Math. Res. Lett.

Volume 22, Number 5, 1541–1562, 2015

A new Fourier transform

Jonathan Wang

In order to define a geometric Fourier transform, one usually works

with either -adic sheaves in characteristic p > 0 or with D-modules
in characteristic 0. If one considers -adic sheaves on the stack quo-
tient of a vector bundle V by the homothety action of Gm , how-
ever, Laumon provides a uniform geometric construction of the
Fourier transform in any characteristic. The category of sheaves
on [V /Gm ] is closely related to the category of (unipotently) mon-
odromic sheaves on V . In this article, we introduce a new func-
tor, which is defined on all sheaves on V in any characteristic,
and we show that it restricts to an equivalence on monodromic
sheaves. We also discuss the relation between this new functor and
Laumon’s homogeneous transform, the Fourier-Deligne transform,
and the usual Fourier transform on D-modules (when the latter
are defined).

1. Introduction

In order to define a geometric Fourier transform, one usually works with

either -adic sheaves in characteristic p > 0 or with D-modules in character-
istic 0 (under these conditions one has a rank 1 local system on A1 which
plays the role of the function eix in classical Fourier analysis). If one only
needs to consider homogeneous sheaves, however, Laumon [10] provides a
uniform geometric construction of the Fourier transform for -adic sheaves
in any characteristic. Laumon considers homogeneous sheaves as sheaves
on the stack quotient of a vector bundle V by the homothety Gm action.
This category is closely related to the category of (unipotently) monodromic
sheaves on V (cf. [5]). While it has been well known to experts that a similar
uniform construction of the Fourier transform exists for monodromic sheaves
(Beilinson suggests a definition in [4, footnote 2]), the details have not been
exposited in the literature. In this note, we fill in this gap. We also introduce
a new functor, which is defined on all sheaves in any characteristic, and show
that it agrees with the usual Fourier transform on monodromic sheaves.

1541
1542 Jonathan Wang

We define the new Fourier transform FourB in §2 and show that the
“square” Four2B has a simple formula. In §3, we use this formula to prove the
main result that FourB induces an equivalence of bounded derived categories
of monodromic (étale) sheaves. We also discuss the relation between FourB
and Laumon’s homogeneous Fourier transform. In §4, we compare FourB and
the Fourier-Deligne transform in characteristic p > 0. Our study of FourB
reveals several surprising facts about a certain object j ∗ B of the monoidal
category Dctf (Gm ). In §5, we prove the analogous facts about j ∗ B in the
D-module setting by considering the Mellin transform. We use this to show
that FourB agrees with the Fourier transform on monodromic D-modules.

1.1. Notation and terminology

Let k be an arbitrary base field and fix an algebraic closure k̄. Choose a

prime  not equal to the characteristic of k. Let R be a finite commuta-
tive Z/r -algebra for a positive integer r. Fix a base scheme S of finite
type over k. Let π : V → S be a vector bundle of rank d and π ∨ : V ∨ → S
the dual vector bundle. We will work with the bounded derived category
Dcb (V ) = Dcb (V, R) of étale sheaves of R-modules with constructible coho-
mologies. Our results are also true when Dcb (V, R) is replaced by the full
subcategory Dctf (V, R) consisting of complexes with finite Tor-dimension,
or by Dcb (V, Q ). All functors will be assumed to be derived.
We say a complex M ∈ Dcb (V ) is monodromic if M is monodromic in
the sense of Verdier [13] after base change to k̄. This is equivalent to the
existence of an integer n coprime to p and an isomorphism θ(n)∗ M ∼ = pr∗2 M
where θ(n) : Gm × V → V sends (λ, v) to λn v, and pr2 : Gm × V → V is the
projection [13, Proposition 5.1]. We denote the monodromic subcategory by
b
Dmon (V ). We recall the fact that π! ∼
= 0! on monodromic complexes (cf. [13,
Lemme 6.1] or [12, Proposition 1] for two different methods of proof), where
0 : S → V is the zero section.
The category Dctf (Gm ) of étale sheaves is monoidal with respect to
convolution with compact support, which is defined by

L ∗ K = m! (L  K)

where m : Gm × Gm → Gm is multiplication, and L, K ∈ Dctf (Gm ). This

monoidal category acts on Dcb (V ) by

L ∗ M = θ(1)! (L  M )

where θ(1) : Gm × V → V is the action map, L ∈ Dctf (Gm ), and M ∈ Dcb (V ).

 A new Fourier transform 1543

2. The functor FourB and its square

Let u : A1 − {1} → A1 be the open embedding removing 1 ∈ A1 (k), and let

j : A1 − {0} → A1 be the open embedding removing zero. Define

B = u∗ R ∈ Dctf (A1 ).

One observes that h! B = 0 where h : A1 → Spec k is the structure map, and

0∗ B ∼
= R where 0 : Spec k → A1 .
Define FourV /S,B : Dcb (V ) → Dcb (V ∨ ) by

FourV /S,B (M ) = pr∨ ∗ ∗

! (pr M ⊗ μ B)[d]

where pr∨ : V ∨ ×S V → V ∨ and pr : V ∨ ×S V → V are the projections and

μ : V ∨ ×S V → A1 is the natural pairing (ξ, v) → v, ξ . This is the new
Fourier transform that we will consider. Our goal in this section is to prove
the following theorem.

Theorem 2.1. There is a canonical isomorphism

FourV ∨ /S,B ◦ FourV /S,B (M ) ∼

= j ∗ B ∗ M (−d)[1].

for M ∈ Dcb (V ).

Let pr , pr : V ×S V → V be the first and second projections, respec-

tively, and prij the projection from V ×S V ∨ ×S V to the product of the
i’th and j’th factor. The usual formal argument shows that FourV ∨ /S,B ◦
FourV /S,B is isomorphic to the functor M → pr! (pr∗ M ⊗ K) where

K = pr13! (pr∗12 μ∗ B ⊗ pr∗23 μ∗ B)[2d].

We claim there exists a canonical isomorphism

(2.1.1) K∼
= ρ! pr∗1 j ∗ B(−d)[1]

where ρ : Gm × V → V ×S V is defined by (λ, v) → (λv, v), and pr1 : Gm ×

V → Gm is the natural projection. This claim implies the theorem since
pr! (pr∗ M ⊗ ρ! pr∗1 j ∗ B) ∼
= j ∗ B ∗ M by the projection formula.
We first establish two lemmas which will help us prove the claim.

Lemma 2.2. If v, w ∈ V (k̄) are not in the same Gm -orbit, then the stalk
K(v,w) equals 0.
1544 Jonathan Wang

Proof. We can assume S = Spec k̄. Clearly v and w cannot both be zero;
we will assume v = 0. Since v, w are not in the same Gm -orbit, there exists
ξ ∈ V ∨ (k̄) such that w, ξ = 0 and v, ξ =
 0. Let v : V ∨ → A1k̄ denote the
evaluation by v map. Split V ∨ as k̄ξ ⊕ Hv where Hv = (k̄v)⊥ . With respect
to this decomposition, v ∗ B ⊗w ∗ B ∼ = B (w |Hv )∗ B. Then by Künneth
formula,

π!∨ (v ∗ B ⊗w ∗ B) ∼

= h! B ⊗(π ∨ |Hv )! (w |Hv )∗ B = 0

Therefore K(v,w) = 0. 

Lemma 2.3. There is a canonical isomorphism

J ∗K ∼
= J ∗ ρ! pr∗1 j ∗ B(−d)[1]

where J : V ×S V − 0(S) → V ×S V is the open embedding removing zero.

Proof. We use V ◦ to denote V − 0(S). In this proof we will use ρ to denote

the restricted morphism Gm × V ◦ → V ×S V , which is an immersion, and
pr1 : Gm × V ◦ → Gm to denote the projection. From Lemma 2.2 we know
that J ∗ K is supported on the image of ρ. Thus it suffices to consider ρ∗ J ∗ K.
Define
ω : G m × V ∨ × V ◦ → G m × A1 × V ◦
S
by sending (λ, ξ, v) to (λ, v, ξ , v). Then

ρ∗ J ∗ K ∼
= pr13! ω! ω ∗ pr∗12 (m∗ B ⊗ p∗2 B)[2d]

where pr13 , pr12 are projections from Gm × A1 × V ◦ and m, p2 : Gm × A1 →

A1 are the multiplication and projection maps. Since ω is in fact a vector
bundle of rank d − 1, we see that ω! R is isomorphic to R(1 − d)[2 − 2d].
Therefore the projection formula implies that

ρ∗ J ∗ K ∼
= pr13! pr∗12 (m∗ B ⊗ p∗2 B)(1 − d)[2].

We have a Cartesian square

pr12
G m × A1 × V ◦ / G m × A1

pr13 id × h
 pr1 
Gm × V ◦ / Gm
 A new Fourier transform 1545

so proper base change gives pr13! pr∗12 ∼

= pr∗1 (id × h)! . We have an exact tri-
angle
R → p∗2 B → (id × 1)! R(−1)[−1]

where 1 : Spec k → A1 is the complement of u. Since (id × h)! (m∗ B) = 0 by

a change of variables, we deduce that

(id × h)! (m∗ B ⊗ p∗2 B) ∼

= (id × h)! (m∗ B ⊗(id × 1)! R)(−1)[−1]
∼
= j ∗ B(−1)[−1].

Now it follows that ρ∗ J ∗ K ∼

= pr∗1 j ∗ B(−d)[1]. 

Proof of Theorem 2.1. The case d = 0 is obvious since h! B = 0 and 0∗ B ∼ =

R. From now on we will assume that d > 0. We will show that both sides
of (2.1.1) are in the essential image of the functor τ≤0 J∗ J ∗ , i.e., there are
isomorphisms

K∼
= τ≤0 J∗ J ∗ (K) and ρ! pr∗1 j ∗ B(−d)[1] ∼
= τ≤0 J∗ J ∗ (ρ! pr∗1 j ∗ B(−d)[1]).

The claimed existence of an isomorphism (2.1.1) will then follow from

Lemma 2.3.
A stalk computation shows that ρ! pr∗1 j ∗ B(−d)[1] lives in non-positive
degrees. We claim that the natural morphism

(2.3.1) ρ! pr∗1 j ∗ B(−d)[1] → τ≤0 (J∗ J ∗ ρ! pr∗1 j ∗ B(−d)[1])

is an isomorphism. Let 0 : S → V ×S V denote the zero section. From the

exact triangle 0! 0! → id → J∗ J ∗ , it suffices to show that 0! ρ! pr∗1 j ∗ B ∈ Dc>2 (S).
Observe that ρ is Gm -equivariant with respect to the Gm -action on the
second coordinate of Gm × V and the diagonal action of Gm on V ×S V .
This implies that ρ! pr∗1 j ∗ B is monodromic. Thus

0! ρ! pr∗1 j ∗ B ∼
= h! j! j ∗ B(−d)[−2d] ∼
= R(−d)[−2d − 1].

Therefore 0! ρ! pr∗1 j ∗ B ∈ Dc>2d (S).

One easily sees that K(0,0) ∼ = R(−d). Thus K lives in non-positive coho-
mological degrees. To show that the natural morphism K → τ≤0 J∗ J ∗ K
is an isomorphism, it suffices by the same argument as above to prove
0! K ∈ Dc≥2d (S). One observes from the definition of K that K is monodromic
1546 Jonathan Wang

with respect to the diagonal Gm -action on V ×S V . Therefore

0! K ∼
=π! (pr∗12 μ∗ B ⊗ pr∗23 μ∗ B)[2d]

 : V ×S V ∨ ×S V → S is the structure map. By projection formula

where π
and proper base change, the right hand side is isomorphic to

π! (μ∗ B ⊗ pr∨,∗ pr∨ ∗

! μ B)[2d]

for π  : V ×S V ∨ → S the structure map. The fact that h! B = 0 implies

that pr∨ ∗ ∨ ∗ ∨ ∗ ∼
! μ B is supported at 0(S) ⊂ V , and 0 pr! μ B = R(−d)[−2d]. We
deduce that
0! K ∼
= π! R(−d) ∼
= R(−2d)[−2d],
which proves the claim, and hence the theorem. 

3. Properties of FourB

Remark 3.1. The functor FourV /S,B is not an equivalence on Dcb (V ) →

Dcb (V ∨ ). Consider the one-dimensional case V = A1S . Then FourV /S,B (0! R) =
R[1] and FourV /S,B (1! R) = B[1]. We have Hom(R, B) = 0 but Hom(0! R,
1! R) = 0, so FourV /S,B is not fully faithful.

3.2. Relation to quotient stacks

Let p : V → V = [V /Gm ] and p∨ : V ∨ → V∨ = [V ∨ /Gm ] denote the canoni-

cal projections to the quotient stacks. By [10, Lemme 3.2], Laumon’s homo-
geneous transform FourV/S : Dcb (V) → Dcb (V∨ ) is canonically isomorphic to
the functor

(3.2.1) K → pr∨ ∗ ∗
! (pr K ⊗ μ f! BS )[d]

where f : A1S → AS is the quotient morphism and BS denotes the base

change of B from A1k to A1S . We abuse notation and use pr∨ : V∨ ×S V →
V∨ , pr : V∨ ×S V → V, and μ : V∨ ×S V → AS to also denote the induced
maps on stacks.

Proposition 3.3. The composed functors

(p∨ )∗ ◦ FourV/S and FourV /S,B ◦p∗ : Dcb (V) → Dcb (V ∨ )

are canonically isomorphic.

 A new Fourier transform 1547

Proof. The proposition follows from (3.2.1) by applying proper base change
to the Cartesian squares

[V ∨ ×S V /Gm ] / A1 V ∨ ×S V / [V ∨ ×S V /Gm ]

f
   
V∨ ×S V
μ
/ AS V∨ / V∨

where Gm acts on V ∨ ×S V anti-diagonally. 

Proposition 3.4. Let V  = V × A1 and let Gm act on both V and A1 . We

have a canonical open embedding ν : V → [V  /Gm ] : v → (v, 1). Similarly,
we have ν ∨ : V ∨ → [(V  )∨ /Gm ] defined by ν ∨ (ξ) = (ξ, −1). The composed
functor
Four[V  /Gm ]/S ∨ ∗

Dcb (V )
ν! / D b ([V  /Gm ]) / D b ([(V  )∨ /Gm ]) (ν ) / D b (V ∨ )
c c c

is isomorphic to FourV /S,B .

Proof. Observe that ν factors into the composition of an open affine chart
V → P(V  ) and the open embedding P(V  ) = [(V  − 0(S))/Gm ] → [V  /Gm ].
Similarly, we have a factorization of ν ∨ . The proposition now follows from
[10, Proposition 1.6], since the restriction of the incidence hyperplane in
P((V  )∨ ) ×S P(V  ) to V ∨ ×S V is μ−1 ({1}). 

3.5. An equivalence induced by FourV /S,B

Let p : V → V be as in the previous subsection.

Proposition 3.6. Let CV denote the full subcategory of Dcb (V ) consist-

ing of complexes M such that p! M = 0. The functor FourV /S,B induces an
equivalence CV → CV ∨ .

Proof. Proper base change and projection formula imply that FourV /S,B
sends CV to CV ∨ and vice versa. We also see by proper base change that
p∗ p ! M ∼
= R ∗ M for M ∈ Dcb (V ), where R is the constant sheaf on Gm .
From the exact triangle 1! R(−1)[−2] → R → B we deduce that j ∗ B ∗ M ∼
=
M (−1)[−1] for M ∈ CV . Therefore Theorem 2.1 implies that

FourV ∨ /S,B ◦ FourV /S,B (M ) ∼

= M (−d − 1)

for M ∈ CV , and we deduce the proposition. 

1548 Jonathan Wang

3.7. Monodromic complexes

We will show that FourV /S,B also induces an equivalence on the subcategories
of monodromic complexes. We use the notation and results of Appendix A.

Theorem 3.8. (i) The functor FourV /S,B preserves monodromicity, and
b
the restriction defines an equivalence Dmon (V ) → Dmon
b (V ∨ ).
(ii) For N ∈ Dmon
b (V ∨ ), the pro-object

(3.8.1) pr! (pr∨,∗ N ⊗ μ∗ j∗ I 0 )(d + 1)[d + 1]

is essentially constant1 .
b
(iii) The functor from Dmon (V ∨ ) to Dmon
b (V ) defined by (3.8.1) is quasi-
inverse to FourV /S,B .

Since B is not monodromic, our first step is to compute the “mon-

odromization” of B.

Lemma 3.9. There is an isomorphism of pro-objects

I0 ∗ B ∼
= j∗ I 1 (−1)[−1].

Proof. First we show that the restriction I 0 ∗j ∗ B is isomorphic to I 1 (−1)[−1].

The exact triangle 1! R(−1)[−2] → R → B induces by convolution exact tri-
angles
In0 (−1)[−2] → In0 ∗ R → In0 ∗ j ∗ B
for p  n. Taking “lim” and using Lemma A.4, the first arrow is isomorphic
←−
to the augmentation map I 0 (−1)[−2] → R(−1)[−2]. Therefore we deduce
that the pro-object I 0 ∗ j ∗ B is isomorphic to I 1 (−1)[−1].
To complete the proof, it suffices to show that the canonical morphism

I 0 ∗ B → j∗ j ∗ (I 0 ∗ B)

is an isomorphism. This is equivalent to proving that 0! (I 0 ∗ B) = 0. Since

I 0 ∗ B is monodromic, 0! (I 0 ∗ B) ∼
= h! (I 0 ∗ B). By the Kunneth formula,
h! (I ∗ B) ∼
0
= h! j! I ⊗ h! B = 0.
0 

1 b
A pro-object is essentially constant if it is isomorphic to an object of Dmon (V ),
which is considered as a pro-object via the constant embedding.
 A new Fourier transform 1549

Proof of Theorem 3.8. One easily sees that FourV /S,B preserves monodromic-
ity. Theorem 2.1 and Lemma A.4 together imply that for M ∈ Dmon b (V ), we
have
FourV ∨ /S,B ◦ FourV /S,B (M ) ∼
= I 1 ∗ M (−d)[2].
Since I −1 ∗ I 1 ∼
= I 0 (−1)[−2] by Corollary A.9, we deduce that FourV /S,B is
an equivalence, with inverse functor I −1 ∗ FourV ∨ /S,B (d + 2)[2]. Lemmas 3.9
and A.4 imply that for N ∈ Dmon b (V ∨ ), we have isomorphisms

I −1 ∗ FourV ∨ /S,B (N ) ∼
= I −1 ∗ pr! (pr∨,∗ N ⊗ μ∗ j∗ I 1 )[d + 1].

Applying Corollary A.9 again, we get (iii). 

Remark 3.10. Observe that the formula (3.8.1) is very similar to Beilin-
son’s suggested definition of the monodromic Fourier transform in [4].

Proposition 3.11. The object j ∗ B ∈ Dctf (Gm ) satisfies the following prop-
erties:
1) j ∗ B is not invertible in the monoidal category Dctf (Gm ).
2) j ∗ B is invertible in the quotient of Dctf (Gm ) by the ideal generated by
the constant sheaf R.
3) There are canonical isomorphisms In0 ∗ j ∗ B ∼
= In1 (−1)[−2] for p  n.

Proof. We showed in Remark 3.1 that FourA1 ,B is not an equivalence on

Dcb (A1 ). Since Four2A1 ,B (M ) is isomorphic to j ∗ B ∗ M (−1)[1], we deduce
that j ∗ B is not invertible in the monoidal category Dctf (Gm ).
From the exact triangle 1! R(−1)[−2] → R → j ∗ B on Gm , we see that
in the quotient of Dctf (Gm ) by the ideal generated by R, the object j ∗ B is
isomorphic to 1! R(−1)[−1], which is invertible.
Lemma 3.9 gives an isomorphism I 0 ∗ j ∗ B ∼ = I 1 (−1)[−2]. Convolving
with In0 , we get an isomorphism In0 ∗ j ∗ B ∼
= In0 ∗ I 1 . One observes that In0 ∗
1 ∼ 1
I = In (−1)[−2] by Corollary A.9. 

4. Relation to Fourier-Deligne transform

Suppose that k has characteristic p > 0. Assume that R contains a primitive

p-th root of unity ζ (where “primitive” means that ζ − 1 is invertible). Let
ψ : Fp → R× be the corresponding additive character with ψ(1) = ζ, and
1550 Jonathan Wang

let Lψ denote the Artin-Schreier sheaf. The usual Fourier-Deligne transform

FourV /S,Lψ : Dcb (V ) → Dcb (V ∨ ) is defined by

FourV /S,Lψ (M ) = pr∨ ∗ ∗

! (pr M ⊗ μ Lψ )[d].

Lemma 4.1. There is a canonical isomorphism

ι ∗ j ∗ Lψ ∗ Lψ ∼
= B[−1]

where ι : Gm → Gm sends λ → −λ−1 .

Proof. By a change of variables, ι∗ j ∗ Lψ ∗ Lψ is isomorphic to

FourA1 ,Lψ−1 (j! j ∗ Lψ )[−1].

We have an exact triangle

j ! j ∗ Lψ → Lψ ∼
= FourA1 ,Lψ (1! R[−1]) → 0∗ R.

Applying FourA1 ,Lψ−1 and using the Fourier-Deligne inversion formula on

the middle term, we have an exact triangle

FourA1 ,Lψ−1 (j! j ∗ Lψ ) → 1! R(−1)[−1] → R[1].

This induces an isomorphism FourA1 ,Lψ−1 (j! j ∗ Lψ ) → u∗ R = B. This isomor-

phism is unique since Hom(1! R(−1)[−1], R) = 0. 

Corollary 4.2. In characteristic p > 0, we have a canonical isomorphism

FourV /S,B (M ) ∼
= ι∗ j ∗ Lψ ∗ FourV /S,Lψ (M )[1].

4.3. Monodromization of Lψ over k̄

We use the notation and results of Appendix A. Suppose that k is alge-

braically closed, so A0 is simply a ring instead of a sheaf of rings (i.e., there
is no Galois action).

Lemma 4.4. There exists a (non-canonical) isomorphism of pro-objects

I 0 ∗ Lψ ∼
= j∗ I 0 [−1].
 A new Fourier transform 1551

Proof. As in the proof of Lemma 3.9, it suffices to prove the isomorphism

after restriction to Gm . Let n be coprime to p. By proper base change,

1∗ (In0 ∗ j ∗ Lψ ) ∼
= Γc (Gm , In0 ⊗ j ∗ Lψ )
R

where we observe that the pullback of In0 under the multiplicative inverse
map Gm → Gm is isomorphic to In0 . Since In0 is tamely ramified at ∞ ∈ P1 (k),
the canonical map

Γc (A1 , j! In0 ⊗ Lψ ) → Γ(A1 , j! In0 ⊗ Lψ )

is an isomorphism (cf. proof of [9, Lemma 7.1(1)]). In particular Γc (Gm , In0 ⊗

j ∗ Lψ ) lives in cohomological degrees 0 and 1. Since In0 ⊗ j ∗ Lψ is locally
constant and Gm is not complete, Hc0 (Gm , In0 ⊗ j ∗ Lψ ) = 0. Thus Γc (Gm , In0 ⊗
j ∗ Lψ ) lives only in cohomological degree 1.
We now consider In0 as a locally free sheaf of A0n -modules of rank 1.
If we let ψ  denote the composition Fp → R× → (A0n )× , then Lψ ⊗R A0n ∼ =
Lψ , where the latter is the Artin-Schreier sheaf with respect to ψ  as a
locally free sheaf of A0n -modules of rank 1. Hence F := In0 ⊗A0n j ∗ Lψ , which
is isomorphic to In0 ⊗R j ∗ Lψ , is a locally free sheaf of A0n -modules of rank 1.
In particular, F ∈ Dctf (Gm , A0n ) and Γc (Gm , F)[1] is quasi-isomorphic to a
finite projective A0n module P . Applying the Grothendieck-Ogg-Shafarevich
formula [2, Exposé X, Corollaire 7.2], one checks that the fiber of P over any
point of Spec A0n has dimension 1. So there exists an isomorphism P ∼ = A0n
0
of An -modules. Observe from the Cartesian square

θ(n) × idGm
Gm × Gm × Gm // G × G
m m
pr2 × idGm
idGm × m m
 θ(n) 
Gm × Gm // G
m
pr2

that In0 ∗ j ∗ Lψ is monodromic, and the monodromy action is induced by the

monodromy action on In0 . Hence by Corollary A.7, there exists an isomor-
phism In0 ∗ j ∗ Lψ [1] ∼
= In0 .
Suppose n is a multiple of n and p  n . The kernel K of the surjection
In  In0 is tamely ramified, so Hc2 (Gm , K ⊗ j ∗ Lψ ) = 0 by the same argu-
0

ment as above. We deduce that

In0 ∗ j ∗ Lψ [1] → In0 ∗ j ∗ Lψ [1]

1552 Jonathan Wang

is a surjection of sheaves. Since (A0n )× → (A0n )× is also surjective, we can

find a projective system of isomorphisms In0 ∗ j ∗ Lψ [1] ∼
= In0 inducing an iso-
morphism of pro-sheaves. 

Corollary 4.5. When k is algebraically closed, there exists a (non-canonical)

isomorphism between the functors FourV /S,B and FourV /S,Lψ restricted to
b
Dmon (V ) → Dmon
b (V ∨ ).

Proof. Lemma 3.9 and Remark A.3 imply that there exists an isomorphism
I0 ∗ B ∼
= j∗ I 0 [−1]. The latter is also isomorphic to I 0 ∗ Lψ by Lemma 4.4.
One easily sees that the Fourier-Deligne transform preserves monodromicity,
and the isomorphism of restricted functors follows from Lemma A.4. 

4.6. The universal Gauss sum

Let k once again be arbitrary. Define the pro-object

G = I 0 ∗ j ∗ Lψ (1)[1].

Lemma 4.4 implies that G is a monodromic pro-sheaf, and there exists a

trivialization G ∼
= I 0 after base changing from k to k̄. Under the equivalence
of abelian categories in Corollary A.7, we see that G corresponds to an
invertible (locally free of rank 1) A0 -module on Spec k. We are motivated
by [6, Exposé VI, §4] to think of G as a “universal Gauss sum”.
Define ι : Gm → Gm by ι(λ) = −λ−1 . Then Lemmas 3.9 and 4.1 give a
canonical isomorphism
ι∗ G ∗ G ∼
= I 1 [−2].
We also see that the Fourier-Deligne transform on monodromic complexes
is isomorphic to the functor M → pr∨ ∗ ∗
! (pr M ⊗ μ j∗ G)[d + 1] on Dmon (V ) →
b
b
Dmon (V ∨ ). By Corollary 4.2, we have

FourV /S,B (M ) ∼
= ι∗ G ∗ FourV /S,Lψ (M )[2].

for M monodromic.

5. Relation to Fourier transform on D-modules

Let k be algebraically closed of characteristic 0. We use M(V ) to denote the

abelian category of quasicoherent right D-modules on V . Let L = DA1 /(1 −
 A new Fourier transform 1553

∂x )DA1 be the exponential D-module on A1 = Spec k[x]. The Fourier trans-

form is the functor DM(V ) → DM(V ∨ ) defined by

!
FourV /S,L (M ) = pr∨
∗ (pr M ⊗ μ L)[1 − d].
! !

It is well known [8, Lemme 7.1.4] that this functor can also be described using
the isomorphism between the algebras of polynomial differential operators
DV ∨ → DV defined in local coordinates by

k[ξ1 , . . . , ξd , ∂ξ1 , . . . , ∂ξd ] → k[v1 , . . . , vd , ∂v1 , . . . , ∂vd ] : ξi → −∂vi , ∂ξi → vi .

In the D-module situation, the analog of B is u! u! (ωA1 ), where ωA1 is

the sheaf of differentials on A1 viewed as a right D-module. We will also call
this D-module B. A simple calculation shows that2

B = k[x, ∂x ]/∂x (x − 1)k[x, ∂x ].

We define FourV /S,B : DM(V ) → DM(V ∨ ) by

!
FourV /S,B (M ) = pr∨
∗ (pr M ⊗ μ B)[1 − d].
! !

Consider DM(Gm ) with the monoidal structure induced by convolution

without compact support L ∗ K := m∗ (L  K). This monoidal category acts
on DM(V ) by L ∗ M = θ(1)∗ (L  M ). The proof of Lemma 4.1 can be easily
modified to prove the following analog of the lemma and Corollary 4.2.

Proposition 5.1. There is a canonical isomorphism

ι∗ j ∗ L ∗ L ∼
=B

where ι : Gm → Gm sends λ → −λ−1 . Consequently, we have a canonical

isomorphism
FourV /S,B (M ) ∼
= ι∗ j ∗ L ∗ FourV /S,L (M ).

2
Beilinson observed that B essentially describes the differential equation for a
shift of the Heaviside step function.
1554 Jonathan Wang

5.2. Mellin transform of j ∗ B

Let B denote the Mellin transform of j ∗ B, viewed as a Z-equivariant qua-

sicoherent O-module on A1 = Spec k[s]. The Mellin transform functor

M : M(Gm ) → QCoh(A1 )Z

is defined by considering D(Gm ) as the algebra of difference operators

D = k[s]T, T −1 /(sT − T (s + 1))

under the identifications s = x∂x and T = x. We consider the derived cate-

gory of Z-equivariant OA1 -modules D(QCoh(A1 )Z ) with monoidal structure
induced by the usual derived tensor product over k[s]. This monoidal struc-
ture corresponds to the convolution product on DM(Gm ). More precisely,
M(L ∗ K) ∼ = M(L) ⊗k[s] M(K).
We start by proving the following proposition, which is an analog of
Proposition 3.11 in the D-module setting.

Proposition 5.3. The module B satisfies the following properties:

1) B is not invertible in D(QCoh(A1 )Z ).
2) The restriction of B to A1 − Z := Spec k[s][s−1 , (s ± 1)−1 , . . . ] is invert-
ible.
3) For any χ ∈ k and n ∈ N, there exists an isomorphism
 
k[s]/(s − χ − i)n ∼
= B ⊗ k[s]/(s − χ − i)n
k[s]
i∈Z i∈Z

of D-modules, where T acts on k[s] by translation.

In order to prove the proposition, we will need an explicit description

of B. Consider k(s) as a right D-module where T acts by translation. Let
B denote the D-submodule of k(s) generated by 1s , or equivalently, the
1
k[s]-submodule generated by s+i for all i ∈ Z.

Lemma 5.4. There exists an isomorphism of D-modules B ∼

= B .
 A new Fourier transform 1555

Proof. We have ∂x x = x∂x + 1 so ∂x (x − 1) = (s + 1) − T −1 s in D. There-

fore
B = D/((s + 1) − T −1 s)D.
Let 1 denote the generator of B. Conjugating sT = T (s + 1) in D by T −1
gives T −1 s = (s + 1)T −1 in D. Using this equality, 1(s + 1) = 1T −1 s = 1(s +
1)T −1 in B, and acting on the right by T gives 1(s + 1)T = 1(s + 1). Using
these relations, we deduce that B is generated over k by 1T i for i ∈ Z and
1sj for j > 0. Then 1 → s+1 1
defines a morphism of D-modules B → k(s).
Since s+i for i ∈ Z and s for j ≥ 0 are k-linearly independent in k(s), we
1 j

see that this morphism is an injection B → k(s). The image is B . 

Proof of Proposition 5.3. Suppose that B is invertible in D(QCoh(A1 )Z ),
i.e., there exists an object N of this monoidal category such that B ⊗k[s] N ∼
=
k[s]. Then N ∼ = Homk[s] (k[s], N ) ∼
= Homk[s] (B, k[s]). There are no nonzero
morphisms from B to k[s], so H 0 N = 0. On the other hand, since k(s) ⊗k[s]
B ∼ = k(s), we have k(s) ⊗k[s] N ∼ = k(s), which implies that H 0 N = 0. We
thus get a contradiction, so B is not invertible.
Since O(A1 − Z) = k[s][s−1 , (s ± 1)−1 , . . . ] ⊂ k(s), we see that

O(A1 − Z) ⊗ B = O(A1 − Z) ⊂ k(s)

k[s]

is the identity object, proving (2).

The direct sums in (3) only depend on the class χ of χ in k/Z. If χ = 0 +
Z we will assume that χ = 0. Let Bi ⊂ B denote the k[s]-submodule gener-
ated by s−i1
. Then B /Bi is isomorphic to the direct sum of skyscraper mod-
ules k[s]/(s − j) for integers j = i. Thus (B /Bi ) ⊗k[s] k[s]/(s − χ − i)n = 0.
On the other hand Bi is free, so B ⊗k[s] k[s]/(s − χ − i)n is free with gen-
1
erator s−i ⊗ 1. These basis elements give our desired isomorphism, which
evidently commutes with the action of T . 

5.5. Monodromization

The Gm -action on V induces an algebra map k[s] → DV , where s = x∂x is

the invariant vector field on Gm . We say that M ∈ M(V ) is monodromic
if every local section m ∈ M is killed by some nonzero polynomial in s =
x∂x . In other words, M is monodromic if it is a torsion module over k[s].
This definition of monodromic was introduced by Verdier [14]. Define an
object of DM(V ) to be monodromic if each of its cohomology D-modules is
monodromic. We denote this full subcategory by Dmon M(V ) ⊂ DM(V ).
1556 Jonathan Wang

For any χ ∈ k and n ∈ N, let Aχ,n ⊂ k(s) consist of those rational func-
tions with poles of order ≤ n at χ + Z and no other poles. Define Iχ0,n ∈
M(Gm ) to be the inverse Mellin transform M−1 (Aχ,n /k[s]). The inclusions
Aχ,n → Aχ,n+1 induce morphisms Iχ0,n → Iχ0,n+1 , which form an inductive
system of D-modules. Define

I0 = lim Iχ0,n ∈ M(Gm )
−→
χ∈k/Z n

where χ ∈ k is any lift of χ. It follows that M(I 0 ) = k(s)/k[s].

Let 1 be the unit object in the monoidal category DM(Gm ), so M(1) =
k[s]. The canonical extension of k(s)/k[s] by k[s] defines an extension of I 0
by 1 and therefore a morphism

ε : I 0 → 1[1].

The monoidal category DM(Gm ) acts on DM(V ) by convolution (without

compact support).

Lemma 5.6. An object M ∈ DM(V ) is monodromic if and only if the mor-

phism I 0 ∗ M → M [1] induced by ε is an isomorphism.

Proof. A calculation using the relative de Rham complex with respect to

the action map Gm × V → V shows that for any M ∈ DM(V ) and N ∈
DM(Gm ), there is a canonical isomorphism N ∗ M ∼ = M(N ) ⊗k[s] M in the
derived category of (sheaves of) k[s]-modules. This implies that the cocone
of the morphism I 0 ∗ M → M [1] is isomorphic (in the derived category of
k[s]-modules) to k(s) ⊗k[s] M . But k(s) is flat over k[s], so the vanishing
of the cohomologies of k(s) ⊗k[s] M is equivalent to the cohomologies of M
being torsion modules over k[s]. 
See [3], [11], and [7, C.2] for further details in the unipotently mon-
odromic case (when χ = 1).

Lemma 5.7. There exists an inductive system of isomorphisms

= j! Iχ0,n ∼
Iχ0,n ∗ B ∼ = Iχ0,n ∗ L.

Proof. Since h∗ B = h∗ L = 0, it suffices as in Lemma 3.9 to give isomor-

phisms of the above objects after restriction to Gm . In fact, it suffices to
construct isomorphisms between the Mellin transforms of these restrictions,
 A new Fourier transform 1557

i.e., isomorphisms M(Iχ0,n ∗ j ∗ B) ∼

= M(Iχ ) ∼
0,n
= M(Iχ ∗ j ∗ L). This is equiv-
0,n

alent to constructing isomorphisms

(5.7.1) M(Iχ0,n ) ⊗ B ∼
= M(Iχ0,n ), B := M(j ∗ B),
k[s]
(5.7.2) M(Iχ0,n ) ⊗E∼
= M(Iχ0,n ), E := M(j ∗ L).
k[s]

Note that we have isomorphisms


(5.7.3) M(Iχ0,n ) = Aχ,n /k[s] ∼
= k[s]/(s − χ − i)n .
i∈Z

Combining (5.7.3) and Proposition 5.3(3), one gets (5.7.1). Let us con-
struct (5.7.2). We have

E = D/(1 − T −1 s)D.

Let 1 be the generator of E. Let Ei ⊂ E denote the free k[s]-submodule gen-

erated by 1T −i−1 for i ∈ Z. If χ ∈ Z, set χ = 0. From the relation 1T −i =
1T −i−1 (s − i), we deduce that E/Ei is supported away from χ + i, so
(E/Ei ) ⊗k[s] k[s]/(s − χ − i)n = 0. Hence E ⊗k[s] k[s]/(s − χ − i)n is freely
generated by 1T −i−1 ⊗ 1, and this gives us (5.7.2). 
Lemma 5.7 implies in particular that I 0 ∗ B ∼
= I 0 ∗ L. We deduce from
Lemma 5.6 that FourV /S,B agrees with FourV /S,L on Dmon M(V ).

Corollary 5.8. There is an isomorphism

FourV /S,B ∼
= FourV /S,L

of functors Dmon M(V ) → Dmon M(V ∨ ).

Appendix A. The monodromic subcategory

In this appendix we prove the facts we need about (non-unipotently) mon-

odromic complexes. For a more complete account of the unipotently mon-
odromic story, see [3, 5].

A.1. Free monodromic objects

Let p be the characteristic of k, which may be 0. For p  n, let A0n be the

group algebra R[μn ] considered as a sheaf on Spec k, i.e., a Gal(k̄/k)-module.
1558 Jonathan Wang

Put
A0 = lim A0n .
←−
p n

Consider T := limpn μn (k̄) the tame fundamental group of Gm,k̄ . For any
←−
γ ∈ T, let γ
 denote the corresponding invertible element in A0 (k̄). Pick a
topological generator t ∈ T. Note that t − 1 is not a zero divisor in A0 , so
0 0
A injects to the localization A = (A )t−1 . Define

Ai = (
t − 1)i A0 ⊂ A

for i ∈ Z and set Ain = Ai ⊗A0 A0n for p  n. The definition of Ai is indepen-
dent of the choice of t, and Ai is a Gal(k̄/k)-module. Note that A1 is the
kernel of the quotient map A0 → A01 = R.

Remark A.2. The ring A0 (k̄) is isomorphic to the product of the comple-
tions of R[t, t−1 ] at all maximal ideals m such that tn ≡ 1 mod m for some
p  n. The maximal ideals m correspond to the eigenvalues of the monodromy
action.

For i ∈ Z and p  n, let Ini be the local system on Gm such that the fiber
at 1 ∈ Gm (k) is Ain and the monodromy action of γ ∈ T is multiplication by
. We define I i to be the pro-sheaf
γ

“lim” Ini ,
←−
p n

where we use “lim” to denote pro-objects, following the notation of [1,

←−
Exposé I, (8.5.3.2)].

Remark A.3. After base change from Spec k to Spec k̄, the local systems
In0 and Ini are isomorphic via multiplication by (t − 1)i , and this induces an
isomorphism I 0 ∼ = I i . The isomorphism is not canonical as it depends on the
choice of t.

Lemma A.4. There is a canonical isomorphism of pro-objects

I0 ∗ M ∼
= M (−1)[−2]

for M ∈ Dmon
b (V ) considered as a constant pro-object.
 A new Fourier transform 1559

Proof. Let en : Gm → Gm denote the nth power map. Note that en! R ∼ = In0
for p  n. Since M is monodromic, there exists n0 coprime to p such that
θ(n0 )∗ M ∼= pr∗2 M . Then

“lim”(en! R) ∗ M ∼
= “lim ” θ(n)! pr∗2 M ∼
= M (−1)[−2],
←− ←−
where we use the fact that the pro-object “lim” Γc (Gm , R) is essentially
←−
constant and isomorphic to R(−1)[−2] (cf. [13, Lemme 5.2]). 

A.5. Monodromic sheaves as A0 -modules

Let Modτ (A0 ) denote the abelian category of sheaves of discrete A0 -modules
on Spec k, where A0 is equipped with the projective limit topology, and let
Sh(Gm ) denote the abelian category of sheaves of R-modules on Gm . We
have a canonical exact functor

Loc : Modτ (A0 ) → Sh(Gm ).

Define another functor M : Sh(Gm ) → Modτ (A0 ) by

M(F) = lim h∗ en,∗ e∗n F

−→
where h : Gm → Spec k is the structure map and A0 acts on en,∗ e∗n F by
transport of structure. We deduce from étale descent that Loc is left adjoint
to M. Passing to derived categories, the derived functors are still adjoint,
and we also denote them by

Loc : DModτ (A0 )  D(Gm ) : M.

Note that M : D(Gm ) → DModτ (A0 ) is equal to the composition of the

exact functor lim en,∗ e∗n with the derived functor h∗ .
−→
Proposition A.6. The derived functor Loc : DModτ (A0 ) → D(Gm ) is fully
faithful.

Proof. We need to show that the unit of adjunction L → M ◦ Loc(L) is

an isomorphism for L ∈ DModτ (A0 ). We can assume that k is algebraically
closed. Since Loc and M both commute with filtered colimits, we may further
suppose that L is a finite module concentrated in degree 0. Then there exists
n0 not divisible by p such that the action of A0 on L factors through R[μn0 ].
If n is a multiple of n0 then e∗n Loc(L) ∼
= L, where L is the constant sheaf
1560 Jonathan Wang

on Gm with stalk L. The proposition now follows from the fact that for any
finite abelian group L of order prime to p, one has lim H 0 Γ(Gm , en,∗ e∗n L) ∼
=L
−→
and lim H i Γ(Gm , en,∗ e∗n L) = 0 for i = 0. 
−→
Corollary A.7. The restriction of Loc induces an equivalence between
the subcategory of Db Modτ (A0 ) consisting of complexes whose cohomology
b
sheaves have finite stalks and Dmon (Gm ). Taking hearts with respect to the
standard t-structures of the above triangulated categories, we get an isomor-
phism between the abelian category of sheaves of A0 -modules on Spec k with
finite stalk and the abelian category of monodromic sheaves on Gm .

The monoidal structure on DModτ (A0 ) with respect to (derived) tensor

product over A0 corresponds under Loc to convolution on D(Gm ).

Lemma A.8. For L, K ∈ DModτ (A0 ) there exists a canonical isomor-

phism
Loc(L) ∗ Loc(K) ∼
= Loc(L ⊗ K)(−1)[−2].
A0

 R A0 ) → D(Gm × Gm ),
Proof. Consider the functor LocGm × Gm : DModτ (A0 ⊗
which is defined similarly to the above functor Loc = LocGm . Applying
LocGm × Gm to the natural map L ⊗R K → L ⊗A0 K, we get a map

Loc(L)  Loc(K) → m∗ Loc(L ⊗ K)

A0

in D(Gm × Gm ). Recall that since m is smooth, m∗ Loc(L ⊗A0 K) is isomor-

phic to m! Loc(L ⊗A0 K)(−1)[−2]. Therefore the (m! , m! )-adjunction induces
a morphism
Loc(L) ∗ Loc(K) → Loc(L ⊗ K)(−1)[−2].
A0
To check this is an isomorphism, we can assume k is algebraically closed
and take L = K = A0n for p  n since the functors on both sides commute
with filtered colimits and have finite cohomological amplitude. Under these
assumptions, the isomorphism is an easy computation. 

Corollary A.9. There is a canonical projective system of isomorphisms

I i ∗ Inj ∼
= Ini+j (−1)[−2]

for p  n and any integers i and j. Consequently there is an isomorphism of

pro-objects
Ii ∗ Ij ∼
= I i+j (−1)[−2].
 A new Fourier transform 1561

Proof. Fix p  n. By Lemma A.8, the first isomorphism is equivalent to an

isomorphism

“lim” Aim ⊗ Ajn ∼

= Ai+j
←− A 0
n
p m

as pro-objects in DModτ (A0 ). Remark A.3 and Lemma A.4 imply that it
suffices to consider the cohomology in degree 0, i.e., we consider the non-
derived tensor product on the LHS. Then H 0 (Aim ⊗A0 Ajn ) ∼ i+j
= An for n | m
by definition. These isomorphisms are evidently compatible with changes in
n, so the rest of the corollary follows. 

Acknowledgements

The research was partially supported by the Department of Defense (DoD)

through the NDSEG fellowship. The author is very thankful to Sasha Beilin-
son and Vladimir Drinfeld for many helpful discussions. The definition of
FourB was first suggested by Drinfeld.

References

[1] Théorie des topos et cohomologie étale des schémas. Tome 1: Théorie
des topos, Lecture Notes in Mathematics, Vol. 269, Springer-Verlag,
Berlin-New York (1972). Séminaire de Géométrie Algébrique du Bois-
Marie 1963–1964 (SGA 4), Dirigé par M. Artin, A. Grothendieck, et
J. L. Verdier. Avec la collaboration de N. Bourbaki, P. Deligne et B.
Saint-Donat.
[2] Cohomologie l-adique et fonctions L, Lecture Notes in Mathematics,
Vol. 589, Springer-Verlag, Berlin (1977), ISBN 3-540-08248-4. Séminaire
de Géometrie Algébrique du Bois-Marie 1965–1966 (SGA 5), Edité par
Luc Illusie.
[3] A. Beilinson, How to glue perverse sheaves. In: K-theory, arithmetic and
geometry (Moscow, 1984–1986), Vol. 1289 of Lecture Notes in Math.,
42–51, Springer, Berlin (1987).
[4] A. Beilinson, A remark on primitive cycles and Fourier-Radon trans-
form. In: Regulators, Vol. 571 of Contemp. Math., 19–23, Amer. Math.
Soc., Providence, RI (2012).
[5] R. Bezrukavnikov and Z. Yun, On Koszul duality for Kac-Moody groups.
1562 Jonathan Wang

[6] P. Deligne, Cohomologie étale, Lecture Notes in Mathematics, Vol. 569,

Springer-Verlag, Berlin (1977). Séminaire de Géométrie Algébrique
du Bois-Marie SGA 4 12 , Avec la collaboration de J. F. Boutot, A.
Grothendieck, L. Illusie et J. L. Verdier.
[7] V. Drinfeld and D. Gaitsgory, Compact generation of the category of
D-modules on the stack of G-bundles on a curve. arXiv:1112.2402v6.
[8] N. M. Katz and G. Laumon, Transformation de Fourier et majoration
de sommes exponentielles. Inst. Hautes Études Sci. Publ. Math., (1985),
no. 62, 361–418.
[9] R. Kiehl and R. Weissauer, Weil conjectures, perverse sheaves and
l’adic Fourier transform. Vol. 42 of Ergebnisse der Mathematik und
ihrer Grenzgebiete. 3. Folge. A Series of Modern Surveys in Mathemat-
ics [Results in Mathematics and Related Areas. 3rd Series. A Series of
Modern Surveys in Mathematics], Springer-Verlag, Berlin (2001), ISBN
3-540-41457-6.
[10] G. Laumon, Transformation de Fourier homogène. Bull. Soc. Math.
France, 131 (2003), no. 4, 527–551.
[11] S. Lichtenstein, Vanishing cycles for algebraic D-modules. Harvard
senior thesis, March 2009.
[12] T. A. Springer, A purity result for fixed point varieties in flag manifolds.
J. Fac. Sci. Univ. Tokyo Sect. IA Math., 31 (1984), no. 2, 271–282.
[13] J.-L. Verdier, Spécialisation de faisceaux et monodromie modérée. In:
Analysis and topology on singular spaces, II, III (Luminy, 1981), Vol.
101 of Astérisque, 332–364, Soc. Math. France, Paris (1983).
[14] J.-L. Verdier, Prolongement des faisceaux pervers monodromiques.
Astérisque (1985), no. 130, 218–236. Differential systems and singu-
larities (Luminy, 1983).

Department of Mathematics, University of Chicago

5734 S. University Avenue, Room 208C, Chicago, Illinois 60637, USA
E-mail address: [email protected]

Received April 17, 2014