You are on page 1of 16

PRODUCT BASES AND ENDOMORPHISMS

OF PRODUCTS OF INTEGERS

Copyright c 2007-2009 All Rights Reserved

E. F. CORNELIUS, JR.

College of Engineering and Science, University of Detroit Mercy,


Detroit, MI, USA 48221-3038
Email: efcornelius@comcast.net, frankcornelius@alum.mit.edu

Abstract

Let C(k; N) denote the k-tuples of nonnegative integers N. The endomor-


phism ring of the group of all integer-valued functions on C(k; N) is isomorphic
to the ring of k-dimensional row-…nite in…nite matrices over the integers. The
product bases of that group are represented by the multiplicative group of in-
vertible elements in that matrix ring. In particular, all endomorphisms of the
Baer-Specker group are multiplication by row-…nite in…nite integral matrices.
All products in the Baer-Specker group are characterized, and a lemma of Lás-
zló Fuchs regarding such products is revisited.

1. Introduction

The integral root basis < = f n (x) : n 2 Ng of the group of integral poly-
nomials Z[x] was introduced in [CS07] as 0 (x) 1, n+1 (x) = n (x)(x n)
for n 0. In [CS08] this basis was extended to a basis of the group of inte-
gral polynomials in k variables, Z[Xk ], which then was used to construct the
group of formalP power series with respect to the extended integral root basis,
Z[[Xk ]]< = f m2C(k;N) am m : am 2 Zg. The integral root basis has the
property that the valuation map v : Z[[Xk ]]< ! ZC(k;N) de…ned by vf (b) =
f (b), f 2 Z[[Xk ]]< , b 2 C(k; N), is a monomorphism, the image of which is the
group of multinomial points, M P (k; N) ZC(k;N) .

A principal result of [CS08, Theorem 5.4] is that ZC(k;N) has a subset A =


f m : m 2 C(k; N)g and M P (k; N) has a subset G = f m : m 2 C(k; N)g which
satisfy:

(1) For all a, m 2 C(k; N), m (a) = m (a).


P (2) Each f 2 ZC(k;N) has a unique representation of the form f =
m2C(k;N) bm m : bm 2 Z.

P (3) Each g 2 M P (k; N) has a unique representation of the form g =


m2C(k;N) cm m : cm 2 Z.

1
(4) For all m 2 C(k; N), m = pm m where pm is a positive integer.
(5) Let m = (m0 ; : : : ; mk 1 ), n = (n0 ; : : : ; nk 1 ) 2 C(k; N). If mi ni
for 0 i k 1, then from (4), pm j pn .

The structure of ZC(k;N) =M P (k; N) also was described in [CS08, Theorem 5.4]
and a test for inclusion in M P (k; N) was formulated in [CS08, Theorem 5.7].

From such observations evolved the de…nition of a product P basis. Let U =


fum : m 2 C(k; N)g be a subset of ZC(k;N) . If each expression m2C(k;N) bm um ,
bm 2 Z, represents an element Q of ZC(k;N) , the set of all such elements forms a
C(k;N)
subgroup of Z , denoted m2C(k;N) hum i, which is called a product in
P
ZC(k;N) . If the representations m2C(k;N) bm um , bm 2 Z, are unique, then U is
Q
said to be a product basis of m2C(k;N) hum i. The principal objective of this
paper is to characterize products and product bases in term of endomorphisms
of ZC(k;N) .

Abelian group notation and terminology are standard [F70 & 73]. N denotes
the nonnegative integers and C(k; N) the k-tuples of nonnegative integers for the
positive integer k. Elements of C(k; N) are denoted by bold lower-case letters,
m, n, etc. C(k; n) C(k; N) denotes all k-tuples with entries < n, n 2 N+ ,
the positive integers. Z denotes the ring of integers. The group of all functions
from C(k; N) to Z is denoted ZC(k;N) . Although ZC(k;N) is in fact a ring under
pointwise multiplication, that fact is merely noted here.
Q
Let P = ZC(1;N) = ZN = N Z, the group of all integer sequences, known as
the Baer-Specker group [B37, S50]. Although it is true that ZC(k;N) = Pk = P,
there is a de…nite di¤erence when attempting to …t polynomials. For exam-
ple, no permutation of the sequence (1; 0; 0; 0) can be generated by an integral
polynomial in one variable, because in any such permutation (a0 ; a1 ; a2 ; a3 ), the
expression a0 =6 a1 =2 + a2 =2 a3 =6 can never be an integer, a prerequisite
for such generation. However, all arrangements of (1; 0; 0; 0) indexed by the ele-
ments of C(2; 2) can be generated by integral polynomials in 2 variables. [CS08,
Section 2]. Groups are studied from the perspective of ZC(k;N) for that reason,
so that this paper is a continuation of [CS08].

2. k-Dimensional In…nite Integral Matrices And Their Operations

A k-dimensional in…nite integral matrix is a function M : C(k; N) C(k; N) !


Z, the value of which at (i; j) 2 C(k; N) C(k; N) is written as M (i; j). The
group of all k-dimensional in…nite integral matrices is denoted MZ (k; !), with
addition being de…ned pointwise.

F 2 MZ (k; !) is said to be row-…nite if, for each i, F (i; j) = 0 for all but
…nitely many j; i.e., if for each i, there exists li such that F (i; j) = 0 unless j li

2
(at each of the k coordinates). A column-…nite matrix is de…ned analogously.
The group of all row-…nite matrices in MZ (k; !) is denoted FMZ (k; !). T 2
MZ (k; !) is said to be lower-triangular if T (i; j) = 0 unless i j. The group
of all lower-triangular matrices in MZ (k; !) is denoted T MZ (k; !). Clearly
T MZ (k; !) FMZ (k; !).

For F 2 FMZ (k; !) and G 2 MZ (k; !), the product of F and G, denoted
F G 2 MZ (k; !), is de…ned as

P
F G(i; j) = l2C(k;N) F (i; l)G(l; j)

Under multiplication thus de…ned, FMZ (k; !) and T MZ (k; !) are unital rings
with identity I(i; j) = (i; j), the Kronecker delta.

The jth column of M 2 MZ (k; !) (or column j of M ), denoted M (j), is the


function C(k; N) ! Z de…ned by M (j)(i) = M (i; j). Given = ffj 2 ZC(k;N) :
j 2 C(k; N)g, the matrix with columns fj is de…ned to be M 2 MZ (k; !)
such that M (i; j) = fj (i) = M (j)(i), i, j 2 C(k; N). A column operation on
M 2 MZ (k; !) is an operation of the following type:

(a) Multiply a column of M by 1; for example, multiply the lth column


of M by 1 to obtain M 0 2 MZ (k; !) satisfying M 0 (i; j) = M (i; j) for j 6= l and
M 0 (i; l) = M (i; l) = M (l)(i).

(b) Interchange two columns of M ; for example, interchange columns l1


and l2 of M to obtain M 0 2 MZ (k; !) satisfying M 0 (i; j) = M (i; j) for j 6= l1
and l2 and M 0 (i; l1 ) = M (i; l2 ), M 0 (i; l2 ) = M (i; l1 ).

(c) Add an integral multiple of one column of M to another column of M ;


for example, for any integer n, add n times column l1 of M to column l2 of M
to obtain M 0 2 MZ (k; !) satisfying M 0 (i; j) = M (i; j) for j 6= l2 , and M 0 (i; l2 ) =
nM (i; l1 ) + M (i; l2 ) = nM (l1 )(i) + M (l2 )(i). Subtraction is accomplished with
n.

F 2 FMZ (k; !) is said to be invertible if there exists G 2 FMZ (k; !)


satisfying F G = I = GF . It clear that such a G is a unique two-sided inverse, so
that F 1 = G is well de…ned [C50, pp 21-25]. The set of invertible matrices in
FMZ (k; !) is denoted GLZ (k; !) and forms a group under matrix multiplication.

Proposition 2.1. If F 0 2 FMZ (k; !) is obtained from F 2 FMZ (k; !)


through one of the column operations described in (a), (b) or (c) above, then F 0
= F E where E is obtained by performing the same operation on I. E is both
row- and column-…nite and invertible.

3
Proof. Case (c) above; the others are far simpler. Suppose that for the
integer n, n times column l1 of F has been added to column l2 of F to obtain
F 0 2 FMZ (k; !) satisfying F 0 (i; j) = F (i; j) for j 6= l2 and F 0 (i; l2 ) = nF (i; l1 )
+ F (i; l2 ) = nF (l1 )(i) + F (l2 )(i). Let E(i; j) = I(i; j) for
Pj 6= l2 and E(i; l2 ) =
nI(i; l1 ) + I(i; l2 ) = n (i; l1 ) + (i; l2 ). Then F E(i; j) = l2C(k;N) F (i; l)E(l; j)
P
= ( l2C(k;N);l6=l2 F (i; l)E(l; j)) + F (i; l2 )E(l2 ; j). When j 6= l2 , this expression
becomes
0 1 0 1
X X
@ F (i; l) (l; j)A + F (i; l2 ) (l2 ; j) = @ F (i; l) (l; j)A + 0
l2C(k;N);l6=l2 l2C(k;N);l6=l2

= F (i; j) (j; j) = F (i; j) = F 0 (i; j):


When j = l2 , the
0 expression becomes 1
X
@ F (i; l)E(l; l2 )A + F (i; l2 )E(l2 ; l2 ) =
l2C(k;N);l6=l2
0 1
X
@ F (i; l)[n (l; l1 ) + (l; l2 )]A + F (i; l2 )[n (l2 ; l1 ) + (l2 ; l2 )] =
l2C(k;N);l6=l2

F (i; l1 )n (l1 ; l1 )+ F (i; l2 ) = nF (i; l1 ) + F (i; l2 ) = F 0 (i; l2 ):


It is clear that the inverse of E is the matrix which subtracts n times column
l1 of E from column l2 of E. E and its inverse obviously are row- and column-
…nite.

The corollary is immediate:

Corollary 2.2. If F 0 2 FMZ (k; !) is obtained from F 2 FMZ (k; !)


through a …nite sequence of operations described in (a), (b) or (c) above, then
F 0 = F E where E is obtained by performing the same sequence of operations
on I. E is both row- and column-…nite and invertible.

For F 2 FMZ (k; !) and f 2 ZC(k;N) , the matrix-vector product


P of F and f ,
denoted simply F f , is de…ned as F f : C(k; N) ! Z, F f (i) = l2C(k;N) F (i; l)f (l).
Each such F induces an endomorphism of ZC(k;N) . A product in ZC(k;N) ,
Q
m2C(k;N) hum i, is simply the image of such an endomorphism where the matrix
F is constructed with columns um , m 2 C(k; N).

Some of the results below can be proved through topological techniques.


Background material may be found in [BI01, BI96(1) & (2)] and the references
cited therein. For f 6= g 2 ZC(k;N) , the distance between them, d(f; g), is de…ned
to be 2 n , where n is the least positive integer for which there is an x 2 C(k; n)
with f (x) 6= g(x). d(f; f ) = 0, of course. It is easy to check that d is a metric
on ZC(k;N) , which comports with the product topology on ZC(k;N) , when Z is
discrete.

4
ZC(k;N) is a separable metric space having as a countable dense subset those
functions f which are 0 outside some C(k; nf ), nf 2 N+ . This countable
dense subset also is a free subgroup, which is denoted by (k); (1) =
P, the sequences of integers which are 0 after a while. (k) has a standard
basis, fej (i) = (i; j) : i, j 2C(k; N)g, which also the standard product basis for
ZC(k;N) . Finally, ZC(k;N) is a complete metric space in which Cauchy sequences
eventually become constant pointwise.

The important aspects of this topology are that all endomorphisms of ZC(k;N)
are continuous and are determined by their values on (k) and hence by their
values on the standard basis. As a result, with each element f in the endo-
morphism ring of ZC(k;N) , End(ZC(k;N) ), is associated a unique element Mf 2
MZ (k; !), given by Mf (i; j) = [f (ej )](i), i, j 2 C(k; N). Recall that each F 2
FMZ (k; !) acts on ZC(k;N) to produce an endomorphism of ZC(k;N) ,
P
(F g)(i) = l2C(k;N) F (i; l)g(l) for g 2 ZC(k;N) .

Because F is row-…nite, the sum reduces to


P
(F g)(i) = l2C(k;ni ) F (i; l)g(l) where ni 2 N+ .

The immediate goal is to prove the converse— that every endomorphism of


ZC(k;N) takes this form.

3. Product Bases And Endomorphisms

Analyses of the endomorphisms of ZC(k;N) demonstrate that its products


all stem from T MZ (k; !) and its product bases from GLZ (k; !). Of particular
interest is the fact that every endomorphism of P is nothing more than multi-
plication by a row-…nite matrix, and every product basis of P is produced by
an integrally invertible matrix.

Theorem 3.1. Every endomorphism of ZC(k;N) is induced by the action


of a k-dimensional row-…nite in…nite integral matrix.

Proof. Let f 2 End(ZC(k;N) ), and let Mf 2 MZ (k; !) be de…ned by Mf (i; j)


= [f (ej )](i), i, j 2 C(k; N). Suppose that Mf is not row-…nite; i.e., suppose there
are i and distinct ji 2 C(k; N) such that Mf (i; ji ) 6= 0, i 2 N. De…ne g 2 ZC(k;N)
by g(m) = Mf (i; ji ) if m = ji , i 2 N, and g(m) = 0 otherwise, for all m 2
C(k; N). Further, for n 2 N+ , de…ne gn (m) = g(m) for m 2 C(k; n) and gn (m)
= 0 for m 2 C(k; N)nC(k; n). By P design, gn convergesPto g, so that f (gn )
must converge to f (g). Now gn = l2C(k;n) gn (l)el = ji 2C(k;n) gn (ji )eji =
P P
ji 2C(k;n) Mf (i; ji )eji so that f (gn ) = ji 2C(k;n) Mf (i; ji )f (eji ) and
X
[f (gn )](i) = Mf (i; ji )[f (eji )](i) =
ji 2C(k;n)

5
X X
Mf (i; ji )Mf (i; ji ) = Mf (i; ji )2 :
ji 2C(k;n) ji 2C(k;n)
P
Now [f (g)](i) = limn!1 [f (gn )](i) = limn!1 ji 2C(k;n) Mf (i; ji )2 . Since
[f (g)](i) 2 Z, it is not possible that an in…nite number of the Mf (i; ji ) are
nonzero; i.e., Mf must be row-…nite.

To see that Mf acts on ZC(k;N) to produce f , it su¢ ces to


Pcheck agreement on
the standard basis of (k). For i, j 2 C(k; N), (Mf ej )(i) = l2C(k;N) Mf (i; l)ej (l)
P
= l2C(k;N) [f (el )](i) (j; l) = [f (ej )](i).

Corollary 3.2. (i) The endomorphism ring of ZC(k;N) , End(ZC(k;N) ), is


isomorphic to the ring of k-dimensional row-…nite in…nite integral matrices,
FMZ (k; !).
(ii) The automorphism group of ZC(k;N) , Aut(ZC(k;N) ), is isomorphic to
GLZ (k; !), the multiplicative group of invertible matrices in FMZ (k; !).

Proof. For f 2 End(ZC(k;N) ), f ! M f is a ring isomorphism with


FMZ (k; !), which maps Aut(ZC(k;N) ) isomorphically onto GLZ (k; !). The row-
…niteness of the elements of FMZ (k; !) ensures that all sums are …nite and that
matrix multiplication is associative.

Corollary 3.3. There is a one-to-one correspondence between the matrices


of GLZ (k; !) and the product bases of ZC(k;N) .

Proof. Let fum : m 2 C(k; N)g be a product basis for ZC(k;N) , and let
C(k;N)
A
P 2 Aut(Z ) be de…ned by [A(f )](m) = am for f 2 ZC(k;N) , when f =
m2C(k;N) am um , am 2 Z. The uniqueness of expression of the elements of
C(k;N)
Z in terms of the product basis guarantees that A is an automorphism.
By Corollary 3.2(ii), A corresponds to a unique element of GLZ (k; !). The
converse is clear.

C(k;N)
Q Corollary 3.4. If U is a subgroup of Z with product basis U =
m2C(k;N) hu m i and = fu m : m 2 C(k; N)g, then the matrix with columns
C(k;N)
um , M , is row-…nite and U = M Z .

Proof.
P The mapping : ZC(k;N) ! U ZC(k;N) de…ned for f 2 ZC(k;N) by
C(k;N)
f !
7 m2C(k;N) f (m)um is an endomorphism of Z , so according to The-
orem 3.1, it is induced by a row-…nite matrix M satisfyingP M (i; j) = [ (ej )](i)
for all i, j 2 C(k; N), and M ZC(k;N) = U . Now (ej ) = m2C(k;N) ej (m)um
= uj , so that [ (ej )](i) = uj (i) = M (i; j) for all i, j 2 C(k; N), demonstrating
that M = M .

6
4. All Products In ZC(k;N)
Are Generated By Lower-Triangular Matrices

Recall that when k = 1, C(k; N) reduces to N and ZC(k;N) reduces to ZN ,


the Baer-Specker group P. Further, FMZ (1; !) reduces to the familiar ring of
2-dimensional row-…nite in…nite integral matrices. With this understanding, a
product in P is just a product in the sense of Fuchs [F73, p 164].

Lemma 4.1. Every product in P is generated by a lower-triangular matrix,


the columns of which form a basis of the product.

Proof. Let F 2 FMZ (1; !). The proof proceeds by transforming F via
column operations into a lower-triangular matrix T satisfying T P = F P. Since
F = 0 is trivial, assume F 6= 0. Let y 2 F P, 0 6= y = (y(0); y(1); :::)t , where t
denotes transpose, and suppose y = F x for x = (x(0); x(1); :::)t 2 P.

If row 0 of F consists entirely of 0’s, simply set F1 = F , I1 = I, and x1 =


x and proceed to the next step. Otherwise, perform column operations on the
nonzero elements of row 0 of F to produce a new matrix F1 in which F1 (0; j) =
0 for all j > 0. This is tantamount to …nding the gcd of the elements in row 0,
columns j 0 of F . Perform the same column operations on I to produce I1 ,
so that F1 = F I1 with I1 an invertible row- and column-…nite matrix. Since F
= F1 I1 1 , y = F x = F1 (I1 1 x). Set x1 = (x1 (0); x1 (1); :::)t = I1 1 x so that y =
F1 x1 .

If F1 (1; j) = 0 for all j 1, simply set F2 = F1 , x2 = x1 , and I2 = I and


proceed. Otherwise perform column operations on the nonzero elements of row
1 in columns j 1 of F1 to produce a new matrix F2 in which F2 (1; j) = 0 for
all j > 1. As before, this is tantamount to …nding the gcd of the elements in
row 1, columns j 1 of F1 . F2 (i; 0) = F1 (i; 0) for all i; i.e., column 0 of F2 is
the same as column 0 of F1 .

Perform the same column operations on I to produce I2 , so that F2 =


F1 I2 with I2 an invertible row- and column-…nite matrix. Note that since no
operation was performed on column 0 of F1 , none was performed on column 0
of I. Moreover rows 0 of F1 and I remained unchanged (because of 0 entries),
so that I2 (0; 0) = 1 and the rest of the row 0 and column 0 entries of I2 are 0.
Thus the same is true of I2 1 . Since F = F2 I2 1 I1 1 , y = F x = F2 I2 1 (I1 1 x0 ) =
F2 (I2 1 x1 ). Set x2 = I2 1 x1 so that y = F2 x2 . Because row 0 of I2 1 is 1; 0; : : :,
it follows that x2 (0) = x1 (0).

Suppose that after n > 1 iterations, row-…nite matrices F1 , . . . , Fn and


invertible row- and column-…nite matrices I1 , . . . , In and x1 , . . . , xn 2 P have
been obtained, such that

7
(a) Fn (i; j) = 0 for all j > i, 0 i n 1; i.e., the matrices Fn are
becoming increasingly lower-triangular.
(b) Fn (j) = Fn 1 (j) for 0 j < n 1; i.e., after the nth iteration, rows
and columns < n do not change.
(c) Fn = Fn 1 In = F I1 In .
(d) In (i; j) = (i; j) = In 1 (i; j), 0 i < n 1, j 2 N; and 0 j < n 1,
i 2 N; i.e., the In ’s and their inverses are becoming increasingly diagonal with
1’s down the diagonal, tending toward the identity matrix.
(e) xn = In 1 xn 1 , y = Fn xn ; i.e., y is always in the image of Fn .
(f) xn (i) = xn 1 (i), 0 i < n 1; i.e., after the nth iteration, the row
elements < n of the xn ’s do not change.

If Fn (n; j) = 0 for all j n, simply set Fn+1 = Fn , In+1 = I, and xn+1 =


xn . Otherwise, perform column operations on the nonzero elements of the nth
row in columns j n of Fn to produce a new matrix Fn+1 in which Fn+1 (n; j)
= 0 for all j > n. Because no operations are performed on columns 0-(n-1),
Fn+1 (i; j) = Fn (i; j), i 2 N, j = 0, . . . , n 1; i.e., columns 0-(n-1) of Fn+1 are
the same as columns 0-(n-1) of Fn .

Perform the same column operations on I to produce In+1 , so that Fn+1 =


Fn In+1 with In+1 an invertible row- and column-…nite matrix. Note that since
no operation was performed on columns 0-(n-1) of Fn , none was performed
on columns 0-(n-1) of I. Moreover, rows 0-(n-1) of Fn were not changed (be-
1
cause of 0’s), and the same is true of I and In+1 . (i) As a result, In+1 (i; j)
1
= (i; j) = In+1 (i; j), i < n, j 2 N, and j < n, i 2 N. Since Fn+1 =
1
Fn In+1 = (F I1 In )In+1 , F = Fn+1 In+1 I1 1 , y = F x = Fn+1 In+11
I1 1 x
1 1
= Fn+1 (In+1 xn ). (ii) Set xn+1 = In+1 xn so that y = Fn+1 xn+1 . (i) and (ii)
imply that xn+1 (i) = xn (i), i = 0, . . . , n 1.

The Cauchy sequence xn converges to some x0 2 P because P is a complete


metric space. Since all endomorphisms of P are continuous, the sequence Fn (xj )
converges to Fn (x0 ) for each n. Because of (b) and (f), the convergence is
uniform. The Cauchy sequence of matrices Fn thus obtained converges to a
lower-triangular matrix T and T x0 = y so that T P F P. From (b) it follows
that T P F P and so T P = F P.

It may, of course, happen that some of the diagonal terms of T are 0 so that
the …rst nonzero entry in a column may be below the diagonal. Should that
occur, the …rst nonzero entry in the next column will be in a lower row. If T
has only …nitely many nonzero columns, the result is elementary. If all columns
of T are nonzero, it is clear that an in…nite linear combination of its columns
can be 0 only if all coe¢ cients are 0, so that the columns of T form a product
basis.

Remark 4.2. From the results in Section 3, a row-…nite matrix can be


classi…ed as epic or monic if the endomorphism which it induces is epic or monic,

8
respectively. (1) If F 2 FMZ (1; !) has a 0 row m, it cannot be epic because
em cannot be reached. If it has a 0 column n, it cannot be monic because en is in
the kernel. Here E = fei : i 2 Ng denote the usual basis of P; E is also the
standard product basis of P. (2) Similarly, if a row (column) of F is a multiple
of another row (column), then F cannot be epic (monic). (3) T 2 T MZ (1; !)
is invertible (2-sided) over the rationals if and only if each diagonal entry is
nonzero [C50, pp 19-20 ]. T is integrally invertible if and only if each diagonal
entry = 1.

Lemma 4.3. T 2 T MZ (1; !) is epic if and only if each diagonal entry is


1.

Proof. If each diagonal entry of T is 1, then T 2 GLZ (1; !) and so is epic.


Conversely let y = (y(0); y(1); :::)t be a nonzero element of P, where t denotes
transpose. To solve T x = T (x(0); x(1); :::)t = y, induct on n. Let Tn = (T [i; j]),
i, j = 0, 1, . . . , n. If n = 0 and T (0; 0) 6= 1, then T (0; 0)x(0) = y(0) is not
solvable in general. If T (0; 0) = 1, then x(0) is uniquely determined. Suppose
that the diagonals T (m; m) = 1 for m = 0, 1, . . . , n, and suppose further
that Tn (x(0); :::; x(n))t = (y(0); :::; y(n))t is uniquely solvable, (x(0); :::; x(n))t
Tn 0
= Tn 1 (y(0); :::; y(n))t . Express Tn+1 = so
T (n + 1; 0)::: T (n + 1; n + 1)
Tn (x(0); :::; x(n))t
that Tn+1 (x(0); :::; x(n); x(n + 1))t = Pj=n+1 , with some
j=0 T (n + 1; j)x(j)
abuse of notation for the sake of simplicity. Now by the induction hypothe-
sis, (x(0); :::; x(n))t = Tn 1 (y(0); :::; y(n))t uniquely. Thus the only solution to
Pj=n+1
j=0 T (n + 1; j)x(j) = y(n + 1) must come from T (n + 1; n + 1)x(n + 1) =
Pj=n
y(n + 1) – j=0 T (n + 1; j)x(j), which is impossible in general unless T (n +
1; n + 1) = 1.

Invertible lower-triangular in…nite integral matrices may therefore be char-


acterized as follows:

Theorem 4.4. For T 2 T MZ (1; !) the following are equivalent:


(1) T is epic.
(2) T (i; i) = 1 for all i 2 N.
(3) T 2 GLZ (1; !).

The following lemma is a generalization of Lemma 4.1.

Lemma 4.5. Every F 2 FMZ (k; !) can be triangularized in the sense


that there exists F0 2 FMZ (k; !) satisfying F0 [(i; i); (j; j)] = 0 for all i; j 2
C(k 1; N); i < j 2 N, and F0 ZC(k;N) = F ZC(k;N) .

Proof. De…ne Fij 2 FMZ (1; !) for i; j 2 C(k 1; N) and i; j 2 N by


Fij (i; j) = F [(i; i); (j; j)]. As in the proof of Lemma 4.1, there exists Tij 2

9
T MZ (1; !), obtained by performing column operations on the 2-dimensional
matrix Fij , which satis…es Tij P = Fij P. For i; j 2 C(k 1; N) and i; j 2 N,
de…ne F0 [(i; i); (j; j)] = Tij (i; j). Since each Tij is lower-triangular, it is clear
that F0 [(i; i); (j; j)] = 0 for all i < j.

To see that F0 is row-…nite, …x (i; i). Tij (i; j) = 0 for all j > i; i.e., Tij (i; j) 6=
0 for only …nitely many j, namely j = 0; : : : ; i. For each such j, there are only
…nitely many j where F [(i; i); (j; j)] 6= 0. Thus, since the entries of each Tij
were obtained from those of F , for each (i; i), only …nitely many Tij (i; j) 6= 0,
so F0 2 FMZ (k; !).
C(k;N)
Finally, to see that F0 ZP = F ZC(k;N) , let f 2 C(k;N)
PZ P . For i 2 C(k 1; N)
and i 2 N, (F f )(i; i) = F [(i; i); j]f (j) = F [(i; i); (j; j)]f (j; j)
j2C(k;N) j2C(k 1;N)j2N
P P
= Fij (i; j)yj (j), with yj 2 P de…ned by yj (j) = f (j; j) for j 2 C(k
j2C(k 1;N)j2N
1; N) and j 2 N. Since Tij P = Fij P, there exists P xj 2 P satisfying P Tij xj = Fij yj
or, for i 2 N, (Tij xj )(i) = (Fij yj )(i); i.e., Tij (i; j)xj (j) = Fij (i; j)yj (j),
P P j2N P Pj2N
so that (F f )(i; i) = Fij (i; j)yj (j) = Tij (i; j)xj (j) =
j2C(k 1;N)j2N j2C(k 1;N)j2N
P P
F0 [(i; i); (j; j)]xj (j) =
j2C(k 1;N)j2N
P P
F0 [(i; i); (j; j)]g(j; j) = (F0 g)(i; i) when g 2 ZC(k;N) is de…ned as
j2C(k 1;N)j2N
g(j; j) = xj (j) for j 2 C(k 1; N) and j 2 N.

In the Proof of Lemma 4.5, notation has been abused for the sake of sim-
plicity, but attention to each index set should clear up any confusion.

A principal result is the following

Theorem 4.6. Every product in ZC(k;N) is generated by a lower-triangular


matrix, the columns of which form a basis of the product.

Proof. By induction on k. Lemma 4.1 establishes the case k = 1, so the


result holds for some k. Let F 2 FMZ (k + 1; !) so that F ZC(k+1;N ) is a
product in ZC(k+1;N) . F = 0 is trivial so assume that F 6= 0. Per Lemma
4.5, triangularize F ; i.e., …nd F0 2 FMZ (k + 1; !) such that F0 [(i; i); (j; j)] =
0 for all i, j 2 C(k; N), i < j 2 N, and such that F0 ZC(k+1;N) = F ZC(k+1;N) .
When i = 0, by the induction hypothesis, there exists T00 2 T MZ (k; !) such
that T00 ZC(k;N) = F0 [(i; 0); (j; 0)]ZC(k;N) , where by F0 [(i; 0); (j; 0)] is meant the
matrix G0 2 FMZ (k; !), de…ned as G0 (i; j) = F0 [(i; 0); (j; 0)] and similarly
below; again notation is abused for the sake of simplicity. De…ne F1 2 FMZ (k+
1; !) as follows:
For all i; j 2 C(k; N) and i; j 2 N;
T00 (i; j) if i = 0 = j
F1 [(i; i); (j; j)] =
F0 [(i; i); (j; j)] otherwise

10
Observe that F1 [(i; 0); (j; 0)] = 0 unless i j, and that F1 ZC(k+1;N) = F0 ZC(k+1;N)
= F ZC(k+1;N) .

When i = 1, by the induction hypothesis, there exist T10 and T11 2 T MZ (k; !)
such that

T10 ZC(k;N) = F1 [(i; 1); (j; 0)]ZC(k;N) and T11 ZC(k;N) = F1 [(i; 1); (j; 1)]ZC(k;N) .

De…ne F2 2 FMZ (k + 1; !) by:

For all i; j 2 C(k; N) and i; j 2 N;


8
< T10 (i; j) if i = 1 and j = 0
F2 [(i; i); (j; j)] = T11 (i; j) if i = 1 and j = 1
:
F1 [(i; i); (j; j)] otherwise

Observe that 0 = F2 [(i; 0); (j; 0)] = F2 [(i; 1); (j; 0)] = F2 [(i; 1); (j; 1)] unless i j,
and that F2 ZC(k+1;N) = F1 ZC(k+1;N) = F0 ZC(k+1;N) = F ZC(k+1;N) .

Proceeding in this fashion, obtain Fn 2 FMZ (k+1; !) and Tij 2 T MZ (k; !)


with these properties: For i, j 2 C(k; N) and i, j 2 N,

(a) Tij ZC(k;N) = Fn 1 [(i; i); (j; j)]ZC(k;N) , n i j;


(b) Fn [(i; i); (j; j)] = 0 unless (i; i) (j; j) for i, j n; and
(c) Fn ZC(k+1;N) = F ZC(k+1;N) .

Then as in Lemma 4.1, limn!1 Fn = T uniformly and T ZC(k+1;N) = F ZC(k+1;N)


with T 2 T MZ (k + 1; !).

By use of the foregoing methodology, the following readily can be proved:

Proposition 4.7. Let T 2 T MZ (k; !).


(1) These are equivalent: (i) T is epic; (ii) T (i; i) = 1 for all i 2
C(k; N); (iii) T 2 GLZ (k; !).
(2) T is monic if and only if T (i; i) 6= 0 for all i 2 C(k; N).

11
5. In…nite Matrix Examples And Counterexamples

The next examples and counterexamples illustrate results from Section 4 and
further illustrate some of the pitfalls and limitations of working with in…nite
matrices. They are not new but continue to be discussed in various contexts.
[BM05, sec 2]; [CC-CS01, p 437]; [BI96(1), pp 189-191].

As before, the elements of P are viewed as column vectors and E = fej : j 2


Ng denotes the usual basis of P; E is also the standard product basis of
P. The same symbol E is used for the matrix with the ej as columns, i.e., the
identity matrix. The same liberty is taken with other bases and their matrices.

Example 5.1. E1 = fe0 + e1 + : : :+ ei : i 2 Ng also is a basis of ,


having matrix

0 1
1 1 1 1 :::
B0 1 1 1 ::: C
B C
B ::: C
E 1 = B0 0 1 1 C
B0 0 0 1 ::: C
@ A
.. .. .. .. ..
. . . . .

Because the matrix is not row-…nite, E1 cannot be a product basis of P.

Example 5.2. E2 = fe0 , ei –ei 1 : i 2 N+ g also is a basis of , having


matrix

0 1
1 1 0 0 :::
B0 1 1 0 ::: C
B C
B ::: C
E 2 = B0 0 1 1 C
B0 0 0 1 ::: C
@ A
.. .. .. .. ..
. . . . .

Note that E2 is not lower triangular.

Example 5.3. Let 1 2 P denote the column vector having all entries =
1 and let E3 = fe0 + e2 + : : :+ ei 1 –1 : i 2 N+ g. Note that E3 is not a basis
of ; indeed, no element is even contained in . The accompanying matrix is

12
0 1
0 0 0 0 :::
B 1 0 0 0 ::: C
B C
B 1 1 0 0 ::: C
E3 = B C
B 1 1 1 0 ::: C
@ A
.. .. .. .. ..
. . . . .

The following relationships hold:

E1 E2 = E = E2 E1 E 2 E3 = E

Observe that although both E2 and E3 are row-…nite and even pure independent,
E2 is not monic (1 is in the kernel), and E3 is not epic (e0 is not attained). In
particular, (E1 E2 )1 = E1 = 1, but E1 (E2 1) = E1 (0) = 0, so that multiplication
is not associative. Despite the fact that E2 P = P, E2 nevertheless cannot serve
as a product basis for P. The problem, of course, is that E2 is not invertible
in FMZ (k; !). Restriction to multiplicatively associative matrix rings helps
eliminate much of this in…nite matrix pathology [C50, pp 19-22].

In the proof of Lemma 4.1, although the columns of the resulting lower-
triangular matrix T obviously are independent, T nevertheless may be singular,
as illustrated by E3 above. Moreover, it may not be the case that that the
products I1 In in that proof will converge to a row-…nite matrix, as further
illustrated by E2 . Even though E2 itself is not monic, the resulting lower-
triangular matrix E, which results from column operations on E2 , is monic.

6. Fuchs’Lemma 95.1 Redux

Products in P were introduced in [F73] by beginning with a countable subset


fxnP: n 2 Ng P, with no condition other than that all sums of the form x
1
= n=0 sn xn be well de…ned. This, of course, means that the in…nite matrix
with columns xn must be row-…nite. Lemma 95.1 in [F73] then was stated as
follows:

Let X be a product in P. Q There are elements aQ


n 2 P and integers
1 1
kn (n 2 N) such that P = n=0 han i and X = n=0 hkn an i where
( ) kn j kn+i if kn 6= 0 for all n, i 0.

The goal was to establish a product analog of the well known result for stacked
bases of free groups of …nite rank [F70, Lemma 15.4]. However, a counterexam-
ple has been produced [G81].

Perhaps another approach is to utilize the weaker de…nition of stacked bases


found in [CG70], which does not require divisibility as in ( ). From Lemma

13
4.1, there would be no loss of generality in assuming that the xn 2 P form a
lower-triangular matrix.
Q1
PLemma 695.1. Let n=0 hxn i be a product of in…nite rank in P, let S
1
= n=0 hxn i , and let S be the pure
Q1subgroup which the xn generate. Then
there existQa product basisQof P = n=0 han i and positive integers kn (n 2 N)
1 1
such that n=0 hkn an i = n=0 hxn i if and only if there exist stacked bases of
S and S such that the matrix F formed by the basisQof S is invertible and
1
the matrix F formed by the basis of S satis…es F P = n=0 hxn i.

Proof. If fan : n 2 Ng is a product basis of P, then the Q1an ’s are pure


independent. If fkn : n 2 N)g are positive integers, then n=0 hkn an i is a
product in P and if S = 1 n=0 hkn an i, then S =
1
n=0 han i so that fkn an :
n 2 Ng and fan : n 2 Ng are stacked bases for the respective subgroups of P.
The matrix formed by the an ’s is in GLZ (1; !) by Corollary
Q1 3.4; the matrix F
formed by the xn =Qkn an certainly satis…es F P = n=0 hxn i.
1
Conversely, let n=0 hxn i be a product in P, let S be the subgroup of P
which the xn ’s generate, and let S be the pure subgroup which they generate.
Suppose there are stacked bases fkn an : kn , n 2 Ng of S and fan : n 2 Ng
of S ; because S =S is torsion, all kn ’s must be positive. Suppose further that
the matrix
Q1 formed by the an ’s is invertible; then the an ’s are a product basis of
P and n=0 hkn an i is Q a product in P. Finally, if the matrix F formed by the
1
kn an ’s satis…es F P = n=0 hxn i, the proof is concluded.
Q1
Remark 6.2. (1) If n=0 hxn i is a product in P and fan : n 2 Ng is
a basis of the subgroup generated by the xn ’s, the matrix formed by the an ’s
may not be row-…nite. See examples E and E1 in Section 5 above. (2) In
Section 5, although all positive multiples of the basis elements in example E2
will yield subgroups of P with stacked bases and even though E2 P = P, E2
cannot serve as a product basis for P. (3) As previously noted, the lower-
triangular matrix T , obtained from the column operations in Lemma 4.1, may
be singular. However, Nunke [BI96, p 199] has shown that every endomorphic
image of P is either free of …nite rank or isomorphic to P itself. In the latter
case, there then exists monic F 2 FMZ (k; !) satisfying F P = T P. Column
operations will, of course, reveal matrix rank, …nite or in…nite.

All products in P may therefore be characterized as follows:

Theorem 6.3. Every product in P is generated by a lower-triangular


in…nite integral matrix, the rank of which determines whether the product is
free of …nite rank or isomorphic to P.

7. Obiter Dicta

14
Although the methods employed and results ontained in this paper and [CS07
& 08] could be extended to more complex structures, e.g., to …nite and countably
in…nite sequences of polynomials in a countably in…nite number of variables,
and to some larger cardinals/ordinals, e.g., along the lines of the earlier works
of Zeeman [F73, Corollary 94.6] (see also [F70, Theorem 43.2]) and the more
recent ones of Eda [E83(1) & E83(2)], the author does not envision that such
extensions will render results signi…cantly di¤erent from those already attained.

References

[B37] Baer, R., Abelian groups without elements of …nite order, Duke Math. J.
3 (1937) 68-122

[BI96(1)] Blass, A. & Irwin, J., On the group of eventually divisible integer
sequences. In: Abelian groups and modules, Arnold, D. M. & Rangaswamy, K.
M., eds., Lecture Notes in Pure and Applied Mathematics, Marcel Dekker, Inc.,
182 (1996) 181-192

[BI96(2)] Blass, A. & Irwin, J., Baer meets Baire: applications of category
arguments and descriptive set theory to Z@0 . In: Abelian groups and modules,
Arnold, D. M. & Rangaswamy, K. M., eds., Lecture Notes in Pure and Applied
Mathematics, Marcel Dekker, Inc., 182 (1996) 193-202

[BI01] Blass, A. & Irwin, J., Free subgroups of the Baer-Specker group, Comm.
Algebra 29(12) (2001) 5769-5794

[BM05] Bartosiewicz, Z. & Mozyrska, D., Observability of row–…nite countable


systems of linear di¤erential equations, http://katmet.pb.bialystok.pl/mat.barz/
Infsys4.pdf (2005)

[C50] Cooke, R. G., In…nite Matrices and Sequence Spaces, McMillan, London
(1950)

[CC-CS01] Camillo, V.; Costa-Cano, F. J. & Simon, J. J., Relating properties


of a ring and its ring of row and column …nite matrices, J. Algebra 244 (2001)
435-449

[CG70] Cohen, J. M. & Gluck, H., Stacked bases for modules over principal
ideal domains, J. Algebra 14 (1970) 493-505

[CS07] Cornelius, E. F., Jr. & Schultz, P., Polynomial points, J. Integer Se-
quences 10 (2007) Article 07.3.6

15
[CS08] Cornelius, E. F., Jr. & Schultz, P., Multinomial points, Houston J.
Math. 34(3) (2008) 661-676

[E83(1)] Eda, K., On a Boolean power of a torsion free abelian group, J. Algebra
82(1) (1983) 84-93

[E83(2)] Eda, K., Almost-slender groups and Fuchs-44 groups, Comment. Math.
Uni. St. Paui., 32(2) (1983) 131-135

[F70] Fuchs, L., I In…nite Abelian Groups, Academic Press, New York & London
(1970)

[F73] Fuchs, L., II In…nite Abelian Groups, Academic Press, New York &
London (1973)

[G81] Goldsmith, B., A note on products of in…nite cyclic groups, Rend. Semin.
Mat. Univ. Padova, 64 (1981) 243-246

[S50] Specker, E., Additive gruppen von folgen ganzer Zahlen, Portugaliae
Math. 9 (1950) 131-140

16