Sei sulla pagina 1di 19

Bhargavas Early Work: The Genesis of P-Orderings

Author(s): Paul-Jean Cahen, Jean-Luc Chabert and Kiran S. Kedlaya


Source: The American Mathematical Monthly, Vol. 124, No. 9 (November 2017), pp. 773-790
Published by: Mathematical Association of America
Stable URL: http://www.jstor.org/stable/10.4169/amer.math.monthly.124.9.773
Accessed: 17-11-2017 02:19 UTC

JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide
range of content in a trusted digital archive. We use information technology and tools to increase productivity and
facilitate new forms of scholarship. For more information about JSTOR, please contact support@jstor.org.

Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at
http://about.jstor.org/terms

Mathematical Association of America is collaborating with JSTOR to digitize, preserve and


extend access to The American Mathematical Monthly

This content downloaded from 128.135.98.165 on Fri, 17 Nov 2017 02:19:44 UTC
All use subject to http://about.jstor.org/terms
Bhargavas Early Work:
The Genesis of P-Orderings
Paul-Jean Cahen, Jean-Luc Chabert, and Kiran S. Kedlaya

Abstract. As an undergraduate student, Manjul Bhargava gave a full answer to a question on


polynomial functions on the integers. He immediately generalized this study to finite principal
ideal rings, thanks to the amazingly simple notion of P-ordering. This tool, together with a
beautiful generalization of factorials, allowed him to generalize many classical theorems. It
turned out to be extremely useful for the study of integer-valued polynomials on subsets.

INTRODUCTION. Born in 1974 in Hamilton, Ontario, Canada, Manjul Bhargava


received the Fields Medal in 2014. In fact, he had already been distinguished as an
undergraduate student. In 1995, he participated in the ten-week summer Research
Experience for Undergraduates (REU) program at the University of Minnesota Duluth
under the supervision of Joseph Gallian. He was handed a number-theoretic question
raised by Chen [21] in Discrete Mathematics. Not only did he give a full answer to
it, soon published in the same journal [4] (in fact, received by the editor as early as
August 21, 1995), but he also immediately undertook a generalization. To do so, he
developed the notion of P-ordering, giving rise to a beautiful generalization of fac-
torials. He soon found this tool applied to other situations: In a letter to Joe Gallian
(March 22, 1996), he wrote

My thesis is on Duluth-related stuff! . . . It turned out that the P-orderings could be used not
only to count polynomial mappings on finite principal ideal rings (as I did in Duluth), but also
to construct what are called regular bases for Dedekind domains, as well as to generalize the
notions of factorial and binomial coefficient to rings other than the integers. Amazing how far
one little combinatorial idea can go (in algebra!).

This undergraduate senior thesis,1 under the supervision of Persi Diaconis and Barry
Mazur at Harvard University, was presented on April 1, 1996 [3]. It was published,
almost as such, in Crelles journal (1997) [5] and soon led to another article in the Jour-
nal of Number Theory (1998) [6]. He was honored with the Hoopes Prize for Excel-
lence in Scholarly Work and Research by Harvard University in 1996 and next, on a
national basis, with the AMS-MAA-SIAM Frank and Brennie Morgan Prize for Out-
standing Undergraduate Research in Mathematics (1996).2 Let us quote Barry Mazur
from two letters of recommendation, written for each of these prizes:

It is a rare thing for a senior thesis in Mathematics to have this level of originality, but (as
good as it is), I feel that it is only the bearest [sic] indication of the level of mathematical
contributions that Bhargava will make in the future.

http://dx.doi.org/10.4169/amer.math.monthly.124.9.773
MSC: Primary 13F20, Secondary 13F10; 13M10
1 His Ph.D. thesis, under the supervision of Andrew Wiles, Higher Composition Laws (Princeton 2001) is

on a totally different topic.


2 Bhargava was also recognized for another work of his REU program: a full answer to a tiling problem for

which there seems to be only a manuscript version [2].

November 2017] BHARGAVAS EARLY WORK 773

This content downloaded from 128.135.98.165 on Fri, 17 Nov 2017 02:19:44 UTC
All use subject to http://about.jstor.org/terms
Our goal in this paper is to focus on Manjul Bhargavas undergraduate work and
hence on the genesis of P-orderings and generalized factorials. It is not to give but
another exposition of these notions. There are already plenty, and the best one, in fact,
is an expository paper by Bhargava himself, The factorial function and generaliza-
tions, published in THIS M ONTHLY [7] (2000), that gained him The Mathematical
Association of Americas Merten M. Hasse Prize (2003).
We devote the first section to Bhargavas REU work. It contains a brief account of
his successful answer to Chens question. We also describe Chens results on polyno-
mial functions from {0, 1, . . . , n 1} to Z/mZ [21], especially to make understand-
able Bhargavas generalization to polynomial functions f : X R, with R a finite
principal ideal ring and X a subset of R. We thus underline Bhargavas idea of order-
ing X in a way that would generalize the natural ordering of {0, 1, . . . , n 1}. This
is how the very simple and powerful notion of P-ordering appeared. We present the
notions of P-orderings and generalized factorials in Section 2. We give an expository
definition of P-orderings as in Bhargavas paper in THIS M ONTHLY [7], but first quote
what we believe is their very first definition, as in [3]. We also present his results in a
style as close as possible to their original formulations. Section 3 is on Duluth-related
stuff. Classically, integer-valued polynomials are polynomials mapping Z into Z and,
more generally, a domain D into D. The beauty of P-orderings is to allow easy gen-
eralizations, in the case D is a Dedekind domain, to polynomials mapping a subset S
of D into D. These polynomials form a D-module, and we sketch his construction of
regular bases for Dedekind domains.
In the first three sections, we thus present what we believe is necessary, from Bhar-
gavas senior thesis, to shed some light on the birth of P-orderings and generalized
factorials. We skip some proofs and leave out many results. For more details, we refer
the reader to the two papers [5, 6] he immediately published from his work. In the
very short last section, we briefly consider further applications. We focus on one ana-
lytical joint result with Kedlaya [9], especially because it can be considered under-
graduate work: It was sent to Acta Arithmetica as early as August 1997! We also hint
to further analytical developments. Although Bhargavas senior thesis is mostly alge-
braic, his paper in THIS M ONTHLY [7] already flows from algebra to analysis and
back.
With a few exceptions (in particular, for factorial ideals), we most often use Bhar-
gavas notations and terminology. For instance, we say that an ideal I of a ring R
divides an element a of R to mean that I contains a. Also, for each positive integer m,
we denote by Zm (rather than Z/mZ) the quotient ring of Z formed by the congruence
classes modulo m. However, in the last section, Z p (for p prime) denotes as usual the
ring of p-adic integers.

1. POLYNOMIAL FUNCTIONS. A polynomial F Z[x] naturally induces a


function a  F(a) from Z to Z, thus also from Zm to Zm , as a b (mod m) obvi-
ously implies F(a) F(b) (mod m). Conversely, a function f : Zm Zm need not
be induced by a polynomial, unless m is prime (in which case Zm is a field and one can
use Lagrange interpolation). Characterizing and counting the polynomial functions
from Zm to Zm is an old problem that dates back from Kempner in 1921 [30]. It was
extended by Zhibo Chen in 1995 in a paper entitled On polynomial functions from
Zn to Zm [21].
The title of Chens paper is misleading. What he really considers are functions from
{0, 1, . . . , n 1} to Zm . Denoting by x the class of x Z modulo m, let us formalize
the definition of polynomial functions.

774 
c THE MATHEMATICAL ASSOCIATION OF AMERICA [Monthly 124

This content downloaded from 128.135.98.165 on Fri, 17 Nov 2017 02:19:44 UTC
All use subject to http://about.jstor.org/terms
Definition 1. Let f : {0, 1, . . . , n 1} Zm be a function. The function f is said
to be polynomial if there is a polynomial F Z[x] such that f (a) = F(a) for all
a {0, 1, . . . , n 1}.

In his paper, Chen characterized and counted all such functions. We first summarize
his result. If a function f is represented by a polynomial F Z[x] of degree d, rather
than writing

F(x) = a0 + a1 x + + ad x d ,

it is a much better idea in this context to write F in the form

F(x) = c0 + c1 x + c2 x(x 1) + + cd x(x 1) (x d + 1).

For k n, it is obvious that x(x 1) (x k + 1) is zero on {0, 1, . . . , n 1}.


Hence, f is also represented by


n1
F(x) = ck x(x 1) (x k + 1).
k=0

It is well known (and obvious) that, for each k, the binomial polynomial
 
x x(x 1) (x k + 1)
=
k k!
is integer-valued. In other words, the values of x(x 1) (x k + 1) are divisible
by k!. On the other hand, its value at x = k, is precisely k!. Hence, the values of
ck x(x 1) (x k + 1) are zero modulo m if and only if m divides k!ck . Therefore,
m
each coefficient ck can be uniquely replaced by an equivalent integer modulo gcd{m,k!}
that one can choose to be such that 0 ck < gcd{m,k!}
m
. This is Chens result.

Theorem 1. Let n, m be two positive integers.


(i) Every polynomial function f : {0, 1, . . . , n 1} Zm can be uniquely repre-
sented by a polynomial


n1
m
F(x) = ck x(x 1) (x k + 1) where 0 ck < . (1)
k=0
gcd{m, k!}

(ii) The number of such polynomial functions is


n1
m
(n, m) = . (2)
k=0
gcd{m, k!}

Formula (1) was obtained for m = n by Singmaster (1974) [35]. Formula (2) fol-
lows immediately and is known as Kempners formula. It was first obtained, quite
laboriously, still for m = n, by Kempner (1921) [30].

Chens Question. We now turn to the question raised by Chen to which Bhargava
gave an answer, quickly published in [4]. We need another definition.

November 2017] BHARGAVAS EARLY WORK 775

This content downloaded from 128.135.98.165 on Fri, 17 Nov 2017 02:19:44 UTC
All use subject to http://about.jstor.org/terms
Definition 2. Let f : {0, 1, . . . , n 1} Zm be a function. The function f is said to
be congruence preserving if a b (mod d) implies f (a) f (b) (mod d), for all d
dividing m and all a, b {0, 1, . . . , n 1}.

It is not difficult to see that a polynomial function is congruence preserving. It turns


out the converse does not hold, even in the case m = n. Here is an easy example.

Example. Let m = n = 8. Observe that, for k 4, gcd{8, k!} = 8, thus ck = 0 in (1).


A polynomial function f : {0, 1, . . . , 7} Z8 is thus entirely determined by its values
on {0, 1, 2, 3}. In particular, if f (k) = k, for k 3, then f is represented by F = x.
Yet, the function g defined by g(k) = k for k 3, and g(k) = k 4, for 4 k 7,
is clearly congruence preserving, while not represented by F = x.

So here is the question raised by Chen:


Characterize all pairs (n, m) for which all congruence preserving functions are
polynomial.
We summarize Bhargavas solution, as an illustration of his skill in computations.
We skip the proofs (they are in [4]).
Knowing the number (n, m) of polynomials functions by Kempners formula
[Theorem 1], Bhargavas idea was to compute the number M(n, m) of congruence  pre-
e
serving functions and merely compare the two numbers. Writing m = ri=1 pi i , the
prime factorization of m, he first showed that a function f : {0, 1, . . . , n 1} Zm
is congruence preserving (respectively, a polynomial function) if and only if, for each
i, the same is true of the function f i : {0, 1, . . . , n 1} Z pei obtained by taking the
i
e
values of f modulo pi i . He could thus reduce to the case where m = pe is a prime
power. Proceeding by induction on n, he obtained the formula
n1
M(n, p e ) = pen {
k=1 min e, log p (k)
}. (3)

He then rewrote (n, p e ) in a manner similar to (3), using the fact that the largest h
such that p h divides k! is h = log p (k!)
:3


n1
pe n1
(n, p e ) = = pen k=1 min{e, log p (k!)
} . (4)
k=0
gcd{ p , k!}
e

Comparing the two formulas, one can see the only difference is that, where log p (k)
appears in (3), there is log p (k!) in (4)! It is then not too difficult to obtain Bhargavas
answer to Chens question, first for m = pe and then, by globalization, for any m.
 e
Theorem 2. Let m = ri=1 pi i be the prime factorization of m. Then every congruence
preserving function from {0, 1, . . . , n 1} to Zm is a polynomial function if and only
if, for each i, either n 2 pi , or ei = 1 in the case pi is odd and ei 2 in the case
pi = 2.

The result becomes much simpler for n = m. In addition, Bhargava obtained a den-
sity formula.
3 One can also recall that h = log p (k!)
= kp
+ pk2
+ , as in the product k! there are kp
factors
divisible by p, pk2
factors divisible by p2 , and so on.

776 
c THE MATHEMATICAL ASSOCIATION OF AMERICA [Monthly 124

This content downloaded from 128.135.98.165 on Fri, 17 Nov 2017 02:19:44 UTC
All use subject to http://about.jstor.org/terms
Corollary 3. The polynomial functions from Zm to itself are exactly the congruence
preserving functions if and only if 23  m and p2  m for all odd primes p. The density
in N of the set of all such m is 72 .

Finite principal ideal rings. Bhargava went beyond the question he was handed in
his REU program, with a generalization to finite principal ideal rings. This is Chapter 4
of his senior thesis [3]. In the original setting, one can assume that n m, as indeed
a congruence preserving function f : {0, 1, . . . , n 1} Zm is determined by its
values on a complete set of representatives modulo m. In other words, one can consider
{0, 1, . . . , n 1} as a subset of Zm . Bhargava replaced Zm by a commutative finite
principal ideal ring R and {0, 1, . . . , n 1} by a subset X of R. With these notations,
he first adapted Definitions 1 and 2 as follows.

Definition 3. Let f : X R be a function.


The function f is said to be polynomial if there is a polynomial F R[x] such that
f (a) = F(a) for all a X.
f is said to be congruence preserving if a b (mod I ) implies f (a) f (b)
(mod I ) for every ideal I of R and all a, b X.

Here, a polynomial function is represented by a polynomial in R[x], while in the


original setting, it was represented by a polynomial in Z[x]. But if two polynomials in
Z[x] have their coefficients congruent modulo m, they represent the same function.
Considering a finite sequence {ai }i=0
n1
of length n, one can write a polynomial F
R[x] in the form


n1
F= ck (x a0 ) (x ak1 ) + Q(x a0 ) (x an1 ) (5)
k=0

with ck R and Q R[x]. Bhargavas illuminating idea was to choose a sequence


|X |1
{ai }i=0 of elements of X (where |X | denotes the cardinality of X ) in such a way that
one can easily tell when F vanishes on X and hence when two polynomials represent
the same function f : X R, as in the proof of Theorem 1. In the original setting,
one uses the fact that the usual ordering of the integers is such that, as stated in the
introduction of Bhargavas senior thesis [3] if we take any k consecutive elements in
this natural ordering, then their product must always be an element of the ideal (k!).
Equivalently, for each prime p,
() the highest power of p that divides k! also divides n(n + 1) (n + k 1) for all
k and all n.
As R is a finite ring, it is isomorphic to a product of local rings:


r
e
R R/P j j , (6)
j=1

where the P j s are the prime ideals of R, each of them maximal. Bhargava first con-
|X |1
structed, for each prime P j of R, an ordering {ai, j }i=0 of X, with a property similar
to ():
k1
() for every positive k < |X |, the highest power of P j dividing i=0 (ak, j ai, j ) also
k1
divides i=0 (a ai, j ) for all a X.

November 2017] BHARGAVAS EARLY WORK 777

This content downloaded from 128.135.98.165 on Fri, 17 Nov 2017 02:19:44 UTC
All use subject to http://about.jstor.org/terms
|X |1
In the next section, we explain how Bhargava constructed each sequence {ai, j }i=0
in an amazingly simple way. This is how P-orderings first appeared. For the moment,
all we need to know is that P-orderings exist. From all these local sequences, he then
obtained a global sequence {ai }i=0
n1
using the Chinese remainder theorem.

The global sequence. With the previous notations, for each prime ideal P j of R we
|X |1
let {ai, j }i=0 be a P j -ordering of X, that is, a sequence with property (). By the Chi-
|X |1
nese remainder theorem, there exists a sequence {ai }i=0 that satisfies the following
congruence relations for all i and j, 0 i < n, 1 j r :
 e

ai ai, j mod P j j . (7)

Although this sequence is not necessarily in X, it behaves somewhat like a P j -ordering


for each j.
|X |1
Lemma 4. There exists a sequence {ai }i=0 in R such that
k1 k1
(i) i=0 (ak ai ) divides i=0 (a ai ) for all k {1, . . . , n 1} and a X,
(ii) each polynomial F R[x] that vanishes on X vanishes also on this sequence.

As is often the case, Bhargava is a bit fast and intuitive. He uses the first property
with no proof and tacitly assumes the second one. For both, we prefer to give a short
justification.
k1 k1
Proof. By (7), we have i=0 (ak ai ) i=0 (ak, j ai, j ) (mod P e j ), and also
k1 k1
i=0 (a ai ) i=0 (a ai, j ) (mod P ) for all a R. By property (), as
ej
ej k1
R/P is a principal local ring, the class of i=0 (ak, j ai, j ) divides the class of
k1j k1 k1 ej
i=0 (a ai, j ) in this ring. Thus, i=0 (ak ai ) divides i=0 (a ai ) modulo P j .
As this holds for each prime P j , that proves (i).
Now, let F R[x]. If F vanishes on X, then F(ai, j ) = 0, for all i and j, since
e
each ai, j is in X. A fortiori, F(ai, j ) is zero modulo P j j and, by (7), so is F(ai ). As this
holds for each prime P j , that proves (ii).

|X |1
The sequence {ai }i=0 plays the role of the sequence {0, 1, . . . , n 1} and the prod-
k1
uct i=0 (ak ai ) the role of k!. This is how the generalization of the factorial function
first appeared, with the following definitions and notations.
For 0 < k < n, the kth factorial ideal of X , denoted by k! X ,4 is the ideal generated
k1
by the product i=0 (ak ai ):


k1
k! X = (ak ai )R. (8)
i=0

The cofactorial ideal of X, denoted by k (X ), is the annihilator of k! X :





k1
k (X ) = Ann R (k! X ) = a R | a (ak ai ) = 0 .
i=0
4 In [5] Bhargavas less intuitive notation was k (X ). He changed it to k! X in [7].

778 
c THE MATHEMATICAL ASSOCIATION OF AMERICA [Monthly 124

This content downloaded from 128.135.98.165 on Fri, 17 Nov 2017 02:19:44 UTC
All use subject to http://about.jstor.org/terms
Also, it may be that k! X = (0), for some k < |X |. Moreover, we shall later (easily)
see that the factorial ideals form a decreasing sequence [Proposition 14]. In particular,
k! X = (0) implies (k + 1)! X = (0). Bhargava thus sets a last notation.
(X ) denotes the smallest k, if any, such that k! X = (0), otherwise (X ) = |X |.
|X |1
Structure theorem. Given a sequence {ai }i=0 as in Lemma 4, and writing a polyno-
mial F R[x] as in (5), Bhargava first determined under which conditions F vanishes
on X. He used several lemmas that we summarize with a single one. Note that the
cofactorial ideal k (X ) plays the role of gcd{m,k!}
m
(annihilator of k! in Zm ) appearing in
Theorem 1.

Lemma 5. Let X be a subset of a finite principal ideal ring R and F R[x] be a


polynomial of degree d. Then one can write
|X |1

F= ck (x a0 ) (x ak1 ) + Q(x a0 ) (x a|X |1 ),
k=0

and F vanishes on X if and only if ck k (X ) for all k {0, . . . , |X | 1}.

Proof. Suppose that ck k (X ) for all k {0, . . . , |X | 1}. By definition of k (X ),


k1 k1
we have ck i=0 (ak ai ) = 0. Thus, by Lemma 4 (i), we also have ck i=0 (a ai ) =
|X |1
0 for all a X. To prove that F vanishes on X , it remains to show that i=0 (x ai )
|X |1 |X |1
vanishes on X. As {ai, j }i=0 is an ordering of X, we have i=0 (a ai, j ) = 0, for
|X |1
all j and a X. A fortiori, i=0 (a ai, j ) 0 (mod P ). By (7), we also have
ej
|X |1
i=0 (a ai ) 0 (mod P ). As this holds for each j, we are done.
ej

Conversely, suppose that F vanishes on X. By (ii) in Lemma 4, F also van-


|X |1
ishes on the sequence {ai }i=0 . Now F(a0 ) = 0 implies c0 = 0. Thus, F(a1 ) = 0
implies c1 (a1 a0 ) = 0, and hence, c1 1 (X ). By induction, F(ak ) = 0 implies
k1
ck i=0 (ak ai ) = 0, and hence, ck k (X ).

Theorem 6. Let X be a subset of a finite principal ideal ring R. Given a sequence


|X |1
{ai }i=0 in R as in Lemma 4, a polynomial function f : X R can be represented
by a polynomial
(X )1

F= ck (x a0 )(x a1 ) (x ak1 ), ck R,
k=0

where the ck are uniquely determined modulo k (X ).


Hence, the set of all polynomial functions from X to R and the direct sums
(X )1 (X )1

R/k (X ) k! X
k=0 k=0

are isomorphic as R-modules. Moreover, the number (X, R) of polynomial functions


from X to R is
(X )1

(X, R) = |k! X |.
k=0

November 2017] BHARGAVAS EARLY WORK 779

This content downloaded from 128.135.98.165 on Fri, 17 Nov 2017 02:19:44 UTC
All use subject to http://about.jstor.org/terms
The first two statements are Theorem 3 in Chapter 4 of [3]. The last one, generaliz-
ing Kempners formula, is Theorem 4 in the same chapter.

Proof. As (x a0 ) (x a|X |1 ) vanishes on X, one can represent a function by


 |1
a polynomial F = |X k=0 ck (x a0 ) (x ak1 ). In fact, there is no need to take
the sum from 0 to |X | 1 but only from 0 to (X ) 1, as k! X = (0) implies that
(x a0 ) (x ak1 ) vanishes on X by Lemma 4. Finally, it follows immediately
from Lemma 5 that the coefficients ck are uniquely determined modulo k (X ). The
two last statements follow easily.

Bhargava even gave an explicit formula for (R, R) [3, Corollary 2] in terms, as he
puts it, of the basic constants associated to the ring R. He thus recovered a result of
Brawley and Mullen [11] obtained in the particular case of Galois rings (already with
a somewhat laborious generalization of factorials). The computation is very technical,
and we skip it. As a corollary [3, Corollary 7], he also easily generalized an old result
of Hensel (1896) [27]:
All values of a polynomial F(x) Z[x] of degree d are divisible by m if and only
if m divides F(i) for i = 0, 1, . . . , d.

Corollary 7. A polynomial F R[x] of degree d vanishes on X if and only if F(ai ) =


0 for i min{d, (X ) 1}.

Chapter 5 of the senior thesis. Bhargava went further about polynomial and congru-
ence preserving functions over a finite principal ideal ring. For instance, he gave nec-
essary and sufficient conditions for a given function f : X R to be representable by
a polynomial in R[x], with a beautiful generalization of a result of Carlitz [14] (1964).
As nice as it is, we nevertheless skip that part but comment on the last two results of
Chapter 5 between which, in his own words, there is a striking resemblance in the
case where R is local. Here is the first one [3, Theorem 8].

Theorem 8. Let X be a subset of a finite principal ideal ring R. Then all functions
from X to R are polynomial functions if and only if, for each each prime P R, no
two elements of X are congruent modulo P.

In fact, whatever the commutative ring R, the condition is obviously necessary,


as a polynomial function must be congruence preserving. Conversely, it implies that
the difference of any two elements is a unit, hence that R is a field. Therefore, this
condition is sufficient, provided X is finite: By Lagrange interpolation, every function
f : X R can then be represented by a (unique) polynomial (of degree d < |X |).
The second result [3, Theorem 9] generalizes Bhargavas answer to Chens question
to finite principal ideal rings.

Theorem 9. Let X be a subset of a finite principal ideal ring R. Then all congruence
preserving functions from X to R are polynomial functions if and only, for each prime
P of R, no three elements of X that are distinct modulo a sufficiently high power of P
are congruent modulo P.

This is Bhargavas somewhat elliptic wording! What the condition says is that, if three
elements are congruent modulo P, at least two of them must then be congruent modulo
P r for all r 1. In the local case, r=1 P r = 0, and this condition becomes much
simpler: No three elements of X are congruent modulo P. This sounds indeed very
much like the condition in Theorem 8! We skip the proof.

780 
c THE MATHEMATICAL ASSOCIATION OF AMERICA [Monthly 124

This content downloaded from 128.135.98.165 on Fri, 17 Nov 2017 02:19:44 UTC
All use subject to http://about.jstor.org/terms
2. P-ORDERINGS AND FACTORIAL IDEALS. Bhargavas main achievement is
probably the tool he forged to solve his problem. This did not escape Barry Mazur who
wrote the following in a letter of support for the Hoopes Prize (May 1996):

He has a construction in his senior thesis which seems to me to be wonderfully clean, surpris-
ing, completely original (as far as I know) and which really shed light on the subject.

Definitions of P-orderings. Having first in mind finite principal ideal rings, Bhar-
gava soon grasped that P-orderings had a much larger scope. He then considered
Dedekind-type rings that he defined as Noetherian, locally principal and with dimen-
sion at most one. (In fact, the last condition is superfluous as it follows from the second
one.) Dedekind-type rings include finite principal ideal rings, Dedekind domains, and
more, as for instance Z Z/2Z. Let us quote verbatim what we believe is the very
first occurrence of P-orderings, as in [3, Chapter 2]:

Let R be a Dedekind-type ring, X an arbitrary subset of R, and P any fixed prime ideal of
R. For an element a R, let w P (a) denote the highest power of P containing the ideal (a).
By assigning labels a0 , a1 , . . . to elements of X, we obtain what we call a P-ordering of X as
follows. Choose a0 to be any element of X, and for 0 < k < |X |, inductively choose ak to be
an element of X which minimizes the highest power of P dividing

(ak a0 )(ak a1 ) (ak ak1 ).


|X |1
Given such a P-ordering {ai }i=0 , we refer to the associated descending chain of ideals
|X |1
{k (X, P)}k=0 , defined by

k (X, P) = w P ((ak a0 )(ak a1 ) (ak ak1 )),


|X |1
as the associated P-sequence of X corresponding to the P-ordering {ai }i=0 .

|X |1
If X is a finite set, a P-ordering {ai }i=0 is then truly an ordering of X. If X is
infinite, Bhargava noted that a P-ordering may not include all elements of X (adding
in [5] that, if X is countable, it could certainly be constructed to do so5 ). Later on,
he defined a P-ordering to be an infinite sequence {an }n0 of X, for arbitrary X, even
finite. This point of view is now standard. We thus give a definition, always for a
subset X of Dedekind-type ring R, in the style of a game called p-ordering in THIS
M ONTHLY (2000) [7] (where X is a subset of the ring Z of integers, and p a prime
number).

Definition 4. A P-ordering of X is a sequence {an }n0 in X that is formed as follows:


Choose any element a0 X ;
Choose an element a1 X that minimizes the highest power of P dividing
a1 a0 ;
Choose an element a2 X that minimizes the highest power of P dividing
(a2 a0 )(a2 a1 );
and in general, at the kth step,
Choose an element ak X that minimizes the highest power of P dividing
(ak a0 )(ak a1 ) (ak ak1 ).
5 This affirmation was proved much later by Youssef Fares [23].

November 2017] BHARGAVAS EARLY WORK 781

This content downloaded from 128.135.98.165 on Fri, 17 Nov 2017 02:19:44 UTC
All use subject to http://about.jstor.org/terms
Remark 1. In the case of a Dedekind domain D, it may be convenient to consider
valuations. To a (nonzero) prime ideal P of D is associated a discrete valuation v. A
sequence {an }n0 is a P-ordering of X when, a0 being arbitrarily chosen, ak is induc-

tively defined to minimize the valuation of the product k1
j=0 (ak ai ):


k1 
k1
a X, v(ak ai ) v(a ai ). (9)
i=0 i=0
k1
One then has k (X, P) = P k (X ) where k (X ) = i=0 v(ak ai ). As such, the notion
of P-ordering can be extended to various contexts where valuations appear, under the
name of v-ordering, provided X is precompact in the v-adic topology [17].

The P-sequence. We next formalize the definition of the associated P-sequence.

Definition 5. Given a P-ordering {an }n0 of X, the associated P-sequence is the


sequence {k (X, P)}k0 of powers of P, where k (X, P) is the power of P mini-
mized at the kth step of the P-ordering process: More precisely, 0 (X, P) = R, and
for k 1,
 
k (X, P) = w P (ak a0 )(ak 1) (ak ak1 ) ,
where w P (a) denotes the highest power of P dividing a.

In his senior thesis, Bhargava made some tacit assumptions. First, he assumed P to
be a nonzero prime (probably because he first had in mind finite ringsfortunately,
this became explicit in [5]). Also, he did not explicitly mention 0 (X, P). Finally,
he said that the P-sequence is a descending chain of ideals with no justification.
Although this is more or less trivial, we provide here a short argument. In fact,
k (X, P) is some power of P, and we prefer to say (as he did himself in [7], again
with no justification) that these powers form a monotone increasing sequence.

Lemma 10. The P-sequence {k (X, P)} k0 is a monotone increasing sequence of


powers of P (equivalently, {k (X, P)}
k0 is a descending chain of ideals).

Proof. By definition of the associated P-sequence, k (X, P) is the highest power of



P dividing k1 j=0 (ak a j ). By the minimizing property that defines a P-ordering,

k (X, P) also divides k1 (ak+1 a j ). A fortiori k (X, P) is a power of P that
k j=0 k1
divides the product j=0 (ak+1 a j ) = (ak+1 ak ) j=0 (ak+1 a j ). This power
must be less than (or equal to) the highest such power, that is, k+1 (X, P).

The descending P-sequence starts with 0 (X, P) = R = (1). It may go down to



k (X, P) = (0) for some k. This is the case if k1j=0 (ak a j ) = 0 as every power of
P divides 0. This certainly occurs for k |X |.

Lemma 11. If X is finite, then k (X, P) = (0) for k |X |.

We made explicit Lemmas 10 and 11 for further reference, but Bhargava preferred
to jump to his first main result. Although P-orderings are obviously not unique, as a0 is
arbitrary and one can make choices at some further steps, they have, in his own words,
several nice properties that are completely independent of these choices. In particular,
here is his Theorem 1 in [3].

782 
c THE MATHEMATICAL ASSOCIATION OF AMERICA [Monthly 124

This content downloaded from 128.135.98.165 on Fri, 17 Nov 2017 02:19:44 UTC
All use subject to http://about.jstor.org/terms
Theorem 12. Let P be a prime ideal of a Dedekind-type ring R, and let X be any
subset of R. Then any two P-orderings of X necessarily result in the same associated
P-sequence of X.

He postponed the proof and so do we (until the next section). He then gave three
examples. We quote the first two (almost verbatim for Example 1) but skip the last
one, clearly inspired by Polya [34] and Carlitz [13], as it is quite elaborate but given
without any explanation.

Example 1. For an odd prime p, let X be the subset {1, p, 2 p, p 2 , p 2 + 1} of Z.


Then the orderings 1, p, 2 p, p2 , p 2 + 1 and 1, p, p 2 , p 2 + 1, 2 p are both seen to be
p-orderings of X. They both lead to the same associated p-sequence of X , which is
(1), (1), ( p), ( p2 ), ( p2 ).

Example 2. The natural sequence {0, 1, 2, 3, . . .} is a p-ordering of Z (and also of the


subset N of natural integers) for all primes p. The associated p-sequence is given by
k (Z, p) = w P (k!) (the highest power of p dividing k!).

Factorial ideals. A generalization of the factorial function immediately follows the


introduction of P-orderings. Bhargava first needs a lemma that does in fact depend
on the structure of the type of rings we are considering. Specifically, it relies on the
fact that every ideal I in a Dedekind-type ring has a factorization into prime ideals. . .
We quote verbatim this lemma and its proof [3, Lemma 3, Chapter 2].

Lemma 13. Let R be a Dedekind-type ring, and let X R be any subset. Then for
any integer 0 k < |X |, k (X, P) = R for only finitely many primes P.

Proof. The condition k (X, P) = R is equivalent to the existence of k elements in


X that are all distinct modulo
 P. However, given any k elements of X, the principal
ideals generated by the k2 pairwise differences of these elements are each contained
in at most finitely many primes, namely the primes that appear in the factorization of
the ideal into primes. It follows that any k elements of X must be distinct modulo P
for all but finitely many primes P, implying the lemma.

This result allows Bhargava to set the following definition.

Definition 6. Let R be a Dedekind-type ring, and let X R be any subset. For each
k 0, the ideal

k! X = k (X, P) (10)
P prime

is called the kth factorial ideal of X .

If k |X |, then the previous lemma does not apply, but from Lemmas 10 and 11,
we have the following easy properties that are implicit in Bhargavas thesis.

Proposition 14. Let R be a Dedekind-type ring, and let X R be any subset. The
factorial ideals of X form a descending chain of ideals. Moreover, 0 X ! = R and k! X =
(0) for k |X |.

November 2017] BHARGAVAS EARLY WORK 783

This content downloaded from 128.135.98.165 on Fri, 17 Nov 2017 02:19:44 UTC
All use subject to http://about.jstor.org/terms
The factorial ideals enjoy many of those properties which are commonly used
when working over sets of consecutive integers [3, Chapter 2]. Many of these beau-
tiful properties are to be found in the expository paper Bhargava later wrote for THIS
M ONTHLY [7]. For instance, it is known that, for any nonnegative integers r and s, the
product r !s! divides (r + s)!. Using Bhargavas results on integer-valued polynomials,
it will be fairly easy to generalize this property as follows.

Theorem 15. Let D be a Dedekind domain, and let X D be any subset. For any
nonnegative integers r and s, the product r ! X s! X divides (r + s)! X .
Simultaneous orderings. The natural sequence {0, 1, 2, 3, . . .} is a p-ordering of Z
for all primes p [Example 2]. At the end of his senior thesis, Bhargava raised the
following question [3, Question 3].
Question. Which Dedekind domains are simultaneously P-orderable for all P?
He added Simultaneous P-orderability is such a strong condition on a Dedekind
domain that it would seem at first that perhaps no domains other than Z have this
property. It is indeed conjectured that there is no simultaneous P-ordering for the
ring of integers of a proper number field. In the 2000 Duluth REU program, Melanie
Wood, who would a few years later become a Ph.D. student of Bhargavas, proved
this conjecture for imaginary quadratic fields [36] (and, as her famous adviser did, she
received the Morgan Prize). More recently, it was proved to also hold for real quadratic
number fields, with at most finitely many exceptions [1].
On the other hand, Bhargava noted that if D is semi-local (that is, with finitely many
maximal ideals), then a direct application of the Chinese Remainder Theorem may
be used to give a simultaneous P-ordering. He also gave a less trivial example with
the ring D = Fq [t] of polynomials over the finite field Fq , referring to Carlitz [15]
in [5] and relating the factorial ideals to what are known as the Carlitz factorials.
Finally, he showed that, for a Dedekind domain D that is an extension of Z, the natural
sequence {0, 1, 2, 3, . . .} is a simultaneous P-ordering of D if and only if every prime
of Z not invertible in D splits completely in D [3, Theorem 21]. (Not invertible in
D was omitted in [3] but correctly stated in [5] where Bhargava mentioned this to
be a result of Chabert and Gerboud [20] independently obtained by Halter-Koch and
Narkiewicz [26].)
Later, Bhargava extended the question to subsets [7]. He gave two examples of
subsets of Z, each equipped with an (obviously suggested) simultaneous ordering:
{2n | n N} and {n 2 | n N}. Also, in the case of a subset X of a finite principal ideal
|X |1
ring R, there was a sequence {ai }i=0 in R (but not necessarily in X ) that behaved
like a P-ordering of X for each prime P of R [Lemma 4]. For k < |X |, we thus had
k1
from (8) that k! X = i=0 (ak ai )R. Similarly, for a subset of a Dedekind-type ring,
we have the following that we record for further use.

Lemma 16. Let R be a Dedekind-type ring, and let X R be any subset. If {ai }i0 is
k1
a simultaneous P-ordering of X , then k! X = i=0 (ak ai )R.

3. INTEGER-VALUED POLYNOMIALS. We now show how Bhargava applied


P-orderings to the study of integer-valued polynomials6 in Chapter 6 of his senior
thesis. Throughout this section, D is a Dedekind domain with quotient field K and X
a subset of D. The ring Int(X, D) of integer-valued polynomials over X is formed by
the polynomial functions in K [x] that map X into D:
6 This topic appeared in the AMS subject classification (2000): 13F20 Polynomial rings and ideals; rings of

integer-valued polynomials.

784 
c THE MATHEMATICAL ASSOCIATION OF AMERICA [Monthly 124

This content downloaded from 128.135.98.165 on Fri, 17 Nov 2017 02:19:44 UTC
All use subject to http://about.jstor.org/terms
Int(X, D) = {F(x) K [x] | F(X ) D} .

In the case X = D, one simply denotes by Int(D) the ring of integer-valued polyno-
mials over D (mapping D into D).
In 1996, D. L. McQuillan wrote the following about Bhargavas contribution to this
topic in a letter supporting the nomination of Bhargava for the Morgan Prize.

There have been several recent attempts to shed light on Int(X, D) for arbitrary X , but no one
has come close to achieving results as powerful as those presented in this paper. . . . Several
French colleagues7 will be interested.

Fixed divisor. Let us recall some standard terminology and notations.


The content c(F) of a polynomial F D[x] is the ideal of D generated by its
coefficients.
F D[x] is said to be primitive if its coefficients generate the unit ideal of D, that
is, c(F) = D.
The fixed divisor d(X, F) of F over X is the ideal of D generated by the image set
F(X ), that is, generated by the the values of F on X.
In order to determine the leading coefficients of degree k polynomials in Int(X, D),
Bhargava first determined the fixed divisor of primitive polynomials [3, Theorem 11].
In fact, he needed a more general result, as follows.

Theorem 17. If F D[x] is a polynomial of degree k, then

c(F) k! X d(X, F). (11)

In particular, if F is primitive, then k! X d(X, F). Moreover, for each k!, the equality
k! X = d(X, F) is achieved by the polynomial

Sk = (x a0,k )(x a1,k ) (x ak1,k ), (12)

where {ai,k }i0 is a sequence in D that, for each prime P containing k! X , is termwise
congruent modulo k (X, P) to some P-ordering.

Bhargava notes in [6] that the particular case of primitive polynomials generalizes
a classical result of Polya [33]:
If F Z[x] is a primitive polynomial of degree k, then the fixed divisor d(F, Z) of
F divides k!. Moreover, for each k, there is a primitive polynomial of degree k such
that d(F, Z) = k!.
It is easy to see that a case of equality is given by Sk = x(x 1) (x k + 1).
Indeed, Sk (k) = k!, and k! divides Sk (n) = n(n 1) (n k + 1) for all n Z. As
P-orderings generalize the natural ordering of the integers, it follows similarly that
a case of equality in (11) is achieved by the polynomial Sk defined in (12). Let us
note that, given an integer k 0, there are finitely many primes P dividing k! X and
hence that the Chinese remainder theorem allows us to find the sequence {ai,k }i0 . This
sequence is similar to the global sequence of Lemma 4 (but for the fact that it depends
on k). Bhargava was probably not pleased with his proof (neither are we) in [3] as he
shortly after gave a totally different one in [6]. Thus, we skip the proof of Theorem 17.
7 These French colleagues are probably the authors of a monograph on integer-valued polynomials [12].

November 2017] BHARGAVAS EARLY WORK 785

This content downloaded from 128.135.98.165 on Fri, 17 Nov 2017 02:19:44 UTC
All use subject to http://about.jstor.org/terms
The interested reader could, however, see that his later argument can easily be gener-
alized from primitive polynomials to polynomials with coefficients in D.
Bhargava gave much stronger results on fixed divisors in another chapter of his
senior thesis [3, Chapter 7]. In particular, given a primitive polynomial F D[x], he
gave a formula to compute d(X, F) in terms of a unimodular matrix. He also showed
that, for every ideal I containing k! X , there exists a primitive degree k polynomial
F D[x] with d(X, F) = I . For these results, we refer the reader to [6] (1998).

Leading coefficients. Bhargava next determined the leading coefficients of degree k


polynomials in Int(X, D). Recall that the inverse I 1 of a nonzero ideal I of D is the
fractional ideal

I 1 = {b K | bI D}.

Here is Theorem 12 in [3].

Corollary 18. The set formed by the leading coefficients of polynomials of degree k in
Int(X, D), together with 0, is the fractional ideal (k! X )1 .

The adjunction of 0, obviously necessary to obtain an ideal, is in fact tacit both


in the senior thesis [3] and the following publication [5]. To use the previous results,
Bhargava writes a polynomial F of degree k in the form

c0 + c1 x + + ck x k G
F= = ,
b b

with G D[x], b D. The leading coefficient of F is cbk . If F is integer-valued, G


must map X into the principal ideal (b). From Theorem 17, one can derive

ck k! X c(G)k! X d(X, G) (b).

The proof follows. As we said above, Bhargava dealt only with the fixed divisor of
primitive polynomials. However, given F Int(X, D), it is not always possible to
write F = Gb with G primitive.

Regular bases. Bhargava opened his study on integer-valued polynomials with an


elementary but classic result of G. Polya (Theorem 10 in [3]):
A polynomial in Q[x] maps Z into
 Z if and only if it can be written as a Z-linear
combination of the polynomials xi = x(x1)(xi+1)
i!
.
He added that Polya thus showed that the ring Int(Z) is a free Z-module with a
basis consisting of exactly one polynomial of each degree. For this reason, Z is said to
have a regular basis. More generally, Int(X, D) is said to have a regular basis if it is
a free D-module with a D-basis consisting of exactly one polynomial of each degree.
Bharagavas aim was to give an explicit description of Int(X, D) with necessary and
sufficient conditions for the existence of a regular basis. As he says himself, his work
unifies much of the previous work on this topic, as in [24, 25, 32, 33], and [34], mostly
done in the case of Int(D). What P-orderings allow us to achieve is the consideration
of subsets, finite or infinite!
In his 1996 Morgan Prize nomination letter, Willam W. Smith wrote

786 
c THE MATHEMATICAL ASSOCIATION OF AMERICA [Monthly 124

This content downloaded from 128.135.98.165 on Fri, 17 Nov 2017 02:19:44 UTC
All use subject to http://about.jstor.org/terms
I recently worked on a problem which involved constructing an algorithm for a basis for a
specific ring of integer-valued polynomials.8 I wish I had been given the opportunity to read
Mr. Bhargavas work first.

With the notations of Theorem 17, one can uniquely represent a degree n polyno-
mial in the form
n 
n
F= bk Sk = bk (x a0,k ) (x ak1,k ), (13)
k=0 k=0

with coefficients bk K . Bhargava could then derive the conditions for F to be


integer-valued, with a structure theorem somewhat similar to Theorem 6.

Theorem 19. A polynomial F is in Int(X, D) if and only if it can be represented in


the form

n 
n
F= bk Sk = bk (x a0,k ) (x ak1,k ),
k=0 k=0

 bk 1
with coefficients (k! X )1 uniquely determined by F. Thus, Int(X, D) and the
direct sum k=0 (k! X ) are isomorphic as D-modules.

Proof. By Theorem 17, if each bk is in (k! X )1 , then each bk Sk maps X into D. Thus,
F is integer-valued. Conversely, if F is integer-valued, its leading coefficient bn must
be in (n! X )1 , by Corollary 18, and so must be the difference F bn Sn with leading
coefficient bn1 , thus bn1 must be in ((n 1)! X )1 , and so on!

If k! X is a nonzero principal ideal with generator k , then (k! X )1 is a principal


fractional ideal with generator 1/k . If this holds for all k! X , it follows easily from
Theorem 19 that the polynomials Sk /k form a regular basis of Int(X, D). The con-
verse is not too difficult to prove. Here is Theorem 14 in [3].

Corollary 20. The D-module Int(X, D) has a regular basis if and only if k! X is a
nonzero principal ideal for all k 0. In that case, a regular basis of Int(X, D) is
given by
(x a0,k )(x a1,k ) (x ak1,k )
(k 0),
k
where k is a generator of the ideal k! X .

By Lemma 16, if a sequence {ai }i0 is a P-ordering of X for all (nonzero) primes
P of D, the associated factorial ideal k! X is then a principal ideal generated by
k1
i=0 (ak ai ). If X is an infinite subset, such a simultaneous P-ordering provides a
regular basis of Int(X, D), an analog to Polyas regular basis in the case of Int(Z). In
fact, given any sequence {ai }i0 in X , the terms of which are distinct, one can consider
the associated generalized binomial polynomials
   x ai
x
= . (14)
ak 0i<k
a k ai

We then have the following (Bhargavas Theorem 20 in [3], for X = D).


8 Most probably, Smith refers to a work on Int(P, Z), published in [18].

November 2017] BHARGAVAS EARLY WORK 787

This content downloaded from 128.135.98.165 on Fri, 17 Nov 2017 02:19:44 UTC
All use subject to http://about.jstor.org/terms
Corollary 21. Let X be an infinite subset of D, and let {a  i }i0 be a sequence in X ,
the terms of which are distinct. The generalized binomials axk form a regular basis of
Int(X, D) if and only if {ai }i0 is a simultaneous P-ordering of X.

Postponed proofs. Bhargava proved Theorem 12 in the case of finite principal ideal
rings, using Theorem 6. We give here a very similar proof in the case of Dedekind
domains, using Theorem 19.

Proof of Theorem 12. By the fundamental structure of Int(X, D) [Theorem 19],




Int(X, D) (k! X )1 .
k=0

In particular, the D-module


 formed by the integer-valued polynomial of degree at most
n is isomorphic to nk=0 (k! X )1 . It follows by induction on n that the 
ideal (k! X )1
depends only on k, X , and D. The same can be said of its inverse k! X = P k (X, P).

Proof of Theorem 15. If two polynomials f and g are of respective degree r and s,
then their product f g is of degree r + s. Comparing the leading coefficients, one
immediately derives from Corollary 18 that (r ! X )1 (s! X )1 ((r + s)! X )1 , and
therefore that (r + s)! X r ! X s! X .

4. FURTHER DEVELOPMENTS. Let us quote another excerpt from a Morgan


Prize support letter, this one from Eric Brussel (a reader of Bhargavas senior thesis).

In addition to several new results, Bhargava captures as corollaries published theorems by


Kempner, Singmaster, Hensel, Dueball, Brawley-Mullen, Carlitz, Polya, Ostrowski, and
Gilmer . . . His work will very likely lead to further developments in the field.

Bhargava indeed undertook a tremendous amount of reading for his senior thesis
and was thus already a very serious scholar. Also, the final sentence is prophetic: His
work did lead to many further developments! Papers on P-orderings and generalized
factorial ideals appeared abundantly in the literature. The notion of factorial ideals was
further generalized in [16]. Techniques were developed for effective computations, as
in [10] or [28]. These notions were extended to some noncommutative domains, such
as quaternion rings [29]; they even appeared in dynamical systems [19].
On the other hand, as immediately foreseen by Bhargava, the properties of P-
orderings could also be viewed more generally as statements about certain sequences
in ultrametric spaces [3, Chapter 2]. By a classical theorem of Mahler [31] (1958),
denoting by Q p the field of p-adic numbers and by Z p the ring of p-adic integers (that
is, the completions of Q and Z with respect to the p-adic valuation):
Every continuous function C (Z p , Q p ) can be uniquely represented in the form
  
x
(x) = ck with ck Q p and ck 0 as k .9
k=0
k

Mahlers result was generalized by Bhargava, still an undergraduate, in collabora-


tion with Kedlaya [9], replacing Q p by any local field K (that is, the quotient field
9 In particular, every continuous function is thus uniformly approximated by polynomials. Such an ultra-

metric version of the StoneWeierstrass theorem had first been obtained by Dieudonne [22] (1944).

788 
c THE MATHEMATICAL ASSOCIATION OF AMERICA [Monthly 124

This content downloaded from 128.135.98.165 on Fri, 17 Nov 2017 02:19:44 UTC
All use subject to http://about.jstor.org/terms
of a complete discrete valuation domain V, with finite residue field), and Z p by any
compact subset S of K . Observing that the natural sequence {0, 1, 2, 3, . . .} is a p-
ordering of Z p , they had the great idea to consider a -ordering {ai }i0 of S (where 
is a generator of the maximal ideal of V ). They could then replace the binomials xk
  
by the generalized binomials axk = 0i<k axa i
k ai
, as in (14):

Theorem 22. Every continuous function C (S, K ) can be uniquely represented in


the form
  
x
(x) = ck with ck K and ck 0, as k . (15)
k=0
ak

Bhargava planned to extend his methods to obtain similar results for various ultra-
metric spaces, considering appropriate subrings of integer-valued polynomials that
may also be of interest in their own right, as he said in a talk he gave in 2000 at a
meeting organized in Marseille (France).10 With the previous notations, Bhargava con-
sidered a subspace B (S, K ) of C (S, K ), the appropriate subring being the intersection
B (S, K ) Int(S, V ). In [8], published in 2009, he reached his goal in two cases: the
space of locally analytic functions on the one hand, and the space of r -times continu-
ously differentiable functions on the other. For each case, he made a variation on the
theme of P-orderings. Once again, the main interest is probably with this tool. Soon
after the Marseille meeting, a young French student, Julie Yeramian, undertook the
study of the rings that occur in the case of locally analytic functions, indeed of some
interest in their own right, under the name of Bhargava rings [37].
To conclude about P-orderings, as Bhargava himself said (but not only for algebra):

Amazing how far one little combinatorial idea can go (in algebra!).

ACKNOWLEDGMENT. The authors wish to thank the referee and the editors for their numerous very help-
ful comments.

REFERENCES

1. D. Adam, P.-J. Cahen, Newtonian and Schinzel quadratic fields, J. Pure and Appl. Algebra 215 (2011)
19021918.
2. M. Bhargava, Mistiling of the plane with rectangles, preprint (1995).
3. , On P-Orderings and Polynomial Functions on Arbitrary Subsets of Dedekind-Type Rings,
Senior thesis, Harvard Univ., Cambridge, MA, 1996.
4. , Congruence preservation and polynomial functions from Zn to Zm , Discrete Math. 173 (1997)
1521.
5. , P-orderings and polynomial functions on arbitrary subsets of Dedekind rings, J. Reine Angew.
Math. 490 (1997) 101127.
6. , Generalized factorials and fixed divisors over subsets of a Dedekind domain, J. Number Theory
72 (1998) 6775.
7. , The factorial function and generalizations, Amer. Math. Monthly 107 (2000) 783799.
8. , On P-orderings, rings of integer-valued polynomials, and ultrametric analysis, J. Amer. Math.
Soc. 22 (2009) 963993.
9. M. Bhargava, K. S. Kedlaya, Continuous functions on compact subsets of local fields, Acta Arith. 91
(1999) 191198.
10. J. Boulanger, J.-L. Chabert, Asymptotic behavior of characteristic sequences of integer-valued polyno-
mials, J. Number Theory 80 (2000) 238259.
11. J. V. Brawley, G. L. Mullen, Functions and polynomials over Galois rings, J. Number Theory 41 (1992)
156166.
10 Second Encounter on Integer-Valued Polynomials.

November 2017] BHARGAVAS EARLY WORK 789

This content downloaded from 128.135.98.165 on Fri, 17 Nov 2017 02:19:44 UTC
All use subject to http://about.jstor.org/terms
12. P.-J. Cahen, J.-L. Chabert, Integer-Valued Polynomials. Vol. 48, Amer. Math. Soc. Surveys and Mono-
graphs, American Mathematical Society, Providence, RI, 1997, http://dx.doi.org/10.1090/surv/
048.
13. L. Carlitz, A class of polynomials, Trans. Amer. Math. Soc. 43 (1938) 167182.
14. , Functions and polynomials (mod p n ), Acta Arith. 9 (1964) 6778.
15. , A set of polynomials, Duke Math. J. 6 (1940) 486504.
16. J.-L. Chabert, Generalized factorial ideals, Arab. J. Sci. Eng. 26 (2001) 5168.
17. J.-L. Chabert, P.-J. Cahen, Old Problems and new questions around integer-valued polynomials and fac-
torial sequences, in Multiplicative Ideal Theory in Commutative Algebra, a Tribute to the Work of Robert
Gilmer. Ed. J. W. Brewer, S. Glaz, W. Heinzer, and B. Olberding. Springer, New York, 2006. 89108,
http://dx.doi.org/10.1007/978-0-387-36717-0.
18. J.-L. Chabert, S. Chapman, W. W. Smith, A basis for the ring of polynomials integer-valued on prime
numbers, in Factorization in Integral Domains. Ed. D. D. Anderson. Vol. 89, Lecture Notes in Pure and
Appl. Math., Dekker, New York, 1997. 271284.
19. J.-L. Chabert, A.-H. Fan, Y. Fares, Minimal dynamical systems on a discrete valuation domain, Discrete
Contin. Dyn. Syst. 35 (2009) 777795.
20. J. L. Chabert, G. Gerboud, Polynomes a valeurs entieres et binomes de Fermat, Canad. J. Math. 45 (1993)
621.
21. Z. Chen, On polynomial functions from Zn to Zm , Discrete Math. 137 (1995) 137145.
22. J. Dieudonne, Sur les fonctions continues p-adiques, Bull. Sci. Math. 68 (1944) 7995.
23. Y. Fares, v-ordering sequences and countable sets, in Commutative Algebra and Its Applications. Ed.
M. Fontana, S.-E. Kabbaj, and B. Olberding, I. Swanson. De Gruyter, New York, 2009. 239246.
24. G. Gerboud, Polynomes a valeurs entieres sur lanneau des entiers de Gauss, C. R. Acad. Sc. Paris, Ser.
A 307 (1988) 375378.
25. H. Gunji, D. L. McQuillan, Polynomials with integral values, Proc. Roy. Irish Acad. A78 (1978) 17.
26. F. Halter-Koch, W. Narkiewicz, Finiteness properties of polynomial mappings, Monatsh. Math. 114
(1992) 107110.
27. K. Hensel, Uber den grossten gemeinsamen Theiler aller Zahlen, welche durch eine ganze Funktion von
n Veranderlichen darstellbar sind, J. reine angew. Math. 116 (1896) 350356.
28. K. Johnson, Limits of characteristic sequences of integer-valued polynomials on homogeneous sets,
J. Number Theory 129 (2009) 29332942.
29. , P-orderings of noncommutative rings, Proc. Amer. Math. Soc. 143 (2015) 32653279.
30. A. J. Kempner, Polynomials and their residue systems, Amer. Math. Soc. Trans. 22 (1921) 240288.
31. K. Mahler, An interpolation series for continuous functions of a p-adic variable, J. Reine Angew. Math.
199 (1958) 2334 and 208 (1961) 7072.
32. A. Ostrowski, Ueber ganzwertige Polynome in algebraischen Zahlkorpern, J. Reine Angew. Math. 149
(1919) 117124.
33. G. Polya, Ueber ganzwertige Polynome in algebraischen Zahlkorpern, Rend. Circ. Mat. Palermo 40
(1915) 116.
34. , Ueber ganzwertige Polynome in algebraischen Zahlkorpern, J. Reine Angew. Math. 149 (1919)
97116.
35. D. Singmaster, On polynomial functions (mod n), J. Number Theory 6 (1974) 345352.
36. M. Wood, P-orderings: A metric viewpoint and the non-existence of simultaneous orderings, J. Number
Theory 99 (2003) 3656.
37. J. Yeramian, Anneaux de Bhargava, Comm. Algebra 32 (2004) 30433069.

PAUL-JEAN CAHEN received his Ph.D. in 1973 at the University Paris XI-Orsay with Pierre Samuel as
adviser. He is currently retired.
12 Traverse du lavoir de grand mere, 13100 Aix-en-Provence, France
pauljean.cahen@gmail.com
JEAN-LUC CHABERT received his Ph.D. in 1973 at the University Paris XI-Orsay with Pierre Samuel as
adviser. He is currently professor emeritus of the University of Picardie.
Department of Mathematics, LAMFA CNRS UMR 7352, 33 Rue Saint-Leu, 80039 Amiens, France
jean-luc.chabert@u-picardie.fr

KIRAN S. KEDLAYA received his Ph.D. in 2000 at MIT with Johan de Jong as adviser. He is currently a
professor at the University of California, San Diego.
Department of Mathematics, University of California, San Diego, 9500 Gilman Drive # 0112, La Jolla, CA
92093
kedlaya@ucsd.edu

790 
c THE MATHEMATICAL ASSOCIATION OF AMERICA [Monthly 124

This content downloaded from 128.135.98.165 on Fri, 17 Nov 2017 02:19:44 UTC
All use subject to http://about.jstor.org/terms

Potrebbero piacerti anche