• Nie Znaleziono Wyników

Kum hev (Columbia, S.C.) and T

N/A
N/A
Protected

Academic year: 2021

Share "Kum hev (Columbia, S.C.) and T"

Copied!
10
0
0

Pełen tekst

(1)

On an equation with prime numbers

by

A. Kum hev (Columbia, S.C.) and T. Nedeva (Plovdiv)

1. Introdu tion. B. I. Segal ([13℄, [14℄) was the rst to onsider in

1933 additiveproblemswith non-integerdegrees. He studiedtheinequality

(1) jx

1 +x

2

+:::+x

k

Nj<"

and theequation

(2) [x

1

℄+[x

2

℄+:::+[x

k

℄=N;

where > 1 is not an integer, and proved in both ases that there exists

k

0

( ) su h that the orresponding problem has solutions if k  k

0 and

N is suÆ iently large. Later Deshouillers [4℄ and Arkhipov and Zhitkov

[1℄ improved Segal's result on (2). One may also mention the papers of

Deshouillers [5℄ and Gritsenko [7℄, where the equation (2) intwo variables

was onsidered.

In 1952 I. I. Piatetski-Shapiro [12℄ onsidered (1) with x

1

;:::;x

k re-

stri ted to prime numbers. Let H( ) denote the least k su h that the

inequality (1) with xed " > 0 has solutions in prime numbers for every

suÆ ientlylargereal N. Piatetski-Shapiroproved that

limsup

!1

H( )

log

4:

He also proved that H( )5 for1 < <3=2. The theorem of Goldba h{

Vinogradov [16℄ motivates the onje ture that for lose to 1, H( )  3.

This was proved byD. I. Tolev [15℄. He showed that if1 < <15=14 and

"=N

(1= )(15=14 )

log 9

N thenthe quantity

D(N):=

X

jp

1 +p

2 +p

3 Nj<"

logp

1 logp

2 logp

3

1991 Mathemati sSubje tClassi ation: 11P05,11P32.

Resear hofthe rst-namedauthorsupportedbyBulgarianMinistryofEdu ationand

S ien e,grantMM-430.

(2)

is positive for a suÆ ientlylarge N. Re ently Y. C. Cai[3℄ improved the

upperboundfor to 13/12.

In [10℄ Laporta and Tolev onsideredthe orresponding equation of the

type(2). For1< <17=16 theyprovedanasymptoti formulaforthesum

R (N):=

X

[p

1

℄+[p

2

℄+[p

3

℄=N logp

1 logp

2 logp

3 :

In thepresent paperwe improvethe rangeof they obtained.

Theorem 1.Assume that 1< <12=11 and Æ >0 is arbitrary small.

Then for any suÆ iently large integerN we have the asymptoti formula

R (N)= 3

(1+1= )

(3= ) N

3= 1

+O(N 3= 1

exp( (logN) 1=3 Æ

)):

We also improve theresult from [3℄. We obtain an asymptoti formula

forthesumD(N). Sin etheproofis similarto theproofof Theorem1,we

omitit.

Theorem 2.Assume that 1< <11=10 and Æ >0 is arbitrary small.

Thenfor any suÆ ientlylarge real N and "N

(1= )(11=10 )+

for some

 >0 we have the asymptoti formula

D(N)=2"

3

(1+1= )

(3= ) N

3= 1

+O("N 3= 1

exp( (logN) 1=3 Æ

)):

Therangeof inbothproblemsdependsontheestimateofanexponen-

tialsum over primes. In [10℄ and [15℄ Vaughan's identityand theexponent

pair(1=2;1=2)areused. WederiveTheorem1fromamorepre iseestimate

of this sum (Lemma 5 below). To prove it we use the identity of Heath-

Brown[8℄,vanderCorput'smethodasdes ribedinChapters2and3 of[6℄

and theestimateof a doubleexponentialsum dueto Kolesnik[9℄.

2. Notation. Sin e for 1 < < 17=16 Theorem 1 is proved in [10℄,

we an assume that 17=16  < 12=11. In this paper  > 0 is a xed

small number depending only on ; P = N 1=

; ! = P 1 

; p, p

1

;::: are

primes; 2 (0;1); " is an arbitrary smallpositive number, not ne essarily

thesame indi erent appearan es. We use[x℄,fxgand kxkfortheintegral

partofx,fra tionalpartofx andthedistan efrom xtothenearestinteger

respe tively. (n) isvon Mangoldt'sfun tion. Moreover,

 e(x)=exp(2ix);

 f(x)g(x)meansthat f(x)=O(g(x));

 f(x)g(x) meansthatf(x)g(x)f(x);

 xX meansthatx runsthrougha subintervalof[X;2X℄;

 f(x ;:::;x ) g(x ;:::;x ) meansthat

(3)



j1+:::+jn

x j

1

1 :::x

j

n

n f(x

1

;:::;x

n )=



j1+:::+jn

x j

1

1 :::x

j

n

n g(x

1

;:::;x

n

)(1+O())

forall n-tuples(j

1

;:::;j

n

) forwhi hit makessense.

We usesumsof two types,whi hwede ne inthefollowingway:

 type Isums:

XX

mM;nL

mnX a

m

F(mn);

 type II sums:

XX

mM;nL

mnX a

m b

n

F(mn);

wherethe oeÆ ientssatisfythe onditionsa

m

m

"

,b

n

n

"

.

We de ne

 =exp((logN) 1=3 Æ

):

We alsoset

S( )= X

pP

logpe( [p

℄);

R

i

=

\

i S

3

( )e( N)d (i=1;2)

where

1

=( !;!) and

2

=(!;1 !).

3. Some preliminary results

Lemma 1.Let D bea subdomain of the re tangle f(x;y)jXx2X,

Y  y  2Yg (X  Y) su h that any line parallel to any oordinate axis

interse ts it in O(1) line segments. Let , be real numbers, 6= 0,

+ 6= 1, + 6= 2, and let f(x;y) be a real suÆ iently many times

di erentiable fun tion su h that f(x;y) 

 Ax

y

throughout D. Setting

N =XY,F =AX

Y

, we have

X

(x;y)2D

e(f(x;y))

(NF)

"

(F 1=3

N 1=2

+NY 1=2

+N 5=6

+NF 1=4

+NF 1=8

X 1=8

+ 2=5

F 1=5

N 9=10

X 2=5

+ 1=4

NX 1=4

):

Proof. This isa version of Theorem1 of [9℄. The proof maybe found

in[11℄.

Lemma 2. Let 3 < U < V < Z < X and suppose that Z 1=2 2 N,

2 2 3

(4)

valued fun tionsu h that jF(n)j1. Thenthe sum

X

nX

(n)F(n)

maybede omposed intoO(log 10

X) sums, ea h eitheroftype Iwith L>Z,

or of type II with U <L<V.

Proof. This isLemma 3of [8℄.

Lemma 3.Let x notbe an integer, 2(0;1), H3. Then

e( fxg)= X

jhjH

h

( )e(hx)+O



min



1;

1

Hkxk



where

h ( )=

1 e( )

2i(h+ ) :

Proof. See Lemma12 of[2℄.

In the following lemma we estimate the number N() of quadruples

(h

1

;h

2

;n

1

;n

2

) forwhi h h

1

;h

2

H,n

1

;n

2

N and

j(h

1 + )n

1 (h

2 + )n

2 j:

Lemma 4. Supposethat 6=0, 2(0;1),>0,H3and N islarge.

Then

N()HN 2

+H 3=2

N log (2HN):

Proof. We follow the approa h of D. R. Heath-Brown [8℄. We de ne

thequantity

N(;a;b)=#f(h

1

;h

2

;n

1

;n

2 )jh

1

;h

2

H; (h

1

;h

2

)=a; n

1

;n

2

N;

(n

1

;n

2

)=b; j(h

1 + )n

1 (h

2 + )n

2 jg

whi h we are going to estimate. If h

1

;h

2

 H, n

1

;n

2

 N and j(h

1 +

)n

1 (h

2 + )n

2

jwe have



n

1

n

2



h

2 +

h

1 +





HN

;

h

2

h

1 h

2 +

h

1 +

 1

H

;

hen e

(3)

h

2

h

1



n

1

n

2



 1

H +



HN

:

We also have

(4)

n

1

n



h

2 +

h +



1=





HN

:

(5)

From (3)and (4), arguingason pp. 256{257 of[8℄, we obtain

N(;a;b)



HN

 H

2

N 2

a 2

b 2

+min



H 2

a 2

; N

2

b 2

+ HN

2

a 2

b 2



:

Sin e

N() X

a2H X

b2N

N(;a;b) ;

theproofof thelemmais omplete.

4. The main lemma

Lemma 5.Suppose that X >P 9=10

,H=X 1

and

h

( ) are omplex

numbers su h that j

h

( )j (1+jhj) 1

. Then, uniformly with respe t to

2(!;1 !), we have

T( )= X

jhjH

h ( )

X

nX

(n)e((h+ )n

)X 2 %

for some suÆ ientlysmall %>0, depending only on .

Proof. We use Lemma 2 with F(n) = e((h+ )n

) to redu e the

estimationof T( )to theestimationof thesums

T

i ( )=

X

jhjH

h ( )

X

i

(i=1;2)

where P

1 ,

P

2

are type I and type II sums, respe tively. We hoose the

parameters U,V,Z asfollows:

U =X

2 2+2%

=256; V =4X 1=3

and

Z = 8

<

: [X

(16 16)=3+3%

℄+1=2 if17=16 <14=13,

[X

(13 13)=3+3%

℄+1=2 if14=13 <13=12,

[X

(20 21)=2+5%

℄+1=2 if13=12 <12=11 .

Let us onsiderT

2

( ). We have

(5) T

2

( ) max

!2 jT

(1)

2

()j+(logX) max

2JH jT

(2)

2

( ;J)j

where

T (1)

2

()= X

mM X

nL a

m b

n

e((mn)

);

T (2)

2

( ;J)= X

h ( )

X X

a

m b

n

e((h+ )(mn)

) :

(6)

Firstwe estimate T (2)

2

( ;J). We obtain

T (2)

2

( ;J) X

"

J X

mM X

qQ

X

(h;n)2I

q

d(h;n)e((h+ )(mn)

)

wherejd(h;n)j1,Q>1 isaparametertobede nedlaterandforq Q,

I

q

=f(h;n)jhJ; nL; 5(q 1)JL

<Q(h+ )n

5qJL

g:

So,usingtheCau hyinequality,we get

jT (2)

2

( ;J)j 2

 X

"

MQ

J 2

XX

h

1

;h

2

J

n

1

;n

2

L

jj5JL

=Q

X

mM e(m

)

where=(h

1 + )n

1 (h

2 + )n

2

. Weestimatetheinnermostsumtrivially

ifjjM

,and usingtheexponent pair (13=40;11=20) otherwise. From

Lemma4 we nowobtain

jT (2)

2

( ;J)j 2

 X

"

MQ

J 2

(MN(M

)

+ max

M

5JL

=Q (

13=40

M

(9+13 )=40

+ 1

M 1

)N())

X

"

(J 1=2

M 2

LQ+J 13=40

M

(49+13 )=40

L

(80+13 )=40

Q 13=40

+J 1

M 2

L 2

Q+J 7=40

M

(49+13 )=40

L

(40+13 )=40

Q 27=40

):

We hooseQviaLemma2.4of[6℄and the onditionsonJ,M and Limply

(6) max

2JH jT

(2)

2

( ;J)jX

2 %+"

:

Let us nowestimate T (1)

2

(). Using theCau hy inequalityand Lemma

2.5of [6℄ we get

jT (1)

2 ()j

2

X

"



M 2

L 2

Q +

ML

Q X

qQ X

nL

X

mM

e(((n+q)

n

)m

)



where Q  L is a positive integer. We apply the exponent pair (13=40;

11=20) to theinnermostsum and hoose Qvia Lemma2.4 of[6℄ to obtain

jT (1)

2 ()j

2

X

"

(M 2

L+ 13=40

M

(49+13 )=40

L

(67+13 )=40

+

13=53

M

(75+13 )=53

L

(93+13 )=53

)

and usingthe onditionson M,L and we get

(7) max

!2 jT

(1)

2

()jX

2 %+"

:

The needed estimateforT ( ) follows from(5){(7).

(7)

Let usnow onsiderT

1

( ). We have

(8) T

1

( )X

"

max

jj2(!;H+1) X

mM

X

nL

e((mn)

)

:

If LX

(57 49)=23+3%

we estimate thesum over n using theexponent

pair (8=41;26=41) to obtain

(9) jT

1

( )jX

2 %+"

:

Otherwise we rst use the Cau hy inequality and Lemma 2.5 of [6℄ to

thesum on theright-handsideof (8)and obtain

jT

1 j

2

X

"



M 2

L 2

Q +

ML

Q X

qJ X

nL X

mM

e(f(m;n;q))



wheref(m;n;q)=((n+q)

n

)m

,J Q=2 andQLisaparameter

tobe hosenlater. ThenweapplythePoissonsummationformula(Lemma

3.6of[6℄)tothesumsovermand nsu essivelyandAbel'stransformation:

X

q X

m;n

e(f(m;n;q))

= X

q;n X





 2

f(m



;n;q)

m 2



1=2

e(1=8+f(m



;n;q) m

 )

+O(MLJF 1=2

+LJlogX)

MF 1=2

X

q;

X

n e(f

1

(;q;n))

+XJF 1=2

+LJlogX

MF 1=2

X

q;

X





 2

f

1 (;q;n

 )

n 2



1=2

e(1=8+f

1 (;q;n

 ) n

 )

+MF 1=2

JFM 1

(LF 1=2

+logX)+XJF 1=2

+LJlogX

MLF 1

X

q;;

e(g(; ;q))

+F

1=2

JlogX+LJlogX+XJF 1=2

whereF =JM

L 1

,f

1

(;q;n)=f(m



;n;q) m

 ,

g(; ;q)=f

1 (;q;n

 ) n







0 (q)

1=(2 2 )

 1=2



=(2 2)

F;

0

isa onstant dependingonlyon ,=J=L, FL 1

,FM 1

.

Hen e

X

"

jT

1 j

2

X 2

Q 1

+X 2

F 1

Q 1

X

qJ

X

FM 1

X

FL 1

e(g(; ;q))

(10)

2 1=2 1=2

(8)

If X 1=2

 L < X

(57 49)=23+3%

we estimate the sum over ; in (10)

usingLemma 1 with X =FM 1

,Y =FL 1

and f(x;y) =g(; ;q). We

get

X

"

jT

1 j

2

X 2

Q 1

+F 1=3

X 3=2

+XF 1=2

L 1=2

+X 7=6

F 2=3

+X 3=2

F 3=5

J 2=5

L 4=5

+XF 3=4

M 1=8

+J 1=4

X 5=4

F 3=4

L 1=2

+X 2

F 1=2

+XL:

Now we substitute the expressionfor F in thelast estimate and hoose Q

via Lemma2.4 of[6℄. We obtain(9).

IfZ L<X 1=2

weinter hangetherolesofand andprove(9)again.

This ompletes theproofof thelemma.

5. Proof of Theorem 1. Itis easy to seethat

R (N)= 1

\

0 S

3

( )e( N)d =R

1 +R

2 :

TheintegralR

1

isstudiedbyLaportaandTolev[10℄,pp. 928{929. They

proved that if1< <17=16 then

R

1

= 3

(1+1= )

(3= ) N

3= 1

+O(

1

N 3= 1

)

but the same argument shows that this asymptoti formula holds for 1 <

<3=2. Hen e thetheorem follows fromtheestimate

(11) R

2

 1

P 3

:

It is notdiÆ ultto prove that

R

2

PlogP max

2

2

jS( )j:

To prove (11) itremainsto show that

max

2

2

jS( )j 1

P 2

:

We have

S( )= X

nP

(n)e( n

)e( fn

g)+O(P 1=2

):

So,it issuÆ ient to prove thatforX satisfyingP 9=10

<X P,

S

1 ( )=

X

(n)e( n

)e( fn

g) 1

X 2

:

(9)

UsingLemma3 withx=n

and H = X 1

weobtain

S

1 ( )=

X

jhjH

h ( )

X

nX

(n)e((h+ )n

)

+O



logX X

nX min



1;

1

Hkn

k



:

The estimationof theerrorterm intheabove equalityis standard(see [8℄,

pp. 245{246). Hen e (11) followsfrom Lemma5.

The proof of Theorem1is omplete.

A knowledgements. We would like to thank D. I. Tolev for intro-

du ing us into this problem and the regular attention to our work and to

M. B. L.LaportafortellingusaboutCai'spaper[3℄.

Referen es

[1℄ G. I. Arkhipov and A. N. Zhitkov, On the Waring problem with non-integer

degrees,Izv.Akad.NaukSSSR48(1984),1138{1150(inRussian).

[2℄ K.Buriev,Additiveproblemswithprimenumbers,thesis,Mos owUniversity,1989

(inRussian).

[3℄ Y.C.Cai,Onadiophantineinequalityinvolvingprimenumbers,A taMath.Sini a

39(1996),733{742(inChinese).

[4℄ J.M. Deshouillers, Probleme de Waringave exposants nonentiers,Bull.So .

Math.Fran e101(1973),285{295.

[5℄ |,Un problemebinaire entheorieadditive,A taArith.25(1974),393{403.

[6℄ S. W. Graham and G.A. Kolesnik, Van der Corput's Method of Exponential

Sums,LondonMath.So .Le tureNoteSer.126,CambridgeUniv.Press,1991.

[7℄ S. A. Gritsenko, Three additive problems, Izv. Ross. Akad. Nauk 56 (1992),

1198{1216(inRussian).

[8℄ D. R. Heath-Brown, The Pjate ki-



Sapiro prime number theorem, J. Number

Theory16(1983),242{266.

[9℄ G.A.Kolesnik,Onthenumberofabeliangroupsofagivenorder,J.ReineAngew.

Math.329(1981),164{175.

[10℄ M.LaportaandD.I.Tolev,Onanequationwithprimenumbers,Mat.Zametki

57(1995),926{929(inRussian).

[11℄ H.-Q.Liu,On square-fullnumbersinshortintervals,A taMath. Sini a(N.S.)65

(1993),148{164.

[12℄ I.I.Piatetski-Shapiro,OnavariantoftheWaring{Goldba hproblem,Mat.Sb.

30(1952),105{120(inRussian).

[13℄ B.I.Segal,OnatheoremsimilartotheWaringtheorem,Dokl.Akad.NaukSSSR

1(1933),47{49(inRussian).

[14℄ |, The Waring theorem with fra tionaland irrational degrees, Trudy Mat. Inst.

Steklov.5(1933),73{86(inRussian).

[15℄ D.I.Tolev,Onadiophantineinequalityinvolvingprimenumbers,A taArith.61

(10)

[16℄ I.M.Vinogradov,Representation ofanoddnumber asthesum ofthree primes,

Dokl.Akad.NaukSSSR15(1937),291{294(inRussian).

DepartmentofMathemati s Hrabretz22, apt.11

UniversityofSouthCarolina Plovdiv4019,Bulgaria

Columbia,SouthCarolina 29208

U.S.A.

E-mail:koumt hemath.s .edu

Re eivedon17.7.1996 (3024)

Cytaty

Powiązane dokumenty

Wska˙z przyk

Wska˙z przyk

The examples in Section 4 are expository in nature, and it is not suggested that all the results are new; Section 5 contains all the results we have managed to prove for values of

Research supported in part by the Hungarian Academy of Sciences, by Grants 16975, 19479 and 23992 from the Hungarian National Foundation for Scientific

In the absence of noise, the numerical results are satisfactory, taking into account that they have been obtained with a small number of nodes.. As one could predict, the presence of

The southern margin of the Magura Ocean is des- ignated as the northern boundary of the Pieniny Klip- pen Belt (PKB), which separates the Central Western Carpathians (CWC,

A random sample of 167 people who own mobile phones was used to collect data on the amount of time they spent per day using their phones.. The results are displayed in the

Ecalle used for some nonlinear ordinary differential equations we construct first a formal power series solution and then we prove the convergence of the series in the same class as