The American Mathematical Monthly

ISSN: 0002-9890 (Print) 1930-0972 (Online) Journal homepage: https://www.tandfonline.com/loi/uamm20

An Interesting Family of Symmetric Polynomials

Jennifer Brooks

To cite this article: Jennifer Brooks (2019) An Interesting Family of Symmetric Polynomials, The
American Mathematical Monthly, 126:6, 527-540, DOI: 10.1080/00029890.2019.1584514

To link to this article: https://doi.org/10.1080/00029890.2019.1584514

Published online: 29 May 2019.

Submit your article to this journal

Article views: 321

View Crossmark data

Full Terms & Conditions of access and use can be found at

https://www.tandfonline.com/action/journalInformation?journalCode=uamm20
 An Interesting Family of Symmetric
Polynomials
Jennifer Brooks

Abstract. We discuss two natural extremal problems for homogeneous polynomials. These
problems have simple solutions for polynomials in one or two variables but become interesting
for polynomials in three or more variables. We introduce a family of homogeneous symmet-
ric polynomials in three variables that solve one of these problems and have a number of
other interesting properties. For example, their coefficients are integers that can be expressed
as sums of binomial coefficients and possess a certain divisibility property. Furthermore,
these polynomials are connected in a simple way to a family of polynomials arising as sharp
examples in the study of proper polynomial mappings between balls in complex Euclidean
space.

1. INTRODUCTION AND STATEMENT OF RESULTS. Many beautiful results

in mathematics illustrate the heuristic principle that those objects satisfying some
extremal condition possess additional special properties. We encounter an elementary
illustration of this principle in calculus when we observe that, among all rectangles of
a fixed perimeter, the one of maximum area is a square. The isoperimetric inequality
gives a similar but less trivial example. It states that the length L of a closed curve in
the plane and the enclosed area A satisfy 4πA ≤ L2 . The extremal curve, for which
equality is achieved, is the circle. We also find many illustrations of this principle in
extremal graph theory. For example, given a fixed number n of vertices, the graph with
the maximum number of edges and no clique of size r + 1 is the lovely Turán graph
T (n, r).
In this article, we explore the properties of a family of symmetric polynomials that
arise as solutions to a certain simple extremal problem. Perhaps not surprisingly, in
light of the discussion above, their coefficients possess a number of beautiful combi-
natorial and number-theoretic properties. In order to state the extremal problem, we
make several definitions. A polynomial p is said to be symmetric if it is invariant
under any permutation of the variables. For example, p1 (x, y) = x 2 + xy + y 2 is a
symmetric polynomial in two variables because p1 (y, x) = p1 (x, y), but p2 (x, y) =
x 2 + xy − y 2 is not symmetric.
Let s be the elementary symmetric polynomial of degree 1 in n variables. Thus for
n = 2, s(x, y) = x + y and for n = 3, s(x, y, z) = x + y + z. For k a nonnegative
integer, let Vk denote the vector space of homogeneous polynomials of degree k in n
variables with complex coefficients, together with the zero polynomial. We say q ∈ Vk
is full if every monomial of degree k in n variables appears in q with nonzero coeffi-
cient. For example, if n = 2, then q1 (x, y) = x 3 − x 2 y + xy 2 − y 3 is a full polynomial
of degree 3, whereas q2 (x, y) = x 3 + y 3 is not full.
For any polynomial p, the rank of p, denoted by R(p), is the number of distinct
monomials appearing in p with nonzero coefficient. We consider the following basic
question about homogeneous polynomials:
doi.org/10.1080/00029890.2019.1584514
MSC: Primary 32H35, Secondary 11C08

June–July 2019] SYMMETRIC POLYNOMIALS 527

Question 1. Among all homogeneous symmetric polynomials p of degree m with p =
sq for full quotient q, what is the minimum possible rank? What are the polynomials
with this minimum rank?
We call such polynomials extreme symmetric polynomials. Question 1 is related to
an even simpler question (which, itself, has connections to an important problem in
several complex variables; see Section 2):
Question 2. Among all homogeneous polynomials p of degree m with p = sq for full
quotient q, what is the minimum possible rank? What are the polynomials with this
minimum rank?
We call such polynomials sharp polynomials.
For polynomials in one variable, these questions are uninteresting because any
nonzero homogeneous polynomial of degree m is just a multiple of x m and thus has
rank 1. For polynomials in two variables, both Questions 1 and 2 have simple answers.
For any degree m ≥ 1, there exists a homogeneous polynomial p with p = sq, with
full quotient q, and with only two terms. Indeed, if
p(x, y) = x m + (−1)m−1 y m ,

then
p(x, y)
q(x, y) = = x m−1 − x m−2 y + · · · + (−1)m−1 y m−1 ,
x+y

which is full. Thus p is a sharp polynomial in two variables. When m is odd, p is

symmetric, and it is thus also an extreme symmetric polynomial. When m is even, p is
not symmetric, and, in fact, it is easy to see that there is no homogeneous symmetric
polynomial divisible by x + y having only two terms. We can, however, find such a
polynomial with three terms; for m = 2r,
p(x, y) = x 2r + 2(−1)r−1 x r y r + y 2r

is such a polynomial. One verifies that its quotient when divided by x + y is


r−1
(x r + (−1)r−1 y r ) (−1)j x r−1−j y j ,
j =0

which is full. Thus for m even, the extreme symmetric polynomials in two variables
have three terms.
In this article, we explore the above questions for polynomials in three variables.
As far as we know, the questions have not been considered for polynomials in more
than three variables. In Section 2, we describe previous research that has identified a
family {Fm } of sharp polynomials with surprising properties and with connections to
questions in several complex variables. We also describe a new family {Sm } of extreme
symmetric polynomials (one for each positive degree m) in three variables. The first
few odd-degree members of the family are
S1 = x + y + z
S3 = x 3 + y 3 + z3 − 3xyz
S5 = x 5 + y 5 + z5 + 5xyz(xy + xz + yz)

528 
c THE MATHEMATICAL ASSOCIATION OF AMERICA [Monthly 126
 S7 = x 7 + y 7 + z7 − 7xyz(x 2 y 2 + x 2 z2 + y 2 z2 )
S9 = x 9 + y 9 + z9 + 9xyz(x 3 y 3 + x 3 z3 + y 3 z3 ) − 30x 3 y 3 z3
S11 = x 11 + y 11 + z11 − 11xyz(x 4 y 4 + x 4 z4 + y 4 z4 )
+ 55x 3 y 3 z3 (xy + xz + yz)
S13 = x 13 + y 13 + z13 + 13xyz(x 5 y 5 + x 5 z5 + y 5 z5 )
− 91x 3 y 3 z3 (x 2 y 2 + x 2 z2 + y 2 z2 ),

and the first few even-degree members are

S2 = x 2 + y 2 + z2 + 2(xy + xz + yz)
S4 = x 4 + y 4 + z4 − 2(x 2 y 2 + x 2 z2 + y 2 z2 )
S6 = x 6 + y 6 + z6 + 2(x 3 y 3 + x 3 z3 + y 3 z3 ) − 9x 2 y 2 z2
S8 = x 8 + y 8 + z8 − 2(x 4 y 4 + x 4 z4 + y 4 z4 ) + 16x 2 y 2 z2 (xy + xz + yz).

We give general formulas for the Sm below. In addition to this symbolic represen-
tation of these polynomials, we can give a visual representation by using a Newton
diagram. Although it is possible to construct a Newton diagram for a homogeneous
polynomial in any number of variables, we will only consider diagrams for polynomi-
als in three variables. The Newton diagram for a homogeneous polynomial g(x, y, z)
is a graph with one vertex for each monomial appearing in g with nonzero coefficient.
We introduce an edge between the vertices associated with two monomials m1 and m2
if there are two unequal degree one monomials λ1 and λ2 such that λ1 m1 = λ2 m2 . We
illustrate by constructing the Newton diagram for the degree three homogeneous poly-
nomial g(x, y, z) = x 3 + 3x 2 y − 2x 2 z + xyz − z3 . The Newton diagram will have
five vertices because there are five nonzero coefficients; see Figure 1.

Figure 1. Newton diagram for g(x, y, z).

We find the Newton diagrams easier to read if we simply label each vertex with the
coefficient of the corresponding term and if we superimpose the diagram on a dashed
grid that shows all possible monomials of the degree of our homogeneous polynomial.
We always orient the diagram so that higher rows correspond to higher powers of z,
higher powers of x are to the left, and higher powers of y are to the right. Figure 2
shows the Newton diagram for g again, incorporating these conventions.

June–July 2019] SYMMETRIC POLYNOMIALS 529

 Figure 2. Modified Newton diagram for g(x, y, z).

The Newton diagrams for the extreme symmetric polynomials are particularly
lovely. These diagrams are quite sparse, of course, because the extreme symmetric
polynomials are, by definition, the symmetric polynomials of minimum rank divisible
by x + y + z with full quotient. Figures 3, 4, and 5 show the Newton diagrams for S3 ,
S5 , and S7 .

Figure 3. Newton diagram for S3 .

Our main theorem, of course, states that the polynomials Sm satisfy one or both of
the extremal conditions.

Theorem 1. If Sm is defined as in (2)–(5), then Sm is an extreme symmetric polynomial.

Furthermore, if m ≡ 1, 3 mod 6, then Sm is a sharp polynomial as well.

Remark 1. If m is not congruent to 1 or 3 mod 6, then Sm is not sharp. In these

cases, the rank of Sm exceeds the rank of sharp polynomials of that degree by either
one or two.

The polynomials Sm possess other interesting properties. Their coefficients are inte-
gers and can be expressed simply in terms of binomial coefficients. One of the most
interesting properties is number-theoretic in nature.

530 
c THE MATHEMATICAL ASSOCIATION OF AMERICA [Monthly 126
 Figure 4. Newton diagram for S5 .

Figure 5. Newton diagram for S7 .

Proposition 1. Let Zm [x, y, z] be the polynomial ring in three variables over Zm , the
ring of integers modulo m. Then Sm (x, y, z) = x m + y m + zm in Zm [x, y, z] if and
only if m is prime.

In other words, with the exception of the coefficients of the pure terms, all coeffi-
cients of Sm are divisible by m if and only if m is prime. We are led to this proposition

June–July 2019] SYMMETRIC POLYNOMIALS 531

partly by examining the polynomials Sm listed above. We have, however, a deeper
reason to suspect that this proposition is true; the family {Fm } of sharp polynomials
mentioned above satisfies an analogue of Proposition 1. We will establish an explicit
relationship between our family {Sm } of extreme symmetric polynomials and the fam-
ily {Fm } of sharp polynomials.
Before we can explore this connection, we must define Sm for general m. Let r =
 m2 . Set Km,0 = 1, and set
   
m−k m−1−k
Km,k = + for 1 ≤ k ≤ r. (1)
k k−1

The form of Sm depends on the congruence class modulo 6 of the degree m. Suppose
first that m ≡ 1 mod 6 or m ≡ 5 mod 6. In the first case, we may write m = 6L + 1
for some nonnegative integer L and in the second case we may write m = 6L − 1 for
some positive integer L. In either case, set

Sm = x m + y m + z m

L−1
+ (−1)r−j Km,r−j (xyz)2j +1 [(xy)r−3j −1 + (xz)r−3j −1 + (yz)r−3j −1 ]. (2)
j =0

If m ≡ 3 mod 6, we write m = 6L + 3 and set

Sm = right-hand side of (2) + (−1)r−L Km,r−L (xyz)2L+1 . (3)

Similarly, if m ≡ 4 mod 6, we write m = 6L − 2, and if m ≡ 2 mod 6, we write

m = 6L − 4. In either case, we set

Sm = x m + y m + z m

L−1
− (−1)r−j Km,r−j (xyz)2j [(xy)r−3j + (xz)r−3j + (yz)r−3j ]. (4)
j =0

Finally, if m ≡ 0 mod 6, we write m = 6L and set

Sm = right-hand side of (4) − (−1)r−L Km,r−L (xyz)2L . (5)

2. CONNECTION WITH SEVERAL COMPLEX VARIABLES. Because Ques-

tions 1 and 2 seem purely algebraic, it is surprising that the sharp polynomials {Fm }
were discovered in the course of exploring an important and long-standing open prob-
lem in the theory of functions of several complex variables. In this section, we describe
this connection and state the properties of sharp polynomials. J. D’Angelo is primarily
credited with the discovery of these polynomials and their properties (see [1] and [2]).
A continuous mapping f : X → Y between topological spaces is proper if f −1 (K)
is compact in X whenever K is compact in Y . An important problem is to understand
proper holomorphic mappings between balls in complex Euclidean spaces of different
dimensions. We focus here on proper polynomial mappings. Let Bn denote the unit
ball in Cn and consider a polynomial mapping p = (p1 , . . . , pN ) : Cn → CN . Then p
maps Bn to BN properly if and only if it maps the unit sphere in Cn to the unit sphere

532 
c THE MATHEMATICAL ASSOCIATION OF AMERICA [Monthly 126
in CN , i.e., if and only if


N 
n
p(z) 2
= |pk (z)|2 = 1 whenever z 2
= |zj |2 = 1. (6)
k=1 j =1

Example 1. Consider the mapping from C2 to C3 given by

√
(z, w) → (z3 , 3zw, w3 ).

If |z|2 + |w|2 = 1, then

√
|z3 |2 + | 3zw|2 + |w3 |2 = |z3 |2 + 3|zw|2 (|z|2 + |w|2 ) + |w3 |2
= |z|6 + 3|z|4 |w|2 + 3|z|2 |w|4 + |w|6
= (|z|2 + |w|2 )3 = 1,

and thus this mapping is a proper polynomial mapping from B2 to B3 .

We further focus on the case n = 2. The complexity of a proper polynomial map-

ping p : B2 → BN is connected to the target dimension N; a conjecture of D’Angelo
[2] states that the degree m of such a proper mapping must satisfy

m ≤ 2N − 3. (7)

D’Angelo, Kos, and Riehl [3] prove this inequality in the case in which all components
of p are monomials and provide a family of monomial mappings for which m = 2N −
3. For each odd m, the squared norm of their sharp mapping is
(m−1)/2

2 m 2 m
p(z, w) 2
= (|z| ) + (|w| ) + Km,k (|z|2 )m−2k (|w|2 )k , (8)
k=1

where the coefficients Km,k are the same as those defined in (1). Replacing (|z|2 , |w|2 )
with (x, y) gives a polynomial in two real variables with nonnegative coefficients, with
value 1 on the line x + y = 1, and with rank N = m+3 2
. Clearly there is a bijection
between the class of proper monomial mappings from B2 to the unit ball in some CN
and the class P of polynomials in two variables with nonnegative coefficients taking
value 1 when x + y = 1.
We modify (8) to define a family {fm } of polynomials of both even and odd degrees,
although only those of odd degree are in class P . We set
m
2

m m
fm (x, y) = x − (−y) + Km,k x m−2k y k . (9)
k=1

Note that each fm is invariant under the map (x, y) → (ηx, η2 y) for η an mth root of
unity. Our goal is to use the family {fm } to define a family {Fm } of sharp polynomials.
Lebl and Peters [5] have made some important contributions to the study of proper
monomial mappings between balls. They show that when n = 2, the degree estimate
(7) holds for a class of polynomials with class P as a proper subset. They begin by con-
sidering a projectivized version of the problem; for each polynomial p ∈ P of degree

June–July 2019] SYMMETRIC POLYNOMIALS 533

m, we homogenize p(x, y) − 1 with z and then replace z with −z to obtain a homoge-
neous polynomial P (x, y, z) of degree m such that P (x, y, z) = q(x, y, z)(x + y + z)
for a quotient polynomial q that is homogeneous of degree m − 1. The condition that
p has nonnegative coefficients becomes a rather awkward condition on the signs of the
coefficients of P . Lebl and Peters show, however, that the degree estimate R(P ) ≥ m+5
2
can be proved under weaker hypotheses than that P arises from an element of P . Their
first hypothesis says simply that the terms of P have no common monomial factor of
positive degree. Their second hypothesis involves the Newton diagram of the quotient
q. They prove the following theorem.

Theorem 2 (Lebl and Peters [5]). Let P (x, y, z) be a homogeneous polynomial of

degree m such that P (x, y, z) = q(x, y, z)(x + y + z) for a homogeneous polynomial
q of degree m − 1. Suppose that the terms of P have no common monomial factor of
positive degree and that the Newton diagram of q is a connected graph. Then

m+5
R(P ) ≥ ,
2
and the inequality is sharp.

We obtain a family {Fm } for which equality holds by homogenizing fm (x, y) − 1

with z and then replacing z with −z, as above:
m
2

m m m
Fm (x, y, z) = x − (−y) − (−z) + (−1)k Km,k x m−2k y k zk . (10)
k=1

The polynomials fm and Fm have many interesting properties. We collect those we

will use in the following proposition.

Proposition 2 (Properties of fm and Fm ). Let fm and Fm be defined as in (9) and

(10). Thus Fm (x, y, −1) + 1 = fm (x, y).
1. fm (x, y) is the unique polynomial satisfying the following:
(a) fm (0, 0) = 0;
(b) fm (x, y) = 1 when x + y = 1;
(c) fm has degree m;
(d) fm (ηx, η2 y) = fm (x, y) for η a primitive mth root of unity.
2. Fm (x, y, z) is divisible by x + y + z, and the quotient is a full polynomial.
3. If m is odd, R(Fm ) = m+5 . Thus Fm is a sharp polynomial.
2
 
4. The coefficients Km,k are integers. If m is prime, for 1 ≤ k ≤ m2 , Km,k ≡ 0
mod m, whereas if m is not prime, there exists a k ≥ 1 for which Km,k is not
congruent to 0 modulo m. That is, Fm (x, y, z) = x m + y m + zm in Zm [x, y, z] if
and only if m is prime.

Proof. Part (1) is the special case of Proposition 2.5 in [1] with q = 2. The first state-
ment in (2) simply restates property (b) in (1). Part (3) follows from the results proved
in [3] since Fm (x, y, z) is obtained by homogenizing fm (x, y) − 1 and R(fm ) = m+3 2
.
Part (4) is essentially Corollary 2.8 in [1].

534 
c THE MATHEMATICAL ASSOCIATION OF AMERICA [Monthly 126
 It thus remains only to prove that Fm has full quotient. Although this fact about Fm
seems to be known, no proof appears in the literature. We claim that the quotient is
min{m−1−j,j −1}  

m−1  m − 1 − j m−1−k−j  j k 
j
Qm = (−1) x z y + y j zk
j =1 k=0
k
m
2
  
m − 1 − j m−1−2j j j
j
+ (−1) x y z , (11)
j =0
j

where, by convention, 01 = 0. One must verify that Qm (x, y, z)(x + y + z) indeed

equals Fm (x, y, z). This argument is similar to the argument below that the Sm have
full quotient and we omit it.

Our proof of Theorem 1 will show that, if m ≡ 1, 3 mod 6, then R(Sm ) = m+5 2
.
Therefore it follows from Proposition 2 that, for such m, both Fm and Sm are sharp
polynomials of degree m. Thus, in general, there is not a unique sharp polynomial
of each degree. Ours is not the first result bearing on the question of uniqueness;
D’Angelo and Lebl [4] address the question of whether fm (x, y) is the unique polyno-
mial in two variables of degree m in class P satisfying (7). This question is equivalent
to asking whether, for each degree m, Fm (x, y, z) is the unique polynomial divisible
by x + y + z with rank m+5 2
for which Fm (x, y, −1) + 1 has positive coefficients. Even
with this additional restriction, they prove that, for infinitely many m, uniqueness fails.
(They do not, however, address the question of whether other sharp polynomials have
full quotient.)
D’Angelo and Lebl give a concrete method for constructing new sharp examples
from the fm by replacing expressions in fm with expressions congruent modulo x +
y − 1. We use a similar process to go from Fm to Sm . A key congruence that we will
use repeatedly comes immediately
 from the statement that Fm (x, y, z) ≡ 0 mod (x +
y + z). Suppose r = m2 . Then

r
Gm (x, y, z) := x m + (−1)k Km,k x m−2k y k zk
k=1

≡ (−1) [y m + zm ]
m
mod (x + y + z). (12)

Proposition 3 describes in all cases how to use the fundamental congruence (12) to
obtain Sm from Fm . Because it is nonobvious and notation intensive, we illustrate the
process with an example.
Example 2. Take m = 13. Then
F13 = x 13 + y 13 + z13 − 13x 11 yz + 65x 9 y 2 z2 − 156x 7 y 3 z3 + 182x 5 y 4 z4
− 91x 3 y 5 z5 + 13xy 6 z6 .

Observe that
156x 7 y 3 z3 = 65x 7 y 3 z3 + 91x 7 y 3 z3 .

Making this substitution, reordering terms, and factoring gives

F13 = x 13 + y 13 + z13 − 13x 11 yz + 65x 9 y 2 z2 − 65x 7 y 3 z3 + 13xy 6 z6
− 91x 7 y 3 z3 + 182x 5 y 4 z4 − 91x 3 y 5 z5

June–July 2019] SYMMETRIC POLYNOMIALS 535

 = x 13 + y 13 + z13 + 13xyz[x 5 (−x 5 + 5x 3 yz − 5xy 2 z2 ) + y 5 z5 ]
− 91x 3 y 3 z3 [x 2 (x 2 − 2yz) + y 2 z2 ].

We recognize that −x 5 + 5x 3 yz − 5xy 2 z2 = −G5 and that x 2 − 2yz = G2 . We may

therefore use the fundamental congruence (12) to obtain

F13 = x 13 + y 13 + z13 + 13xyz[x 5 (−G5 ) + y 5 z5 ] − 91x 3 y 3 z3 [x 2 (G2 ) + y 2 z2 ]

≡ x 13 + y 13 + z13 + 13xyz[x 5 (y 5 + z5 ) + y 5 z5 ]
− 91x 3 y 3 z3 [x 2 (y 2 + z2 ) + y 2 z2 ] mod (x + y + z)
= S13 .
m
Proposition 3. Let Gm = Fm + (−y)m + (−z)m for m ≥ 1 and let r = 2
.
1. If m = 6L + 1 or m = 6L − 1,

Fm = x m + y m + z m

L−1
+ (−1)r−j Km,r−j (xyz)2j +1 [(−x)r−3j −1 Gr−3j −1 + (yz)r−3j −1 ], (13)
j =0

whereas if m = 6L + 3,

Fm = right-hand side of (13) + (−1)r−L Km,r−L (xyz)2L+1 . (14)

We obtain Sm by replacing (−1)r−3j −1 Gr−3j −1 with the expression y r−3j −1 +

zr−3j −1 , which is congruent to it modulo x + y + z.
2. If m = 6L − 2 or m = 6L − 4,

Fm = −x m − y m − zm

L−1
+ (−1)r−j Km,r−j (xyz)2j [(−x)r−3j Gr−3j + (yz)r−3j ], (15)
j =0

whereas if m = 6L,

Fm = right-hand side of (15) + (−1)r−L Km,r−L (xyz)2L . (16)

We obtain −Sm by replacing (−1)r−3j Gr−3j with the expression y r−3j + zr−3j ,
which is congruent to it modulo x + y + z.

We needn’t do any tedious algebra to prove the proposition; instead, we use the
characterization of fm given in part (1) of Proposition 2.

Proof. Suppose m is odd. The argument for even m is almost identical and is omit-
ted. Replacing (−1)r−3j −1 Gr−3j −1 with the expression y r−3j −1 + zr−3j −1 clearly gives
the formula for Sm . Thus by the fundamental congruence (12), the expression on the
right-hand side of each of (13) and (14) above is congruent to Sm modulo x + y + z.
Consequently, each vanishes when x + y + z = 0. Hence if we replace z with −1 in

536 
c THE MATHEMATICAL ASSOCIATION OF AMERICA [Monthly 126
each and add the constant 1, we obtain a polynomial that takes the value 1 on the
line x + y = 1. The resulting polynomial also clearly vanishes at (0, 0) and is invari-
ant under (x, y) → (ηx, η2 y) for η a primitive mth root of unity. Thus it is equal to
fm (x, y) by Proposition 2, part (1). Subtracting 1, homogenizing, and replacing z with
−z gives the result.

3. PROOF OF THE MAIN THEOREM. It is apparent from the explicit formu-

las (2) and (3) that if m ≡ 1, 3 mod 6, then R(Sm ) = m+5 2
. Furthermore, since Sm
includes the monomials x m , y m , and zm , the terms of Sm have no common monomial
factor. Because a full polynomial has a connected Newton diagram, if we can show
that Sm = sqm for a full polynomial qm , then it will follow from the theorem of Lebl
and Peters (Theorem 2) that when m ≡ 1, 3 mod 6, Sm is a sharp polynomial. We
define the quotient polynomial qm explicitly, so that the remainder of the proof is a
simple verification that Sm is the resulting product polynomial.

Proof. Suppose first that m = 2r + 1. Let q2r+1 (x, y, z) = γ (a, b, c)x a y b zc , where
 
j j
γ (2r − j, k, j − k) = (−1) (17)
k

for 2r − j ≥ k ≥ j − k ≥ 0, with the same value for γ (σ (2r − j, k, j − k)) for any
other permutation σ (2r − j, k, j − k) of these numbers. As with the polynomials Sm ,
the structure of the quotient polynomials is perhaps better seen by looking at the New-
ton diagram than by looking at the formula. Figure 6 shows the Newton diagram for
q7 .

Figure 6. Newton diagram for q7 .

June–July 2019] SYMMETRIC POLYNOMIALS 537

 The proof is somewhat tedious and consists of demonstrating that the coefficients
of q2r+1 s agree with those of S2r+1 . Let α(A, B, C) be the coefficient of x A y B zC in
q2r+1 (x, y, z)s(x, y, z). Then A + B + C = 2r + 1 and

α(A, B, C) = γ (A − 1, B, C) + γ (A, B − 1, C) + γ (A, B, C − 1), (18)

where we take γ (a, b, c) = 0 if one of the arguments is negative. We must show that
the coefficients α(A, B, C) agree with the coefficients of the polynomial S2r+1 . In
principle, there are many cases to consider. We note, however, that q2r+1 and q2r+1 s
are symmetric. Thus we need only consider α(A, B, C) for A ≥ B ≥ C.

Case 1. If B = C = 0, then

α(2r + 1, 0, 0) = γ (2r, 0, 0) = 1,

which agrees with the coefficient of x 2r+1 in S2r+1 .

Case 2. Suppose next that A ≥ B > C = 0. Observe that, because A + B = 2r + 1,

we must in fact have A > B. Thus

α(A, B, 0) = γ (A − 1, B, 0) + γ (A, B − 1, 0) = (−1)B + (−1)B−1 = 0.

We have now dealt fully with the cases in which at least one of A, B, or C is zero.

Case 3. Next consider A > B ≥ C > 0. Write (A, B, C) = (2r + 1 − j, k, j − k)

for natural numbers j and k with j > k. Thus

α(2r + 1 − j, k, j − k)
= γ (2r − j, k, j − k) + γ (2r − (j − 1), k − 1, (j − 1) − (k − 1))
+ γ (2r − (j − 1), k, (j − 1) − k)
     
j j j −1 j − 1 j −1 j − 1
= (−1) + (−1) + (−1) = 0. (19)
k k−1 k

Case 4. We next treat the case in which A = B > C > 0. Write (A, B, C) =
(2r + 1 − j, 2r + 1 − j, 2j − 2r − 1) for some j . Thus

α(2r + 1 − j, 2r + 1 − j, 2j − 2r − 1)
= γ (2r − j, 2r − (j − 1), 2(j − 1) − 2r + 1)
+ γ (2r − (j − 1), 2r − j, 2(j − 1) − 2r + 1)
+ γ (2r − (j − 1), 2r + 1 − j, 2(j − 1) − 2r)
     
j −1 j − 1 j −1 j − 1 j −1 j −1
= (−1) + (−1) + (−1)
2r − j 2r − j 2r + 1 − j
   
j −1 j
= (−1)j −1 +
2r − j 2r + 1 − j
= (−1)j −1 K2r+1,2r+1−j .

538 
c THE MATHEMATICAL ASSOCIATION OF AMERICA [Monthly 126
Case 5. Finally, if 2r + 1 = 6L + 3, then A = B = C = 2L + 1 is possible. In this
case,

α(2L + 1, 2L + 1, 2L + 1)
= γ (2L, 2L + 1, 2L + 1) + γ (2L + 1, 2L, 2L + 1)
+ γ (2L + 1, 2L + 1, 2L)
= 3γ (2L + 1, 2L + 1, 2L)
= 3γ (2r − (4L + 1), 2L + 1, 2L)
 
4L+1 4L + 1
= 3(−1)
2L + 1
     
4L + 1 4L + 1 4L + 1
= (−1) 4L+1
+ +
2L + 1 2L + 1 2L
   
4L + 1 4L + 2
= (−1)4L+1 +
2L + 1 2L + 1
= (−1)4L+1 K6L+3,2L+1 .

Thus if r ≡ 0, 2 mod 3,

q2r+1 s
 4r+1
3 

=x 2r+1
+y 2r+1
+z 2r+1
+ (−1)j −1 K2r+1,2r+1−j
j =r+1
 2r+1−j 2(j −r)−1

· (xy) z + (xz)2r+1−j y 2(j −r)−1 + x 2(j −r)−1 (yz)2r+1−j
 r−2
3 

= x 2r+1 + y 2r+1 + z2r+1 + (−1)r+J K2r+1,r−J
J =0
 r−J 2J +1 r−J

· (xy) z + (xz) y 2J +1 + x 2J +1 (yz)r−J .

If 2r + 1 = 6L + 3, we get the last expression above together with the additional term
−K2r+1,2L+1 (xyz)2L+1 . Thus in either case q2r+1 s = S2r+1 , as claimed.
If m is even, we write m = 2r and let q2r (x, y, z) = γ (a, b, c)x a y b zc , with
 
j +1 j
γ (2r − 1 − j, k, j − k) = (−1)
k

for 2r − 1 − j ≥ k ≥ j − k ≥ 0, with the same value for γ (σ (2r − 1 − j, k, j − k))

for any other permutation σ (2r − 1 − j, k, j − k) of these numbers. The proof that
q2r s = S2r is analogous to the above and we omit it.

4. PROOF OF PROPOSITION 1. We turn now to Proposition 1. Since the coeffi-

cients of Sm are a subset of the coefficients of Fm , one direction follows immediately
from part (4) of Proposition 2. We may also give a direct proof of both directions using
the idea of D’Angelo’s proof.

June–July 2019] SYMMETRIC POLYNOMIALS 539

Proof. Suppose m = 6L + 1 or m = 6L − 1. The cases for m = 6L + 3 and for m
even are analogous. Taking z = −x − y gives

Sm (x, y, −(x + y)) + (x + y)m


L−1
= xm + ym + (−1)r+j +1 Km,r−j
j =0

· [xy(x + y)]2j +1 [(xy)r−1−3j + (−x − y)r−1−3j (x r−1−3j + y r−1−3j )]


L−1
= xm + ym + (−1)r+j +1 Km,r−j qj (x, y).
j =0

Because Sm (x, y, −(x + y)) = 0, we obtain the identity


L−1
(x + y)m = x m + y m + (−1)r+j +1 Km,r−j qj (x, y).
j =0

It is well known that, as elements of the vector space Zm [x, y], (x + y)m = x m + y m
if and only if m is prime. Therefore the same is true of the equivalent expression
x m + y m + (−1)r+j +1 Km,r−j qj (x, y). Because the degree of qj in x is 2(2j + 1) +
2(r − 1 − 3j ) = 2r − 2j , the qj have different degrees in x and are hence linearly
independent in Zm [x, y]. Since (−1)r+j +1 Km,r−j qj (x, y) is the zero polynomial if
and only if m is prime, we conclude that if m is prime, then Km,r−j ≡ 0 mod m for
all j , whereas if m is not prime, then there exists j for which Km,r−j is not congruent
to zero modulo m.

ACKNOWLEDGMENTS. The author acknowledges support from NSF grant DMS 1200815. The author
would also like to thank John D’Angelo for his encouragement.

REFERENCES

[1] D’Angelo, J. (2004). Number-theoretic properties of certain CR mappings. J. Geom. Anal. 14(2): 215–
229.
[2] D’Angelo, J. P. (1993). Several Complex Variables and the Geometry of Real Hypersurfaces. Boca Raton,
FL: CRC Press.
[3] D’Angelo, J., Kos, S., Riehl, E. (2003). A sharp bound for the degree of proper monomial mappings
between balls. J. Geom. Anal. 13(4): 581–593.
[4] D’Angelo, J., Lebl, J. (2009). Complexity results for CR mappings between spheres. Int. J. Math. 29(2):
149–166.
[5] Lebl, J., Peters, H. (2011). Polynomials constant on a hyperplane and CR maps of hyperquadrics. Mosc.
Math. J. 11(2): 285–315.

JENNIFER BROOKS earned her Ph.D. in Mathematics from the University of Wisconsin in 2005. In 2005,
she joined the faculty at the University of Montana. Her research interests include harmonic analysis and the
theory of functions of several complex variables.
Department of Mathematical Sciences, University of Montana, Missoula, MT 59812, USA
jennifer.brooks@mso.umt.edu

540 
c THE MATHEMATICAL ASSOCIATION OF AMERICA [Monthly 126