ON THE BEHAVIOR OF MULTIPLE ZETA-FUNCTIONS WITH IDENTICAL

ARGUMENTS ON THE REAL LINE II

KOHJI MATSUMOTO, TOSHIKI MATSUSAKA, AND ILIJA TANACKOV

Abstract. In the present series of papers, we study the behavior of r-fold zeta-functions of Euler-Zagier
type with identical arguments ζr (s, s, . . . , s) on the real line. In this second part, we consider the behavior on
the interval (1, +∞) and (−∞, 0]. We carry out numerical computations, and draw graphs of ζr (s, s, . . . , s)
on those intervals, for several small values of r. We prove several theorems suggested by such graphs. For
example, we prove asymptotic formulas for ζr (−k, −k, . . . , −k), where k takes odd positive integer values and
arXiv:2012.01720v1 [math.NT] 3 Dec 2020

tends to +∞. Moreover, on the number of real zeros of ζr (s, s, . . . , s), we prove that there are exactly (r − 1)
real zeros on the interval (−2n, −2(n − 1)) for any n ≥ 2.

1. Introduction
Define the Euler-Zagier multiple zeta-function by
X 1 1 1
(1.1) ζr (s1 , s2 , ..., sr ) = · · · · sr ,
ms11 ms22 mr
1≤m1 <m2 <...<mr

where s1 , . . . , sr are complex variables. The series (1.1) is convergent absolutely when <sj (1 ≤ j ≤ r) are
sufficiently large, but it can be continued meromorphically to the whole space Cr .
The aim of the present series of papers is to investigate the behavior of
ζr (s) = ζr (s, s, . . . , s)
for real s. In the first part [5], we studied the behavior of ζr (s) on the interval [0, 1]. In the present paper,
we consider the behavior of ζr (s) outside this interval.
Our investigation is based on numerical computations on the behavior of ζr (s) for s ∈ (1, +∞) and
s ∈ (−∞, 0]. Such a numerical study was already done in a paper [4] by the first-named author and M.
Shōji, in the case r = 2. Our study is a generalization of [4].
We prove several theorems suggested by such numerical computations. For example, we prove asymptotic
formulas for ζr (−k, −k, . . . , −k), where k takes odd positive integer values and tends to +∞. Moreover, on
the number of real zeros of ζr (s, s, . . . , s), we prove that there are exactly (r − 1) real zeros on the interval
(−2n, −2(n − 1)) for any n ≥ 2. The latter result may be regarded as an analogue of the second-named
author’s result [6] on the real zeros of Hurwitz zeta-functions.
The basic tool of our study is the following identity:
X r
(1.2) rζr (s) = (−1)j−1 ζr−j (s)ζ(js) (r ∈ N),
j=1

2010 Mathematics Subject Classification. Primary 11M32, Secondary 11B83, 11M35.

Key words and phrases. multiple zeta-function, real zeros, asymptotic behavior, Newton’s identities.
Research of the first author is supported by Grants-in-Aid for Science Research no. 18H01111, JSPS, that of the second
author is by JP20K14292, JSPS, and that of the third author is by Ministry of Science and Technological Development of Serbia
no. TR 36012.
1
2 KOHJI MATSUMOTO, TOSHIKI MATSUSAKA, AND ILIJA TANACKOV

where ζ(s) = ζ1 (s) is the Riemann zeta-function, and we understand that ζ0 = 1. This identity is a kind
of “infinite” version of classical Newton’s identity, and was proved in [5] (and had been known by Kamano
[3]).
For several small values of r, this formula implies:
1
ζ(s)2 − ζ(2s) ,

(1.3) ζ2 (s) =
2

1
(1.4) ζ3 (s) = {ζ2 (s)ζ(s) − ζ(s)ζ(2s) + ζ(3s)} ,
3

1
(1.5) ζ4 (s) = {ζ3 (s)ζ(s) − ζ2 (s)ζ(2s) + ζ(s)ζ(3s) − ζ(4s)} .
4
Originally, the idea of applying Newton’s identity to the investigation of multiple zeta-functions was
inspired by the third-named author’s ”new-nacci” method for solving polynomial roots (see Tanackov et
al. [7]). The new-nacci method also gives the relation
ζr (k + 1) 1
(1.6) lim = ,
k→∞ ζr (k) r!
which is useful when we consider the behavior of ζr (s) when s → +∞. A proof of this relation will be given
in Section 3.

2. The double and the triple zeta-functions

We first recall the graphs (Figure 1 and Figure 2) of ζr (s) on the real line for r = 2 and 3, which we gave
in [5].

2.1. The double case. From (1.3) it is clear that all trivial zeros of the Riemann zeta-function s =
−2, −4, −6, . . . are trivial zeros of the double zeta-function. The values of ζ2 (s) at s = −1, −3, −5, . . . were
already computed by Kamano [3]. For example, since ζ1 (−1) = −12−1 , we find
!
1 2
 
1 2 1 1
(2.1) ζ2 (−1) = (ζ1 (−1) − ζ1 (−2)) = − −0 = .
2 2 12 288

Let ` ∈ N. For any `, there should be at least one “inter-trivial” zero (ITZ for brevity) between two trivial
zeros s = −2` and s = −2(` + 1). The proof is as follows. Since all trivial zeros of ζ(s) are simple, the value
of ζ 0 (−2`) is positive for even ` and negative for odd `. From (1.3) we have ζ20 (s) = ζ(s)ζ 0 (s) − ζ 0 (2s), hence
ζ20 (−2`) = −ζ 0 (−4`), which is negative for all `. Therefore ζ2 (s) is decreasing at s = −2`, which implies the
claim.
When s → +∞, the graph is smoothly going down to 0 because of (1.6), while when s → −∞ the graph
is strongly oscillated; indeed, ζ2 (−2` − 1) → +∞ as ` → ∞.
All the stated values, the graph, the location of zeros for ζ2 (s) mentioned above were analyzed in detail
by the first-named author and Shōji [4].

2.2. The triple case. All trivial zeros of the Riemann zeta-function are zeros of ζ3 (s). It seems from the
graph that, on the interval between consecutive trivial zeros of the Riemann zeta-function, ζ3 (s) has two
ITZs. The behavior of ζ3 (s) when s → +∞ can be explained by (1.6).
 MULTIPLE ZETA-FUNCTIONS 3

Figure 1. The double zeta-function (Note that the vertical scale in the negative half-plane
is different from that in the positive half-plane)

3. Multiple zeta-functions on the interval s ∈ (1, +∞)

In this section we discuss the behavior of ζr (s) for s ∈ (1, +∞). From the definition it is obvious that
ζr (s) is monotonically decreasing when s increases in this interval, and


1 (r = 1)
lim ζr (s) =
s→+∞ 0 (r ≥ 2).

We can refine this to the following asymptotic formula.

Theorem 3.1. For any r ≥ 1, we have ζr (s) ∼ (r!)−s as s → +∞. In particular, (1.6) holds.

Proof. Let s > 1. We first notice the inequality

∞ ∞ Z ∞  
X 1 1 X 1 1 dt 1 k
(3.1) = s+ ≤ s+ = s 1+ .
ms k ms k k ts k s−1
m=k m=k+1
4 KOHJI MATSUMOTO, TOSHIKI MATSUSAKA, AND ILIJA TANACKOV

Figure 2. The triple zeta-function (Note that the vertical scale in the negative half-plane
is different from that in the positive half-plane)

Therefore
X 1 X 1 X 1
ζr (s) = s
+ s
(m2 · · · mr ) m1 (m2 · · · mr )s
2≤m2 <···<mr m1 ≥2 (m1 <)m2 <···<mr
 
X 1   
X 1 1 2 X 1
≤ 1+
  ≤ 1+ s 1+ .
ms1 (m2 · · · mr )s 2 s−1 (m2 · · · mr )s
m1 ≥2 2≤m2 <···<mr 2≤m2 <···<mr

Using (3.1) again, we see that the last sum is

1 X 1 X 1 X 1
= s s
+ s
2 (m3 · · · mr ) m2 (m3 · · · mr )s
3≤m3 <···<mr m2 ≥3 (m2 <)m3 <···<mr
  
1 1 3 X 1
≤ + 1 + .
2s 3s s−1 (m3 · · · mr )s
3≤m3 <···<mr

Repeating this procedure, we arrive at

     
1 2 1 1 3
(3.2) ζr (s) ≤ 1 + s 1 + + 1+
2 s−1 2s 3s s−1
    
1 1 r 1 r
× ··· × + 1+ × s 1+
(r − 1)s rs s−1 r s−1
     s  
1 1 2 2 3
= 1+ s 1+ 1+ 1+
(r!)s 2 s−1 3 s−1
  s    
r−1 r r
× ··· × 1 + 1+ × 1+ .
r s−1 s−1
When s → +∞,  s  
k−1 k
1+ →0 (2 ≤ k ≤ r),
k s−1
 MULTIPLE ZETA-FUNCTIONS 5

hence ζr (s) ≤ (r!)−s (1 + o(1)) as s → +∞. On the other hand it is obvious that ζr (s) ≥ (r!)−s . Therefore
the assertion of the theorem follows. 
Remark 3.2. When r = 2, Theorem 3.1 is included in the proof of [4, Proposition 2.1].
From (3.2), we can also prove:
Theorem 3.3. For any fixed s > 1, we have limr→∞ ζr (s) = 0.
Proof. Since ζr (s) is monotonically decreasing with respect to s, it is enough to show the theorem for
s ∈ (1, 2). Then
k−1 s
     
k k k 2k
1+ 1+ ≤1+ 1+ =2+ ≤ (2 ≤ k ≤ r)
k s−1 s−1 s−1 s−1
and 1 + r/(s − 1) ≤ 2r/(s − 1). Therefore from (3.2) we have
 r  r
1 2 2 r
ζr (s) ≤ · r! · r = .
(r!)s s−1 s−1 (r!)s−1
Here we apply the classical formula of Stirling
(3.3) r! ∼ (2πr)1/2 rr e−r (r → +∞)
to get
r
r1−(s−1)/2 2es−1

1
ζr (s) .
(2π)s−1 s−1 r(s−1)r
which, for any fixed s > 1, tends to 0 as r → ∞. 
The assertions of Theorem 3.1 and Theorem 3.3 can be illustrated in Figure 3, where we symbolically
write ζ∞ (s) as the limit r → ∞.

Figure 3. Dynamics of the r-fold zeta-function ζr (s) in the interval s ∈ (1, +∞)

4. Multiple zeta-functions on the interval s ∈ (−∞, 0]

Now we consider the behavior of ζr (s) for s ∈ (−∞, 0]. We already know [5, Theorem 5.1] that the value
ζr (0) is positive for even r, and negative for odd r. Here we list up the values of ζr (0) and ζr (−1) for
1 ≤ r ≤ 14 .
How about the zeros? First of all, we note the following fact, which was conjectured by [1] and first
proved by Kamano [3].
Proposition 4.1. For any n ∈ N, we have ζr (−2n) = 0.
Proof. This is an immediate consequence of the fact that ζ(−2n) = 0 for any n ∈ N. In fact, since there is
no pole of ζr−j (s) (1 ≤ j ≤ r) on the negative real line [5, Theorem 5.2(i)], all terms on the right-hand side
of (1.2) are zero at s = −2n. 
6 KOHJI MATSUMOTO, TOSHIKI MATSUSAKA, AND ILIJA TANACKOV

r ζr (0) ζr (−1) r ζr (0) ζr (−1)

1 −0,500000000000000 −0,08333333333 8 +0,196380615234375 −0,00005171790
2 +0,375000000000000 +0,00347222222 9 −0,185470581054687 −0,00083949872
3 −0,312500000000000 +0,00268132716 10 +0,176197052001953 +0,00007204895
4 +0,273437500000000 −0,00022947209 11 −0,168188095092773 +0,00191443849
5 −0,246093750000000 −0,00078403922 12 +0,161180257797241 −0,00016251626
6 +0,225585937500000 +0,00006972813 13 −0,154981017112732 −0,00640336283
7 −0,209472656250000 +0,00059216643 14 +0,149445980787277 +0,00054016476

Numerical computations show that there are more zeros on the negative real line, which we call inter-
trivial zeros (ITZs). This phenomenon was first noticed by the first-named author and Shōji [4] in the
double zeta case. We presented the graphs for the double and the triple cases in Section 2. We can compute
further, and consequently, we may guess that the following conjecture is plausible:
Conjecture 4.2. There would be (r − 1) ITZs of ζr (s) on the interval (−2n, −2(n − 1)) for any n ∈ N.
(Therefore, more and more ITZs appear when r increases.)
In fact, we can prove this conjecture with only one exception:
Theorem 4.3. There are exactly (r − 1) ITZs of ζr (s) on the interval (−2n, −2(n − 1)) for any n ≥ 2.
We will give the proof of this theorem in the next section.
The method of our proof of Theorem 4.3 cannot be applied to the case n = 1, and hence this case is still
open. We show the graphs of r-fold zeta-functions (2 ≤ r ≤ 10) on the interval [−2, 0] (see Figure 4), and
furthermore we present the data of ITZs for the r-fold zeta-functions ζr (s), 2 ≤ r ≤ 10, on the same interval
(see Figure 5). This data clearly agree with the case n = 1 of Conjecture 4.2.
When s → −∞, the values of ζr (s) become highly oscillating, and take larger and larger absolute values.
This situation can be well illustrated in the following theorem.
Theorem 4.4. Let r ≥ 1. When k takes odd integer values and k → +∞, we have
(i) If r is odd,
1/2 k rk
 
1 (rk+1)/2 2k rk−1/2
(4.1) ζr (−k) ∼ ζ(−rk) ∼ (−1) · √ ·r .
r 2π 2πe
(ii)
 2k
1 k k
(4.2) ζ2 (−k) = ζ(−k)2 ∼ .
2 π 2πe
(iii) If r ≥ 4 is even,
 rk
1 2k k
(4.3) ζr (−k) ∼ ζ(−k)ζ(−(r − 1)k) ∼ (−1)(r/2)−1 · · (r − 1)(r−1)k−1/2 .
r−1 π 2πe
To prove this theorem, we first prepare:
Lemma 4.5. Let r ≥ 6 and 3 ≤ j ≤ r − 3. Then (r − j)r−j j j < (r − 1)r−1 .
Proof. The claim is equivalent to the inequality
(4.4) (r − j) log(r − j) + j log j < (r − 1) log(r − 1)
 MULTIPLE ZETA-FUNCTIONS 7

Figure 4. The behavior of r-fold zeta-functions (2 ≤ r ≤ 10) on the interval [−2, 0]

for 3 ≤ j ≤ r − j. Since the function x log x is convex downward for x > 0, it is sufficient to prove (4.4) for
j = 3:
(4.5) (r − 3) log(r − 3) + 3 log 3 < (r − 1) log(r − 1).
When r = 6, this is 6 log 3 < 5 log 5, which is true. When r ≥ 7, we have 3 log 3 < 2 log(r − 1), and so
(r − 3) log(r − 3) + 3 log 3 < (r − 3) log(r − 1) + 2 log(r − 1) = (r − 1) log(r − 1).


Proof of Theorem 4.4. First consider the case r = 1. Then it is well-known that
Bk+1
ζ1 (−k) = ζ(−k) = −
k+1
for odd k, where Bk+1 is the (k + 1)-th Bernoulli number (see [2, Theorem 12.16]). Since the asymptotic
formula B2k ∼ (−1)k+1 2(2k)!/(2π)2k , as k → ∞, is known (see [2, Theorem 12.18]), combining with (3.3)
we obtain
2k 3/2 k k
Bk+1 ∼ (−1)(k−1)/2 ,
(2π)1/2 (2πe)k
8 KOHJI MATSUMOTO, TOSHIKI MATSUSAKA, AND ILIJA TANACKOV

Figure 5. Inter-trivial zeros of multiple zeta-functions ζr (s) (2 ≤ r ≤ 10) in the interval [−2, 0]

and hence
k
2k 1/2 k k 1/2

(k+1)/2 (k+1)/2 2k k
(4.6) ζ(−k) ∼ (−1) = (−1) √ ,
(2π)1/2 (2πe)k 2π 2πe
which is the case r = 1 of (4.1).
Then, from (1.3) we have
 2k
1
1 k k
ζ2 (−k) = ζ(−k)2 − ζ(−2k) = ζ(−k)2 ∼ ,
2 2 π 2πe
which is (4.2). Next, using (1.4) and applying (4.6) and (4.2), we obtain
1
ζ3 (−k) = (ζ2 (−k)ζ(−k) + ζ(−3k))
3(
 )
k 2k 1/2 k k 1/2 3k 3k
    
1 k (k+1)/2 2k (3k+1)/2 2(3k)
∼ (−1) √ + (−1) √
3 π 2πe 2π 2πe 2π 2πe
( )
k 3k k 3/2
 
1 (k+1)/2 2
= (−1) √ − k 1/2 33k+1/2
3 2π 2πe π
1/2 k 3k
 
(3k+1)/2 2k 3k−1/2
∼ (−1) √ 3 ,
2π 2πe
which is the case r = 3 of (4.1). This calculation also shows ζ3 (−k) ∼ (1/3)ζ(−3k), and hence, from (1.5),
1 1
ζ4 (−k) = (ζ3 (−k)ζ(−k) + ζ(−k)ζ(−3k)) ∼ ζ(−k)ζ(−3k).
4 3
Therefore using (4.6) we obtain
 4k
2k k
ζ4 (−k) ∼ − 33k−1/2 ,
π 2πe
which is the case r = 4 of (4.3).
 MULTIPLE ZETA-FUNCTIONS 9

Now we prove the general case by induction on r. The formula (1.2) implies
1 X
(4.7) ζr (−k) = (−1)j−1 ζr−j (−k)ζ(−jk).
r
i≤j≤r
j:odd

When r is odd, we separate the last term from the above sum to get
1 X 1
(4.8) ζr (−k) = (−1)j−1 ζr−j (−k)ζ(−jk) + ζ(−rk).
r r
i≤j≤r−1
j:odd

Note that r − j is even in each term in the above sum. Therefore, By the induction assumption, each term
in the above sum is

k (r−j)k jk jk
   
(r−j−1)k−1/2 1/2
(4.9) r k(r − j − 1) · (jk)
2πe 2πe
 rk
k
= (r − j − 1)(r−j−1)k−1/2 k 3/2 j jk+1/2 .
2πe

Since
(r − j − 1)(r−j−1)k−1/2 j jk+1/2 ≤ r(r−j−1)k−1/2 rjk+1/2 = r(r−1)k ,
the order of the right-hand side of (4.9) with respect to k is less than the order of the term (1/r)ζ(−rk)
(which is k 1/2 (rk/2πe)rk ). Therefore from (4.8) we have ζr (−k) ∼ (1/r)ζ(−rk), and hence (4.1) follows by
applying (4.6) to ζ(−rk).
We proceed to the case of even r. In this case the term corresponding to j = r does not appear on the
right-hand side of (4.7). We divide the sum as
1
ζr (−k) = (ζr−1 (−k)ζ(−k) + ζ(−k)ζ(−(r − 1)k)) + Er ,
r
where Er consists of the terms corresponding to 3 ≤ j ≤ r − 3, j:odd. Then, since r − 1 is odd, we use the
induction assumption ζr−1 (−k) ∼ (r − 1)−1 ζ(−(r − 1)k) to obtain
1
(4.10) ζr (−k) ∼ ζ(−k)ζ(−(r − 1)k) + Er .
r−1
If the order of Er is less than the order of ζ(−k)ζ(−(r − 1)k), then ζr (−k) ∼ (r − 1)−1 ζ(−k)ζ(−(r − 1)k),
and hence (4.3) follows by applying (4.6). The order of ζ(−k)ζ(−(r − 1)k) is given by the right-hand side
of (4.3), while the order of each term ζr−j (−k)ζ(−jk) (3 ≤ j ≤ r − 3, j:odd) in the sum Er is

(r − j)k (r−j)k jk jk
   
1/2 −1/2 1/2
 k (r − j) (jk)
2πe 2πe
 rk
k
r k(r − j)(r−j)k j jk .
2πe

Since the order of (r − j)(r−j)k j jk with respect to k is less than the order of (r − 1)(r−1)k by Lemma 4.5, we
obtain the desired assertion. 
10 KOHJI MATSUMOTO, TOSHIKI MATSUSAKA, AND ILIJA TANACKOV

5. Proof of Theorem 4.3

In this last section we provide a proof of Theorem 4.3. For positive integers a, b, k > 0, we let
  
 2k + 1 2k − 1 1 1
Rk = Rk (a + b) := s ∈ C − 2 −
 ≤ <s ≤ −2 − ,− ≤ =s ≤ ,
a+b a+b a+b a+b
and
  
 1 1 1
R0 = R0 (a + b) := s ∈ C − 2 − ≤ <s ≤ −2, − ≤ =s ≤ .
a+b a+b a+b
First, we prepare the following three lemmas.

Lemma 5.1. Fix positive integers a, b. For any k ≥ 0 and any s ∈ ∂Rk on the boundary, we have
 
 sin π (a + b)s
 2
2

 > .

π π
 sin 2 as sin 2 bs  a + b


Proof. We divide the proof into three cases.

i
1. For s = x ± a+b (x ≤ −2), by the fact that | sin(x + iy)|2 = sin2 x + sinh2 y, we have

 sin π (a + b)s
2
 
sin2 π
(a + b)x + sinh2 π2



2
2
(5.1)
 = 

 sin π2 as sin π2 bs 
    
sin2 π
sinh2 π2 a+b
a
sin2 π2 bx + sinh2 π b



2 ax + 2 a+b

sinh2 π2


≥    .
cosh2 π2 a+b
a
cosh2 π2 a+b
b

The addition formula for sinh y gives

π   
π a π b
sinh = sinh +
2 2a+b 2a+b
       
π a π b π a π b
= sinh cosh + cosh sinh .
2a+b 2a+b 2a+b 2a+b

Therefore the right-hand side of (5.1) is equal to

    2
π a π b
tanh + tanh ,
2a+b 2a+b
which implies
 
 sin π (a + b)s
 
π a
 
π b
 
π 1

2

≥ tanh + tanh ≥ 2 tanh .
 
 sin π2 as sin π2 bs 


2a+b 2a+b 2a+b
 

Since tanh(π/2y) > 1/y holds for y ≥ 2, we now obtain

 
 sin π (a + b)s
 2
2

 > .

π π
 sin 2 as sin 2 bs  a + b

 MULTIPLE ZETA-FUNCTIONS 11

2. For s = −2 − 2k−1 it
a+b + a+b (k > 0, −1 ≤ t ≤ 1), in a similar manner, we get
 sin π (a + b)s
2
 
1 + sinh2 π2 t


 2

 = 

 sin π2 as sin π2 bs 
         
sin2 π(2k−1)
2
a
a+b + sinh 2 π a
2 a+b t sin 2 π(2k−1) b
2 a+b + sinh 2 π b
2 a+b t
2 π 2 π



cosh 2 t 4 cosh 2 t 4
>    = 2 ≥ 1 ≥ .
(a + b)2

cosh2 π2 a+b
a
t cosh2 π2 a+b
b
t cosh π2 t + cosh π2 a−b


a+b t
Here, the first inequality is valid because the equation
   
2 π(2k − 1) a 2 π(2k − 1) b
sin = sin =1
2 a+b 2 a+b
does not happen.
it
3. For s = −2 + a+b (−1 ≤ t ≤ 1), again using the addition formula, we have
     
 sin π (a + b)s
  sinh π
t

 
1 1

2
2
2
  

π π

 = 
      =   +    > 2 ≥ .
a+b

 sin 2 as sin 2 bs   sinh π a t sinh π b t   tanh π a t tanh π b
t
2 a+b 2 a+b 2 a+b 2 a+b


Note that this argument is valid even when t = 0; in this case we understand that the left-hand side is +∞.
The proof of Lemma 5.1 is thus complete. 
We now replace R0 with a smaller rectangle
  
(ε)  1 1 1
R0 = s ∈ C − 2 −
 ≤ <s ≤ −2 − ε, − ≤ =s ≤
a+b a+b a+b
(ε) (ε)
for a small enough ε > 0 such that Lemma 5.1 also holds for s ∈ ∂R0 . For simplicity, we refer to R0 as
R0 .
S
Lemma 5.2. Fix positive integers a, b. For any s ∈ k≥0 Rk , we have
  
 Γ(1 − (a + b)s)   ζ(1 − (a + b)s)  1 2
 Γ(1 − as)Γ(1 − bs)   ζ(1 − as)ζ(1 − bs)  > 2π (a + b) .
  

Proof. First, the zeta-factor is evaluated as follows. For x ≥ 3 and −1 ≤ y ≤ 1,

∞  ∞
X  X
||ζ(x + iy)| − 1| ≤  n−x−iy  ≤ n−x ≤ ζ(3) − 1,
 
 
n=2 n=2
hence 2 − ζ(3) ≤ |ζ(x + iy)| ≤ ζ(3). Therefore we get
 
 ζ(1 − (a + b)s)  2 − ζ(3)
 ζ(1 − as)ζ(1 − bs)  ≥ ζ(3)2 .
 

Secondly, since it is known that

∞  −1/2
Y y2
|Γ(x + iy)| = |Γ(x)| 1+ ,
(x + j)2
j=0
t
for s = −x + i a+b (x ≥ 2, −1 ≤ t ≤ 1) we have
 
 Γ(1 − (a + b)s) 
 Γ(1 − as)Γ(1 − bs)  ≥ G(x)H(x),
 
12 KOHJI MATSUMOTO, TOSHIKI MATSUSAKA, AND ILIJA TANACKOV

where
∞  −1/2
Γ((a + b)x + 1) Y 1
G(x) = , H(x) = 1+ .
Γ(ax + 1)Γ(bx + 1) ((a + b)x + 1 + j)2
j=0
Write z := (a + b)x + 1. Then z ≥ 5, and so
∞   X ∞ ∞
X 1 1 X 1
(5.2) log 1 + < ≤ =: A,
(z + j)2 (z + j)2 (5 + j)2
j=0 j=0 j=0

say. Note that

4
X 1 π 2 205
A = ζ(2) − = − .
l2 6 144
l=1
From (5.2) we obtain H(x) > e−A/2 .
We next claim that the G(x) is a monotonically increasing function in x ≥ 2. This follows from the fact
that its logarithmic derivative is always non-negative, that is,
(a + b)ψ((a + b)x + 1) ≥ aψ(ax + 1) + bψ(bx + 1),
where ψ(x) is the digamma function. Since ψ(x + 1) is convex and monotonically increasing, we get
 2
a + b2

aψ(ax + 1) + bψ(bx + 1)
≤ψ x + 1 ≤ ψ((a + b)x + 1)
a+b a+b
(the convexity gives the first inequality, and the monotonicity gives the second inequality). Therefore the
claim follows.
We evaluate G(x) at x = 2:
   
2a + 2b 2a + 2b (2a + 2b)(2a + 2b − 1) 3
G(2) = ≥ = ≥ (a + b)2 .
2a 2 2 2π
Thus we have
  
 Γ(1 − (a + b)s)   ζ(1 − (a + b)s)  3 2 −A/2 2 − ζ(3)
 Γ(1 − as)Γ(1 − bs)   ζ(1 − as)ζ(1 − bs)  ≥ 2π (a + b) · e ·
  
ζ(3)2
1 1
= 1.483 · · · × (a + b)2 > (a + b)2 ,
2π 2π
which concludes the proof. 
Lemma 5.3. Let r ≥ 2 be a fixed integer. For any 0 < j < r, k ≥ 0 and any s ∈ ∂Rk (r) on the boundary,
we have  
 ζ(rs) 
 ζ((r − j)s)ζ(js)  > r.
 

Proof. By the functional equation,

 
sin π2 rs
 
 

 ζ(rs)  
 = π 

  Γ(1 − rs) 
  ζ(1 − rs) 
.
 sin π2 (r − j)s sin π
 ζ((r − j)s)ζ(js) 

 
2 js  Γ(1 − (r − j)s)Γ(1 − js)   ζ(1 − (r − j)s)ζ(1 − js) 

Applying Lemma 5.1 and 5.2, we get the lemma. 

Now we prove the following key proposition.
Proposition 5.4. Let r ≥ 2 and k ≥ 0. For any s ∈ ∂Rk (r), we have |ζr (s)| < |ζ(rs)|.
 MULTIPLE ZETA-FUNCTIONS 13

Proof. We prove it by induction on r. Note that ζ(rs) is never equal to 0 on the boundary ∂Rk . For r = 2,
by Lemma 5.3 (with r = 2) and (1.2), we have
   
1 2 1 1
|ζ2 (s)| ≤ |ζ(s)| + |ζ(2s)| ≤ |ζ(2s)| + |ζ(2s)| < |ζ(2s)|.
2 2 2
Now suppose that the proposition holds for less than r. Then by the induction assumption and lemma 5.3
we have
r r−1
1X 1 1X
|ζr (s)| ≤ |ζr−j (s)||ζ(js)| < |ζ(rs)| + |ζ((r − j)s)||ζ(js)|
r r r
j=1 j=1
1 r−1
≤ |ζ(rs)| + 2 |ζ(rs)| < |ζ(rs)|,
r r
which finishes the proof. 
We finally provide a proof of Theorem 4.3. By (1.2) and the above proposition, for s ∈ ∂Rk (r),
 
 r−1  r−1 r−1
X  X X
 (−1)j−1 ζr−j (s)ζ(js) ≤ |ζr−j (s)||ζ(js)| < |ζ((r − j)s)||ζ(js)|
 
 j=1  j=1 j=1
r−1
(5.3) < |ζ(rs)| < |ζ(rs)|.
r
Therefore, by Rouché’s theorem, ζ(rs) and ζr (s) have the same number of zeros in the interior of each
Rk (r). Since ζ(rs) has exactly one zero in the interior of each Rk (r) (k > 0), ζr (s) has the unique zero in
the interior of Rk for k > 0. Since ζr (s) is symmetric with respect to the real-axis, the zero is real. By
taking ε → 0+ , we also see that there is no zero in the interior of R0 (r).
By replacing Rk (r) (k > 0) with a slightly larger rectangle such that (5.3) also holds, we can check that
there is no zero of ζr (s) on the end points of Rk (r) ∩ R. Thus we complete the proof of the theorem.

References
[1] S. Akiyama, S. Egami and Y. Tanigawa, Analytic continuation of multiple zeta-functions and their valuers at non-positive
integers, Acta Arith. 98 (2001), 107–116.
[2] T. M. Apostol, Introduction to Analytic Number Theory, Springer, 1976.
[3] K. Kamano, The multiple Hurwitz zeta function and a generalization of Lerch’s formula, Tokyo J. Math. 29 (2006), 61–73.
[4] K. Matsumoto and M. Shōji, Numerical computations on the zeros of the Euler double zeta-function I, Moscow J. Combin.
Number Theory 4 (2014), 295–313.
[5] K. Matsumoto and I. Tanackov, On the behavior of multiple zeta-functions with identical arguments on the real line I,
preprint.
[6] T. Matsusaka, Real zeros of the Hurwitz zeta function, Acta Arith. 183 (2018), 53–62.
[7] I. Tanackov, I. Pavkov and Ž. Stević, The new-nacci method for calculating the roots of a univariate polynomial and
solution of quintic equation in radicals, Mathematics 2020 8(5), 746; 18pp, https://doi.org/10.3390/math8050746.

K. Matsumoto: Graduate School of Mathematics, Nagoya University, Chikusa-ku, Nagoya 464-8602, Japan
Email address: kohjimat@math.nagoya-u.ac.jp

T. Matsusaka: Institute for Advanced Research, Nagoya University, Chikusa-ku, Nagoya 464-8602, Japan
Email address: matsusaka.toshiki@math.nagoya-u.ac.jp

I. Tanackov: Faculty of Technical Sciences, University of Novi Sad, Trg Dositeja Obradovića 6, 21000 Novi
Sad, Serbia
Email address: ilijat@uns.ac.rs