ROBBA’S METHOD ON EXPONENTIAL SUMS

PEIGEN LI
arXiv:2109.02825v1 [math.AG] 7 Sep 2021

Abstract. In the present paper, we use Robba’s method to give

an estimation of the Newton polygon for the L-function.

Contents
1. Introduction 1
2. p-adic estimations 3
References 10

1. Introduction
The basic objects of this study are exponential sums on a torus of
dimension n defined over a finite field k with char(k) = p. Our methods
are p-adic and are based on the work of Dwork, Adolphson and Sperber.
Let ζp be a primitive root of unity. Let ψ be the additive character
Trk/F (t)
p
of k given by ψ(t) = ζp . Let f be a Laurent polynomial and
write
XN
f= ai xwi ∈ k[x1 , · · · , xn , x−1 −1
1 , · · · , xn ].
i=1
We assume that ai 6= 0 for all i. Defines exponential sums
X
Si (f ) = ψ(Trki /k (f (x))),
x∈Tn (ki )

where ki is the extension of k of degree i. The L-function is defined by

X ∞ 
L(f, t) = exp Si (f )ti /i .
i=1

According to the theorems of Grothendieck and Dwork, it is a rational

function.
In [1, section 2], Adolphson and Sperber use Dwork’s method to
n−1
prove that L(f, t)(−1) is a polynomial when f is nondegenerate.
n−1
Moreover, they give a low bound of the Newton polygon of L(f, t)(−1)
1
in [1, section 3], which we call Hodge polygon in this article. In our
study, we want to give a more precise result about the Newton poly-
gon when f has only n terms, that is N = n. Note that if we as-
sume that J = (w1 , · · · , wn ) is invertible, we can found a solution
b = (b1 , · · · , bn ) ∈ k̄ × such that ai bwi = 1 for all i. From now on, we
assume that (p, det J) = 1, k = Fp and
n
X
f= xwi .
i=1

Let ∆(f ) be the Newton polyhedron at ∞ of f which is defined to

be the convex hull in Rn of the set {wj }nj=1 ∪ {(0, · · · , 0)} and let C(f )
be the convex cone generated by {wj }nj=1 in Rn . Let Vol(∆(f )) be the
volume of ∆(f ) with respect to Lebesgue measure on Rn . For any face
σ of ∆(f ), set
X
fσ = aj xwj .
wj ∈σ

We say f is nondegenerate with respect to ∆(f ) if for any face σ of

∆(f ) not containing the origin, the Laurent polynomials
∂fσ ∂fσ
,··· ,
∂x1 ∂xn
have no common zero in (k̄ × )n . Note that (p, det J) = 1 imply that f is
nondegenerate. But conversely it is not right. Set M(f ) = C(f ) ∩ Zn .
Since we have assumed that J is invertible, any element u ∈ M(f ) can
be uniquely written
n
X
(1.1) u= ri w i .
i=1

We define a weight on M(f )

n
X
w(u) := ri .
i=1

Note that the set of all elements u ∈ M(f ) such that all 0 ≤ ri < 1 in
the expression (1.1) form a fundamental domain of the lattice M(f ).
We denote it by S(∆). Note that S(∆) has a natural p-action. For any
u = r1 w1 + · · · + rn wn ∈ S(∆), define
n
X
p.u = {pri } wi ,
i=1
2
where {pri } is the fractional part of pri for each i. We say S(∆) is p-
stable under weight function if w(u) = w(p.u) for any u ∈ S(∆). Now
we give our main result.
Theorem 1.1. Suppose that f = xw1 + · · · + xwn where wi ∈ Zn for
n−1
each i, p ≥ 3 and (p, det J) = 1. The Newton polygon of L(f, t)(−1)
coincides with the Hodge polygon of ∆(f ) if and only if S(∆) is p-stable.
Wan uses the Gauss sum to give an explicit formula of the L-function
for the diagonal Laurent polynomial. Then he uses Stickelberger’s the-
orem to give a proof of above theorem. See [5, Theorem 3.4]. In this
article, we use Robba’s method to prove above theorem.

2. p-adic estimations
In [4], Robba gives an explicit calculation of one variable twisted
exponential sums. In fact, his method can be applied to the case of
multi-variables. Let Zp be the p-adic integers and Qp the p-adic num-
bers. Let Ω be the completion of the algebraic closure of Qp .
Note that there is a integer M such that w(M(f )) ⊂ M1 Z. In [1,
section 1], Adolphson and Sperber introduce a filtration on R(f ) :=
n
k[xδ∩Z ] given by
( )
X
R(f )i/M = bu xu |w(u) ≤ i/M for all u with bu 6= 0 .
u∈δ∩Zn

The associated graded ring is

M
R̄ = R̄i/M ,
i∈Z≥0

where
R̄i/M = R(f )i/M /R(f )(i−1)/M .
For 1 ≤ i ≤ n, let f¯i be the image of xi ∂x ∂f
i
∈ R(f )1 in ∈ R̄1 . Let I¯
be the ideal generated by f¯1 , . . . , f¯n in R̄. By [1, Theorem 2.14] and
[1, Theorem 2.18], f¯1 , . . . , f¯n in R̄ form a regular sequence in R̄ and
dimk R̄/I¯ = n! Vol(∆(f )). For each integer i, we have a decomposition
(2.1) ¯
R̄i/M = V̄ i/M ⊕ (R̄i/M ∩ I).
Set ai =dimk V̄ i/M .
For a non-negative integer l, set
 
n l
W (l) = card u ∈ Z |w(u) = .
M
3
When u ∈
/ M(f ), set w(u) = +∞. This is a finite number for each l.
Set
Xn  
l n
H(i) = (−1) W (i − lM).
l
l=0

Lemma 2.1. With the notation above. Suppose that f is nondegener-

ate. Then H(i) = ai for all integer i ≥ 0. Moreover, we have
nM
X
H(k) = 0 for k > nM, H(k) = n! Vol(∆(f )).
k=0
 n
Proof. By [1, Theorem 2.14], f¯i i=1 form a regular sequence in R̄. So

PR̄/I¯(t) = PR̄ (t)(1 − tM )n ,

where PR̄/I¯ (resp. PR̄ ) is the Poincaré series of R̄/I¯ (resp. R̄). On the
other hand, we have
∞
X ∞
X
i
PR̄/(f¯1 ,...,f¯n ) = ai t , PR̄ (t) = W (i)ti .
i=0 i=0

Hence
Xn  
l n
ai = (−1) W (i − lM) = H(i).
l=0
l
The second assertion follows from [3, Lemma 2.9]. 

Note that R̄/I¯ has a finite basis S = {xu |u ∈ S(∆)} and card(S(∆)) =
n! Vol(∆(f )).
Definition 2.2. The Hodge polygon HP (∆) of ∆(f ) is defined to be
the convex polynomial in R2 with vertices (0, 0) and
X
m
1 X
m 
H(k), kH(k) .
k=0
M k=0

Let π be a solution of tp−1 + p = 0 such that ζp ≡ 1 + π mod π 2 and

consider
X∞
p
θ1 (t) = exp(π(t − t )) = λj tj ∈ Ω[[t]].
j=1

The series θ1 (t) is a splitting function in Dwork’s terminology. By [2,

§4a], we have ord(λj ) ≥ p−1 p2
j, θ1 (t) ∈ Qp (ζp ) and θ1 (1) = ζp . We fix
4
a choice π
e of M-th root of π in Ω. Set K = Qp (eπ ). Define a space of
p-adic function
 
 X 
B= eM w(u) xu |Au ∈ K, Au → 0 as u → 0 .
Au π
 
u∈M (f )

Note
P that B Misw(u)complete under the norm defined as follows: If ξ =
u∈M (f ) Au π
e xu ∈ B, set
||ξ|| = sup |Au |.
u∈M (f )

Define an operator ϕ on formal Laurent series by

X X
ϕ( Au xu ) = Apu xu .
u∈M (f ) u∈M (f )

Set α1 = ϕ ◦ exp(π(fˆ(x) − fˆ(xp ))), here fˆ = xw1 + · · · + xwn is regarded

as an element in Qp [x±1 ±1
1 , · · · , xn ]. Note that α1 is an operator on B
when p > 2. For each i, define operator Di = Ei + πEi (fˆ), where
∂ fˆ
Ei = xi ∂x i
. By [1, Corollary 2.19], when p > 2, we have
n
X
(−1)n−1
L(f, t) = det(1 − tα1 |B/ Di B)
i=1

By [1, Therorem 2.18] and [1, Theorem A.1], S = {xu }u∈S(∆) is a free
P
basis of B/ ni=1 Di B. For any u ∈ M(f ), u′ ∈ S(∆), define A(u, u′)
by the relation
X n
X
′
xu ≡ A(u, u′)xu mod Di B.
u′ ∈S(∆) i=1

For any u, u′ ∈ S(∆), define γ(u, u′) by the relation

X n
X
′
α1 (xu ) ≡ γ(u, u′)xu mod Di B.
u′ ∈S(∆) i=1

The main purpose of is to give estimations for the p-adic valuations of

the coefficients γ(u, u′).
P
Lemma 2.3. For any u = ni=1 ri wi ∈ M(f ), u′ ∈ S(∆) we have
Pn Xn
′ w(u′ ) − w(u) i=1 [ri ] ′
ord(A(u, u )) ≥ =− if u = {ri } wi
p−1 p−1 i=1
5
and
n
X
A(u, u′) = 0 if u′ 6= {ri } wi ,
i=1
where {ri } (resp. [ri ]) denotes the fractional part (resp. integral part)
of ri for each i.
Proof. For any 1 ≤ i ≤ n, we have
Di (xu ) = ui xu + πEi (fˆ)xu
and
   ˆ
xw1 E1 f
 
J  ..  =  ...  .
.
xwn En fˆ
So    
xw1 +u −u1 xu Xn
1
J  ...  ≡  ...  mod Di B.
wn +u π u
x −un x i=1

Since we assume that (p, det J) = 1, we have

 J1 +u   u
x ∗x Xn
 ...  ≡ 1  ...  mod Di B,
Jn +u π u
x ∗x i=1

where ∗ is used to indicate a coefficient which is a p-adic integer. There-

fore, if ri0 ≥ 1 for some i0 , we have
Xn
u ∗ u−wi
x ≡ x 0 mod Di B
π i=1

which implies the results immediately. 

j
Lemma 2.4. (i) ord(λj ) = p−1 when j ≤ p − 1.
j
(ii) ord(λj ) ≥ p−1 when j ≤ p2 − 1.
1
(iii) ord(λj ) ≥ ( p−1 − p22 )j when j ≤ p3 − 1.
Proof. (i) We have
θ1 (t) = exp(π(t − tp )) ≡ exp(πt) mod tp .
j j
Note that ord(λj ) = ord( πj! ) = p−1 for any j ≤ p − 1.
(ii) Let E(t) be the Artin-Hasse exponential series:
X+∞ pi 
t
E(t) = exp
i=0
pi
6
Note that E(t) ∈ Zp [[t]]. We have
∞
Y i
(πt)p p2
θ1 (t) = E(πt) exp(− ) ≡ E(πt) mod t
i=2
pi
j
Hence, ord(λj ) ≥ for any j ≤ p2 − 1.
p−1
(iii) By the proof of (ii), we have
2
(πt)p
θ1 (t) ≡ E(πt) exp(− 2 )
p
p 2 p 2
 ( (πt)
p2
)1 ( (πt)
p2
)p−1  3
≡ E(πt) 1 + +···+ mod tp .
1! (p − 1)!
Set
p2
p−1 (πt)
X ( p 2 )k
g(t) = .
k=0
k!
Note that
2
 ( πp2 )k 
 p2   1 2
p
ord =k−2 = − 2 · kp2
k! p−1 p−1 p
for any k ≤ p − 1. Combined with E(t) ∈ Zp [[t]], we have ord(λj ) ≥
1
( p−1 − p22 )j when j ≤ p3 − 1. 
Proposition 2.5. For any u, u′ ∈ S(∆), when p ≥ 3, we have

′
+∞ if p.u′ − u 6= 0,
ord(γ(u, u )) = ′
pw(u )−w(u)
p−1
if p.u′ − u = 0.
ord(γ(u, u′)) = +∞ means that γ(u, u′) = 0.
Proof. We have
X
(2.2) γ(u, u′) = hpu′ −u + hpv−u A(v, u′).
v∈M (f )−S(∆)
P
By Lemma 2.3, A(v, u′) = P0 when v ∈ / u′ + ni=1 Z≥0 wi . Thus we only
consider the case v = u′ + ni=1 si wi where si ∈ Z≥0 . Note that
n
Y
hpv−u = ck j ,
j=1

where (k1 , . . . , kn ) ∈ Zn≥0 satisfying the equation

n
X n
X
′
(2.3) ki wi = pv − u = pu − u + p si w i .
i=1 i=1
7
If p.u′ −u 6= 0, the above equation has no integer solution which implies
γ(u, u′) = 0. Assume
Pn that pu′ − u =Pr1 w1 + · · · + rn wn where ri ∈ Z for
all i. Set u = i=1 ri,u wi and u = ni=1 ri,u′ wi where 0 ≤ ri,u , ri,u′ < 1
′

for all i. Thus

−1 < ri = pri,u′ − ri,u < p
for all i. So 0 ≤ ri ≤ p − 1 for all i and w(pu′ − u) = r1 + · · · + rn . By
(2.3) we have ki = ri + psi for each i. Hence, by Lemma 2.3, we have
n n
′ 1 X 1 X
ord(A(v, u )) ≥ − si = − (ki − ri )
p − 1 i=1 p(p − 1) i=1
n
X X ri X ri n n
ki
= − − + .
i=1
p(p − 1) i=1 p i=1
p−1

By Lemma 2.4, we have

X ki  1 2 X X p−1
ord(hpv−u ) ≥ + − 2 ki + ki .
2
p−1 p−1 p p2
ki <p p2 ≤ki <p3 3
ki ≥p

w(pu′ − u)
ord(hpv−u A(v, u′)) ≥ A + B + C + ,
p−1
where
X ki ki ri X k i ri X
A := − − = − = si ,
p − 1 p(p − 1) p p p
ki <p2 2 2
ki <p ki <p

X  1 2 ki ri X p−2 ri
B := − 2 ki − − = k i −
p−1 p p(p − 1) p p2 p
p2 ≤ki <p 3 p2 ≤ki <p3
X
> p − 3,
p2 ≤k i <p3

X p−1 ki ri X p − 1 1  ri
C := 2
ki − − = 2
− ki −
3
p p(p − 1) p 3
p p(p − 1) p
ki ≥p ki ≥p
X p2
> p(p − 1) − −1
p−1
ki ≥p3
X 1
= (p − 1)2 − 3 − ,
p−1
ki ≥p3
8
 / S(∆), there exists some i0 such that si0 > 0. If ki0 < p2 , we
When v ∈
have X
A= si > 0.
ki <p2

If ki0 ≥ p2 , we have
B+C >0
when p ≥ 3. Hence,
n
X
′ ri pw(u′) − w(u)
ord(hpv−u A(v, u )) > =
i=1
p−1 p−1
for any v ∈/ S(∆). Consider v = u′ ∈ S(∆). Thus ki = ri ≤ p − 1 for
all i. By Lemma 2.4, we have
Xn n
1 X pw(u′) − w(u)
ord(hpu −u ) =
′ ord(cki ) = ki = .
i=1
p − 1 i=1 p−1
By (2.2), we have
pw(u′) − w(u)
ord(γ(u, u′)) = ord(hpu′ −u ) = .
p−1

Pn
Theorem 2.6. Suppose that f = j=1 xwj and (p, det J) = 1, p ≥
n−1
3. The Newton polygon of L(Tn , f, t)(−1) coincides with the Hodge
polygon HP (∆) if and only if S(∆) is p-stable under weight function.
n−1
Proof. By [1, Corollary 3.11], the Newton polygon of L(Tn , f, t)(−1)
lies above the Hodge polygon of HP (∆) with same endpoints and
the matrix Γ := (γ(u, u′))u,u′ ∈S(∆) is invertible. By Proposition 2.5,
γ(u, u′) 6= 0 if and only if p.u′ − u = 0. Note that p.u′ − u1 = p.u′ − u2
if and only is u1 = u2 . Thus there is exactly one non zero element in
every column and row of Γ. Let S(d, u) be the orbit of u under the
p-action with exactly d elements. Suppose that S(d, u) = {u1 , · · · , ud},
where ui = pi−1 .u. By Proposition 2.5, we have
 
0 γ21 . . .
 0 0 ... 
α1 (xu1 , · · · , xud ) = (xu1 , · · · , xud ) 
 ... .. 

. γ dd−1
γ1d · · · 0
where γij = γ(ui , uj ). Thus
Y Y
det(1 − α1 t) = (1 − td λu′ )1/d ,
d≥1 u′ ∈S(d,u)
9
where λu′ = λu and
ord(λu′ ) = ord(λu ) = ord(γ1d γ21 · · · γdd−1 )
pw(ud ) − w(u1 ) pw(ud−1 ) − w(ud )
= +···+
p−1 p−1
d−1
X
= w(pi .u).
i=0
Set Y
fu,d = (1 − td λu′ )1/d = 1 − td λu
u′ ∈S(d,u)
and
d−1
Y i
gu,d = (1 − tpw(p .u) ).
i=0
Then the Newton polygon of fu,d always lies above the Newton Q polygon
of gu,d. Moreover, the Newton polygon of the polynomial d≥1 gu,d is
HP (∆). Then HP (∆) coincide with the Newton polygon of det(1 −
α1 t) if and only if the Newton polygons of gu,d and fu,d coincide for
each u.
When S(∆) is p-stable under weight function. We have w(u) =
w(p.u) = · · · = w(pd−1 .u) for each u. Thus, the Newton polygons of
gu,d and fu,d coincide for each u.
Conversely, if the Newton polygons of gu,d and fu,d coincide for each
u. Since both polygons have same endpoints, thus w(u) = w(p.u) =
· · · = w(pd−1 .u) for each u. Thus S(∆) is p-stable under weight func-
tion. 
References
[1] Alan Adolphson and Steven Sperber. Exponential sums and Newton polyhedra:
Cohomology and estimates. Annals of Mathematics, 130(2):367–406, 1989.
[2] Bernard Dwork. On the zeta function of a hypersurface. Publications
Mathématiques de l’IHÉS, 12:5–68, 1962.
[3] Anatoli G Kouchnirenko. Polyedres de Newton et nombres de Milnor. Inven-
tiones mathematicae, 32(1):1–31, 1976.
[4] Philippe Robba. Index of p-adic differential operators III. Application to twisted
exponential sums. Astérisque, 119(120):191–266, 1984.
[5] Daqing Wan. An Introduction to the theory of Newton polygons for L-functions
of exponential sums, to appear. Preprint available at http://www. math. uci.
edu/dwan/Overview. html, 1999.

Yau Mathematical Sciences Center, Tsinghua University, Beijing

100084, P. R. China
Email address: lpg16@mails.tsinghua.edu.cn
10