-
8/6/2019 Ivic Divisor Problem
1/121
CEJM 2(4) 2004 494508
On the Riemann zeta-function and the divisorproblem
Aleksandar Ivic Katedra Matematike RGF-a,
Universiteta u Beogradu,
usina 7, 11000 Beograd, Serbia (Yugoslavia)
Received 14 April 2004; accepted 19 July 2004
Abstract: Let ( x) denote the error term in the Dirichlet
divisor problem, and E (T ) the errorterm in the asymptotic formula
for the mean square of | ( 12 + it )|. If E (t) = E (t) 2 (t/
2)with (x) = ( x) + 2(2 x) 12 (4 x), then we obtain
T
0 (E
(t))4
dt T 16/ 9+
.
We also show how our method of proof yields the bound
R
r =1
t r + G
t r G | (12 + it )|2 dt
4
T 2+ G 2 + RG 4T ,
where T 1/ 5+ G T, T < t 1 < < t R 2T, tr +1 t r 5G (r
= 1 , . . . , R 1).c Central European Science Journals. All rights
reserved.
Keywords: Dirichlet divisor problem, Riemann zeta-function, mean
square and twelfth moment of | ( 12 + it )|, mean fourth power of E
(t)MSC (2000): 11N37, 11M06
1 Introduction and statement of results
Let, as usual,( x) =
n x
d(n) x(log x + 2 1) 14 , (1)
E-mail: [email protected], [email protected]
-
8/6/2019 Ivic Divisor Problem
2/121
A. Ivic / Central European Journal of Mathematics 2(4) 2004
494508 495
and
E (T ) = T
0 | (12 + it )|2 dt T log
T 2
+ 2 1 , (2)where d(n) is the number of divisors of n, = (1) = 0
.577215. . . is Eulers constant.Thus ( x) denotes the error term in
the classical Dirichlet divisor problem, and E (T ) isthe error
term in the mean square formula for | ( 12 + it )|. An interesting
analogy betweend(n) and | ( 12 + it )|2 was pointed out by F.V.
Atkinson [1] more than sixty years ago. Inhis famous paper [2],
Atkinson continued his research and established an explicit
formulafor E (T ) (see also the authors monographs [7, Chapter 15]
and [8, Chapter 2]). The mostsignicant terms in this formula, up to
the factor ( 1)n , are similar to those in Voronoisformula (see [7,
Chapter 3]) for ( x). More precisely, in [13] M. Jutila showed that
E (T )should be actually compared to 2 (T/ (2)), where
(x) := ( x) + 2(2 x) 12 (4 x). (3)Then the arithmetic
interpretation of (x) (see T. Meurman [16]) is
12
n 4x
(1)n d(n) = x(log x + 2 1) + (x). (4)
We have the explicit, truncated formula (see e.g., [7] or
[18])
( x) =1
2x1
4
n N d(n)n
3
4 cos(4nx 14 ) + O(x1
2 + N 1
2 ) (2 N x). (5)
One also has (see [7, eq. (15.68)]), for 2N x, (x) =
12x
14
n N
(1)n d(n)n34 cos(4nx 14 ) + O(x
12 + N
12 ), (6)
which is completely analogous to (5).
M. Jutila, in his works [13] and [14], investigated both the
local and global behaviourof
E (t) := E (t) 2 t
2.
He proved the mean square bound
T + H
T H (E (t))2 dt HT 1/ 3 log3 T + T 1+ (1 H T ), (7)
which in particular yields
T
0(E (t))2 dt T 4/ 3 log3 T. (8)
Here and later denotes positive constants which are arbitrarily
small, but are not nec-essarily the same at each occurrence. The
bound (8) shows that, on the average, E (t)
-
8/6/2019 Ivic Divisor Problem
3/121
496 A. Ivic / Central European Journal of Mathematics 2(4) 2004
494508
is of the order t1/ 6 log3/ 2 t, while both E (x) and ( x) are
of the order x1/ 4. Thisfollows from the mean square formulas (see
e.g., [8])
T
0 2(x) dx = (6 2) 1
n =1
d2(n)n 3/ 2T 3/ 2 + O(T log4 T ), (9)
and
T
0E 2(x) dx = 23 (2)
1/ 2
n=1
d2(n)n 3/ 2T 3/ 2 + O(T log4 T ). (10)
The mean square formulas (9) and (10) also imply that the
inequalities < 1/ 4 and < 1/ 4 cannot hold, where and are,
respectively, the inma of the numbers a and bfor which the
bounds
( x) xa , E (x) xb (11)
hold. For upper bounds on , see e.g., M.N. Huxley [5]. Classical
conjectures are that = = 1 / 4 holds, although this is notoriously
difficult to prove. M. Jutila [13] succeededin showing the
conditional estimates: if the conjectural = 1 / 4 holds, then this
impliesthat 3/ 10. Conversely, = 1 / 4 implies that (x) x+ holds
with 3/ 10.Although one expects the maximal orders of ( x) and (x)
to be approximately of thesame order of magnitude, this does seem
difficult to establish.
In what concerns the formulas involving higher moments of ( x)
and E (t), we referthe reader to the authors works [6], [7] and
[10] and D.R. Heath-Brown [4]. In particular,
note that [10] contains proofs of
T
0E 3(t) dt = 164
T 2
0( (t))3 dt + O(T 5/ 3 log3/ 2 T ),
T
0E 4(t) dt = 325
T 2
0( (t))4 dt + O(T 23/ 12 log3/ 2 T ). (12)
In a recent work by P. Sargos and the author [12], the
asymptotic formulas of K.-M.Tsang [19] for the cube and the fourth
moment of ( x) were sharpened to
X1 3(x) dx = BX 7/ 4 + O(X + ) (B > 0) (13)and
X
1 4(x) dx = CX 2 + O(X + ) (C > 0) (14)
with = 75 , =2312 . This improves on the values =
4728 , =
4523 , obtained in [19].
Moreover, (13) and (14) remain valid if ( x) is replaced by (x),
since their proofs usednothing more besides (5) and the bound d(n)
n . Hence from (12) and the analoguesof (13)(14) for (x), we infer
then that
T
0E 3(t) dt = B1T 7/ 4 + O(T 5/ 3 log3/ 2 T ) (B1 > 0),
T
0E 4(t) dt = C 1T 2 + O(T 23/ 12+ ) (C 1 > 0). (15)
-
8/6/2019 Ivic Divisor Problem
4/121
A. Ivic / Central European Journal of Mathematics 2(4) 2004
494508 497
The main aim of this paper is to provide an estimate for the
upper bound of the fourthmoment of E (t), which is the rst
non-trivial upper bound for a higher moment of E (t).The result is
the following
Theorem 1.1. We have T
0(E (t))4 dt T 16/ 9+ . (16)
Note that the bounds (8) and (16) do not seem to imply each
other. For the proof of (16) we shall need several lemmas, which
will be given in Section 2. The proof of Theorem 1.1 will be given
in Section 3. Finally, in Section 4, it will be shown how themethod
of proof of Theorem 1.1 can give a proof of
Theorem 1.2. Let T 1/ 5+
G T, T < t 1 < < t R 2T, tr +1 t r 5G (r =1, , R 1).
Then R
r =1
t r + G
t r G | (12 + it )|2 dt
4
T 2+ G 2 + RG 4T . (17)
The bound in (17) easily gives the well-known bound (see Section
4)
T
0 | ( 12 + it )
|12 dt T 2+ , (18)
due to D.R. Heath-Brown [3] (who had log 17 T instead of the T
-factor). It is still es-sentially the sharpest result concerning
high moments of | ( 12 + it )|. General sums of zeta-integrals over
short intervals, analogous to the one appearing in (17), were
treatedby the author in [9].
2 The necessary lemmas
Lemma 2.1. (O. RobertP. Sargos [17]). Let k 2 be a xed integer
and > 0 be given.Then the number of integers n1, n2, n 3, n 4
such that N < n 1, n 2, n3, n4 2N and
|n1/k1 + n
1/k2 n
1/k3 n
1/k4 | < N 1/k
is, for any given > 0, N (N 4 + N 2). (19)
Lemma 2.2. Let 1 G T/ log T . Then we have
E (T ) 2
G
0E (T + u) e u
2 /G 2 du + O(G log T ), (20)
and E (T )
2G
0E (T u) e u
2 /G 2 du + O(G log T ). (21)
-
8/6/2019 Ivic Divisor Problem
5/121
498 A. Ivic / Central European Journal of Mathematics 2(4) 2004
494508
Proof of Lemma 2.2 . The proofs of (20) and (21) are analogous,
so only the formerwill be treated in detail. From (2) we have, for
0 u T ,
0
T + uT | ( 12 + it )|2 dt = ( T + u) log T + u2 + 2 1 (22)
T log T 2 + 2 1 + E (T + u) E (T ). (23)This gives
E (T ) E (T + u) + O(u log T ),hence
G log T
0E (T ) e u
2 /G 2 du G log T
0(E (T + u) + O(u log T )) e u
2 /G 2 du.
The proof of (20) is completed when we extend the integration to
[0 , ) making a smallerror, and recall that
0 e
u 2 /G 2 du = 12G,
0 ue u 2 /G 2 du = 12 G.
Lemma 2.3. Let 1 G T . Then we have
T 2
=2
G
0
T 2
u2
e u2 /G 2 du + O(GT ). (24)
Proof of Lemma 2.3. Both the cases of the + and sign in (24) are
treated analogously.For example, we have12G T 2
0
T 2 +
u2 e
u 2 /G 2 du=
0
(T ) T 2 + u2 e u2 /G 2 du
= G log T
0 T
2 ) ( T 2 + u2 e u2 /G 2 du + O(1)
G log T
0 2 T n 2 (T + u)
(1)n d(n) + O((1 + |u|)log T ) du G2T ,
where we used (4) and d(n) n. This establishes (24).
The next lemma is F.V. Atkinsons classical explicit formula for
E (T ) (see [2], [7] or[8]).
Lemma 2.4. Let 0 < A < A be any two xed constants such
that AT < N < A T , and let N = N (T ) = T/ (2) + N/ 2 (N 2/
4 + NT/ (2))1/ 2. Then
E (T ) = 1(T ) + 2(T ) + O(log2 T ), (25)
where 1(T ) = 2 1/ 2(T/ (2))1/ 4
n N
(1)n d(n)n 3/ 4e(T, n) cos(f (T, n)) , (26)
2(T ) = 2n N
d(n)n 1/ 2(log T/ (2n )) 1 cos(T log T/ (2n ) T + / 4), (27)
-
8/6/2019 Ivic Divisor Problem
6/121
-
8/6/2019 Ivic Divisor Problem
7/121
500 A. Ivic / Central European Journal of Mathematics 2(4) 2004
494508
We combine now (20) with (24) with the + sign (when E (T ) 0) or
(21) with (24) withthe sign (when E (T ) 0), to obtain by the
Cauchy-Schwarz inequality
(E (t))2 G 1
G log T
G log T
e u2 /G 2 (E (t + u))2 du + G2T , (36)
provided that T t 2T, T G T 5/ 12. Keeping in mind the preceding
discussionwe thus have (replacing ( t + u)1/ 4 with t1/ 4 by
Taylors formula, with the error absorbedby the last term in (37))
by using (6), (25), (35) and (36),
(E (t))2 G 1 G log T
G log T e u
2 /G 2 ( 23(X ; u) + 24(X, N ; u) +
25(X, N ; u)) d u
+ T 1+ N 1, (37)
where we set
3(X ; u) := t1/ 4 n X
(1)n d(n)n 3/ 4
e(t + u, n ) cos(f (t + u, n )) cos( 8n (t + u) / 4) , (38) 4(X,
N ; u) := t1/ 4
X
-
8/6/2019 Ivic Divisor Problem
8/121
A. Ivic / Central European Journal of Mathematics 2(4) 2004
494508 501
left-hand side of (33) is majorized by the maximum taken over
|u| G log T times
2T
T (E (t))2( 24(X, N ; u) +
25(X, N ; u) +
26(X ; u) + T
1+ N 1) dt
2T
T (E (t))4 dt
2T
T 44(X, N ; u) + 45(X, N ; u) + 46(X ; u) dt
1/ 2
+ T 7/ 3+ N 1, (42)
where we used the Cauchy-Schwarz inequality for integrals and
(8). Thus from (42) wehave the key bound
2T
T (E (t))4 dt max
|u | G log T 2T
T 44(X, N ; u) +
45(X, N ; u) +
46(X ; u) dt
+ T 7/ 3+ N 1. (43)
To evaluate the integrals on the right-hand side of (43) we note
rst that
2T
T 44(X, N ; u) + . . . dt
5T/ 2
T/ 2(t) 44(X, N ; u) + . . . dt, (44)
where (t) is a smooth, nonnegative function supported in [ T/ 2,
5T/ 2] , such that (t) =1 when T t 2T . The integrals of 44(X, N ;
u), 45(X, N ; u) and 46(X ; u) are allestimated analogously. The
sums over n are divided into log T subsums of the form
K 0), where := 8(m + n k l ). (45)
Therefore, in the case of 5(X, N ; u), there remains the
estimate
2T
T 45(X, N ; u) dt 1 + T
1+ max|u | G log T
supX K N
5T/ 2
T/ 2(t)
K
-
8/6/2019 Ivic Divisor Problem
9/121
502 A. Ivic / Central European Journal of Mathematics 2(4) 2004
494508
Proceeding analogously as in (47), we obtain that
2T
T 46(X ; u) dt T
1+ max1 K X, | | T 1 / 2
T 1K 3(K 4K 1/ 2| |+ K 2)
T
(T 1/ 2
X 13/ 2
+ X 5
), (48)since instead of (mnkl ) 3/ 4 in (46) now we shall have
(mnkl )3/ 4t 1 (see (41)).
The estimation of 4(X, N ; u) (see (39)) presents a technical
problem, since the cosinescontain the function f (t, n ), and Lemma
2.1 cannot be applied directly. First we notethat, by using (28),
we can expand the exponential in power series to get rid of the
termsa5n5/ 2t 3/ 2 + . . . . In this process the main term will be
1, and the error terms will makea contribution which will be (for
shortness we set a = 8, b = 1623 and = t + u)
max|u | G log T
supX K N
T 5T/ 2
T/ 2(t)
K
-
8/6/2019 Ivic Divisor Problem
10/121
A. Ivic / Central European Journal of Mathematics 2(4) 2004
494508 503
integrations by parts over t, only the portion of the sum for
which | | T 1/ 2 willremain, where is given by (45). In the
remaining sum we use (49) (once with A = 14 and once with A = 34 ),
noting that e
iz = cos z + i cos(z 12 ). We remark that, for|v| u0, we can use
the crude estimate (u) T 1/ 6, hence for this portion the
estimationwill be quite analogous to the preceding case. Next we
note that
u 1
u 0 1/ 8v 1/ 4 exp( ib 1/ 4v3/ 2 8nv ) dv log T ( = t + u, |u| G
log T ),
writing the integral as a sum of log T integrals over [U, U ]
with u0 U < U 2U u1, and applying the second derivative test to
each of these integrals. We also remarkthat the contribution of the
O-term in (32) will be, by trivial estimation,
u 0T 1/ 8u 7/ 4 du T 1/ 8u 3/ 40 1
if we suppose that (50) is satised. It remains yet to deal with
the integral with v > u 1in (49), when we note that
v
b 1/ 4v3/ 2 8nv T 1/ 4v1/ 2 (v > u 1),provided that C in (50)
is sufficiently large. Hence by the rst derivative test
u 1(v) cos(8n ( + v)
14 ) dv
1 + T 1/ 8u 1/ 41 T 1/ 4u 1/ 21
1 + T 1/ 4K 3/ 4 1 + T 1/ 4X 3/ 4 1,
since K X T 1/ 3. Thus the contribution of the integrals on the
right-hand side of (49) is log T .
Then we can proceed with the estimation as in the case of 5(X, N
; u) to obtain
2T
T 44(X, N ; u) dt T
3/ 2+ N 1/ 2 + T 2+ X 1.
Gathering together all the bounds, we see that the integral in
(33) is
T T 3/ 2N 1/ 2 + T 2X 1 + T 1/ 2X 13/ 2 + X 5 + T 7/ 3+ N 1 ,
(51)
provided that (40) holds. Finally we choose
X = T 1/ 3 , N = T 5/ 9,
so that (40) is fullled. The above terms are then T 16/ 9+ , and
the proof of Theorem1 is complete. The limit of the method is the
bound T 2X 1 T 5/ 3, which wouldyield the exponent 5 / 3 + in (16).
The true order of the integral in (16), and in generalthe order of
the k-th moment of E (t), is elusive. This comes from the denition
E (t) =E (t) 2 (t/ (2)), which makes it difficult to see how much
the oscillations of thefunctions E and cancel each other.
-
8/6/2019 Ivic Divisor Problem
11/121
504 A. Ivic / Central European Journal of Mathematics 2(4) 2004
494508
4 The proof of Theorem 1.2
We shall show now how the method of proof of our Theorem 1.1 may
be used to yieldTheorem 1.2. Our starting point is an expression
for the integral
tr +2 G
t r 2Gr (t)| ( 12 + it )|2 dt, (52)
where t r is as in the formulation of Theorem 1.2, and r C is a
non-negative function
supported in [ t r 2G, t r + 2 G] that equals unity in [ t r G,
t r + G]. The integral in (52)majorizes the integral
t r + G
t r G | (12 + it )|2 dt, (53)
which is of great importance in zeta-function theory (see K.
Matsumoto [15] for an ex-tensive account on mean square theory
involving (s)). One can treat the integral in (52)by any of the
following methods.(a) Using exponential averaging (or some other
smoothing like r above), namely the
Gaussian weight exp( 12 x2), in connection with the function E
(T ), in view of F.V.Atkinsons well-known explicit formula (cf.
Lemma 2.4). This is the approach em-ployed originally by D.R.
Heath-Brown [3].
(b) One can use the Voronoi summation formula (e.g., see [8,
Chapter 3]) for the explicitexpression (approximate functional
equation) for
| ( 12 + it )
|2 = 1( 12 + it )
2( 12 + it ),where (s) = (s) (1 s), namely
(s) = 2 s s 1 sin( 12 s )(1 s).Voronois formula is present
indirectly in Atkinsons formula, so that this approachis more
direct. The effect of the smoothing function r in (53) is to
shorten the sumapproximating | |2 to the range T 2 (1 G 1T ) n T 2
(T = t r ). After this nointegration is needed, and proceeding as
in [7, Chapters 7-8] one obtains that theintegral in (53) equals
O(GT ) plus a multiple of
t r +2 G
t r 2Gr (t)
k T 1+ G 2(1)kd(k)k 1/ 2
14
+t
2k
1/ 4
sin f (t, k ) dt, (54)
where f (t, k ) is given by (28).(c) Instead of the Voronoi
summation formula one can use the (simpler) Poisson sum-
mation formula, namely
n =1
f (n) =
0f (x) dx + 2
n =1
0f (x) cos(2nx ) dx,
provided that f (x) is smooth and compactly supported in (0 , ).
In [11] a sketchof this approach is given.
We begin now the derivation of (17), simplifying rst in (54) the
factor (1 / 4 +t/ (2k)) 1/ 4 by Taylors formula, and then raising
the expression in (54) to the fourth
-
8/6/2019 Ivic Divisor Problem
12/121
A. Ivic / Central European Journal of Mathematics 2(4) 2004
494508 505
power, using Holders inequality for integrals. It follows that
the sum in (17) is boundedby
RG 4T + T 1G3R
r =1
t r +2 G
t r 2Gr (t)
k T 1+ G 2(
1)kd(k)k 1/ 4 sin f (t, k )
4dt
RG 4T + T 1G3 5T/ 2
T/ 2(t)
k T 1+ G 2(1)kd(k)k 1/ 4 sin f (t, k )
4dt, (55)
where (t) is a non-negative, smooth function supported in [ T/
2, 5T/ 2] such that (t) = 1for T t 2T , hence (m )(t) m T m .
Therefore it suffices to bound the expression
I K := 5T/ 2
T/ 2(t)
K
-
8/6/2019 Ivic Divisor Problem
13/121
506 A. Ivic / Central European Journal of Mathematics 2(4) 2004
494508
which is (17).It remains to show how (17) gives the twelfth
moment estimate (18). Write
2T
T | ( 12 + it )|12 dt
r T +1| ( 12 + i r )|12, (57)
where for r = 1 , 2, . . . we set
| ( 12 + i r )| := maxT + r 1 t T + r | ( 12 + it )|.Let {t r,V
}be the subset of { r }such that
V | ( 12 + it r,V )| 2V (r = 1 , . . . , R V ),where clearly V
may be restricted to O(log T ) values of the form 2m such that log
T V T 1/ 6, since ( 12 + it ) = o(t1/ 6) (see [7, Chapter 7]). Now
since we have (see e.g., [8,Theorem 1.2]), for xed k
N ,
| ( 12 + it )|k log t t+ 12
t 12| ( 12 + it + iu)|k du + 1 ,
it follows that, for some points t r ([T, 2T ]) with r = 1 , . .
. , R , R RV , 1 GT, t
r +1 t
r 5G,
RV V 2 R V
r =1| ( 12 + it r,V )|2
R V
r =1
log T t r,V + 12
t r,V 12| ( 12 + it )|2 dt + 1
R
r =1
log T t r + G
t r G | (12 + it )|2 dt + RV log T
log T (R )3/ 4R
r =1
t r + G
t r G | (12 + it )|2 dt
1/ 4+ RV log T
T (RV G + R3/ 4V T
1/ 2G 1/ 2),
where the estimate of Theorem 1.2 was used, with RV replacing R.
If we take G = V 2T 2 ,then we obtain
R1/ 4V T 1/ 2+ G 3/ 2,
which givesRV T 2+ G 6 T 2+ V 12 .
Then the portion of the sum in (57) for which | ( 12 + i r )| T
1/ 10+ islog T max
V T 1 / 10+ RV V 12 T 2+ .
-
8/6/2019 Ivic Divisor Problem
14/121
A. Ivic / Central European Journal of Mathematics 2(4) 2004
494508 507
But for values of V such that V T 1/ 10+ , the above bound
easily follows from the largevalues estimate (the fourth moment) R
T 1+ V 4. This shows that the integral in (57)is T 2+ , and proves
(18). Note that the author [9, Corollary 1] proved the bound
R
r =1
t r + G
t r G| ( 12 + it )|4 dt
2RG 2 log8 T + T 2G 1 logC T (58)
for some C > 0, where T < t 1 < . . . < t R 2T , t r
+1 t r 5G for r = 1 , . . . , R 1 and1 G T . The bound (58), which
is independent of Theorem 1.2, was proved by amethod different from
the one used in this work. Like (17), the bound (58) also leads
tothe twelfth moment estimate (18).
References[1] F.V. Atkinson: The mean value of the zeta-function
on the critical line, Quart. J.
Math. Oxford , Vol. 10, (1939), pp. 122128.
[2] F.V. Atkinson: The mean value of the Riemann zeta-function,
Acta Math. , Vol. 81,(1949), pp. 353376.
[3] D.R. Heath-Brown: The twelfth power moment of the Riemann
zeta-function,Quart. J. Math. (Oxford) , Vol. 29, (1978), pp.
443462.
[4] D.R. Heath-Brown: The distribution of moments in the
Dirichlet divisor problems,Acta Arith. , Vol. 60, (1992), pp.
389415.
[5] M.N. Huxley: Area, Lattice Points and Exponential Sums ,
Oxford SciencePublications, Clarendon Press, Oxford, 1996.
[6] A. Ivic: Large values of the error term in the divisor
problem, Invent. Math. , Vol.71, (1983), pp. 513520.
[7] A. Ivic: The Riemann zeta-function , John Wiley & Sons,
New York, 1985.
[8] A. Ivic: The mean values of the Riemann zeta-function , LNs
82, Tata Inst. of
Fundamental Research, Bombay (distr. by Springer Verlag, Berlin
etc.), 1991.[9] A. Ivic: Power moments of the Riemann zeta-function
over short intervals, Arch.
Mat. , Vol. 62, (1994), pp. 418424.
[10]A. Ivic: On some problems involving the mean square of | (
12 + it )|, Bull. CXVI Acad.Serbe, Classe des Sciences
mathematiques , Vol. 23, (1998), pp. 7176.[11]A. Ivic: Sums of
squares of | ( 12 + it )| over short intervals, Max-Planck-Institut
f ur Mathematik, Preprint Series , Vol. 52, (2002), pp. 12.[12]A.
Ivic and P. Sargos: On the higher moments of the error term in the
divisor problem ,to appear.
[13]M. Jutila: Riemanns zeta-function and the divisor problem,
Arkiv Mat., Vol. 21,(1983), pp. 7596; ibid. Vol. 31, (1993), pp.
6170.
[14]M. Jutila: On a formula of Atkinson, In: Proc. Coll. Soc. J.
BolyaiBudapest81 ,Vol. 34, North-Holland, Amsterdam, 1984, pp.
807823.
http://www.ingentaconnect.com/content/external-references?article=0003-889X(1983)21L.75[aid=6407442]http://www.ingentaconnect.com/content/external-references?article=0003-889X(1983)21L.75[aid=6407442]http://www.ingentaconnect.com/content/external-references?article=0003-889X(1994)62L.418[aid=6407443]http://www.ingentaconnect.com/content/external-references?article=0003-889X(1994)62L.418[aid=6407443]http://www.ingentaconnect.com/content/external-references?article=0020-9910(1983)71L.513[aid=6407444]http://www.ingentaconnect.com/content/external-references?article=0020-9910(1983)71L.513[aid=6407444]
-
8/6/2019 Ivic Divisor Problem
15/121
508 A. Ivic / Central European Journal of Mathematics 2(4) 2004
494508
[15]K. Matsumoto: Recent developments in the mean square theory
of the Riemann zetaand other zeta-functions, In: Number Theory ,
Birkhauser, Basel, 2000, pp. 241286.
[16]T. Meurman: A generalization of Atkinsons formula to
L-functions, Acta Arith. ,Vol. 47, (1986), pp. 351370.
[17]O. Robert and P. Sargos: Three-dimensional exponential sums
with monomials, J.reine angew. Math. (in print).
[18]E.C. Titchmarsh: The theory of the Riemann zeta-function ,
2nd Ed., UniversityPress, Oxford, 1986.
[19]K.-M. Tsang: Higher power moments of ( x), E (t) and P (x),
Proc. London Math.Soc.(3) , Vol. 65, (1992), pp. 6584.
-
8/6/2019 Ivic Divisor Problem
16/121
CEJM 2(4) 2004 509515
On the Apostol-Bernoulli Polynomials
Qiu-Ming Luo
Department of Mathematics,Jiaozuo University,
Jiaozuo City, Henan 454003, The Peoples Republic of China
Received 19 July 2001; accepted 23 May 2003
Abstract: In the present paper, we obtain two new formulas of
the Apostol-Bernoullipolynomials (see On the Lerch Zeta function.
Pacic J. Math., 1 (1951), 161167 .), using theGaussian
hypergeometric functions and Hurwitz Zeta functions respectively,
and give certainspecial cases and applications.c Central European
Science Journals. All rights reserved.
Keywords: Bernoulli numbers, Bernoulli polynomials,
Apostol-Bernoulli numbers, Apostol-Bernoulli polynomials, Gaussian
hypergeometric functions, Stirling numbers of the second
kind,Hurwitz Zeta functions, Lerch functional equation MSC (2000):
Primary: 11B68; Secondary: 33C05, 11M35, 30E20
1 Introduction
An analogue of the classical Bernoulli polynomials were dened by
T. M. Apostol (see[1]) when he studied the Lipschitz-Lerch Zeta
functions. We call this polynomials theApostol-Bernoulli
polynomials. First we rewrite Apostols denitions below
Denition 1.1. Apostol-Bernoulli polynomials B n (x, ) are dened
by means of thegenerating function (see [1, p.165 (3.1)] or [4,
p.83])
zexz
ez
1=
n =0
B n (x, )zn
n!, (|z + ln | < 2) (1)
setting = 1 in (1), Bn (x) = B n (x, 1) are classical Bernoulli
polynomials.
[email protected]
-
8/6/2019 Ivic Divisor Problem
17/121
510 Q.-M. Luo / Central European Journal of Mathematics 2(4)
2004 509515
Denition 1.2. Apostol-Bernoulli numbers B n () := B n (0, ) are
dened by means of the generating function
z
ez
1=
n =0
B n ()zn
n!, (|z + ln | < 2) (2)
setting = 1 in (2), Bn = B n (1) are classical Bernoulli
nunbers.
T.M.Apostol not only gave elementary properties of polynomials B
n (x, ) in [1], butalso obtained the recursion formula for the set
of numbers B n () using the Stirling num-bers of the second kind
(see [1, p. 166 (3.7)]) as follows
Bn() = n
n 1
k=1
(
1)k k!k (
1) 1 k S (n
1, k), (n
N0; () > 0, = 1) (3)
where S (n, k ) denote the Stirling numbers of the second kind
which are dened by meansof the following expansion (see [3, p.207,
Theorem B])
xn =n
k =0
xk
k!S (n, k ). (4)
By applying binomial series expansion and Leibnizs rule, we rst
obtain the repre-sentation of the polynomials B n (x, ) involving
the Gaussian hypergeometric functions,and thereout deduce Apostols
formula (3); afterward we prove Theorem 3.1 using Lerchfunctional
equation with related Hurwitz Zeta function. Furthermore we show
that themain result in [9, p.1529, Theorem A] is only a special
case of Theorem 3.1.
2 Apostol-Bernoulli Polynomials and GaussianHypergeometric
functions
Theorem 2.1. If n is a positive integer and () > 0, = 1 are
complex numbers, thenwe have
B n (x, ) = nn 1
l=0
n 1l
l( 1) l 1l
j =0
( 1) jl
j j l(x + j )n l 1
F [l n + 1 , l; l + 1; j/ (x + j )] (5)
where F [a, b; c; z] denotes Gaussian hypergeometric functions
dened by (cf. [5, p.44 (4)])
F [a, b; c; z] :=
n =0
(a)n (b)n(c)n
zn
n!, |z| < 1 (6)
where ()0 = 1 , ()n = ( + 1) ( + n 1) = ( + n )( ) , (n 1).
-
8/6/2019 Ivic Divisor Problem
18/121
Q.-M. Luo / Central European Journal of Mathematics 2(4) 2004
509515 511
Proof 2.2. We differentiate both sides of (1) with respect to
the variable z. ApplyingLeibnizs rule yields
B n (x, ) = D nzzexz
e z 1z =0
, Dz =ddz
.
=( 1) 1n
k=0
nk
kxn k D k 1z
1(ez 1) + 1
1
z =0.
(7)
Since binomial series expansion
(1 + w) 1 =
l=0
( w)l , |w| < 1 (8)
setting w =
1 (ez
1), we have
B n (x, ) = ( 1) 1n
k=0
nk
kx n kk 1
l=0
1
l
D k 1z {(ez 1)l}| z =0 . (9)
By the denition of Stirling numbers of the second kind (see [5,
p.58 (15)])
(ez 1)l = l!
r = l
S (r, l )zr
r !(10)
yields
B n (x, ) =n
k=1
nk
kk 1
l=0
( 1)l l( 1) l 1l!S (k 1, l)xn k . (11)
We change sum order of k and l, and using the formula below (see
[5, p.58 (20)])
S (n, k ) =1k!
k
j =0( 1)
k j k j j
n
(12)
we obtain
B n (x, ) = nn 1
l=0
l( 1) l 1xn k l 1l
j =0
( 1) jl
j j l
n l 1
k=0
n 1n k 2
jx
k
. (13)
Applying (6) to (13) readily yields
B n (x, ) = nn 1
l=0
n 1l
l( 1) l 1xn k l 1
l
j =0
( 1) jl
j j l .F [l n + 1 , 1; l + 1; j/x ] (14)
-
8/6/2019 Ivic Divisor Problem
19/121
512 Q.-M. Luo / Central European Journal of Mathematics 2(4)
2004 509515
Finally, we apply the known transformation [10, 15.3.4]
F [a, b; c; z] = (1 z) a F [a, c b; c; z/ (z 1)],
and (2.2) immediately obtain (5).
Remark 2.3. H .M. Srivastava and P. G Todorov considered earlier
the formula of thegeneralized Bernoulli polynomials (see [6, p.510
(3)]), for = 1, which is a complemen-tarity of our result (5), for
= 1, as follows
Bn (x) = nk =0nk
k!(2k)!
k
j =0
( 1) jk j
j 2k (x + j )n k
F [k n, k 1; 2k + 1; j/ (x + j )]. (15)
Remark 2.4. We will also apply the representation (5) in order
to derive an interestingspecial case considered by T. M. Aspotol in
(3).
By the well-known formula [10, 15.1.20]
F [a, b; c; 1] =(c)(c a b)(c a)(c b)
(c = 0 , 1, 2, . . . , (c a b) > 0),
upon setting a = l n + 1 , b = l, and c = l + 1 yields
F [l n + 1 , l; l + 1; 1] =n 1
l
1
, (0 l n). (16)
In view of (16), the special case of our formula (5) when x = 0
gives Apostols represen-tation (3).
Remark 2.5. The formula of classical Bernoulli numbers
considered by H. W. Gould in[7, p.49, Eq.(17)] is also a
complementarity of the Apostols formula (3), for = 1
Bn
=n
k =0
( 1)kn + 1
n k
n + k
k
1
S (n + k, k). (17)
Remark 2.6. There is a relationship between Apostol-Bernoulli
polynomials B n (x, )and Stirling numbers of the second kind:
B n (x, ) =n
k =1
knk
k 1
j =0
( 1) j j ( 1) j 1 j !S (k 1, j )xn k , ( = 1). (18)
Remark 2.7. Recently, Luo also obtained the relation between the
classical Bernoullipolynomials and the Stirling numbers of the
second kind [8], which is a complementarityof (18), for = 1:
Bn (x) =n
k =0
n k
s =0
ns + k
k + s + 1s
s + 2 kk
1
S (s + 2 k, k)xn s k . (19)
-
8/6/2019 Ivic Divisor Problem
20/121
Q.-M. Luo / Central European Journal of Mathematics 2(4) 2004
509515 513
3 Apostol-Bernoulli polynomials and the Hurwitz ZetaFunction
It is well-known that Hurwitz-Lerch Zeta functions are dened by
the innite series (see
[5, p.121, (1)])
(z,a,s ) =
k =0
zk
(a + k)s, (20)
(a C \ Z 0 ; s C , when |z| < 1; (s) > 1, when |z| = 1)
.
Setting z = e2iz gives the Lipschitz-Lerch Zeta function (z,a,s
) (see [1, p.161])
(z,a,s ) =
k=0
e2kiz
(a + k)s, (21)
(a C \ Z 0 ; z R \ Z , (s) > 0; z Z , (s) > 1).
When s = n is negative integer, the Lipschitz-Lerch Zeta
function (z,a,s ) was evalu-ated by T. M. Apostol using polynomials
B n (x, ) (see [1, p.164]):
(z,a, n) = B n +1 (a, e 2iz )
n + 1. (22)
If we set z = 0 in (22), then we have the formula (see [2,
p.264, Theorem 12.13])
(a, n) = Bn +1 (a)n + 1
(23)
where (a, s ) are Hurwitz Zeta functions dened by (a, s ) :=
k=01
(a + k)s(see [2,
p.249]) and Bn (x) are classical Bernoulli polynomials.Further,
if we set a = 0 in (23), then we have the known formula (see [2,
p.266,
Theorem 12.16]):
( n) = Bn +1
n + 1(24)
where (s) are Riemann Zeta functions dened by (s) := k =01ks
(see [2, p.249]) andBn := Bn (0) are classical Bernoulli
numbers.
In this section, we will apply the Lerch functional equation to
obtain the representa-tion of Apostol-Bernoulli polynomials B n (x,
) at rational points x =
pq
. Clearly, it is a
generalized form of [9, p.1529, Theorem A].
Theorem 3.1. For n N \ { 1}; pZ , qN , z C we have
B n pq
, e2iz = n!
(2q)n
q
j =1
z + j 1
q, n exp
n2
2(z + j 1) p
qi
+q
j =1
j z
q, n exp
n2
+2( j z) p
qi . (25)
-
8/6/2019 Ivic Divisor Problem
21/121
514 Q.-M. Luo / Central European Journal of Mathematics 2(4)
2004 509515
Proof 3.2. In view of the Lipschitz-Lerch Zeta functions (z,a,s
) (21), and using theknown series identity
k=1
f (k) =q
j =1
k =0
f (qk + j ), (qN ) (26)
we obtain that
(z,a,s ) = q sq
j =1
a + j 1
q, s exp(2(kq + j 1)iz ), (k N 0). (27)
Setting z =pq
yields
p
q,a ,s = q s
q
j =1
a + j 1
q, s exp
2( j 1) pi
q, ( pZ , qN ). (28)
On the other hand, by the Lerch functional equation (see [5,
p.125, (29)]), for 0 < z 0 and B(r / a, w) is a beta
function.
Distribution (5) is a special limit case of the Bessel
distribution investigated by T.Srodka (1973). It was also analyzed
by J. Seweryn (1986) and by W. Ogi nski (1979) and
was applied in reliability theory.Let us consider the GNBD (4)
that depends on cy :
P (x; n,cy) =n
n + xn + x
x(cy)x (1 cy)n + x x , x = 0 , 1, 2, . . . , (6)
where 0 < cy < 1, n = 1, 2, . . . , cy < 1, 1, and Y is
a random variable withGBD (5).
Theorem 3.3. The probability function of the compound
distribution
GNBD Y
GBD
is given by the formula
P GB (x) = D
k=0
( c)kn + x x
k
(bw)ka B
x + r + ka
, w , (7)
where
D =n
n + x
n + x
xcx (bw)
xa
B(r/a,w ),
x = 0 , 1, 2, . . . ; a, b, w, r > 0, where cy < 1, 0 <
cy < 1, n > 0, 1.
Proof. From formulas (3), (5) and (6) we have
P GB (x) = aD 1
(bw)1 /a
0
yx+ r 1 1 ya
bw
w 1
(1 cy)n + x xdy
= aD 1
k=0
( c)kn + x x
k
(bw)1 /a
0
yx+ r + k 1 1 ya
bw
w 1
dy,
where
D1 =n
n + x
n + x
xcx
(bw)r/a B(r/a, w ).
-
8/6/2019 Ivic Divisor Problem
38/121
T. Gerstenkorn / Central European Journal of Mathematics 2(4)
2004 527537 531
Substituting ya /bw = t, we get
P GB (x) = D
k=0
( c)kn + x x
k(bw)
ka
1
0
tx + r + k
a 1(1 t)w 1dt,
x = 0 , 1, 2, . . . , a, b, w, r > 0, n > 0, 1, 0 < c
1
(bw)1/a
and, from the denition of the beta function, we obtain (7). In
the special case whenn, Nwe have a simpler formula given by
P GB (x) = Dn + x x
k=0
( c)kn + x x
k
(bw)ka B
x + r + ka
, w . (8)
Case (ii) of (4) the case of = 0 is discussed next in Section
3.2.
3.2 Special cases
In the case when = 0 (the binomial distribution) the proof of
Theorem 3.3 is similarto, but even simpler than, the proof given
above. More specically, we get
P GB (x) = D
n x
k=0
( c)kn x
k(bw)
ka B
x + r + ka
, w
where
D =
n
xcx (bw)
xa
B(r/a,w ).
Here we get a compound of the binomial distribution with the
generalized beta dis-tribution:
binomial Y
generalized beta ( = 0)
P 0GB (x) =
n
xcx (bw)
xa
B(r/a,w )
n x
k=0
( c)kn x
k
(bw)ka B
x + r + ka
, w , (9)
x = 0 , 1, 2, . . . , n .
T. Gerstenkorn (1982, p. 90, (6)) obtained the same result.
Moreover, Gerstenkorndetermined some special cases of this
compoundcompounds of the binomial distributionwith the beta
distribution, generalized gamma, exponential, arcsine, Erlang, 2,
gamma,
-
8/6/2019 Ivic Divisor Problem
39/121
532 T. Gerstenkorn / Central European Journal of Mathematics
2(4) 2004 527537
Rayleigh, Maxwell, , normal truncated, and Weibull. In the
Gerstenkorn paper, factorialmoments and ordinary moments of order l
of the distributions are also determined.
From (8) we also have the compound of the generalized negative
binomial with thebeta distribution
GNBD Y beta ( b = 1/ w, a = 1)
P B(x) =n ck
n + x
n + x
x
B(r, w )
k=0
( 1)kn + x x
kB(x + r + k, w).
It can be demonstrated that
k=0
( 1)kn + x x
kB(x + r + k, w) = B(x + r, w + n + x x)
holds. Hence, we get
P B(x) =n ck
n + xn + x
x
B(x + r, w + n + x x)B(r, w )
, (10)
a result that, for = 1 and c =1, yieldsnegative binomial
Y beta
P 1B(x) =n + x 1
x
B(x + r, w + n)B(r, w )
(11)
or, after using the formula
n [x,
1] = n(n + 1)( n + 2) . . . (n + x 1),
P 1B(x) =n [x, 1]r [x, 1]w[n, 1]
x!(r + w)[x+ n, 1].
Furthermore, a consequence of Theorem 3.3 is for = 1, n = 1 , 2,
. . . the compounddistribution
Pascal binomial Y
generalized beta
P 1GB (x) =n + x 1
x
cx (bw) xaB(r/a,w )
n
k=0
( c)kn
k(bw)
ka B
x + r + ka
, w . (12)
Some special compounds, as mentioned before for
binomialgeneralized beta, can beobtained for distribution (12).
-
8/6/2019 Ivic Divisor Problem
40/121
T. Gerstenkorn / Central European Journal of Mathematics 2(4)
2004 527537 533
4 Factorial moments and ordinary (crude) moments of a com-pound
of the negative binomial distribution with the gener-alized beta
distribution
Denition 4.1. Let X y and X be random variables with
distribution functions F (x|y)and H (x), respectively (see (1)),
and let parameter y have distribution G(y). Then, whenone keeps in
mind the formula for the so-called factorial polynomial
x[l] = x(x 1)(x 2) . . . (x (l 1)),
m[l] = E (X [l]) =
E (X [l]y )dG(y) (13)
is called a factorial moment of order l of the variable X with
compound distribution (1).
Relation (13) will be symbolized as follows:
E (X [l]y ) Y G(y). (14)
Theorem 4.2. The factorial moment of order l of the compound
distribution negativebinomial
Y generalized beta is given by the formula
mNB GB[l] = n[l, 1]
cl
(bw)l
a
B(r/a, w )
k=0
lk
( c)k(bw) ka B l + r + ka , w (15)
Proof. The factorial moment of order l of the negative binomial
distribution is given bythe formula
mNB[l] (n, p) = pq
l
n [l, 1], 0 < p < 1, q = 1 p
(W. Dyczka (1973, p. 223, (63))). Consequently, by (14), the
factorial moment of this
order of distribution (12) is, if we let p = cy, the
following:mNB GB[l] = m
NB[l] (n,cy) Y GB (y; a,b,r,w )
=an [l, 1]
(bw)r/a B(r/a,w )
(bw)1 /a
0
cy1 cy
l
yr 1 1 ya
bw
w 1
dy
=an [l, 1]cl
(bw)r/a B(r/a,w )
k=0
l
k( c)k
(bw)1 /a
0
yl+ r + k 1 1 ya
bw
w 1
dy.
Substituting ya /bw = t, we get
m [l] =n [l, 1]cl(bw)
la
B(r/a,w )
k=0
l
k( c)k(bw)
ka
1
0
tl+ r + k
a 1(1 t)w 1dt,
-
8/6/2019 Ivic Divisor Problem
41/121
534 T. Gerstenkorn / Central European Journal of Mathematics
2(4) 2004 527537
a result that yields (15).
A special case of Theorem 4.2 is the following theorem.
Theorem 4.3. The factorial moment of order l of the compound
distribution negativebinomial Y
beta ( b = 1 /w, a = 1 , c = 1) is given by the formula (for w
> l)
mNB B[l] = n[l, 1] B(l + r, w l)
B(r, w )(16)
or by the formula
mNB B[l] =n [l, 1]r [l, 1]
(w 1)[l]. (17)
Proof. In this case we have
mNB B[l] =n [l, 1]
B(r, w )
k=0
l
k( 1)kB(l + r + k, w).
It can be demonstrated that
k=0
lk
( 1)kB(l + r + k, w) = B(l + r, w l),
thus giving formula (16). Formula (17) is then evident.Further
special cases can easily be obtained by taking account of the
remarks included
in Section 3.
The ordinary (crude) moments of the compound distributions under
consideration are
obtained by using the formulam l =
l
k=0
S lkm [k],
where S lk stands for the so-called Stirling numbers of the
second kind. Bohlmann (1913)seems to be the rst to give this
formula; the tables for these numbers can be found, forinstance, in
A. Kaufmann (1968) or in J. Lukasiewicz and M. Warmus (1956).
References[1] G. Bohlmann: Formulierung und Begr undung zweier
Hilfssatze der mathematischen
Statistik, Math. Ann. , Vol. 74, (1913), pp. 341409.
[2] M.E. Cansado: On the compound and generalized Poisson
distributions, Ann.Math. Stat. , Vol. 19, (1948), pp. 414416.
http://www.ingentaconnect.com/content/external-references?article=0025-5831(1913)74L.341[aid=6407460]
-
8/6/2019 Ivic Divisor Problem
42/121
T. Gerstenkorn / Central European Journal of Mathematics 2(4)
2004 527537 535
[3] D.S. Dubey: A compound Weibull distribution, Naval Res.
Logist. Quart , Vol.15(2), (1968), pp. 179182.
[4] D.S. Dubey: Compound gamma, beta and F distributions,
Metrika , Vol. 16(1),(1970), pp. 2731.
[5] W. Dyczka: Zastosowanie sk ladania rozk ladow do wyznaczania
moment ow(Application of compounding of distribution to
determination of moments; inPolish), Zeszyty Nauk. Politech. L
odzkiej Scient. Bull. L odz Techn. Univ. , Vol.168, (1973);
Matematyka , Vol. 3, (1973), pp. 205230.
[6] W. Dyczka: A generalized negative binomial distribution as a
distribution of PSDclass, Zeszyty Nauk. Politech. L odzkiej Scient.
Bull. L odz Techn. Univ. , Vol. 272,(1978); Matematyka , Vol. 10,
(1978), pp. 514.
[7] W. Feller: On general class of contagious distributions,
Ann. Math. Stat. , Vol. 14,(1943), pp. 389400.
[8] T. Gerstenkorn and T. Srodka: Kombinatoryka i Rachunek
Prawdopodobie nstwa (Combinatorics and Probability Theory; in
Polish) , 7th Ed., PWN, Warsaw, 1983.
[9] T. Gerstenkorn: The compounding of the binomial and
generalized betadistributions, In: W. Grossmann, G.Ch. Pug and W.
Wertz (Eds.): Proc. 2 ndPannonian Conf. on Mathem. Statistics
(Tatzmannsdorf, Austria, 1981) , D. ReidelPub., Dordrecht, Holland,
1982, pp. 8799.
[10]T. Gerstenkorn: A compound of the generalized gamma
distribution with theexponential one, Bull. Soc. Sci. Letters Lodz
, Vol. 43(1), (1993); Recherches sur les deformations , Vol. 16(1),
(1993), pp. 510.
[11]T. Gerstenkorn: A compound of the P olya distribution with
the beta one, Random Oper. and Stoch. Equ. , Vol. 4(2), (1996), pp.
103110.
[12]I. Ginzel: Die konforme Abbildung durch die Gammafunktion,
Acta Mathematica ,Vol. 56, (1931), pp. 273353.
[13]M. Greenwood and G.U. Yule: An inquiry into the nature of
frequency distributionrepresentative of multiple happenings with
particular reference to the occurrence of multiple attacks of
disease or of repeated accidents, J. Roy. Stat. Soc. , Vol.
83,(1920), pp. 255279.
[14]J. Gurland: Some interrelations among compound and
generalized distributions,Biometrika , Vol. 44, (1957), pp.
265268.
[15]M.L. Huang and K.Y. Fung: D compound Poisson distribution,
Statistical Papers-Statistische Hefte , Vol. 34, (1993), pp.
319338.
[16]G.C. Jain and P.C. Consul: A generalized negative binomial
distribution, SIAM J.Appl. Math , Vol. 21(4), (1971), pp.
507513.
[17]H. Jakuszenkow: Nowe z lozenia rozkladow (New compounds of
distributions; inPolish), Przegl ad Statystyczny , Vol. 20(1),
(1973), pp. 6773.
[18]N.L. Johnson, S. Kotz and A. Kemp: Univariate Discrete
Distributions , 2nd Ed., J.Wiley, New York, 1992.
[19]S.K. Katti and J. Gurland: The Poisson-Pacal distribution,
Biometrics , Vol. 17,(1967), pp. 527538.
http://www.ingentaconnect.com/content/external-references?article=0026-1335(1970)16:1L.27[aid=6407472]http://www.ingentaconnect.com/content/external-references?article=0026-1335(1970)16:1L.27[aid=6407472]
-
8/6/2019 Ivic Divisor Problem
43/121
536 T. Gerstenkorn / Central European Journal of Mathematics
2(4) 2004 527537
[20]A. Kaufmann: Introduction `a la Combinatorique en Vue des
Applications , Dunod,Paris, 1968.
[21]O. Lundberg: On Random Processes and Their Application to
Sickness and Accident Statistics , Almquist and Wiksells
BoktryckeriA. B., Uppsala, 1940.
[22]J. Lukaszewicz and M. Warmus: Metody Numeryczne i Graczne
(Numerical and Graphical Methods; in Polish) , PWN, Warszawa,
1956.
[23]W. Molenaar: Some remarks on mixtures of distributions,
Bull. Intern. Statist.Institute; 35 th Session of the Intern.
Statist. Inst., Beograd , Vol. 2, (1965), pp. 764765.
[24]W. Molenaar and W.R. van Zwet: On mixtures of distributions,
Ann. Math. Stat. ,Vol. 37(1), (1966), pp. 281283.
[25]L. Oginski: Zastosowanie pewnego rozk ladu typu Bessela w
teorii niezawodnosci(Application of a distribution of the Bessel
type in the reliability theory; in Polish),Zeszyty Nauk. Politech.
L odzkiej Scient. Bull. L odz Techn. Univ. , Vol. 324,
(1979);Matematyka , Vol. 12, (1979), pp. 3142.
[26]G.P. Patil, S.W. Joshi and C.R. Rao: A Dictionary and
Bibliography of DiscreteDistributions , Oliver and Boyd, Edinburgh,
1968.
[27]E.S. Pearson: Bayess theorem, examined in the light of
experimental sampling,Biometrika , Vol. 17, (1925), pp. 388442.
[28]F.E. Satterthwaite: Generalized Poisson distribution, Ann.
Math. Stat. , Vol. 134,(1942), pp. 410417.
[29]J.G. Seweryn: Some probabilistic properties of Bessel
distributions, Zeszyty Nauk.Politechn. L odzkiej Scient. Bull. L
odz Techn. Univ. , Vol. 466, (1986); Matematyka ,Vol. 19, (1986),
pp. 6987.
[30]J.G. Skellam: A probability distribution derived from the
binomial distribution byregarding the probability of success as
variable between the sets of trials, J. Roy.Statist. Soc., Ser. B ,
Vol. 10(2), (1948), pp. 257261.
[31]E.W. Stacy: A generalization of the gamma distribution, Ann.
Math. Stat. , Vol.33(3), (1962), pp. 11871192.
[32]T. Srodka: Nouvelles compositions de certaines
distributions, Bull. Soc. Sci. Lettres, L odz , Vol. 15(5), (1964),
pp. 111.
[33]T. Srodka: Zlozenie rozk ladu Laplacea z pewnym uogolnionym
rozkladem gamma,Maxwella i Weibulla (Compounding of the Laplace
distribution with a generalizedgamma, Maxwell and Weibull
distributions; in Polish), Zeszyty Nauk Politech.
L odzk.- Scient. Bull., L odz Techn. Univ. , Vol. 77, (1966); W
l okiennictwo (Textileindustry) , Vol. 14, (1966), pp. 2128.
[34]T. Srodka: On some generalized Bessel-type probability
distribution, Zeszyty Nauk.
Politechn. L odzk. Scient. Bull. L odz. Techn. Univ. , Vol. 179,
(1973); Matematyka ,Vol. 4, (1973), pp. 531.
[35]H. Teicher: On the mixtures of distributions, Ann. Math.
Stat. , Vol. 31, (1960), pp.5572.
[36]H. Teicher: Identiability of mixtures, Ann. Math. Stat. ,
Vol. 32, (1962), pp. 244248..
-
8/6/2019 Ivic Divisor Problem
44/121
T. Gerstenkorn / Central European Journal of Mathematics 2(4)
2004 527537 537
[37]G.E. Willmot: A remark on the Poisson-Pascal and some other
contagiousdistributions, Statist.Probab. Lett. , Vol. 7(3), (1989),
pp. 217220.
[38]G. Wimmer and G. Altmann: Thesaurus of Discrete Probability
Distributions ,Stamm, Essen, 1999.
-
8/6/2019 Ivic Divisor Problem
45/121
CEJM 2(4) 2004 538560
Zero-dimensional subschemes of ruled varieties
Edoardo Ballico 1, Cristiano Bocci 2, Claudio Fontanari 1
1 Department of Mathematics,University of Trento,
38050 Povo (TN), Italy 2 Department of Mathematics,University of
Milano,
I-20133 Milano, Italy
Received 22 September 2003; accepted 23 August 2004
Abstract: Here we study zero-dimensional subschemes of ruled
varieties, mainly Hirzebruchsurfaces and rational normal scrolls,
by applying the Horace method and the Terracini method.c Central
European Science Journals. All rights reserved.
Keywords: Ruled varieties, Hirzebruch surfaces, fat points,
Horace lemma, Grassmann defectivevarietiesMSC (2000): 14N05
1 Introduction
Let Z Pr
be a closed subscheme. For any integer t, let Z,t,r be the
restriction mapH 0(P r , OP r (t)) H 0(Z, OZ (t)). We say that Z
has maximal rank if for every integert > 0 the linear map Z,t,r
has maximal rank, i.e. it is either injective or surjective. Weare
going to investigate this property for a zero-dimensional Z .A very
classical motivation for such an interest is provided by the
following
Denition 1.1. Let X P r be an integral nondegenerate projective
variety of dimensionn . The h-secant variety Sech (X ) of X is the
Zariski closure of the set { pP r : p lies in
This research is part of the T.A.S.C.A. project of I.N.d.A.M.,
supported by P.A.T. (Trento) andM.I.U.R. (Italy). E-mail:
[email protected] E-mail: [email protected] E-mail:
[email protected]
-
8/6/2019 Ivic Divisor Problem
46/121
E. Ballico et al. / Central European Journal of Mathematics 2(4)
2004 538560 539
the span of h + 1 independent points of X }. The expected
dimension of Sech (X ) is
expdim Sech (X ) = min {(n + 1)( h + 1) 1, r }
and X is h-defective with h-defect h (X ) if
h (X ) = expdim Sec h (X ) dim Sech (X ) > 0.
A basic tool for understanding defective varieties is the
classical Lemma of Terracini(see [33] for the original version and
[18] or [1] for a modern proof), which says that, forP 1, . . . , P
h +1 general points and P general in their span, one has
T P (Sech (X )) < T P 1 (X ), . . . , T P h +1 (X ) >
with equality holding in characteristic zero. Therefore, at
least over the eld C , X is not h-defective if and only if the
restriction map
Z, 1,r : H 0(P r , OP r (1)) H 0(Z, OZ (1))
has maximal rank, whereZ := h +1i=1 2P i X.
More generally, for Z := h +1i=1 m i P i X with m i 2, the
restriction map Z, 1,r fails tohave maximal rank if and only if the
union of the ( m i 1)-osculating space at P i fori = 1 , . . . , n
spans a linear space of dimension less than the expected one. In
other words,the linear system of hyperplane sections having a
contact of order m i at P i has dimensiongreater than the expected
one, i.e. it is a special linear system.
Here we focus on ruled varieties, by applying as an essential
tool the so-called Horacemethod. We recall its set-up in Section 2,
while in Section 3 we collect the main results of the present
paper. In particular, Horace method provides a quick proof of the
followingwell-known
Proposition 1.2. Fix integers x > 0, a, b 3. Let Z F 0 be a
general union of x doublepoints of F 0. Then the restriction map
Z,a,b : H 0(F 0, OF 0 (ah + bF )) H 0(Z, OZ (ah +bF )) has maximal
rank.
We stress that the above result is sharp: indeed, it is easy to
verify that homogeneousspecial systems arise when we consider the
value a = 2. This fact is true not onlyfor F 0, but for F e with e
0. For example, consider the system L := |2h + (2 d +2e)F 2d+
e+1i=1 2P i | on F e , e 0; we have h0(F e , OF e (2h + (2 d + 2
e)F ) = 6 d + 3 e + 3,hence the virtual dimension of L is 1 then we
would expect that L is empty. Sinceh0(F e , OF e (h +( d+ e)F ) = 2
d+ e+2, there is a unique curve C in the system |h +( d+ e)F
|passing through P i , i = 1 , . . . , 2d + e + 1: its double 2 C
is a divisor on F e of type2h + (2 d + 2 e)F with 2d + e + 1 double
points, hence 2 C L and L is special.
We also remark that, at least in characteristic zero,
Proposition 1.2 may be provedby using [5], Lemma 4.2, or [5],
Proposition 4.1, [16], Lemma 3, [32], Lemma 2.2, and
-
8/6/2019 Ivic Divisor Problem
47/121
540 E. Ballico et al. / Central European Journal of Mathematics
2(4) 2004 538560
certain reducible curves with negative intersection with K F e .
From this point of viewthe cases e = 0 , 1 are quite easier,
because in this case K F e is ample. However, Horacemethod allows
us to handle also fat points of higher multiplicity. For instance,
we obtainthe following
Theorem 1.3. Fix integers b 4, a 11, s > 0 and m i , 1 i s,
such that 1 m i 3for every i. Fix s general points P i F 0, 1 i s,
and set Z := m i P i . Then therestriction map Z,a,b : H 0(F 0, OF
0 (ah + bF )) H 0(Z, OZ (ah + bF )) has maximal rank.
We do not have a unied way to handle all cases with 3 a 10 left
open byTheorem 1.3, hence we omit the few cases with low a which
can be easily checked by hand.Indeed, the paper [26] by Antonio
Laface contains the complete list of all homogeneousspecial systems
with multiplicity two or three. We point out that such a list gives
someinformation also in the non-homogeneous case, just by applying
the Horace method (seeProposition 3.6).
For F e , e > 0, the proof of Theorem 1.3 gives a similar,
but weaker, statement(Theorem 3.11); roughly speaking, we have to
replace b with b ea in our assumptions.
Our work suggests several open questions, which are collected in
Section 4. In par-ticular, Question 4.4 should be viewed as the
analogue for Hirzebruch surfaces of thecelebrated Segre and
Harbourne-Hirschowitz conjectures for the projective plane.
Next, in Section 5 we consider higher dimensional scrolls and we
apply the samemethods to the study of a zero-dimensional subscheme
which is curvilinear, i.e. its Zariskitangent space at a reduced
point has dimension at most one.
Finally, in Section 6 we turn to another geometrically
meaningful kind of defectivity,the so-called Grassmann defectivity,
which is dened as follows.
Denition 1.4. Let X P r be an integral nondegenerate projective
variety of dimensionn dened over the eld C . The (k, h)-Grassmann
secant variety Seck,h (X ) of X is theZariski closure of the set {l
G (k, r ) : l lies in the span of h + 1 independent points of
X }. The expected dimension of Seck,h (X ) isexpdim Seck,h (X )
= min {(h + 1) n + ( k + 1)( h k), (k + 1)( r k)}
and X is (k, h)-defective with (k, h)-defect k,h (X ) if
k,h (X ) = expdim Sec k,h (X ) dim Seck,h (X ) > 0.
This property was classically investigated by Alessandro
Terracini in his beautifulpaper [34], going back to 1915. Recent
advances in the subject are due to Ciro Ciliberto,
Luca Chiantini, and Marc Coppens ([10], [11], [12], and [17]);
see also the papers [19],[21], and [20], all three strongly
inspired by the pioneering work of Terracini.Here we give a new
application of the same method in order to answer the following
Question 1.5. (Ciliberto, [15]) Can a smooth projective variety
be both defective andGrassmann defective?
-
8/6/2019 Ivic Divisor Problem
48/121
E. Ballico et al. / Central European Journal of Mathematics 2(4)
2004 538560 541
After describing a natural generalization of Terracinis
approach, we derive explicitnumerical conditions in order to be
Grassmann defective for a projective variety. A closerinspection to
the case of rational normal scrolls makes it easy to produce innite
seriesof smooth varieties (even surfaces!) being both defective and
Grassmann-defective, for
instance:
Example 1.6. For every h 4 the smooth rational scroll
X (1, 2h + 2) P 2h +4
is both h-defective and (1 , h)-defective.
We work over an algebraically closed eld K of characteristic
zero. Indeed, our proof
of Proposition 1.2 is characteristic free, but in the proof of
Theorem 1.3 and in all relatedproofs we need char(K ) = 0 because
we quote [13]. We point out that in the case p := char( K ) > 0
it would be sufficient to assume p > max{a, 4}, but we stress
that wedo not know any characteristic free proof of Theorem
1.3.
2 Preliminaries
Let Y be any algebraic variety, Z Y a closed subscheme and D Y
an effective Cartier
divisor of Y . For any P Y let 2P or 2{P, Y } be the closed
subscheme of Y with I 2P,Y as
its ideal sheaf. Thus, if P Y reg , then length(2 {P, Y }) =
1+dim( Y ). The same notationwill be used when P is an arbitrary
integral closed subvariety of Y . The residual schemeResD (Z ) of Z
with respect to D is the closed subscheme of Y with Hom ( I D,Y , I
Z,Y )as the ideal sheaf. Thus Res D (Z ) Z . If P D reg (and hence
P Y reg ) we have2{P, Y } D = 2 {P, D } and Res D (2{P, Y }) = {P }
(with its reduced structure).
For any L Pic( Y ) we have an exact sequence on Y :
0 I ResD
(Z ) ,Y L( D) I Z,Y L I Z D,D (L |D ) 0 (1)
From (1) we obtain the following very elementary form of Horace
Lemma.
Lemma 2.1. For any L Pic( Y ) we have h0(Y, I Z,Y L) h0(Y, I
ResD (Z ) ,Y L( D))+h0(D, I Z D,D (L |D )) and h1(Y, I Z,Y L) h1(Y,
I ResD (Z ) ,Y L( D))+ h
1(D, I Z D,D (L |D )).
The next remark will be called the double residue trick.
Remark 2.2. Fix a projective variety W , a closed subscheme T of
W , a line bundleL Pic( W ), an effective Cartier divisor D on W ,
a point P D reg with P /T red . LetZ W be a zero-dimensional scheme
such that Z red = {P }. Set Z 1 := Z . For anyinteger i 2 dene
inductively the scheme Z i by the formula Z i = Res D (Z i 1).
HenceZ i Z i 1 with strict inequality if Z i 1 = . Set a i :=
length( Z i D). The non-increasing
-
8/6/2019 Ivic Divisor Problem
49/121
542 E. Ballico et al. / Central European Journal of Mathematics
2(4) 2004 538560
sequence (a1, . . . ) will be called the associated sequence of
Z with respect to D. We havelength( Z ) = i 1 a i . Fix an integer
i 1. By [4], Lemma 2.3 (see in particular Fig. 1 atp. 308) there is
a scheme U Z whose associated sequence is obtained from the
associatedsequence (a1, . . . ) of Z omitting the term a i and a
subscheme E D with E red = {P }
and length( E ) = a i such that to check h1(W, I Z T L) = 0
(resp. h
0(W, I Z T L) = 0) it
is sufficient to prove h1(D, I (T D )E,D (L |D )) = h1(W, I ResD
(T )U L( D)) = 0 (resp.h0(D, I (T D )E,D (L |D )) = h0(W, I ResD (T
)U L( D)) = 0).If a1 = 3, a2 = 2, a3 = 1, a j = 0 for j > 3 and
i = 3, then we will call the (1 , 3, 2)-trick this application of
[4], Lemma 2.3.
Z = -(1,3,2)-trick
E =
U =
If a1 = 3, a2 = 2, a1 = 1, a j = 0 for j > 3 and i = 2, then
we will call the (2 , 3, 1)-trick this application of [4], Lemma
2.3.
Z = -(2,3,1)-trick
E =
U =
If a1 = 2, a1 = 1, a j = 0 for j > 2 and i = 2, then we will
call the (1 , 2)-trick thisapplication of [4], Lemma 2.3.
Z = -(1,2)-trickE =U =
Now assume that we want to prove h i (W, I B T L) = 0 ( i = 0 or
i = 1) for a general zero-dimensional scheme B with B red = {P 1}
and with associated sequence (3 , 1, 0, . . . ),where P 1 is a
general point of a sufficiently general irreducible Cartier divisor
D1; inour set-up, D1 will be a general ber of the ruling. Set M :=
Res D 1 (B). To prove thesought-for vanishing it is sufficient to
prove h i (D1, I (T D 1 ){P },D 1(L |D 1 )) = h i (W, I T M L( D1))
= 0. This application of [4], Lemma 2.3, will be called the (1 ,
3)-trick . Andso on for other associated sequences.
Example 2.3. Fix P D such that D is smooth at P , hence Y is
smooth at P . LetZ Y be a zero-dimensional scheme such that Z red =
{P }. Assume Z curvilinear, i.e.assume that the Zariski tangent
space of Z at P has dimension at most one. There isa germ C at P of
a smooth curve containing Z . Let be the intersection
multiplicityof C with D at P . Set a := length( Z ). Hence Z is the
divisor aP of C and henceOZ = OC / OC ( aP ). Set b := [a/ ] 0 and
c = a b. We have a i (Z ) = for
1 i b, ab+1 (Z ) = c and a i (Z ) = 0 for all i b + 2.
Remark 2.4. Let Y be an irreducible variety, L Pic( Y ) and V H
0(Y, L) a nite-dimensional vector space. Assume V = {0}. For a
general P Y we have V H 0(Y, I {P }L) = V and hence dim( V H 0(Y, I
{P } L)) = dim( V ) 1. In several proofs, we are
-
8/6/2019 Ivic Divisor Problem
50/121
E. Ballico et al. / Central European Journal of Mathematics 2(4)
2004 538560 543
going to use this obvious remark without any further mention. By
this observation, inall the statements of this paper we could
assume m i 2 for every i without any lossof generality.
3 Ruled surfaces
Let F e , e 0, be the Hirzebruch surface with invariant e, i.e.
such that e is the minimalself-intersection of a section of the
ruling of F e . Hence F 0 = P 1 P 1 is isomorphic to asmooth
quadric surface and F 1 is isomorphic to the blowing-up of P 2 at
one point. Wex a ruling f : F e P 1; f is unique if and only if e
> 0. We have Pic( F e) = Z2 and wetake, as a basis of Pic( F e),
a section h of f with h2 = e and a class, F , of the ruling f .Thus
h F = 1 and F 2 = 0. We will use both the additive and the
multiplicative notationfor line bundles and divisors on F e . We
have F e = 2h (2 + e)F , and it is easy tocheck h0(F e , OF e (ah +
bF )) = 0 if a < 0. Moreover, by the projection formula we havef
(OF e (ah + bF )) =
ai=0 OP 1 (b ie) for every a 0. Thus h0(F e , OF e (ah + bF )) =
0 if
a 0 and b < 0, h0(F e , OF e (ah + bF )) =t 1i=0 (b ie +1) if
0 (t 1)e b < te for some
integer t with 0 < t a, h0(F e , OF e (ah + bF )) =ai=0 (b
ie+1) = (2 b+2 ae)(a +1) / 2 if
a 0 and b ae 1 and h1(F e , OF e (ah + bF )) = 0 if a 0 and b ae
1. Now we will seehow to translate any statement concerning the
postulation of a zero-dimensional schemeZ F e in a statement
concerning an interpolation problem for suitable polynomials
(seeRemark 3.1 for the case e = 0 and Remark 3.2 for the case e
> 0).
Remark 3.1. Fix integers 0 and 0. Since F 0 = P 1 P 1, the
vector spaceH 0(F 0, OF e (h + F )) may be parametrized by the set
of all bihomogeneous polynomialsin the variables x0, x1, y0, y1
whose monomials have degree with respect to x0, x1 anddegree with
respect to y0, y1.
Remark 3.2. Fix integers e > 0, a 0 and b. Take variables z0,
z1 and w and assign the
weight one to z0 and z1 and the weight e to w. Since f (OF e (ah
+ bF )) =ai=0 OP 1 (b ie),the vector space H 0(F e , OF e (ah + bF
)) corresponds to the vector space of all weighted
homogeneous polynomials in z0, z1 and w with b as total weight
degree and with degree a as a polynomial in w.
Remark 3.3. Fix integers e 0, a 0, b ea and t 0. Fix A R F e
with R |F |,and A a zero-dimensional scheme such that length( A) =
t. Since h1(F e , OF e (ah + ( b 1)F )) = 0, the restriction map :
H 0(F e , OF e (ah + bF )) H 0(R, OR (a)) is surjective.We have
h0(F e , I A (ah + bF )) = h0(F e , OF e (ah + bF )) min{t, a +1 },
hence h1(F e , I A (ah +bF )) = 0 if and only if t a + 1.
Lemma 3.4. Let Z F 0 be the general union of 5 (resp. 6) double
points. Thenthe restriction map H 0(F 0, OF 0 (3h + 3 F )) H 0(Z,
OZ (3h + 3 F )) is surjective (resp.injective).
-
8/6/2019 Ivic Divisor Problem
51/121
544 E. Ballico et al. / Central European Journal of Mathematics
2(4) 2004 538560
Proof 3.5. We will do only the surjectivity part, the
injectivity part being similar andeasier. Let T F 0 be a general
element of |h + 2 F |. Hence T = P 1. Notice thatT (3h+3 F ) = 9
and h1(F 0, OF 0 (2h+ F )) = 0. Hence the restriction map H 0(F 0,
OF 0 (3h+3F )) H 0(T, OT (3h + 3 F )) is surjective. Let Z F 0 and
union of 5 double points with
Z red T . Thus length( Z T ) = 10. By Horace Lemma we have
hi
(F 0, I Z (3h + 3 F )) =h i (F 0, I Z red (2h + F )), i = 0 , 1.
Since there is a unique C |2h + F | containing Z red , weare
done.
Proof of Proposition 1.2. If 3x (a +1)( b+1) 1 we need to prove
the surjectivity of Z,a,b . If 3x > (a +1)( b+ 1) 1 we need to
prove the injectivity of Z,a,b . We will only dothe case 3x (a +1)(
b+ 1) 1, the case 3x > (a +1)( b+ 1) 1 requiring only
notationalmodications. Hence we assume 3 x (a + 1)( b+ 1) 1. Just
taking a suitable union of more double points, it is sufficient to
prove the case ( a+1)( b+1) 3 3x (a+1)( b+1) 1. We will assume that
these inequalities are satised. If a = b = 3, then x = 5 and
theresult follows from Lemma 3.4. Hence we may assume a + b 7 and,
say, b a. Suppose,for the moment, b 5. Thus, by induction on the
integer a + b, we may assume that theresult is true for the pair (
a, b 1) and the pair ( a, b 2), b 5. Fix D |F |. First assumea odd.
Let W F 0 be a general union of x (a +1) / 2 double points of F 0.
Let Z F 0 bea general union of (a+1) / 2 double points of F 0 with
(Z )red D. Thus Res D (Z ) = ( Z )redand length( Z D) = a + 1 =
h0(D, OD (ah + bF )). By Horace Lemma 2.1 it is sufficientto show
h1(F 0, I W (Z ) red (ah + ( b 1)F )) = 0. By the inductive
assumption and theinequality 3( x (a + 1) / 2) (a + 1) b 1 we have
h1(F 0, I W (ah + ( b 1)F )) = 0. Noticethat ( a + 1) / 2 + 3(x (a
+ 1) / 2) = 3 x (a + 1)( b + 1) + ( a + 1) b. Hence to showthat
h1(F 0, I W (Z ) red (ah + ( b 1)F )) = 0 and the generality of ( Z
)red it is sufficient toshow that h0(F 0, I W D (ah + ( b 1)F ))
h0(F 0, I W (ah + ( b 1)F )) (a + 1) / 2. Sinceh0(F 0, I W D (ah +(
b 1)F )) = h0(F 0, I W (ah +( b 2)F )), it is sufficient to use the
inductiveassumption. Indeed, in all cases with 3 x (a + 1)( b + 1)
3, the inductive assumptionimplies h0(F 0, I W (ah +( b 2)F )) = 0.
Now assume a even. Let A F 0 be a general unionof x a/ 2 1 double
points of F 0. Take a general union B of a/ 2 double points of F 0
withsupport on D and a general P D. Apply the double residue trick
(Remark 2.2) withrespect to P . In order to conclude, it is
sufficient to prove h1(F 0, I AB red 2{P,D }(ah + ( b1)F )) = 0.
Indeed, we have h1(F 0, I 2{P,D }(ah +( b 1)F )) = 0 because a >
0 and b 1 > 0,and h1(F 0, I B red 2{P,D }(ah + ( b 1)F )) = 0
because h1(F 0, I 2{P,D }(ah + ( b 1)F )) = 0,h0(F 0, I 2{P,D }(ah
+ ( b 1)F )) card( B red ) and B red is general in D. Hence we see
thath1(F 0, I AB red 2{P,D }(ah + ( b 1)F )) = 0 because h1(F 0, I
B red 2{P,D }(ah + ( b 1)F )) = 0,h0(F 0, I B red 2{P,D }(ah + ( b
1)F )) = ( a + 1) b card( B red ) 3 card( A) and A is generalin F
0.
Consider now the case b = 4. We can restrict our attention to 3
a b (in fact, if a > b = 4 then we can apply the previous
argument replacing a by b and b by a). Thusthe surjectivity of the
map Z,a,b is given by h1(F 0, I Z (ah + 4 F )) = 0 where Z = xi=1
2P iwith 5a + 2 3x 5a + 4 and a 4. By the same argument of Lemma
3.4 or by adirect computation (for example with a computer algebra
system) we see that the systems
-
8/6/2019 Ivic Divisor Problem
52/121
E. Ballico et al. / Central European Journal of Mathematics 2(4)
2004 538560 545
I 6i =1 2P i (3h + 4 F ) and I 8i =1 2P i (4h + 4 F ) are
non-special and the claim follows.
Proof of Theorem 1.3. Let Z = s 3i=1 3P i s 3 + s 2i= s 3 +1 2P
i be a zero-dimensional scheme of
fat points, where s3 and s2 are respectively the number of
triple points and double points.
By Laface we can easily treat the cases s2 = 0 or s3 = 0. In
fact, if s2 = 0, then Z is theunion of triple points and I Z (ah +
bF ) is never special if a, b 4. When s3 = 0, Z is theunion of
double points. In this situation I Z (ah + bF ) is never special if
a, b 3.
Notice that we will never be able use induction on the
parameters appearing in thestatement of Theorem 1.3. We will just
try to control the postulation of a certain zero-dimensional scheme
in nitely many steps; if at one step, we cannot go on, then
wefailed. If all the steps can be done and is exhausted (resp. the
line bundle OF 0 (ah + bF )is exhausted), then we obtain h1(F 0, I
(ah + bF )) = 0 (resp. h0(F 0, I (ah + bF )) = 0).
By semicontinuity we will obtain Theorem 1.3 for certain data s,
m i , 1 i s, relatedto . Even more, instead of is some of the steps
we will use a virtual scheme and applyRemark 2.2.
Step 1: First of all we study the case b = 4 and a arbitrary. We
consider the case withs3 > 0 and s2 > 0 and we may assume 5(
a + 1) 2 6s3 + 3 s2 5(a + 1) + 2 (allother cases are reduced to
this one). We assume s3 2. We x a ber A and twodifferent points P,
Q A. We use 3P and the (2 , 3, 1)-trick with respect to Q.
Sincelength( Z A) = 3 + 2 we have h i ( I Z A (4F )) = 0, for i = 0
, 1. Thus h i ( I Z (ah + 4 F )) =
hi( I ResA (Z )((a 1)h + 4 F )). The residual scheme Z := Res A
(Z ) has s3 1 triple
points, s2 double points, another double points 2 P and a
virtual scheme Q with layers(3, 1). Moreover, we observe that P and
Q are general in A. Since on A we haveagain a virtual scheme of
length 2 + 3 = 5, then h i ( I Z A (4F )) = 0, for i = 0 , 1 andh i
( I Z (ah + 4 F )) = h i ( I Z ((a 2)h + 4 F )). Z := Res A Z
consists of s3 2 triple points,s2 double points and 2 simple
points.
If s3 is an even number we apply the previous argument s3 times
and we obtainh i ( I Z (ah + 4 F )) = h i ( I Z ( s 3 ) ((a s3)h +
4 F )), where Z (s 3 ) = N M with N union of s2
double points and M union of s3 simple points. One big inductive
step consists in thereduction of an assertion on |(a x)h+4 F |,
where x is an even integer with 0 x s3 2,to an assertion on |(a x
2)h + 4 F | by making two smaller inductive steps, rst from|(a
x)h+4 F | to |(a x 1)h+4 F | and then from |(a x 1)h+4 F | to |(a x
2)h+4 F |,each time using A |h| . In this part of the proof, for
every big inductive step we take adifferent Ai general in |h| , 1 i
s3/ 2. Since the points in N are general, from [26] itfollows that
h1(F 0, I N ((a s3)h +4 F )) = 0; we also have h0(F 0, I N ((a s3)h
+4 F )) s2.This inequality would be sufficient to obtain h1(F 0, I
N M ((a s3)h+4 F )) = 0 if the pointsin M were general. Unluckily
this is not the case, but the only restriction on M is theexistence
of s3/ 2 general |h |, 1 i s3/ 2, such that card( Ai M ) = 2 for
all i; inparticular, the points in M i := M Ai are general in Ai .
Dene M i := i j =1 M j and setM 0 := . We claim that if h1(F 0, I N
M i ((a s3)h +4 F )) = 0 for some i {0, . . . , s 3/ 2},then also
h1(F 0, I N M i +1 ((a s3)h + 4 F )) = 0. Indeed, the claim follows
from theinequality h0(F 0, I N M i ((a s3)h + 4 F )( Ai+1 )) 2 and
the fact that M i+1 is general
-
8/6/2019 Ivic Divisor Problem
53/121
546 E. Ballico et al. / Central European Journal of Mathematics
2(4) 2004 538560
in Ai+1 . From this claim we inductively obtain h1(F 0, I Z ( s
3 ) ((a s3)h + 4 F )) = 0, asdesired.
If s3 is an odd number we apply the previous argument s3 1
times. Then we applyan Horace method using the last triple point
and a double point. Again, the length on
A is 3 + 2 = 5 and we obtain hi
( I Z (ah + 4F )) = hi
( I Z ( s 3 ) ((a s3)h + 4F )), where Z (s
3)
consists of s2 double points and s3 simple points.Finally, when
s3 = 1 we just apply an Horace method using the triple point and
a
double point obtaining again h i ( I Z (ah + 4 F )) = h i ( I Z
(1) ((a 1)h + 4 F )), where Z (1)
consists of s2 double points and 1 simple points.If a s3 then Z
has maximal rank for bidegree ( a, 4). Moreover, by the
assumption
6s3 + 3 s2 5(a + 1) + 2, we see that a < s 3 only if a
10.
Step 2: Fix an effective divisor R |F | and P 1 R. We assume the
existence of
an integer k such that 1 k s and integers j i , 1 i k, 1 j i s,
j 1 = 1, j i = j r if i = r , such that ki=1 m j i = a + 1. Choose
P j i R, with P j i = P j r if i = r . Choose general P j F 0 for j
= { j1, . . . , j k }. Set Z 1 := ki=1 m j i P j i and Z 2 = j = {
j 1 ,...,j k }m j P j . By semicontinuity it is sufficient to show
h1(F 0, I Z 1Z 2 (ah + bF )) = 0.Set Z := ki=1 (m j i 1)P j i Z 2.
Thus Z = Res R (Z 1 Z 2). By Horace Lemma we have
h i (F 0, I Z 1Z 2 (ah + bF )) = hi (F 0, I Z (ah + ( b 1)F )) i
= 0 , 1
and hence we may continue taking again R as Cartier divisor and
trying to insert someof the points P j , j = { j1, . . . , j k },
into R. Notice that the set ( Z 2)red is general in F 0, butthat
the points P j 1 , . . . , P j k are not general in F 0.
Step 3: Now assume that there are no such integers 1 k s, j i ,
1 i k, 1 j i s, j1 = 1, j i = j r if i = j , such that
ki=1 m j i = a + 1. We take the integers k and j i so that
ki=1 m j i a +1 and
ki=1 m j i is maximal under this restriction. We claim that we
may
apply the double residue trick (Remark 2.2) with respect to many
P j with j = { j1, . . . , j k }so that the total sum of the new
virtual intersection with R is again a +1. If k = s, thenwe cannot
do that, but in this subcase we immediately get h1(F 0, I Z (ah +
bF )) = 0 byHorace Lemma and the assumption m i 3 b + 1.
If k < s by the maximality of ki=1 m j i and the assumption m
i 3 for all i we have
a 1 k
i=1
m j i a.
If outside R we still have a fat point with multiplicity three,
then we may alwaysincrease by one (case ki=1 m j i = a) or by two
(case
ki=1 m j i = a 1) as we want the
length of the restriction to R of a subscheme by using Remark
2.2 and respectively the(1, 3, 2)-trick and the (2 , 3,
1)-trick.
If outside R we have a double point, then we use the (1 ,
2)-trick to increase by onethe length on R and we specialize it to
a double point 2 Q, Q general in R, to increase bytwo the length on
R. In each step we apply the double residue trick only at one
pointbecause m i 3. Furthermore, at each step we need to consider
at most one scheme
-
8/6/2019 Ivic Divisor Problem
54/121
E. Ballico et al. / Central European Journal of Mathematics 2(4)
2004 538560 547
having support on R and coming from a double residue trick of
some previous steps,because m i 3 for every i (and, indeed, if
there is any such scheme, then it comes fromthe previous step).
More in detail we obtain the following situations. Let k3 and k2
be respectively the
number of triple points and double points such that k = k2 + k3.
Case ki=1 m j i = a 1
i) If (k3 = s3) or (k3 s3 and k2 < s 2) then we specialize a
double point Q toR. The residual scheme consists of s3 k3 triple
points, s2 + k3 k2 1 doublepoints, k2 + 1 simple points.
ii) If (k3 < s 3 and s2 = k2) then we use the (2 , 3,
1)-trick on a triple point Q. Theresidual scheme consists of s3 k3
1 triple points, s2 + k3 k2 double points,k2 simple points and a
virtual scheme with layers (3 , 1).
Casek
i=1 m j i = aiii) If (k3 = s3) or (k3 s3 and k2 < s 2) then
we use the (1 , 2)-trick on a doublepoint Q. The residual scheme
consists of s3 k3 triple points, s2 + k3 k2 1double points, k2
simple points and a virtual scheme with layers (2).
iv) If (k3 < s 3 and s2 = k2) then we use the (1 , 3,
2)-trick on a triple point Q. Theresidual scheme consists of s3 k3
1 triple points, s2 + k3 k2 double points,k2 simple points and a
virtual scheme with layers (3 , 2).
Step 4: We use the set-up of Remark 2.2. If we apply the (1 ,
3)-trick or the (1 , 3, 2)-
trick (resp., (1 , 2)-trick, (2 , 3)-trick, or (1 , 2)-trick) at
the next step (i.e. for the integerb := b 1) we obtain a connected
virtual scheme whose intersection with R has lengthlarger by two
(resp., by one) than the one of the original scheme. If P R and m
> 0,ResR (mP ) = ( m 1)P (with the convention 0 P = ), hence
length( R ResR (mP )) =m 1 = length( R mP ) 1. Since in our set-up
we may always take m 3 anda + 1 4 3, we are always sure that at the
next step we have a virtual scheme whoseintersection with R has
length at most a + 1. It follows that we can always apply
thegeneral machinery of Step 3 also with the integer b < b.
Step 5: After b 4 steps we arrive at b = 4. We are going to
mimic the proof of the case b = 4 given in Step 1 with suitable
modications. In the present case wehave a scheme Z = Z 1 Z 2 Z 3
with Z 1 union of general fat points with multiplicity1, 2 or 3; Z
2 union of general fat points with multiplicity 1 or 2, supported
by pointsof R; Z 3 either the empty set or the disjoint union of at
most 2 schemes each of whichbelong to the following list: (2 , 0, .
. .); (3, 1, 0, . . .); (3, 0, . . .); (3, 2, 0, . . .), where we
havespecied the length of the intersection with R obtained via
residual schemes. Furthermore,we have length( Z 2 Z 3) 2a 1 and
length( R (Z 2 Z 3)) a; notice also thatlength( A (Z 2 Z 3)) 2 for
any A |h| and length( A (Z 2 Z 3)) = 2 if and only if atthe point A
R the scheme Z 2Z 3 is either a double point or a point of type (3
, 1, 0, . . .)with respect to R.
If length( A (Z 2Z 3)) = 2, then we insert a triple point on A.
Let U be the union of all Ai |h | such that length( Ai (Z 2Z 3)) =
2. Since outside U (Z 2Z 3)) the schemeZ 2Z 3 (not just its
reduction) is contained in R, it follows that E := Res U (Z 2Z 3)
R.
-
8/6/2019 Ivic Divisor Problem
55/121
548 E. Ballico et al. / Central European Journal of Mathematics
2(4) 2004 538560
We claim that if h1(F 0, I S (eh + 4 F )) = 0 for a suitable
scheme S Z 1 Z 2 Z 3 anda suitable e a, then h1(F 0, I S E (eh + 4
F )) = 0. Indeed, by [13] applied to the linearsystem induced by H
0(F 0, I S (eh + 4 F )) on R, for general E red we have h0(F 0, I S
E (eh +4F )) = max {0, h0(F 0, I S (eh + 4 F )) h0(F 0, I S (eh + 4
F )( R))}. Hence in order to
check h1(F 0, I S E (eh + 4 F )) = 0 it is sufficient to prove
that h
0(F 0, I S (eh + 4 F )( R))
h0(F 0, I S (eh + 4 F )) length( E ). Since OF 0 (eh + 4 F )( R)
= OF 0 ((e 1)h + 4 F ), weget the above inequality by a standard
inductive argument (we refer to the proof of Proposition 1.2 and to
Step 1 for more details in a similar situation).
If length( A (Z 2 Z 3)) = 1, then the connected component C of Z
2 Z 3 withC red = A R is contained in R. Set := length( C ) and
notice that 3. If = 1, thenwe insert a triple point on A and
simultaneously apply the (1 , 3, 2)-trick. After this smallstep,
there is a length 5 scheme on A (a length 2 connected component
comes from the
triple point, a length 3 connected component comes from the (1 ,
3, 2)-trick). By applyingHorace Lemma 2.1 we reduce to a case with
a length 3 scheme on A (one reduced pointcoming from the triple
point and one length 2 connected component coming from the(1, 3,
2)-trick). Then either we apply a (2 , 3, 1)-trick or (if no triple
point is left) we inserton A a double point. Assume now = 2. In
this case, we insert a triple point on Aand simultaneously apply
the (1 , 2, 3)-trick. After this small step, there is a length
5scheme on A, hence by applying Horace Lemma 2.1 we can reduce to a
case with on Aa connected scheme of length 1 (coming from the
triple point) and a connected schemeof length 3 (coming from the (1
, 2, 3)-trick). Then we apply either the (1 , 3, 2)-trick or(if no
triple point is left) the (1 , 2)-trick. Finally, in the case = 3,
we insert a triplepoint on A and simultaneously apply the (1 , 2,
3)-trick. After this small step, there is alength 5 scheme on A
(one reduced component coming from C , one length 2
connectedcomponent coming from the triple point and another one of
length 2 coming from the(1, 2, 3)-trick). By Horace Lemma 2.1 we
reduce to a case with a length 5 scheme on A(one reduced point
coming from C , another reduced point coming from the triple
pointand one length 3 connected component coming from the (1 , 2,
3)-trick). After applyingHorace Lemma 2.1 once again, the proof is
over.
Next we classify all exceptional cases on F 0 with m i 3 for all
indices i, m i = 2 forat least one index i and m i = 3 for at least
one index i (see [26] for the other exceptionalcases with m i
3).
Proposition 3.6. Fix integers b 2, u > 0, c > 0. take u +
c general points P i F 0,1 i u + c, and set Z := ui=1 3P i
u + c j = u +1 2P i . We have h0(F 0, I Z (2h + bF ))
h1(F 0, I Z (2h + bF )) = 0 if and only if b = 2 u + c 1 and u +
c 1 0 (mod 2).
Proof 3.7. Set W := u + ci=1 2P i . Apply Horace Lemma u times
with respect to the divisorsR i |OF 0 (0, 1)| such that P i R i , 1
i u. Since W = Res R 1R u (Z ), we obtain bothh0(F 0, I Z (2h+ bF
)) = h0(F 0, I W (2h+( b u)F )) and h1(F 0, I Z (2h+ bF )) = h1(F
0, I W (2h+(b u)F )). Thus we restrict to analyze the homogeneous
system I W (2h + ( b u)F ). By[26], this system is special if it
has the form I T (2h + (2 d)F ) with T = 2d+1i=1 2P i . This
can
-
8/6/2019 Ivic Divisor Problem
56/121
E. Ballico et al. / Central European Journal of Mathematics 2(4)
2004 538560 549
happen only if we require 2d = b u and 2d + 1 = u + c or
equivalently, b = 2 u + c 1and u + c 1 0 (mod 2).
The proof of Theorem 1.3 may be applied to many other cases. We
give, for example,
the following result.
Theorem 3.8. Fix integers x 0, c j > 0, 1 j x, b 4 + x j =1 c
j a 11 +x j =1 c j ,
s > 0 and m i , 1 i s, such that 1 m i 3 for every i. Fix s +
x general pointsP i , Q j F 0, 1 i s, 1 j x and set Z := si=1 m i P
ix j =1 c j Q j . Then the restrictionmap Z,a,b : H 0(F 0, OF 0 (ah
+ bF )) H 0(Z, OZ (ah + bF )) has maximal rank.
Proof 3.9. The case x = 0 is just Theorem 1.3. Fix R |F | and
take as Q j x generalpoints of R. Hence ResR (x j =1 c j Q j )
=
x j =1 (c j 1)Q j (with the convention 0 P j = ) and
length( R (x j =1 c j Q j )) = c1 + + cx . Then we repeat the
proof of Theorem 1.3 takingthis ber R to apply Horace Lemma (see
the details of Step 4). After at most x j =1 c jsteps we have
exhausted all fat points with multiplicity 4, hence we are again in
theset-up of Theorem 1.3. We never use the double residue trick
with respect to any pointQ j , but only with respect to the other s
points. Hence in our new situation the samebounds apply.
Our method can also be applied in several situations in which
the points are not
general. In this way one obtains results for complete linear
systems on a (non-general)blown - up surface. Here we give only a
very easy example. The interested readermay construct in the same
way thousands of other examples. Following the proof of Theorem 1.3
it is easy to construct examples with maximal rank.
Example 3.10. Fix integers e 2, b > e and y such that 1 y b.
Fix y distinct pointsP i F e \ h such that f (P i ) = f (P j ) if i
= j and set Z := yi=1 2P i , T :=
yi=1 f 1(f (P i )).
We have h0(F e , OF e (h+ zF )) = z+1 if 0 z e 1, while h0(F e ,
OF e (h+ zF )) = 2 z e+2
if z e. Now assume y > b e and 2y b. Since ResT (Z ) = y
i=1 {P i }, every m H 0(F e , I Z (h+ bf )) vanishes on T , i.e.
h0(F e , I Z (h+ bF )) = h0(F e , I si =1 {P i }(h+( b y)F ))
>h0(F e , OF e (h + bF )) 3y.
The proofs of Theorem 1.3 and Theorem 3.8 give the following
result.
Theorem 3.11. Fix integers e > 0, s > 0, 1 i s, m i , a 11
and b such thatb 4 + ae > 0, 1 m i 3 for every i and si=1 m i (m
i + 1) / 2 (b ae 1)(a + 1). Fixgeneral P i F e and set Z := si=1 m
i P i . Then h1(F e , I Z (ah + bF )) = 0.
Indeed, just use |F | as a ruling and take the curves A, Ai in
|h + eF | instead of |h|;notice however that the prescribed bound
is certainly not sharp for e > 0.
We also obtain the following result.
-
8/6/2019 Ivic Divisor Problem
57/121
550 E. Ballico et al. / Central European Journal of Mathematics
2(4) 2004 538560
Proposition 3.12. Fix integers e > 0, s > 0, a > 0, b
ae, m i , 1 i s. Set := max 1 i s {m i } and assume
si=1 m i (m i + 1) / 2 (b ea 1)(a + 1 ). Fix general
P i F e , 1 i s, and set Z := si=1 m i P i . Then h1(F e , I Z
(ah + bF )) = 0.
Proof 3.13. Fix an effective divisor R |F |. We take the
integers k and j i , 1 i k,so that ki=1 m j i a + 1 andki=1 m j i
is maximal with this restriction. We have
a + 1 ki=1 m j i a + 1 by the denition of . Choose Q j i R, with
Q j i = Q j rif i = r . Choose general O j F e for j = { j1, . . .
, j k }. Set Z 1 := ki=1 m j i Q j i and Z 2 = j = { j 1 ,...,j k
}m j O j . By semicontinuity it is sufficient to show h1(F 0, I Z
1Z 2 (ah + bF )) = 0.Set Z := ki=1 (m j i 1)Q j i Z 2. Thus Z = Res
R (Z 1 Z 2). By Horace Lemma and theinequality ki=1 m j i a + 1 it
is sufficient to prove h1(F e , I Z (ah + ( b 1)F )) = 0. Inorder
to do so, we may continue taking again R as Cartier divisor and
specializing ontoR the support of some of the fat points m j O j ,
j = { j
1, . . . , j k }. Since s
i=1m i (m i +
1)/ 2 (b ea 1)(a + 1 ), after at most b ea 1 steps we have
inserted all zero-dimensional schemes. Hence with another step the
claim is reduced to the vanishing of h1(F e , OF e (ah + zF )) for
some z b ea 1 and the proof is over.
The same proof gives the following result.
Proposition 3.14. Fix integers s > 0, b a > 0, m i > 0,
1 i s. If s = 1, thenset := 0 and assume m1 a + 1. If s 2, then set
:= max 2 i s {m i} and assume
m1 a + 1. Assume si=1 m i (m i + 1) / 2 (b + 1)( a + 1 ). Fix
general P i F 0,1 i s, and set Z := si=1 m i P i . Then h1(F 0, I Z
(ah + bF )) = 0.
Finally we turn to more general ruled varieties. Let C be a
smooth projective curveof genus g 0 and E a rank n vector bundle on
C . Set X := P (E ) and call f : X C the natural projection. If L
Pic( X ), then there is a unique line bundle R on C suchthat L = OX
(t) f (R). Assume h1(C, E R) = 0 and set (L) the maximal integerb 0
such that h1(C, E R( bP )) = 0 (i.e. h0(C, E R( bP )) = h0(C, E R)
bn)
for a general P C .The proof of Proposition 3.12 gives verbatim
the following results.
Theorem 3.15. Fix a ruled surface X = P (E ) and L Pic( X ) with
L = OX (a)f (R).Assume a 11 and h1(C, E R) = 0. Fix integers s >
0 and m i , 1 i s, such that1 m i 3. Choose general P i X , 1 i s,
and set Z = si=1 m i P i . Assume
si=1 m i (m i + 1) / 2 (a + 1) (L). Then h1(X, I Z L) = 0.
Proposition 3.16. Fix a ruled surface X = P (E ) and L Pic( X )
with L = OX (a) f (R). Assume a > 0 and h1(C, E R) = 0. Fix
integers s > 0 and m i > 0, 1 i s,and general P i X , 1 i s.
Set Z := si=1 m i P i . If s = 1, set = 0 and assumem1 a + 1. If s
2, set := max 2 i s {m i }. Assume m1 a + 1 and
si=1 m i (m i + 1) / 2 (a + 1 ) (L). Then h1(X, I Z L) = 0.
-
8/6/2019 Ivic Divisor Problem
58/121
E. Ballico et al. / Central European Journal of Mathematics 2(4)
2004 538560 551
Remark 3.17. Let X P r be an integral variety and P 1, . . . , P
h general points of X .By the characteristic free part of
Terracinis lemma ([1], part (1) of Corollary 1.11, or[18]) if Z,
1,r has maximal rank, then X is not h-defective, the converse being
in generaltrue only in characteristic zero. Hence Horace method can
be used to prove the non
h-defectivity of certain projective varieties even in positive
characteristic.
4 Open questions
In this section we raise two questions concerning the
postulation of general unions of fatpoints on a scroll over a
curve.Let C be a smooth projective curve of genus g 0 and E a rank
n vector bundle on C .Set X := P (E ) and call f : X C the natural
projection. Let OX (1) the tautologicalquotient line bundle on X ,
i.e. the only line bundle on X such that f (OX (1)) = E .The
restriction of OX (1) to any ber F = P n 1 of f has degree one. We
have Pic( X ) =f (Pic( C )) Z OX (1). For any R Pic( C ) and any t
0 we have f (OX (t)f (R)) =S t (E ) R. For all integers i, x with 1
i n 2 we have R i f (OX (x) f (R)) = 0.For every integer b n + 1 we
have Rn 1f (OX (b) f (R)) = 0.
Remark 4.1. Fix positive integers n, m and t with n 2. If t m 1,
set (n,t ,m ) :=n + m 1
n . Hence if t m 1 then the integer (n,t,m ) is the length of a
fat point mP
of multiplicity m on any n-fold. Let X = P (E ) be an
n-dimensional scroll over a smoothcurve and L Pic( X ) such that L
|F = OF (t) for any ber F of the ruling of X . Fixany P X and let F
be the ber of the