A Polynomial Ring Sampler

J. W. Brewer
Department of Mathematical Sciences, Florida Atlantic University, 777 Glades Road, Boca Raton,
FL 33431
BrewerQf au.edu

1. Introduction
In this retrospective, we survey some results of Robert Gilmer and others concerning
polynomial rings in a single indeterminate over a commutative ring. It is with
respect and admiration that I dedicate the paper to Robert.

2. ^-automorphisms of R[X]
Let R be a commutative ring and consider the polynomial ring R[X}. An R-
algebra endomorphism ip of R[X] is completely determined by its action on the
indeterminate X. That is, ii(p(X) /, then tp(ro + r\X + + rtXi) = r 0 + n / +
+ ftf*- It follows that the image of <p is the subring R[f] of .RptT]. In particular,
if is surjective if and only if R[f] = R[X}.
Notice that R[f] = R[f a] for any element a R and that, if u is a unit of
R, then R[f] = R[uf]. These observations will be useful in what follows.
In the paper [4], Robert considered the following question.
Let R be a ring and let / = ao + a\X + + amXm be a polynomial
in R[X]. If (ff denotes the .R-endomorphism of R[X] determined by
sending X to /, what are necessary and sufficient conditions on the
o's in order that the endomorphism ipf be an automorphism?
Robert settled this question in a most satisfactory manner. Let's see what he
did.
We begin by proving the following well known result, the proof of which we
shall use in the proposition below.
Lemma 1. Let T be a Noetherian ring'with a : T > T a ring endomorphism. If a
is a surjection, then a is an injection.

55
56 J. W. Brewer

Proof. Since a is surjective, cr(T) = T and it follows by induction that crl(T) = T

for i > 1. Now consider the sequence
ker a C ker a2 C
of ideals of T. Since T is Noetherian, the sequence stabilizes, say ker an+1 = ker a".
Let x kercr. Since an(T) = T, there exists an element y T such that <Jn(y) = x.
Then an+1(y) = <r(z) = 0 and so y e ker<7 +1 = kera". Hence, 0 = u n (y) = x and
ker cr = 0. D
We next show that <p/ is a surjection if and only if it is an automorphism. The
proof is adapted from [1], which was published four years after Gilmer's paper.
Proposition 2. With the notation as above, iff is a surjection if and only if it is
an automorphism.
Proof. If iff is surjective, then R[X] = R[f] and there exist elements bo,bi, . . . ,bn
R such that
X = b0 + b l f + - - - + bnfn.
Let -Ro be the prime subring of R and consider the subring

of R.. Now, -RipCj = Ri[f] and so the restriction, ip'^, of iff to .Ri[X] is a surjective
endomorphism of the Noetherian ring I?i[X]. By Lemma 1 above, tp'f is an R\-
automorphism, say with inverse (f'g. We can extend ip'g to an .R-endomorphism, say
<pg, of R[X]. But then (<pg o ipf)(X] = 9 g ( f ) = X and (Vj o ifg)(f] = iff(X] = f.
Thus, iff is injective as claimed. D
So, to determine1 necessary and sufficient conditions on the coefficients of / in
order that iff be an automorphism, it suffices to determine conditions in order that
iff be a surjection. Thus, we seek to know when R[X] = R[f}- As before, write
/ = o0 + aiX H ----- h amXm. Now, R[X] = R[f] if and only if there exist elements
bo, &j, . . . , bn R, bn / 0, such that
n
X = b0 + bif + + bn
We first treat the case when R := D is an integral domain. As noted earlier, we
may assume that 60 0. After expanding the equality above, we equate coefficients.
The coefficient of Xmn on the left hand side of the equality is 0 and, on the right
hand side, it is bno^- Now, n > 1 and, if m > 2, then mn > 2 and hence am = 0.
Thus, m = I , from which it follows that n = 1 as well. But then, bi(a0 + a\X) X
and ai is a unit of D. Consequently, in the case of an integral domain, with
/ = CQ + a\X + + amXm, iff is an automorphism if (and, as noted earlier, only
if) a\ is a unit and QJ = 0 for i > 2. That is, / = a0 + a\X with ai a unit.
This can easily be extended to give necessary conditions on / in the case of
general -R.
If P is an arbitrary prime ideal of 'R, the map p : R[X] > (R/P)[X\ given
by reducing coefficients modulo F is a surjective homomorphism with kernel P[-X"],
the ideal of R[X] consisting of those polynomials having all their coefficients in P.
A Polynomial Ring Sampler 57

If R[X] = R[f], then (R/P)[X] = (R/P)[f], where "/" denotes the image of / in
(R/P)[X\. By the integral domain case, a\ is a unit of R/P and 0^ = 0 for i > 1.
Since P was arbitrary, ai is in no prime ideal of R and a,, for i > 2, is in every
prime ideal of R. Thus, ai is a unit of R and a^, for i > 2, is nilpotent.
We can now give Gilmer's main result, Theorem 3 of [4]. The proof is essentially
verbatim from the original.
Theorem 3. Let / = ao + a\X + h amXm e .HpsT]. The ring R-endomorphism
(ff : R[X] > R[X] determined by ipf(X) = f is an automorphism if and only if a\
is a unit and a^ is nilpotent for i > 2.

Proof. We have proved the necessity of the conditions. It remains to prove the
sufficiency. By Proposition 2, we have only to prove that (ff is a surjection. We use
induction on k, the order of nilpotence of the ideal cz(f) (2i a 3i i am)- Fr
any / for which k 1, each a* = 0 for i > 1 and ipf is surjective. We assume the
truth of the theorem for all elements / of R[X] of the desired form for which the
order k of nilpotence of the associated ideal c 2 (/) is less than r, where r > 2, and
we consider an element h = h0 + hiX H ---- + hvXv of R[x], where hi is a unit of
R and the order of nilpotence of c^Cn) = (h%, . . . , hv) is r. Since
R[h] = R[h!(h - ho)},
it suffices to consider the case in which hi = 1 and h0 = 0. We then consider,
Ui

for i > 2, hih1. Now, h1 has the form X1 + ^3 u

ij^ > where each u^ G C2(h).
J=i+l
n
i
Hence, hih1 h^X1 + ]P v^X^ , where each Vij & [c2(h)}2. Therefore, if g =

h /i2/i2 h3h3 --- hvhv, then g has the form X + g^X^ H ----- h gmXm, where
each 9i e [c2(/i)]2. It follows that [c2(5)]r-1 C [c2(h')}^r-1\ and

because 2(r 1) > r. Consequently, the order of nilpotence of c 2 (<?) is less than r.
The induction hypothesis applies to g showing that R[X} R[g}. But, g e R[h},
so that R[X} = R[h], and the proof is complete. D

In the spirit of the times, Robert devoted several additional pages to the con-
sideration of similar questions in case the "ring" R failed to contain an identity
element. We shall omit those considerations.
We conclude this section with a quote from the final paragraph of Robert's
paper [4]. It gives an idea of the culture of those same times.
"When we originally obtained Theorem 3, it seemed most likely to us
that this result was known. But a rather thorough search of the lit-
erature did not uncover the result. We then contacted several knowl-
edgeable persons in the area of commutative-ring theory, one of whom
had heard of the result, but knew no reference to it in the literature.
Finally, we submitted the results as an abstract to the Notices of the
58 J- W. Brewer

American Mathematical Society, and after receiving no correspon-

dence concerning the results for several months, we concluded at last
that if Theorem 3 is known it certainly is not well known, and that
such a basic result ought to appear in some widely read journal."

3. Polynomial rings over von Neumann regular rings

We begin by recalling the definition of a von Neumann regular ring. As there are
different versions, we list several of them in the following result.
Proposition 4. Let R be a commutative ring The following are equivalent:
1. If r R, there exists an idempotent element e S R and a unit u e R such
that r = ue.
2. If r & R, there exists an element s R such that r = sr2.
3. Ifr e R, then rR = r2R.
4. The Krull dimension of R is zero and R is reduced, that is, R has no non-
zero nilpotent elements.
5. For each prime ideal M of R., RM is " field.
Proof. (1) => (2): Let r e R and write r = ue, where u is a unit and e is an
idempotent. Then r 2 = u2e2 = u2e. Thus, u~lr2 = ue = r.
(2) => (1): Let r R. Then there exists an element 5 R such that r sr2.
Now,
(sr)2 = s2r2 = s(sr2) = sr.
Thus, sr is idempotent. Set e := sr and let u = 1 e + r. An easy computation
shows that u(l - e + se) = 1 and hence that u is a unit. Finally,
ue = (I e + r)e = e e + re = re = rsr = sr2 = r
(3) & (2): This is obvious.
(1) => (4): Let x e R and write x = ue, where u is a unit and e is an idempotent.
Then x2 = u2e2 = u2e. Thus, xn = une for n > 1. If xn = 0, then e = 0 and so
x = 0. It follows that 0 is the only nilpotent element of R.
Let P be a prime ideal of R. If a; e P, then there exists an element y S R such
that x = yx2 and so, x(l yx) = 0. Thus, in RP, I yx is a unit and so, x = 0.
Since PRp = (0) and PRp is the unique maximal ideal of Rp, it follows that P is
a maximal ideal of R. Hence, the Krull dimension of R is zero,
(4) => (5): Recall that if a ring T is reduced, then so is any localization T$.
(If (j) n = 0 in Ts, there exists an element s S such that san 0. But then
snan = (sa)n = 0 and so sa = 0. Thus, ( f ) = 0.) In our case, Rp is reduced
for each prime ideal P of R. But, R is zero-dimensional and so PRp is both the
nilradical and the unique maximal ideal of -Rp. Therefore, PRp = (0) and Rp is a
field.
(5) => (3): We have to show that for each element x & R, xR = x2R. Well,
x R C xR and for P a prime ideal of R, x P if and only if x2 P. If x ^ P,
2

then x2Rp = Rp = xRp. If x e P, then xRP = (0) = x2Rp. In either case,

xRp = x2Rp at each prime ideal of R and so xR = x2R. D
A Polynomial Ring Sampler 59

If one, and hence all, of these conditions is satisfied, then R is called a von
Neumann regular ring.
Remark 5. Since it follows easily from condition (5) that each module over a von
Neumann regular ring must be flat, these are also the "absolutely flat rings" of
Bourbaki.
Remark 6. It follows from (2) that each principal ideal of R is generated by an
idempotent element. In fact, any finitely generated ideal is generated by an idem-
potent. (In the case of two idempotent elements e\and e% of R, e\R + e-^R =
(BI + 62 e^e^R. The general case follows by induction.)
The paper [5] is a note following up the paper [9] by Paul McCarthy. The
principal result of Paul's was:
If R is a von Neumann regular ring, then the ring of polynomials
R[X] in a single indeterminate X is a semihereditary ring.
This oft-quoted result is now referred to as "McCarthy's Theorem." Robert
notes that a useful lemma for McCarthy was: If f ( X ) = OQ + aiX + + anXn
is in R[X] and if e is an idempotent generator of the ideal a,iR for each i, then
the annihilator of f ( X ) is the principal ideal of R[X] generated by (1 - e 0 )(l
e i ) . . . ( l -en).
The purpose of Robert's paper was to extend this result to the case of finitely
many polynomials in the ring .R[{Xx}], where {X\} is a family of indeterminates
over the von Neumann regular ring R. With hindsight, this was a straightforward
effort. In fact, the essence of McCarthy's lemma is simply the fact noted above
that a finitely generated ideal in a von Neumann regular ring is generated by an
idempotent.
McCarthy's Theorem itself is an interesting result and here is a short proof
using a lovely result of Endo [3].
Theorem 7 (McCarthy). // R is a von Neumann regular ring, then R[X] is
semihereditary.
Proof. Endo's Theorem says the following:
A commutative ring S is semihereditary if and only if the total quo-
tient ring of S is von Neumann regular and SM is a valuation domain
for each maximal ideal M of S.
We shall show that the hypotheses of Endo's theorem are satisfied by R[X}.
Since the nilradical TV of R is zero and since the nilradical of R[X] = N[X],
R[X] has no nonzero nilpotent elements. It follows that the same is true for
T(R[X]), the total quotient ring of R[X}.
If / is a regular element of R[X], let c(f) denote the content ideal of /, that
is, the ideal of R[X] generated by the coefficients of /. Since R is a von Neumann
regular ring, c(f) = eR, for some nonzero idempotent element e e R. If 1 ^ e,
then (1 - e) ^ 0 and (1 - e) / = 0. Thus, c(f) = R and it follows that a
polynomial / is a regular element of R[X] if and only if / has unit content. Hence,
T(R[X}) = (R[X])S = R(X), where S is the set of polynomials in R[X] having unit
60 J. W. Brewer

content. It is well known [6] Proposition 33.1 that the map M > MR(X) is always
a homeomorphism from the maximal spectrum of R to the maximal spectrum of
R(X). Consequently, T(R[X\) is zero-dimensional. It follows that T(R[X]) is a
von Neumann regular ring.
If OT is a maximal ideal of R[X], then (R[X])m is a localization of RfmnniX]
at a nonzero prime ideal. Since -R is a von Neumann regular ring, RmnR is a field
and so (J?[J\r])jrn is a valuation domain.
Thus, the hypotheses of Endo's theorem are satisfied and R[X] is a semihered-
itary ring. D

Gilmer, in collaboration with T. Parker [7], made an additional contribution

to the topic by proving the following nice result. We give their proof virtually
unchanged.
Theorem 8. If R is a von Neumann regular ring, then R[X] is a Bezout ring.

Proof. We have to show that every finitely generated ideal of R[X] is principal
and for that, it suffices to prove that if / and g are nonzero elements of -R[X],
then (/, g) is principal. We assume that deg/, the degree of /, is less than or
equal to the degree of g, and we use induction on deg/. If deg/ = 0, that is,
if / R, then we can assume that / is an idempotent of R. Then (/, g) = (/i),
where h = f + (1 f ) g , because / = fh and g = (1 / f g ) h . We assume
that (/, g) is principal if deg/ < n, and we consider the case where deg/ = n + 1.
Let t be the leading coefficient of / and let e be an idempotent generator of the
ideal tR of R. Then R = eR < (1 - e)R, R[X] = eR[X] (1 - e)R[X], and
(/, 9) = (e/, eg) ((1 - e)f, (I - e)g)). The ideal ((1 - e)f, (1 - e)g)) is principal by
the induction hypothesis. And, as an element of e.R[X], the leading coefficient et
of ef is a unit, since et eR = eR. Hence, eg = q ef + r for some q, r S eJR[.X"] with
r = 0 or degr < dege/. It follows that (ef,eg) = ( e f , r ) , and again the induction
hypothesis implies that (ef, eg) is principal. Since each of the summands of the
ideal (/, g) in the decomposition

is principal, it follows that (/, g) is also principal. D

Motivated by altogether different considerations, J. Brewer, D. Katz, and

W. Ullery [2] strengthened this result by proving that R[X] is an elementary di-
visor ring. Recall from [8] that a ring R is called an elementary divisor ring if
each matrix A over R admits a diagonal reduction that is, there exist invertible
matrices P and Q such that PAQ is diagonal.
Proposition 9. If R is a von Neumann regular ring, then R[X] is an elementary
divisor ring.

Proof. Let A be a matrix over f?[JT]. Set / = {a R \ A admits a diagonal

reduction over _R a [X]}, where Ra denotes the localization of R at the element a,
and let /* be the ideal of R generated by /. For each maximal ideal M of R, RM[X]
A Polynomial Ring Sampler 61

is a principal ideal domain and so A admits a diagonal reduction over RM [X]. It

follows easily from this that /* = R so there exist elements o,j I and r^ e R with
riai H \-rnan = 1.
Since .R is a von Neumann regular ring, we may assume that each a^ is idempo-
tent. Replacing ai, a 2 , . . . , an by <H, (1 - oi)a 2 ,.. , (1 - ai)(l - a 2 ) . . . (1 -a n _!)a n
respectively, we may assume that the a;'s are pairwise orthogonal idempotents.
Thus,
# = .Rai 8 Ran S* Rai Ran
via a "natural" isomorphism. It follows that A admits a diagonal reduction over
R[X]. D

There is a nice converse to these results whose proof can also be found in the
paper of Gilmer and Parker. For ease of readability, we record the following well
known fact.
Remark 10. If S is a ring and a.b.c S, with c not a zero-divisor, then (a, b] = (c)
implies that (a,b)2 = (a 2 ,b 2 ). (In fact, the result remains true if the ideal is only
assumed to be invertible, as can be seen by localizing.}

Proof. Since (a,6) 2 = (a 2 ,a6,b 2 ), (a,6) 2 = (a2,b2} if and only if ab e (a 2 ,6 2 ). We

have that a ac, 6 = /3c, and c = xa + yb for some a/3, x, y 5. Then
i1 a x\-b-
, ?, y-
c c
and so, ab = ab 1 = a (6 f) + b (a *) = a2(3x + b2ay.
2 2
D

Proposition 11. If R[X] is a Bezout ring, then R is a von Neumann regular ring.

Proof. Let r e R. The ideal (r, X} of R[X] is principal, generated by a non-zero-

divisor / of R[X]. Then (r,X)2 = (r2,X2) = (/ 2 ) and so, rX = or2 + /3X2 for
a, /3 -R[J^]. Since the degree of /3X2 is greater than or equal to 2, the coefficient
of X on the right hand side is r2a, where a is the coefficient of X in the polynomial
a. Thus, r = r2a and R is von Neumann regular. D

Remark 12. The proof actually shows that if R is a ring and if r R, then there
exists an element a & R such that r = r2a if and only if the ideal (r, X) of R[X] is
such that (r, X)2 = (r2,X2).
Before stating and proving the main result of this section, we need to make two
additional remarks.
Remark 13. If S is a commutative ring of Krull dimension n, and if X is an
indeterminate, then n + 1 < dimS'piT] < In + 1. Hence, if R is a von Neumann
regular ring, dimR[X] = 1. And, as noted above, the nilradical extends to the
nilradical of R[X]. Thus, R is a von Neumann regular ring if (and only if), R[X]
is one-dimensional and reduced.
62 J. W. Brewer

Remark 14. The usual proof that a projective ideal in an integral domain is in-
vertible can be used to prove that, over any ring, a projective ideal which contains
a non-zero-divisor must be invertible. Thus, projective ideals containing non-zero-
divisors in semihereditary rings must be invertible.
Putting all this together, we have the following lovely result.
Theorem 15. Let R be a commutative ring with X an indeterminate. The follow-
ing are equivalent:
1. R is a von Neumann regular ring.
2. R[X] is one-dimensional and reduced.
3. R[X] is an elementary divisor ring.
4. R[X] is a Bezout ring.
5. R[X] is a semihereditary ring.
6. Each ideal I of R[X] generated by two elements is invertible.
7. //*P is a prime ideal ofR[X], then (R[X])<p is a valuation ring (zero-divisors
allowed).
Contrasting this result with the following one on fields adds credence to an
observation Fred Richman made to me: "von Neumann regular rings are fields
with idempotents."
Theorem 16. Let D be an integral domain with X an indeterminate. The following
are equivalent:
1. D is a field.
2. D[X] is one-dimensional.
3. -D[X] is a principal ideal domain.
4. D[X] is a Bezout domain.
5. D[X] is a Prufer domain.

References
[1] J. Brewer and E. Rutter, Isomorphic polynomial rings, Arch. Math., 23 (1972), 484-488.
[2] J. Brewer, D. Katz, and W. Ullery, Pole assignability in polynomial rings, power series rings,
and Prufer domains, J. Algebra, 106 (1987), 265-286.
[3] S. Endo, On semi-hereditary rings, J. Math. Soc. Japan, 13 (1961), 109-119.
[4] R. Gilmer, R-automorphisms of R[X], Proc. London Math. Soc., 18 (1968), 328-336.
[5] R. Gilmer, Polynomial rings over a commutative von Neumann regular ring, Proc. Amer.
Math. Soc., 49 (1975), 294-296.
[6] R. Gilmer, Multiplicative Ideal Theory, Marcel Dekker, New York, 1972.
[7] R. Gilmer and T. Parker, Semigroup rings as Prufer rings, Duke Math. Journal, 21 (1974),
65-86.
[8] I. Kaplansky, Elementary divisors and modules, Trans. Amer. Math. Soc., 66 (1949), 464-491.
[9] P. J. McCarthy, The ring of polynomials over a von Neumann regular ring, Proc. Amer.Math.
Soc., 39 (1973), 253-254.