An on-line talk (June 19, 2020) Introduction to Combinatorial Number Theory (V) – Combinatorial Nullstellensatz and its Applications Zhi-Wei Sun Nanjing University Nanjing 210093, P. R. China [email protected]http://maths.nju.edu.cn/∼zwsun June 19, 2020
58
Embed
Introduction to Combinatorial Number Theory (V ...maths.nju.edu.cn › ~zwsun › CNT-talk5.pdf · An on-line talk (June 19, 2020) Introduction to Combinatorial Number Theory (V)
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
An on-line talk (June 19, 2020)
Introduction to Combinatorial Number Theory (V)
– Combinatorial Nullstellensatzand its Applications
In this talk we introduce Alon’s Combinatorial Nullstellensatz (i.e.,the so-called polynomial method), and its applications to Snevily’sconjectures and restricted sumsets.
2 / 58
Part I. Combinatorial Nullstellensatz and its Backgrounds
3 / 58
Cramer’s conjecture
Let n ∈ Z+ = {1, 2, 3, . . .}. Any cyclic group of order n isisomorphic to the additive group Zn = Z/nZ of residue classesmodulo n. If n is odd, then
1 + 1, 2 + 2, . . . , n + n
are pairwise incongruent modulo n and hence they form a completesystem of residues modulo n.Let a1, . . . , an ∈ Z. If a1 + 1, . . . , an + n form a complete system ofresidues modulo n, then
n∑i=1
(ai + i) ≡ 1 + · · ·+ n (mod n)
and hence∑n
i=1 ai ≡ 0 (mod n).
Cramer’s Conjecture. Let a1, . . . , an ∈ Z with n |∑n
i=1 ai . Thenthere is a permutation σ ∈ Sn such that aσ(1) + 1, . . . , aσ(n) + nform a complete system of residues mod n.
4 / 58
M. Hall’s Theorem
In 1952 M. Hall [Proc. Amer. Math. Soc.] obtained an extensionof Cramer’s conjecture.
M. Hall’s Theorem. Let G = {b1, . . . , bn} be an additive abeliangroup of order n, and let a1, . . . , an be elements of G witha1 + . . .+ an = 0. Then there exists a permutation σ ∈ Sn suchthat
{aσ(1) + b1, . . . , aσ(n) + bn} = G .
Remark. Hall used induction argument and his method is verytechnique. Up to now there are no other proofs of this theorem.
Observation. If a1, . . . , an ∈ Z are incongruent modulo n witha1 + · · ·+ an ≡ 0 (mod n), then n divides
0 + 1 + · · ·+ (n − 1) =n(n − 1)
2
and hence n is odd.5 / 58
A conjecture of Snevily
Snevily’s Conjecture for Abelian Groups [Amer. Math.Monthly, 1999]. Let G be an additive abelian group of odd order.Then for any two subsets A = {a1, . . . , ak} and B = {b1, . . . , bk}of G with |A| = |B| = k , there is a permutation σ ∈ Sk such thataσ(1) + b1, . . . , aσ(k) + bk are (pairwise) distinct.
Remark. The result does not hold for any group G of even order.In fact, there is an element g ∈ G of order 2, and A = B = {0, g}gives a counterexample.
Difficulty. No direct construction. Induction also does not work!
Snevily’s conjecture looks simple, beautiful and difficult!
6 / 58
Latin transversal
Let M be an n × n matrix. A line of M is a row or a column of M.M is called a Latin square over a set S of cardinality n if all itsentries come from the set S and no line of which contains anelement more than once. A transversal of the matrix M is acollection of n cells no two of which lie in the same line. A Latintransversal of M is a transversal whose cells contain no repeatedelement.
If G = {a1, . . . , an} is an additive group of order n, then thematrix M = (ai + aj)16i ,j6n formed by the Cayley addition table isa Latin square over G .
Another Form of Snevily’s Conjecture. Let G = {a1, . . . , aN}be an additive abelian group with |G | = N odd, and let M be theLatin square (ai + aj)16i ,j6N formed by the Cayley addition table.Then any n × n submatrix of M contains a Latin transversal.
7 / 58
Another Conjecture of Snevily
Snevily’s Conjecture on Addition modulo n [Amer. Math.Monthly, 1999]. Let 0 < k < n and a1, . . . , ak ∈ Z. Then thereexists π ∈ Sk such that a1 + π(1), . . . , ak + π(k) are distinctmodulo n.
Remark. A. E. Kezdy and H. S. Snevily [Combin. Probab.Comput. 2002] proved the conjecture for k 6 (n + 1)/2 and foundan application to tree embeddings.
8 / 58
Jager-Alon-Tarsi Conjecture
In 1982, motivated by his study of graph theory, F. Jager posedthe following conjecture in the case |F | = 5.
Jager-Alon-Tarsi Conjecture. Let F be a finite field with at least4 elements, and let A be an invertible n × n matrix with entries inF . There there exists a vector ~x ∈ F n such that both ~x and A~xhave no zero component.
In 1989 N. Alon and M. Tarsi [Combinarorica, 9(1989)] confirmedthe conjecture in the case when |F | is not a prime. Moreovertheir method resulted in the initial form of the CombinatorialNullstellensatz which was refined by Alon in 1999.
9 / 58
Sumsets with distinct summandsFor subsets A1, . . . ,An of an additive group G , we define
A1 u · · ·u An = {a1 + · · ·+ an : ai ∈ Ai , and ai 6= aj if i 6= j}.
When A1 = · · · = An = A, we denote A1 u · · ·u An by n∧A. Notethat
nA = (n − 1)A + A, but n∧A 6= (n − 1)∧A u A.
For A = [0, k − 1] = {0, 1, . . . , k − 1} and 0 < n 6 k, clearly
Let A be any finite subset Z. By construction, one can show
|n∧A| ≥ n|A| − n2 + 1.
M.B. Nathanson [Trans. AMS 347(1995)] proved that if2 6 n < |A| − 2 and |n∧A| = n|A| − n2 + 1 then A must be an AP.
10 / 58
Erdos-Heilbronn ConjectureErdos-Heilbronn Conjecture (1964). Let p be a prime and letA ⊆ Zp = Z/pZ. Then
|2∧A| > min{p, 2|A| − 3}.
Difficulty. Unlike Z, the field Zp has no suitable ordering. Directconstruction does not work! Also, Dyson’s g -transformation doesnot work for sumsets with distinct summands.
Dias da Silva-Hamidoune Theorem [Bull. London Math. Soc.,1994]. Let F be any field and let p(F ) be the additive order of themultiplicative identity of F . For any finite A ⊆ F , we have
|n∧A| > min{p(F ), n(|A| − n) + 1}.
Method: Exterior algebras and the representation theory ofsymmetric groups!
In 1995-1996 N. Alon, M. B. Nathanson and I. Z. Ruzsa were ableto prove this via the so-called polynomial method related toCombinatorial Nullstellensatz.
11 / 58
Erdos-Heilbronn ConjectureErdos-Heilbronn Conjecture (1964). Let p be a prime and letA ⊆ Zp = Z/pZ. Then
|2∧A| > min{p, 2|A| − 3}.
Difficulty. Unlike Z, the field Zp has no suitable ordering. Directconstruction does not work! Also, Dyson’s g -transformation doesnot work for sumsets with distinct summands.
Dias da Silva-Hamidoune Theorem [Bull. London Math. Soc.,1994]. Let F be any field and let p(F ) be the additive order of themultiplicative identity of F . For any finite A ⊆ F , we have
|n∧A| > min{p(F ), n(|A| − n) + 1}.
Method: Exterior algebras and the representation theory ofsymmetric groups!
In 1995-1996 N. Alon, M. B. Nathanson and I. Z. Ruzsa were ableto prove this via the so-called polynomial method related toCombinatorial Nullstellensatz.
12 / 58
Usual form of Alon’s Combinatorial Nullstellensatz
Usual Form of the Combinatorial Nullstellensatz (CN) [Alon,Combin. Probab. Comput. 8(1999)]. Let A1, . . . ,An be finitenonempty subsets of a field F and let f (x1, . . . , xn) ∈ F [x1, . . . , xn].Suppose that 0 6 ki < |Ai | for i = 1, . . . , n, k1 + · · ·+ kn = deg fand
[xk11 · · · x
knn ]f (x1, . . . , xn) (the coefficient of xk1
1 · · · xknn in f )
does not vanish. Then there are a1 ∈ A1, . . . , an ∈ An such thatf (a1, . . . , an) 6= 0.
Advantage: This advanced algebraic tool enables us to establishexistence via computation. It has many applications.
13 / 58
Strong form of the Combinatorial NullstellensatzStrong Form of the Combinatorial Nullstellensatz [Alon,Combin. Probab. Comput. 8(1999)]. Let A1, . . . ,An be finitenonempty subsets of a field F and let f (x1, . . . , xn) ∈ F [x1, . . . , xn].Set gi (x) =
∏a∈Ai
(x − a) for i = 1, . . . , n. Then
f (a1, . . . , an) = 0 for all a1 ∈ A1, . . . , an ∈ An
with deg hi 6 deg f − deg gi for i = 1, . . . , n, such that
f (x1, . . . , xn) =n∑
i=1
gi (xi )hi (x1, . . . , xn).
Remark: Let I be the ideal of F [x1, . . . , xn] generated byg1(x1), . . . , gn(xn). Then the strong form of CN tells us thatf (x1, . . . , xn) ∈ F [x1, . . . , xn] vanishes on Z (I ) = A1 × · · · × An ifand only if f ∈ I , where
Z (I ) = {(x1, . . . , xn) ∈ F n : P(x1, . . . , xn) = 0 for all P ∈ I}.14 / 58
Strong Form implies the Usual Form
Suppose that f vanishes on A1 × · · · × An. Then, by the StrongForm, we can write
f (x1, . . . , xn) =n∑
i=1
gi (xi )hi (x1, . . . , xn)
with hi (x1, . . . , xn) ∈ F [x1, . . . , xn] and deg hi 6 deg f − deg gi .Since k1 + · · ·+ kn = deg f and ki < |Ai | for i = 1, . . . , n, we have
[xk11 · · · x
knn ]f (x1, . . . , xn) =
n∑i=1
[xk11 · · · x
knn ]x
|Ai |i hi (x1, . . . , xn) = 0,
which contradicts the condition that the coefficient is nonzero.
15 / 58
A Lemma
Lemma [Alon, Nathanson and Ruzsa, Amer. Math. Monthly1995; J. Number Theory 1996] Let F be a field and A1, . . . ,An itssubsets which are finite and nonempty. Letf (x1, . . . , xn) ∈ F [x1, . . . , xn] have degree less than ki = |Ai | in xifor each i = 1, . . . , n. If f (a1, . . . , an) = 0 for alla1 ∈ A1, . . . , an ∈ An, then f (x1, . . . , xn) is identically zero.
Proof. The case n = 1 is easy since a nonzero polynomialP(x) ∈ F [x ] of degree less than a positive integer k can’t have kdistinct zeroes in F .
Let n > 1 and assume that the Lemma holds with n replaced byn − 1. Write f (x1, . . . , xn) =
∑kn−1i=0 fi (x1, . . . , xn−1)x i
n. For anya1 ∈ A1, . . . , an−1 ∈ An−1, as f (a1, . . . , an−1, xn) = 0 for allxn ∈ An we have fi (a1, . . . , an−1) = 0 for all i = 0, . . . , kn − 1. Bythe induction hypothesis, all the fi (x1, . . . , xn−1) are the zeropolynomial. So f (x1, . . . , xn) is also identically zero.
16 / 58
Proof of the Strong FormIf there are h1(x1, . . . , xn), . . . , hn(x1, . . . , xn) ∈ F [x1, . . . , xn] suchthat
f (x1, . . . , xn) =n∑
i=1
gi (xi )hi (x1, . . . , xn),
then for any a1 ∈ A1, . . . , an ∈ An we have
f (a1, . . . , an) =n∑
i=1
gi (ai )hi (a1, . . . , an) = 0.
Now we consider the converse. Write
f (x1, . . . , xn) =∑
j1,...,jn>0
fj1,...,jnx j11 . . . x
jnn
andx j = gi (x)qij(x) + r
(j)i (x),
where qij(x), r(j)i (x) ∈ F [x ] and deg r
(j)i (x) < deg gi (x) = |Ai |.
Note that both r(j)i (x) and gi (x)qij(x) = x j − r
(j)i (x) have degree
not exceeding j .17 / 58
Continue the ProofClearly
f (x1, . . . , xn) =∑
j1,...,jn>0j1+···+jn6deg f
fj1,...,jn
n∏i=1
(gi (xi )qiji (xi ) + r
(ji )i (xi )
)
=f (x1, . . . , xn) +n∑
i=1
gi (xi )hi (x1, . . . , xn),
where
f (x1, . . . , xn) =∑
j1,...,jn>0
fj1,...,jn
n∏i=1
r(ji )i (xi )
and each hi (x1, . . . , xn) is a suitable polynomial over F withdeg gi + deg hi 6 deg f . If a1 ∈ A1, . . . , an ∈ An, then
f (a1, . . . , an) =∑
j1,...,jn>0
fj1,...,jn
n∏i=1
ajii = f (a1, . . . , an) = 0.
As the degree of f (x1, . . . , xn) with respect to xi is smaller than|Ai |, by the Lemma the polynomial f (x1, . . . , xn) is identically zero. 18 / 58
Part II. Applications of Combinatorial Nullstellensatzto Snevily’s Conjectures
19 / 58
For σ ∈ Sk let sign(σ) be the sign of σ which takes 1 or −1according as the permutation σ is even or odd.For a matrix A = (aij)16i ,j6k over a field, its determinant andpermanent are given by
det A =∑σ∈Sk
sign(σ)k∏
i=1
ai ,σ(i) and perA =∑σ∈Sk
k∏i=1
ai ,σ(i).
Observe that
[xk−11 · · · xk−1
k ]∏
16i<j6k
(xj − xi )2
=[xk−11 · · · xk−1
k ](det(x i−1j )16i ,j6k)2
=[xk−11 · · · xk−1
k ]∑σ∈Sk
sign(σ)k∏
j=1
xσ(j)−1j
∑τ∈Sk
sign(τ)k∏
j=1
xτ(j)−1j
=∑σ∈Sk
sign(σ)sign(σ′) =∑σ∈Sk
(−1)(k2) = k!(−1)(k2)
where σ′(j) = k − σ(j) + 1 for j = 1, . . . , k . (For 1 6 i < j 6 k,we clearly have σ(i) > σ(j) ⇐⇒ σ′(i) < σ′(j).) 20 / 58
Attack Snevily’s conjecture on addition modulo n
A. E. Kezdy and H. S. Snevily [Combin. Probab. Comput.2002] Let k and n be positive integers with k 6 (n + 1)/2. Then,for any a1, . . . , ak ∈ Z, there exists π ∈ Sk such thata1 + π(1), . . . , ak + π(k) are distinct modulo n.
Proof. For xi , xj ∈ A = {1, . . . , k}, we have|xi − xj | 6 k − 1 6 n−1
2 < n2 , and
xi + ai 6≡ xj + aj (mod n)⇔ xj − xi 6≡ ai − aj (mod n)⇔ xj − xi 6= rij
where rij denotes the residue of ai − aj in the interval (−n/2, n/2].Thus, we only need to show that there are distinctx1, . . . , xk ∈ A = {1, . . . , k} such that xj − xi 6= rij for all1 6 i < j 6 k . By the Combinatorial Nullstellensatz for the realfield R, it suffices to note that
[xk−11 · · · xk−1
k ]∏
16i<j6k
(xj − xi )(xj − xi − rij)
=[xk−11 · · · xk−1
k ]∏
16i<j6k
(xj − xi )2 = k!(−1)(k2) 6= 0.
21 / 58
Alon’s contribution for cyclic groups of prime orders
Alon’s Result [Israel J. Math. 2000]. Let p be an odd prime andlet A = {a1, . . . , ak} be a subset of Zp with cardinality k < p.Given (not necessarily distinct) b1, . . . , bk ∈ Zp there is apermutation σ ∈ Sk such that aσ(1) + b1, . . . , aσ(k) + bk are(pairwise) distinct.
Remark. This result is slightly stronger than Snevily’s conjecturefor cyclic groups of prime order.
Proof. Let A1 = · · · = Ak = {a1, . . . , ak}. We need to show thatthere exist x1 ∈ A1, . . . , xk ∈ Ak such that∏
16i<j6k(xj − xi )(xj + bj − (xi + bi )) 6= 0. By the CombinatorialNullstellensatz for the filed Zp, it suffices to note that
[xk−11 · · · xk−1
k ]∏
16i<j6k
(xj − xi )(xj + bj − (xi + bi ))
=[xk−11 · · · xk−1
k ]∏
16i<j6k
(xj − xi )2 = k!(−1)(k2) 6= 0 (in Zp).
22 / 58
An extension of Alon’s result by Hou and Sun
Theorem. (Qing-Hu Hou and Z.-W. Sun [Acta Arith. 102(2002)])Let k > n > 1 be integers, and left F be a field whosecharacteristic is zero or greater than max{n, (k − n)n}. LetA1, . . . ,An be subsets of F with cardinality k, and letb1, . . . , bn ∈ F . Then the sumset
{a1 + . . .+ an : ai ∈ Ai , ai 6= aj and ai + bi 6= aj + bj if i 6= j}
have more than (k − n)n elements.
Actually, Hou and Sun proved a much more general resultincluding the above theorem as a special case.
23 / 58
Snevily’s Conjecture for cyclic groups
For odd composite number n, Zn = Z/nZ is not a field. How toprove Snevily’s conjecture for the cyclic group Zn?
Dasgupta, Karolyi, Serra and Szegedy [Israel J. Math., 2001]:Snevily’s conjecture holds for any cyclic group of odd order.
Their key observation is that a cyclic group of odd order n canbe viewed as a subgroup of the multiplicative group of thefinite field F2ϕ(n) . (Note that n divides 2ϕ(n) − 1 by Euler’stheorem.) Thus, it suffices to show that
c := [xk−11 · · · xk−1
k ]∏
16i<j6k
(xj − xi )(bjxj − bixi ) 6= 0.
Now c depends on b1, . . . , bk so that the condition∏16i<j6k(bj − bi ) 6= 0 might be helpful.
24 / 58
Computing c
c =[xk−11 . . . xk−1
k ]∏
16i<j6k
(xj − xi )(bjxj − bixi )
=[xk−11 . . . xk−1
k ]|(bixi )j−1|16i ,j6k × |x j−1
i |16i ,j6k
=[xk−11 . . . xk−1
k ]∑σ∈Sk
sign(σ)k∏
i=1
(bixi )σ(i)−1
∑τ∈Sk
sign(τ)k∏
i=1
xτ(i)−1i
=∑σ∈Sk
sign(σ)sign(σ′)k∏
i=1
bσ(i)−1i =
∑σ∈Sk
(−1)(k2)k∏
i=1
bσ(i)−1i ,
where σ′(i) = k − σ(i) + 1 for all i = 1, . . . , k . As ch(F ) = 2, wehave 1 = −1 in F . Therefore
c =∑σ∈Sk
sign(σ)k∏
i=1
bσ(i)−1i
=|bi−1j |16i ,j6k =
∏16i<j6k
(bj − bi ) 6= 0 (Vandermonde).25 / 58
My related work
Theorem [Z. W. Sun, J. Combin. Theory Ser A 103(2003)]. LetG be an additive abelian group whose finite subgroups are allcyclic. Let b1, . . . , bn be pairwise distinct elements of G , and letA1, . . . ,An be finite subsets of G with cardinality k > m(n− 1) + 1where m is a positive integer.(i) There are at least (k − 1)n −m
(n2
)+ 1 multisets {a1, . . . , an}
such that ai ∈ Ai for i = 1, . . . , n and all the mai + bi are pairwisedistinct.(ii) If b1, . . . , bn are of odd order, then the sets
{{a1, . . . , an}: ai ∈ Ai , ai 6= aj and mai + bi 6= maj + bj if i 6= j}
and
{{a1, . . . , an}: ai ∈ Ai , mai 6= maj and ai + bi 6= aj + bj if i 6= j}
have more than (k − 1)n − (m + 1)(n2
)> (m − 1)
(n2
)elements.
26 / 58
My related work
In the proof I used Alon’s Combinatorial Nullstellensatz andDirichlet’s unit theorem in Algebraic Number Theory. The theoremfollows from my stronger results on sumsets with polynomialrestrictions for which we need the following auxiliary result.
Lemma (Sun). Let R be a commutative ring with identity. LetA = (aij)16i ,j6n be a matrix over R, and let k ,m1, . . . ,mn ∈ N.(i) If m1 6 · · · 6 mn 6 k, then we have
[xk1 · · · xk
n ]|aijxmij |16i ,j6n
( k∑s=1
xs)kn−∑n
i=1 mi =(kn −
∑ni=1 mi )!∏n
i=1(k −mi )!det(A).
(ii) If m1 < · · · < mn 6 k then
[xk1 · · · xk
n ]|aijxmij |16i ,j6n
∏16i<j6n
(xj − xi ) ·( n∑
s=1
xs
)kn−(n2)−∑n
i=1 mi
= (−1)(n2)(kn −
(n2
)−∑n
i=1 mi )!∏ni=1
∏mi<j6k
j 6=mi+1,...,mn
(j −mi )per(A).
27 / 58
3-Dimensional Analogy of Snevily’s Conjecture
In Snevily’s conjecture the condition that |G | is odd cannot beomitted. For general abelian groups, what can we say?
Theorem [Z. W. Sun, Math. Res. Lett. 15(2008)]. Let G be anyadditive abelian group with
Tor(G ) = {g ∈ G : g has a finite order}cyclic, and let A, B and C be finite subsets of G with cardinalityn > 0. Then there is a numbering {ai}ni=1 of the elements of A, anumbering {bi}ni=1 of the elements of B and a numbering {ci}ni=1
of the elements of C , such that ai + bi + ci (1 6 i 6 n) are(pairwise) distinct. Consequently, each subcube of the Latin cubeformed by the Cayley addition table of Z/NZ contains a Latintransversal.
Remark. We don’t require that |G | is odd. The theorem fails forthe noncyclic Klein group Z2 ⊕ Z2.
Conjecture [Z. W. Sun, Math. Res. Lett. 15(2008)]. Anyn × n × n Latin cube contains a Latin transversal.
28 / 58
The DKSS Conjecture
The DKSS Conjecture (Dasgupta, Karolyi, Serra and Szegedy[Israel J. Math., 2001]). Let G be a finite abelian group with|G | > 1, and let p(G ) be the smallest prime divisor of |G |. Letk < p(G ) be a positive integer. Assume that A = {a1, a2, . . . , ak}is a k-subset of G and b1, b2, . . . , bk are (not necessarily distinct)elements of G . Then there is a permutation π ∈ Sk such thata1bπ(1), . . . , akbπ(k) are distinct.
Remark. When G = Zp, the DKSS conjecture reduces to Alon’sresult. DKSS proved their conjecture for Zpn and Zn
p via theCombinatorial Nullstellensatz.
W. D. Gao and D. J. Wang [Israel J. Math. 2004]: TheDKSS conjecture holds when k <
√p(G ), or G is an abelian
p-group and k <√
2p.
Tool of Gao and Wang: The DKSS method combining withgroup rings.
29 / 58
A Result of Feng, Sun and Xiang
Theorem [T. Feng, Z. W. Sun & Q. Xiang, Israel J. Math.,182(2011)]. Let G be a finite abelian group with |G | > 1. LetA = {a1, . . . , ak} be a k-subset of G and let b1, . . . , bk ∈ G , wherek < p = p(G ). Then there is a permutation π ∈ Sk such thata1bπ(1), . . . , akbπ(k) are distinct, provided either of (i)-(iii).
(i) A or B is contained in a p-subgroup of G .
(ii) Any prime divisor of |G | other than p is greater than k!.
(iii) There is an a ∈ G such that ai = ai for all i = 1, . . . , k.
Remark. By this result, the DKSS conjecture holds for any abelianp-group!
Tools: Characters of abelian groups, exterior algebras.
30 / 58
Key lemmas
a1, . . . , ak (in a field) are distinct ⇐⇒k∏
i=1
(aj − ai ) 6= 0.
Let a1, . . . , ak be elements of a finite abelian group G . How tocharacterize that a1, . . . , ak are distinct ?We need the character group
G = {χ : G → K \ {0}| χ(ab) = χ(a)χ(b) for any a, b ∈ G} ∼= G ,
where K is a field having an element of multiplicative order |G |.Lemma 1 (Feng-Sun-Xiang). a1, . . . , ak ∈ G are distinct if andonly if there are χ1, . . . , χk ∈ G such that det(χi (aj))1≤i ,j≤k 6= 0.
Also, there exist χ1, . . . , χk ∈ G with per(χi (aj))1≤i ,j≤k 6= 0provided that a1, . . . , ak are distinct.Lemma 2 (Feng-Sun-Xiang). Let a1, . . . , ak , b1, . . . , bk ∈ G andχ1, . . . , χk ∈ G . If det(χi (aj))16i ,j6k) and per(χi (bj))16i ,j6k) arenonzero, then for some π ∈ Sk the products a1bπ(1), . . . , akbπ(k)are distinct.
31 / 58
Open Problem
How to prove the DKSS conjecturefor general finite abelian groups?
In particular,
how to prove the DKSS conjecturefor the cyclic group Z/nZ?
32 / 58
Arsovski solved the Snevily conjectureIn 2010 B. Arsovski [Israel J. Math. 182(2011)] proved Snevily’sconjecture fully! A key lemma is closely related to the condition in
a result of Feng, Sun and Xiang.
Combinatorial Lemma of Arsovski. Let A = {a1, . . . , ak} andB = {b1, . . . , bk} be k-subsets of an arbitrary abelian group G .Then, there exists a permutation π ∈ Sk such that for anypermutation σ ∈ Sk \ {π}, the multisets
“Snevily’s conjecture was finally solved by Arsovski, aided by thepreparatory work of Feng, Sun and Xiang who had already shownthat Snevily’s conjecture could be deduced from a weakenedversion of Theorem 18.2, which remained a conjecture at thetime.”
33 / 58
Part III. Applications of Combinatorial Nullstellensatzto Restricted Sumsets
34 / 58
A lemma for restricted sumsetsLemma (Alon, Nathanson & Ruzsa [J. Number Theory56(1996)]). Let A1, . . . ,An be finite nonempty subsets of a field Fand let f (x1, . . . , xn) ∈ F [x1, . . . , xn] \ {0}. Suppose thatdeg f 6 k1 + . . .+ kn where ki = |Ai | − 1, and
n nonzero.Applying the Combinatorial Nullstellensatz, we find thatP(a1, . . . , an) 6= 0 for some a1 ∈ A1, . . . , an ∈ An. This isimpossible since a1 + · · ·+ an ∈ C if f (a1, . . . , an) 6= 0.
35 / 58
Alon-Nathanson-Ruzsa TheoremAlon-Nathanson-Ruzsa Theorem [Amer. Math. Monthly102(1995); J. Number Theory 56(1996)]. For finite nonemptysubsets A1, . . . ,An of a field F with |A1| < · · · < |An|, we have
|A1 u · · ·u An| > min
{p(F ),
n∑i=1
(|Ai | − i) + 1
}.
Remark. When |A1| = · · · = |An| = k > n, we can choose A′i ⊆ Ai
with |A′i | = k − n + i and then apply the ANR theorem to get
|A1 u · · ·u An| > |A′1 u · · ·u A′n|
>min
{p(F ),
n∑i=1
(|Ai |′ − i) + 1
}= min{p(F ), (k − n)n + 1}.
Via the Combinatorial Nullstellensatz, the ANR theorem reduces to
[xk11 · · · x
knn ]
∏16i<j6n
(xj − xi )× (x1 + · · ·+ xn)∑n
i=1 ki−(n2)
=(k1 + · · ·+ kn −
(n2
))!
k1! · · · kn!
∏16i<j6n
(kj − ki ).36 / 58
Some other resultsQ. H. Hou and Z. W. Sun [Acta Arith. 102(2002)] studied therestricted sumset
C = {a1+· · ·+an : a1 ∈ A1, . . . , an ∈ An, and ai−aj 6∈ Sij if i < j}
with |A1| = . . . = |An| = k + 1 via
[xk1 · · · xk
n ]∏
16i<j6n
(xj − xi )2m × (x1 + · · ·+ xn)(k−m(n−1))n
=(−1)m(n2)((k −m(n − 1))n)!
m!n
n∏j=1
(jm)!
(k − (j − 1)m)!.
Z. W. Sun and Y. N. Yeh [J. Number Theory 114(2005)]determined
[xk−n+11 · · · xk
n ]∏
16i<j6n
(xj − xi )2m−1 × (x1 + · · ·+ xn)(k−m(n−1))n
37 / 58
Zhao’s supplement to the Hou-Sun resultLilu Zhao [Finite Fields Appl. 28(2014)]: Let k ,m, n be positiveintegers with k > m(n − 1), and let ki ∈ {k , k + 1} for alli = 1, . . . , n. Then
[xk1−11 . . . xkn−1
n ]∏
16i<j6n
(xi − xj)2m × (x1 + . . .+ xn)
∑ni=1(ki−1)−mn(n−1)
=
∑ni=1(ki − 1)−mn(n − 1)
m!n∏s−1
j=0 (k − jm)
n∏j=1
(jm)!
(k − 1−m(j − 1))!.
Zhao deduced this from Aomoto’s identity and a result ofGessel-Lv-Xin-Zhou [JCTA 115(2008)].
Theorem (Hou-Sun; Zhao). Let Sij (1 6 i 6= j 6 n) be finitesubsets of a field F with |Sij | = m. Let A1, . . . ,An be finite subsetsof F with |A1| = . . . = |An| = k ∈ Z+. Suppose that p(F ) > mn.Then, for
C = {a1 + . . .+ an : a1 ∈ A1, . . . , an ∈ An, ai − aj 6∈ Sij if i 6= j},we have |C | > min{p(F ), (k − 1)n −mn(n − 1) + 1}.
38 / 58
A Result of Liu and Sun
J.-X. Liu and Z.-W. Sun [J. Number Theory 97(2002)]. LetA1, . . . ,An be finite subsets of a field F with |Ai+1| − |Ai | ∈ {0, 1}for i = 1, . . . , n − 1, and |An| = k > m(n − 1). Suppose thatP(x) ∈ F [x ], deg P = m and p(F ) > (k − 1)n− (m + 1)
(n2
). Then
|{a1 + · · ·+ an : ai ∈ Ai , P(ai ) 6= P(aj) if i 6= j}|
>(k − 1)n − (m + 1)
(n
2
)+ 1.
Lemma: For positive integers k ,m, n with k − 1 > m(n − 1) wehave
[xk−n1 · · · xk−1
n ]∏
16i<j6n
(xmj − xm
i )× (x1 + · · ·+ xn)(k−1)n−(m+1)(n2)
=(−m)(n2)((k − 1)n − (m + 1)
(n2
))!1!2! · · · (n − 1)!
(k − 1)!(k − 1−m)! · · · (k − 1− (n − 1)m)!.
39 / 58
Solving a conjecture of Erdos and SelfridgeApplying the Liu-Sun result with P(x) = x2 and usingGessel-Viennot’s evaluation (see [Adv. in Math. 1985]) of somebinomial determinants, E. Balandraud obtained the following resulton subset sums.
E. Balandraud [Israel J. Math. 188(2012)]. Let p be a prime andlet A ⊆ Zp with 0 6∈ A + A. Then∣∣∣∣{∑
a∈Ba : ∅ 6= B ⊆ A
}∣∣∣∣ > min
{p,|A|(|A|+ 1)
2
}.
Corollary (conjectured by Erdos and Selfridge). Let p be aprime. Then
max
{|A| :
∑a∈B
a 6= 0 for any ∅ 6= B ⊆ A
}= max
{k ∈ Z :
k(k + 1)
2< p
}=
⌊√8p − 7− 1
2
⌋40 / 58
A Result of Z. W. Sun [J. Combin. Theory Ser. A 103(2003)]:Let A1, . . . ,An be finite subsets of a field F with cardinalityk > m(n − 1). Suppose p(F ) > max{n, (k − 1)n − (m + 1)
(n2
)}.
For any dij ∈ F (1 6 i < j 6 n) and P(x) ∈ F [x ] with degree m,we have
|{a1 + · · ·+ an : ai ∈ Ai , P(ai ) 6= P(aj) and ai − aj 6= dij if i 6= j}|
> (k − 1)n − (m + 1)
(n
2
)+ 1.
Lemma (Z.-W. Sun): For positive integers k ,m, n withk − 1 > m(n − 1), we have
[xk−11 · · · xk−1
n ]∏
16i<j6n
(xj − xi )(xmj − xm
i )× (x1 + · · ·+ xn)K
=(−m)(n2)K !1!2! · · · n!
(k − 1)!(k − 1−m)! · · · (k − 1− (n − 1)m)!,
where K = (k − 1)n − (m + 1)(n2
).
41 / 58
Erdos-Heilbronn conjecture for finite groups
The original Erdos-Heilbronn conjecture is only concerned withcyclic groups of prime order.
P. Balister & J. P. Wheeler [Acta Arith. 140(2009)]:Let G be a finite group written additively with |G | > 1. Then
|2∧A| > min{p(G ), 2|A| − 3} for any A ⊆ G ,
where p(G ) is the least order of a nonzero element of G , i.e., p(G )is the smallest prime divisor of |G |.
Remark. (a) One auxiliary result needed is the Feit-Thompsontheorem: Any group of odd order is solvable.
(b) It is not clear how to extend the result to n∧A or A u B.
42 / 58
Linear extension of the Erdos-Heilbronn conjectureFor a prime p, Zp is an additively cyclic group. On the other hand,Zp is a field which involves both addition and multiplication.
A Conjecture of Z. W. Sun [Finite Fields Appl. 14(2008)]. Leta1, . . . , an be nonzero elements of a field F . If p(F ) 6= n + 1, thenfor any finite A ⊆ F we have
|{a1x1 + · · ·+ anxn : x1, . . . , xn are distinct elements of A}|> min{p(F )− δ, n(|A| − n) + 1},
where
δ = [[n = 2 & a1 + a2 = 0]] =
{1 if n = 2 & a1 + a2 = 0,
0 otherwise.
Difficulty: We cannot apply the Combinatorial Nullstellensatzdirectly, for, the related coefficient involving a1, . . . , an might bezero.
Prizes: I’d like to offer 200 US dollars for a complete proof.43 / 58
Sun and Zhao’s resultsTheorem (Z.-W. Sun and L.-L. Zhao [JCTA 119(2012)]). Theconjecture (posed by Sun) holds if p(F ) > n(3n − 5)/2.
Remark. Zhao and Sun also noted that the conjecture holds forn = 3.
An Auxiliary Theorem (Z.-W. Sun and L.-L. Zhao [JCTA119(2012)]).Let n be a positive integer, and let F be a field withp(F ) > (n − 1)2. Let a1, . . . , an ∈ F ∗ = F \ {0}, and suppose thatAi ⊆ F and |Ai | > 2n − 2 for i = 1, . . . , n. Then, for the set
C = {a1x1+· · ·+anxn : x1 ∈ A1, . . . , xn ∈ An, and xi 6= xj if i 6= j}we have
|C | > min{p(F )− [[n = 2 & a1+a2 = 0]], |A1|+ · · ·+ |An|−n2+1}.Corollary. Let p > 7 be a prime and let A ⊆ F = Z/pZ with|A| >
√4p − 7. Let n = b|A|/2c and a1, . . . , an ∈ F ∗. Then every
element of F can be written in the linear form a1x1 + · · ·+ anxnwith x1, . . . , xn ∈ A distinct. 44 / 58
Sumsets with polynomial restrictions
Theorem (Z.-W. Sun and L.-L. Zhao [JCTA 119(2012)]). LetP(x1, . . . , xn) be a polynomial over a field F . Suppose thatk1, . . . , kn are nonnegative integers with k1 + · · ·+ kn = deg P and[xk1
1 · · · xknn ]P(x1, . . . , xn) 6= 0. Let A1, . . . ,An be finite subsets of
F with |Ai | > ki for i = 1, . . . , n. Then, for the restricted sumset
C = {x1 + · · ·+xn : x1 ∈ A1, . . . , xn ∈ An, and P(x1, . . . , xn) 6= 0},
we have
|C | > min{p(F )− deg P, |A1|+ · · ·+ |An| − n − 2 deg P + 1}.
Remark. In the case P(x1, . . . , xn) = 1 this theorem gives theCauchy-Davenport theorem. When F is of characteristic zero (i.e.,p(F ) = +∞), this theorem extends a result of Sun [Acta Arith.99(2001)] on sums of subsets of Z with various linear restrictions.
45 / 58
ProofIf p(F ) 6 deg P or
∑ni=1 |Ai | < n + 2 deg P, then the desired
inequality holds trivially. Below we assume that p(F ) > deg P and∑ni=1 |Ai | > n + 2 deg P. Write
06i<j(x − ie) with e the identity of F . Note that
[xk11 . . . xkn
n ]P∗(x1, . . . , xn)
=[xk11 . . . xkn
n ]∑
j1+...+jn=degP
cj1,...,jnx j11 . . . x
jnn = ck1,...,kn 6= 0.
46 / 58
ProofFor i = 1, . . . , n let
Bi = {me : m ∈ [|Ai | − ki − 1, |Ai | − 1]}.As ki < deg P < p(F ), we have |Bi | = ki + 1 > ki . By theCombinatorial Nullstellensatz, there are mi ∈ [|Ai |− ki − 1, |Ai |− 1](i = 1, . . . , n) such that P∗(m1e, . . . ,mne) 6= 0.Let M =
Value sets of polynomials over a fieldTheorem (Z.-W. Sun [Finite Fields Appl. 14(2008)]). Let F be afield, and let
f (x1, . . . , xn) = a1xk1 + . . .+ anxk
n + g(x1, . . . , xn) ∈ F (x1, . . . , xn)
with k ∈ Z+ = {1, 2, 3, . . .}, a1, . . . , an ∈ F ∗ = F \ {0} anddeg g < k . Then, for any finite nonempty subsets A1, . . . ,An of F ,we have
|{f (x1, . . . , xn) : x1 ∈ A1, . . . , xn ∈ An}|
>min
{p(F ),
n∑i=1
⌊|Ai | − 1
k
⌋+ 1
}.
Remark. This theorem includes several known results as specialcases. When F = Z/pZ (with p prime) andf (x1, . . . , xn) = x1 + . . .+ xn, this theorem yields theCauchy-Davenport theorem.
50 / 58
ProofLet m be the largest nonnegative integer not exceeding n such that∑
0<i6mb(|Ai | − 1)/kc < p(F ). For each 0 < i 6 m let A′i be asubset of Ai with cardinality kb(|Ai | − 1)/kc+ 1. In the casem < n, p = p(F ) is a prime and we let A′m+1 be a subset of Am+1
with
|A′m+1| =k
(p − 1−
∑0<i6m
⌊|Ai | − 1
k
⌋)+ 1
< k
⌊|Am+1| − 1
k
⌋+ 1 6 |Am+1|.
If m + 1 < j 6 n then we let A′j ⊆ Aj be a singleton. Whetherm = n or not, we have
∑ni=1(|A′i | − 1) = k(N − 1), where
N = min
{p(F ),
n∑i=1
⌊|Ai | − 1
k
⌋+ 1
}.
51 / 58
Continue the proofSet
C = {f (x1, . . . , xn) : x1 ∈ A′1, . . . , xn ∈ A′n}.Suppose that |C | 6 N − 1. Then
[x|A′1|−11 · · · x |A
′n|−1
n ]f (x1, . . . , xn)N−1−|C |∏c∈C
(f (x1, . . . , xn)− c)
=[x|A′1|−11 · · · x |A
′n|−1
n ](a1xk1 + · · ·+ anxk
n )N−1
=(N − 1)!∏n
i=1((|A′i | − 1)/k)!a(|A′1|−1)/k1 · · · a(|A
′n|−1)/k
n 6= 0.
By the Combinatorial Nullstellensatz, for somex1 ∈ A′1, . . . , xn ∈ A′n we have
f (x1, . . . , xn)N−1−|C |∏c∈C
(f (x1, . . . , xn)− c) 6= 0
which contradicts the fact f (x1, . . . , xn) ∈ C .In view of the above,
Theorem [Z. W. Sun, Finite Fields Appl. 14(2008)]. Let F be afield, and let
f (x1, . . . , xn) = a1xk1 + · · ·+ anxk
n + g(x1, . . . , xn) ∈ F [x1, . . . , xn]
with a1, . . . , an ∈ F ∗ = F \ {0} and deg g < k . If A1, . . . ,An arefinite subsets of F with |Ai | > i for i = 1, . . . , n, andn 6 k = deg f , then
|{f (x1, . . . , xn) : xi ∈ Ai , and xi 6= xj if i 6= j}|
>min
{p(F ),
n∑i=1
⌊|Ai | − i
k
⌋+ 1
}.
53 / 58
A General Conjecture
Conjecture (Z. W. Sun [Finite Fields Appl. 14(2008)]). Let F bea field, and let
f (x1, . . . , xn) = a1xk1 + · · ·+ anxk
n + g(x1, . . . , xn) ∈ F [x1, . . . , xn]
with a1, . . . , an ∈ F ∗ = F \ {0} and deg g < k . If A is a finitesubset of F with |A| > n > k and p(F ) 6= n + 1, then
|{f (x1, . . . , xn) : x1, . . . , xn are distinct elements of A}|
>min
{p(F )− δ, n(|A| − n)
k− k
{n
k
}{ |A| − n
k
}+ 1
},
where δ = [[n = 2 & & a1 + a2 = 0]].
Remark. In the case k = 1 this reduces to the conjecture of Sunon linear extension of the Erdos-Heilbronn conjecture.
Theorem (H. Pan and Z.-W. Sun [J. Combin. Theory Ser. A116(2009)]). The above conjecture holds when a1 = · · · = an.
54 / 58
Lev’s Conjecture
Let A and B be finite nonempty subsets of an additive abeliangroup G . In contrast with the Cauchy-Davenport theorem, J.H.B.Kemperman (1960) and P. Scherk (1955) proved that
|A + B| > |A|+ |B| − minc∈A+B
νA,B(c),
whereνA,B(c) = |{(a, b) ∈ A× B: a + b = c}|;
in particular, we have |A + B| > |A|+ |B| − 1 if some c ∈ A + Bcan be uniquely written as a + b with a ∈ A and b ∈ B.
Motivated by the Kemperman-Scherk theorem and theErdos-Heilbronn conjecture, V. F. Lev (2005) proposed thefollowing interesting conjecture.
Lev’s Conjecture. Let A and B be finite nonempty subsets of anabelian group G . Set A u B = {a + b : a ∈ A, b ∈ B, a 6= b}.Then
|A u B| > |A|+ |B| − 2− minc∈A+B
νA,B(c).55 / 58
Progress due to H. Pan and Z. W. SunH. Pan and Z.-W. Sun [Israel J. Math. 154(2006)]:Let A and B be finite nonempty subsets of a field F . LetP(x , y) ∈ F [x , y ] and
C = {a + b: a ∈ A, b ∈ B, and P(a, b) 6= 0}.
If C is nonempty, then
|C | > |A|+ |B| − deg P −minc∈C
νA,B(c).
H. Pan and Z.-W. Sun [Israel J. Math. 154(2006)]:Let A and B be finite nonempty subsets of an abelian group Gwith cyclic torsion subgroup. For i = 1, . . . , l let mi and ni benonnegative integers and let di ∈ G . Suppose that
C = {a+b: a ∈ A, b ∈ B, and mia−nib 6= di for all i = 1, . . . , l} 6= ∅.
Then |C | > |A|+ |B| −∑l
i=1(mi + ni )−minc∈C νA,B(c).
Remark. The proofs involve the Strong Form of the CombinatorialNullstellensatz.
56 / 58
Difference-restricted sumsetsH. Pan and Z.-W. Sun [Israel J. Math. 154(2006)]. Let G be anabelian group, and let A,B, S be finite nonempty subsets of G with
C = {a + b: a ∈ A, b ∈ B, and a− b 6∈ S} 6= ∅.(i) If G is torsion-free or elementary abelian, then
|C | > |A|+ |B| − |S | −minc∈C
νA,B(c).
(ii) If Tor(G ) (the torsion subgroup of G ) is cyclic, then
|C | > |A|+ |B| − 2|S | −minc∈C
νA,B(c).
Remark. Clearly minc∈C νA,B(c) > minc∈A+B νA,B(c) sinceC ⊆ A + B. So, when Tor(G ) is cyclic and S = {0}, part (ii) givesa result slightly weaker than Lev’s conjecture.
H. Pan and Z.-W. Sun [JCTA 100(2002)]. Let F be a finite fieldwith ch(F ) = p 6= 2. Let A,B and S be finite nonempty subsets ofF , and let q be the largest power of p with q 6 |S |. Then
|{a+b: a ∈ A, b ∈ B, and a−b 6∈ S}| ≥ min{p, |A|+|B|−|S |−q−1}.57 / 58
2. Zhi-Wei Sun, A survey of problems and results on restrictedsumsets, in: Number Theory (S. Kanemitsu & J.-Y. Liu, eds.),World Sci., Singapore, 2007, pp. 190–213.
3. Zhi-Wei Sun and Lilu Zhao, Linear exntension of theErdos-Heilbronn conjecture, J. Combin. Theory Ser. A 119(2012), 364–381.