LCM splittings in GCD rings
The notion of gcd for integers is classic, giving rise to arguably the first proper known algorithm, the Euclidean algorithm from at least 300 BC computing it. GCD rings generalize this notion to rings, which in this paper are always associative, commutative and unital.
Definition 1.1. Let R R R be a ring and a , b ∈ R a,b \in R a , b ∈ R two elements.
(a) A greatest common divisor g = gcd ( a , b ) ∈ R g = \gcd(a,b) \in R g = g cd( a , b ) ∈ R for the pair a , b a,b a , b is a common divisor (g ∣ a , g ∣ b ⟺ ∃ a ′ , b ′ ∈ R : a = a ′ g , b = b ′ g g \mid a, g \mid b \Longleftrightarrow\exists a',b' \in R: a = a'g, b = b'g g ∣ a , g ∣ b ⟺ ∃ a ′ , b ′ ∈ R : a = a ′ g , b = b ′ g ) such that any other common divisor d ∈ R d \in R d ∈ R of a a a and b b b necessarily divides g g g , i.e. d ∣ g d \mid g d ∣ g .
(b) R R R is a GCD ring if every pair of elements admits a gcd.
(c) A least common multiple l = lcm ( a , b ) ∈ R l = \operatorname{lcm}(a,b) \in R l = lcm ( a , b ) ∈ R is a common multiple (a ∣ l a \mid l a ∣ l and b ∣ l b \mid l b ∣ l ) such that any other common multiple m ∈ R m \in R m ∈ R of a a a and b b b is necessarily a multiple of l l l , i.e. l ∣ m l \mid m l ∣ m .
Gcds, when they exist, are only defined up to associates , i.e. only up to multiplication by a unit u ∈ R × u \in R^{\times} u ∈ R × . Explicitly, for a , b ∈ R a,b \in R a , b ∈ R , I will write a ∼ R b a \sim_R b a ∼ R b if ∃ u ∈ R × : u a = b \exists u \in R^{\times}: ua = b ∃ u ∈ R × : u a = b . I will simply write a ∼ b a \sim b a ∼ b when the ring is clear from the context. It is a standard result that this is an equivalence relation, i.e. it is reflexive, symmetric, and transitive. I assume that gcd ( a , b ) \gcd(a,b) g cd( a , b ) is some fixed choice among all valid associates. Consequently, for u , v ∈ R × u,v \in R^{\times} u , v ∈ R × , we have gcd ( u a , v b ) ∼ gcd ( a , b ) \gcd(ua,vb) \sim\gcd(a,b) g cd( u a , v b ) ∼ g cd( a , b ) , but equality need not hold. Lcms behave similarly.
Motivated by a question on Mathematics Stack Exchange by Ash Seifert [18 ] , I define the following notions, which I could not find in the literature despite their seemingly obvious nature.
Definition 1.2. Let R R R be a GCD ring and a , b ∈ R a,b \in R a , b ∈ R two elements.
(a) An lcm splitting of a , b a,b a , b , if it exists, consists of a pair c , d ∈ R c,d \in R c , d ∈ R such that c d ∼ lcm ( a , b ) cd \sim\operatorname{lcm}(a,b) c d ∼ lcm ( a , b ) , c ∣ a c \mid a c ∣ a , d ∣ b d \mid b d ∣ b and gcd ( c , d ) ∼ 1 \gcd(c,d) \sim1 g cd( c , d ) ∼ 1 . The fourth and last condition is called the splitting condition .
(b) R R R is lcm splitting if every pair of elements admits an lcm splitting.
The main contributions of this paper are
the definitions of lcm splitting just given,
a theorem that adequate (defined below) implies lcm splitting for domains, and
two explicit examples of Bézout domains that are not lcm splitting.
Some of the presented ideas were developed in dialog with chatbots, cf. § 6 .
Preliminaries, related notions and results
Most rings considered will be integral domains , i.e., they have no nontrivial zero divisors. A GCD domain is a GCD ring that is an integral domain. Domains are canonically embedded in a fraction field R ⊆ Frac ( R ) R \subseteq\operatorname{Frac}(R) R ⊆ Frac ( R ) . This allows one to write certain elements of R R R as fractions with nonzero denominator, simplifying notation. E.g., if gcd ( a , b ) ≠ 0 \gcd(a,b) \ne0 g cd( a , b ) = 0 and a ′ ∈ R a' \in R a ′ ∈ R satisfies a = a ′ gcd ( a , b ) a = a'\gcd(a,b) a = a ′ g cd( a , b ) , then the fraction a ′ = a gcd ( a , b ) ∈ Frac ( R ) a' = \frac{a}{\gcd(a,b)} \in\operatorname{Frac}(R) a ′ = g c d ( a , b ) a ∈ Frac ( R ) is well-defined and, moreover, lies in R R R .
In Appendix A, I collect some standard identities for gcd and lcm for a , b , c a,b,c a , b , c elements of a GCD domain R R R , assumed to be nonzero when in doubt. Most references deal with the special case R = Z R = \mathbb{Z} R = Z , so I will supply (references to) proofs in the general context to be self-contained.
Remark 2.1. I will apply the identities and properties of Prop. A.1 by labeling steps with the corresponding numbers throughout the paper. Numbers in parentheses in equation-like statements without further explanation will always refer to Prop. A.1 .
Associativity (7) lets us define multiple gcds and lcms iteratively. E.g. gcd ( a , b , c ) = gcd ( gcd ( a , b ) , c ) \gcd(a,b,c) = \gcd(\gcd(a,b),c) g cd( a , b , c ) = g cd( g cd( a , b ) , c ) or lcm ( a , b , c , d ) = lcm ( lcm ( lcm ( a , b ) , c ) , d ) \operatorname{lcm}(a,b,c,d) = \operatorname{lcm}(\operatorname{lcm}(\operatorname{lcm}(a,b),c),d) lcm ( a , b , c , d ) = lcm ( lcm ( lcm ( a , b ) , c ) , d ) .
Next, I recall some classical notions used in the subsequent sections, followed by references for details. Then I recall several related facts, introducing further classical terminology.
Definition 2.2. (1) A GCD ring R R R is a Bézout ring if all its finitely generated ideals are principal. In particular (and ultimately equivalently), for any a , b ∈ R a,b \in R a , b ∈ R the ideal identity ( a , b ) = ( gcd ( a , b ) ) (a,b)=(\gcd(a,b)) ( a , b ) = ( g cd( a , b )) holds. In general, only “⊆ \subseteq ⊆ ” is guaranteed. In the language of the generalized Euclidean algorithm, the ideal identity (and thus the Bézout property) can equivalently be stated as ∃ c , d ∈ R : a d − b c = gcd ( a , b ) \exists c,d \in R: ad-bc=\gcd(a,b) ∃ c , d ∈ R : a d − b c = g cd( a , b ) .
A Bézout domain is a Bézout ring that is also an integral domain.
(2) A GCD ring R R R has good factorization (gf) if for all a , b ∈ R ∖ { 0 } a,b \in R \setminus\{0\} a , b ∈ R ∖ { 0 } there exist c , d ∈ R c,d \in R c , d ∈ R with a = c d a=cd a = c d , gcd ( c , b ) ∼ 1 \gcd(c,b)\sim1 g cd( c , b ) ∼ 1 and b ∈ ( d ) b \in(d) b ∈ ( d ) .
(3) A ring R R R is adequate if for all a , b ∈ R ∖ { 0 } a,b \in R \setminus\{0\} a , b ∈ R ∖ { 0 } there exist c , d ∈ R c,d \in R c , d ∈ R with a = c d a=cd a = c d , b R + c R = R bR+cR=R b R + c R = R and, for all d ′ ∈ R d' \in R d ′ ∈ R , the inclusions d R ⊆ d ′ R ⊊ R dR \subseteq d'R \subsetneq R d R ⊆ d ′ R ⊊ R imply b R + d ′ R ⊊ R bR+d'R \subsetneq R b R + d ′ R ⊊ R .
If R R R is a Bézout ring, this is equivalent to: For all a , b ∈ R ∖ { 0 } a,b \in R \setminus\{0\} a , b ∈ R ∖ { 0 } there exist c , d ∈ R c,d \in R c , d ∈ R with a = c d a=cd a = c d , gcd ( c , b ) ∼ 1 \gcd(c,b)\sim1 g cd( c , b ) ∼ 1 and for all non-units d ′ ∈ R ∖ R × d' \in R \setminus R^\times d ′ ∈ R ∖ R × the divisibility d ′ ∣ d d' \mid d d ′ ∣ d implies gcd ( d ′ , b ) ≁ 1 \gcd(d',b)\not\sim1 g cd( d ′ , b ) ∼ 1 . Note the similarity to the definition of gf.
(4) For an m × n m \times n m × n -matrix A ∈ Mat m , n ( R ) A \in\operatorname{Mat}_{m,n}(R) A ∈ Mat m , n ( R ) with coefficients in a ring R R R (m , n ∈ N > 0 m,n \in\mathbb{N}_{>0} m , n ∈ N > 0 ), we say that A A A admits a diagonal reduction if there are invertible (over R R R ) matrices P ∈ Mat m , m ( R ) P \in\operatorname{Mat}_{m,m}(R) P ∈ Mat m , m ( R ) , Q ∈ Mat n , n ( R ) Q \in\operatorname{Mat}_{n,n}(R) Q ∈ Mat n , n ( R ) such that P A Q = Diag ( ( d i ) i = 1 min ( m , n ) ) PAQ=\operatorname{Diag}((d_i)_{i=1}^{\min(m,n)}) P A Q = Diag (( d i ) i = 1 m i n ( m , n ) ) is diagonal with d i ∣ d i + 1 d_i \mid d_{i+1} d i ∣ d i + 1 for all 1 ≤ i ≤ min ( m , n ) − 1 1 \leq i \leq\min(m,n)-1 1 ≤ i ≤ min ( m , n ) − 1 . R R R has the elementary divisor property if all (finite-dimensional) matrices admit diagonal reductions.
(5) For A ∈ Mat m , n ( R ) A \in\operatorname{Mat}_{m,n}(R) A ∈ Mat m , n ( R ) over a ring R R R (m , n ∈ N > 0 m,n \in\mathbb{N}_{>0} m , n ∈ N > 0 ), we say that A A A admits a triangular reduction if there is an invertible (over R R R ) matrix Q ∈ Mat n , n ( R ) Q \in\operatorname{Mat}_{n,n}(R) Q ∈ Mat n , n ( R ) such that A Q AQ A Q is lower triangular (arbitrary entries on the diagonal are allowed). R R R is K-Hermite if all (finite-dimensional) matrices admit triangular reductions.
I will say that a pair ( a , b ) (a,b) ( a , b ) satisfies resp. violates gf or adequate if the respective property holds resp. fails for ( a , b ) (a,b) ( a , b ) . Anderson [1 ] provides a survey of divisibility theory for GCD domains . Cohn [3 ] gives results about Bézout domains (Cohn calls GCD domains HCF domains and Bézout domains Bézout rings). Gillman and Henriksen [7 ] (they call Bézout rings F-rings and K-Hermite Hermite), Lam [13 Ch. 1, Def. 4.23] and L’Innocente and Point [14 Sec. 5.2] elaborate on the properties K-Hermite, elementary divisor, adequate and good factorization . A textbook by Cohn [4 Ch. 6.5] discusses some elementary properties of divisibility, gcds and lcms.
In the literature, gf and adequate are often used only in the narrower context of Bézout domains. Quoting van den Dries and Macintyre [20 Sec. 2.10] and adjusting notation: “In intuitive terms, writing prime factor of x x x for maximal ideal containing x x x , this implies that the prime factors of c c c are those of a a a that do not occur in b b b , while the prime factors of d d d are exactly the common prime factors of a a a and b b b .”
Good factorization resp. adequate allow one to split a ring element into factors multiplicatively well aligned with another ring element. This is close in spirit to the idea of an lcm splitting. Indeed, in § 3 I prove adequate (and gf) implies lcm splitting. [20 Sec. 2.10] also sketches an example, due to R. Wiegand, in which gf fails. In § 4 , this is my starting point for the first counterexample: a Bézout domain that is not lcm splitting. Note that gf and adequate are not symmetric: ( a , b ) (a,b) ( a , b ) can be gf resp. adequate while ( b , a ) (b,a) ( b , a ) is not, cf. Prop. 4.14 . However, the definition of lcm splitting is clearly symmetric.
Remark 2.4. I give some first observations on the lcm splitting property. I remind the reader of the convention in Rem. 2.1 about numbers in parentheses.
In Def. 1.2 , replacing c d ∼ lcm ( a , b ) cd \sim\operatorname{lcm}(a,b) c d ∼ lcm ( a , b ) by c d = lcm ( a , b ) cd = \operatorname{lcm}(a,b) c d = lcm ( a , b ) does not change the concept.
For a = 0 a = 0 a = 0 or b = 0 b = 0 b = 0 (or both), lcm splitting is always satisfied. Assume a = 0 a = 0 a = 0 and set c = 0 , d = 1 c = 0,d = 1 c = 0 , d = 1 . Then c d = 0 ⋅ 1 = 0 cd = 0 \cdot1 = 0 c d = 0 ⋅ 1 = 0 (4), (6) = lcm ( 0 , b ) = lcm ( a , b ) = \operatorname{lcm}(0,b) = \operatorname{lcm}(a,b) = lcm ( 0 , b ) = lcm ( a , b ) , c = 0 ∣ 0 = a c = 0 \mid0 = a c = 0 ∣ 0 = a , d = 1 ∣ b d = 1 \mid b d = 1 ∣ b since 1 1 1 divides every ring element, and gcd ( c , d ) = gcd ( 0 , 1 ) \gcd(c,d) = \gcd(0,1) g cd( c , d ) = g cd( 0 , 1 ) (4) ∼ 1 \sim1 ∼ 1 . Similarly for a = 0 = b a = 0 = b a = 0 = b . We can wlog assume that a ≠ 0 ≠ b a \ne0 \ne b a = 0 = b . E.g., below gcd ( a , b ) ≠ 0 \gcd(a,b) \ne0 g cd( a , b ) = 0 is assumed in terms like b gcd ( a , b ) \frac{b}{\gcd(a,b)} g c d ( a , b ) b .
LCM splittings need not be unique, not even up to associates. E.g., in R = Z R = \mathbb{Z} R = Z , for a = 2 ⋅ 3 = 6 a = 2 \cdot3 = 6 a = 2 ⋅ 3 = 6 , b = 2 ⋅ 5 = 10 b = 2 \cdot5 = 10 b = 2 ⋅ 5 = 10 , any of ( c , d ) ∈ { ( ± 6 , ± 5 ) , ( ± 3 , ± 10 ) } (c,d) \in\{(\pm6,\pm5),(\pm3,\pm10)\} ( c , d ) ∈ {( ± 6 , ± 5 ) , ( ± 3 , ± 10 )} is an lcm splitting.
Without the splitting condition, the other three properties can always easily be satisfied, e.g., by ( c , d ) = ( a , b gcd ( a , b ) ) (c,d) = \left(a,\frac{b}{\gcd(a,b)}\right) ( c , d ) = ( a , g c d ( a , b ) b ) . Use the product formula (5) for the first condition. The two divisibility conditions are trivially satisfied. The following example in R = Z R = \mathbb{Z} R = Z (with gcds chosen nonnegative) demonstrates that this approach does not guarantee the splitting condition: for a = 2 2 ⋅ 3 = 12 a = 2^2 \cdot3 = 12 a = 2 2 ⋅ 3 = 12 , b = 2 ⋅ 3 2 = 18 b = 2 \cdot3^2 = 18 b = 2 ⋅ 3 2 = 18 , we have b gcd ( a , b ) = 18 6 = 3 \frac{b}{\gcd(a,b)} = \frac{18}{6} = 3 g c d ( a , b ) b = 6 18 = 3 and thus gcd ( a , b gcd ( a , b ) ) = gcd ( 12 , 3 ) = 3 ≁ 1 \gcd\left(a,\frac{b}{\gcd(a,b)}\right) = \gcd(12,3) = 3 \not\sim1 g cd( a , g c d ( a , b ) b ) = g cd( 12 , 3 ) = 3 ∼ 1 .
One could try to iterate this division by common factors until the evolving pair ( c , d ) (c,d) ( c , d ) becomes coprime. For this, start with ( c 0 , d 0 ) = ( a , b gcd ( a , b ) ) (c_0,d_0) = \left(a,\frac{b}{\gcd(a,b)}\right) ( c 0 , d 0 ) = ( a , g c d ( a , b ) b ) . If gcd ( c 0 , d 0 ) ≁ 1 \gcd(c_0,d_0) \not\sim1 g cd( c 0 , d 0 ) ∼ 1 , intuitively speaking, a a a and b b b have a “common irreducible (atomic) factor” that has higher multiplicity in b b b than in a a a , so it is present in a a a , but also “survives” the division by gcd ( a , b ) \gcd(a,b) g cd( a , b ) in d 0 d_0 d 0 . The idea is to shift enough factors back from c i c_i c i to d i d_i d i until ( c i , d i ) (c_i,d_i) ( c i , d i ) become coprime. For this, define ( c i + 1 , d i + 1 ) = ( c i gcd ( c i , d i ) , d i ⋅ gcd ( c i , d i ) ) (c_{i+1},d_{i+1}) = \left(\frac{c_i}{\gcd(c_i,d_i)},d_i \cdot\gcd(c_i,d_i)\right) ( c i + 1 , d i + 1 ) = ( g c d ( c i , d i ) c i , d i ⋅ g cd( c i , d i ) ) – note that this also makes sense if ( c i , d i ) (c_i,d_i) ( c i , d i ) are already coprime.
This algorithm1 gives us an lcm splitting if the sequence of ( c i , d i ) (c_i,d_i) ( c i , d i ) stabilizes after finitely many steps (a rigorous proof of this claim is left to the reader). This is guaranteed, for instance, if the ascending chain of principal ideals ( c 0 ) ⊆ ( c 1 ) ⊆ ⋯ ⊆ ( c i ) ⊆ … (c_0) \subseteq(c_1) \subseteq\dots\subseteq(c_i) \subseteq\dots ( c 0 ) ⊆ ( c 1 ) ⊆ ⋯ ⊆ ( c i ) ⊆ … always stabilizes, i.e. if the ACCP holds for enough sequences. As discussed above, GCD domains satisfying the ACCP are exactly UFDs. Hence UFDs are lcm splitting. The ring of entire functions R 2 = O ( C ) R_2 = \mathcal{O}(\mathbb{C}) R 2 = O ( C ) is an lcm splitting domain, for which this procedure does not terminate after finitely many steps, cf. App. D.
We can see that UFDs are lcm splitting in a different way, using the fact that UFDs admit a unique factorization into irreducible (atomic) elements as was intuitively alluded to before. For a , b a,b a , b , such unique factorizations can be written in the following form:
a = ∏ i = 1 l p i e i ⋅ ∏ i = 1 m q i f i ⋅ ∏ i = 1 n r i g i , b = ∏ i = 1 l p i e i ′ ⋅ ∏ i = 1 m q i f i ′ ⋅ ∏ i = 1 n r i g i ′ , a = \prod_{i=1}^{l} p_i^{e_i} \cdot\prod_{i=1}^{m} q_i^{f_i} \cdot\prod_{i=1}^{n} r_i^{g_i}, \qquad b = \prod_{i=1}^{l} p_i^{e_i'} \cdot\prod_{i=1}^{m} q_i^{f_i'} \cdot\prod_{i=1}^{n} r_i^{g_i'}, a = i = 1 ∏ l p i e i ⋅ i = 1 ∏ m q i f i ⋅ i = 1 ∏ n r i g i , b = i = 1 ∏ l p i e i ′ ⋅ i = 1 ∏ m q i f i ′ ⋅ i = 1 ∏ n r i g i ′ ,
where all the p i , q i , r i p_i,q_i,r_i p i , q i , r i are pairwise coprime, e i > e i ′ ≥ 0 e_i > e_i' \ge0 e i > e i ′ ≥ 0 , f i = f i ′ > 0 f_i = f_i' > 0 f i = f i ′ > 0 , and 0 ≤ g i < g i ′ 0 \le g_i < g_i' 0 ≤ g i < g i ′ . The atoms p i p_i p i need to go to c c c , the r i r_i r i to d d d , and the (full power of each) q i q_i q i can be distributed arbitrarily. Two possibilities for an lcm splitting (that coincide if m = 0 m = 0 m = 0 ) are given by
( c , d ) = ( ∏ i = 1 l p i e i ⋅ ∏ i = 1 m q i f i ⋅ ∏ i = 1 n r i g i ′ , ) , ( c ′ , d ′ ) = ( ∏ i = 1 l p i e i ⋅ ∏ i = 1 m q i f i ′ ⋅ ∏ i = 1 n r i g i ′ ) . (c,d) = \left(\prod_{i=1}^{l} p_i^{e_i} \cdot\prod_{i=1}^{m} q_i^{f_i} \cdot\prod_{i=1}^{n} r_i^{g_i'},\right), \qquad(c',d') = \left(\prod_{i=1}^{l} p_i^{e_i} \cdot\prod_{i=1}^{m} q_i^{f_i'} \cdot\prod_{i=1}^{n} r_i^{g_i'}\right). ( c , d ) = ( i = 1 ∏ l p i e i ⋅ i = 1 ∏ m q i f i ⋅ i = 1 ∏ n r i g i ′ , ) , ( c ′ , d ′ ) = ( i = 1 ∏ l p i e i ⋅ i = 1 ∏ m q i f i ′ ⋅ i = 1 ∏ n r i g i ′ ) .
It is easily verified that these pairs constitute lcm splittings. This is a direct way to see that UFDs are lcm splitting. However, the other approach also has its merits. It gives an algorithm for UFDs whenever gcds are computable, which is often the case, while a direct algorithm for unique factorization is often unavailable or more expensive.
Adequate implies lcm splitting
I remind the reader of the convention in Rem. 2.1 about numbers in parentheses.
Lemma 3.1. Let R R R be a G C D GCD GC D domain. Given a , b ∈ R ∖ { 0 } a,b \in R \setminus\{0\} a , b ∈ R ∖ { 0 } and some c , d ∈ R c,d \in R c , d ∈ R satisfying the adequate conditions, i.e. a = c d a=cd a = c d , gcd ( c , b ) ∼ 1 \gcd(c,b)\sim1 g cd( c , b ) ∼ 1 and ∀ d ′ ∈ R ∖ R × : d ′ ∣ d ⇒ gcd ( d ′ , b ) ≁ 1 \forall d' \in R\setminus R^{\times}:d'\mid d\Rightarrow\gcd(d',b)\not\sim1 ∀ d ′ ∈ R ∖ R × : d ′ ∣ d ⇒ g cd( d ′ , b ) ∼ 1 , we have gcd ( c , d ) ∼ 1 \gcd(c,d)\sim1 g cd( c , d ) ∼ 1 .
Proof. Assume for contradiction that g = gcd ( c , d ) ≁ 1 g=\gcd(c,d)\not\sim1 g = g cd( c , d ) ∼ 1 . By the definition of the gcd, g = gcd ( c , d ) ∣ d g=\gcd(c,d)\mid d g = g cd( c , d ) ∣ d . Hence the adequate property for d ′ = g ≁ 1 d'=g\not\sim1 d ′ = g ∼ 1 implies gcd ( g , b ) ≁ 1 \gcd(g,b)\not\sim1 g cd( g , b ) ∼ 1 . But also g = gcd ( c , d ) ∣ c g=\gcd(c,d)\mid c g = g cd( c , d ) ∣ c , so
gcd ( g , b ) = gcd ( gcd ( c , d ) , b ) ∣ gcd ( c , b ) ∼ 1 ⇒ gcd ( g , b ) ∼ 1. (1) \gcd(g,b)=\gcd(\gcd(c,d),b)\mid\gcd(c,b)\sim1\Rightarrow\gcd(g,b)\sim1. \tag*{(1)} g cd( g , b ) = g cd( g cd( c , d ) , b ) ∣ g cd( c , b ) ∼ 1 ⇒ g cd( g , b ) ∼ 1. (1)
This contradiction shows that the assumption is false. Thus g = gcd ( c , d ) ∼ 1 g=\gcd(c,d)\sim1 g = g cd( c , d ) ∼ 1 . □
Theorem 3.2. An adequate Bézout domain R R R is lcm splitting.
Proof. Let a , b ∈ R ∖ { 0 } a,b\in R\setminus\{0\} a , b ∈ R ∖ { 0 } be arbitrary (cf. Rem. 2.4). Set g = gcd ( a , b ) , l = lcm ( a , b ) g=\gcd(a,b),l=\operatorname{lcm}(a,b) g = g cd( a , b ) , l = lcm ( a , b ) . From (5), a b ∼ g l ab\sim gl ab ∼ g l , we deduce l ∼ a b g = a g b g l\sim\frac{ab}{g}=\frac{a}{g}\frac{b}{g} l ∼ g ab = g a g b .
Now apply the adequate property to the pair ( g , b g ) \left(g,\frac{b}{g}\right) ( g , g b ) to get g = g 1 g 2 g=g_{1}g_{2} g = g 1 g 2 with gcd ( g 1 , b g ) ∼ 1 \gcd\left(g_{1},\frac{b}{g}\right)\sim1 g cd( g 1 , g b ) ∼ 1 and ∀ g ‾ 2 ∈ R ∖ R × : g ‾ 2 ∣ g 2 ⇒ gcd ( g ‾ 2 , b g ) ≁ 1 \forall\overline{g}_{2}\in R\setminus R^{\times}:\overline{g}_{2}\mid g_{2}\Rightarrow\gcd\left(\overline{g}_{2},\frac{b}{g}\right)\not\sim1 ∀ g 2 ∈ R ∖ R × : g 2 ∣ g 2 ⇒ g cd( g 2 , g b ) ∼ 1 . Apply adequate a second time to ( g 2 , a g ) \left(g_{2},\frac{a}{g}\right) ( g 2 , g a ) to get g 2 = g 2 ′ g ‾ 2 ′ g_{2}=g'_{2}\overline{g}'_{2} g 2 = g 2 ′ g 2 ′ with gcd ( g 2 ′ , a g ) ∼ 1 \gcd\left(g'_{2},\frac{a}{g}\right)\sim1 g cd( g 2 ′ , g a ) ∼ 1 and ∀ g ‾ 1 ∈ R ∖ R × : g ‾ 1 ∣ g 1 ′ ⇒ gcd ( g ‾ 1 , a g ) ≁ 1 \forall\overline{g}_{1}\in R\setminus R^{\times}:\overline{g}_{1}\mid g'_{1}\Rightarrow\gcd\left(\overline{g}_{1},\frac{a}{g}\right)\not\sim1 ∀ g 1 ∈ R ∖ R × : g 1 ∣ g 1 ′ ⇒ g cd( g 1 , g a ) ∼ 1 . Overall, we have l ∼ g g a g b g = a g g 1 ′ g 2 ′ b g l\sim\frac{g}{g}\frac{a}{g}\frac{b}{g}=\frac{a}{g}g'_{1}g'_{2}\frac{b}{g} l ∼ g g g a g b = g a g 1 ′ g 2 ′ g b , and we will show some coprimality properties among these factors:
(A) gcd ( a g , g 2 ′ ) ∼ 1 \gcd\left(\frac{a}{g},g'_{2}\right)\sim1 g cd( g a , g 2 ′ ) ∼ 1 follows directly from the 2nd application of adequate and (1) .
(B) Setting c = g c=g c = g in (13), we get gcd ( a g , b g ) ∼ gcd ( a , b ) g = g g = 1 \gcd\left(\frac{a}{g},\frac{b}{g}\right)\sim\frac{\gcd(a,b)}{g}=\frac{g}{g}=1 g cd( g a , g b ) ∼ g g c d ( a , b ) = g g = 1 .
(C) gcd ( g 1 , g 2 ′ ) ∣ gcd ( g 1 , g 1 ′ g 2 ′ ) = gcd ( g 1 , g 2 ) ∼ 1 \gcd(g_{1},g'_{2})\mid\gcd(g_{1},g'_{1}g'_{2})=\gcd(g_{1},g_{2})\sim1 g cd( g 1 , g 2 ′ ) ∣ g cd( g 1 , g 1 ′ g 2 ′ ) = g cd( g 1 , g 2 ) ∼ 1 , where the last step follows from Lem. 3.1 . By (9) we get gcd ( g 1 , g 2 ′ ) ∼ 1 \gcd(g_{1},g'_{2})\sim1 g cd( g 1 , g 2 ′ ) ∼ 1 .
(D) gcd ( g 1 , b g ) ∼ 1 \gcd\left(g_{1},\frac{b}{g}\right)\sim1 g cd( g 1 , g b ) ∼ 1 follows directly from the 1st application of adequate.
(E) Assume 1 ≁ g 1 ′ 1\not\sim g'_{1} 1 ∼ g 1 ′ . From g 1 ′ ∣ g 1 ′ g'_{1}\mid g'_{1} g 1 ′ ∣ g 1 ′ and the 2nd application of adequate (setting g ‾ 1 = g 1 ′ \overline{g}_{1}=g'_{1} g 1 = g 1 ′ ), we have f = gcd ( g 1 ′ , a g ) ≁ 1 f=\gcd\left(g'_{1},\frac{a}{g}\right)\not\sim1 f = g cd( g 1 ′ , g a ) ∼ 1 .
Since 1 ≁ f = gcd ( g 1 ′ , a g ) ∣ g 1 ′ ∣ g 1 ′ g 2 ′ = g 2 1\not\sim f=\gcd\left(g'_{1},\frac{a}{g}\right)\mid g'_{1}\mid g'_{1}g'_{2}=g_{2} 1 ∼ f = g cd( g 1 ′ , g a ) ∣ g 1 ′ ∣ g 1 ′ g 2 ′ = g 2 , the 1st application of adequate (setting g ‾ 2 = f \overline{g}_{2}=f g 2 = f ) yields f ′ = gcd ( f , b g ) ≁ 1 f'=\gcd\left(f,\frac{b}{g}\right)\not\sim1 f ′ = g cd( f , g b ) ∼ 1 .
But f ′ = gcd ( f , b g ) ∣ b g f'=\gcd\left(f,\frac{b}{g}\right)\mid\frac{b}{g} f ′ = g cd( f , g b ) ∣ g b and f ′ = gcd ( f , b g ) ∣ f = gcd ( g 1 ′ , a g ) ∣ a g f'=\gcd\left(f,\frac{b}{g}\right)\mid f=\gcd\left(g'_{1},\frac{a}{g}\right)\mid\frac{a}{g} f ′ = g cd( f , g b ) ∣ f = g cd( g 1 ′ , g a ) ∣ g a . By (1), we get 1 ≁ f ′ ∣ gcd ( a g , b g ) ∼ 1 1\not\sim f'\mid\gcd\left(\frac{a}{g},\frac{b}{g}\right)\sim1 1 ∼ f ′ ∣ g cd( g a , g b ) ∼ 1 , which by (9) is a contradiction. Therefore g 1 ′ ∼ 1 g'_{1}\sim1 g 1 ′ ∼ 1 .
By (E), only four nontrivial factors remain: l ∼ a g g 1 ′ g 2 ′ b g l\sim\frac{a}{g}g'_{1}g'_{2}\frac{b}{g} l ∼ g a g 1 ′ g 2 ′ g b . We set c = a g g 1 c=\frac{a}{g}g_{1} c = g a g 1 and d = g 2 ′ b g d=g'_{2}\frac{b}{g} d = g 2 ′ g b . Now we apply (12) several times based on the coprimality results (A)–(D):
( α ) gcd ( a g , g 2 ′ ) ∼ ( A ) 1 ⇒ ( 12 ) gcd ( a g , g 2 ′ b g ) ∼ gcd ( a g , b g ) ∼ ( B ) 1 , (\alpha)\ \gcd\left(\frac{a}{g},g'_{2}\right)\mathrel{\overset{(A)}{\sim}}1\mathrel{\overset{(12)}{\Rightarrow}}\gcd\left(\frac{a}{g},g'_{2}\frac{b}{g}\right)\sim\gcd\left(\frac{a}{g},\frac{b}{g}\right)\mathrel{\overset{(B)}{\sim}}1, ( α ) g cd( g a , g 2 ′ ) ∼ ( A ) 1 ⇒ ( 12 ) g cd( g a , g 2 ′ g b ) ∼ g cd( g a , g b ) ∼ ( B ) 1 ,
( β ) gcd ( g 1 , g 2 ′ ) ∼ ( C ) 1 ⇒ ( 12 ) gcd ( g 1 , g 2 ′ b g ) ∼ gcd ( g 1 , b g ) ∼ ( D ) 1 , (\beta)\ \gcd(g_{1},g'_{2})\mathrel{\overset{(C)}{\sim}}1\mathrel{\overset{(12)}{\Rightarrow}}\gcd\left(g_{1},g'_{2}\frac{b}{g}\right)\sim\gcd\left(g_{1},\frac{b}{g}\right)\mathrel{\overset{(D)}{\sim}}1, ( β ) g cd( g 1 , g 2 ′ ) ∼ ( C ) 1 ⇒ ( 12 ) g cd( g 1 , g 2 ′ g b ) ∼ g cd( g 1 , g b ) ∼ ( D ) 1 ,
( γ ) gcd ( a g , g 2 ′ b g ) ∼ ( α ) 1 ⇒ ( 6 ) , ( 12 ) gcd ( c , d ) = gcd ( a g g 1 , g 2 ′ b g ) ∼ gcd ( g 1 , g 2 ′ b g ) ∼ ( β ) 1. (\gamma)\ \gcd\left(\frac{a}{g},g'_{2}\frac{b}{g}\right)\mathrel{\overset{(\alpha)}{\sim}}1\mathrel{\overset{(6),(12)}{\Rightarrow}}\gcd(c,d)=\gcd\left(\frac{a}{g}g_{1},g'_{2}\frac{b}{g}\right)\sim\gcd\left(g_{1},g'_{2}\frac{b}{g}\right)\mathrel{\overset{(\beta)}{\sim}}1. ( γ ) g cd( g a , g 2 ′ g b ) ∼ ( α ) 1 ⇒ ( 6 ) , ( 12 ) g cd( c , d ) = g cd( g a g 1 , g 2 ′ g b ) ∼ g cd( g 1 , g 2 ′ g b ) ∼ ( β ) 1.
Having shown gcd ( c , d ) ∼ 1 \gcd(c,d)\sim1 g cd( c , d ) ∼ 1 , we verify three further properties. c d = a g g 1 g 2 ′ b g ∼ l = lcm ( a , b ) cd=\frac{a}{g}g_{1}g'_{2}\frac{b}{g}\sim l=\operatorname{lcm}(a,b) c d = g a g 1 g 2 ′ g b ∼ l = lcm ( a , b ) by construction. From a = a g g = a g g 1 g 2 a=\frac{a}{g}g=\frac{a}{g}g_{1}g_{2} a = g a g = g a g 1 g 2 , we get c = a g g 1 ∣ a c=\frac{a}{g}g_{1}\mid a c = g a g 1 ∣ a , and finally, from b = g b g = g 1 g 1 ′ g 2 ′ b g b=g\frac{b}{g}=g_{1}g'_{1}g'_{2}\frac{b}{g} b = g g b = g 1 g 1 ′ g 2 ′ g b , we get d = g 2 ′ b g ∣ b d=g'_{2}\frac{b}{g}\mid b d = g 2 ′ g b ∣ b . Since a , b a,b a , b were arbitrary, R R R is an lcm splitting Bézout domain. □
Remark 3.3. An effective method for computing the pair ( c , d ) (c,d) ( c , d ) in the adequate property would yield an effective method for computing lcm splittings. By Thm. C.2 , the same holds for gf.
Bézout domains that are not lcm splitting
First example
4.1. First example. I will give an explicit Bézout domain R R R and elements a , b ∈ R ∖ { 0 } a,b \in R \setminus\{0\} a , b ∈ R ∖ { 0 } that fail lcm splitting. This first example is based on a Bézout domain sketched by van den Dries and Macintyre [20 Sec. 2.10] , which they attribute to Wiegand. The example exhibits a Bézout domain R R R that is not gf, which by Thm. 3.2 and “gf ⇒ \Rightarrow ⇒ adequate” (cf. Rem. 2.3) is a requirement for a counterexample to lcm splitting. The example relies on several constructions of medium complexity from the literature, which are worth making explicit. The failure of gf can be established essentially by topological properties, so [20 Sec. 2.10] can avoid going into details about the constructions, essentially giving the gist of what is needed in a short paragraph. For lcm splitting, one has to dig deeper. As it is quite illustrative, I will present a failed attempt at a counterexample to lcm splitting. I hope this sheds some light on the property and serves as a partial guide to constructing further examples of failure of lcm splitting.
As a starting point, I use a construction of Bézout domains with prescribed topological data; it proceeds in several steps scattered across the literature. Outline:
(A) Fischer [6 ] gives criteria for when a topological space can be the maximal spectrum mSpec ( R ) \operatorname{mSpec}(R) mSpec ( R ) of a Bézout domain R R R , and presents constructions of spectral trees that encode the (prime) spectrum Spec ( R ) \operatorname{Spec}(R) Spec ( R ) of a Bézout domain R R R together with topological information on Spec ( R ) \operatorname{Spec}(R) Spec ( R ) .
(B) Fischer [5 ] earlier gave criteria for when a topological space can be the (prime) spectrum Spec ( R ) \operatorname{Spec}(R) Spec ( R ) of a Bézout domain R R R , and gave a construction of a lattice ordered (abelian) $^{2 }g r o u p group g r o u p G, w h o s e a s s o c i a t e d ∗ ∗ g r o u p s p e c t r u m ∗ ∗ , whose associated **group spectrum** , w h ose a ssoc ia t e d ∗ ∗ g r o u p s p ec t r u m ∗ ∗ \operatorname{Spec}(G) \cong (\operatorname{Spec}(R) \setminus {(0)}) is homeomorphic to the ring spectrum except for the generic point^{3 }$.
(C) These lattice ordered groups G G G serve as demivalue groups and groups of divisibility for a Bézout domain R R R . Jaffard [12 Ch. III.2.2, pp. 78 ff.] gave a construction of a k k k -algebra R R R , for any field k k k , that is a Bézout domain$^{4 }w i t h t h e d e s i r e d with the desired w i t h t h e d es i r e d \operatorname{Spec}(R) \setminus {(0)} \cong \operatorname{Spec}(G)$ as topological spaces. An English textbook version of this part of the construction was given by Gilmer [8 Thm. 18.6, pp. 215 ff.] .
In the construction of a non-gf Bézout domain in [20 Sec. 2.10] , the goal is a topological space in which strong separation properties fail. The idea is to have a countable set of isolated points with two distinct but inseparable “limit points” attached to it.
Definition 4.1. Explicitly, V = { z i : i ∈ N } ∪ { x , y } V = \{z_i : i \in\mathbb{N}\} \cup\{x,y\} V = { z i : i ∈ N } ∪ { x , y } carries the spectral topology in which each z i z_i z i is an isolated point and a basis of open neighborhoods of x x x resp. y y y is given by V m , i = { z n : n ≥ i } ∪ { m } V_{m,i} = \{z_n : n \ge i\} \cup\{m\} V m , i = { z n : n ≥ i } ∪ { m } for each i ∈ N , m ∈ { x , y } i \in\mathbb{N}, m \in\{x,y\} i ∈ N , m ∈ { x , y } separately.
I illustrate the topology in Figure 1 and describe it explicitly. A subset { z i : i ∈ Z } ∪ L ⊆ V \{z_i:i\in Z\}\cup L\subseteq V { z i : i ∈ Z } ∪ L ⊆ V for Z ⊆ N , L ⊆ { x , y } Z\subseteq\mathbb{N},L\subseteq\{x,y\} Z ⊆ N , L ⊆ { x , y } in the spectral topology on V V V is
Figure 1. The space V V V and several open sets (circles and “tubes”) in the spectral topology.
closed iff ( L = { x , y } (L=\{x,y\} ( L = { x , y } and Z Z Z arbitrary) or ( L ≠ { x , y } (L\ne\{x,y\} ( L = { x , y } and Z Z Z finite),
open iff ( L = ∅ (L=\varnothing ( L = ∅ and Z Z Z arbitrary) or ( L ≠ ∅ (L\ne\varnothing ( L = ∅ and Z Z Z cofinite),
quasi-compact open iff ( L = ∅ (L=\varnothing ( L = ∅ and Z Z Z finite) or ( L ≠ ∅ (L\ne\varnothing ( L = ∅ and Z Z Z cofinite).
Fischer [6 Prop. 2.1] shows that the maximal spectrum of any Bézout domain is a B-space, so I repeat the definition here. In fact, the whole construction will make use of several topological notions; for easier reference, I define the less familiar ones here.
Definition 4.2.
A spectral space is a topological space homeomorphic to a ring spectrum Spec ( R ) \operatorname{Spec}(R) Spec ( R ) for some ring R R R . All spectral spaces are sober, in particular T 0 T_0 T 0 .
Given any topology (e.g. a spectral topology), the associated patch topology has as a basis of closed sets all sets that are either (a) closed in the original topology or (b) quasi-compact open in the original topology. A closed set in the patch topology is called a patch . Often patches are not viewed with the induced subspace patch topology, but with the subspace original topology.
We define the specialization order for points x , y ∈ X x,y\in X x , y ∈ X of a topological space by
x ≥ s y ⟺ x ∈ { y } ‾ . x\geq_s y\Longleftrightarrow x\in\overline{\{y\}}. x ≥ s y ⟺ x ∈ { y } . ≥ s \geq_s ≥ s is a partial order and for x ≥ s y x\geq_s y x ≥ s y we say x x x is a specialization of y y y resp. y y y is a generization of x x x . The specialization order adds no new information to a topology, but sometimes allows for more concise descriptions. We set the generization set for x x x to be G ( x ) = { y ∈ X : x ≥ s y } G(x)=\{y\in X:x\geq_s y\} G ( x ) = { y ∈ X : x ≥ s y } . If, for all x ∈ X x\in X x ∈ X , the generization set G ( x ) G(x) G ( x ) is totally ordered by the specialization order, we say that X X X is a tree .5 A spectral space that is a tree is called a spectral tree .
A topological basis of quasi-compact open sets is called a QCI basis (quasi-compact with intersection property) if the intersection of any finitely many members of the basis again lies in the basis.
The elements of the power set P ( X ) \mathcal{P}(X) P ( X ) form the Boolean algebra of subsets of X X X (with operations union, intersection and complement). Consider the Boolean subalgebra A ( X ) ⊆ P ( X ) A(X)\subseteq\mathcal{P}(X) A ( X ) ⊆ P ( X ) generated by quasi-compact open sets for the (original) topology on X X X . We call elements of A ( X ) A(X) A ( X ) constructible . If X X X is a spectral tree, the constructible sets (for the spectral topology) are exactly the clopen (closed and open) subsets of X X X with respect to the patch topology.
Let X X X be a topological space. For every integer n ≥ 1 n\geq1 n ≥ 1 , a sequence ( x 1 , … , x n ) ∈ X n (x_1,\ldots,x_n)\in X^n ( x 1 , … , x n ) ∈ X n is said to be sticky provided ⋂ i = 1 n N i ≠ ∅ \bigcap_{i=1}^{n}N_i\ne\varnothing ⋂ i = 1 n N i = ∅ whenever N i N_i N i is a neighborhood of x i x_i x i . A quasi-compact T1 space is called a B-space provided every closed subspace Y Y Y has the following property: If ( y i − 1 , y i ) ∈ Y 2 (y_{i-1},y_i)\in Y^2 ( y i − 1 , y i ) ∈ Y 2 is sticky for 2 ≤ i ≤ n 2\leq i\leq n 2 ≤ i ≤ n , then ( y 1 , … , y n ) (y_1,\ldots,y_n) ( y 1 , … , y n ) is sticky.
I restate Fischer [6 Prop. 3.3] for a topological space X X X :
Proposition 4.3. Suppose X X X has only finitely many non-isolated points. Then X X X is homeomorphic to the maximal ideal space of a finite-dimensional Bézout domain if and only if
( 1 ) (1) ( 1 ) X X X is a B-space, and
( 2 ) (2) ( 2 ) X X X has an open basis consisting of quasi-compact sets.
My intermediate aim is to follow [20 Sec. 2.10] and make the failure of g f gf g f explicit. We need
Lemma 4.4. Let R R R be a Bézout ring and X = mSpec ( R ) X = \operatorname{mSpec}(R) X = mSpec ( R ) its maximal spectrum as a topological space. If C ⊂ X C \subset X C ⊂ X is closed with quasi-compact complement X ∖ C X \setminus C X ∖ C , then C = Z ( r ) C = Z(r) C = Z ( r ) is the zero set of a single ring element r ∈ R r \in R r ∈ R (resp. of its principal ideal ( r ) (r) ( r ) ).
Proof. In the Zariski topology (for the maximal spectrum as well as the prime spectrum), the complements of zero sets of single ring elements D ( s ) = X ∖ Z ( s ) D(s) = X \setminus Z(s) D ( s ) = X ∖ Z ( s ) form an open basis. In a Bézout ring where any sum of two principal ideals is again principal, we have ( a , b ) = ( a ) + ( b ) = ( g ) (a,b) = (a) + (b) = (g) ( a , b ) = ( a ) + ( b ) = ( g ) for g = gcd ( a , b ) g = \operatorname{gcd}(a,b) g = gcd ( a , b ) ; note that this requires exactly Bézout and will fail for general GCD rings. In particular, the union of two basic open sets is again a basic open set:
D ( a ) ∪ D ( b ) = ( X ∖ Z ( a ) ) ∪ ( X ∖ Z ( b ) ) = X ∖ ( Z ( a ) ∩ Z ( b ) ) = X ∖ Z ( a , b ) = X ∖ Z ( g ) = D ( g ) . D(a) \cup D(b) = (X \setminus Z(a)) \cup(X \setminus Z(b)) = X \setminus(Z(a) \cap Z(b)) = X \setminus Z(a,b) = X \setminus Z(g) = D(g). D ( a ) ∪ D ( b ) = ( X ∖ Z ( a )) ∪ ( X ∖ Z ( b )) = X ∖ ( Z ( a ) ∩ Z ( b )) = X ∖ Z ( a , b ) = X ∖ Z ( g ) = D ( g ) .
By induction, for any finite collection ( a i ) i = 1 n ∈ R n (a_i)_{i=1}^n \in R^n ( a i ) i = 1 n ∈ R n , we can find an r ∈ R r \in R r ∈ R such that ⋃ i = 1 n D ( a i ) = D ( r ) \bigcup_{i=1}^n D(a_i) = D(r) ⋃ i = 1 n D ( a i ) = D ( r ) .
Now we take the closed C C C from the assumptions. Its complement X ∖ C X \setminus C X ∖ C is quasi-compact open. Since the sets of the form D ( a ) D(a) D ( a ) are an open basis in the spectral topology, there is some index set I \mathcal{I} I and a family of open D ( a i ) ⊆ X ∖ C D(a_i) \subseteq X \setminus C D ( a i ) ⊆ X ∖ C , one for each i ∈ I i \in\mathcal{I} i ∈ I , such that their union ⋃ i ∈ I D ( a i ) = X ∖ C \bigcup_{i\in\mathcal{I}} D(a_i) = X \setminus C ⋃ i ∈ I D ( a i ) = X ∖ C . Since X ∖ C X \setminus C X ∖ C is quasi-compact, we can choose a finite subset of I \mathcal{I} I to do the job, i.e., after relabeling we can assume I = { 1 , … , n } \mathcal{I} = \{1,\ldots,n\} I = { 1 , … , n } for some n ∈ N n \in\mathbb{N} n ∈ N and X ∖ C = ⋃ i = 1 n D ( a i ) = D ( r ) X \setminus C = \bigcup_{i=1}^n D(a_i) = D(r) X ∖ C = ⋃ i = 1 n D ( a i ) = D ( r ) , where the last equality uses the Bézout ring property derived in the previous paragraph. But X ∖ C = D ( r ) X \setminus C = D(r) X ∖ C = D ( r ) is by definition equivalent to C = Z ( r ) C = Z(r) C = Z ( r ) . This finishes the proof. □
Proposition 4.5. For the space V V V from Def. 4.1 there is a Bézout domain R V R_V R V with mSpec ( R V ) ≅ V \operatorname{mSpec}(R_V) \cong V mSpec ( R V ) ≅ V such that g f gf g f fails for some a ~ , b ~ ∈ R V ∖ { 0 } \tilde{a},\tilde{b} \in R_V \setminus\{0\} a ~ , b ~ ∈ R V ∖ { 0 } . In particular, g f gf g f fails in some Bézout domain.
Proof. We check that V V V satisfies all requirements of Prop. 4.3 .
V V V is a T1 space. Any z i z_i z i is isolated, which proves separation for any pair of distinct z i z_i z i . Separating a z i z_i z i from some m ∈ { x , y } m \in\{x,y\} m ∈ { x , y } follows the same argument. To separate m ∈ { x , y } m \in\{x,y\} m ∈ { x , y } from z i z_i z i use V m , i + 1 V_{m,i+1} V m , i + 1 . x x x is separated from y y y by V x , 0 V_{x,0} V x , 0 , and y y y is separated from x x x by V y , 0 V_{y,0} V y , 0 .
V V V is quasi-compact. To cover x ∈ V x \in V x ∈ V , any open cover needs to contain an open neighborhood containing some V x , i V_{x,i} V x , i , leaving only finitely many points in V V V uncovered. A finite subcover is immediate.
Since all z i z_i z i are isolated, any sticky sequence can only involve x x x and y y y , and ( x , y ) (x,y) ( x , y ) is indeed sticky. The implication in the B-space condition, which is only non-vacuous if more than two distinct points can form sequences from pairwise sticky sequences, is trivially satisfied.
Having described all open and quasi-compact open sets above, it is straightforward to see that all quasi-compact opens are a topological basis. A more manageable quasi-compact open basis is given by ⋃ i = 0 ∞ { { z i } , V x , i , V y , i } \bigcup_{i=0}^{\infty}\{\{z_i\},V_{x,i},V_{y,i}\} ⋃ i = 0 ∞ {{ z i } , V x , i , V y , i } . Note that V V V does not have a QCI basis.
The set of non-isolated points is { x , y } \{x,y\} { x , y } and thus is finite.
Therefore, Fischer’s proposition asserts that there is a Bézout domain R V R_V R V with mSpec ( R V ) ≅ V \operatorname{mSpec}(R_V) \cong V mSpec ( R V ) ≅ V .
We look at the two sets A = V ∖ { z 0 } A = V \setminus\{z_0\} A = V ∖ { z 0 } and B = { y } B = \{y\} B = { y } . By construction, z 0 z_0 z 0 is isolated, which means that its complement A A A is closed. B = { y } = V ∖ ( { z i : i ∈ N } ∪ { x } ) = V ∖ V x , 0 B = \{y\} = V \setminus(\{z_i : i \in\mathbb{N}\} \cup\{x\}) = V \setminus V_{x,0} B = { y } = V ∖ ({ z i : i ∈ N } ∪ { x }) = V ∖ V x , 0 where V x , 0 V_{x,0} V x , 0 is a basic open. Thus B B B also is closed. The complement of A A A is V ∖ A = V ∖ ( V ∖ { z 0 } ) = { z 0 } V \setminus A = V \setminus(V \setminus\{z_0\}) = \{z_0\} V ∖ A = V ∖ ( V ∖ { z 0 }) = { z 0 } , an open singleton, hence clearly quasi-compact open. The complement of B B B is V ∖ B = V ∖ { y } = V x , 0 V \setminus B = V \setminus\{y\} = V_{x,0} V ∖ B = V ∖ { y } = V x , 0 . Any cover by basic open sets needs to cover x ∈ V x , 0 x \in V_{x,0} x ∈ V x , 0 . Hence the cover needs to contain at least one neighborhood containing some V x , n V_{x,n} V x , n that form the basic open neighborhoods of x x x . Take such an n n n . But V x , n V_{x,n} V x , n already covers cofinally much of V x , 0 V_{x,0} V x , 0 , i.e., starting from that cover and using V x , n V_{x,n} V x , n , we only need to choose at most one more open set of the cover for each of the z i z_i z i for i < n i < n i < n . Hence we can select a subcover of at most 1 + n 1+n 1 + n open sets, i.e. a finite cover. Thus B B B also has quasi-compact complement. In summary, A A A and B B B are both closed with quasi-compact complement.
By Lem. 4.4 , there are ring elements a ~ , b ~ \tilde{a},\tilde{b} a ~ , b ~ such that Z ( a ~ ) = A Z(\tilde{a})=A Z ( a ~ ) = A , Z ( b ~ ) = B Z(\tilde{b})=B Z ( b ~ ) = B . Van den Dries and Macintyre [20 Sec. 2.10] imply that these a ~ , b ~ \tilde{a},\tilde{b} a ~ , b ~ are not compatible with good factorization. Indeed, assume there were c ~ , d ~ ∈ R V \tilde{c},\tilde{d}\in R_V c ~ , d ~ ∈ R V such that a ~ = c ~ d ~ \tilde{a}=\tilde{c}\tilde{d} a ~ = c ~ d ~ , gcd ( c ~ , b ~ ) ∼ 1 \gcd(\tilde{c},\tilde{b})\sim1 g cd( c ~ , b ~ ) ∼ 1 and b ~ ∈ ( d ~ ) \tilde{b}\in\sqrt{(\tilde{d})} b ~ ∈ ( d ~ ) . Topologically, this means Z ( c ~ ) ∪ Z ( d ~ ) = Z ( a ~ ) = A = V ∖ { z 0 } Z(\tilde{c})\cup Z(\tilde{d})=Z(\tilde{a})=A=V\setminus\{z_0\} Z ( c ~ ) ∪ Z ( d ~ ) = Z ( a ~ ) = A = V ∖ { z 0 } , ∅ = Z ( c ~ ) ∩ Z ( b ~ ) = Z ( c ~ ) ∩ B = Z ( c ~ ) ∩ { y } \varnothing=Z(\tilde{c})\cap Z(\tilde{b})=Z(\tilde{c})\cap B=Z(\tilde{c})\cap\{y\} ∅ = Z ( c ~ ) ∩ Z ( b ~ ) = Z ( c ~ ) ∩ B = Z ( c ~ ) ∩ { y } and { y } = B = Z ( b ~ ) ⊇ Z ( d ~ ) \{y\}=B=Z(\tilde{b})\supseteq Z(\tilde{d}) { y } = B = Z ( b ~ ) ⊇ Z ( d ~ ) . Taking everything together, this implies Z ( c ~ ) = V x , 1 Z(\tilde{c})=V_{x,1} Z ( c ~ ) = V x , 1 is a basic open set, which, however, is not closed (see the description of the topology above) – this is a contradiction, since a set of the form Z ( c ~ ) Z(\tilde{c}) Z ( c ~ ) by definition must be closed. Thus a ~ , b ~ \tilde{a},\tilde{b} a ~ , b ~ do not have a good factorization, finishing the example from [20 Sec. 2.10] . □
It is remarkable that one does not need any explicit algebraic knowledge of the ring R V R_V R V to prove that it does not have gf. In dialog with an LLM, I came up with the following approach in the same spirit of avoiding explicit algebraic descriptions of R V R_V R V . Although this approach fails to demonstrate the failure of lcm splitting, it is nevertheless instructive.
In Rem. 2.3, I mentioned that a domain R R R is Bézout if and only if it is a Prüfer GCD domain. Since all localizations of Prüfer domains are valuation rings, we get for each maximal ideal m ∈ mSpec ( R ) \mathfrak{m}\in\operatorname{mSpec}(R) m ∈ mSpec ( R ) a valuation v m v_{\mathfrak{m}} v m on R R R by restriction. These valuations behave well with respect to gcds and lcms. Explicitly, for a , b ∈ R ∖ { 0 } a,b\in R\setminus\{0\} a , b ∈ R ∖ { 0 } , we have
v m ( gcd ( a , b ) ) = min ( v m ( a ) , v m ( b ) ) and v m ( lcm ( a , b ) ) = max ( v m ( a ) , v m ( b ) ) . v_{\mathfrak{m}}(\gcd(a,b))=\min\left(v_{\mathfrak{m}}(a),v_{\mathfrak{m}}(b)\right)\quad\text{and}\quad v_{\mathfrak{m}}(\operatorname{lcm}(a,b))=\max\left(v_{\mathfrak{m}}(a),v_{\mathfrak{m}}(b)\right). v m ( g cd( a , b )) = min ( v m ( a ) , v m ( b ) ) and v m ( lcm ( a , b )) = max ( v m ( a ) , v m ( b ) ) .
Lemma 4.6. Let R R R be a Bézout domain, let a , b ∈ R ∖ { 0 } a,b\in R\setminus\{0\} a , b ∈ R ∖ { 0 } , and let c , d ∈ R c,d\in R c , d ∈ R satisfy lcm splitting for a , b a,b a , b , i.e. lcm ( a , b ) ∼ c d \operatorname{lcm}(a,b)\sim cd lcm ( a , b ) ∼ c d , c ∣ a c\mid a c ∣ a , d ∣ b d\mid b d ∣ b , gcd ( c , d ) ∼ 1 \gcd(c,d)\sim1 g cd( c , d ) ∼ 1 . Then, for any m ∈ mSpec ( R ) \mathfrak{m}\in\operatorname{mSpec}(R) m ∈ mSpec ( R ) , we have
v m ( a ) > v m ( b ) ⇒ ( v m ( c ) = v m ( a ) ∧ v m ( d ) = 0 ) and v m ( a ) < v m ( b ) ⇒ ( v m ( c ) = 0 ∧ v m ( d ) = v m ( b ) ) . v_{\mathfrak{m}}(a)>v_{\mathfrak{m}}(b)\Rightarrow\left(v_{\mathfrak{m}}(c)=v_{\mathfrak{m}}(a)\land v_{\mathfrak{m}}(d)=0\right)\quad\text{and}\quad v_{\mathfrak{m}}(a)<v_{\mathfrak{m}}(b)\Rightarrow\left(v_{\mathfrak{m}}(c)=0\land v_{\mathfrak{m}}(d)=v_{\mathfrak{m}}(b)\right). v m ( a ) > v m ( b ) ⇒ ( v m ( c ) = v m ( a ) ∧ v m ( d ) = 0 ) and v m ( a ) < v m ( b ) ⇒ ( v m ( c ) = 0 ∧ v m ( d ) = v m ( b ) ) .
Proof. Wlog, we assume v m ( a ) > v m ( b ) v_{\mathfrak{m}}(a)>v_{\mathfrak{m}}(b) v m ( a ) > v m ( b ) ; the second part follows by symmetry. Assumption gcd ( c , d ) ∼ 1 \gcd(c,d)\sim1 g cd( c , d ) ∼ 1 implies
0 = v m ( 1 ) = v m ( gcd ( c , d ) ) = min ( v m ( c ) , v m ( d ) ) 0=v_{\mathfrak{m}}(1)=v_{\mathfrak{m}}(\gcd(c,d))=\min\left(v_{\mathfrak{m}}(c),v_{\mathfrak{m}}(d)\right) 0 = v m ( 1 ) = v m ( g cd( c , d )) = min ( v m ( c ) , v m ( d ) )
and lcm ( a , b ) ∼ c d \operatorname{lcm}(a,b)\sim cd lcm ( a , b ) ∼ c d implies
0 ≤ v m ( b ) < v m ( a ) = max ( v m ( a ) , v m ( b ) ) = v m ( lcm ( a , b ) ) = v m ( c d ) = v m ( c ) + v m ( d ) . 0\leq v_{\mathfrak{m}}(b)<v_{\mathfrak{m}}(a)=\max\left(v_{\mathfrak{m}}(a),v_{\mathfrak{m}}(b)\right)=v_{\mathfrak{m}}(\operatorname{lcm}(a,b))=v_{\mathfrak{m}}(cd)=v_{\mathfrak{m}}(c)+v_{\mathfrak{m}}(d). 0 ≤ v m ( b ) < v m ( a ) = max ( v m ( a ) , v m ( b ) ) = v m ( lcm ( a , b )) = v m ( c d ) = v m ( c ) + v m ( d ) .
From the first sequence of equalities, we have either v m ( c ) = 0 v_{\mathfrak{m}}(c)=0 v m ( c ) = 0 or v m ( d ) = 0 v_{\mathfrak{m}}(d)=0 v m ( d ) = 0 or both; the second inequality rules out the third option. If we had v m ( c ) = 0 v_{\mathfrak{m}}(c)=0 v m ( c ) = 0 , then v m ( d ) = v m ( a ) v_{\mathfrak{m}}(d)=v_{\mathfrak{m}}(a) v m ( d ) = v m ( a ) , but d ∣ b d\mid b d ∣ b implies v m ( d ) ≤ v m ( b ) < v m ( a ) v_{\mathfrak{m}}(d)\leq v_{\mathfrak{m}}(b)<v_{\mathfrak{m}}(a) v m ( d ) ≤ v m ( b ) < v m ( a ) , a contradiction. Thus the only remaining possibility is v m ( d ) = 0 v_{\mathfrak{m}}(d)=0 v m ( d ) = 0 and v m ( c ) = v m ( a ) v_{\mathfrak{m}}(c)=v_{\mathfrak{m}}(a) v m ( c ) = v m ( a ) . This proves the lemma. □
Now I come back to the ring R V R_V R V with mSpec ( R V ) ≅ V \operatorname{mSpec}(R_V)\cong V mSpec ( R V ) ≅ V and a ~ , b ~ ∈ R V \tilde{a},\tilde{b}\in R_V a ~ , b ~ ∈ R V as constructed in the proof of Prop. 4.5 . Since mSpec ( R V ) ≅ V \operatorname{mSpec}(R_V)\cong V mSpec ( R V ) ≅ V , we will write the maximal ideals as x , y , z i x,y,z_i x , y , z i . Clearly a ~ , b ~ \tilde{a},\tilde{b} a ~ , b ~ are nonzero; otherwise their vanishing varieties would not be proper subsets of V V V . Suppose that c ~ , d ~ \tilde{c},\tilde{d} c ~ , d ~ form an lcm splitting of a ~ , b ~ \tilde{a},\tilde{b} a ~ , b ~ . Recall that the vanishing sets are Z ( a ~ ) = A = { z i : i ∈ N 0 } ∪ { x , y } Z(\tilde{a})=A=\{z_i:i\in\mathbb{N}_0\}\cup\{x,y\} Z ( a ~ ) = A = { z i : i ∈ N 0 } ∪ { x , y } and Z ( b ~ ) = B = { y } Z(\tilde{b})=B=\{y\} Z ( b ~ ) = B = { y } . We need to consider three cases.
m = z 0 \mathfrak{m}=z_0 m = z 0 . Here we have v m ( a ~ ) = 0 = v m ( b ~ ) v_{\mathfrak{m}}(\tilde{a})=0=v_{\mathfrak{m}}(\tilde{b}) v m ( a ~ ) = 0 = v m ( b ~ ) . The lemma does not say anything about this case, but it is also irrelevant.
m ∈ { z i : i ∈ N 0 } ∪ { x } \mathfrak{m}\in\{z_i:i\in\mathbb{N}_0\}\cup\{x\} m ∈ { z i : i ∈ N 0 } ∪ { x } . We have v m ( a ~ ) > 0 = v m ( b ~ ) v_{\mathfrak{m}}(\tilde{a})>0=v_{\mathfrak{m}}(\tilde{b}) v m ( a ~ ) > 0 = v m ( b ~ ) , and the lemma implies v m ( c ~ ) = v m ( a ~ ) > 0 = v m ( d ~ ) v_{\mathfrak{m}}(\tilde{c})=v_{\mathfrak{m}}(\tilde{a})>0=v_{\mathfrak{m}}(\tilde{d}) v m ( c ~ ) = v m ( a ~ ) > 0 = v m ( d ~ ) .
m = y \mathfrak{m}=y m = y . We have v m ( a ~ ) > 0 < v m ( b ~ ) v_{\mathfrak{m}}(\tilde{a})>0<v_{\mathfrak{m}}(\tilde{b}) v m ( a ~ ) > 0 < v m ( b ~ ) and the lemma is not directly applicable. If we had v m ( a ~ ) ≥ v m ( b ~ ) > 0 v_{\mathfrak{m}}(\tilde{a})\geq v_{\mathfrak{m}}(\tilde{b})>0 v m ( a ~ ) ≥ v m ( b ~ ) > 0 , this would imply that b ~ \tilde{b} b ~ has strictly smaller valuations than a ~ \tilde{a} a ~ at most places, with possible equality at z 0 z_0 z 0 and y y y . In any case, this would imply b ~ ∣ a ~ \tilde{b}\mid\tilde{a} b ~ ∣ a ~ , and an lcm splitting would trivially be given by ( c ~ , d ~ ) = ( a ~ , 1 ) (\tilde{c},\tilde{d})=(\tilde{a},1) ( c ~ , d ~ ) = ( a ~ , 1 ) . Therefore, in order to have a chance of demonstrating failure of lcm splitting, we would need 0 < v m ( a ~ ) < v m ( b ~ ) 0<v_{\mathfrak{m}}(\tilde{a})<v_{\mathfrak{m}}(\tilde{b}) 0 < v m ( a ~ ) < v m ( b ~ ) at the only place where this is at all possible, i.e. m = y \mathfrak{m}=y m = y .
By the abstract description used so far, we only know v m ( a ~ ) > 0 < v m ( b ~ ) v_m(\tilde{a}) > 0 < v_m(\tilde{b}) v m ( a ~ ) > 0 < v m ( b ~ ) for sure. If 0 < v m ( a ~ ) < v m ( b ~ ) 0 < v_m(\tilde{a}) < v_m(\tilde{b}) 0 < v m ( a ~ ) < v m ( b ~ ) , then we are en route to the failure of lcm splitting. If v m ( a ~ ) ≥ v m ( b ~ ) > 0 v_m(\tilde{a}) \geq v_m(\tilde{b}) > 0 v m ( a ~ ) ≥ v m ( b ~ ) > 0 , a dialog with an LLM suggested taking a high enough power of b ~ \tilde{b} b ~ , i.e. some N ∈ N N \in\mathbb{N} N ∈ N with 0 < v m ( a ~ ) < v m ( b ~ N ) 0 < v_m(\tilde{a}) < v_m(\tilde{b}^{N}) 0 < v m ( a ~ ) < v m ( b ~ N ) , and using b ~ N \tilde{b}^{N} b ~ N instead of b ~ \tilde{b} b ~ . That looked very promising. However, for non-Noetherian domains R R R like the ones we are interested in, in particular R V R_V R V , the value groups need not be Archimedean (i.e. embeddable in R \mathbb{R} R , also called rank-1 valuations) . Instead, we could have higher-rank valuations for which taking high enough powers is not guaranteed to succeed.
At this point, the abstract approach got stuck. In retrospect, it turned out that R V R_V R V , explicitly constructed following Fischer’s general approach, does have a valuation of rank 2 at m = y \mathfrak{m}=y m = y and that a ~ , b ~ \tilde{a},\tilde{b} a ~ , b ~ as described by van den Dries and Macintyre actually satisfy lcm splitting, as we will see later. The elegant abstract approach succeeded in showing failure of gf, but to exhibit failure of lcm splitting, one has to work through the explicit algebraic construction of R V R_V R V , which I will do now. I try to keep the general construction and the effect on our example as separate as possible using ●V _V V indication (and later ●M _M M ) unless the notation becomes too cluttered.
For a topological space X X X , denote the set of isolated points by X i s o ⊆ X X^{\mathrm{iso}} \subseteq X X iso ⊆ X and set X ′ = X ∖ X i s o X'=X\setminus X^{\mathrm{iso}} X ′ = X ∖ X iso . The assumptions of [6 Prop. 3.3] include in particular that X X X may have only finitely many non-isolated points, i.e. that N = ∣ X ′ ∣ < ∞ N=\lvert X'\rvert<\infty N = ∣ X ′ ∣ < ∞ , and the construction proceeds by induction on N N N with base cases being N ∈ { 0 , 1 } N\in\{0,1\} N ∈ { 0 , 1 } . For V V V , we have V ′ = V ∖ V i s o = { x , y } V'=V\setminus V^{\mathrm{iso}}=\{x,y\} V ′ = V ∖ V iso = { x , y } , so we need to follow one induction step. The induction distinguishes between all elements of the current exceptional induction set X ′ = { x 1 , x 2 , … , x N } X'=\{x_1,x_2,\ldots,x_N\} X ′ = { x 1 , x 2 , … , x N } forming a sticky sequence, as defined in Def. 4.2 , or not. For us, ( x , y ) (x,y) ( x , y ) forms a sticky sequence, as any open neighborhoods of x x x resp. y y y intersect, since the basic open neighborhoods intersect nontrivially: V x , m ∩ V y , n = { z i : i ≥ max ( m , n ) } ≠ ∅ V_{x,m}\cap V_{y,n}=\{z_i:i\geq\max(m,n)\}\neq\varnothing V x , m ∩ V y , n = { z i : i ≥ max ( m , n )} = ∅ .
The construction then lets one choose quasi-compact open neighborhoods for each x i ∈ X ′ x_i\in X' x i ∈ X ′ . I take them as large as possible: α x = V x , 0 , α y = V y , 0 \alpha_x=V_{x,0},\alpha_y=V_{y,0} α x = V x , 0 , α y = V y , 0 , which yields a subset X ⊇ Ω X\supseteq\Omega X ⊇ Ω , in our case Ω V = { z i : i ∈ N } \Omega_V=\{z_i:i\in\mathbb{N}\} Ω V = { z i : i ∈ N } . The complement X ∖ Ω X\setminus\Omega X ∖ Ω , which in our case is { x , y } \{x,y\} { x , y } and coincides by coincidence with V ′ V' V ′ (in general only X ′ ⊆ X ∖ Ω X'\subseteq X\setminus\Omega X ′ ⊆ X ∖ Ω is guaranteed), then loses stickiness (fewer pairs in X ′ X' X ′ are sticky in the subspace topology of $X\setminus\Omega than in the topology of X X X itself), and the other case of the induction is employed to incorporate the set X ∖ Ω X\setminus\Omega X ∖ Ω into a suitable spectral tree Y 1 Y_1 Y 1 . In our case, Y 1 = V ∖ Ω V = { x , y } Y_1=V\setminus\Omega_V=\{x,y\} Y 1 = V ∖ Ω V = { x , y } already works without having to add anything additionally (recall that forests with several connected components are possible for spectral trees). Now, back in the sticky branch of the induction, we add points Γ = { p x : x ∈ X ∖ X ′ } ∪ { p } \Gamma=\{p_x:x\in X\setminus X'\}\cup\{p\} Γ = { p x : x ∈ X ∖ X ′ } ∪ { p } , which will be generalizations of the established points of X X X . In our case, we write p i p_i p i for p z i p_{z_i} p z i for simplicity. For the suitably extended topology, we will then have z i > s p i z_i>_s p_i z i > s p i for all i ∈ N i\in\mathbb{N} i ∈ N as well as x > s p < s y x>_s p<_s y x > s p < s y (recall the specialization order from Def. 4.2 ). Fischer also introduces the notation Θ = Ω ∪ Γ \Theta=\Omega\cup\Gamma Θ = Ω ∪ Γ . The complete set for the induction step is given by Y = Θ ∪ Y 1 Y=\Theta\cup Y_1 Y = Θ ∪ Y 1 .
Fischer’s construction would in general proceed inductively like this, but since our exceptional set at the start V ′ = { x , y } V'=\{x,y\} V ′ = { x , y } had only two elements, the construction already stops here. At the end, an additional generic point ( 0 ) (0) ( 0 ) is added, which is the generalization of all points constructed so far – this is described in the proof of [5 Thm. 2.1] . This makes sure that the spectral space belongs to an integral domain and not an arbitrary ring. I summarize the resulting space M ′ = M ∪ { ( 0 ) } M' = M \cup\{(0)\} M ′ = M ∪ {( 0 )} in the following definition, describe the topology and illustrate the space in Figure 2 .
Figure 2. The spectral tree M ′ = M ∪ { ( 0 ) } M'=M\cup\{(0)\} M ′ = M ∪ {( 0 )} with various named sets and specialization order indicated.
Definition 4.7. The space M = { z i , p i : i ∈ N } ∪ { x , y , p } ⊃ V M = \{z_i,p_i : i \in\mathbb{N}\} \cup\{x,y,p\} \supset V M = { z i , p i : i ∈ N } ∪ { x , y , p } ⊃ V is endowed with the topology induced by the following open basis B \mathcal{B} B (the numbering follows the general construction in [6 p. 252] and seems to be off at iii. for this reason):
i. { p i } \{p_i\} { p i } for each i ∈ N i \in\mathbb{N} i ∈ N ,
ii. { p i , z i } \{p_i,z_i\} { p i , z i } for each i ∈ N i \in\mathbb{N} i ∈ N ,
iii. no further open sets of this type,
iv. { p , w } ∪ { p i : i ≥ m } ∪ { z i : i ≥ n } \{p,w\} \cup\{p_i : i \ge m\} \cup\{z_i : i \ge n\} { p , w } ∪ { p i : i ≥ m } ∪ { z i : i ≥ n } for each pair of integers 0 ≤ m ≤ n 0 \le m \le n 0 ≤ m ≤ n and w ∈ { x , y } w \in\{x,y\} w ∈ { x , y } ,
v. { p } ∪ { p i : i ≥ m } ∪ { z i : i ≥ n } \{p\} \cup\{p_i : i \ge m\} \cup\{z_i : i \ge n\} { p } ∪ { p i : i ≥ m } ∪ { z i : i ≥ n } for each pair of integers 0 ≤ m ≤ n 0 \le m \le n 0 ≤ m ≤ n .
Set M ′ = M ∪ { ( 0 ) } M' = M \cup\{(0)\} M ′ = M ∪ {( 0 )} with open basis B ′ = { U ∪ { ( 0 ) } : U ∈ B } \mathcal{B}' = \{U \cup\{(0)\} : U \in\mathcal{B}\} B ′ = { U ∪ {( 0 )} : U ∈ B } .
It will be useful to describe several aspects of the topology explicitly. A subset { z i : i ∈ Z } ∪ { p i : i ∈ P } ∪ L ⊆ M \{z_i : i \in Z\} \cup\{p_i : i \in P\} \cup L \subseteq M { z i : i ∈ Z } ∪ { p i : i ∈ P } ∪ L ⊆ M for Z , P ⊆ N Z, P \subseteq\mathbb{N} Z , P ⊆ N , L ⊆ { x , y , p } L \subseteq\{x,y,p\} L ⊆ { x , y , p } in the spectral topology on M M M is
closed iff ( L = { x , y , p } , Z ⊇ P (L = \{x,y,p\}, Z \supseteq P ( L = { x , y , p } , Z ⊇ P and otherwise arbitrary) or ( p ∉ L (p \notin L ( p ∈ / L and Z ⊇ P Z \supseteq P Z ⊇ P finite),
open iff ( L = ∅ , Z ⊆ P (L = \varnothing, Z \subseteq P ( L = ∅ , Z ⊆ P and otherwise arbitrary) or ( p ∈ L (p \in L ( p ∈ L and Z ⊆ P Z \subseteq P Z ⊆ P cofinite),
quasi-compact open iff ( L = ∅ (L = \varnothing ( L = ∅ and Z ⊆ P Z \subseteq P Z ⊆ P finite) or ( p ∈ L (p \in L ( p ∈ L and Z ⊆ P Z \subseteq P Z ⊆ P cofinite).
In particular, all { z i } \{z_i\} { z i } are closed, as are { x } \{x\} { x } , { y } \{y\} { y } , and all { p i } \{p_i\} { p i } are quasi-compact open (type i. in Def. 4.7). Therefore, these singletons are closed in the patch topology. Similarly, their complements are also closed in the patch topology. Thus every point of M M M except p p p is isolated in the patch topology. A subset { z i : i ∈ Z } ∪ { p i : i ∈ P } ∪ L ⊆ M \{z_i : i \in Z\} \cup\{p_i : i \in P\} \cup L \subseteq M { z i : i ∈ Z } ∪ { p i : i ∈ P } ∪ L ⊆ M for Z , P ⊆ N Z, P \subseteq\mathbb{N} Z , P ⊆ N , L ⊆ { x , y , p } L \subseteq\{x,y,p\} L ⊆ { x , y , p } in the patch topology on M M M is
closed iff p ∈ L p \in L p ∈ L or ( p ∉ L (p \notin L ( p ∈ / L and Z , P Z, P Z , P finite),
open iff p ∉ L p \notin L p ∈ / L or ( p ∈ L (p \in L ( p ∈ L and Z , P Z, P Z , P cofinite),
quasi-compact open iff ( p ∉ L (p \notin L ( p ∈ / L and Z , P Z, P Z , P finite) or ( p ∈ L (p \in L ( p ∈ L and Z , P Z, P Z , P cofinite).
The constructible sets for the spectral topology on M M M are the clopen sets for the patch topology. Thus a set in the above presentation is constructible iff p ∈ L p \in L p ∈ L and Z , P Z, P Z , P cofinite or p ∉ L p \notin L p ∈ / L and Z , P Z, P Z , P finite. In this special case, for X = M X = M X = M , a set is constructible in the spectral topology on M M M iff it is quasi-compact open in the patch topology on M M M .
Remark 4.8. The proof of [6 Prop. 3.3] guarantees that M ′ ∩ V ≅ V M' \cap V \cong V M ′ ∩ V ≅ V are homeomorphic as topological spaces. The reader is invited to check this directly from the explicit descriptions of the spectral topologies on M ′ M' M ′ resp. M M M and V V V given above.
Next I construct from this spectral space an associated lattice ordered group. I repeat Fischer [5 Thm. 2.1] as the key result around this construction:
Proposition 4.9. Let X X X be a spectral tree such that the closure (in the spectral topology) of every constructible set is constructible. Then X X X is homeomorphic to the prime spectrum of a Bézout ring R R R , i.e. Spec ( R ) ≅ X \operatorname{Spec}(R) \cong X Spec ( R ) ≅ X .
Prop. 4.9 does not assert that R R R is a domain. Fischer comments in [5 p. 1727] that the resulting ring is a domain if X X X has a unique minimal element, which is the case in our example.
Using the description of the topology above, we can check the conditions of Prop. 4.9. Figure 2 suggests that M M M is a spectral tree, but one can also check this formally by determining the closures of all singletons with the help of the description of closed sets above. E.g., any closed set containing p p p must also contain { x , y } \{x,y\} { x , y } . Thus x , y x,y x , y are specializations of p p p , and since { p , x , y } \{p,x,y\} { p , x , y } is closed, they are also the only specializations, etc.
To see that – all in the spectral topology – the closure of any constructible set C ⊆ M C \subseteq M C ⊆ M is constructible, we distinguish two cases. In the first case, p ∉ C p \notin C p ∈ / C , we have $C = {z_i : i \in Z} \cup {p_i : i \in P } ∪ L P\} \cup L P } ∪ L with L ⊆ { x , y } L \subseteq\{x,y\} L ⊆ { x , y } , and C C C is finite. The closure is C ‾ c l = { z i : i ∈ Z ∪ P } ∪ { p i : i ∈ P } ∪ L \overline{C}^{\mathrm{cl}} = \{z_i : i \in\mathbb{Z} \cup P\} \cup\{p_i : i \in P\} \cup L C cl = { z i : i ∈ Z ∪ P } ∪ { p i : i ∈ P } ∪ L , again finite, and easily seen to be constructible. If p ∈ C p \in C p ∈ C , then C ‾ = { z i : i ∈ Z } ∪ { p i : i ∈ P } ∪ L \overline{C} = \{z_i : i \in\mathbb{Z}\} \cup\{p_i : i \in P\} \cup L C = { z i : i ∈ Z } ∪ { p i : i ∈ P } ∪ L with L ⊆ { p , x , y } L \subseteq\{p,x,y\} L ⊆ { p , x , y } such that Z Z Z , P P P are cofinite. The closure is C ‾ c l = { z i : i ∈ Z ∪ P } ∪ { p i : i ∈ P } ∪ { p , x , y } \overline{C}^{\mathrm{cl}} = \{z_i : i \in\mathbb{Z} \cup P\} \cup\{p_i : i \in P\} \cup\{p,x,y\} C cl = { z i : i ∈ Z ∪ P } ∪ { p i : i ∈ P } ∪ { p , x , y } , again with Z ∪ P Z \cup P Z ∪ P , P P P cofinite, which is clearly constructible by the list above.
According to Fischer’s proof [5 Thm. 2.1] , the lattice ordered group G = C ( X , Z ) = C ( X p a t c h , Z d i s c ) G = C(X,\mathbb{Z}) = C(X^{\mathrm{patch}},\mathbb{Z}^{\mathrm{disc}}) G = C ( X , Z ) = C ( X patch , Z disc ) consists of integer-valued functions that are continuous for the patch topology on X X X and the discrete topology on Z \mathbb{Z} Z . For f ∈ G f \in G f ∈ G , denote by Supp ( f ) ⊆ X \operatorname{Supp}(f) \subseteq X Supp ( f ) ⊆ X the support of f f f defined by Supp ( f ) = { x ∈ X : f ( x ) ≠ 0 } \operatorname{Supp}(f) = \{x \in X : f(x) \ne0\} Supp ( f ) = { x ∈ X : f ( x ) = 0 } , and by mSupp ( f ) ⊆ Supp ( f ) ⊆ X \operatorname{mSupp}(f) \subseteq\operatorname{Supp}(f) \subseteq X mSupp ( f ) ⊆ Supp ( f ) ⊆ X the minimal support of f f f (with respect to the specialization order) defined by mSupp ( f ) = { x ∈ X : f ( x ) ≠ 0 ∧ ∀ y ∈ X : y < s x ⇒ f ( y ) = 0 } \operatorname{mSupp}(f) = \{x \in X : f(x) \ne0 \land\forall y \in X : y <_{\mathrm{s}} x \Rightarrow f(y) = 0\} mSupp ( f ) = { x ∈ X : f ( x ) = 0 ∧ ∀ y ∈ X : y < s x ⇒ f ( y ) = 0 } . The positive cone is G + = { f ∈ G : ∀ x ∈ mSupp ( f ) : f ( x ) > 0 } G^{+} = \{f \in G : \forall x \in\operatorname{mSupp}(f) : f(x) > 0\} G + = { f ∈ G : ∀ x ∈ mSupp ( f ) : f ( x ) > 0 } . Note that since mSupp ( 0 ) ⊆ Supp ( 0 ) = ∅ \operatorname{mSupp}(0) \subseteq\operatorname{Supp}(0) = \varnothing mSupp ( 0 ) ⊆ Supp ( 0 ) = ∅ , the positivity condition is vacuous for the 0 0 0 function and thus 0 ∈ G + 0 \in G^{+} 0 ∈ G + . The lattice order, which is a partial order in general (i.e. some pairs of elements may be incomparable), is defined via the positive cone: f ≥ G g ⇔ f − g ∈ G + f \geq_{G} g \Leftrightarrow f-g \in G^{+} f ≥ G g ⇔ f − g ∈ G + .
In our case, X = M X=M X = M , we describe G M G_{M} G M and G M + G_{M}^{+} G M + explicitly. For f ∈ G M f \in G_{M} f ∈ G M , consider f ( p ) ∈ Z f(p) \in\mathbb{Z} f ( p ) ∈ Z . In the discrete topology, { f ( p ) } \{f(p)\} { f ( p )} is open. Continuity forces U p = f − 1 ( { f ( p ) } ) ⊆ M U_{p} = f^{-1}(\{f(p)\}) \subseteq M U p = f − 1 ({ f ( p )}) ⊆ M to be open (in the patch topology), and the only way this can happen according to the description of the topology above is with ∣ M ∖ U p ∣ < ∞ |M \setminus U_{p}| < \infty ∣ M ∖ U p ∣ < ∞ , i.e. if U p U_{p} U p is cofinite. So, except for finitely many (but otherwise arbitrary) points from { z i , p i : i ∈ N } ∪ { x , y } \{z_i,p_i : i \in\mathbb{N}\} \cup\{x,y\} { z i , p i : i ∈ N } ∪ { x , y } , f f f takes the value f ( p ) f(p) f ( p ) . This also implies that the image of f f f is finite: ∣ f ( M ) ∣ < ∞ |f(M)| < \infty ∣ f ( M ) ∣ < ∞ .
Note that f ∈ G + f \in G^{+} f ∈ G + need not take only nonnegative values. f ( x ) < 0 f(x) < 0 f ( x ) < 0 is allowed, as long as the minimal generization y < s x y <_{\mathrm{s}} x y < s x with f ( y ) ≠ 0 f(y) \ne0 f ( y ) = 0 satisfies f ( y ) > 0 f(y) > 0 f ( y ) > 0 . In this case mSupp ( f ) ⊊ Supp ( f ) \operatorname{mSupp}(f) \subsetneq\operatorname{Supp}(f) mSupp ( f ) ⊊ Supp ( f ) , i.e., the minimal support is a proper subset of the support. This also has consequences for the lattice operations inf , sup \inf,\sup inf , sup , which correspond directly to gcd, lcm in the constructed Bézout domain. For f , g ∈ G f,g \in G f , g ∈ G and x ∈ X x \in X x ∈ X , we set
inf ( f , g ) ( x ) = { f ( x ) , if G ( x ) ∩ mSupp ( f − g ) = ∅ , f ( x ) , if { y } = G ( x ) ∩ mSupp ( f − g ) ≠ ∅ ∧ f ( y ) < g ( y ) , g ( x ) , if { y } = G ( x ) ∩ mSupp ( f − g ) ≠ ∅ ∧ f ( y ) > g ( y ) . \inf(f,g)(x) =
\begin{cases}
f(x), & \text{if } G(x) \cap\operatorname{mSupp}(f-g) = \varnothing,\\
f(x), & \text{if } \{y\} = G(x) \cap\operatorname{mSupp}(f-g) \ne\varnothing\land f(y) < g(y),\\
g(x), & \text{if } \{y\} = G(x) \cap\operatorname{mSupp}(f-g) \ne\varnothing\land f(y) > g(y).
\end{cases} inf ( f , g ) ( x ) = ⎩ ⎨ ⎧ f ( x ) , f ( x ) , g ( x ) , if G ( x ) ∩ mSupp ( f − g ) = ∅ , if { y } = G ( x ) ∩ mSupp ( f − g ) = ∅ ∧ f ( y ) < g ( y ) , if { y } = G ( x ) ∩ mSupp ( f − g ) = ∅ ∧ f ( y ) > g ( y ) .
Note that since each G ( x ) G(x) G ( x ) is a totally ordered set (X X X is assumed to be a spectral tree), the intersection with the minimal support G ( x ) ∩ mSupp ( f − g ) = G x , f , g G(x) \cap\operatorname{mSupp}(f-g) = G_{x,f,g} G ( x ) ∩ mSupp ( f − g ) = G x , f , g is either a singleton or empty. We can describe G x , f , g G_{x,f,g} G x , f , g as the minimal generization y ≤ s x y \leq_{\mathrm{s}} x y ≤ s x of x x x where f f f and g g g differ, if they differ at all. If they do not differ (on the generizations of x x x – outside these generizations, f f f and g g g may still differ), we take the only meaningful value f ( x ) = g ( x ) f(x) = g(x) f ( x ) = g ( x ) . If they differ, the smaller value on the minimal differing y y y determines which value to take, i.e., we take f ( x ) f(x) f ( x ) if f ( y ) < g ( y ) f(y) < g(y) f ( y ) < g ( y ) , and g ( x ) g(x) g ( x ) if f ( y ) > g ( y ) f(y) > g(y) f ( y ) > g ( y ) . If the space X X X has proper generizations, this inevitably leads to higher-rank valuations. This is actually the case for X = M X=M X = M , giving a posteriori insight into the failure of the initial strategy for a counterexample. Similarly,
sup ( f , g ) ( x ) = { f ( x ) , if G ( x ) ∩ mSupp ( f − g ) = ∅ , f ( x ) , if { y } = G ( x ) ∩ mSupp ( f − g ) ≠ ∅ ∧ f ( y ) > g ( y ) , g ( x ) , if { y } = G ( x ) ∩ mSupp ( f − g ) ≠ ∅ ∧ f ( y ) < g ( y ) . \sup(f,g)(x) =
\begin{cases}
f(x), & \text{if } G(x) \cap\operatorname{mSupp}(f-g) = \varnothing,\\
f(x), & \text{if } \{y\} = G(x) \cap\operatorname{mSupp}(f-g) \ne\varnothing\land f(y) > g(y),\\
g(x), & \text{if } \{y\} = G(x) \cap\operatorname{mSupp}(f-g) \ne\varnothing\land f(y) < g(y).
\end{cases} sup ( f , g ) ( x ) = ⎩ ⎨ ⎧ f ( x ) , f ( x ) , g ( x ) , if G ( x ) ∩ mSupp ( f − g ) = ∅ , if { y } = G ( x ) ∩ mSupp ( f − g ) = ∅ ∧ f ( y ) > g ( y ) , if { y } = G ( x ) ∩ mSupp ( f − g ) = ∅ ∧ f ( y ) < g ( y ) .
I summarize these findings in the following definition for M M M .
Definition 4.10. The lattice ordered group G M ⊂ Map ( M , Z ) G_{M} \subset\operatorname{Map}(M,\mathbb{Z}) G M ⊂ Map ( M , Z ) consists of those functions f : M → Z f : M \to\mathbb{Z} f : M → Z for which all but finitely many values agree with f ( p ) f(p) f ( p ) , i.e. for which ∣ f − 1 ( Z ∖ { f ( p ) } ) ∣ < ∞ |f^{-1}(\mathbb{Z} \setminus\{f(p)\})| < \infty ∣ f − 1 ( Z ∖ { f ( p )}) ∣ < ∞ . The positive cone G M + G_{M}^{+} G M + consists of those f ∈ G M f \in G_{M} f ∈ G M such that, for all i ∈ N i \in\mathbb{N} i ∈ N , we have ( f ( p i ) = 0 ∨ f ( p i ) ≠ 0 ∧ f ( z i ) ≥ 0 ) (f(p_i) = 0 \lor f(p_i) \ne0 \land f(z_i) \geq0) ( f ( p i ) = 0 ∨ f ( p i ) = 0 ∧ f ( z i ) ≥ 0 ) and additionally f ( p ) > 0 ∨ ( f ( p ) = 0 ∧ f ( x ) ≥ 0 ≤ f ( y ) ) f(p) > 0 \lor(f(p) = 0 \land f(x) \geq0 \leq f(y)) f ( p ) > 0 ∨ ( f ( p ) = 0 ∧ f ( x ) ≥ 0 ≤ f ( y )) . This defines the (partial) lattice order via f ≥ G M g ⇔ f − g ∈ G M + f \geq_{G_M} g \Leftrightarrow f-g \in G_{M}^{+} f ≥ G M g ⇔ f − g ∈ G M + .
We also have, for f , g ∈ G M f,g\in G_M f , g ∈ G M and m ∈ M m\in M m ∈ M ,
inf ( f , g ) ( m ) = { min ( f ( m ) , g ( m ) ) if m ∈ { p i : i ∈ N } ∪ { p } f ( m ) if m = z i ∧ f ( p i ) < g ( p i ) min ( f ( m ) , g ( m ) ) if m = z i ∧ f ( p i ) = g ( p i ) g ( m ) if m = z i ∧ f ( p i ) > g ( p i ) f ( m ) if m ∈ { x , y } ∧ f ( p ) < g ( p ) min ( f ( m ) , g ( m ) ) if m ∈ { x , y } ∧ f ( p ) = g ( p ) g ( m ) if m ∈ { x , y } ∧ f ( p ) > g ( p ) \operatorname{inf}(f,g)(m)=
\begin{cases}
\min(f(m),g(m)) & \text{if }m\in\{p_i:i\in\mathbb{N}\}\cup\{p\}\\
f(m) & \text{if }m=z_i\land f(p_i)<g(p_i)\\
\min(f(m),g(m)) & \text{if }m=z_i\land f(p_i)=g(p_i)\\
g(m) & \text{if }m=z_i\land f(p_i)>g(p_i)\\
f(m) & \text{if }m\in\{x,y\}\land f(p)<g(p)\\
\min(f(m),g(m)) & \text{if }m\in\{x,y\}\land f(p)=g(p)\\
g(m) & \text{if }m\in\{x,y\}\land f(p)>g(p)
\end{cases} inf ( f , g ) ( m ) = ⎩ ⎨ ⎧ min ( f ( m ) , g ( m )) f ( m ) min ( f ( m ) , g ( m )) g ( m ) f ( m ) min ( f ( m ) , g ( m )) g ( m ) if m ∈ { p i : i ∈ N } ∪ { p } if m = z i ∧ f ( p i ) < g ( p i ) if m = z i ∧ f ( p i ) = g ( p i ) if m = z i ∧ f ( p i ) > g ( p i ) if m ∈ { x , y } ∧ f ( p ) < g ( p ) if m ∈ { x , y } ∧ f ( p ) = g ( p ) if m ∈ { x , y } ∧ f ( p ) > g ( p )
sup ( f , g ) ( m ) = { max ( f ( m ) , g ( m ) ) if m ∈ { p i : i ∈ N } ∪ { p } f ( m ) if m = z i ∧ f ( p i ) > g ( p i ) max ( f ( m ) , g ( m ) ) if m = z i ∧ f ( p i ) = g ( p i ) g ( m ) if m = z i ∧ f ( p i ) < g ( p i ) f ( m ) if m ∈ { x , y } ∧ f ( p ) > g ( p ) max ( f ( m ) , g ( m ) ) if m ∈ { x , y } ∧ f ( p ) = g ( p ) g ( m ) if m ∈ { x , y } ∧ f ( p ) < g ( p ) \operatorname{sup}(f,g)(m)=
\begin{cases}
\max(f(m),g(m)) & \text{if }m\in\{p_i:i\in\mathbb{N}\}\cup\{p\}\\
f(m) & \text{if }m=z_i\land f(p_i)>g(p_i)\\
\max(f(m),g(m)) & \text{if }m=z_i\land f(p_i)=g(p_i)\\
g(m) & \text{if }m=z_i\land f(p_i)<g(p_i)\\
f(m) & \text{if }m\in\{x,y\}\land f(p)>g(p)\\
\max(f(m),g(m)) & \text{if }m\in\{x,y\}\land f(p)=g(p)\\
g(m) & \text{if }m\in\{x,y\}\land f(p)<g(p)
\end{cases} sup ( f , g ) ( m ) = ⎩ ⎨ ⎧ max ( f ( m ) , g ( m )) f ( m ) max ( f ( m ) , g ( m )) g ( m ) f ( m ) max ( f ( m ) , g ( m )) g ( m ) if m ∈ { p i : i ∈ N } ∪ { p } if m = z i ∧ f ( p i ) > g ( p i ) if m = z i ∧ f ( p i ) = g ( p i ) if m = z i ∧ f ( p i ) < g ( p i ) if m ∈ { x , y } ∧ f ( p ) > g ( p ) if m ∈ { x , y } ∧ f ( p ) = g ( p ) if m ∈ { x , y } ∧ f ( p ) < g ( p )
The final step is the construction of the actual Bézout domain R R R from a lattice ordered group G G G . Here I follow Gilmer [8 Thm. 18.6] . Also note that the notions of semi-/demivaluation resp. semi-/demivaluation ring employed at the end of its proof can be found in [8 Ch. 16] . We start with any field k k k , for concreteness, k = Q k=\mathbb{Q} k = Q , and form the group ring k [ G ] k[G] k [ G ] . Associate to each g ∈ G g\in G g ∈ G a formal symbol x g x^g x g . Then elements of k [ G ] k[G] k [ G ] have a canonical form ∑ i = 1 n a g i x g i \sum_{i=1}^{n}a_{g_i}x^{g_i} ∑ i = 1 n a g i x g i , unique up to the order of summation, with a g i ∈ k ∖ { 0 } a_{g_i}\in k\setminus\{0\} a g i ∈ k ∖ { 0 } and ( g i ) i = 1 n ∈ G n (g_i)_{i=1}^{n}\in G^n ( g i ) i = 1 n ∈ G n a finite sequence of pairwise distinct elements of G G G for some integer n ∈ N n\in\mathbb{N} n ∈ N . Note that for n = 0 n=0 n = 0 , we get the zero 0 = 0 k [ G ] ∈ k [ G ] 0=0_{k[G]}\in k[G] 0 = 0 k [ G ] ∈ k [ G ] , an empty sum. For g 1 = 0 : X → Z g_1=0:X\to\mathbb{Z} g 1 = 0 : X → Z , x ↦ 0 x\mapsto0 x ↦ 0 and a g 1 = 1 ∈ k a_{g_1}=1\in k a g 1 = 1 ∈ k , we obtain the unit 1 = 1 k [ G ] = 1 ⋅ x 0 ∈ k [ G ] 1=1_{k[G]}=1\cdot x^0\in k[G] 1 = 1 k [ G ] = 1 ⋅ x 0 ∈ k [ G ] as a one-element sum. For canonical r = ∑ i = 1 n a g i x g i ∈ k [ G ] ∖ { 0 } r=\sum_{i=1}^{n}a_{g_i}x^{g_i}\in k[G]\setminus\{0\} r = ∑ i = 1 n a g i x g i ∈ k [ G ] ∖ { 0 } , we set its value w ( r ) = inf ( ( g i ) i = 1 n ) w(r)=\operatorname{inf}((g_i)_{i=1}^{n}) w ( r ) = inf (( g i ) i = 1 n ) – recall that, similar to Rem. 2.1, multiple infima in a lattice ordered group can be well-defined through iteration in the same way as multiple gcds in a GCD domain.
After establishing the relevant properties, one forms the fraction field Frac ( k [ G ] ) \operatorname{Frac}(k[G]) Frac ( k [ G ]) and extends w w w as follows: any nonzero element of Frac ( k [ G ] ) \operatorname{Frac}(k[G]) Frac ( k [ G ]) can be written (non-uniquely) as a quotient of canonical polynomials, i.e. r = ∑ i = 1 m a f i x f i ∑ i = 1 n b g i x g i r=\frac{\sum_{i=1}^{m}a_{f_i}x^{f_i}}{\sum_{i=1}^{n}b_{g_i}x^{g_i}} r = ∑ i = 1 n b g i x g i ∑ i = 1 m a f i x f i , and w ( r ) = w ( ∑ i = 1 m a f i x f i ) − w ( ∑ i = 1 n b g i x g i ) = inf ( ( f i ) i = 1 m ) − inf ( ( g i ) i = 1 n ) ∈ G w(r)=w\left(\sum_{i=1}^{m}a_{f_i}x^{f_i}\right)-w\left(\sum_{i=1}^{n}b_{g_i}x^{g_i}\right)=\operatorname{inf}((f_i)_{i=1}^{m})-\operatorname{inf}((g_i)_{i=1}^{n})\in G w ( r ) = w ( ∑ i = 1 m a f i x f i ) − w ( ∑ i = 1 n b g i x g i ) = inf (( f i ) i = 1 m ) − inf (( g i ) i = 1 n ) ∈ G yields a well-defined demivaluation. Its demivaluation ring R = Frac ( k [ G ] ) w R=\operatorname{Frac}(k[G])_w R = Frac ( k [ G ] ) w consists of those r ∈ Frac ( k [ G ] ) ∖ { 0 } r\in\operatorname{Frac}(k[G])\setminus\{0\} r ∈ Frac ( k [ G ]) ∖ { 0 } with w ( r ) ≥ G 0 w(r)\ge_G0 w ( r ) ≥ G 0 , i.e. w ( r ) ∈ G + w(r)\in G^+ w ( r ) ∈ G + , together with 0 0 0 (one could set w ( 0 ) = ∞ w(0)=\infty w ( 0 ) = ∞ ). The proof also shows that it is always possible to write a nonzero ring element r ∈ R ∖ { 0 } r\in R\setminus\{0\} r ∈ R ∖ { 0 } in the form r = 1 k ⋅ x w ( r ) 1 k [ G ] ⋅ r ′ r=\frac{1_k\cdot x^{w(r)}}{1_{k[G]}}\cdot r' r = 1 k [ G ] 1 k ⋅ x w ( r ) ⋅ r ′ for some unique r ′ ∈ R r'\in R r ′ ∈ R with w ( r ′ ) = 0 w(r')=0 w ( r ′ ) = 0 , i.e. with r ′ r' r ′ a unit. We shorten the notation to r = x w ( r ) ⋅ r ′ r=x^{w(r)}\cdot r' r = x w ( r ) ⋅ r ′ . For G = G M G=G_M G = G M , this finally realizes the Bézout domain R V R_V R V explicitly in which the counterexample to gf abstractly took place. To summarize:
Definition 4.11. The ring R V ⊂ Frac ( Q [ G M ] ) R_V\subset\operatorname{Frac}(\mathbb{Q}[G_M]) R V ⊂ Frac ( Q [ G M ]) consists of quotients (with nonzero denominator) of canonical group ring polynomials r = ∑ i = 1 m a f i x f i ∑ i = 1 n b g i x g i r=\frac{\sum_{i=1}^{m}a_{f_i}x^{f_i}}{\sum_{i=1}^{n}b_{g_i}x^{g_i}} r = ∑ i = 1 n b g i x g i ∑ i = 1 m a f i x f i with m , n ∈ N m,n\in\mathbb{N} m , n ∈ N , f i , g i ∈ G M = C ( M , Z ) f_i,g_i\in G_M=C(M,\mathbb{Z}) f i , g i ∈ G M = C ( M , Z ) , and a f i , b g i ∈ Q ∖ { 0 } a_{f_i},b_{g_i}\in\mathbb{Q}\setminus\{0\} a f i , b g i ∈ Q ∖ { 0 } , such that its demivaluation w ( r ) ≥ G M 0 w(r)\ge_{G_M}0 w ( r ) ≥ G M 0 is nonnegative, i.e. such that inf ( ( f i ) i = 1 m ) − inf ( ( g i ) i = 1 n ) ∈ G M + \operatorname{inf}((f_i)_{i=1}^{m})-\operatorname{inf}((g_i)_{i=1}^{n})\in G_M^+ inf (( f i ) i = 1 m ) − inf (( g i ) i = 1 n ) ∈ G M + , and additionally a zero 0 ∈ R V 0\in R_V 0 ∈ R V .
The ring operations are the usual ones for such quotients. The theorems used in the construction guarantee that R V R_V R V is a Bézout domain with Spec ( R V ) ≅ Spec ( G M ) ∪ { ( 0 ) } ≅ M ′ \operatorname{Spec}(R_V)\cong\operatorname{Spec}(G_M)\cup\{(0)\}\cong M' Spec ( R V ) ≅ Spec ( G M ) ∪ {( 0 )} ≅ M ′ and mSpec ( R V ) ≅ V \operatorname{mSpec}(R_V)\cong V mSpec ( R V ) ≅ V . Any nonzero ring element r ∈ R V ∖ { 0 } r\in R_V\setminus\{0\} r ∈ R V ∖ { 0 } can be written as r = x w ( r ) ⋅ r ′ r=x^{w(r)}\cdot r' r = x w ( r ) ⋅ r ′ for a unique unit r ′ ∈ R V × r'\in R_V^\times r ′ ∈ R V × with $w(r')=0. While r ′ r' r ′ is unique, its presentation as a quotient of canonical group ring polynomials is not.
Before presenting the counterexample to lcm splitting, I revisit the counterexample to gf from Prop. 4.5 . Recall that we showed the existence of a ~ , b ~ ∈ R V ∖ { 0 } \tilde{a},\tilde{b}\in R_V\setminus\{0\} a ~ , b ~ ∈ R V ∖ { 0 } with Z ( a ~ ) = A = V ∖ { z 0 } Z(\tilde{a})=A=V\setminus\{z_0\} Z ( a ~ ) = A = V ∖ { z 0 } , Z ( b ~ ) = B = { y } Z(\tilde{b})=B=\{y\} Z ( b ~ ) = B = { y } , where Z Z Z denotes the vanishing subspace in the maximal spectrum. Since Z ( a ~ ) Z(\tilde{a}) Z ( a ~ ) is infinite, our description of closed sets in the spectral topology of M M M forces the vanishing subspace in the prime spectrum to contain p p p , i.e. p ∈ V ( a ~ ) p\in V(\tilde{a}) p ∈ V ( a ~ ) or w ( a ~ ) ( p ) > 0 w(\tilde{a})(p)>0 w ( a ~ ) ( p ) > 0 , while Z ( b ~ ) Z(\tilde{b}) Z ( b ~ ) finite forces p ∉ V ( b ~ ) p\notin V(\tilde{b}) p ∈ / V ( b ~ ) resp. w ( b ~ ) ( p ) = 0 w(\tilde{b})(p)=0 w ( b ~ ) ( p ) = 0 . Since y ∈ { p } cl y\in\{p\}^{\operatorname{cl}} y ∈ { p } cl , no power of b ~ \tilde{b} b ~ can ever surpass a ~ \tilde{a} a ~ in the rank-2 valuation at y y y , i.e. v y ( a ~ ) ≮ v y ( b ~ N ) v_y(\tilde{a})\not< v_y(\tilde{b}^N) v y ( a ~ ) < v y ( b ~ N ) for all N ∈ N N\in\mathbb{N} N ∈ N . Thus, counterexamples to gf of the type in [20 Sec. 2.10] will never yield a counterexample to lcm splitting.
Explicitly, take g A , g B : M → Z g_A,g_B:M\to\mathbb{Z} g A , g B : M → Z such that g A ( m ) = 1 g_A(m)=1 g A ( m ) = 1 for all m ∈ M ∖ { z 0 , p 0 , y } m\in M\setminus\{z_0,p_0,y\} m ∈ M ∖ { z 0 , p 0 , y } , g B ( m ) = 0 g_B(m)=0 g B ( m ) = 0 for all m ∈ M ∖ { y } m\in M\setminus\{y\} m ∈ M ∖ { y } , and g A ( z 0 ) = g A ( p 0 ) = g A ( y ) = 0 g_A(z_0)=g_A(p_0)=g_A(y)=0 g A ( z 0 ) = g A ( p 0 ) = g A ( y ) = 0 , while g B ( y ) = 1 g_B(y)=1 g B ( y ) = 1 . Set a ~ = 1 ⋅ x g A 1 = x g A \tilde{a}=\frac{1\cdot x^{g_A}}{1}=x^{g_A} a ~ = 1 1 ⋅ x g A = x g A , b ~ = 1 ⋅ x g B 1 = x g B ∈ R V ∖ { 0 } \tilde{b}=\frac{1\cdot x^{g_B}}{1}=x^{g_B}\in R_V\setminus\{0\} b ~ = 1 1 ⋅ x g B = x g B ∈ R V ∖ { 0 } . Then b ~ ∣ a ~ \tilde{b}\mid\tilde{a} b ~ ∣ a ~ . Thus, by (2), we have lcm ( a ~ , b ~ ) ∼ a ~ , gcd ( a ~ , b ~ ) ∼ b ~ \operatorname{lcm}(\tilde{a},\tilde{b})\sim\tilde{a},\operatorname{gcd}(\tilde{a},\tilde{b})\sim\tilde{b} lcm ( a ~ , b ~ ) ∼ a ~ , gcd ( a ~ , b ~ ) ∼ b ~ . Hence c = a ~ , d = 1 R V c=\tilde{a},d=1_{R_V} c = a ~ , d = 1 R V are easily checked to witness lcm splitting of ( a ~ , b ~ ) (\tilde{a},\tilde{b}) ( a ~ , b ~ ) .
Note that the topological description of Prop. 4.5 neither fixes the ring R V R_V R V (e.g. replace k = Q k=\mathbb{Q} k = Q by another field) nor ( a ~ , b ~ ) (\tilde{a},\tilde{b}) ( a ~ , b ~ ) uniquely, e.g. ( 4 a ~ 5 , − 6 7 b ~ 8 ) (4\tilde{a}^{5},-\frac{6}{7}\tilde{b}^{8}) ( 4 a ~ 5 , − 7 6 b ~ 8 ) would be equally valid as a counterexample to gf. I will use the choice just described in the following when referring to Prop. 4.5 .
Theorem 4.12. For G M G_M G M as in Def. 4.10 , take g a , g b : M → Z g_a,g_b:M\to\mathbb{Z} g a , g b : M → Z such that g a ( m ) = 1 = g b ( m ) g_a(m)=1=g_b(m) g a ( m ) = 1 = g b ( m ) for all m ∈ M ∖ { x , y } m\in M\setminus\{x,y\} m ∈ M ∖ { x , y } and g a ( x ) = 0 = g b ( y ) g_a(x)=0=g_b(y) g a ( x ) = 0 = g b ( y ) , g a ( y ) = − 1 = g b ( x ) g_a(y)=-1=g_b(x) g a ( y ) = − 1 = g b ( x ) . With the notation of Def. 4.11 , set a = x g a , b = x g b ∈ R V ∖ { 0 } a=x^{g_a},b=x^{g_b}\in R_V\setminus\{0\} a = x g a , b = x g b ∈ R V ∖ { 0 } . Then a , b a,b a , b admit no lcm splitting, and lcm splitting fails in R V R_V R V .
Proof. We first verify that all quantities are well-defined in the sets we asserted above. Since g a , g b g_a,g_b g a , g b take the value 1 1 1 outside the finite set { x , y } \{x,y\} { x , y } , in particular at p p p , these functions belong to G M G_M G M . They even belong to G M + G_M^+ G M + : they are nonnegative except for g a ( y ) = − 1 = g b ( x ) g_a(y)=-1=g_b(x) g a ( y ) = − 1 = g b ( x ) , and since x > s p < s y x>_s p<_s y x > s p < s y and g a ( p ) = g b ( p ) = 1 > 0 g_a(p)=g_b(p)=1>0 g a ( p ) = g b ( p ) = 1 > 0 , this is compatible with being in G M + G_M^+ G M + . In particular, w ( a ) = g a − 0 = g a , w ( b ) = g b − 0 = g b ∈ G M + w(a)=g_a-0=g_a,w(b)=g_b-0=g_b\in G_M^+ w ( a ) = g a − 0 = g a , w ( b ) = g b − 0 = g b ∈ G M + . Thus a , b ∈ R V ⊆ Frac ( Q [ G M ] ) a,b\in R_V\subseteq\operatorname{Frac}(\mathbb{Q}[G_M]) a , b ∈ R V ⊆ Frac ( Q [ G M ]) are nonzero ring elements.
Applying the description of the supremum in Def. 4.10 , we see that w ( lcm ( a , b ) ) = sup ( w ( a ) , w ( b ) ) = g l w(\operatorname{lcm}(a,b))=\sup(w(a),w(b))=g_l w ( lcm ( a , b )) = sup ( w ( a ) , w ( b )) = g l with g l : M → Z g_l:M\to\mathbb{Z} g l : M → Z where g l ( m ) = 1 g_l(m)=1 g l ( m ) = 1 for all m ∈ M ∖ { x , y } m\in M\setminus\{x,y\} m ∈ M ∖ { x , y } and g l ( x ) = 0 = g l ( y ) g_l(x)=0=g_l(y) g l ( x ) = 0 = g l ( y ) . This lcm can be represented by l = x g l l=x^{g_l} l = x g l (uniquely up to units).
Now suppose that ( c , d ) ∈ R V 2 (c,d)\in R_V^2 ( c , d ) ∈ R V 2 is an lcm splitting of ( a , b ) (a,b) ( a , b ) . Up to units, we can represent them by monomials c = x g c , d = x g d c=x^{g_c},d=x^{g_d} c = x g c , d = x g d , and by the properties of G M + G_M^+ G M + , we have g c ( p ) ≥ 0 ≤ g d ( p ) g_c(p)\geq0\leq g_d(p) g c ( p ) ≥ 0 ≤ g d ( p ) . Since c d ∼ lcm ( a , b ) = l cd\sim\operatorname{lcm}(a,b)=l c d ∼ lcm ( a , b ) = l and multiplication in R V R_V R V corresponds to addition in G M G_M G M , this implies g c ( p ) + g d ( p ) = 1 g_c(p)+g_d(p)=1 g c ( p ) + g d ( p ) = 1 . By symmetry, we may assume g c ( p ) = 1 , g d ( p ) = 0 g_c(p)=1,g_d(p)=0 g c ( p ) = 1 , g d ( p ) = 0 . Since c ∣ a c\mid a c ∣ a and g c ( p ) = 1 = g a ( p ) g_c(p)=1=g_a(p) g c ( p ) = 1 = g a ( p ) , this forces g c ( y ) ≤ g a ( y ) = − 1 g_c(y)\leq g_a(y)=-1 g c ( y ) ≤ g a ( y ) = − 1 . Since c d ∼ l cd\sim l c d ∼ l , we have at y y y that − 1 + g d ( y ) ≥ g c ( y ) + g d ( y ) = g l ( y ) = 0 -1+g_d(y)\geq g_c(y)+g_d(y)=g_l(y)=0 − 1 + g d ( y ) ≥ g c ( y ) + g d ( y ) = g l ( y ) = 0 , i.e. g d ( y ) ≥ 1 g_d(y)\geq1 g d ( y ) ≥ 1 . We also need to satisfy the splitting condition gcd ( c , d ) ∼ 1 \gcd(c,d)\sim1 g cd( c , d ) ∼ 1 , i.e. inf ( g c , g d ) ( m ) = 0 \inf(g_c,g_d)(m)=0 inf ( g c , g d ) ( m ) = 0 for all m ∈ M m\in M m ∈ M . In particular, for m = y m=y m = y we need inf ( g c , g d ) ( y ) = 0 \inf(g_c,g_d)(y)=0 inf ( g c , g d ) ( y ) = 0 . Since y > s p y>_s p y > s p , we first check at p p p and have g c ( p ) = 1 > 0 = g d ( p ) g_c(p)=1>0=g_d(p) g c ( p ) = 1 > 0 = g d ( p ) . Thus inf ( g c , g d ) ( y ) = g d ( y ) ≥ 1 \inf(g_c,g_d)(y)=g_d(y)\geq1 inf ( g c , g d ) ( y ) = g d ( y ) ≥ 1 , which implies inf ( g c , g d ) ( y ) ≠ 0 \inf(g_c,g_d)(y)\neq0 inf ( g c , g d ) ( y ) = 0 , a contradiction. □ \square □
I summarize some further properties of R V R_V R V .
Corollary 4.13. R V R_V R V is a Bézout domain that is neither gf, nor adequate, nor lcm splitting, but it is an elementary divisor domain and K K K -Hermite.
Proof. By construction (cf. Def. 4.11 ), R V R_V R V is a Bézout domain. By Thm. 4.12 , it is not lcm splitting. By Thm. 3.2 , a Bézout domain such as R V R_V R V that is not lcm splitting cannot be adequate. By “gf ⇒ \Rightarrow ⇒ adequate” (cf. Rem. 2.3), R V R_V R V cannot be gf either. By [16 Thm. 3.1.2] , a Bézout domain with countably many maximal ideals is an elementary divisor domain. Clearly, V = { z i : i ∈ N } ∪ { x , y } V=\{z_i:i\in\mathbb{N}\}\cup\{x,y\} V = { z i : i ∈ N } ∪ { x , y } is countable. Thus R V R_V R V is an elementary divisor domain.
:::
By “elementary divisor ⇒ \Rightarrow ⇒ K-Hermite” (cf. Rem. 2.3), R V R_V R V is K-Hermite. Additionally, R V R_V R V is a Bézout domain, which is equivalent to K-Hermite for domains (cf. Rem. 2.3). □
Next, I examine some properties of pairs of elements of R V R_V R V .
Proposition 4.14. The counterexample to g f gf g f from Prop. 4.5 is also a counterexample to adequate. Let g A , g B ∈ G M g_A,g_B\in G_M g A , g B ∈ G M be given by g A ( m ) = 1 g_A(m)=1 g A ( m ) = 1 for all m ∈ M ∖ { z 0 , p 0 , y } m\in M\setminus\{z_0,p_0,y\} m ∈ M ∖ { z 0 , p 0 , y } and g A ( z 0 ) = g A ( p 0 ) = g A ( y ) = 0 g_A(z_0)=g_A(p_0)=g_A(y)=0 g A ( z 0 ) = g A ( p 0 ) = g A ( y ) = 0 , and g B ( m ) = 0 g_B(m)=0 g B ( m ) = 0 for all m ∈ M ∖ { y } m\in M\setminus\{y\} m ∈ M ∖ { y } and g B ( y ) = 1 g_B(y)=1 g B ( y ) = 1 . Then a ~ = x g A , b ~ = x g B \tilde{a}=x^{g_A},\tilde{b}=x^{g_B} a ~ = x g A , b ~ = x g B form a counterexample.
However, in reversed order, ( b ~ , a ~ ) (\tilde{b},\tilde{a}) ( b ~ , a ~ ) satisfies g f gf g f and adequate.
Proof. Assume we had a pair ( c , d ) (c,d) ( c , d ) satisfying the adequate conditions, wlog of the form c = x w ( c ) = x g C c=x^{w(c)}=x^{g_C} c = x w ( c ) = x g C , d = x w ( d ) = x g D d=x^{w(d)}=x^{g_D} d = x w ( d ) = x g D . Necessarily g C ( p ) = 0 g_C(p)=0 g C ( p ) = 0 , because g C ( p ) ≥ 1 g_C(p)\ge1 g C ( p ) ≥ 1 would force inf ( g C , g B ) ( y ) = g B ( y ) = 1 > 0 \inf(g_C,g_B)(y)=g_B(y)=1>0 inf ( g C , g B ) ( y ) = g B ( y ) = 1 > 0 , contradicting gcd ( c , b ~ ) ∼ 1 = x 0 \gcd(c,\tilde{b})\sim1=x^0 g cd( c , b ~ ) ∼ 1 = x 0 . Since all but finitely many values of g C g_C g C must agree with g C ( p ) = 0 g_C(p)=0 g C ( p ) = 0 (by continuity of g C g_C g C ), there is some i ≠ 0 i\ne0 i = 0 with g C ( z i ) = g C ( p i ) = 0 g_C(z_i)=g_C(p_i)=0 g C ( z i ) = g C ( p i ) = 0 . The condition a ~ = c d \tilde{a}=cd a ~ = c d implies 1 = g A ( z i ) = g C ( z i ) + g D ( z i ) = 0 + g D ( z i ) 1=g_A(z_i)=g_C(z_i)+g_D(z_i)=0+g_D(z_i) 1 = g A ( z i ) = g C ( z i ) + g D ( z i ) = 0 + g D ( z i ) , i.e. g D ( z i ) = 1 g_D(z_i)=1 g D ( z i ) = 1 . Now define g D ′ g_D' g D ′ by g D ′ ( m ) = 0 g_D'(m)=0 g D ′ ( m ) = 0 for all m ∈ M ∖ { z i } m\in M\setminus\{z_i\} m ∈ M ∖ { z i } and g D ′ ( z i ) = 1 g_D'(z_i)=1 g D ′ ( z i ) = 1 . By construction, 0 G M < G M g D ′ ≤ G M g D 0_{G_M}<_{G_M}g_D'\le_{G_M}g_D 0 G M < G M g D ′ ≤ G M g D . Thus d ′ = x g D ′ ∈ R V ∖ R V × d'=x^{g_D'}\in R_V\setminus R_V^\times d ′ = x g D ′ ∈ R V ∖ R V × is a non-unit with d ′ ∣ d d'\mid d d ′ ∣ d . On the other hand, g D ′ ( m ) = 0 g_D'(m)=0 g D ′ ( m ) = 0 for all m ≠ z i m\ne z_i m = z i forces inf ( g D ′ , g B ) ( m ) = 0 \inf(g_D',g_B)(m)=0 inf ( g D ′ , g B ) ( m ) = 0 at the same places m m m . Since also g B ( z i ) = 0 = g B ( p i ) = g D ′ ( p i ) g_B(z_i)=0=g_B(p_i)=g_D'(p_i) g B ( z i ) = 0 = g B ( p i ) = g D ′ ( p i ) , we get inf ( g D ′ , g B ) ( z i ) = 0 \inf(g_D',g_B)(z_i)=0 inf ( g D ′ , g B ) ( z i ) = 0 ; hence gcd ( b ~ , d ′ ) ∼ 1 \gcd(\tilde{b},d')\sim1 g cd( b ~ , d ′ ) ∼ 1 . This contradicts the adequate property and establishes the claim.
To prove that ( b ~ , a ~ ) (\tilde{b},\tilde{a}) ( b ~ , a ~ ) satisfies g f gf g f , set c = 1 c=1 c = 1 , d = b ~ d=\tilde{b} d = b ~ . Clearly b ~ = c d \tilde{b}=cd b ~ = c d and gcd ( c , a ~ ) = gcd ( 1 , a ~ ) ∼ ( 6 ) , ( 9 ) 1 \gcd(c,\tilde{a})=\gcd(1,\tilde{a})\mathrel{\overset{(6),(9)}{\sim}}1 g cd( c , a ~ ) = g cd( 1 , a ~ ) ∼ ( 6 ) , ( 9 ) 1 . Let g D ′ ′ ∈ G M g_{D''}\in G_M g D ′′ ∈ G M be given by g D ′ ′ ( m ) = 1 g_{D''}(m)=1 g D ′′ ( m ) = 1 for all m ∈ M ∖ { z 0 , p 0 , y } m\in M\setminus\{z_0,p_0,y\} m ∈ M ∖ { z 0 , p 0 , y } , g D ′ ′ ( z 0 ) = g D ′ ′ ( p 0 ) = 0 g_{D''}(z_0)=g_{D''}(p_0)=0 g D ′′ ( z 0 ) = g D ′′ ( p 0 ) = 0 and g D ′ ′ ( y ) = − 1 g_{D''}(y)=-1 g D ′′ ( y ) = − 1 . One checks that g D ′ ′ ∈ G M + g_{D''}\in G_M^+ g D ′′ ∈ G M + and d ′ ′ = x g D ′ ′ ∈ R V d''=x^{g_{D''}}\in R_V d ′′ = x g D ′′ ∈ R V is well-defined. By construction, a ~ = d ′ ′ b ~ = d ′ ′ d ∈ ( d ) \tilde{a}=d''\tilde{b}=d''d\in(d) a ~ = d ′′ b ~ = d ′′ d ∈ ( d ) , which shows the g f gf g f property. Lem. B.1 also implies the adequate property. □
Second example
I found the second example after finishing the first. It is arguably simpler and gives a counterexample pair ( a , b ) (a,b) ( a , b ) with properties similar to those in § 4.1 . The previous construction might, however, be more flexible and give rise to modified examples with new properties, cf. § 5 for some concrete ideas. I use the concepts and terminology developed for the first example for easier comparison and presentation.
Let S = Z + x Q [ x ] ⊂ Q [ x ] S=\mathbb{Z}+x\mathbb{Q}[x]\subset\mathbb{Q}[x] S = Z + x Q [ x ] ⊂ Q [ x ] . By [1 Thm. 4.6 (6)] , it is a Bézout domain. By Roitman [17, Ex. 3.3], S S S is even an elementary divisor domain. Properties of this ring can be found in Hutchins [11, Ex. 32, p. 67]. According to [19 Thm. 1.13] , irreducible elements of S S S are prime and they are classified (up to units) as follows:
(a) prime numbers { p i : i ∈ N > 0 } ⊂ Z \{p_i:i\in\mathbb{N}_{>0}\}\subset\mathbb{Z} { p i : i ∈ N > 0 } ⊂ Z or
(b) irreducible polynomials with 1 1 1 as constant coefficient { f j : j ∈ N > 0 } ⊂ Q [ x ] \{f_j:j\in\mathbb{N}_{>0}\}\subset\mathbb{Q}[x] { f j : j ∈ N > 0 } ⊂ Q [ x ] .
This gives rise to four different types of prime ideals (see Figure 3 for an illustration):
Figure 3. The spectral tree Spec(S S S ) for S = Z + x Q [ x ] S=\mathbb{Z}+x\mathbb{Q}[x] S = Z + x Q [ x ] with indicated specialization order.
(a) height 2 2 2 prime ideals of the form ( p i ) (p_i) ( p i ) ,
(b) height 1 1 1 prime ideals of the form ( f j ) (f_j) ( f j ) ,
(c) the height 1 1 1 prime ideal I x = ( x , x 2 , x 3 , x 4 , … ) I_x=\left(x,\frac{x}{2},\frac{x}{3},\frac{x}{4},\ldots\right) I x = ( x , 2 x , 3 x , 4 x , … ) ,
(d) the height 0 0 0 prime ideal ( 0 ) (0) ( 0 ) .
Note that the prime ideal I x I_x I x is not finitely generated, reflecting that S S S is non-Noetherian. We also have I x ⊊ ( p i ) I_x\subsetneq(p_i) I x ⊊ ( p i ) for every prime number p i p_i p i . The ideals ( p i ) (p_i) ( p i ) show that Krull’s principal ideal theorem need not hold in non-Noetherian rings. None of the infinitely many maximal ideals are isolated in Spec ( S ) \operatorname{Spec}(S) Spec ( S ) , thus Prop. 4.3 and the whole construction from § 4.1 does not apply here.
S S S is not a UFD, since, e.g., x = 2 ⋅ ( 1 2 x ) = 2 2 ⋅ ( 1 2 2 x ) = 2 3 ⋅ ( 1 2 3 x ) = … x=2\cdot\left(\frac{1}{2}x\right)=2^2\cdot\left(\frac{1}{2^2}x\right)=2^3\cdot\left(\frac{1}{2^3}x\right)=\ldots x = 2 ⋅ ( 2 1 x ) = 2 2 ⋅ ( 2 2 1 x ) = 2 3 ⋅ ( 2 3 1 x ) = … acquires arbitrarily many prime factors 2 2 2 . This links to I x I_x I x not being finitely generated. Nevertheless, there is “almost a unique prime factorization”. If a ∈ S a\in S a ∈ S has a nonvanishing constant coefficient, it can be written as a product of powers of irreducibles a = ± ∏ i p i r i ⋅ ∏ j f j s j a=\pm\prod_i p_i^{r_i}\cdot\prod_j f_j^{s_j} a = ± ∏ i p i r i ⋅ ∏ j f j s j . If its constant coefficient is 0 0 0 , we have a = m n x ⋅ x u ⋅ ∏ j f j t j a=\frac{m}{n}x\cdot x^u\cdot\prod_j f_j^{t_j} a = n m x ⋅ x u ⋅ ∏ j f j t j . Also note that there are only two units, S × = { ± 1 } S^\times=\{\pm1\} S × = { ± 1 } .
For the remainder of this subsection, fix ( a , b ) = ( x q a , x q b ) (a,b)=\left(\frac{x}{q_a},\frac{x}{q_b}\right) ( a , b ) = ( q a x , q b x ) , where q a ≠ q b q_a\ne q_b q a = q b are prime numbers. For concreteness, take q a = 2 , q b = 3 q_a=2,q_b=3 q a = 2 , q b = 3 , but note the structural symmetry of this choice. We have gcd ( a , b ) ∼ x 6 \gcd(a,b)\sim\frac{x}{6} g cd( a , b ) ∼ 6 x and lcm ( a , b ) ∼ x \operatorname{lcm}(a,b)\sim x lcm ( a , b ) ∼ x . Assume for contradiction that ( c , d ) (c,d) ( c , d ) witnessed lcm splitting for ( a , b ) (a,b) ( a , b ) . Then c d ∼ lcm ( a , b ) ∼ x cd\sim\operatorname{lcm}(a,b)\sim x c d ∼ lcm ( a , b ) ∼ x ; wlog c d = x cd=x c d = x . For degree reasons and since the elements of degree 0 0 0 are integers, any factorization has the form x = n ⋅ x n x=n\cdot\frac{x}{n} x = n ⋅ n x for n ∈ Z ∖ { 0 } n\in\mathbb{Z}\setminus\{0\} n ∈ Z ∖ { 0 } . By the structural symmetry, we may assume c = n , d = x n c=n,d=\frac{x}{n} c = n , d = n x . Since d = x n = x n 2 ⋅ n = x n 2 ⋅ c d=\frac{x}{n}=\frac{x}{n^2}\cdot n=\frac{x}{n^2}\cdot c d = n x = n 2 x ⋅ n = n 2 x ⋅ c , i.e. c ∣ d c\mid d c ∣ d , we get gcd ( c , d ) ∼ ( 2 ) c = n \gcd(c,d)\mathrel{\mathop{\sim}^{(2)}}c=n g cd( c , d ) ∼ ( 2 ) c = n . The condition gcd ( c , d ) ∼ 1 \gcd(c,d)\sim1 g cd( c , d ) ∼ 1 forces c = n ∈ S × = { ± 1 } c=n\in S^\times=\{\pm1\} c = n ∈ S × = { ± 1 } , wlog c = 1 c=1 c = 1 . But then d = x n = x ∤ x 3 = b d=\frac{x}{n}=x\nmid\frac{x}{3}=b d = n x = x ∤ 3 x = b , violating one of the lcm splitting conditions. Therefore ( a , b ) (a,b) ( a , b ) does not admit any lcm splitting.
( a , b ) = ( x 2 , x 3 ) (a,b)=\left(\frac{x}{2},\frac{x}{3}\right) ( a , b ) = ( 2 x , 3 x ) satisfies gf. Set ( c , d ) = ( 1 , x 2 ) (c,d)=\left(1,\frac{x}{2}\right) ( c , d ) = ( 1 , 2 x ) . Clearly a = c d a=cd a = c d and gcd ( c , b ) = gcd ( 1 , b ) ∼ ( 6 ) , ( 9 ) 1 \gcd(c,b)=\gcd(1,b)\mathrel{\mathop{\sim}^{(6),(9)}}1 g cd( c , b ) = g cd( 1 , b ) ∼ ( 6 ) , ( 9 ) 1 . Setting B = 2 x 9 B=\frac{2x}{9} B = 9 2 x , we have b 2 = x 2 9 = B d b^2=\frac{x^2}{9}=Bd b 2 = 9 x 2 = B d , i.e. b ∈ ( d ) b\in\sqrt{(d)} b ∈ ( d ) . By Lem. B.1, this implies adequate. By the structural symmetry, also ( b , a ) (b,a) ( b , a ) is gf and adequate, too.
Theorem 4.16. For a = x 2 , b = x 3 ∈ S ∖ { 0 } a=\frac{x}{2}, b=\frac{x}{3}\in S\setminus\{0\} a = 2 x , b = 3 x ∈ S ∖ { 0 } , both ( a , b ) (a,b) ( a , b ) and ( b , a ) (b,a) ( b , a ) satisfy the gf and adequate conditions but fail lcm splitting. In particular, S S S is a non-adequate, elementary divisor domain that fails lcm splitting.
Proof. All claims except S S S being non-adequate were proven or referenced in the discussion above. S S S being non-adequate follows from the failure of lcm splitting and Thm. 3.2 .
Final remarks and open questions
I give further observations and state some open questions.
Let R R R be a Bézout domain that fails lcm splitting. The ring R [ x ] R[x] R [ x ] is a GCD domain that is not Bézout (for a nontrivial principal ideal ( r ) ⊂ R (r) \subset R ( r ) ⊂ R , the ideal ( r , x ) ⊂ R [ x ] (r,x) \subset R[x] ( r , x ) ⊂ R [ x ] is not principal), and the same pair ( a , b ) (a,b) ( a , b ) that witnessed failure of lcm splitting before again shows failure of lcm splitting in R [ x ] R[x] R [ x ] . On the other hand, let k k k be any field. Then the polynomial ring P = k [ ( x i ) i = 1 n ] P = k[(x_i)_{i=1}^n] P = k [( x i ) i = 1 n ] for n ≥ 2 n \ge2 n ≥ 2 is a UFD, hence a GCD domain, non-Bézout (e.g. ( x 1 , x 2 ) (x_1,x_2) ( x 1 , x 2 ) is not principal), but, as a UFD, P P P satisfies lcm splitting. Thus non-Bézout GCD domains may or may not be lcm splitting.
We have seen that adequate domains are always lcm splitting. I have given two non-adequate, elementary divisor domains that fail lcm splitting. I leave it open whether there are non-adequate, elementary divisor domains that satisfy lcm splitting. Recently, an example of a Bézout domain that is not an elementary divisor domain was announced. Is the example of Hägg and Mörtberg [9 ] lcm splitting? Does failure of the elementary divisor property more generally imply failure of lcm splitting? Does lcm splitting imply adequate? Note that lcm splitting is symmetric in ( a , b ) (a,b) ( a , b ) , while adequate (and gf) are not. It would be surprising to me if lcm splitting implied adequate, although I do not have a good idea what an example of an lcm splitting, non-adequate Bézout domain should look like.
One could investigate lcm splitting in GCD rings with zero divisors. Gillman and Henriksen [7 ] already gave examples of K-Hermite GCD rings that fail the elementary divisor property.
There are other properties that might be relevant to lcm splitting. How does lcm splitting relate to them? Anderson et al. [2 ] consider almost splitting sets and refer to work on t-splitting sets , which, superficially, seem to be connected to lcm splitting. Anderson [1 Sec. 5] mentions that more general classes of rings can be inertly embedded into Bézout domains — cf. also Cohn [3 Thm. 3.3] . Since inert embeddings leave divisibility properties and gcds mostly unchanged, it is natural to ask how lcm splitting behaves with respect to such embeddings.
There is a correspondence between Bézout domains and lattice ordered (abelian) groups. Is there an analog of lcm splitting for lattice ordered groups? Some research brought up the (strongly) projectable lattice ordered groups , which might be worth investigating further.
Unrelated to lcm splitting, but still interesting: Are there adequate domains that are not gf? Van den Dries and Macintyre [20 Sec. 2.10] used essentially only topology for their counterexample. Could one reformulate gf and/or adequate in purely topological terms of spectral spaces/trees? If so, this might open a route to resolve the relation between gf and adequate.
I exhibited a pair of elements ( a ~ , b ~ ) (\tilde{a},\tilde{b}) ( a ~ , b ~ ) in the ring R V R_V R V that violates gf and adequate but is lcm splitting, and two pairs ( a , b ) (a,b) ( a , b ) , one in R V R_V R V and the other in S S S , that violate lcm splitting although both ( a , b ) (a,b) ( a , b ) and ( b , a ) (b,a) ( b , a ) satisfy gf and adequate. In general, what are the possible combinations of properties that (pairs of) elements can have in Bézout domains? It is not hard to see that as soon as 1 ∈ { a , b } 1 \in\{a,b\} 1 ∈ { a , b } , gf, adequate and lcm splitting can be satisfied by suitable and obvious choices of ( c , d ) (c,d) ( c , d ) . Therefore, it is impossible for all pairs to behave badly. See also Rem. 2.4 for the role of 0 0 0 in lcm splitting.
Is it possible to have a pair of elements that violates all three properties, gf, adequate and lcm splitting? One idea is to take the disjoint union V ′ ′ V'' V ′′ of two copies of V V V . The conditions of Prop. 4.3 are also satisfied by V ′ ′ V'' V ′′ , so there is a Bézout domain R V ′ ′ R_{V''} R V ′′ with V ′ ′ V'' V ′′ as its maximal spectrum. An associated spectral tree M ′ ′ M'' M ′′ should consist of two disjoint copies of M M M and a single generic point ( 0 ) (0) ( 0 ) . One should be able to set up a pair ( a ^ , b ^ ) (\hat{a},\hat{b}) ( a ^ , b ^ ) that behaves like ( a ~ , b ~ ) (\tilde{a},\tilde{b}) ( a ~ , b ~ ) on one copy of M M M in M ′ ′ M'' M ′′ and like ( a , b ) (a,b) ( a , b ) on the other copy. This should provide a Bézout domain R V ′ ′ R_{V''} R V ′′ with a pair ( a ^ , b ^ ) (\hat{a},\hat{b}) ( a ^ , b ^ ) that violates all three properties, although I have not checked all relevant details. In order to get a pair ( a ^ , b ^ ) (\hat{a},\hat{b}) ( a ^ , b ^ ) such that it and ( b ^ , a ^ ) (\hat{b},\hat{a}) ( b ^ , a ^ ) violate all three properties, one could try an R V ′ ′ ′ R_{V'''} R V ′′′ associated with three disjoint copies of M M M .
Finding a Bézout domain with a pair that violates gf but satisfies adequate could also turn out to be helpful for the question whether there are domains that are adequate but not gf.
Closing thoughts. It is quite remarkable that lcm splitting seems to be new in commutative algebra, as it appears to be such an obvious and naïve property. At least, I was not able to find substantial literature in commutative algebra discussing it. I would greatly appreciate any pointers from the community to work treating this property, possibly under a different name.
Acknowledgment and use of AI
I thank Ash Seifert for the question that inspired lcm splitting, and Andrew Kresch and Elisa Gorla for help with publishing.
Some of the results were developed in dialog with LLMs: the free versions, current at the time of writing, of Anthropic’s Sonnet 5 (https://claude.ai/share/21cb9a53-c0db-4ca9-af3e-746fb22aad83 ) and OpenAI’s ChatGPT (https://chatgpt.com/share/6ac74f32-08e0-83eb-ad30-f7940cf31be5 ).
I take full responsibility for the content of this paper.
Appendix A. Proofs for basic identities and properties in GCD domains
Here I give proofs of the claims in Prop. A.1. For convenience, I restate the proposition.
Proposition A.1. Let R R R be a GCD domain and a , b , c ∈ R a,b,c \in R a , b , c ∈ R . Then:
(1) definitional divisibility: c ∣ a ∧ c ∣ b ⇒ c ∣ gcd ( a , b ) c \mid a \land c \mid b \Rightarrow c \mid\gcd(a,b) c ∣ a ∧ c ∣ b ⇒ c ∣ g cd( a , b ) , and a ∣ c ∧ b ∣ c ⇒ lcm ( a , b ) ∣ c a \mid c \land b \mid c \Rightarrow\operatorname{lcm}(a,b) \mid c a ∣ c ∧ b ∣ c ⇒ lcm ( a , b ) ∣ c ,
(2) mutual divisibility: a ∣ b ∧ b ∣ a ⇒ a ∼ b a \mid b \land b \mid a \Rightarrow a \sim b a ∣ b ∧ b ∣ a ⇒ a ∼ b , a ∣ b ⇒ ( gcd ( a , b ) ∼ a ∧ lcm ( a , b ) ∼ b ) a \mid b \Rightarrow(\gcd(a,b) \sim a \land\operatorname{lcm}(a,b) \sim b) a ∣ b ⇒ ( g cd( a , b ) ∼ a ∧ lcm ( a , b ) ∼ b ) ,
(3) compatibility with ∼ \sim ∼ : ( a ∼ c ∧ b ∼ d ) ⇒ ( gcd ( a , b ) ∼ gcd ( c , d ) ∧ lcm ( a , b ) ∼ lcm ( c , d ) ) (a \sim c \land b \sim d) \Rightarrow(\gcd(a,b) \sim\gcd(c,d) \land\operatorname{lcm}(a,b) \sim\operatorname{lcm}(c,d)) ( a ∼ c ∧ b ∼ d ) ⇒ ( g cd( a , b ) ∼ g cd( c , d ) ∧ lcm ( a , b ) ∼ lcm ( c , d )) ,
(4) zero behavior: ( a ∼ 0 ⇔ a = 0 ) (a \sim0 \Leftrightarrow a = 0) ( a ∼ 0 ⇔ a = 0 ) , gcd ( a , 0 ) ∼ a \gcd(a,0) \sim a g cd( a , 0 ) ∼ a , ( 0 ∼ gcd ( a , b ) ⇔ ( a = 0 ∧ b = 0 ) ) (0 \sim\gcd(a,b) \Leftrightarrow(a = 0 \land b = 0)) ( 0 ∼ g cd( a , b ) ⇔ ( a = 0 ∧ b = 0 )) , lcm ( a , 0 ) = 0 \operatorname{lcm}(a,0) = 0 lcm ( a , 0 ) = 0 , and ( 0 ∼ lcm ( a , b ) ⇔ ( a = 0 ∨ b = 0 ) ) (0 \sim\operatorname{lcm}(a,b) \Leftrightarrow(a = 0 \lor b = 0)) ( 0 ∼ lcm ( a , b ) ⇔ ( a = 0 ∨ b = 0 )) ,
(5) product formula: a b ∼ gcd ( a , b ) lcm ( a , b ) ab \sim\gcd(a,b)\operatorname{lcm}(a,b) ab ∼ g cd( a , b ) lcm ( a , b ) ,
(6) commutativity: gcd ( a , b ) ∼ gcd ( b , a ) \gcd(a,b) \sim\gcd(b,a) g cd( a , b ) ∼ g cd( b , a ) , and lcm ( a , b ) ∼ lcm ( b , a ) \operatorname{lcm}(a,b) \sim\operatorname{lcm}(b,a) lcm ( a , b ) ∼ lcm ( b , a ) ,
(7) associativity: gcd ( a , gcd ( b , c ) ) ∼ gcd ( gcd ( a , b ) , c ) \gcd(a,\gcd(b,c)) \sim\gcd(\gcd(a,b),c) g cd( a , g cd( b , c )) ∼ g cd( g cd( a , b ) , c ) , and lcm ( a , lcm ( b , c ) ) ∼ lcm ( lcm ( a , b ) , c ) \operatorname{lcm}(a,\operatorname{lcm}(b,c)) \sim\operatorname{lcm}(\operatorname{lcm}(a,b),c) lcm ( a , lcm ( b , c )) ∼ lcm ( lcm ( a , b ) , c ) ,
(8) idempotence: gcd ( a , a ) ∼ a \gcd(a,a) \sim a g cd( a , a ) ∼ a , and lcm ( a , a ) ∼ a \operatorname{lcm}(a,a) \sim a lcm ( a , a ) ∼ a ,
(9) unity: c ∣ 1 ⇒ c ∼ 1 c \mid1 \Rightarrow c \sim1 c ∣ 1 ⇒ c ∼ 1 , gcd ( a , b ) ∣ 1 ⇒ gcd ( a , b ) ∼ 1 \gcd(a,b) \mid1 \Rightarrow\gcd(a,b) \sim1 g cd( a , b ) ∣ 1 ⇒ g cd( a , b ) ∼ 1 , $1 \nmid a \mid b \Rightarrow 1 \nmid b$, gcd ( a , 1 ) ∼ 1 \gcd(a,1) \sim1 g cd( a , 1 ) ∼ 1 , and lcm ( a , 1 ) ∼ a \operatorname{lcm}(a,1) \sim a lcm ( a , 1 ) ∼ a ,
(10) common factor: gcd ( c a , c b ) ∼ c gcd ( a , b ) \gcd(ca,cb) \sim c\gcd(a,b) g cd( c a , c b ) ∼ c g cd( a , b ) , and lcm ( c a , c b ) ∼ c lcm ( a , b ) \operatorname{lcm}(ca,cb) \sim c\operatorname{lcm}(a,b) lcm ( c a , c b ) ∼ c lcm ( a , b ) ,
(11) additional factor: gcd ( a , b ) ∣ gcd ( a , b c ) \gcd(a,b) \mid\gcd(a,bc) g cd( a , b ) ∣ g cd( a , b c ) , and lcm ( a , b ) ∣ lcm ( a , b c ) \operatorname{lcm}(a,b) \mid\operatorname{lcm}(a,bc) lcm ( a , b ) ∣ lcm ( a , b c ) ,
(12) coprime elements: gcd ( a , b ) ∼ 1 ∧ a ∣ b c ⇒ a ∣ c \gcd(a,b) \sim1 \land a \mid bc \Rightarrow a \mid c g cd( a , b ) ∼ 1 ∧ a ∣ b c ⇒ a ∣ c , gcd ( a , b ) ∼ 1 ⇒ gcd ( a , b c ) ∼ gcd ( a , c ) \gcd(a,b) \sim1 \Rightarrow\gcd(a,bc) \sim\gcd(a,c) g cd( a , b ) ∼ 1 ⇒ g cd( a , b c ) ∼ g cd( a , c ) , gcd ( a , b ) ∼ 1 ⇒ lcm ( a , b c ) ∼ b lcm ( a , c ) \gcd(a,b) \sim1 \Rightarrow\operatorname{lcm}(a,bc) \sim b\operatorname{lcm}(a,c) g cd( a , b ) ∼ 1 ⇒ lcm ( a , b c ) ∼ b lcm ( a , c ) , and ( a ∣ c ∧ b ∣ c ∧ gcd ( a , b ) ∼ 1 ) ⇒ a b ∣ c (a \mid c \land b \mid c \land\gcd(a,b) \sim1) \Rightarrow ab \mid c ( a ∣ c ∧ b ∣ c ∧ g cd( a , b ) ∼ 1 ) ⇒ ab ∣ c ,
(13) division by common factor (assume 0 ≠ c ∣ a 0 \ne c \mid a 0 = c ∣ a and c ∣ b c \mid b c ∣ b , which implies that all following fractions in Frac ( R ) \operatorname{Frac}(R) Frac ( R ) are actually in R R R ): gcd ( a c , b c ) ∼ gcd ( a , b ) c \gcd\left(\frac{a}{c},\frac{b}{c}\right) \sim\frac{\gcd(a,b)}{c} g cd( c a , c b ) ∼ c g c d ( a , b ) , and lcm ( a c , b c ) ∼ lcm ( a , b ) c \operatorname{lcm}\left(\frac{a}{c},\frac{b}{c}\right) \sim\frac{\operatorname{lcm}(a,b)}{c} lcm ( c a , c b ) ∼ c lcm ( a , b ) — of particular interest will be the special case c = gcd ( a , b ) c = \gcd(a,b) c = g cd( a , b ) ,
(14) powers (n ∈ N n \in\mathbb{N} n ∈ N , c 0 = 1 R c^0 = 1_R c 0 = 1 R ): gcd ( a n , b n ) ∼ gcd ( a , b ) n \gcd(a^n,b^n) \sim\gcd(a,b)^n g cd( a n , b n ) ∼ g cd( a , b ) n , and lcm ( a n , b n ) ∼ lcm ( a , b ) n \operatorname{lcm}(a^n,b^n) \sim\operatorname{lcm}(a,b)^n lcm ( a n , b n ) ∼ lcm ( a , b ) n ,
(15) absorption: gcd ( a , lcm ( a , b ) ) ∼ a \gcd(a,\operatorname{lcm}(a,b)) \sim a g cd( a , lcm ( a , b )) ∼ a , and lcm ( a , gcd ( a , b ) ) ∼ a \operatorname{lcm}(a,\gcd(a,b)) \sim a lcm ( a , g cd( a , b )) ∼ a ,
(16) distributivity: gcd ( a , lcm ( b , c ) ) ∼ lcm ( gcd ( a , b ) , gcd ( a , c ) ) \gcd(a,\operatorname{lcm}(b,c)) \sim\operatorname{lcm}(\gcd(a,b),\gcd(a,c)) g cd( a , lcm ( b , c )) ∼ lcm ( g cd( a , b ) , g cd( a , c )) , and lcm ( a , gcd ( b , c ) ) ∼ gcd ( lcm ( a , b ) , lcm ( a , c ) ) \operatorname{lcm}(a,\gcd(b,c)) \sim\gcd(\operatorname{lcm}(a,b),\operatorname{lcm}(a,c)) lcm ( a , g cd( b , c )) ∼ g cd( lcm ( a , b ) , lcm ( a , c )) .
Proof. (1) This is really in the definition of gcd resp. lcm, so trivial.
(2) From a ∣ b ∧ b ∣ a a \mid b \land b \mid a a ∣ b ∧ b ∣ a we have a ′ , b ′ ∈ R a',b' \in R a ′ , b ′ ∈ R such that a ′ a = b a'a = b a ′ a = b , b ′ b = a b'b = a b ′ b = a . Assume first that (at least) one of a , b a,b a , b is zero, wlog a = 0 a = 0 a = 0 . Then, by the general behavior of zero in rings, b = a ′ a = a ′ ⋅ 0 = 0 b = a'a = a' \cdot0 = 0 b = a ′ a = a ′ ⋅ 0 = 0 and b = 0 = a b = 0 = a b = 0 = a , thus a ∼ b a \sim b a ∼ b .
Otherwise 0 ≠ a = b ′ b = b ′ a ′ a 0 \ne a = b'b = b'a'a 0 = a = b ′ b = b ′ a ′ a resp. 1 ⋅ a = b ′ a ′ a 1 \cdot a = b'a'a 1 ⋅ a = b ′ a ′ a . By cancellation in domains, we have 1 = b ′ a ′ 1 = b'a' 1 = b ′ a ′ . Thus b ′ , a ′ ∈ R × b',a' \in R^{\times} b ′ , a ′ ∈ R × are units, and a ∼ b a \sim b a ∼ b .
Since a ∣ a a \mid a a ∣ a is trivial, the assumption a ∣ b a \mid b a ∣ b yields, by part (1), a ∣ g = gcd ( a , b ) a \mid g = \gcd(a,b) a ∣ g = g cd( a , b ) . But by the other part of the definition of a gcd, g g g has to divide each of the two arguments of the gcd as well, i.e. g ∣ a g \mid a g ∣ a . By the first part of mutual divisibility, gcd ( a , b ) = g ∼ a \gcd(a,b) = g \sim a g cd( a , b ) = g ∼ a . The proof for lcm follows the same lines.
(3) Set g 1 = gcd ( a , b ) , g 2 = gcd ( c , d ) g_1 = \gcd(a,b), g_2 = \gcd(c,d) g 1 = g cd( a , b ) , g 2 = g cd( c , d ) . By definition of ∼ \sim ∼ there are a ′ , b ′ ∈ R × a',b' \in R^{\times} a ′ , b ′ ∈ R × with a ′ a = c , b ′ b = d a'a = c, b'b = d a ′ a = c , b ′ b = d . Thus g 1 ∣ a ∣ c g_1 \mid a \mid c g 1 ∣ a ∣ c and g 1 ∣ b ∣ d g_1 \mid b \mid d g 1 ∣ b ∣ d . By part (1), g 1 ∣ gcd ( c , d ) = g 2 g_1 \mid\gcd(c,d) = g_2 g 1 ∣ g cd( c , d ) = g 2 . By symmetry of ∼ \sim ∼ , we also have g 2 ∣ g 1 g_2 \mid g_1 g 2 ∣ g 1 and thus by part (2) g 1 ∼ g 2 g_1 \sim g_2 g 1 ∼ g 2 . The proof for lcm follows mutatis mutandis.
(4) We use that in any ring d ⋅ 0 = 0 d \cdot0 = 0 d ⋅ 0 = 0 for any d ∈ R d \in R d ∈ R .
a ∼ 0 a \sim0 a ∼ 0 implies the existence of d ∈ R × d \in R^{\times} d ∈ R × with d a = 0 da = 0 d a = 0 . In particular, as a unit in an integral domain, d ≠ 0 d \ne0 d = 0 . Since integral domains have no nontrivial zero divisors, we deduce a = 0 a = 0 a = 0 and finish “⇒ \Rightarrow ⇒ ”. “⇐ \Leftarrow ⇐ ” follows from the assumption and since “∼ \sim ∼ ” is reflexive as an equivalence relation.
It is straightforward to see gcd ( 0 , 0 ) = 0 \gcd(0,0) = 0 g cd( 0 , 0 ) = 0 , so we may assume a ≠ 0 a \ne0 a = 0 .
We have g = gcd ( a , 0 ) ∣ a g = \gcd(a,0) \mid a g = g cd( a , 0 ) ∣ a by definition, but a = 1 ⋅ a a = 1 \cdot a a = 1 ⋅ a and 0 = 0 ⋅ a 0 = 0 \cdot a 0 = 0 ⋅ a show that a a a is also a common divisor of the pair ( a , 0 ) (a,0) ( a , 0 ) ; hence a ∣ g a \mid g a ∣ g by definition. By mutual divisibility from (2), we get gcd ( a , 0 ) = g ∼ a \gcd(a,0) = g \sim a g cd( a , 0 ) = g ∼ a . For the third part, “⇐ \Leftarrow ⇐ ” follows from gcd ( 0 , 0 ) = 0 \gcd(0,0)=0 g cd( 0 , 0 ) = 0 . For “⇒ \Rightarrow ⇒ ”, we argue by contraposition. Assume that wlog a ≠ 0 a \ne0 a = 0 . Since by definition g = gcd ( a , b ) g=\gcd(a,b) g = g cd( a , b ) is in particular a divisor of a a a , there is some g 1 ∈ R g_1 \in R g 1 ∈ R such that g 1 g = a ≠ 0 g_1g=a \ne0 g 1 g = a = 0 . By the first part, g ∼ 0 g \sim0 g ∼ 0 implies g = 0 g=0 g = 0 . We had 0 = g 1 ⋅ 0 = g 1 g = a ≠ 0 0=g_1 \cdot0=g_1g=a \ne0 0 = g 1 ⋅ 0 = g 1 g = a = 0 , a contradiction. Thus g = gcd ( a , b ) ≁ 0 g=\gcd(a,b) \not\sim0 g = g cd( a , b ) ∼ 0 demonstrating the third property.
Again, in any ring d ⋅ 0 = 0 d \cdot0=0 d ⋅ 0 = 0 for any d ∈ R d \in R d ∈ R , thus the only multiple of 0 0 0 is 0 0 0 . Hence 0 0 0 is the only candidate for an lcm with 0 0 0 , and since 0 ⋅ a = 0 0 \cdot a=0 0 ⋅ a = 0 also proves that 0 0 0 is a multiple of a a a , we establish 0 0 0 as the only common multiple of a a a and 0 0 0 , hence the claimed lcm ( a , 0 ) = 0 \operatorname{lcm}(a,0)=0 lcm ( a , 0 ) = 0 .
From the previously proven identity and by commutativity (proven below without recourse to this part), we immediately get “⇐ \Leftarrow ⇐ ”. For “⇒ \Rightarrow ⇒ ”, assume that a ≠ 0 ≠ b a \ne0 \ne b a = 0 = b . Then 0 ≠ a b 0 \ne ab 0 = ab (integral domain) is a common multiple of a a a and b b b , thus l = lcm ( a , b ) ∣ a b ≠ 0 l=\operatorname{lcm}(a,b)\mid ab \ne0 l = lcm ( a , b ) ∣ ab = 0 , i.e. there is an l 1 l_1 l 1 with l 1 l = a b ≠ 0 l_1l=ab \ne0 l 1 l = ab = 0 . As was already argued several times, this can only happen if l ≠ 0 l \ne0 l = 0 ; hence, by the first part, l ≁ 0 l \not\sim0 l ∼ 0 . By contraposition, l ∼ 0 l \sim0 l ∼ 0 needs to imply the negation of a ≠ 0 ≠ b a \ne0 \ne b a = 0 = b , which yields the fifth and last claim.
(5) A proof of a generalized product formula valid for domains that are not necessarily GCD, with the gcd replaced by the v-gcd, may be found in [1 Thm. 2.1 (2)] . Since for GCD domains both versions agree, i.e. v-gcd is the gcd, this proves the product formula in the above form as a corollary.
(6) By the definition of the gcd (logical symmetry in a a a and b b b ), we get gcd ( a , b ) ∣ gcd ( b , a ) \gcd(a,b)\mid\gcd(b,a) g cd( a , b ) ∣ g cd( b , a ) and gcd ( b , a ) ∣ gcd ( a , b ) \gcd(b,a)\mid\gcd(a,b) g cd( b , a ) ∣ g cd( a , b ) . Part (2) finishes the proof for gcds. Lcms are dealt with similarly.
(7) Set g 1 = gcd ( a , gcd ( b , c ) ) g_1=\gcd(a,\gcd(b,c)) g 1 = g cd( a , g cd( b , c )) , g 2 = gcd ( gcd ( a , b ) , c ) g_2=\gcd(\gcd(a,b),c) g 2 = g cd( g cd( a , b ) , c ) , and assume that a , b , c a,b,c a , b , c are nonzero. By part (4), all gcds involved and in particular g 1 , g 2 g_1,g_2 g 1 , g 2 are nonzero as well. We first show that g 1 ∣ g 2 g_1\mid g_2 g 1 ∣ g 2 . By definition we have g 1 ∣ a , g 1 ∣ gcd ( b , c ) g_1\mid a,g_1\mid\gcd(b,c) g 1 ∣ a , g 1 ∣ g cd( b , c ) , thus g 1 g_1 g 1 divides all three, “g 1 ∣ a , b , c g_1\mid a,b,c g 1 ∣ a , b , c ” for short. Therefore g 1 ∣ gcd ( a , b ) g_1\mid\gcd(a,b) g 1 ∣ g cd( a , b ) by definition together with g 1 ∣ c g_1\mid c g 1 ∣ c , which in turn by definition yields g 1 ∣ gcd ( gcd ( a , b ) , c ) = g 2 g_1\mid\gcd(\gcd(a,b),c)=g_2 g 1 ∣ g cd( g cd( a , b ) , c ) = g 2 . By symmetry, g 2 ∣ g 1 g_2\mid g_1 g 2 ∣ g 1 as well, and mutual divisibility in part (2) yields g 1 ∼ g 2 g_1\sim g_2 g 1 ∼ g 2 .
If one of a , b , c a,b,c a , b , c is zero, the claim follows by a straightforward computation using part (4). E.g., for a = 0 a=0 a = 0 , we have gcd ( a , gcd ( b , c ) ) = gcd ( 0 , gcd ( b , c ) ) ∼ ( 6 ) , ( 4 ) gcd ( b , c ) \gcd(a,\gcd(b,c))=\gcd(0,\gcd(b,c))\mathrel{\overset{(6),(4)}{\sim}}\gcd(b,c) g cd( a , g cd( b , c )) = g cd( 0 , g cd( b , c )) ∼ ( 6 ) , ( 4 ) g cd( b , c ) and gcd ( gcd ( a , b ) , c ) = gcd ( gcd ( 0 , b ) , c ) ∼ ( 6 ) , ( 4 ) , ( 3 ) gcd ( b , c ) \gcd(\gcd(a,b),c)=\gcd(\gcd(0,b),c)\mathrel{\overset{(6),(4),(3)}{\sim}}\gcd(b,c) g cd( g cd( a , b ) , c ) = g cd( g cd( 0 , b ) , c ) ∼ ( 6 ) , ( 4 ) , ( 3 ) g cd( b , c ) . The claim follows by transitivity of ∼ \sim ∼ . The cases b = 0 b=0 b = 0 and c = 0 c=0 c = 0 follow in the same way.
The property for lcm is proved mutatis mutandis.
(8) By the two parts of the definition, gcd ( a , a ) ∣ a \gcd(a,a)\mid a g cd( a , a ) ∣ a and a ∣ gcd ( a , a ) a\mid\gcd(a,a) a ∣ g cd( a , a ) . Mutual divisibility in part (2) implies association. The same holds for lcms.
(9) From c ∣ 1 c\mid1 c ∣ 1 , we get some c ′ ∈ R c' \in R c ′ ∈ R with c ′ c = 1 c'c=1 c ′ c = 1 . This means that c ′ , c c',c c ′ , c are units and associates of 1 1 1 , in particular c ∼ 1 c\sim1 c ∼ 1 . Replacing c c c by gcd ( a , b ) \gcd(a,b) g cd( a , b ) yields the second statement.
By assumption, there is an a ′ ∈ R a' \in R a ′ ∈ R such that a ′ a = b a'a=b a ′ a = b . Assume, to the contrary, that 1 ∼ b 1\sim b 1 ∼ b . Then there is some b ′ ∈ R × b' \in R^{\times} b ′ ∈ R × with b ′ b = 1 b'b=1 b ′ b = 1 , i.e. a ∣ a ′ a = b ∣ b ′ b = 1 a\mid a'a=b\mid b'b=1 a ∣ a ′ a = b ∣ b ′ b = 1 . By the first part, this yields the contradiction a ∼ 1 a\sim1 a ∼ 1 to the given assumption. Hence b ∤ 1 b\nmid1 b ∤ 1 , the third claim.
By definition, g = gcd ( a , 1 ) g=\gcd(a,1) g = g cd( a , 1 ) divides its constituents, so in particular g ∣ 1 g\mid1 g ∣ 1 . By the previously proven implication, this yields g ∼ 1 g\sim1 g ∼ 1 and thus the claim.
Clearly a a a is a multiple of a a a (via 1 ⋅ a = 1 ⋅ a 1\cdot a=1\cdot a 1 ⋅ a = 1 ⋅ a ), i.e., a common multiple, and thus l = lcm ( a , 1 ) ∣ a l=\operatorname{lcm}(a,1)\mid a l = lcm ( a , 1 ) ∣ a . On the other hand, a a a needs to divide lcm ( a , 1 ) = l \operatorname{lcm}(a,1)=l lcm ( a , 1 ) = l by definition of lcm. By part (2), the combination l ∣ a l\mid a l ∣ a and a ∣ l a\mid l a ∣ l yields l ∼ a l\sim a l ∼ a , the last claim of (9).
(10) Again, we prove the gcd identity. The lcm identity follows by a dual argument. Again, the case of some a , b , c a,b,c a , b , c being zero can be settled by direct calculations in the spirit of previous parts, so we assume all of them to be nonzero.
We first prove a special case of part (13). Set g = gcd ( a , b ) g=\gcd(a,b) g = g cd( a , b ) ; we show that gcd ( a g , b g ) ∼ 1 \gcd\left(\frac{a}{g},\frac{b}{g}\right)\sim1 g cd( g a , g b ) ∼ 1 .
First, since g g g divides both a a a and b b b , there are a ′ , b ′ a',b' a ′ , b ′ with a ′ g = a , b ′ g = b a'g=a,b'g=b a ′ g = a , b ′ g = b , and we have a g = a ′ , b g = b ′ ∈ R ⊆ Frac ( R ) \frac{a}{g}=a',\frac{b}{g}=b' \in R\subseteq\operatorname{Frac}(R) g a = a ′ , g b = b ′ ∈ R ⊆ Frac ( R ) . Now set g ′ = gcd ( a ′ , b ′ ) g'=\gcd(a',b') g ′ = g cd( a ′ , b ′ ) . Then g ′ g' g ′ divides both a ′ , b ′ a',b' a ′ , b ′ , and since divisibility is multiplicative, g ′ g g'g g ′ g divides both a ′ g = a a'g=a a ′ g = a , b ′ g = b b'g=b b ′ g = b . Part (1) yields g ′ g ∣ gcd ( a , b ) = g g'g \mid\gcd(a,b)=g g ′ g ∣ g cd( a , b ) = g , so there is some g ′ ′ g'' g ′′ with g ′ ′ g = g = 1 ⋅ g g''g=g=1\cdot g g ′′ g = g = 1 ⋅ g , implying g ′ g' g ′ and g ′ ′ g'' g ′′ are units by canceling in domains. This shows gcd ( a ′ , b ′ ) = g ′ ∼ 1 \gcd(a',b')=g'\sim1 g cd( a ′ , b ′ ) = g ′ ∼ 1 as claimed.
With the previous notation, a = a ′ g , b = b ′ g a=a'g,b=b'g a = a ′ g , b = b ′ g . Therefore c a = c a ′ g ca=ca'g c a = c a ′ g , c b = c b ′ g cb=cb'g c b = c b ′ g and thus c g cg c g is a common divisor of both c a ca c a and c b cb c b , implying c g ∣ gcd ( c a , c b ) = G cg\mid\gcd(ca,cb)=G c g ∣ g cd( c a , c b ) = G . Hence there is some m m m with m c g = G mcg=G m c g = G . By definition, G ∣ c a G\mid ca G ∣ c a and G ∣ c b G\mid cb G ∣ c b , i.e. for some d a , d b ∈ R ∖ { 0 } d_a,d_b\in R\setminus\{0\} d a , d b ∈ R ∖ { 0 } we have d a m c g = d a G = c a = c a ′ g d_amcg=d_aG=ca=ca'g d a m c g = d a G = c a = c a ′ g , d b m c g = d b G = c b = c b ′ g d_bmcg=d_bG=cb=cb'g d b m c g = d b G = c b = c b ′ g , and canceling c g cg c g on both sides yields d a m = a ′ d_am=a' d a m = a ′ , d b m = b ′ d_bm=b' d b m = b ′ . So m m m divides both a ′ , b ′ a',b' a ′ , b ′ ; therefore m ∣ gcd ( a ′ , b ′ ) ∼ 1 m\mid\gcd(a',b')\sim1 m ∣ g cd( a ′ , b ′ ) ∼ 1 . By part (9), m ∼ 1 m\sim1 m ∼ 1 is a unit, i.e. c g = c gcd ( a , b ) cg=c\gcd(a,b) c g = c g cd( a , b ) and G = gcd ( c a , c b ) G=\gcd(ca,cb) G = g cd( c a , c b ) are associates as claimed.
(11) Set g = gcd ( a , b ) g=\gcd(a,b) g = g cd( a , b ) , so g ∣ a g\mid a g ∣ a , g ∣ b g\mid b g ∣ b , which implies g ∣ a g\mid a g ∣ a , g ∣ b c g\mid bc g ∣ b c , which by part (1) yields g ∣ gcd ( a , b c ) g\mid\gcd(a,bc) g ∣ g cd( a , b c ) which is the first claim.
Let l = lcm ( a , b c ) l=\operatorname{lcm}(a,bc) l = lcm ( a , b c ) , thus a ∣ l a\mid l a ∣ l , b c ∣ l bc\mid l b c ∣ l , which implies a ∣ l a\mid l a ∣ l , b ∣ l b\mid l b ∣ l , which by part (1) yields lcm ( a , b ) ∣ l \operatorname{lcm}(a,b)\mid l lcm ( a , b ) ∣ l which is the other claim.
(12) By assumption, we have some a ′ ∈ R a'\in R a ′ ∈ R with a ′ a = b c a'a=bc a ′ a = b c . By previous parts, we have
gcd ( a , a ) ∣ ( 11 ) gcd ( a , a ′ a ) = gcd ( a , c b ) ∣ ( 11 ) , ( 6 ) gcd ( c a , c b ) ∼ ( 10 ) c gcd ( a , b ) ∼ c ⋅ 1 = c . \gcd(a,a)\mathrel{\overset{(11)}{\mid}}\gcd(a,a'a)=\gcd(a,cb)\mathrel{\overset{(11),(6)}{\mid}}\gcd(ca,cb)\mathrel{\overset{(10)}{\sim}}c\gcd(a,b)\sim c\cdot1=c. g cd( a , a ) ∣ ( 11 ) g cd( a , a ′ a ) = g cd( a , c b ) ∣ ( 11 ) , ( 6 ) g cd( c a , c b ) ∼ ( 10 ) c g cd( a , b ) ∼ c ⋅ 1 = c .
We have g = gcd ( a , c ) ∣ ( 11 ) gcd ( a , b c ) = G g=\gcd(a,c)\mathrel{\overset{(11)}{\mid}}\gcd(a,bc)=G g = g cd( a , c ) ∣ ( 11 ) g cd( a , b c ) = G . By definition, G = gcd ( a , b c ) G=\gcd(a,bc) G = g cd( a , b c ) satisfies G ∣ a G\mid a G ∣ a and G ∣ b c G\mid bc G ∣ b c . Let d = gcd ( G , b ) d=\gcd(G,b) d = g cd( G , b ) . Then we have d ∣ G ∣ a d\mid G\mid a d ∣ G ∣ a , d ∣ b d\mid b d ∣ b , i.e. d ∣ gcd ( a , b ) ∼ 1 d\mid\gcd(a,b)\sim1 d ∣ g cd( a , b ) ∼ 1 , which by part (9) yields gcd ( G , b ) = d ∼ 1 \gcd(G,b)=d\sim1 g cd( G , b ) = d ∼ 1 . The first part applied to ( a , b , c ) (a,b,c) ( a , b , c ) replaced by ( G , b , c ) (G,b,c) ( G , b , c ) (conditions gcd ( G , b ) ∼ 1 \gcd(G,b)\sim1 g cd( G , b ) ∼ 1 and G ∣ b c G\mid bc G ∣ b c are satisfied) yields G ∣ c G\mid c G ∣ c . Combine this with G ∣ a G\mid a G ∣ a to get G ∣ gcd ( a , c ) = g G\mid\gcd(a,c)=g G ∣ g cd( a , c ) = g . Combining g ∣ G g\mid G g ∣ G and G ∣ g G\mid g G ∣ g by part (2) yields g ∼ G g\sim G g ∼ G , i.e. the claim.
For the third part of (12), we use the previous part of (12) just proven and the product formula (5) (treating zero cases separately, as before).
lcm ( a , b c ) ∼ ( 5 ) a b c gcd ( a , b c ) ∼ ( 12 ) a b c gcd ( a , c ) = b ⋅ a c gcd ( a , c ) ∼ ( 5 ) b ⋅ lcm ( a , c ) . \operatorname{lcm}(a,bc)\mathrel{\overset{(5)}{\sim}}\frac{abc}{\gcd(a,bc)}\mathrel{\overset{(12)}{\sim}}\frac{abc}{\gcd(a,c)}=b\cdot\frac{ac}{\gcd(a,c)}\mathrel{\overset{(5)}{\sim}}b\cdot\operatorname{lcm}(a,c). lcm ( a , b c ) ∼ ( 5 ) g cd( a , b c ) ab c ∼ ( 12 ) g cd( a , c ) ab c = b ⋅ g cd( a , c ) a c ∼ ( 5 ) b ⋅ lcm ( a , c ) .
There are a ′ , b ′ ∈ R a',b'\in R a ′ , b ′ ∈ R with a ′ a = c = b ′ b a'a=c=b'b a ′ a = c = b ′ b , i.e. a ∣ b ′ b a\mid b'b a ∣ b ′ b . The first statement of (12) together with the assumption gcd ( a , b ) ∼ 1 \gcd(a,b)\sim1 g cd( a , b ) ∼ 1 yields a ∣ b ′ a\mid b' a ∣ b ′ . Since divisibility is multiplicative, we have a b ∣ b ′ b = c ab\mid b'b=c ab ∣ b ′ b = c , which is the claim.
(13) We already proved a special case of part (13) in (10) for c = gcd ( a , b ) c=\gcd(a,b) c = g cd( a , b ) . Note that an arbitrary common divisor c c c of a , b a,b a , b also satisfies c ∣ gcd ( a , b ) = g c\mid\gcd(a,b)=g c ∣ g cd( a , b ) = g , i.e. there are a ′ , b ′ , c ′ ∈ R a',b',c'\in R a ′ , b ′ , c ′ ∈ R with a ′ g = a a'g=a a ′ g = a , b ′ g = b b'g=b b ′ g = b , c ′ c = g c'c=g c ′ c = g . Then we calculate
gcd ( a c , b c ) = gcd ( a ′ g c , b ′ g c ) = gcd ( a ′ c ′ , b ′ c ′ ) ∼ ( 10 ) c ′ ⋅ gcd ( a ′ , b ′ ) = g c ⋅ gcd ( a ′ , b ′ ) = g ⋅ gcd ( a ′ , b ′ ) c ∼ ( 10 ) gcd ( a ′ g , b ′ g ) c = gcd ( a , b ) c . \gcd\left(\frac{a}{c},\frac{b}{c}\right)=\gcd\left(\frac{a'g}{c},\frac{b'g}{c}\right)=\gcd(a'c',b'c')\mathrel{\overset{(10)}{\sim}}c'\cdot\gcd(a',b')=\frac{g}{c}\cdot\gcd(a',b')=\frac{g\cdot\gcd(a',b')}{c}\mathrel{\overset{(10)}{\sim}}\frac{\gcd(a'g,b'g)}{c}=\frac{\gcd(a,b)}{c}. g cd( c a , c b ) = g cd( c a ′ g , c b ′ g ) = g cd( a ′ c ′ , b ′ c ′ ) ∼ ( 10 ) c ′ ⋅ g cd( a ′ , b ′ ) = c g ⋅ g cd( a ′ , b ′ ) = c g ⋅ g cd( a ′ , b ′ ) ∼ ( 10 ) c g cd( a ′ g , b ′ g ) = c g cd( a , b ) .
Since c ≠ 0 c\ne0 c = 0 , all the fractions are actually defined in R ⊆ Frac ( R ) R\subseteq\operatorname{Frac}(R) R ⊆ Frac ( R ) .
As before, treating zero cases separately and taking care that all fractions stay in R R R , we obtain the other assertion from the product formula (5):
lcm ( a c , b c ) ∼ ( 5 ) a c ⋅ b c gcd ( a c , b c ) ∼ a c ⋅ b c gcd ( a , b ) c = a b c gcd ( a , b ) ∼ ( 5 ) lcm ( a , b ) c . \operatorname{lcm}\left(\frac{a}{c},\frac{b}{c}\right)\mathrel{\overset{(5)}{\sim}}\frac{\frac{a}{c}\cdot\frac{b}{c}}{\gcd\left(\frac{a}{c},\frac{b}{c}\right)}\mathrel{\sim}\frac{\frac{a}{c}\cdot\frac{b}{c}}{\frac{\gcd(a,b)}{c}}=\frac{ab}{c\gcd(a,b)}\mathrel{\overset{(5)}{\sim}}\frac{\operatorname{lcm}(a,b)}{c}. lcm ( c a , c b ) ∼ ( 5 ) g cd( c a , c b ) c a ⋅ c b ∼ c g c d ( a , b ) c a ⋅ c b = c g cd( a , b ) ab ∼ ( 5 ) c lcm ( a , b ) .
(14) Treat the special cases of a , b ∈ R a,b\in R a , b ∈ R and n ∈ N n\in\mathbb{N} n ∈ N not all being nonzero separately. The convention c 0 = 1 c^0=1 c 0 = 1 works out in all cases, even for 0 0 = 1 0^0=1 0 0 = 1 .
Let g = gcd ( a , b ) g=\gcd(a,b) g = g cd( a , b ) . Then a = a ′ g a=a'g a = a ′ g , b = b ′ g b=b'g b = b ′ g for some a ′ , b ′ ∈ R a',b'\in R a ′ , b ′ ∈ R . Hence a n = a ′ n g n a^n=a'^ng^n a n = a ′ n g n , b n = b ′ n g n b^n=b'^ng^n b n = b ′ n g n and therefore g n ∣ gcd ( a n , b n ) g^n\mid\gcd(a^n,b^n) g n ∣ g cd( a n , b n ) , meaning gcd ( a n , b n ) g n \frac{\gcd(a^n,b^n)}{g^n} g n g c d ( a n , b n ) is well-defined in R R R .
By the special case of part (13) proven in (10), we have gcd ( a g , b g ) ∼ 1 \gcd\left(\frac{a}{g},\frac{b}{g}\right)\sim1 g cd( g a , g b ) ∼ 1 . Note that we can assume n ≥ 1 n \ge1 n ≥ 1 . Inductively applying the second identity of part (12), we get
gcd ( a g , b n g n ) = gcd ( a g , ( b g ) n ) ∼ ( 12 ) gcd ( a g , ( b g ) n − 1 ) ∼ ( 12 ) ⋯ ∼ ( 12 ) gcd ( a g , ( b g ) 1 ) ∼ 1. \gcd\left(\frac{a}{g},\frac{b^n}{g^n}\right)=\gcd\left(\frac{a}{g},\left(\frac{b}{g}\right)^n\right)\mathrel{\sim}^{(12)}\gcd\left(\frac{a}{g},\left(\frac{b}{g}\right)^{n-1}\right)\mathrel{\sim}^{(12)}\cdots\mathrel{\sim}^{(12)}\gcd\left(\frac{a}{g},\left(\frac{b}{g}\right)^1\right)\sim1. g cd( g a , g n b n ) = g cd( g a , ( g b ) n ) ∼ ( 12 ) g cd( g a , ( g b ) n − 1 ) ∼ ( 12 ) ⋯ ∼ ( 12 ) g cd( g a , ( g b ) 1 ) ∼ 1.
Similarly, using the association just proven and (6), we get
gcd ( a n g n , b n g n ) = gcd ( ( a g ) n , b n g n ) ∼ ( 12 ) gcd ( ( a g ) n − 1 , b n g n ) ∼ ( 12 ) ⋯ ∼ ( 12 ) gcd ( ( a g ) 1 , b n g n ) ∼ 1. \gcd\left(\frac{a^n}{g^n},\frac{b^n}{g^n}\right)=\gcd\left(\left(\frac{a}{g}\right)^n,\frac{b^n}{g^n}\right)\mathrel{\sim}^{(12)}\gcd\left(\left(\frac{a}{g}\right)^{n-1},\frac{b^n}{g^n}\right)\mathrel{\sim}^{(12)}\cdots\mathrel{\sim}^{(12)}\gcd\left(\left(\frac{a}{g}\right)^1,\frac{b^n}{g^n}\right)\sim1. g cd( g n a n , g n b n ) = g cd( ( g a ) n , g n b n ) ∼ ( 12 ) g cd( ( g a ) n − 1 , g n b n ) ∼ ( 12 ) ⋯ ∼ ( 12 ) g cd( ( g a ) 1 , g n b n ) ∼ 1.
Then we get
gcd ( a n , b n ) g n ∼ ( 13 ) gcd ( a n g n , b n g n ) ∼ 1. \frac{\gcd(a^n,b^n)}{g^n}\mathrel{\sim}^{(13)}\gcd\left(\frac{a^n}{g^n},\frac{b^n}{g^n}\right)\sim1. g n g cd( a n , b n ) ∼ ( 13 ) g cd( g n a n , g n b n ) ∼ 1.
Multiplying both sides by g n g^n g n and canceling yields the first identity. The lcm identity follows by applying the product formula (5), treating zero cases separately, as seen before.
(15) Set l = lcm ( a , b ) , g = gcd ( a , l ) l=\operatorname{lcm}(a,b),g=\gcd(a,l) l = lcm ( a , b ) , g = g cd( a , l ) . Since a a a is an argument of the outer gcd, we have by definition g ∣ a g\mid a g ∣ a . We also have a ∣ a a\mid a a ∣ a and, by definition of lcm, a ∣ l a\mid l a ∣ l , which, by (1), combines to a ∣ g a\mid g a ∣ g . Mutual divisibility (2) yields the claim.
The second absorption statement is proven analogously.
(16) First, we treat the zero cases explicitly:
a = 0 : gcd ( 0 , lcm ( b , c ) ) ∼ ( 4 ) lcm ( b , c ) ∼ ( 3 ) , ( 4 ) , ( 6 ) twice lcm ( gcd ( 0 , b ) , gcd ( 0 , c ) ) , a=0:\gcd(0,\operatorname{lcm}(b,c))\mathrel{\sim}^{(4)}\operatorname{lcm}(b,c)\mathrel{\sim}^{(3),(4),(6)\ \text{twice}}\operatorname{lcm}(\gcd(0,b),\gcd(0,c)), a = 0 : g cd( 0 , lcm ( b , c )) ∼ ( 4 ) lcm ( b , c ) ∼ ( 3 ) , ( 4 ) , ( 6 ) twice lcm ( g cd( 0 , b ) , g cd( 0 , c )) ,
b = 0 : gcd ( a , lcm ( 0 , c ) ) = ( 4 ) , ( 6 ) gcd ( a , 0 ) ∼ ( 15 ) lcm ( a , gcd ( a , c ) ) ∼ ( 3 ) , ( 4 ) lcm ( gcd ( a , 0 ) , gcd ( a , c ) ) , b=0:\gcd(a,\operatorname{lcm}(0,c))\mathrel{=}^{(4),(6)}\gcd(a,0)\mathrel{\sim}^{(15)}\operatorname{lcm}(a,\gcd(a,c))\mathrel{\sim}^{(3),(4)}\operatorname{lcm}(\gcd(a,0),\gcd(a,c)), b = 0 : g cd( a , lcm ( 0 , c )) = ( 4 ) , ( 6 ) g cd( a , 0 ) ∼ ( 15 ) lcm ( a , g cd( a , c )) ∼ ( 3 ) , ( 4 ) lcm ( g cd( a , 0 ) , g cd( a , c )) ,
c = 0 : gcd ( a , lcm ( b , 0 ) ) = ( 4 ) gcd ( a , 0 ) ∼ ( 6 ) , ( 15 ) lcm ( gcd ( a , b ) , a ) ∼ ( 3 ) , ( 4 ) lcm ( gcd ( a , b ) , gcd ( a , 0 ) ) . c=0:\gcd(a,\operatorname{lcm}(b,0))\mathrel{=}^{(4)}\gcd(a,0)\mathrel{\sim}^{(6),(15)}\operatorname{lcm}(\gcd(a,b),a)\mathrel{\sim}^{(3),(4)}\operatorname{lcm}(\gcd(a,b),\gcd(a,0)). c = 0 : g cd( a , lcm ( b , 0 )) = ( 4 ) g cd( a , 0 ) ∼ ( 6 ) , ( 15 ) lcm ( g cd( a , b ) , a ) ∼ ( 3 ) , ( 4 ) lcm ( g cd( a , b ) , g cd( a , 0 )) .
Next, assume g = gcd ( a , gcd ( b , c ) ) ∼ 1 g=\gcd(a,\gcd(b,c))\sim1 g = g cd( a , g cd( b , c )) ∼ 1 and set g b = gcd ( a , b ) , g c = gcd ( a , c ) , L H S = gcd ( a , lcm ( b , c ) ) , R H S = lcm ( g b , g c ) g_b=\gcd(a,b),g_c=\gcd(a,c),LHS=\gcd(a,\operatorname{lcm}(b,c)),RHS=\operatorname{lcm}(g_b,g_c) g b = g cd( a , b ) , g c = g cd( a , c ) , L H S = g cd( a , lcm ( b , c )) , R H S = lcm ( g b , g c ) .
We first show R H S ∣ L H S RHS\mid LHS R H S ∣ L H S . We have g b ∣ a g_b\mid a g b ∣ a , g b ∣ b ∣ lcm ( b , c ) g_b\mid b\mid\operatorname{lcm}(b,c) g b ∣ b ∣ lcm ( b , c ) , thus g b ∣ L H S g_b\mid LHS g b ∣ L H S . From g c ∣ a , g c ∣ c ∣ lcm ( b , c ) g_c\mid a,g_c\mid c\mid\operatorname{lcm}(b,c) g c ∣ a , g c ∣ c ∣ lcm ( b , c ) follows g c ∣ L H S g_c\mid LHS g c ∣ L H S . I.e., L H S LHS L H S is a common multiple of g b , g c g_b,g_c g b , g c , and by part (1), we have R H S = lcm ( g b , g c ) ∣ L H S RHS=\operatorname{lcm}(g_b,g_c)\mid LHS R H S = lcm ( g b , g c ) ∣ L H S .
We prove an intermediate result, which we will use shortly. For x , y , z ∈ R ∖ { 0 } x,y,z\in R\setminus\{0\} x , y , z ∈ R ∖ { 0 } , we have
( x ∣ y z ∧ gcd ( gcd ( x , y ) , gcd ( x , z ) ) ∼ 1 ) ⇒ x ∼ gcd ( x , y ) gcd ( x , z ) . \left(x\mid yz\mathbin{\wedge}\gcd(\gcd(x,y),\gcd(x,z))\sim1\right)\Rightarrow x\sim\gcd(x,y)\gcd(x,z). ( x ∣ y z ∧ g cd( g cd( x , y ) , g cd( x , z )) ∼ 1 ) ⇒ x ∼ g cd( x , y ) g cd( x , z ) .
Set g y = gcd ( x , y ) , g z = gcd ( x , z ) g_y=\gcd(x,y),g_z=\gcd(x,z) g y = g cd( x , y ) , g z = g cd( x , z ) . By the definition of gcds, we have g y ∣ x g_y\mid x g y ∣ x and g z ∣ x g_z\mid x g z ∣ x , and by assumption gcd ( g y , g z ) ∼ 1 \gcd(g_y,g_z)\sim1 g cd( g y , g z ) ∼ 1 , so the last part of (12) implies g y g z ∣ x g_yg_z\mid x g y g z ∣ x , i.e. there is an x ′ ∈ R ∖ { 0 } x'\in R\setminus\{0\} x ′ ∈ R ∖ { 0 } with x ′ g y g z = x x'g_yg_z=x x ′ g y g z = x . By definition of gcds, we have g y ∣ y , g z ∣ z g_y\mid y,g_z\mid z g y ∣ y , g z ∣ z , hence there are y ′ , z ′ ∈ R ∖ { 0 } y',z'\in R\setminus\{0\} y ′ , z ′ ∈ R ∖ { 0 } with y ′ g y = y , z ′ g z = z y'g_y=y,z'g_z=z y ′ g y = y , z ′ g z = z . We have g y = gcd ( x , y ) = gcd ( x ′ g y g z , y ′ g y ) ∼ ( 10 ) g y gcd ( x ′ g z , y ′ ) g_y=\gcd(x,y)=\gcd(x'g_yg_z,y'g_y)\mathrel{\sim}^{(10)}g_y\gcd(x'g_z,y') g y = g cd( x , y ) = g cd( x ′ g y g z , y ′ g y ) ∼ ( 10 ) g y g cd( x ′ g z , y ′ ) , so gcd ( x ′ g z , y ′ ) ∼ 1 \gcd(x'g_z,y')\sim1 g cd( x ′ g z , y ′ ) ∼ 1 and symmetrically gcd ( x ′ g y , z ′ ) ∼ 1 \gcd(x'g_y,z')\sim1 g cd( x ′ g y , z ′ ) ∼ 1 . Now, by assumption, x ′ g y g z = x ∣ y z = y ′ g y z ′ g z x'g_yg_z=x\mid yz=y'g_yz'g_z x ′ g y g z = x ∣ y z = y ′ g y z ′ g z , i.e., after canceling g y g z g_yg_z g y g z , we have x ′ ∣ y ′ z ′ x'\mid y'z' x ′ ∣ y ′ z ′ . Since gcd ( x ′ , y ′ ) ∣ gcd ( x ′ g z , y ′ ) ∼ 1 \gcd(x',y')\mid\gcd(x'g_z,y')\sim1 g cd( x ′ , y ′ ) ∣ g cd( x ′ g z , y ′ ) ∼ 1 , part (9) implies gcd ( x ′ , y ′ ) ∼ 1 \gcd(x',y')\sim1 g cd( x ′ , y ′ ) ∼ 1 , which, together with x ′ ∣ y ′ z ′ x'\mid y'z' x ′ ∣ y ′ z ′ , implies via part (12) that x ′ ∣ z ′ x'\mid z' x ′ ∣ z ′ . Symmetrically, we get gcd ( x ′ , z ′ ) ∼ 1 \gcd(x',z')\sim1 g cd( x ′ , z ′ ) ∼ 1 , which, together with x ′ ∣ z ′ = z ′ ⋅ 1 x'\mid z'=z'\cdot1 x ′ ∣ z ′ = z ′ ⋅ 1 , yields again via (12) that x ′ ∣ 1 x'\mid1 x ′ ∣ 1 , and by (9) we have x ′ ∼ 1 x'\sim1 x ′ ∼ 1 . Therefore, the previous relation x = x ′ g y g z ∼ 1 ⋅ g y g z = g y g z = gcd ( x , y ) gcd ( x , z ) x=x'g_yg_z\sim1\cdot g_yg_z=g_yg_z=\gcd(x,y)\gcd(x,z) x = x ′ g y g z ∼ 1 ⋅ g y g z = g y g z = g cd( x , y ) g cd( x , z ) is exactly the intermediate result.
Using commutativity (6) and associativity (7) repeatedly and finally the intermediate assumption of coprimality g = gcd ( a , gcd ( b , c ) ) ∼ 1 g=\gcd(a,\gcd(b,c))\sim1 g = g cd( a , g cd( b , c )) ∼ 1 , we have:
gcd ( g b , g c ) = gcd ( gcd ( a , b ) , gcd ( a , c ) ) ∼ ( 6 ) , ( 7 ) gcd ( gcd ( a , a ) , gcd ( b , c ) ) ∼ ( 8 ) gcd ( a , gcd ( b , c ) ) ∼ 1. \gcd(g_b,g_c)=\gcd(\gcd(a,b),\gcd(a,c))\mathrel{\sim}^{(6),(7)}\gcd(\gcd(a,a),\gcd(b,c))\mathrel{\sim}^{(8)}\gcd(a,\gcd(b,c))\sim1. g cd( g b , g c ) = g cd( g cd( a , b ) , g cd( a , c )) ∼ ( 6 ) , ( 7 ) g cd( g cd( a , a ) , g cd( b , c )) ∼ ( 8 ) g cd( a , g cd( b , c )) ∼ 1.
Thus R H S ∼ ( 5 ) g b g c gcd ( g b , g c ) ∼ g b g c 1 = g b g c RHS\mathrel{\sim}^{(5)}\frac{g_bg_c}{\gcd(g_b,g_c)}\mathrel{\sim}\frac{g_bg_c}{1}=g_bg_c R H S ∼ ( 5 ) g c d ( g b , g c ) g b g c ∼ 1 g b g c = g b g c . Next L H S = gcd ( a , lcm ( b , c ) ) ∣ lcm ( b , c ) ∣ b c LHS=\gcd(a,\operatorname{lcm}(b,c))\mid\operatorname{lcm}(b,c)\mid bc L H S = g cd( a , lcm ( b , c )) ∣ lcm ( b , c ) ∣ b c , which we call ☆1. By definition of gcd, L H S ∣ a LHS\mid a L H S ∣ a , i.e. there is an a ′ ∈ R a'\in R a ′ ∈ R with a ′ L H S = a a'LHS=a a ′ L H S = a , hence by (6) and (11): gcd ( L H S , b ) ∣ gcd ( a ′ L H S , b ) = gcd ( a , b ) = g b \gcd(LHS,b)\mid\gcd(a'LHS,b)=\gcd(a,b)=g_b g cd( L H S , b ) ∣ g cd( a ′ L H S , b ) = g cd( a , b ) = g b and symmetrically gcd ( L H S , c ) ∣ g c \gcd(LHS,c)\mid g_c g cd( L H S , c ) ∣ g c . By (6) and another twofold application of (11), this yields gcd ( gcd ( L H S , b ) , gcd ( L H S , c ) ) ∣ gcd ( g b , g c ) ∼ 1 \gcd(\gcd(LHS,b),\gcd(LHS,c))\mid\gcd(g_b,g_c)\sim1 g cd( g cd( L H S , b ) , g cd( L H S , c )) ∣ g cd( g b , g c ) ∼ 1 .
Setting x = L H S , y = b , z = c x=LHS,y=b,z=c x = L H S , y = b , z = c in the intermediate result, ⋆ 1 \star_1 ⋆ 1 and ⋆ 2 \star_2 ⋆ 2 are the two conditions needed to apply the intermediate result, which gives L H S ∼ gcd ( L H S , b ) gcd ( L H S , c ) ∣ g b g c = R H S LHS\sim\gcd(LHS,b)\gcd(LHS,c)\mid g_bg_c=RHS L H S ∼ g cd( L H S , b ) g cd( L H S , c ) ∣ g b g c = R H S .
Now mutual divisibility (2) proves the claim L H S ∼ R H S LHS\sim RHS L H S ∼ R H S in the coprime case.
Let g = gcd ( a , gcd ( b , c ) ) g=\gcd(a,\gcd(b,c)) g = g cd( a , g cd( b , c )) (possibly ∤ 1 \nmid1 ∤ 1 ). Since the zero cases are already resolved, by (4) we may assume g ≠ 0 g\ne0 g = 0 . There are a ′ , b ′ , c ′ ∈ R ∖ { 0 } a',b',c'\in R\setminus\{0\} a ′ , b ′ , c ′ ∈ R ∖ { 0 } such that a ′ g = a , b ′ g = b , c ′ g = c a'g=a,b'g=b,c'g=c a ′ g = a , b ′ g = b , c ′ g = c . Set
g ′ = gcd ( a ′ , gcd ( b ′ , c ′ ) ) ≠ ( 4 ) 0. g'=\gcd(a',\gcd(b',c'))\overset{(4)}{\ne}0. g ′ = g cd( a ′ , g cd( b ′ , c ′ )) = ( 4 ) 0.
Tracing the definitions of gcd (details left to the reader), we deduce g ′ g ∣ c ′ g = c g'g\mid c'g=c g ′ g ∣ c ′ g = c and g ′ g ∣ b ′ g = b g'g\mid b'g=b g ′ g ∣ b ′ g = b , hence g ′ g ∣ gcd ( b , c ) g'g\mid\gcd(b,c) g ′ g ∣ g cd( b , c ) , and further g ′ g ∣ a ′ g = a g'g\mid a'g=a g ′ g ∣ a ′ g = a , thus g ′ g ∣ gcd ( a , gcd ( b , c ) ) = g g'g\mid\gcd(a,\gcd(b,c))=g g ′ g ∣ g cd( a , g cd( b , c )) = g , i.e. there is a g ′ ′ g'' g ′′ with g ′ ′ g ′ g = g g''g'g=g g ′′ g ′ g = g . Canceling g ≠ 0 g\ne0 g = 0 in the domain R R R yields g ′ ′ g ′ = 1 g''g'=1 g ′′ g ′ = 1 , hence g ′ g' g ′ is a unit and g ′ ∼ 1 g'\sim1 g ′ ∼ 1 . Thus a ′ , b ′ , c ′ a',b',c' a ′ , b ′ , c ′ satisfy the coprime condition g ′ ∼ 1 g'\sim1 g ′ ∼ 1 , and we may apply the special case that we just proved. We have
gcd ( a , lcm ( b , c ) ) = gcd ( a ′ g , lcm ( b ′ g , c ′ g ) ) ∼ ( 10 ) twice g gcd ( a ′ , lcm ( b ′ , c ′ ) ) ∼ coprime case \gcd(a,\operatorname{lcm}(b,c))=\gcd(a'g,\operatorname{lcm}(b'g,c'g))\overset{(10)\ \text{twice}}{\sim}g\gcd(a',\operatorname{lcm}(b',c'))\overset{\text{coprime case}}{\sim} g cd( a , lcm ( b , c )) = g cd( a ′ g , lcm ( b ′ g , c ′ g )) ∼ ( 10 ) twice g g cd( a ′ , lcm ( b ′ , c ′ )) ∼ coprime case
g lcm ( gcd ( a ′ , b ′ ) , gcd ( a ′ , c ′ ) ) ∼ ( 10 ) thrice lcm ( gcd ( a ′ g , b ′ g ) , gcd ( a ′ g , c ′ g ) ) = lcm ( gcd ( a , b ) , gcd ( a , c ) ) . g\operatorname{lcm}(\gcd(a',b'),\gcd(a',c'))\overset{(10)\ \text{thrice}}{\sim}\operatorname{lcm}(\gcd(a'g,b'g),\gcd(a'g,c'g))=\operatorname{lcm}(\gcd(a,b),\gcd(a,c)). g lcm ( g cd( a ′ , b ′ ) , g cd( a ′ , c ′ )) ∼ ( 10 ) thrice lcm ( g cd( a ′ g , b ′ g ) , g cd( a ′ g , c ′ g )) = lcm ( g cd( a , b ) , g cd( a , c )) .
For the second distributivity law, we again treat the three zero cases explicitly. We may assume that all terms involved are now nonzero. We apply the distributivity result just proven with ( a , b , c ) (a,b,c) ( a , b , c ) replaced by ( lcm ( a , b ) , a , c ) (\operatorname{lcm}(a,b),a,c) ( lcm ( a , b ) , a , c ) and then absorption (15):
gcd ( lcm ( a , b ) , lcm ( a , c ) ) ∼ d i s t lcm ( gcd ( lcm ( a , b ) , a ) , gcd ( lcm ( a , b ) , c ) ) ∼ ( 3 ) , ( 6 ) , ( 15 ) lcm ( a , gcd ( lcm ( a , b ) , c ) ) . \gcd(\operatorname{lcm}(a,b),\operatorname{lcm}(a,c))\overset{\mathrm{dist}}{\sim}\operatorname{lcm}(\gcd(\operatorname{lcm}(a,b),a),\gcd(\operatorname{lcm}(a,b),c))\overset{(3),(6),(15)}{\sim}\operatorname{lcm}(a,\gcd(\operatorname{lcm}(a,b),c)). g cd( lcm ( a , b ) , lcm ( a , c )) ∼ dist lcm ( g cd( lcm ( a , b ) , a ) , g cd( lcm ( a , b ) , c )) ∼ ( 3 ) , ( 6 ) , ( 15 ) lcm ( a , g cd( lcm ( a , b ) , c )) .
Apply the previous distributivity again with ( a , b , c ) (a,b,c) ( a , b , c ) replaced by ( c , a , b ) (c,a,b) ( c , a , b ) :
gcd ( lcm ( a , b ) , c ) ∼ ( 6 ) gcd ( c , lcm ( a , b ) ) ∼ d i s t lcm ( gcd ( c , a ) , gcd ( c , b ) ) ∼ ( 6 ) twice lcm ( gcd ( a , c ) , gcd ( b , c ) ) . \gcd(\operatorname{lcm}(a,b),c)\overset{(6)}{\sim}\gcd(c,\operatorname{lcm}(a,b))\overset{\mathrm{dist}}{\sim}\operatorname{lcm}(\gcd(c,a),\gcd(c,b))\overset{(6)\ \text{twice}}{\sim}\operatorname{lcm}(\gcd(a,c),\gcd(b,c)). g cd( lcm ( a , b ) , c ) ∼ ( 6 ) g cd( c , lcm ( a , b )) ∼ dist lcm ( g cd( c , a ) , g cd( c , b )) ∼ ( 6 ) twice lcm ( g cd( a , c ) , g cd( b , c )) .
Substituting this into the previous equation yields
gcd ( lcm ( a , b ) , lcm ( a , c ) ) ∼ lcm ( a , gcd ( lcm ( a , b ) , c ) ) = s u b s t lcm ( a , lcm ( gcd ( a , c ) , gcd ( b , c ) ) ) ∼ ( 7 ) \gcd(\operatorname{lcm}(a,b),\operatorname{lcm}(a,c))\sim\operatorname{lcm}(a,\gcd(\operatorname{lcm}(a,b),c))\overset{\mathrm{subst}}{=}\operatorname{lcm}(a,\operatorname{lcm}(\gcd(a,c),\gcd(b,c)))\overset{(7)}{\sim} g cd( lcm ( a , b ) , lcm ( a , c )) ∼ lcm ( a , g cd( lcm ( a , b ) , c )) = subst lcm ( a , lcm ( g cd( a , c ) , g cd( b , c ))) ∼ ( 7 )
lcm ( lcm ( a , gcd ( a , c ) ) , gcd ( b , c ) ) ∼ ( 15 ) lcm ( a , gcd ( b , c ) ) . \operatorname{lcm}(\operatorname{lcm}(a,\gcd(a,c)),\gcd(b,c))\overset{(15)}{\sim}\operatorname{lcm}(a,\gcd(b,c)). lcm ( lcm ( a , g cd( a , c )) , g cd( b , c )) ∼ ( 15 ) lcm ( a , g cd( b , c )) .
This is the second distributivity law.
Uniqueness of ( c , d ) (c,d) ( c , d ) up to associates in the gf and adequate properties
First, I repeat an argument from [14 p. 460] for easier reference.
Lemma B.1. Let R R R be a Bézout domain and a , b ∈ R ∖ { 0 } a,b \in R \setminus\{0\} a , b ∈ R ∖ { 0 } . If c , d ∈ R c,d \in R c , d ∈ R satisfy the gf condition for ( a , b ) (a,b) ( a , b ) , then they satisfy the adequate condition.
Proof. Recall the gf condition: a = c d a=cd a = c d , gcd ( c , b ) ∼ 1 \gcd(c,b)\sim1 g cd( c , b ) ∼ 1 , and b ∈ ( d ) b\in\sqrt{(d)} b ∈ ( d ) . The first two properties of gf and adequate coincide. Let d ′ ∈ R ∖ R × d' \in R \setminus R^{\times} d ′ ∈ R ∖ R × with d ′ ∣ d d' \mid d d ′ ∣ d , i.e. there is some d ′ ′ ∈ R d'' \in R d ′′ ∈ R with d ′ d ′ ′ = d d'd''=d d ′ d ′′ = d . From b ∈ ( d ) b\in\sqrt{(d)} b ∈ ( d ) , we get n ∈ N > 0 n\in\mathbb{N}_{>0} n ∈ N > 0 , B ∈ R B\in R B ∈ R with b n = d B b^n=dB b n = d B , and thus b n = d B = d ′ d ′ ′ B b^n=dB=d'd''B b n = d B = d ′ d ′′ B . Assume for contradiction that gcd ( d ′ , b ) ∼ 1 \gcd(d',b)\sim1 g cd( d ′ , b ) ∼ 1 . Then (note that d ′ n − 1 ∈ R d'^{n-1}\in R d ′ n − 1 ∈ R is well-defined since n > 0 n>0 n > 0 )
d ′ ∣ d ′ gcd ( d ′ n − 1 , d ′ ′ B ) ∼ gcd ( d ′ n , d ′ d ′ ′ B ) = gcd ( d ′ n , b n ) ∼ gcd ( d ′ , b ) n ∼ 1. d'\mid d'\gcd(d'^{n-1},d''B)\mathrel{\sim}\gcd(d'^n,d'd''B)=\gcd(d'^n,b^n)\mathrel{\sim}\gcd(d',b)^n\sim1. d ′ ∣ d ′ g cd( d ′ n − 1 , d ′′ B ) ∼ g cd( d ′ n , d ′ d ′′ B ) = g cd( d ′ n , b n ) ∼ g cd( d ′ , b ) n ∼ 1. By (9) we had d ′ ∼ 1 d'\sim1 d ′ ∼ 1 contradicting d ′ ∉ R × d'\notin R^\times d ′ ∈ / R × . Thus gcd ( d ′ , b ) ≁ 1 \gcd(d',b)\not\sim1 g cd( d ′ , b ) ∼ 1 . This finishes the proof. □
Next I prove that ( c , d ) (c,d) ( c , d ) satisfying the adequate (resp. gf) condition for a pair ( a , b ) (a,b) ( a , b ) are unique up to associates.
Lemma B.2. Let R R R be a Bézout domain. Given a , b ∈ R ∖ { 0 } a,b\in R\setminus\{0\} a , b ∈ R ∖ { 0 } and two pairs ( c 1 , d 1 ) , ( c 2 , d 2 ) ∈ R 2 (c_1,d_1),(c_2,d_2)\in R^2 ( c 1 , d 1 ) , ( c 2 , d 2 ) ∈ R 2 satisfying the adequate conditions for ( a , b ) (a,b) ( a , b ) , we have c 1 ∼ c 2 c_1\sim c_2 c 1 ∼ c 2 and d 1 ∼ d 2 d_1\sim d_2 d 1 ∼ d 2 .
Proof. Note that 0 ≠ a = c 1 d 1 = c 2 d 2 0\ne a=c_1d_1=c_2d_2 0 = a = c 1 d 1 = c 2 d 2 implies that all occurring denominators are nonzero. Set d = gcd ( d 1 , d 2 ) d=\gcd(d_1,d_2) d = g cd( d 1 , d 2 ) , d ~ 1 = d 1 d \widetilde{d}_1=\frac{d_1}{d} d 1 = d d 1 , d ~ 2 = d 2 d \widetilde{d}_2=\frac{d_2}{d} d 2 = d d 2 , and c = a d c=\frac{a}{d} c = d a . We start with some auxiliary calculations.
gcd ( d ~ 1 , d ~ 2 ) = gcd ( d 1 d , d 2 d ) ∼ gcd ( d 1 , d 2 ) d = d d = 1 , \gcd(\widetilde{d}_1,\widetilde{d}_2)=\gcd\left(\frac{d_1}{d},\frac{d_2}{d}\right)\mathrel{\sim}\frac{\gcd(d_1,d_2)}{d}=\frac{d}{d}=1, g cd( d 1 , d 2 ) = g cd( d d 1 , d d 2 ) ∼ d g cd( d 1 , d 2 ) = d d = 1 , d ~ 1 d ~ 2 lcm ( c 1 , c 2 ) ∼ lcm ( c 1 d ~ 1 d ~ 2 , c 2 d ~ 1 d ~ 2 ) = lcm ( a d 1 d ~ 1 d ~ 2 , a d 2 d ~ 1 d ~ 2 ) = lcm ( a d ~ 2 d d ~ 1 , a d ~ 1 d d ~ 2 ) \widetilde{d}_1\widetilde{d}_2\operatorname{lcm}(c_1,c_2)\mathrel{\sim}\operatorname{lcm}(c_1\widetilde{d}_1\widetilde{d}_2,c_2\widetilde{d}_1\widetilde{d}_2)=\operatorname{lcm}\left(\frac{a}{d_1}\widetilde{d}_1\widetilde{d}_2,\frac{a}{d_2}\widetilde{d}_1\widetilde{d}_2\right)=\operatorname{lcm}\left(\frac{a\widetilde{d}_2}{d\widetilde{d}_1},\frac{a\widetilde{d}_1}{d\widetilde{d}_2}\right) d 1 d 2 lcm ( c 1 , c 2 ) ∼ lcm ( c 1 d 1 d 2 , c 2 d 1 d 2 ) = lcm ( d 1 a d 1 d 2 , d 2 a d 1 d 2 ) = lcm ( d d 1 a d 2 , d d 2 a d 1 ) a d lcm ( d ~ 2 , d ~ 1 ) ∼ a d ⋅ d ~ 1 d ~ 2 gcd ( d ~ 1 , d ~ 2 ) ∼ c ⋅ d ~ 1 d ~ 2 1 = c d ~ 1 d ~ 2 . \frac{a}{d}\operatorname{lcm}(\widetilde{d}_2,\widetilde{d}_1)\mathrel{\sim}\frac{a}{d}\cdot\frac{\widetilde{d}_1\widetilde{d}_2}{\gcd(\widetilde{d}_1,\widetilde{d}_2)}\mathrel{\sim}c\cdot\frac{\widetilde{d}_1\widetilde{d}_2}{1}=c\widetilde{d}_1\widetilde{d}_2. d a lcm ( d 2 , d 1 ) ∼ d a ⋅ g cd( d 1 , d 2 ) d 1 d 2 ∼ c ⋅ 1 d 1 d 2 = c d 1 d 2 . Canceling in the domain R R R yields c ∼ lcm ( c 1 , c 2 ) c\sim\operatorname{lcm}(c_1,c_2) c ∼ lcm ( c 1 , c 2 ) . Now we check that ( c , d ) (c,d) ( c , d ) also satisfies the adequate conditions for ( a , b ) (a,b) ( a , b ) . The definition of c c c immediately gives a = c d a=cd a = c d .
gcd ( c , b ) ∼ gcd ( lcm ( c 1 , c 2 ) , b ) ∼ lcm ( gcd ( c 1 , b ) , gcd ( c 2 , b ) ) ∼ lcm ( 1 , 1 ) ∼ 1. \gcd(c,b)\mathrel{\sim}\gcd(\operatorname{lcm}(c_1,c_2),b)\mathrel{\sim}\operatorname{lcm}(\gcd(c_1,b),\gcd(c_2,b))\mathrel{\sim}\operatorname{lcm}(1,1)\mathrel{\sim}1. g cd( c , b ) ∼ g cd( lcm ( c 1 , c 2 ) , b ) ∼ lcm ( g cd( c 1 , b ) , g cd( c 2 , b )) ∼ lcm ( 1 , 1 ) ∼ 1. By the definition of gcd, we have d = gcd ( d 1 , d 2 ) ∣ d 1 d=\gcd(d_1,d_2)\mid d_1 d = g cd( d 1 , d 2 ) ∣ d 1 . Given any d ′ ∈ R ∖ R × d'\in R\setminus R^\times d ′ ∈ R ∖ R × with d ′ ∣ d d'\mid d d ′ ∣ d , we immediately get d ′ ∣ d ∣ d 1 d'\mid d\mid d_1 d ′ ∣ d ∣ d 1 and can apply the adequate conditions for ( c 1 , d 1 ) (c_1,d_1) ( c 1 , d 1 ) . We get gcd ( d ′ , b ) ≁ 1 \gcd(d',b)\not\sim1 g cd( d ′ , b ) ∼ 1 as required. Thus ( c , d ) (c,d) ( c , d ) is adequate for ( a , b ) (a,b) ( a , b ) .
Next, consider d 1 ′ = c c 1 d'_1=\frac{c}{c_1} d 1 ′ = c 1 c . Since d 1 ′ = c c 1 ∣ c d'_1=\frac{c}{c_1}\mid c d 1 ′ = c 1 c ∣ c , we have
gcd ( d 1 ′ , b ) ∣ gcd ( c , b ) ∼ 1 ⇒ gcd ( d 1 ′ , b ) ∼ 1. \gcd(d'_1,b)\mid\gcd(c,b)\mathrel{\sim}1\Rightarrow\gcd(d'_1,b)\sim1. g cd( d 1 ′ , b ) ∣ g cd( c , b ) ∼ 1 ⇒ g cd( d 1 ′ , b ) ∼ 1. From c ∣ a c\mid a c ∣ a , we get d 1 ′ = c c 1 ∣ a c 1 = d 1 d'_1=\frac{c}{c_1}\mid\frac{a}{c_1}=d_1 d 1 ′ = c 1 c ∣ c 1 a = d 1 . Assume d 1 ′ ∉ R × d'_1\notin R^\times d 1 ′ ∈ / R × . Then ( c 1 , d 1 ) (c_1,d_1) ( c 1 , d 1 ) adequate for ( a , b ) (a,b) ( a , b ) implies gcd ( d 1 ′ , b ) ≁ 1 \gcd(d'_1,b)\not\sim1 g cd( d 1 ′ , b ) ∼ 1 , a contradiction. Thus d 1 ′ ∼ 1 d'_1\sim1 d 1 ′ ∼ 1 . From d 1 ′ c 1 = c d'_1c_1=c d 1 ′ c 1 = c we then get c 1 ∼ c c_1\sim c c 1 ∼ c . By symmetry, also c ∼ c 2 c\sim c_2 c ∼ c 2 holds, and by transitivity of ∼ \sim ∼ finally c 1 ∼ c 2 c_1\sim c_2 c 1 ∼ c 2 . d 1 = c c 1 ∼ c c 2 = d 2 d_1=\frac{c}{c_1}\sim\frac{c}{c_2}=d_2 d 1 = c 1 c ∼ c 2 c = d 2 finishes the proof. □
Corollary B.3. Let R R R be a Bézout domain. Given a , b ∈ R ∖ { 0 } a,b\in R\setminus\{0\} a , b ∈ R ∖ { 0 } and two pairs ( c 1 , d 1 ) , ( c 2 , d 2 ) ∈ R 2 (c_1,d_1),(c_2,d_2)\in R^2 ( c 1 , d 1 ) , ( c 2 , d 2 ) ∈ R 2 satisfying the gf conditions for ( a , b ) (a,b) ( a , b ) , we have c 1 ∼ c 2 c_1\sim c_2 c 1 ∼ c 2 and d 1 ∼ d 2 d_1\sim d_2 d 1 ∼ d 2 .
Proof. We apply Lem. B.1 to show that the ( c i , d i ) (c_i,d_i) ( c i , d i ) have adequate properties for ( a , b ) (a,b) ( a , b ) and then Lem. B.2 to get the association. □
Good factorization implies lcm splitting
I originally found a proof that gf implies lcm splitting. The global structure is similar to the proof of adequate implies lcm splitting in Thm. 3.2 , but the details about the radical are different and could be interesting for algorithmic versions.
Lemma C.1. Let R R R be a GCD domain. Given a , b ∈ R ∖ { 0 } a,b \in R \setminus\{0\} a , b ∈ R ∖ { 0 } and some c , d ∈ R c,d \in R c , d ∈ R satisfying the gf conditions, i.e. a = c d a=cd a = c d , gcd ( c , b ) ∼ 1 \gcd(c,b)\sim1 g cd( c , b ) ∼ 1 and b ∈ ( d ) b \in\sqrt{(d)} b ∈ ( d ) , we have gcd ( c , d ) ∼ 1 \gcd(c,d)\sim1 g cd( c , d ) ∼ 1 .
Proof. b ∈ ( d ) b \in\sqrt{(d)} b ∈ ( d ) implies by definition of the radical the existence of n ∈ N > 0 n \in\mathbb{N}_{>0} n ∈ N > 0 , B ∈ R B \in R B ∈ R with b n = d B b^n=dB b n = d B .
gcd ( c , d ) ∣ ( 11 ) gcd ( c , d B ) = gcd ( c , b n ) ∣ ( 6 ) , ( 11 ) gcd ( c n , b n ) ∼ ( 14 ) gcd ( c , b ) n ∼ 1 n = 1. \gcd(c,d)\overset{(11)}{\mid}\gcd(c,dB)=\gcd(c,b^n)\overset{(6),(11)}{\mid}\gcd(c^n,b^n)\overset{(14)}{\sim}\gcd(c,b)^n\sim1^n=1. g cd( c , d ) ∣ ( 11 ) g cd( c , d B ) = g cd( c , b n ) ∣ ( 6 ) , ( 11 ) g cd( c n , b n ) ∼ ( 14 ) g cd( c , b ) n ∼ 1 n = 1. This proves the lemma by (9). □
Theorem C.2. A gf Bézout domain R R R is lcm splitting.
Proof. Let a , b ∈ R ∖ { 0 } a,b \in R \setminus\{0\} a , b ∈ R ∖ { 0 } be arbitrary (cf. Rem. 2.4). Set g = gcd ( a , b ) , l = lcm ( a , b ) g=\gcd(a,b), l=\operatorname{lcm}(a,b) g = g cd( a , b ) , l = lcm ( a , b ) . From (5), a b ∼ g l ab\sim gl ab ∼ g l , we deduce l ∼ a b g = a g b g l\sim\frac{ab}{g}=\frac{a}{g}\frac{b}{g} l ∼ g ab = g a g b .
Now apply the gf property to the pair ( g , b g ) \left(g,\frac{b}{g}\right) ( g , g b ) to get g = g 1 g 2 g=g_1g_2 g = g 1 g 2 with gcd ( g 1 , b g ) ∼ 1 \gcd\left(g_1,\frac{b}{g}\right)\sim1 g cd( g 1 , g b ) ∼ 1 and b g ∈ ( g 2 ) \frac{b}{g}\in\sqrt{(g_2)} g b ∈ ( g 2 ) .
Apply gf a second time to ( g 2 , a g ) \left(g_2,\frac{a}{g}\right) ( g 2 , g a ) to get g 2 = g 1 ′ g 2 ′ g_2=g'_1g'_2 g 2 = g 1 ′ g 2 ′ with gcd ( g 2 ′ , a g ) ∼ 1 \gcd\left(g'_2,\frac{a}{g}\right)\sim1 g cd( g 2 ′ , g a ) ∼ 1 and a g ∈ ( g 1 ′ ) \frac{a}{g}\in\sqrt{(g'_1)} g a ∈ ( g 1 ′ ) . Overall, we have l ∼ a g b g = a g g 1 g 1 ′ g 2 ′ b g l\sim\frac{a}{g}\frac{b}{g}=\frac{a}{g}g_1g'_1g'_2\frac{b}{g} l ∼ g a g b = g a g 1 g 1 ′ g 2 ′ g b , and we will show some coprimality properties among these factors:
(A) gcd ( a g , g 2 ′ ) ∼ 1 \gcd\left(\frac{a}{g},g'_2\right)\sim1 g cd( g a , g 2 ′ ) ∼ 1 follows directly from the 2nd application of gf and (1) .
(B) Setting c = g c=g c = g in (13), we get gcd ( a g , b g ) ∼ gcd ( a , b ) g = g g = 1 \gcd\left(\frac{a}{g},\frac{b}{g}\right)\sim\frac{\gcd(a,b)}{g}=\frac{g}{g}=1 g cd( g a , g b ) ∼ g g c d ( a , b ) = g g = 1 .
(C) gcd ( g 1 , g 2 ′ ) ∣ ( 11 ) gcd ( g 1 , g 1 ′ g 2 ′ ) = gcd ( g 1 , g 2 ) ∼ 1 \gcd(g_1,g'_2)\overset{(11)}{\mid}\gcd(g_1,g'_1g'_2)=\gcd(g_1,g_2)\sim1 g cd( g 1 , g 2 ′ ) ∣ ( 11 ) g cd( g 1 , g 1 ′ g 2 ′ ) = g cd( g 1 , g 2 ) ∼ 1 , where the last step follows from Lem. C.1. By (9) we get gcd ( g 1 , g 2 ′ ) ∼ 1 \gcd(g_1,g'_2)\sim1 g cd( g 1 , g 2 ′ ) ∼ 1 .
(D) gcd ( g 1 , b g ) ∼ 1 \gcd\left(g_1,\frac{b}{g}\right)\sim1 g cd( g 1 , g b ) ∼ 1 follows directly from the 1st application of gf.
(E) By the 1st application of gf, we have some n ∈ N > 0 , B ∈ R n\in\mathbb{N}_{>0},B\in R n ∈ N > 0 , B ∈ R such that g 2 B = ( b g ) n g_2B=\left(\frac{b}{g}\right)^n g 2 B = ( g b ) n . By the 2nd application of gf, we have some m ∈ N > 0 , A ∈ R m\in\mathbb{N}_{>0},A\in R m ∈ N > 0 , A ∈ R such that g 1 ′ A = ( a g ) m g'_1A=\left(\frac{a}{g}\right)^m g 1 ′ A = ( g a ) m . Thus
g 1 ′ ∣ g 1 ′ g 2 ′ = g 2 ∣ g 2 B = ( b g ) n ∣ ( b g ) m n and g 1 ′ ∣ g 1 ′ A = ( a g ) m ∣ ( a g ) m n . g'_1\mid g'_1g'_2=g_2\mid g_2B=\left(\frac{b}{g}\right)^n\mid\left(\frac{b}{g}\right)^{mn}
\qquad\text{and}\qquad
g'_1\mid g'_1A=\left(\frac{a}{g}\right)^m\mid\left(\frac{a}{g}\right)^{mn}. g 1 ′ ∣ g 1 ′ g 2 ′ = g 2 ∣ g 2 B = ( g b ) n ∣ ( g b ) mn and g 1 ′ ∣ g 1 ′ A = ( g a ) m ∣ ( g a ) mn . By (1), we have g 1 ′ ∣ gcd ( ( a g ) m n , ( b g ) m n ) ∼ ( 14 ) gcd ( a g , b g ) m n ∼ ( B ) 1 m n = 1 g'_1\mid\gcd\left(\left(\frac{a}{g}\right)^{mn},\left(\frac{b}{g}\right)^{mn}\right)\overset{(14)}{\sim}\gcd\left(\frac{a}{g},\frac{b}{g}\right)^{mn}\overset{(B)}{\sim}1^{mn}=1 g 1 ′ ∣ g cd( ( g a ) mn , ( g b ) mn ) ∼ ( 14 ) g cd( g a , g b ) mn ∼ ( B ) 1 mn = 1 . By (9): g 1 ′ ∼ 1 g'_1\sim1 g 1 ′ ∼ 1 .
By (E), only four nontrivial factors remain: l ∼ a g g 1 g 2 ′ b g l\sim\frac{a}{g}g_1g'_2\frac{b}{g} l ∼ g a g 1 g 2 ′ g b . We set c = a g g 1 c=\frac{a}{g}g_1 c = g a g 1 and d = g 2 ′ b g d=g'_2\frac{b}{g} d = g 2 ′ g b . Now we apply (12) several times based on the coprimality results (A)–(D):
( α ) gcd ( a g , g 2 ′ ) ∼ 1 ⟹ ( 12 ) gcd ( a g , g 2 ′ b g ) ∼ gcd ( a g , b g ) ∼ ( B ) 1 , (\alpha)\ \gcd\left(\frac{a}{g},g'_2\right)\sim1\mathrel{\overset{(12)}{\Longrightarrow}}\gcd\left(\frac{a}{g},g'_2\frac{b}{g}\right)\sim\gcd\left(\frac{a}{g},\frac{b}{g}\right)\overset{(B)}{\sim}1, ( α ) g cd( g a , g 2 ′ ) ∼ 1 ⟹ ( 12 ) g cd( g a , g 2 ′ g b ) ∼ g cd( g a , g b ) ∼ ( B ) 1 , ( β ) gcd ( g 1 , g 2 ′ ) ∼ 1 ⟹ ( 12 ) gcd ( g 1 , g 2 ′ b g ) ∼ gcd ( g 1 , b g ) ∼ ( D ) 1 , (\beta)\ \gcd(g_1,g'_2)\sim1\mathrel{\overset{(12)}{\Longrightarrow}}\gcd\left(g_1,g'_2\frac{b}{g}\right)\sim\gcd\left(g_1,\frac{b}{g}\right)\overset{(D)}{\sim}1, ( β ) g cd( g 1 , g 2 ′ ) ∼ 1 ⟹ ( 12 ) g cd( g 1 , g 2 ′ g b ) ∼ g cd( g 1 , g b ) ∼ ( D ) 1 , ( γ ) gcd ( a g , g 2 ′ b g ) ∼ ( α ) 1 ⟹ ( 6 ) , ( 12 ) gcd ( c , d ) = gcd ( a g g 1 , g 2 ′ b g ) ∼ gcd ( g 1 , g 2 ′ b g ) ∼ ( β ) 1. (\gamma)\ \gcd\left(\frac{a}{g},g'_2\frac{b}{g}\right)\overset{(\alpha)}{\sim}1\mathrel{\overset{(6),(12)}{\Longrightarrow}}\gcd(c,d)=\gcd\left(\frac{a}{g}g_1,g'_2\frac{b}{g}\right)\sim\gcd\left(g_1,g'_2\frac{b}{g}\right)\overset{(\beta)}{\sim}1. ( γ ) g cd( g a , g 2 ′ g b ) ∼ ( α ) 1 ⟹ ( 6 ) , ( 12 ) g cd( c , d ) = g cd( g a g 1 , g 2 ′ g b ) ∼ g cd( g 1 , g 2 ′ g b ) ∼ ( β ) 1. Having shown gcd ( c , d ) ∼ 1 \gcd(c,d)\sim1 g cd( c , d ) ∼ 1 , we verify three further properties. c d = a g g 1 g 2 ′ b g ∼ l = lcm ( a , b ) cd=\frac{a}{g}g_1g'_2\frac{b}{g}\sim l=\operatorname{lcm}(a,b) c d = g a g 1 g 2 ′ g b ∼ l = lcm ( a , b ) by construction. From a = a g g = a g g 1 g 2 a=\frac{a}{g}g=\frac{a}{g}g_1g_2 a = g a g = g a g 1 g 2 , we get c = a g g 1 ∣ a c=\frac{a}{g}g_1\mid a c = g a g 1 ∣ a , and finally, from b = g b g = g 1 g 1 ′ g 2 ′ b g b=g\frac{b}{g}=g_1g'_1g'_2\frac{b}{g} b = g g b = g 1 g 1 ′ g 2 ′ g b , we get d = g 2 ′ b g ∣ b d=g'_2\frac{b}{g}\mid b d = g 2 ′ g b ∣ b . Since a , b a,b a , b were arbitrary, R R R is an lcm splitting Bézout domain. □
Appendix D. Examples of non-UFD Bézout domains that are lcm splitting
I will give three examples of non-UFD Bézout domains that are lcm splitting.
(1) Let k k k be a field and R 1 = ⋃ n = 1 ∞ k [ [ x 1 / n ] ] R_1 = \bigcup_{n=1}^{\infty} k[[x^{1/n}]] R 1 = ⋃ n = 1 ∞ k [[ x 1/ n ]] , i.e. formal power series with rational exponents of bounded denominators , also called Puiseux series . It is not hard to see that any r ∈ R 1 ∖ { 0 } r \in R_1 \setminus\{0\} r ∈ R 1 ∖ { 0 } can be written in the form r = c x q r = cx^q r = c x q for c ∈ R 1 × c \in R_1^{\times} c ∈ R 1 × a unit and q ∈ Q ≥ 0 q \in\mathbb{Q}_{\geq0} q ∈ Q ≥ 0 . In this ring, divisibility is a total order. In fact, R 1 R_1 R 1 is a valuation domain with valuation in Q ≥ 0 \mathbb{Q}_{\geq0} Q ≥ 0 and Bézout. Explicitly, given any a = c a x q a , b = c b x q b ∈ R 1 ∖ { 0 } a = c_a x^{q_a}, b = c_b x^{q_b} \in R_1 \setminus\{0\} a = c a x q a , b = c b x q b ∈ R 1 ∖ { 0 } and assuming wlog q a ≥ q b q_a \geq q_b q a ≥ q b , we have gcd ( a , b ) ∼ b \gcd(a,b) \sim b g cd( a , b ) ∼ b , lcm ( a , b ) ∼ a \operatorname{lcm}(a,b) \sim a lcm ( a , b ) ∼ a . An lcm splitting is given by c = a , d = 1 c = a, d = 1 c = a , d = 1 , which is easy to verify. It is, however, not a UFD, since it lacks atoms. Any candidate for an irreducible element 0 ≠ r = c x q 0 \ne r = cx^q 0 = r = c x q must have q > 0 q > 0 q > 0 (otherwise it is a unit), and r = ( c x q / 2 ) ⋅ x q / 2 r = (cx^{q/2}) \cdot x^{q/2} r = ( c x q /2 ) ⋅ x q /2 contradicts irreducibility. Since there are no irreducibles, there cannot exist a factorization into them.
(2) Let R 2 = O ( C ) R_2 = \mathcal{O}(\mathbb{C}) R 2 = O ( C ) be the ring of entire functions on the complex plane. R 2 R_2 R 2 is a Bézout domain by a well-known result of Helmer [10 ] . By the Weierstraß factorization theorem we can express any entire function as an infinite product of elementary factors E k E_k E k . Hence R 2 R_2 R 2 is not a UFD, as factorizations in a UFD are necessarily finite. Since this factorization is, however, essentially unique, it can also be used to construct lcm splittings. I leave the general details to the reader, but give one illustrative example of the shortcomings of the effective procedure (algorithm for a UFD) presented on p. 5.
By the Weierstraß factorization theorem, we can construct holomorphic functions f , g f,g f , g using elementary factors E k E_k E k as follows:
f ( z ) = ∏ n = 1 ∞ E p n ( z a n ) , g ( z ) = ∏ n = 1 ∞ E p n ( z b n ) , f(z) = \prod_{n=1}^{\infty} E_{p_n}\left(\frac{z}{a_n}\right), \qquad g(z) = \prod_{n=1}^{\infty} E_{p_n}\left(\frac{z}{b_n}\right), f ( z ) = n = 1 ∏ ∞ E p n ( a n z ) , g ( z ) = n = 1 ∏ ∞ E p n ( b n z ) ,
where ( a n ) = ( 1 , 2 , 3 , … ) (a_n) = (1,2,3,\ldots) ( a n ) = ( 1 , 2 , 3 , … ) is the sequence of positive integers, ( b n ) = ( 1 , 2 , 2 , 3 , 3 , 3 , 4 , … , k , … , k ⏟ k -fold , k + 1 , … ) (b_n) = (1,2,2,3,3,3,4,\ldots,\underbrace{k,\ldots,k}_{k\text{-fold}},k+1,\ldots) ( b n ) = ( 1 , 2 , 2 , 3 , 3 , 3 , 4 , … , k -fold k , … , k , k + 1 , … ) and ( p n ) = ( 2 , 2 , 2 , … ) (p_n) = (2,2,2,\ldots) ( p n ) = ( 2 , 2 , 2 , … ) is the constant sequence 2. The idea is that f f f has zeros exactly at the positive integers z i = i ∈ Z > 0 z_i = i \in\mathbb{Z}_{>0} z i = i ∈ Z > 0 with constant multiplicity 1, and g g g has the same zeros but with increasing multiplicities i i i . Set
a = a ( z ) = g ( 1 2 ( z + 1 ) ) ⋅ g ( 1 2 z ) ⋅ f ( 1 2 ( z + 1 ) ) , b = b ( z ) = g ( 1 2 ( z + 1 ) ) ⋅ g ( 1 2 z ) ⋅ f ( 1 2 z ) . a = a(z) = g\left(\frac{1}{2}(z+1)\right) \cdot g\left(\frac{1}{2}z\right) \cdot f\left(\frac{1}{2}(z+1)\right), \qquad b = b(z) = g\left(\frac{1}{2}(z+1)\right) \cdot g\left(\frac{1}{2}z\right) \cdot f\left(\frac{1}{2}z\right). a = a ( z ) = g ( 2 1 ( z + 1 ) ) ⋅ g ( 2 1 z ) ⋅ f ( 2 1 ( z + 1 ) ) , b = b ( z ) = g ( 2 1 ( z + 1 ) ) ⋅ g ( 2 1 z ) ⋅ f ( 2 1 z ) .
The zeros of a , b a,b a , b are still at the positive integers. For an odd positive integer z o z_o z o , the multiplicity of a a a is 1 more than that of b b b , i.e. mult z o ( a ) = mult z o ( b ) + 1 \operatorname{mult}_{z_o}(a) = \operatorname{mult}_{z_o}(b)+1 mult z o ( a ) = mult z o ( b ) + 1 , and vice versa for even positive integers z e z_e z e : mult z e ( a ) + 1 = mult z e ( b ) \operatorname{mult}_{z_e}(a)+1 = \operatorname{mult}_{z_e}(b) mult z e ( a ) + 1 = mult z e ( b ) . Moreover, the multiplicities increase linearly: mult 2 n − 1 ( b ) = mult 2 n ( a ) = n \operatorname{mult}_{2n-1}(b) = \operatorname{mult}_{2n}(a) = n mult 2 n − 1 ( b ) = mult 2 n ( a ) = n . The procedure on p. 5 produces c n , d n c_n,d_n c n , d n in step n n n such that for an even positive integer z k = 2 k z_k = 2k z k = 2 k , we have mult z k gcd ( c n , d n ) > 0 \operatorname{mult}_{z_k}\gcd(c_n,d_n)>0 mult z k g cd( c n , d n ) > 0 if and only if k ≥ 2 n k \geq2^n k ≥ 2 n . Since we can take k k k arbitrarily large, we see that this approach will never terminate globally. However, viewing each “finite place” z k z_k z k in isolation, it does terminate in the sense that mult z k gcd ( c n , d n ) = 0 \operatorname{mult}_{z_k}\gcd(c_n,d_n)=0 mult z k g cd( c n , d n ) = 0 for fixed k k k and all sufficiently large n n n .
Using the factorizations directly, we set
c = g ( 1 2 ( z + 1 ) ) ⋅ f ( 1 2 ( z + 1 ) ) , d = g ( 1 2 z ) ⋅ f ( 1 2 z ) . c = g\left(\frac{1}{2}(z+1)\right) \cdot f\left(\frac{1}{2}(z+1)\right), \qquad d = g\left(\frac{1}{2}z\right) \cdot f\left(\frac{1}{2}z\right). c = g ( 2 1 ( z + 1 ) ) ⋅ f ( 2 1 ( z + 1 ) ) , d = g ( 2 1 z ) ⋅ f ( 2 1 z ) .
We claim that ( c , d ) (c,d) ( c , d ) is an lcm splitting of ( a , b ) (a,b) ( a , b ) . Let g = gcd ( c , d ) g = \gcd(c,d) g = g cd( c , d ) . If g g g were not a unit, then g g g would have a zero at some z 0 ∈ C z_0 \in\mathbb{C} z 0 ∈ C . However, all zeros of c c c are at the odd positive integers and all zeros of d d d are at the even positive integers, so there cannot be a common zero z 0 z_0 z 0 . Thus g g g is a unit and can be chosen to be g = 1 g = 1 g = 1 . The divisibility properties c ∣ a c \mid a c ∣ a and d ∣ b d \mid b d ∣ b are obvious. The lcm condition c d ∼ lcm ( a , b ) cd \sim\operatorname{lcm}(a,b) c d ∼ lcm ( a , b ) follows similarly by comparing orders of vanishing at the relevant points.
(3) Let R 3 = Z ‾ R_3=\overline{\mathbb{Z}} R 3 = Z be the ring of algebraic integers. According to [20 Sec. 2.11] , it is a gf Bézout domain. Thus Thm. 3.2 and “gf ⇒ \Rightarrow ⇒ adequate” (cf. Rem. 2.3) immediately imply that it has lcm splittings. It is, however, instructive to look at gcds and lcm splittings explicitly.
Given explicit a , b ∈ R 3 ∖ { 0 } a,b\in R_3\setminus\{0\} a , b ∈ R 3 ∖ { 0 } , the gcd ( a , b ) \gcd(a,b) g cd( a , b ) might not exist in the ring of integers o K o_K o K of K = Q ( a , b ) K=\mathbb{Q}(a,b) K = Q ( a , b ) . But there is always an explicitly constructible finite extension field L / K L/K L / K such that gcd ( a , b ) \gcd(a,b) g cd( a , b ) exists in o L o_L o L . This does not mean that o L o_L o L is itself a GCD domain or even Bézout, but only that enough prime ideals of o K o_K o K become products of principal ideals of o L o_L o L such that gcd ( a , b ) \gcd(a,b) g cd( a , b ) exists and a , b a,b a , b admit suitable factorizations (not necessarily by irreducibles), facilitating the computation of gcd ( a , b ) \gcd(a,b) g cd( a , b ) . Using these factorizations, it is possible to find good factorization splittings and ultimately suitable c , d c,d c , d for lcm splitting.
Since irreducible elements of some o K o_K o K split further in suitable extensions o L o_L o L , and this does not stop after finitely many steps, R 3 = Z ‾ R_3=\overline{\mathbb{Z}} R 3 = Z is not a UFD. In the spirit of an argument for S S S in § 4.2 , one can also consider any integers a , k ∈ N > 1 a,k\in\mathbb{N}_{>1} a , k ∈ N > 1 and the infinite sequence of refining factors ( a 1 / k , a 1 / k 2 , a 1 / k 3 , … ) (a^{1/k},a^{1/k^2},a^{1/k^3},\ldots) ( a 1/ k , a 1/ k 2 , a 1/ k 3 , … ) to see directly that it is not a UFD.
In this example, the ring R 3 R_3 R 3 has the structure of a union: Z ‾ = ⋃ i o i \overline{\mathbb{Z}}=\bigcup_i o_i Z = ⋃ i o i , where one can even assume that it is an ascending chain, i.e. o i ⊂ o j o_i\subset o_j o i ⊂ o j for i < j i<j i < j (assuming that the index set is suitably ordered). The idea is that, for each pair ( a , b ) (a,b) ( a , b ) , there is some o i o_i o i where a , b a,b a , b factor nicely enough (although possibly not into irreducibles) such that gcd ( a , b ) , c , d \gcd(a,b),c,d g cd( a , b ) , c , d exist in o i o_i o i and such that gcd ( a , b ) , c , d \gcd(a,b),c,d g cd( a , b ) , c , d keep the properties needed for lcm splitting in extension o j o_j o j all the way up to Z ‾ \overline{\mathbb{Z}} Z .
If one wanted to generalize this tower strategy to general GCD domains R R R , one would have to show first that R R R admits suitable towers. I do not see how this could be possible without using anything about the specifics of the ring in question. It might also be necessary to make the tower depend on the given a , b a,b a , b . Note that Hutchins [11 Ex. 31, p. 66] mentions, for an example ring having similar properties as R 1 R_1 R 1 , that any countable tower of PIDs is Bézout. One would need some kind of converse theorem that any Bézout domain, potentially given some not to restrictive conditions, can be expressed as a suitable tower of nice domains, but I am not aware of any result in that direction. Anyway, that tower strategy might be worth looking at in further investigations of lcm splittings.
Institut für Mathematik, UZH, CH-8057 Zürich
Email address: preu@math.uzh.ch