UNIQUENESS OF DECOMPOSITION , FACTORISATIONS , G-GROUPS AND POLYNOMIALS

In this article, we present the classical Krull-Schmidt Theorem for groups, its statement for modules due to Azumaya, and much more modern variations on the theme, like the so-called weak Krull-Schmidt Theorem, which holds for some particular classes of modules. Also, direct product of modules is considered. We present some properties of the category of G-groups, a category in which Remak’s results about the Krull-Schmidt Theorem for groups can be better understood. In the last section, direct-sum decompositions and factorisations in other algebraic structures are considered. Mathematics Subject Classification (2010): 16D70, 16D90, 18E05


Introduction
In this paper, we mainly present the classical Krull-Schmidt Theorem for groups and modules and some of its weak versions.According to the classical Krull-Schmidt Theorem for modules, any module of finite composition length decomposes as a direct sum of indecomposable modules in an essentially unique way, that is, unique up to isomorphism of the indecomposable summands and a permutation of the summands.In Section 2, we present the historical background of that theorem.Section 3 is devoted to commutative monoids, which provide the best algebraic tool to describe finite direct-sum decompositions of modules.

ALBERTO FACCHINI AND SERAP S ¸AHINKAYA
In 1975, Warfield proved that every finitely presented module over a serial ring is a finite direct sum of uniserial modules and posed a problem, essentially asking whether the Krull-Schimdt Theorem holds for finite direct sums of uniserial modules.More precisely, he asked whether the direct-sum decomposition of a finitely presented module over a serial ring into uniserial summands is unique up to isomorphism [37].The negative answer to this question was given by the first author in 1996 [15].He showed that even though the Krull-Schmidt Theorem does not hold for serial modules, a weak version of it holds not only for serial modules, but also for other classes of modules.Some of these classes of modules will be presented in Section 4.
It is natural to ask what happens when one considers arbitrary direct products of modules instead of arbitrary direct sums.In Section 5, we collect some results about direct product of modules and uniqueness of decomposition.In that section, we also present some properties of the category of G-groups, a framework in which the existence of the central automorphism in the Krull-Schmidt Theorem proved by Remak has a natural explanation.
All these results are strongly related to factorisation in a very general sense.
Thus we present some results concerning factorisation into irreducible polynomials of non-commutative polynomials with coefficients in Z, and also factorization of commutative polynomials with non-negative integral coefficients.

Historical background of the Krull-Schmidt-Remak-Azumaya Theorem
In this survey, rings will be associative rings R with an identity, and modules will be unital right R-modules, unless otherwise stated.
In 1879, Frobenius and Stickelberger proved that any finite abelian group is a direct product of cyclic groups whose orders are powers of primes, and this powers of primes are uniquely determined by the group [23].The "classical Krull-Schmidt Theorem for finite groups" was first stated in 1909 by Wedderburn [27], who published the following theorem.
The proof given by Wedderburn is not entirely convincing.It was Robert Erich Remak [32] who proved in his PhD dissertation (1911) that any two direct-product Otto Yulyevich Schmidt [33] then gave a simplified proof of Remak's results.These results were transfered to modules of finite length by Krull and Schimdt, getting a theorem that, in modern terminology, can be stated as follows.
Theorem 2.3.(The Krull-Schmidt Theorem) Let R be a ring and M be a module of finite length.Then there exists a decomposition Arbitrary modules do not decompose in an "essentially unique" way in general.
Here is an example.
Example 2.4.[17] Let R be a commutative integral domain with at least two distinct maximal ideals M and N that are not principal ideals.Then the morphism M ⊕ N → R, defined by (x, y) → x + y, is an R-module epimorphism, which necessarily splits because R is a projective R-module.The kernel of this morphism is isomorphic to M ∩ N , so that there is a splitting short exact sequence 0 → M and N are not principal ideals, so that they are not isomorphic to R. Therefore the two direct-sum decompositions are essentially different.Notice that R is an Rmodule of Goldie dimension 1 since R is a commutative integral domain.Thus R and its submodules M , N and M ∩ N are all modules of Goldie dimension 1, which proves that the two essentially different direct-sum decompositions are direct-sum decompositions into indecomposables.
Let Ω be a set.Recall that an Ω-group is a pair (H, ϕ), where H is a group and ϕ : Ω → End(H) is a mapping.Krull [25] extended the results known for ALBERTO FACCHINI AND SERAP S ¸AHINKAYA groups to the case of abelian operator groups with the ascending and descending chain conditions (operator groups = Ω-groups).Notice that if H is abelian, then End(H) is a ring, so that the mapping ϕ : Ω → End(H) extends uniquely to a ring homomorphism Z Ω → End(H), where Z Ω denotes the free ring with free set Ω of generators (=ring of non-commutative polynomials with coefficients in Z in the set of non-commuting indeterminates Ω).Thus abelian Ω-groups are exactly left Z Ω -modules.The theory was subsequently further deepened by Schmidt [34].
Øystein Ore (Oslo, 1899-1968) unified the proofs from various categories: groups, abelian operator groups, rings and algebras.He showed that the theorem of Wedderburn holds for modular lattices with descending and ascending chain conditions.
where all the submodules M i (i ∈ I) and N j (j ∈ J) are indecomposable, then there exists a bijection ϕ : I → J such that M i ∼ = N ϕ(i) for every i ∈ I.
After this brief historical introduction, our aim now is to describe direct-sum decompositions of a module M R as a finite direct sum Several behaviours are possible.For instance: • There can be uniqueness of direct-sum decomposition into indecomposables.This is the case described, for example, for modules that are direct sums of modules with local endomorphism rings (Krull-Schmidt-Remak-Azumaya).
• There are modules with a direct-sum decomposition into indecomposables, such that this decomposition is not unique in the sense of the Krull-Schmidt-Remak-Azumaya Theorem, but there are only finitely many such direct-sum decompositions up to isomorphism.This happens, for instance, for torsionfree abelian groups of finite rank [26].
• For some classes of modules, direct-sum decompositions into indecomposables are not unique, but they enjoy some kind of regularity.
• But, in general, there is no direct-sum decomposition into indecomposables, and no uniqueness as in Example 2.4.

The reduced monoid V (C)
The best way to describe finite direct-sum decompositions of a module M R is by making use of commutative monoids.A commutative monoid is a semigroup with a binary operation that is associative, commutative and has an identity element.All the monoids in this paper will be commutative and additive, that is, their operation will be denoted as an addition +, and their identity element will be denoted by 0.
For a commutative monoid M , let U (M ) denote the group of all a ∈ M with an opposite −a in M .A commutative additive monoid M is reduced if x, t ∈ M and For every monoid M , the monoid M red := M/U (M ), whose elements are the cosets m + U (M ), is reduced.
We will look at classes of right R-modules as full subcategories of the category Mod-R of all right R-modules.Let C be a category and let V (C) denote a skeleton of C, that is, a class of representatives of the objects of C modulo isomorphism.
In order to avoid set-theoretical problems, in this survey we could consider only skeletally small categories, that is, categories C in which the class Ob(C) contains a set of representatives of the objects up to isomorphism.Equivalently, a category C is skeletally small if it has a skeleton whose class of objects is a set.For every object A in C, there is a unique object A in V (C) isomorphic to A. Thus there is a mapping Ob(C) → V (C), A → A , that associates to every object A of C the unique object A in V (C) isomorphic to A. Assume that a product A × B exists in C for every pair A, B of objects of C. Define an addition + in V (C) by If the category C is not skeletally small, we have that its skeleton V (C) is a class that is not a set.This class becomes therefore a large monoid, that is, a monoid that is not a set but a class, like in the next lemma.
Lemma 3.1.Let C be a category with a terminal object and in which a product It is easy to prove that the Krull-Schmidt property holds in the additive category C if and only if the monoid V (C) is a free monoid, that is, isomorphic to the direct sum N (I) for some class I.
An element u of a commutative monoid M is called an order-unit in M if, for every x ∈ M , there exists an integer n ≥ 0 such that x ≤ nu.Here ≤ denotes the algebraic preorder on the commutative monoid M , that is, the reflexive and transitive relation ≤ on M defined, for every x, y ∈ M , by x ≤ y if there exists z ∈ M with x + z = y.Hence u is an order-unit in M if and only if, for every x ∈ M , there exist an integer n ≥ 0 and an element y ∈ M such that x + y = nu.
It is now easy to define the category of commutative monoids with order-unit.Its objects are the pairs (M, u), where M is any commutative monoid and u ∈ M is an order-unit.Its morphisms f : (M, u) → (M , u ) are the monoid morphisms For any ring R, let C be the class of all finitely generated projective right Rmodules.We will denote by V (R) the monoid V (C).Then V (R) is a commutative reduced monoid with order-unit R R .
The following theorem was first proved by Bergman for finitely generated monoids with order-unit [7, Theorems 6.2 and 6.4].Then it was extended by Bergman and Dicks to arbitrary monoids with order-unit [8, p. 315].Recall that a ring R is hereditary if all its right ideals and all its left ideals are projective modules.
Theorem 3.2.[7,8] Let k be a field and let (M, u) be a commutative reduced monoid with order-unit.Then there exists a hereditary k-algebra R such that (M, u) and (V (R), R R ) are isomorphic monoids with order-unit.
Corollary 3.3.Let k be a field and let M be a commutative reduced monoid.Then there exists a class C of finitely generated projective right modules over a hereditary k-algebra R such that M ∼ = V (C).

Weak Krull-Schmidt Theorem
Let R be any ring.In this section, we will see what happens for some very special classes of right R-modules that are not direct sums of modules M i with The endomorphism ring of a uniserial module has at most two maximal right (left) ideals, as the following theorem shows.
[15] Let U R be a non-zero uniserial module over a ring R, E := End(U R ) its endomorphism ring, In 1975, Warfield proved that every finitely presented module over a serial ring is a finite direct sum of uniserial modules, and asked whether the direct decomposition of a finitely presented module into uniserial summands is unique up to isomorphism [37].Warfield's question was answered completely in [15], by showing that, although there exist serial rings for which the Krull-Schmidt Theorem does not hold for finitely presented modules, it is possible to prove a weak form of the ] e for every i = 1, 2, . . ., n.
Remark 4.3.In [21], Zahra Nazemian and the first author have studied the fac- cally and the right modules R/A 1 , . . ., R/A n uniserial.The main example of such a factorisation is the factorization into powers of prime ideals for non-zero ideals of a Dedekind domain R.
Theorem 4.2 must be modified if we want it to hold for infinite families of uniserial modules [13,Theorem 4.9].This was done by Příhoda, who proved the converse of [13,Theorem 4.7] in [29].First recall a definition.A right R-module U R is said to be quasismall if, for every family and only if there exist a bijection σ : I → J and a bijection τ : If a uniserial module U has local endomorphism ring, then any direct summand of a direct sum U (I) of copies of U is a direct sum of copies of U , because any uniserial module is σ-small (a module is σ-small if it is countable ascending union of small submodules) and one can use [16,Theorem 2.52].(1) If gf = 0 for every monomorphism f : U → U and every epimorphism g : U → U , then every direct summand of a direct sum U (I) of copies of U is a direct sum of copies of U .
(2) If U is quasismall and there exist a monomorphism f : U → U and an epimorphism g : U → U such that gf = 0, then every direct summand of a direct sum U (I) of copies of U is isomorphic to U (J) ⊕ V (K) , where J and K are suitable sets and V is the unique uniserial module in the same monogeny class of U that is not quasismall.
(3) If U is not quasismall, then every direct summand of a direct sum U (I) of copies of U is a direct sum of copies of U .
Recall that a right module over a ring R is cyclically presented if it is isomorphic to R/aR for some element a ∈ R. For any ring R, the endomorphism ring End R (R/aR) of a non-zero cyclically presented module R/aR is isomorphic to  The Weak Krull-Schmidt Theorem for cyclically presented modules has an immediate consequence as far as equivalence of matrices is concerned.Recall that two m × n matrices A and B with entries in a ring R are said to be equivalent matrices, denoted A ∼ B, if there exist an m × m invertible matrix P and an n × n invertible matrix Q with entries in R (that is, matrices invertible in the rings M m (R) and M n (R), respectively) such that B = P AQ.We denote by diag(a 1 , . . ., a n ) the n×n diagonal matrix whose (i, i) entry is a i and whose other entries are zero.
R, because R is commutative, and therefore the endomorphism ring of each R/a i R is local, because R is local.Similarly for the modules R/b j R. Hence the Krull-Schmidt-Remak-Azumaya Theorem implies that there exists a permutation σ of {1, 2, . . ., n} with R/a i R ∼ = R/b σ(i) R for every i = 1, 2, . . ., n. Taking the annihilators of these isomorphic cyclic modules, we find that a i R = b σ(i) R for every i = 1, 2, . . ., n. Hence a i and b σ(i) are associates.
If the ring R is local, but not-necessarily commutative, we have the following result.
As far as infinite direct sums are concerned, the case of cyclically presented modules over local rings is much simpler than that of uniserial modules [3].The reason for this is that cyclically presented modules are finitely generated, hence small, so the pathology of non-quasismall modules can not appear in this setting.
Thus we have seen that, similarly to the endomorphism ring of a uniserial module over an arbitrary ring, the endomorphism ring of a cyclically presented module over a local ring also has at most two maximal right ideals (Theorem 4.6).By this fact, it is not surprising that there is an analogy between the behaviour of a direct sum of uniserial modules over arbitrary rings and the behaviour of a direct sum of cyclically presented modules over local rings.Therefore it makes sense to try to see which further parts of the theory of uniserial modules also hold for cyclically presented modules over local rings.Here are some very natural questions.
(1) Does the weak Krull-Schmidt Theorem hold for a direct sum of infinitely many cyclically presented modules over local rings?
(2) Is every direct summand of a direct sum of cyclically presented modules over a local ring a direct sum of cyclically presented modules?
(3) Is every direct summand of a direct sum of finitely many cyclically presented modules over a local ring a direct sum of cyclically presented modules?
The answer to the first question was given in [3,Theorem 3.1].The answer to the second question is negative.To this end, there is an example given by Puninski Now we need a very standard technique of homological algebra that allows to extend a morphism between two modules to their injective resolutions.Let us present it.Assume that E 0 , E 1 , E 0 , E 1 are indecomposable injective right modules over a ring R, and that ϕ : The morphisms f 0 and f 1 are not uniquely determined by f .The endomorphism ring of the kernel of a morphism between indecomposable injective modules has the same structure as the endomorphism ring of a uniserial module or a cyclically presented module over a local ring.More precisely, the endomorphism ring of the kernel of a morphism between two indecomposable injective modules is either local or has two maximal ideals, the kernel is determined up to isomorphism by its monogeny class and its upper part, and a weak form of the Krull-Schmidt Theorem also holds for direct sums of these kernels, as the following theorem shows.
Theorem 4.11.(Weak Krull-Schmidt Theorem for kernels of morphisms between indecomposable injective modules, [19,Theorem 2.7] and [14]) Let ϕ i : E i,0 → E i,1 (i = 1, 2, . . ., n) and ϕ j : E j,0 → E j,1 (j = 1, 2, . . ., t) be n + t non-injective morphisms between indecomposable injective right modules E i,0 , E i,1 , E j,0 , E j,1 over an arbitrary ring R. Then the direct sums ⊕ n i=0 ker ϕ i and ⊕ t j=0 ker ϕ j are isomorphic R-modules if and only if n = t and there exist two permutations σ, τ of {1, 2, . . ., n} such that [ker There are also some other classes of modules that satisfy the Weak Krull-Schmidt Theorem.One is the class of couniformly presented modules [20].It generalizes the class of cyclically presented modules over a local ring.An R-module M is said to be couniform if it has dual Goldie dimension 1, that is, it is nonzero and the sum of any two proper submodules of M R is a proper submodule of M R .An R-module M is couniformly presented if it is non-zero and there exists an exact sequence with both C R and P R couniform and P R projective.Under these hypotheses, the exact sequence ( 2) is called a couniform presentation of the couniformly presented module M R .
A similar behaviour, as far as direct-sum decompositions are concerned, takes place for the short exact sequences with A R and C R uniserial modules.The endomorphism ring of such a sequence in the category of all short exact sequences has at most four maximal ideals, and the isomorphism types of these sequences (3) are described by four invariants [10,11].Theorem 4.13.[22] Let C be a full subcategory of Mod-R and P, Q be two completely prime ideals of C. Assume that all objects of C are indecomposable right R-modules and that, for every and there exist two permutations σ, τ of {1, 2, . . ., n} such that ] Q for all i = 1, . . ., n.
A further remark: for the classes C of modules described so far, the fact that the weak form of the Krull-Schmidt Theorem holds can be described by saying that the corresponding monoid V (C) is a subdirect product of two free monoids.

Direct products of modules whose endomorphism rings have at most two maximal ideals
In the previous section, we have seen that the Weak Krull-Schmidt Theorem holds not only for uniserial modules, but also for cyclically presented modules over a local ring R, for kernels of morphisms between indecomposable injective modules, for couniformly presented modules, and more generally, for a number of classes of modules with at most two maximal right ideals.In this section, we will see that a similar result can hold not only for direct sums, but also for direct products of modules.
In order to present the main result in the most general setting, that of modules whose endomorphism rings have at most two maximal right ideals, we begin from the Weak Krull-Schimidt Theorem for direct products of uniserial modules.

Factorisation
Everything we've seen until now corresponds, in a broad sense, to study factorisations of elements in suitable monoids, for instance in the commutative monoid V (R).Let's see how what we have learned about direct-sum decompositions (directproduct decompositions) also holds for factorisations in other classes of monoids.
For instance, it can be applied to factorisations of elements in a commutative domain R, because clearly factorisations of elements in an integral domain R are exactly factorisations in the multiplicative monoid of R. We all know that a unique factorisation domain (UFD) is a commutative integral domain R such that: (i) R is atomic, that is, every element a ∈ R, a = 0 and a non-invertible, is a product of finitely many irreducible elements of R.
In an integral domain R, every prime element is irreducible.If R is a UFD, the converse holds.More precisely, an integral domain R is a UFD if and only if every irreducible is prime and R satisfies the ascending chain condition on principal ideals, if and only if R is atomic and every irreducible is prime.
Proposition 7.2.The following conditions are equivalent for two prime elements a, b of a commutative integral domain R: (i) a = bu for some invertible element u ∈ R.
Let's pass to consider, now, factorisation of polynomials into irreducible polynomials.
The case of commutative polynomials in commuting indeterminates is well known: the ring Z[x 1 , . . ., x n ] of all polynomials whose coefficients are in the ring of integers Z and with x 1 , . . ., x n commuting indeterminates is a UFD.
Let's see how the situation changes when we pass to the ring Z x 1 , . . ., x n , the free ring on n objects.The elements of Z x 1 , . . ., x n are non-commutative polynomials with coefficients in Z and with x 1 , . . ., x n non-commuting indeterminates.
The ring Z x 1 , . . ., x n is atomic, in the sense that polynomials do factorise as a product of irreducible polynomials.The invertible elements in Z x 1 , . . ., x n are only 1 and −1.But the factorisation x(yx − 2) = (xy − 2)x in the ring Z x, y shows that a polynomial in Z x 1 , . . ., x n does not necessarily factorise as a product of irreducible polynomials in a unique way up to the sign of the irreducible factors.Nevertheless the following theorem holds: Theorem 7.3.(Brungs' Theorem [9]) Every polynomial in R := Z x 1 , . . ., x n factorises as a product of irreducible polynomials.Moreover, if p 1 , . . ., p n , q 1 , . . ., q m are irreducible polynomials in R and p 1 . . .p n = q 1 . . .q m , then n = m and there exists a permutation σ of {1, 2, . . ., n} such that [R/p For instance, consider the category of finite partially ordered sets.This category has coproducts (disjoint unions), products (direct products with the componentwise order) and a terminal object 1 (the partially ordered set with one element), which is the identity with respect to product.Let L = {0, 1} denote the partially ordered set with two elements 0 < 1.Then, for every n ≥ 0, the direct product L n is a connected partially ordered set with 2 n elements, and its automorphism group is the symmetric group S n .If we compute the identity (5), which is an identity in the semiring N 0 [x], replacing x with L (and the natural number 1 with the partially ordered set 1), we get two essentially different direct-product decompositions of the partially ordered set 1 ∪L ∪L 2 ∪L 3 ∪L 4 ∪L 5 into indecomposable partially ordered sets, that is, we get that (L 3 ∪1) × (L 2 ∪L ∪1) ∼ = (L ∪1) × (L 4 ∪L 2 ∪1).
This example, due to Nakayama and Hashimoto [24,28], shows that Krull-Schmidt fails in the category of finite partially ordered sets.
The possibility of applying the identity (4) in

Theorem 2 . 1 .
(Krull-Schmidt  Theorem for finite groups) If a finite group G has two direct-product decompositions

Theorem 2 . 2 .
UNIQUENESS OF DECOMPOSITION, FACTORISATIONS 109 decompositions of a finite group into indecomposable factors are not only isomorphic, but also centrally isomorphic.More precisely, he proved the following theorem.Recall that a central automorphism of group G is an automorphism of G that induces the identity G/ζ(G) → G/ζ(G), where ζ(G) denotes the center of G.If a finite group G has two direct-product decompositions into in- and there exist a central automorphism ϕ of G and a permutation σ of {1, 2, . . ., t} such that Then I and K are two two-sided completely prime ideals of E, and every proper right ideal of E and every proper left ideal of E is contained either in I or in K.Moreover, (a) either E is a local ring with maximal ideal I ∪ K, or (b) E/I and E/K are division rings, and E/J(E) ∼ = E/I × E/K.
Krull-Schmidt Theorem.Facchini's weak form of the Krull-Schmidt Theorem holds only for finite direct sums of uniserial modules, not for infinite ones.In order to state the theorem, we need the concepts of monogeny class and epigeny class of a module.Two right R-modules U and V are said to have (1) the same monogeny class, denoted [U ] m = [V ] m , if there exist a monomorphism U → V and a monomorphism V → U ; (2) the same epigeny class, denoted [U ] e = [V ] e , if there exist an epimorphism U → V and an epimorphism V → U .The weak Krull-Schmidt Theorem for finite families of uniserial modules can be formulated as follows: Theorem 4.2.(Weak Krull-Schmidt Theorem [15]) Let U 1 , . . ., U n , V 1 , . . ., V t be n + t non-zero uniserial right modules over a ring R. Then the direct sums

Theorem 4 . 6 .
The following theorem is proved in [2, Theorem 2.1].Let a be a non-zero non-invertible element of an arbitrary local ring R, let E be the idealizer of aR, and let E/aR be the endomorphism ring of the cyclically presented right R-module R/aR.Set I := { r ∈ R | ra ∈ aJ(R) } and K := J(R) ∩ E. Then I and K are two two-sided completely prime ideals of E containing aR, the union (I/aR) ∪ (K/aR) is the set of all non-invertible elements of E/aR, and every proper right ideal of E/aR and every proper left ideal of E/aR is contained either in I/aR or in K/aR.Moreover, exactly one of the following two conditions holds: (a) Either I and K are comparable (that is, I ⊆ K or K ⊆ I), in which case E/aR is a local ring, or (b) I and K are not comparable, and in this case E/I and E/K are division rings, J(E/aR) = (I ∩ K)/aR, and (E/aR)/J(E/aR) is canonically isomorphic to the direct product E/I × E/K.For any ring R, let U (R) denote the group of all invertible elements of R. If R/aR and R/bR are cyclically presented modules over a local ring R, we say that R/aR and R/bR have the same lower part, and write [R/aR] l = [R/bR] l , if there exist u, v ∈ U (R) and r, s ∈ R with au = rb and bv = sa.(It is possible to prove that two cyclically presented modules over a local ring have the same lower part if and only if their Auslander-Bridger transposes have the same epigeny class.)Theorem 4.7.(Weak Krull-Schmidt Theorem for Cyclically Presented Modules) Let a 1 , . . ., a n , b 1 , . . ., b t be n + t non-invertible elements of a local ring R. Then the are isomorphic right R-modules if and only if n = t and there exist two permutations σ, τ of {1, 2, . . ., n} such that [R/a i R] l = [R/b σ(i) R] l and [R/a i R] e = [R/b τ (i) R] e for every i = 1, 2, . . ., n.

Remark 4 . 8 .
If R is a commutative local ring and a 1 , . . ., a n , b 1 , . . ., b n are elements of R, then diag(a 1 , . . ., a n ) ∼ diag(b 1 , . . ., b n ) if and only if there exists a permutation σ of {1, 2, . . ., n} with a i and b σ(i) associates for every i = 1, 2, . . ., n.Here a, b ∈ R are associates if they generate the same principal ideal of R. Let's prove this.Assume R commutative and local, and diag(a 1 , . .

in [ 31 ,
Proposition 8.1].For the third question, there exist countably generated indecomposable relatively divisible projective modules over local rings that are not direct sums of cyclically presented modules, and relatively divisible projective UNIQUENESS OF DECOMPOSITION, FACTORISATIONS 117 modules over local rings that are not direct sums of indecomposable modules.But, according to [36, Corollary 2], every relatively divisible projective module over a commutative local ring is a direct sum of cyclically presented modules.Let us pass to consider another class of modules.For a right module A R over a ring R, let E(A R ) denote the injective envelope of A R .We say that two modules A R and B R have the same upper part, and write [

Theorem 4 . 10 .
Let E 0 and E 1 be indecomposable injective right modules over a ring R, and let ϕ : E 0 → E 1 be a non-zero non-injective morphism.Let S := End R (ker ϕ) denote the endomorphism ring of ker ϕ.Set I := { f ∈ S | the endomorphism f of ker ϕ is not a monomorphism } and K := { f ∈ S | the endomorphism f 1 of E 1 is not a monomorphism } = { f ∈ S | ker ϕ ⊂ f −1 0 (ker ϕ) }.Then I and K are two two-sided completely prime ideals of S, and every proper right ideal of S and every proper left ideal of S is contained either in I or in K.Moreover, exactly one of the following two conditions holds: (a) Either I and K are comparable (that is, I ⊆ K or K ⊆ I), in which case S is a local ring with maximal ideal I ∪ K, or (b) I and K are not comparable, and in this case S/I and S/K are division rings and S/J(S) ∼ = S/I × S/K.
We conclude this section by describing one of the general patterns that allow to treat all the previous examples at the same time.Let C be a full subcategory of the category Mod-R for some ring R and assume that every object of C is an indecomposable right R-module.Define a completely prime ideal P of C as an assignment of a subgroup P(A, B) of the additive abelian group Hom R (A, B) to every pair (A, B) of objects of C, with the following two properties: (1) for every A, B, C ∈ Ob(C), every f : A → B and every g : B → C, one has that gf ∈ P(A, C) if and only if either f ∈ P(A, B) or g ∈ P(B, C); (2) P(A, A) is a proper subgroup of Hom R (A, A) for every object A ∈ Ob(C).Let P be a completely prime ideal of C. If A, B are objects of C, we say that A and B have the same P class, and write [A] P = [B] P , if P(A, B) = Hom R (A, B) and P(B, A) = Hom R (B, A).
N 0 [x] to get an isomorphism in the category of finite partially ordered sets is due to the fact that distributivity holds for partially ordered sets: X × (Y ∪Z) ∼ = (X × Y ) ∪(X × Z).More generally, recall that a category C with finite products (−) × (−) and coproducts (−) + (−) is called (finitary) distributive if, for any objects X, Y, Z of C, the canonical morphismX × Y + X × Z → X × (Y + Z)is an isomorphism.It is now easily seen that Nakayama and Hashimoto's technique can be applied to any distributive category.
[5]]central automorphism ϕ in the statement of Theorem 2.2, whose existence was proved by Remak, corresponds exactly to the fact that Remark, in the study of direct-product decompositions of a group G, was studying the decompositions of the regular object G in the category G-Grp and not the decompositions of the group G in the category Grp.Of course, he didn't know what a category is.An interesting role in this setting is played by the full subcategory C G of G-Grp consisting of all the objects (H, ϕ) of G-Grp for which the image of the group homomorphism ϕ : G → Aut(H) contains the group Inn(H) of all inner automorphisms of H[18].Another setting in which the Weak Krull-Schmidt Theorem holds in Group Theory is in the study of abelian normal subgroups of a group G[5].If G is an arbitrary group and H is an abelian normal subgroup of G, then the conjugationα : G → Aut(G) induces an action α : G → Aut(H) because H is normal in G.If Z[G]is the group algebra of G with integer coefficients, then the conjugation α : G → Aut(H) extends to a ring morphism Z[G] → End(H), so that H turns out to be a left module over the ring Z[G].Hence Theorem 4.2 can be applied to this left module whenever H is a direct sum of uniserial left Z[G]-modules.
Thus an integral domain R is a unique factorisation domain if and only if the multiplicative monoid R \ {0} is isomorphic to the direct product of the abelian group U (R) and a free commutative monoid F .