PROBABILITY LAWS CONCERNING ZETA INTEGRALS

GRAYSON PLUMPTON
arXiv:2411.08863v1 [math.NT] 13 Nov 2024

√
Abstract.
√ We give a probabilistic interpretation of the Dedekind zeta functions of Q( −1) and
Q( −2) using zeta integrals and use this to show that the first two Li coefficients of these zeta
functions are positive. This extends a result of Biane, Pitman, and Yor (2001) which considered
the case of the Riemann zeta function.

1. Preliminaries
1.1. Introduction. A central tool when studying the Riemann zeta function is its representation
as a MellinR transform. Recall that the Mellin transform of a Schwartz function f is the integral
∞
transform 0 ts f (t) dtt , where s ∈ C. This integral also has a probabilistic interpretation: if f (t) dtt
is a probability measure on (0, ∞) induced by the random variable Y , then the Mellin transform
of f is E(Y s ). This is the central idea behind [1], in which the Riemann zeta function is studied by
expressing ξ(s) = s(s − 1)π −s/2 Γ(s/2)ζ(s) as a Mellin transform, then interpreting it as being the
sth moment of a random variable. In this paper, we use zeta integrals, essentially Mellin transforms
√
over ×
√ the locally compact abelian group A , to extend this construction to the fields Q( −1) and
Q( −2). Our main result is the following theorem:
√ √
Theorem 1.1. Let K be one of the fields Q, Q( −1), or Q( −2). Then there is a random variable
X such that, for every s ∈ C, E(X s ) = |D|−s/2 ξK (s), where D is the discriminant of K and ξK is
the Dedekind xi function.
Using our random variable X, we apply some basic properties of √ cumulants to
√ prove that the first
two Li coefficients, introduced in [2], are positive for the fields Q( −1) and Q( −2). A remarkable
result of [2] is that the positivity of all Li coefficients of a Dedekind zeta function implies that all
of its nontrivial zeroes lie on the critical line.

1.2. Defining the Dedekind zeta and xi functions (the classical perspective). The follow-
ing is well known background, details can be found in [2] or [3]. Let K be a number field with r1
real and r2 complex embeddings. For Re(s) > 1, define the Dedekind zeta function ζK as
Y 1
ζK (s) = ,
p
1 − N (p)−s

the product being taken over all prime ideals of OK , and N (p) := #OK /p. We define the real and
complex Gamma factors as
ΓR (s) = π −s/2 Γ(s/2),
ΓC (s) = (2π)1−s Γ(s),
so that the function ZK (s) = ΓR (s)r1 ΓC (s)r2 ζK (s) is analytic on C except for simple poles at s = 0
and s = 1, and satisfies the functional equation
ZK (s) = |D|1/2−s ZK (1 − s)

University of Toronto, [email protected].

1
2 GRAYSON PLUMPTON

where D is the discriminant of K. The residues of ZK at s = 0 and s = 1 are −cK and |D|−1/2 cK
respectively, where
2r1 (2π)r2 hR
cK = ,
w
with h being the class number of K, R the regulator, and w the number of roots of unity in K.
We then define
ξK (s) = c−1
K s(s − 1)|D|
s/2
ZK (s)
so that ξK is entire, satisfies the symmetric functional equation ξK (s) = ξK (1 − s), and ξK (0) =
ξK (1) = 1.

1.3. The local zeta integral. We will now try to understand the function ZK from the perspective
of Fourier analysis on locally compact abelian groups. This is of course a rich theory which we
cannot do justice; what is shown here is primarily meant to clarify notation or any conventions we
may use, but for a more complete exposition see almost any introductory text on algebraic number
theory such as [3], or [4]. We begin with the local perspective.
Absolute values on K are of two types. First are those induced by prime ideals p ⊂ OK , and
are called archimedean, finite, or discrete. Second are those induced by embeddings of K into R or
C, and are called nonarchimedean or infinite (in particular, there are r1 + r2 archimedean absolute
values on K). A theorem of Ostrowski says these are (up to equivalence) all the absolute values
on K [3]. We will denote these prime ideals and embeddings in the general by v and call them
the places of K, denoting the absolute value induced by such a place by | · |v , and the completion
of K with respect to | · |v by Kv (hence, if v is a real or complex embedding then Kv is R or C
respectively, and otherwise Kv will be a p-adic field for some prime ideal p). For v nonarchimedean,
the valuation ring of Kv is denoted by Ov .
Each Kv is a local field and hence locally compact, so that its additive group is a locally compact
abelian group and we can define on it a Haar measure dxv , unique up to multiplication by a constant.
To ensure dxv is self-dual, define it to be the usual Lebesgue measure if v is real archimedean, twice
the usual Lebesgue measure if v is complex archimedean, and the unique Haar measure assigning
Ov measure N (dv )−1/2 if v is nonarchimedean, where dv is the different ideal of Kv . When there is
no ambiguity we may denote dxv simply by dx.
A function f : Kv → C is called a Schwartz-Bruhat function if either v is archimedean and f is a
Schwartz function, or v is nonarchimedean and f is a locally constant function of compact support.
The C-vector space of Schwartz-Bruhat functions is denoted S (Kv ).
We can also define a multiplicative Haar measure on the group Kv× by letting
dxv
d× xv = , if v is archimedean
|x|v
N (p) dxp
d× xp = , if v = p is nonarchimedean.
N (p) − 1 |x|p
For f ∈ S (Kv ), we can now define the local zeta integral for Re(s) > 1 as
Z
Zv (f, s) := |x|sv f (x)d× xv .
Kv×

(In full generality the zeta integral will also have a factor η(x) where η is a unitary character of the
group Kv× , but for our application we will simply take η = 1.)

1.4. The global zeta integral. Denote N by A the ring of adeles and by A× the group of ideles of
K. Functions f : A → C of the form f = v fv with fv ∈ S (Kv ) for every v and fp = 1Op for all
but finitely many nonarchimedean places p form a basis for S (A) as a C-vector space. We denote
 PROBABILITY LAWS CONCERNING ZETA INTEGRALS 3

by dx the self-dual
N Haar measure on A formed with the self-dual dxv , so that integrating functions
of the form f = fv factors as
Z YZ
f (x)dx = fv (xv )dxv .
A v Kv

The same can be done for the group A× , for which the Haar measure will be denoted d× x. A
complete account of the construction of these measures can be found in [4].
In general, global constructions on A and A× , such as absolute values, additive characters, the
Fourier transform, and zeta integrals
Q factor through local ones. That is, for x = (xv )v ∈ A× we
define the idelic norm as |x| = v |xv |v , additive characters on A are generated by characters of
the form ψ = (ψv )v , so that for f ∈ S (A) the integral of the Fourier transform factors and we have
N N
fb = v fbv , and similarly the global zeta integral defined for f = v fv as
Z
Z(f, s) = |x|s f (x)d× x
A×
Q
factors as Z(f, s) = v Zv (fv , s). In this case, the main result of Tate’s thesis is the functional
equation
Z(f, s) = Z(fb, 1 − s).
An important step in proving this is noticing that the idelic norm is trivial on the diagonal
K× ⊂ A× (Artin’s product formula), so that when calculating the zeta integral of f we can make
the following change:
 
Z Z X
|x|s f (x)d× x = |x|s  f (xκ) d× x.
A× A× /K × κ∈K ×

This trick will also be of use to us when constructing our density function. It is not immediately
obvious that our sum over K × will converge; however, the nonarchimedean part of f forces the
summands to be zero outside of a fractional ideal of K, so that one is ultimately only summing the
archimedean part of f over a lattice in K ⊗ R ≃ Rr1 × Cr2 .

2. Constructing the random variable

We previously constructed the function ZK as being a product over the finite prime ideals of
OK , and one gamma factor for each archimedean place of K; in other words, a product over each
of the places of K. This is analogous to each of our global constructions arising as products over
the places of K. We will now see that this is not a coincidence, and that the function ZK arises as,
in a sense, the simplest zeta integral on A× .

2.1. The functions ZK and s(s − 1)ZK (s) as zeta integrals. For each finite place p of K, let

gp = N (dp )1/2 1Op

(the constant cancels out the measure of Op ). Then for each real embedding σ, let
2
gσ (x) = e−πx
and for each complex embedding τ , let
2
gτ (z) = e−2πzz = e−2π|z| .
4 GRAYSON PLUMPTON

It is straightforward to check that for each finite prime p, real embedding σ, and complex
embedding τ that
1
Zp (fp , s) = ,
1 − N (p)−s
Zσ (gσ , s) = ΓR (s),
Zτ (gτ , s) = ΓC (s).
N
Hence, with each gv defined above, d× xthe Haar measure on A× , and g = v gv , we get
Y
Z(g, s) = Zv (gv , s)
v
Y 1
= ΓR (s)r1 ΓC (s)r2
p
1 − N (p)−s
= ZK (s).
Had we started here, we could use some basic properties of zeta integrals to prove the functional
equation of ZK mentioned previously. Obtaining the s(s − 1) factor of ξK is more difficult. To do
so, we will now restrict our attention to number fields with precisely one archimedean place (so
that K = Q, or K is imaginary quadratic). Denote this single archimedean place by ∞, set fp = gp
as before, and let ( 2
2πx2 (2πx2 − 3)e−πx if ∞ is real,
f∞ (x) = 2
4π|x|2 (π|x|2 − 1)e−2π|x| if ∞ is complex.

Lemma 2.1. With f∞ defined above and s ∈ C× , we have fb∞ = f∞ and

Z∞ (f∞ , s) = s(s − 1)ΓK∞ (s)
where ΓK∞ is the Gamma factor for K∞ .
2
Proof. First, suppose ∞ is real and let g1 (x) = e−πx . Then
2
f∞ (x) = 2πx2 (2πx2 − 3)e−πx
d  2 ′ 
= x g1 (x) .
dx
A straightforward computation shows that
Z ∞
dx
2 xs g1 (x) = ΓR (s),
0 x
and because g1 is a Schwartz function, it satisfies the rapidly decreasing condition: for any integers
n, m ≥ 0,
dn
lim xm n g1 (x) = 0.
x→∞ dx
Hence, integration by parts gives
Z ∞
d  2 ′  dx
Z∞ (f∞ , s) = 2 xs x g1 (s)
0 dx x
Z ∞
dx
= 2s(s − 1) xs g1 (x)
0 x
= s(s − 1)ΓR (s).
Next suppose ∞ is complex. Then
Z
Z∞ (f∞ , s) = f∞ (z)|z|2s d× z.
C×
 PROBABILITY LAWS CONCERNING ZETA INTEGRALS 5

2drdθ
To evaluate the above integral, do the change of variables z = reiθ . Then d× x = r and
Z
Z∞ (f∞ , s) = f∞ (z)|z|2s d× z
C ×
Z 2π Z ∞
2 2drdθ
= 4π r 2 (πr 2 − 1)e−2πr r 2s
0 r
Z ∞0
2
= 16π 2 r 2s+1 (πr 2 − 1)e−2πr dr.
0
2
Let g2 (r) = e−2πr and note that
d  3 ′  2
r g2 (r) = 16πr 3 (πr 2 − 1)e−2πr .
dr
Hence, the integral we are evaluating becomes
Z ∞
d  3 ′ 
Z∞ (f∞ , s) = π r 2s−2 r g2 (r) dr.
0 dr
g2 satisfies the same rapidly decreasing condition as g1 , so integration by parts gives us
Z ∞
Z∞ (f∞ , s) = π(2s − 2)2s r 2s−1 g(r)dr
0
Z ∞
2
= 4πs(s − 1) r 2s−1 e−2πr dr
0
= s(s − 1)(2π)1−s Γ(s)
= s(s − 1)ΓC (s).
N
To prove the self-duality of f∞ , let f = v fv , and g the standard adelic function for which
Z(g, s) = ZK (s). Then
Z(fb, 1 − s) = Z(f, s) = s(s − 1)Z(g, s) = s(s − 1)Z(b
g , 1 − s).
Because f and g differ only in the ∞ place, we must have
Z(fb∞ , 1 − s) = s(s − 1)Z(b
g∞ , 1 − s) = s(s − 1)Z(g∞ , 1 − s) = Z(f∞ , 1 − s).
By the uniqueness of the zeta integral, we must have fb∞ = f∞ . 
N
Hence, taking f = v fv , we have Z(f, s) = s(s − 1)ZK (s) for every s ∈ C× .

2.2. A probabilistic interpretation of Dedekind √ zeta

√ functions of three number fields.
Suppose now that K is one of the fields Q, Q( −1) or Q( −2). We will use our function f : A× K →
s −1
C defined above to construct a random variable X so that E[X ] = cK s(s − 1)ZK (s).
For t ≥ 0, let A×
t denote the set of ideles of absolute value t, so that
Z
Z(f, s) = |x|s f (x)d× x
A ×
Z ∞ Z !
s × dt
= t f (x)d x .
0 ×
At t
R
Our goal is now to think of the function c−1
K t
−1 ×
A× f (x)d x as the density of some random variable.
t

Lemma 2.2. For each K above and t ≥ 0,

Z
c−1
K t
−1
f (x)d× x ≥ 0.
A×
t
6 GRAYSON PLUMPTON

Proof. Of course the factor t−1 is nonnegative, and it is easy to check that


1 if K = Q,
√
cK = π2 if K = Q( −1),

 √
π if K = Q( −2),
so that c−1
K > 0; it now suffices to show that the integral is nonnegative as well. We can use the
following trick hinted at before:
 
Z Z X
f (x)d× x =  f (xκ) d× x
A× A×
t /K
×
t κ∈K ×

so it suffices to show that X

f (xκ) ≥ 0
κ∈K ×
for all x ∈ A× .
First suppose that K = Q so that
2
f∞ (x∞ ) = 2πx2∞ (2πx2∞ − 3)e−πx∞ ,
and for each finite prime p,
fp = 1Zp .
We are considering the sum X
f (xq).
q∈Q×
For a given x = (xv )v ∈ A× , we have that f (xq) = 0 if |xp q|p > 1 for just one prime p. However,
|xp |p 6= 1 for only finitely many p, say p ∈ S. Letting
Y
r= pvp (xp )
p∈S

we have |xp q|p = |rq|p for every prime p. However, 0 < |xq|p = |rq|p ≤ 1 for every prime p if and
only if rq ∈ Z \ {0}, so q = n/r for n ∈ Z \ {0}. Hence,
X X  n
f (xq) = f ∞ x∞ .
r
q∈Q
× n∈Z\{0}

Let y = x∞ /r. If |y| ≥ 1, then every term in the sum on the right hand side is positive. If
0 < |y| < 1, then by the Poisson summation formula and the fact that fb∞ = f∞ ,
X X
f∞ (yn) = |y|−1 f∞ (y −1 n) ≥ 0.
n∈Z\{0} n∈Z\{0}

Either way the sum is nonnegative, so that

X
f (xq) ≥ 0
q∈Q×

for any x ∈ A× . √ √
Now suppose that K = Q( −d), d = 1, 2, so that OK = Z ⊕ −dZ and
2
f∞ (x∞ ) = 4π|x∞ |2 (π|x∞ |2 − 1)e−2π|x∞ | .
For the same reasons as the rational case, we can reduce our sum to
X X
f (xκ) = f∞ (x∞ m)
κ∈K × m∈M \{0}
 PROBABILITY LAWS CONCERNING ZETA INTEGRALS 7

where M is a fractional ideal of OK . Because both quadratic imaginary fields being discussed have
class number 1 and x∞ is arbitrary, we can suppose without loss of generality that our sum is of
the form
X
f∞ (x∞ ℓ).
ℓ∈OK \{0}

When |x∞ |2 ≥ 1/π, this sum is clearly positive. Suppose then that 0 < |x∞ |2 < 1/π and let
u = x−1 b
∞ , so by Poisson summation and that f∞ = f∞
X X
f∞ (x∞ ℓ) = |u| f∞ (uℓ∗ )
ℓ∈OK \{0} ℓ∗ ∈OK
∗ \{0}

√
∗ =
where OK √1 (Z⊕ −dZ). But |u|2 ≥ π, and |ℓ∗ |2 ≥ 1
so when d = 1, 2 we have π|uℓ∗ |2 −1 ≥
2 −d 4d ,
π2
4d − 1 > 0, so that each term of the sum is positive and therefore
X X
f∞ (x∞ ℓ) = |u| f∞ (uℓ∗ ) ≥ 0.
ℓ∈OK \{0} ∗ \{0}
ℓ∗ ∈OK

Hence for each K above and t ≥ 0,

 
Z Z X
c−1
K t
−1
f (x)d× x = c−1
K t
−1  f (xκ) d× x ≥ 0.
A× A×
t /K
×
t κ∈K ×


Remark. For imaginary
P quadratic fields of discriminant d ≥ 3 and class number 1, we no longer
have that the sum ℓ∗ ∈OK \{0} f (uℓ∗ ) has only positive summands, so it is unclear if the sum would
be positive. Fields which are neither the rationals nor imaginary quadratic have more than one
archimedean place, and again one runs into issues applying the same Poisson summation argument
directly, so a stronger argument would be required to generalize to other number fields.
We can now prove our main theorem:

Proof of Theorem 1.1. For t > 0, and f defined above, let

Z
−1 −1
ψ(t) := cK t f (x)d× x.
A×
t

Then ψ(t) ≥ 0 by Lemma 2.2, for each s ∈ C×

Z Z Z !
∞ ∞
dt
ts ψ(t)dt = c−1
K ts f (x)d× x = c−1
K s(s − 1)ZK (s),
0 0 A×
t
t
R∞
and as s → 0, c−1K s(s − 1)ZK (s) → 1, so that 0 ψ(t)dt = 1. Hence, ψ is the density of a random
variable satisfying the above conditions. 
Remark. We can view the random variable X from a more adelic perspective.
P Note
that in proving
Lemma 2.2 and Theorem 1.1 we have additionally shown that c−1 K κ∈K × f (xκ) d× x is a prob-
ability measure on the idele class group A× /K × . Letting Y be the idelic random variable with
this distribution, and | · | the idele-norm, we then have that E|Y |s = |D|−s/2 ξK (s) = E(X s ) for
every s ∈ C. Hence log X and log |Y | have the same moment generating functions, so that X = |Y |
almost surely.
8 GRAYSON PLUMPTON

2.3. An application to Li coefficients. The nth Li coefficient of ξK is defined in [2] as

1 dn  n−1 
λn := n
s log ξK (s) s=1 .
(n − 1)! ds
Using the probabilistic interpretation of ξK , we prove the following:
√ √
Proposition 2.1. Let K be one of Q, Q( −1), or Q( −2), and λn the Li coefficients for ξK .
Then λ1 , λ2 > 0.
Proof. Let X be the random variable constructed in Theorem 1.1 so that for each s ∈ C,
E(X s ) = |D|−s/2 ξK (s),
and functional equation of ξK (s) gives
E(X 1−s ) = |D|(s−1)/2 ξK (s).
Let L := − log X, so that the first cumulant of L is given by
d h i
κ1 = log E(e(s−1)L )
ds s=1
d  1−s

= log E(X ) s=1
ds
d h  (s−1)/2 i
= log |D| ξK (s)
ds p s=1
= log |D| + λ1 .
But by a basic property of cumulants and Jensen’s inequality,
p
κ1 = E(L) = E(− log X) > − log E(X) = log |D|
so λ1 > 0. Similarly, the second cumulant of L is
d2 h i
κ2 = 2 log |D|(s−1)/2 ξK (s)
ds s=1
d2
= 2 [log ξK (s)]s=1
ds
and κ2 = Var(L) ≥ 0, so that
d2
λ2 = [s log ξK (s)]s=1
ds2 
d d
= log ξK (s) + s log ξK (s)
ds ds s=1
 2 
d
= 2λ1 + s 2 log ξK (s)
ds s=1
= 2λ1 + κ2
> 0.
√ √
Therefore, the first two Li coefficients are positive for Q, Q( −1), and Q( −2). 

Acknowledgements. I am incredibly grateful to Professor M. Ram Murty for his supervision

during this project. His comments, encouragement, and advice had a profound impact on both
the outcome of this paper and my mathematical development. I would also like to thank Profes-
sor Francesco Cellarosi and Professor Daniel Johnstone for their helpful comments and aid while
working on this paper.
 PROBABILITY LAWS CONCERNING ZETA INTEGRALS 9

References
[1] P. Biane, J. Pitman, M. Yor, Probability laws related to the Jacobi theta and Riemann zeta functions, and Brownian
excursions, Bull. Amer. Math. Soc. 38, no. 4, 2001.
[2] X. J. Li, The positivity of a sequence of numbers and the Riemann hypothesis, J. Number Theory 65, 325-333,
1997.
[3] J. Neukirch, Algebraic Number Theory. Volume 322 of Graduate Texts in Mathematics, Springer, New York, 1995.
[4] J. T. Tate, “Fourier analysis in number fields, and Hecke’s zeta-functions.” In Algebraic Number Theory, edited
by J. W. S. Cassels and A. Frohlich, 305-347, 1967.