1 = Coker we have an exact sequence

(,) O--+ M --~F _ 1 ~ C_1--+ 0.

We now want to prove that HomR(C_I, E) is Gorenstein injective. From (*) we

get a short exact sequence

0 ---HomR(C_I,E) --> H o m R ( F _ I , E ) HomR(,E)) H o m R ( M , E ) -~ 0;

where the module HomR(F_I, E) is injective and HomR(M, E) is Gorenstein in-

jective by assumption. To prove that also HomR(C-1, E) is Gorenstein injective
6.4. EXERCISES IN STABILITY 155

it is, by Corollary (6.1.8), sufficient to see that E x t , ( J , HomR(C-1, E)) = 0 for

all injective modules J. Let J be an injective module, E x t , ( J , HomR(C-1, E))
vanishes if and only if the map

HomR(J, HomR(, E)) : HomR(J, HomR(F-1, E)) ~ Homn(J, HomR(M, E))

is surjective (Ext,(J, HomR(F_I, E)) = 0 because HomR(F-1, E) is injective),

so we consider the commutative diagram

HomR (J, HomR(F_x, E)) Hom,,~(J,Homn(,E))~, Homn(J, Homn(M, E))

~~JF--IE ~gJME
HomR(F-1, HomR (J, E)) Homn(,Homn(J,E))) HomR(M, HomR(J, E))

The module HomR(J,E) is flat, and is a flat preenvelope of M, so

the map HomR(,HomR(J,E)) is surjective, cf. (4.3.2), and hence so is
HomR(J, HomR(, E)). This concludes the proof. []

(6.4.3) T h e o r e m . Let X E C(7)(R); if U is a complex of finite injective dimen-

sion, i.e., U E Z(R), then

GidR(RHomR(X, U)) <_ GfdR X + idn U.

Proof. We can assume that U is homologically non-trivial, otherwise the in-

equality is trivial; and we set s = sup U and i = idR U. The inequality is also
trivial if X is homologically trivial or not of finite Gorenstein flat dimension,
so we assume that X ~ 0 and set g -- GfdRX E Z. We can now choose a
complex A E C~F(R) which is equivalent to X and has At = 0 for t > g; we set
v = inf {e E Z I At ~ 0}. By (A.5.1) U is equivalent to a complex I of injective
modules concentrated in degrees s , . . . , - i . Now, RHomR(X, U) is represented
by the complex HomR(A, I) with
g
(t) HomR(A,I)t = H HomR(Ap,Ip+t) = ~ ) H o m n ( A p , Ip+e).
pEZ p=v

The modules HomR(Ap, Ip+t) are Gorenstein injective by Theorem (6.4.2), and
finite sums of Gorenstein injective modules are Gorenstein injective, cf. Corol-
lary (6.1.8)(c), so H o m R ( A , I ) E CGI(R). Furthermore, it is easy to see that
HomR(A, I) is bounded: by (f) we have HomR(A, I)t = 0 for t > s - v; and if
[ < - ( i + g), then either p > g or p + [ < g + ~ < - i , so also for ~ < - ( i + g )
is Homn(A,I)e = 0. That is, HomR(A,I) is a bounded complex of Goren-
stein injective modules concentrated in degrees at least - ( i + g) and, therefore,
GidR(RHomR(X, U)) _< i + g = Gfdn X + idR U as wanted. []
156 6. G-INJECTIVITY

(6.4.4) P r o p o s i t i o n . /f F E CF(R) is a complete fiat resolution, then so is

F R F' for every fiat R-module F'.
In particular: if M is Gorenstein fiat, then M R F' is the same for every
fiat module F'.

Proof. Use the definitions and associativity. []

(6.4.5) T h e o r e m . Let X E C(2)(R); if U is a complex of finite fiat dimension,

i.e., U E Y ( R ) , then

Gfdn(X ~ U) _< Gfdn X + fdn U.

Proof. Apply the technique from the proof of Theorem (6.4.3); only this time
use Proposition (6.4.4) and Corollary (5.1.9). []

Note that (6.4.5) generalizes (2.3.17)(b).

(6.4.6) P r o p o s i t i o n . / f P E CP(R) is a complete projective resolution, then so

is Homn(P', P) for every finite projective R-module P'.
In particular: if M is Gorenstein projective, then Homn(P', M) is the same
for every finite projective module P'.
Proof. Use the definitions and Hom evaluation. []

The next result generalizes (2.3.17)(a).

(6.4.7) T h e o r e m . Let X E C(-~)(R); if U is a complex with finite homology and

finite projective dimension, i.e., U E P(f) (R), then

GpdR(RHomn(U, X)) _< Gpd n X - inf U.

Proof. Apply the technique from the proof of Theorem (6.4.3); only this time
use Proposition (6.4.6) and Corollary (4.3.5). []

(6.4.8) P r o p o s i t i o n . lf I E CI(R) is a complete injective resolution, then so is

HomR(P', I) for every finite projective R-module P'.
In particular: if N is Gorenstein injective, then HomR(P', N) is the same for
every finite projective module P'.
Proof. Use the definitions and swap. []

(6.4.9) T h e o r e m . Let Y E C(z)(R); if U is a complex with finite homology and

finite projective dimension, i.e., U E P(f)(R), then

GidR(RHomR(U, Y)) _< GidR Y + pd a U.

Proof. Apply the technique from the proof of Theorem (6.4.3); only this time
use Proposition (6.4.8) and Corollary (6.1.8). []
6.4. EXERCISES IN STABILITY 157

(6.4.10) T h e o r e m . Let R be a Cohen-Macaulay local ring with a dualizing mo-

dule, and let Y C(E)(R ). If U is a complex of finite injective dimension, i.e.,
U Z(R), then

Gfdn(RHomn(Y, U)) <_ Gidn Y + sup U.

Proof. Apply the technique from the proof of (6.4.3), but use Corollary (6.3.6)
instead of Theorem (6.4.2) and Corollary (5.1.9) instead of (6.1.8). See also
(6.4.12) below. []

(6.4.11) R e m a r k . We do not know if it is, at all, true that an R-module

N is Gorenstein injective if and only if Homn(N, E) is Gorenstein flat for
every injective R-module E; not even if R is local Cohen-Macaulay with
a dualizing module, cf. Corollary (6.3.6). In particular, it is not obvious
that the technique from the proof of Theorem (6.4.2) can be used to solve
the problem. One of the obstructions seems to be that it is not clear
whether H o m R ( y , E ) : H o m R ( N , E ) ~ H o m n ( I , E ) is a flat preenvelope of
HomR(N,E) whenever ~: I ~ N is an injective precover of N and E is in-
jective. The dual is, however, true: if : M--4 F is a flat preenvelope of
M, then Homn(, E): Homn(F, E) ---4 Homn(M, E) is an injective precover
of Homn(M, E) for every injective module E. This can be deduced from the
closing argument in the proof of (6.4.2).

(6.4.12) R e m a r k . A different proof of Theorem (6.4.10) is available: it is

possible (and using derived category methods it is even easy) to prove that
RHomR(Y, U) belongs to the Auslander class when Y E B(R) and U Z(R).
For every T 7~f0(R) we have

sup (T n RHomR(Y, U)) = sup (RHomn(RHomn(T, Y), U))

_< sup U - inf (RHomn(T, Y))
= - inf ( R H o m n ( D ~ T, Y)) + sup U
= - inf (RHomR(D n T, Y)) + sup U
<_ Gidn Y + sup U

by Horn evaluation (A.4.24), (A.4.6.1), Lemma (6.2.14), Theorems (3.4.6) and

(3.4.11), and GID Corollary (6.2.5), so

Gfdn(RHomn(Y, U)) <_ Gidn Y + sup U

by Theorem (5.4.6).

The reader is invited to apply similar methods reestablish special cases of The-
orems (6.4.3), (6.4.5), (6.4.7), and (6.4.9) in the following form:
158 6. G-INJECTIVITY

(6.4.13) T h e o r e m (Stability of A u s l a n d e r Categories). Let R be a Cohen-

Macaulay local ring with a dualizing module. The following hold for complexes
A E A(R), B E B(R), F E .T(R), I E Z(R), and P E ~o(f)(R):
* RHomR(A,I) e B(R) with

Gidn(RHomn(A, I)) _< Gfdn A + idn I;

RHomn(B,I) E A(R) with

Gfda(RHomn(B, I)) _< Gidn B + sup I;

A L F E A(R) with

Gfdn(A ~ F) _< Gfdn A + fdn F;

RHomn(P, A) E A(R) with

GpdR(RHomn(P,A)) _< G p d n A - infP; and

RHomn(P,B) E B(R) with

Gidn(RHomR(P, B)) _< Gidn B + pd n P.

Notes
A special case of Theorem (6.4.2) follows from [27, Lemma 3.4]: a module M
over a Gorenstein ring is Gorenstein flat if and only if the Pontryagin dual,
Homz(M, Q/Z), is Gorenstein injective. Special cases of some of the other sta-
bility results can also be found in [27].
The stability results for Auslander categories, Theorem (6.4.13), are selected
special cases taken from a series of unpublished results by Foxby.