GENERIC SINGULARITIES ON LINEAR COMBINATIONS OF

PRESCRIBED MONOMIALS IN TWO VARIABLES

DARREN J. FINNIGAN, ALAA HAJ ALI, KYUNGYONG LEE, CHRIS M. LOCRICCHIO, THE MINH
TRAN, RAFAL URBANIUK

Abstract. One of the polynomial interpolation problems on the plane is to determine if

general multiple points impose independent conditions on plane algebraic curves with given
restriction(s). This restriction is classically given by a degree. In this paper we use a different
restriction, namely we fix a set of monomials and consider their linear combinations. We
study whether singular points in general position impose independent conditions on linear
combinations of monomials satisfying certain parity properties.

1. Introduction

Let K be an infinite field and R = K[x, y]. It is an important problem to determine if

general multiple points on A2K (= Spec(R)) impose independent conditions on plane curves of
a given degree. This problem has a very long history (for a very small fraction of numerous
references, see [1, 2, 3, 4, 5, 6, 8, 9]). In this paper we continue the characteristic 2 reduction
approach following the spirit of [7].
Before we state the main theorem, we need some notation. For integers i, j, a, b, we say
that (i, j) ≡ (a, b)(mod 2) if i ≡ a(mod 2) and j ≡ b(mod 2). For a, b ∈ {0, 1}, let
Ma,b := {(i, j) ∈ (Z≥0 )2 | (i, j) ≡ (a, b)(mod 2)}.
For any subset L ⊂ (Z≥0 )2 , define
L1 = M0,0 ∩ L;
L2 = M0,1 ∩ L;
L3 = M1,0 ∩ L;
L4 = M1,1 ∩ L.
Also, define xL := {xi y j | (i, j) ∈ L} and let Span(xL ) be the K-vector space spanned by
xL . If no confusion arises, a nonzero polynomial f in Span(xL ) will be identified with the
(possibly non-reduced or non-irreducible) algebraic curve defined by the R-ideal generated
by f . The following is our main theorem.
Theorem 1.1. Fix a positive integer n. Let L ⊂ (Z≥0 )2 be a set of lattice points with
|L| ≤ 3n. Suppose that |Li | ≤ n for all i ∈ {1, 2, 3, 4}. Then no curve in Span(xL ) passes
through n general points with multiplicity ≥ 2.

Applying this to plane curves with given degree, we obtain several corollaries.
1
2 FINNIGAN ET AL

Corollary 1.2. Fix positive integers d ≥ 2 and n. Let n0 = n + d 2d+53

e + 1, and let (mi )1≤i≤n0
be a sequence consisting of positive integers with mi = 2 for i ∈ {n + 1, ..., n0 }. Assume that
no curve of degree d passes through general points p1 , ..., pn with multiplicity ≥ mi at pi .
Then no curve of degree d + 2 passes through general points p1 , ..., pn0 with multiplicity ≥ mi
at pi .
Corollary 1.3. Let t, d and n be positive integers with t ≥ 3 and d ≥ t + 5. If
(d + 1)(d + 2)/2 ≤ 3n + t(t + 1)/2,
then there is no curve of degree d passing through n + 1 general points with multiplicity ≥ 2
at the first n points and with multiplicity ≥ t at the last point.

As another corollary, we give a new proof of the following theorem which is obtained
more than a century ago by Campbell [4], Palatini [8] and Terracini [9].
Corollary 1.4. Fix two positive integers n and d with (n, d) 6= (2, 2) and (n, d) 6= (5, 4).
Then (d + 1)(d + 2)/2 ≤ 3n if and only if no polynomial of degree d vanishes at n general
points with multiplicity ≥ 2.

2. Proof of the main theorem

In this section we prove Theorem 1.1, so we assume that |L| ≤ 3n and |Li | ≤ n for all
1 ≤ i ≤ 4. For two integers a and b, denote {a, a + 1, ..., b} by [a, b].
Let L ⊂ (Z≥0 )2 be a set of lattice points and (x1 , y1 ), ..., (xn , yn ) be n general points on
K 2 . A nonzero element f in Span(xL ) is of the form
X
ci,j xi y j ,
(i,j)∈L
∂f
and f vanishes at a point (xk , yk ) with multiplicity ≥ 2 if and only if f (xk , yk ) = ∂x
(xk , yk ) =
∂f
∂y
(xk , yk ) = 0.
To prove Theorem 1.1, it is enough to prove it for |L| = 3n. If |L| < 3n we can always
0
find a set of lattice points S ⊂ (Z≥0 )2 \ L where |S| = 3n − |L| and L := L ∪ S satisfies
0
|Li | ≤ n for every i ∈ [1, 4]. For the rest of this section, we will assume that |L| = 3n and
|Li | ≤ n for every i ∈ [1, 4]. Let L = {(i1 , j1 ), (i2 , j2 ), ..., (i3n , j3n )}. If the system of equations
 
xi11 y1j1 xi12 y1j2 ··· xi13n y1j3n
 i xi1 −1 y j1 i xi2 −1 y j2 · · · i xi3n −1 y j3n  
 1 1 1 2 1 1 3n 1 1  ci1 ,j1

 j xi1 y j1 −1 j xi2 y j2 −1 · · · j xi3n y j3n −1 
 1 1 1 2 1 1 3n 1 1   ci ,j 
.. .. .. ..  2 2 
=0

 . . . .  ..

 xi 1 y j 1
 .
 n n xin2 ynj2 ··· xin3n ynj3n 
 ci3n ,j3n
i −1 j i −1 j
 i1 xn yn i2 xn yn · · · i3n xn yn 
1 1 2 2 i 3n −1 j 3n

j1 xin1 ynj1 −1 j2 xin2 ynj2 −1 · · · j3n xin3n ynj3n −1

has no nontrivial solution, then no curve in Span(xL ) passes through n general points with
multiplicity≥ 2.
 GENERIC SINGULARITIES ON LINEAR COMBINATIONS OF MONOMIALS 3

We give a total order H on the set of lattice points as follows:

(0, 0) < (0, 1) < (1, 0) < (0, 2) < (1, 1) < (2, 0) < (0, 3) < (1, 2) < (2, 1) < (3, 0) < ...,
i.e. (i, j) < (i0 , j 0 ) if
(
i + j < i0 + j 0 , or
i + j = i0 + j 0 and i < i0 .

Notation. Let
 
xi11 y1j1 xi12 y1j2 ··· xi13n y1j3n

 i1 xi11 −1 y1j1 i2 xi12 −1 y1j2 · · · i3n xi13n −1 y1j3n 


 j1 xi11 y1j1 −1 j2 xi12 y1j2 −1 · · · j3n xi13n y1j3n −1 

L
D = det 
 .. .. .. .. .

 . . . . 

 xin1 ynj1 xin2 ynj2 ··· xin3n ynj3n 

 i1 xin1 −1 ynj1 i2 xin2 −1 ynj2 ··· i3n xin3n −1 ynj3n 
j1 xin1 ynj1 −1 j2 xin2 ynj2 −1 ··· j3n xin3n ynj3n −1
Also for every W = {(u1 , v1 ), (u2 , v2 ), (u3 , v3 )} ⊂ L and for i ∈ [1, n], we let
xui 1 yiv1 xui 2 yiv2 xui 3 yiv3
 

DiW = det  u1 xui 1 −1 yiv1 u2 xui 2 −1 yiv2 u3 xui 3 −1 yiv3  .

v1 xui 1 yiv1 −1 v2 xui 2 yiv2 −1 v3 xui 3 yiv3 −1
Definition 2.1. A triangle decomposition of L is a sequence T = (W (1) , ..., W (n) ) consisting
of subsets of L such that ∪1≤i≤n W (i) = L, W (i) ∩ W (j) = ∅ for all i 6= j, and |W (i) | = 3
for every i. Let TL be the collection of all triangle decompositions of L. For each T =
n
(i)
{W (1) , ..., W (n) } ∈ TL , let P(T ) = DiW .
Q
i=1

Lemma 2.2. The following recursive algorithm yields a (not necessarily unique) triangle
decomposition T0 = {W (1) , W (2) , ..., W (n) }.
• Step 0: set S = L and j = 1.
• Step 1: choose k ∈ [1, 4] so that |Sk | = min{|Si | : i ∈ [1, 4]}.
• Step 2: let W (j) = ∪i∈[1,4]\{k} {the smallest element in Si with respect to H}.
• Step 3: replace S with S \ W (j) and j with j + 1.
• Step 4: if S 6= ∅ go back to step 1; otherwise stop.

Proof. We use induction on n. It suffices to check that Si is nonempty for i ∈ [1, 4] \ {k} in
Step 2. For the base case n = 1, we have |S| = 3 and |Si | ≤ 1 for every i ∈ [1, 4]. So |Sk | = 0
and |Si | = 1 for i ∈ [1, 4] \ {k}.
Suppose that the algorithm works successfully for n = m. Let n = m + 1, then |L| =
3m + 3 and |Li | ≤ m + 1 for every i ∈ [1, 4]. But there exists i0 ∈ [1, 4] such that |Li0 | ≤ m,
because, if not, we would have |L| = 4(m + 1) > 3(m + 1) which is a contradiction. Thus,
4 FINNIGAN ET AL

after carrying Step 1,2,3 out, we have |S| = 3m and |Si | ≤ m for every i ∈ [1, 4]. Then by
the induction the rest of the algorithm works successfully. 

We want to show that P(T0 ) 6= 0 and P(T0 ) cannot be cancelled by (sum of some) P(T )
for all other T ∈ TL . Then we can conclude that DL 6= 0.
Lemma 2.3. Let a, b, c, d, e, f be integers and let
 a b 
x y xc y d xe y f
D = det axa−1 y b cxc−1 y d exe−1 y f  ,
bxa y b−1 dxc y d−1 f xe y f −1
which is equal to [(cf − de) − (af − be) + (ad − bc)]xa+c+e−1+ y b+d+f −1 .
Then (cf − de) − (af − be) + (ad − bc) 6≡ 0 (mod 2) if and only if (a, b) 6≡ (c, d) 6≡
(e, f ) and (a, b) 6≡ (e, f ) (mod 2). In particular if D 6≡ 0 (mod 2) then there exists a
unique (g, h) ∈ {(0, 0), (0, 1), (1, 0), (1, 1)} such that {(a, b), (c, d), (e, f )} ∩ Mg,h = ∅. In fact,
(g, h) ≡ (a + c + e, b + d + f ) (mod 2).

Proof. Straightforward. 
0 0 0 0 0 0 0
Lemma 2.4. Let W = {(u1 , v1 ), (u2 , v2 ), (u3 , v3 )} and W = {(u1 , v1 ), (u2 , v2 ), (u3 , v3 )} be
0 0 0
two distinct subsets of L, and write DiW = CiW xαi yiβ and DiW = CiW xγi yiδ with CiW , CiW ∈ Z.
0
Assume that CiW , CiW 6≡ 0 (mod 2). Let (a, b) and (c, d) ∈ {(0, 0), (0, 1), (1, 0), (1, 1)} such
0
that W ∩ Ma,b = W ∩ Mc,d = ∅. Then the following two statements hold.
(1) if (a, b) 6= (c, d) then α 6= γ or β 6= δ.
0 0
(2) if (ui , vi ) ≤ (ui , vi ) with respect to H for every i ∈ [1, 3], then α 6= γ or β 6= δ.
0 0 0
Proof. (1) We have α = u1 + u2 + u3 − 1, β = v1 + v2 + v3 − 1, γ = u1 + u2 + u3 − 1, and δ =
0 0 0
v1 + v2 + v3 − 1. Thus, by Lemma 2.3, we get α ≡ a − 1 (mod 2), β ≡ b − 1 (mod 2) and
γ ≡ c − 1 (mod 2), δ ≡ d − 1 (mod 2). Therefore the assertion follows.
0 0
(2) It is a consequence of definition of the total order H. More precisely, if ui + vi < ui + vi
3 3 3 3 3 3
P P 0 0 P P 0 P P 0
for some i, then (ui + vi ) < (ui + vi ) which implies that ui 6= ui or vi 6= vi .
i=1 i=1 i=1 i=1 i=1 i=1
3 3
0 0 P P 0
Thus α 6= γ or β 6= δ. If ui + vi = ui + vi for all i, then ui < ui hence α 6= γ. 
i=1 i=1

Proof of theorem 1.1. Let T0 = {W (1) , W (2) , ...W (n) } be a triangle decomposition obtained
from Lemma 2.2. In order to prove that DL 6= 0, it is sufficient to show that P(T0 ) 6= 0 and
P(T0 ) is not cancelled by (sum of some) P(T ) for all other T ∈ TL .
n n
(i) (i) (i)
DiW CiW xαi i yiβi with CiW
Q Q
Write P(T0 ) = = ∈ Z. By construction, we have
i=1 i=1
|W (i) ∩ Ma,b | ≤ 1 for any (a, b) ∈ {0, 1}2 and for any i ∈ {1, 2, .., n}. Thus, by Lemma 2.3,
(i)
CiW 6≡ 0 (mod 2) for any i ∈ {1, 2, .., n} and the coefficient of P(T0 ) is odd. In particular,
P(T0 ) 6= 0.
 GENERIC SINGULARITIES ON LINEAR COMBINATIONS OF MONOMIALS 5

n n
(i) (i)
For any T = {V (1) , V (2) , ...V (n) } ∈ TL with T 6= T0 , let P(T ) = DiV CiV xγi i yiδi
Q Q
=
i=1 i=1
and j the smallest integer such that W (j) 6= V (j) . Let W (j) = {(u1 , v1 ), (u2 , v2 ), (u3 , v3 )} and
0 0 0 0 0 0
V (j) = {(u1 , v1 ), (u2 , v2 ), (u3 , v3 )}.
(j)
If CjV ≡ 0 (mod 2), then the coefficient of P(T ) is even and it cannot cancel P(T0 ) . If
(j)
CjV 6≡ 0 (mod 2), then by Lemma 2.3, |V (j) ∩ Ms,t | ≤ 1 for any (s, t) ∈ {0, 1}2 . Then there
exist (a, b), (c, d) ∈ {0, 1}2 such that W (j) ∩ Ma,b = ∅ and V (j) ∩ Mc,d = ∅.
If (a, b) 6= (c, d) then by Lemma 2.4(1), we get αj 6= γj or βj 6= δj . Thus P(T ) cannot
cancel P(T0 ) . If (a, b) = (c, d) then we can assume that (ui , vi ) ≡ (u0i , vi0 ) (mod 2) for each
0 0
i ∈ [1, 3]. Then by the construction of P(T0 ) , we have (ui , vi ) ≤ (ui , vi ) for every i ∈ [1, 3].
Thus by Lemma 2.4(2), αj 6= γj or βj 6= δj , so P(T ) cannot cancel P(T0 ) .
P
Therefore P(T0 ) cannot be cancelled by P(T ) . 
T ∈TL ,T 6=T0

3. Proofs of Corollaries
0
Proof of Corollary 1.2. Let (x1 , y1 ),P ..., (xn , yn ), (xn+1 , yn+1 ), ..., (xn0 , yn0 ) be n general points.
00
Let n = d 2d+53
e + 1 and P (x, y) = i j
i+j≤d+2 ci,j x y a polynomial of degree d + 2. We have
00
a system of ni=1 mi2+1 + 3n homogeneous linear equations with (d+3)(d+4)/2 unknowns.
P

Let A be the matrix corresponding to this system of linear equations, N = {(i, j), i + j ≤ d}
and L = {(i, j), d + 1 ≤ i + j ≤ d + 2}. Also let B be the matrix formed by choos-
ing the rows corresponding to the first n points and the columns correspoonding to N ;
00
and C be the submatrix formed by choosing the rows corresponding to the last n points
and the columns corresponding to L. To show that our system of equations has no triv-
ial solution it is enough to show that both of the systems of homegeneous linear equations
associated with B and C have no trivial solutions. The claim is true for B by assump-
tion. For C, we try to use theorem 1.1. We have |L| = 2d + 5. On the other hand,
00 00
3n = 3d 2d+53
e + 3 = 2d + 5 − (2d + 5)(mod2) + 3 ≥ 2d + 5. Thus |L| ≤ 3n . To use theorem
00
1.1 we only need to show that |Li | ≤ n for every i ∈ [1, 4]. It is easy to show that

(d + 4)/2 if d is even
|L1 | =
(d + 3)/2 if d is odd;

(d + 2)/2 if d is even
|L2 | = |L3 | =
(d + 3)/2 if d is odd;

(d + 2)/2 if d is even
|L4 | =
(d + 1)/2 if d is odd;
00
Since by assumption we have (d + 4)/2 ≤ (2d + 5)/3; therefore, |Li | ≤ n for every i ∈
[1, 4]. 

Proof of Corollary 1.3. Let (x1 , y1 ), ..., (xn+1 , yn+1 ) P

be n + 1 general points.
As in the previous two corollaries we let P (x, y) = i+j≤d ci,j xi y j be a polynomial of degree
6 FINNIGAN ET AL

d. We have a system of 3n + t(t + 1)/2 homogeneous linear equations with (d+1)(d+2)/2

unknowns.
Let A be the matrix corresponding to this system of linear equations, N = {(i, j), i+j ≤ t−1}
and L = {(i, j), t ≤ i + j ≤ d}. Also let B be the matrix formed by choosing the rows
corresponding to the (n + 1)th point and the columns correspoonding to N ; and C be the
matrix formed by choosing the rows corresponding to the first n points and the columns
corresponding to L.
To proove that our system of equations has no trivial solution it is enough to show that
both of the systems of homogeneous linear equations corresponding to the matrices B and
C have no trivial solutions. The claim is true for B because there is no curve of degree t − 1
that passes through a point with multiplicity ≥ t.
Since |N | = t(t + 1)/2 and |L| = (d + 1)(d + 2)/2 − t(t + 1)/2 ≤ 3n, we try to use theorem
1.1 to proove our claim for the matrix C. We need to show that |Li | ≤ n for every i ∈ [1, 4].
As in the first corollary

(d + 2)(d + 4)/8 if d is even
|L1 ∪ N1 | =
(d − 1)(d + 1)/8 if dis odd;

(d(d + 2)/8 if d is even
|L2 ∪ N2 | = |L3 ∪ N3 | = |L4 ∪ N4 | =
(d + 1)(d + 3)/8 if dis odd;
and 
(t + 1)(t + 3)/8 if d is even
|N1 | =
(t − 2)t/8 if dis odd;

((t − 1)(t + 1)/8 if d is even
|N2 | = |N3 | = |N4 | =
t(t + 2)/8 if dis odd;
Since |Li | = |Li ∪ Ni | − |Ni | for every i ∈ [1, 4], it is suffices to show that the greatest possible
value of |Li ∪ Li | mince the smallest possible values of |Li | is ≤ n. That is it is suffices to
show that (d + 2)(d + 4)/8 − (t − 2)t/8 ≤ n p
(d + 1)(d + 2)/2 ≤ t(t + 1)/2 + 3n is equivalent to d ≤ ( −3 2
+ ( 14 + t2 + t + 6n)) .
p
To show that (d + 2)(d + 4)/8 − (t − 2)t/8 ≤ n is equivalent to show that d ≤ (−3 + (1 +
t2 − 2t + 8n)). p 1 2 p
It can be shown that if n ≥ t + 5, d ≤ ( −3 2
+ ( 4
+ t + t + 6n)) ≤ (−3 + (1 + t2 − 2t + 8n)).
But for d ≥ t + 4, (d + 1)(d + 2)/2 ≤ t(t + 1)/2 + 3n implies that t + 5 ≤ 53 n + 2. We notice
that in this case, n is necessarly ≥ 5. Because otherwise we would have t + 5 ≤ 5 i.e. t ≤ 0.
Thus n ≥ 5 and t + 5 ≤ 35 n + 2 ≤ n which prooves the claim. 

Proof of Corollary 1.4. A polynomial P (x, y) of degree d is of the form i+j≤d ci,j xi y j . Such
P
a polynomial vanishes at n general points with multiplicity ≥ 2 iff the system of 3n homo-
geneous linear equations
P (xk , yk ) = ∂P/∂x(xk , yk ) = ∂P/∂y(xk , yk ) = 0 for any 1 ≤ k ≤ n
has at least one solution.
The number of unknown in this system is the cardinality of the set {(i, j), i + j ≤ d}. This
number is equal to (d + 1)(d + 2)/2.
If (d + 1)(d + 2)/2 > 3n, the number of unknown is strictly greater than the number of
 GENERIC SINGULARITIES ON LINEAR COMBINATIONS OF MONOMIALS 7

equations; therefore, there are an infinitely many curves of degree d vanishing at n general
points with multiplicity ≥ 2.
If (d + 1)(d + 2)/2 ≤ 3n, Let L = {(i, j), i + j ≤ d}. Then |L| = (d + 1)(d + 2)/2 ≤ 3n. To
apply theorem 1.1 it suffices to show that |Li | ≤ n for every i ∈ [1, 4]. It is easy to show that

(d + 2)(d + 4)/8 if d is even
|L1 | =
(d − 1)(d + 1)/8 if d is odd;

(d(d + 2)/8 if d is even
|L2 | = |L3 | = |L4 | =
(d + 1)(d + 3)/8 if d is odd;
If d is even, |Li | ≤ |L1 | for every i ∈ [1, 4]. So it suffices
p 1 to proove that |L1 | ≤ n.
(d + 1)(d + 2)/2 ≤ 3n is equivalent to d ≤ ( −3 2
+ ( 4 + 6n)) .
p
To show that (d + 2)(d + 4)/8 ≤ n is equivalent p 1 to show that d
p ≤ (−3 + (1 + 8n)).
−3
It can be shown that if n ≥ 15 then ( 2 + ( 4 +6n)) ≤ (−3+ (1+8n)). therefore |L1 | ≤ n
in this case. p
If n < 15, the only possible even values of d that satisfies d ≤ ( −3 2
+ ( 14 + 6n)) are 2 and
4; so we still need to study these two cases.
For d = 2 , (d + 1)(d + 2)/2 ≤ 3n with n 6= 2 implies that n ≥ 3. But (d + 2)(d + 4)/8 = 3;
therefore |L1 | ≤ n in this case.
For d = 4, (d + 1)(d + 2)/2 ≤ 3n with n 6= 5 implies that n ≥ 6. But (d + 2)(d + 4)/8 = 6;
therefore |L1 | ≤ n in this case.
For d is odd, |Li | ≤ |L2 | for every i ∈ [1, 4]. So it is sufficient to show that |L2 | ≤ n.
This is true for n ≥ 15 because (d + 1)(d + 3)/8 < (d + 2)(d + 4)/8. So, we only need to
show that the claim is true for n < 15. p 1
If n < 15, the only possible odd values of d that satisfy d ≤ ( −3 2
+ ( 4 + 6n)) are 1 and 3.
For d = 3, (d + 1)(d + 2)/2 ≤ 3n implies that n ≥ 4. But (d + 1)(d + 3)/8 = 3 ≤ n; therefore
|L2 | ≤ n in this case.
For d = 1, (d + 1)(d + 3)/8 = 1 ≤ n for any n ≥ 1. therefore |L2 | ≤ n in this case.

References
[1] J. Alexander, Singularités imposables en position générale à une hypersurface projective, Compositio
Math. 68 (1988), no. 3, 305–354
[2] J. Alexander, A. Hirschowitz, Polynomial interpolation in several variables, J. Alg. Geom. 4 (1995),
no.2, 201–222
[3] J. Alexander, A. Hirschowitz, Generic Hypersurface Singularities, Proc. Indian Acad. Sci. Math. Sci.
107 (1997), no. 2, 139- 154
[4] J.E. Campbell, Note on the maximum number of arbitrary points which can be double points on a
curve, or surface, of any degree, The Messenger of Mathematics, XXI, 158–164 (1891–92)
[5] K. Chandler, A brief proof of a maximal rank theorem for generic double points in projective space,
Trans. Amer. Math. Soc. 353 (2001), no. 5, 1907–1920
[6] K. Chandler, Linear systems of cubics singular at general points of projective space, Compositio Math-
ematica 134 (2002), 269– 282
[7] K. Lee, Characteristic 2 approach to bivariate interpolation problems,
8 FINNIGAN ET AL

[8] F. Palatini, Sulla rappresentazione delle forme ternarie mediante la soma di potenze di forme lineari,
Rend. Accad. Lincei V, t.12 (1903), 378–384
[9] A. Terracini, Sulla rappresentazione delle coppie di forme ternarie mediante somme di potenze di forme
lineari, Annali Mat., 24 (1915), Selecta vol. I

Department of Mathematics, Wayne State University, Detroit, MI 48202