LXXV.1 (1996)
Distribution of lattice points on hyperbolic surfaces
by
Vsevolod F. Lev (Tel-Aviv)
Let two lattices Λ
0, Λ
00⊂ R
shave the same number of points on each hyperbolic surface |x
1. . . x
s| = C. We investigate the case when Λ
0, Λ
00are sublattices of Z
sof the same prime index and show that then Λ
0and Λ
00must coincide up to renumbering the coordinate axes and changing their directions.
1. Introduction. If the opposite is not stated explicitly, we denote vectors by lower-case italic and Greek letters and their coordinates by the same letters with appropriate lower indices. This paper’s aim is to prove the following result.
Theorem 1. Let Λ
0, Λ
00be two sublattices of Z
sof the same prime index , having the same number of points on each surface
|x
1. . . x
s| = C.
Then Λ
00may be obtained from Λ
0(and vice versa) by renumbering the co- ordinate axes and changing their directions.
It would be interesting to extend this theorem to the case of Z
s-sublat- tices of non-prime indices, and also to investigate the case of general R
s- sublattices, when rotations around the bisectrices ±x
1= . . . = ±x
sshould also be taken into account. However, this may present some unexpected difficulties, as shown already for s = 2 by the example of
Λ
0= {m ∈ Z
2| m
1≡ 0 (mod 4)}, Λ
00= {m ∈ Z
2| m
1, m
2≡ 0 (mod 2)}.
Both these lattices are of index 4, and they evidently have the same number of points on each hyperbola |x
1x
2| = C.
1991 Mathematics Subject Classification: Primary 11P21; Secondary 52Cxx, 11Hxx.
Key words and phrases: lattice, hyperbolic surface.
[85]
One may also investigate surfaces of other types, say, spheres. A special class of hyperbolas-like surfaces arises in the theory of uniform distribution, where one often encounters functions of the form
Z(t, Λ) = X
m∈Λ
1 (m
1. . . m
s)
t,
where m
i= max{1, |m
i|}. It is natural to consider those Λ
0, Λ
00for which Z(t, Λ
0) = Z(t, Λ
00) and evidently this means that on each surface of the form x
1. . . x
s= C there is the same number of points of Λ
0and Λ
00. Using the method of this paper, one can prove that Theorem 1 remains valid also for these surfaces instead of pure hyperbolas (though the proof would require substantially greater amount of calculations).
No other results concerning this kind of problems are known to the au- thor.
2. Sublattices of prime index. In this section we establish the struc- ture of Z
s-sublattices of prime index, and thus reduce the considered prob- lem to an algebraic one.
In what follows p will stand for a fixed prime number. For λ ∈ Z (or λ ∈ F
p) define
δ
p(λ) =
1 if λ ≡ 0 (mod p), 0 if λ 6≡ 0 (mod p).
Lemma 1. Let a ∈ F
spand assume a 6= 0. Then the set Λ
a= {m ∈ Z
s| δ
p(a
1m
1+ . . . + a
sm
s) = 1}
is a Z
s-sublattice of index p and moreover , every Z
s-sublattice of index p is of this form for a properly chosen a.
P r o o f. Suppose a
s6= 0, denote by a
0sthe inverse to a
sin F
pand let a
∗i= −a
ia
0s(i = 1, . . . , s − 1). Consider the matrix
E =
1 0 . . . 0 0
0 1 . . . 0 0
. . . .
. . . .
. . . .
0 0 . . . 1 0
a
∗1a
∗2. . . a
∗s−1p
.
Its columns form a basis for some lattice Λ of index |det E| = p, while each
column obviously belongs to Λ
a. Hence Λ ⊆ Λ
a, and therefore either Λ
a= Λ,
or Λ
a= Z
s. Obviously it is the first possibility that really holds, and thus
Λ
a= Λ is a lattice of index p.
Conversely, let Λ ⊂ Z
sbe a lattice of index p generated by some matrix E. Then det E = ±p, and, therefore, there exists a vector a ∈ F
sp, a 6= 0, such that E
Ta ≡ 0 (mod p) (more precisely, all the coordinates of E
Ta are divisible by p). Hence, each column e of E satisfies a
1e
1+ . . . + a
se
s≡ 0 (mod p), thus Λ ⊆ Λ
a. But (by the already proved part of the lemma) both these lattices are of the same index p, so Λ = Λ
a.
Obviously, Λ
adoes not change if we multiply a by a non-zero residue λ ∈ F
p. If we permute the coordinates of a, the new lattice may be obtained from the original one by renumbering the coordinate axes. And if we change the signs of some of the coordinates a
i, this maps the lattice symmetrically relative to the hyperplanes x
i= 0. This shows that if
(1) T
a(C) = X
|m1...ms|=C
δ
p(a
1m
1+ . . . + a
sm
s)
is the number of points of Λ
aon the surface |x
1. . . x
s| = C, then Theorem 1 will follow from (and is actually equivalent to) the following theorem:
Theorem 1
0. Assume T
a(C) = T
b(C) for some fixed non-zero a, b ∈ F
spand every integer C. Then b may be obtained from a (and vice versa) by means of multiplying by a non-zero residue λ ∈ F
p, reordering the coordi- nates and changing the signs of some of them.
We write a ∼ b if a and b are related as described in this theorem.
We show now that without loss of generality we may restrict ourselves to the case when all the coordinates of a, b are non-zero. This will include two steps. As the first step, we suppose T
a(C) = T
b(C) for every C and prove that a and b have the same number of non-zero coefficients. To this end, assume
a = (a
1, . . . , a
s0, 0, . . . , 0), b = (b
1, . . . , b
s00, 0, . . . , 0), where a
1, . . . , a
s0, b
1, . . . , b
s006= 0, and define
α = (a
1, . . . , a
s0), β = (b
1, . . . , b
s00).
If (m
1, . . . , m
s) contributes a non-zero term to the sum T
a(1), then (m
1, . . . . . . , m
s0) contributes a non-zero term to T
α(1), and it is easily seen that every such (m
1, . . . , m
s0) will be induced in this way by exactly 2
s−s0vectors (as the coordinates m
s0+1, . . . , m
scan be independently picked out to be ±1).
The same applies to T
b(1) and T
β(1), and therefore
(2) T
a(1) = 2
s−s0T
α(1), T
b(1) = 2
s−s00T
β(1),
and for brevity we denote this common value by T . Furthermore, if q is prime then
T
a(q) = 2
s−s0T
α(q) + 2
s−s0(s − s
0)T
α(1) = 2
s−s0T
α(q) + (s − s
0)T,
(3)
T
b(q) = 2
s−s00T
β(q) + 2
s−s00(s − s
00)T
β(1) = 2
s−s00T
β(q) + (s − s
00)T.
(4)
By the Dirichlet theorem, for each z = 1, . . . , p − 1 there exists a prime q
zwith q
z≡ z (mod p), and for z = 0 choose q
0= p. Substituting this q
zinto (3), (4) and summing up over all z one obtains
(5) 2
s−s0p−1
X
z=0
T
α(q
z) + p(s − s
0)T = 2
s−s00p−1
X
z=0
T
β(q
z) + p(s − s
00)T.
But the sums in the latter identity may be explicitly evaluated:
p−1
X
z=0
T
α(q
z)
=
p−1
X
z=0
X
1 ε1,...,εs0=0s0
X
i=1
δ
p((−1)
ε1a
1+ . . . + (−1)
εiq
za
i+ . . . + (−1)
εs0a
s0)
= X
1 ε1,...,εs0=0s0
X
i=1 p−1
X
z=0
δ
p((−1)
ε1a
1+ . . . + (−1)
εiza
i+ . . . + (−1)
εs0a
s0)
= 2
s0s
0, and similarly,
p−1
X
z=0
T
β(q
z) = 2
s00s
00. Hence (5) yields
2
ss
0+ p(s − s
0)T = 2
ss
00+ p(s − s
00)T, s
0(2
s− pT ) = s
00(2
s− pT ), and so s
0= s
00, unless p = 2. But if p = 2, then
T = T
a(1) =
2
sif s
0≡ 0 (mod 2), 0 if s
06≡ 0 (mod 2),
therefore 2
s6= pT and s
0= s
00also in this case. Observe that this completely settles Theorem 1
0for p = 2, and in what follows we will assume p to be odd.
Next (and this is our second step), we show that T
α(C) = T
β(C) provided T
a(C) = T
b(C) (for each C). For C = 1 this follows from (2) (in view of s
0= s
00); and for C > 1 we have
T
a(C) = 2
s−s0T
α(C) + 2
s−s0X
d|C, d<C
k
C/dT
α(d),
where k
C/dare some combinatorial coefficients which do not depend on a,
but only on s−s
0and on the system of exponents in the prime decomposition
of C/d. A similar equality holds for T
b(C), and thus the required conclusion follows by induction on C.
Now, if we prove Theorem 1
0for vectors with non-zero coordinates, then T
α(C) = T
β(C) implies α ∼ β, and hence a ∼ b. Thus we arrive at the final form of the theorem to be proved:
Theorem 1
00. Let p be an odd prime, and let a, b ∈ (F
×p)
s. Assume T
a(C) = T
b(C) for all integer C. Then a ∼ b.
3. Even Dirichlet characters. The starting point of our proof is as follows: along with T
a(C), we consider all the values T
c(C) for c ∈ (F
×p)
s, sum them up and then select T
a(C) using characters of the group (F
×p)
s. But characters of (F
×p)
sare closely related to the Dirichlet characters of F
×p. And of all the characters of F
×p, of particular interest for us will be the even ones (those for which χ(−1) = 1). The even characters form a subgroup of index 2 in the group of all characters, and we denote this subgroup by X.
Also, we denote by G the multiplicative group of all squares of F
×p, and by G the dual group of characters of G. Most of the properties of X we use will b be derived from the following simple lemma.
Lemma 2. The mapping b G → X defined by χ 7→ (n 7→ χ(n
2)) (n ∈ F
×p) establishes an isomorphism b G ∼ = X.
P r o o f. The above mapping is obviously an injective homomorphism, and | b G| = |G| = (p − 1)/2 = |X|.
Corollary 1.
(i) X is cyclic.
(ii) For λ ∈ F
×p, X
χ∈X
χ(λ) =
(p − 1)/2 if λ = ±1,
0 if λ 6= ±1.
(iii) Each complex multiplicative function on X (that is, each character of X) is of the form χ 7→ χ(λ) for some fixed λ ∈ F
×p(and even λ ∈ G).
P r o o f.
(i) X ∼ = b G ∼ = G, and G is cyclic.
(ii) follows from the orthogonality relation for group G.
(iii) follows from the canonical homomorphism between G and G (the b b group of characters of b G).
By χ
0we denote the principal character of F
×p, and by χ the character
conjugate to χ.
Lemma 3. Let X
0⊆ X be a set of even characters such that χ
0∈ X
0, and if χ ∈ X
0then necessarily χ ∈ X
0. Let also f : X
0→ C be a complex function on X
0such that χ
1. . . χ
r= χ
0implies f (χ
1) . . . f (χ
r) = 1 for every r ≥ 1 and χ
1, . . . , χ
r∈ X
0. Then f can be extended to a multiplicative function on all X.
P r o o f. First, observe that f can be extended to the subgroup e X
0⊆ X generated by X
0: if χ = χ
1. . . χ
r(χ
1, . . . , χ
r∈ X
0), we put f (χ) = f (χ
1) . . . f (χ
r), and it is easily verified that this correctly defines f as a multiplicative function on e X
0.
Next, if χ is a generating element of X, then all elements of e X
0are powers of χ, and we choose ν to be the smallest positive integer with χ
νin X e
0. Now, we choose the value of f (χ) to satisfy f
ν(χ) = f (χ
ν), and then we define f (χ
n) = f
n(χ) for every integer n.
4. Proof of Theorem 1
00. We start with fulfilling the plan outlined in the previous section: by (1), we have
T
a(C) = (p − 1)
−sX
c
X
m1...ms=±C
δ
p(c
1m
1+ . . . + c
sm
s)
× X
χ
χ
1(c
01a
1) . . . χ
s(c
0sa
s),
where c = (c
1, . . . , c
s) runs over all elements of (F
×p)
s; next, χ runs over all collections (χ
1, . . . , χ
s) of Dirichlet characters; finally, c
0i(i = 1, . . . , s) stands for the inverse of c
iin F
×p. For our purposes, it is sufficient to consider only those C for which C 6≡ 0 (mod p). Then c
im
iruns over F
×ptogether with c
i, and using multiplicativity of characters we get
T
a(C) = (p − 1)
−sX
χ
χ
1(a
1) . . . χ
s(a
s) X
m1...ms=±C
χ
1(m
1) . . . χ
s(m
s)
× X
c
χ
1(c
1m
1) . . . χ
s(c
sm
s)δ
p(c
1m
1+ . . . + c
sm
s)
= (p − 1)
−sX
χ
σ(χ) χ
1(a
1) . . . χ
s(a
s) X
m1...ms=±C
χ
1(m
1) . . . χ
s(m
s), where
σ(χ) = X
c
χ
1(c
1) . . . χ
s(c
s)δ
p(c
1+ . . . + c
s).
Consider the latter sum. If λ ∈ F
×pis fixed, then λc runs over (F
×p)
stogether with c, hence
σ(χ) = X
c
χ
1(λc
1) . . . χ
s(λc
s)δ
p(λ(c
1+ . . . + c
s))
= χ
1. . . χ
s(λ)σ(χ),
and so σ(χ) = 0 if the product χ
1. . . χ
sis not the principal character χ
0. On the other hand, for χ
1. . . χ
s= χ
0and at least one χ
idistinct from χ
0, we use a well-known representation of δ
pwith complex exponents to obtain
σ(χ) = 1 p
p−1
X
z=0
X
p−1 c1,...,cs=1χ
1(c
1) . . . χ
s(c
s)e
2πi(c1+...+cs)z/p= 1 p
p−1
X
z=1
(χ
1. . . χ
s)(z)
p−1
X
c1,...,cs=1
χ
1(zc
1) . . . χ
s(zc
s)e
2πi(zc1+...+zcs)/p= 1 p
p−1
X
z=1
p−1X
c1=1
χ
1(c
1)e
2πic1/p. . .
p−1X
cs=1
χ
s(c
s)e
2πics/p= p − 1
p τ (χ
1) . . . τ (χ
s),
where the Gaussian sums τ (χ
i) are known to be non-zero. Also, if χ
1= . . . = χ
s= χ
0then obviously σ(χ) 6= 0. We see that σ(χ) is not 0 if and only if χ
1. . . χ
s= χ
0, and thus
T
a(C) = (p − 1)
−sX
χ
∗
σ(χ)χ
1(a
1) . . . χ
s(a
s) X
m1...ms=±C
χ
1(m
1) . . . χ
s(m
s), where the sum marked by an asterisk extends over all collections of Dirichlet characters with product χ
0. As to the inner sum, it may be written as
X
m1...ms=C m1,...,ms≥1
(χ
1(m
1) + χ
1(−m
1)) . . . (χ
s(m
s) + χ
s(−m
s)),
and this vanishes if at least one of the characters χ
iis odd. But otherwise χ
i(m
i) + χ
i(−m
i) = 2χ
i(m
i) and therefore
T
a(C) = 2
s(p − 1)
−sX
χ
∗∗
σ(χ)χ
1(a
1) . . . χ
s(a
s) X
m1...ms=C
χ
1(m
1) . . . χ
s(m
s), where the sum marked by two asterisks extends over all collections of even Dirichlet characters with product χ
0, and the m
ihere and in all subsequent sums take only positive values. Group now together those collections χ which differ only by a permutation. If χ
0and χ
00are two such collections then obviously σ(χ
0) = σ(χ
00), and also
X
m1...ms=C
χ
01(m
1) . . . χ
0s(m
s) = X
m1...ms=C
χ
001(m
1) . . . χ
00s(m
s)
for each C, hence
T
a(C) = 2
s(p − 1)
−sX
ϕ
∗∗∗
σ(ϕ) X
χ∈P (ϕ)
χ
1(a
1) . . . χ
s(a
s)
× X
m1...ms=C
ϕ
1(m
1) . . . ϕ
s(m
s), where ϕ runs over all ordered collections of even characters with product χ
0, and χ runs over the set P (ϕ) of all collections which may be obtained by a permutation of ϕ.
A similar equality holds, of course, for T
b(C), and thus
(6) X
ϕ
∗∗∗
u(ϕ) X
m1...ms=C
ϕ
1(m
1) . . . ϕ
s(m
s) = 0, where we set
u(ϕ) = σ(ϕ) X
χ∈P (ϕ)
χ
1(a
1) . . . χ
s(a
s) − X
χ∈P (ϕ)
χ
1(b
1) . . . χ
s(b
s)
.
The next step is to show that the inner sums in (6), considered as func- tions of C, are linearly independent, and so u(ϕ) = 0.
We first derive from (6) that for every r ≥ 1 and every system of residues n
1, . . . , n
r∈ F
×pwe have
(7) X
ϕ
∗∗∗
u(ϕ) X
s 0 i1,...,ir=1ϕ
i1(n
1) . . . ϕ
ir(n
r) = 0;
here and below the indices of the dashed sums ( P
0) run over pairwise distinct values; for example, in the latter sum (i
1, . . . , i
r) runs over all r-element subsets of the set {1, . . . , s}.
To obtain (7) we use induction on r. For r = 1 the result follows immedi- ately if in (6) we choose C = q
1, where q
1≡ n
1(mod p) is prime. For r ≥ 2 we choose pairwise distinct primes q
iwith q
i≡ n
i(mod p) (i = 1, . . . , r) and let C = q
1. . . q
r. Next, let Ω run over all partitions of the set {1, . . . , r}.
Denote by P
(Ω)the sum over all those i
1, . . . , i
rfor which i
ν= i
µif and only if ν and µ fall into the same class of the partition Ω. We have
(8) X
ϕ
∗∗∗
u(ϕ) X
m1...ms=C
ϕ
1(m
1) . . . ϕ
s(m
s)
= X
ϕ
∗∗∗
u(ϕ) X
s i1,...,ir=1ϕ
i1(q
1) . . . ϕ
ir(q
r)
= X
Ω
X
ϕ
∗∗∗
u(ϕ)
X
s (Ω) i1,...,ir=1ϕ
i1(n
1) . . . ϕ
ir(n
r).
Now, if Ω is the partition into r one-element sets, then the inner sum in the latter equality equals
X
s 0 i1,...,ir=1ϕ
i1(n
1) . . . ϕ
it(n
r);
and if Ω = {S
1, . . . , S
t}, where S
1∪ . . . ∪ S
t= {1, . . . , r}, t < r, then this inner sum equals
X
s 0 j1,...,jt=1ϕ
j1(N
1) . . . ϕ
js(N
t), where N
ν= Q
i∈Sν
n
i. Since by the induction hypothesis X
ϕ
∗∗∗
u(ϕ) X
s 0 j1,...,jt=1ϕ
j1(N
1) . . . ϕ
jt(N
t) = 0, (7) follows from (8) and (6).
Specifically, for r = s, (7) yields
(9) X
ϕ
∗∗∗
u(ϕ) X
s 0 i1,...,is=1ϕ
i1(n
1) . . . ϕ
is(n
s) = 0, where (i
1, . . . , i
s) runs over all permutations of {1, . . . , s}.
Let χ be a fixed collection of even characters with χ
1. . . χ
s= χ
0. Mul- tiplying (9) by χ
1(n
1) . . . χ
s(n
s) and summing up over all n ∈ (F
×p)
swe obtain
X
ϕ
∗∗∗
u(ϕ) X
s 0 i1,...,is=1 p−1X
n1=1
(χ
1ϕ
i1)(n
1)
. . .
p−1X
ns=1
(χ
sϕ
is)(n
s)
= 0, thus u(χ) = 0 by the orthogonality relation, since there exists precisely one ϕ and precisely one permutation (i
1, . . . , i
s) with χ
1= ϕ
i1, . . . , χ
s= ϕ
is. This shows that
(10) X
χ∈P (ϕ)
χ
1(a
1) . . . χ
s(a
s) = X
χ∈P (ϕ)
χ
1(b
1) . . . χ
s(b
s), provided that the ϕ
iare even and ϕ
1. . . ϕ
s= χ
0.
Furthermore, it follows from (10) that (11)
X
s 0 i1,...,is=1ϕ
i1(a
1) . . . ϕ
is(a
s) =
X
s 0 i1,...,is=1ϕ
i1(b
1) . . . ϕ
is(b
s).
Indeed, if {ϕ
1, . . . , ϕ
s} break into t groups with l
1, . . . , l
tcoinciding charac-
ters in each group and distinct characters in distinct groups, then (11) is
obtained from (10) by multiplying by l
1! . . . l
t!. Next, (11) may be rewritten
as (12)
X
s 0 i1,...,is=1ϕ
1(a
i1) . . . ϕ
s(a
is) =
X
s 0 i1,...,is=1ϕ
1(b
i1) . . . ϕ
s(b
is).
If now 1 ≤ r ≤ s and ϕ
iare even with ϕ
1. . . ϕ
r= χ
0, we define ϕ
r+1= . . . = ϕ
s= χ
0and get
(13)
X
s 0 i1,...,ir=1ϕ
1(a
i1) . . . ϕ
r(a
ir) =
X
s 0 i1,...,ir=1ϕ
1(b
i1) . . . ϕ
r(b
ir).
We prove that the latter equality remains valid if we extend the sum- mation over all collections i
1, . . . , i
r, probably with coinciding i
νand even with r > s. Using the above introduced notation, write
X
s i1,...,ir=1ϕ
1(a
i1) . . . ϕ
r(a
ir) = X
Ω
X
s (Ω) i1,...,ir=1ϕ
1(a
i1) . . . ϕ
r(a
ir).
For a fixed Ω = {S
1, . . . , S
t} the inner sum on the right-hand side equals X
s 0j1,...,jt=1
Φ
1(a
j1) . . . Φ
t(a
jt), where Φ
ν= Q
i∈Sν
ϕ
i, and by (13) this does not change if we replace a by b (if t > s we can not use (13) but then this last sum obviously equals 0).
Our next observation is that X
si1,...,ir=1
ϕ
1(a
i1) . . . ϕ
r(a
ir) = Y
r j=1(ϕ
j(a
1) + . . . + ϕ
j(a
s)) and hence we have proved that
(14)
Y
r j=1(ϕ
j(a
1) + . . . + ϕ
j(a
s)) = Y
r j=1(ϕ
j(b
1) + . . . + ϕ
j(b
s)) as soon as ϕ
1, . . . , ϕ
rare even characters with ϕ
1. . . ϕ
r= χ
0.
We now use Lemma 3. For X
0we choose the set of all characters ϕ satisfying ϕ(b
1) + . . . + ϕ(b
s) 6= 0, and define f : X
0→ C by
f (ϕ) = ϕ(a
1) + . . . + ϕ(a
s) ϕ(b
1) + . . . + ϕ(b
s) .
Then (14) guarantees that the conditions of Lemma 3 are satisfied, and using also Corollary 1(iii) we conclude that there exists λ ∈ F
×psuch that
(15) ϕ(a
1) + . . . + ϕ(a
s) = (ϕ(b
1) + . . . + ϕ(b
s)) ϕ(λ)
for every ϕ ∈ X
0. But if in (14) we choose r = 2 and ϕ
1= ϕ, ϕ
2= ϕ, where
ϕ is even but ϕ(b
1) + . . . + ϕ(b
s) = 0, we see that also ϕ(a
1) + . . . + ϕ(a
s) = 0
and thus (15) holds for all even ϕ, regardless of whether ϕ ∈ X
0or not. By multiplicativity of ϕ, we can write (15) in the form
(16)
X
s i=1ϕ(a
i) = X
si=1
ϕ(λb
i).
Multiplying (16) by ϕ(n) for n ∈ F
×pand summing up over all even charac- ters ϕ, we obtain
X
s i=1X
ϕ
ϕ(a
in
0)
= X
si=1
X
ϕ