SYSTEMS OF CONGRUENCES AND EQUATIONS OF WARING TYPE
MATHEMATICS
Submitted 1965-01-01 | SovietRxiv: ru-196501.59875 | Translated from Russian

Abstract Generated abstract

This paper studies solution counts for systems of congruences and Waring type equations associated with equal sums of powers, with emphasis on estimates for smaller numbers of variables than those covered directly by Vinogradov’s mean value theorem. It proves that a previously known estimate for a special common modulus holds for almost all admissible sets of moduli when the number of variables satisfies a logarithmic lower bound, and gives lower bounds showing that comparable estimates cannot generally be expected below this range for certain moduli. The paper also derives improved estimates for the number of solutions of equal sums of nth powers and establishes comparison inequalities between corresponding quantities for different exponents, refining earlier results.

Full Text

UDC 511.9

MATHEMATICS

A. A. KARATSUBA

SYSTEMS OF CONGRUENCES AND EQUATIONS OF WARING TYPE

(Presented by Academician I. M. Vinogradov, 8 IV 1965)

We shall use the following notation. All letters, with the exception of \(\varepsilon, \varepsilon_0,\ldots,\delta,\delta_0,\ldots\), denote integers; \(\varepsilon,\varepsilon_0,\ldots,\delta,\delta_0\) are positive real quantities; \(n \ge 10\); \(1 \le r \le n\); \(p\) is a prime number. The notation \(A \gg B\) means that \(A \ge C(n)|B|\), where \(C(n)\) is a constant depending only on \(n\); \(C, C_1,\ldots\) are absolute constants; \(J_{k,n}(\lambda_1,\ldots,\lambda_n)\) is the number of solutions of the system of equations

\[ \begin{gathered} x_1+\ldots-y_k=\lambda_1\\ x_1^n+\ldots-y_k^n=\lambda_n,\\ 1\le x_i,y_i\le P;\quad i=1,2,\ldots,k; \end{gathered} \tag{1} \]

\(I_{k,n}\) is the number of solutions of the equation

\[ x_1^n+\ldots+x_k^n=y_1^n+\ldots+y_k^n, \]

\[ 1\le x_i,y_i\le P;\quad i=1,2,\ldots,k; \]

\(N_{k,n}\) is the number of solutions of the system of congruences

\[ \begin{gathered} x_1+\ldots-y_k=0,\\ x_1^r+\ldots-y_k^r=0,\\ x_1^{r+1}+\ldots-y_k^{r+1}\equiv 0\pmod{q_{r+1}},\\ x_1^n+\ldots-y_k^n\equiv 0\pmod{q_n},\\ 1\le x_i,y_i\le P;\quad i=1,2,\ldots,k. \end{gathered} \tag{1′} \]

The mean value theorem of I. M. Vinogradov \((^1)\) asserts that for
\[ k \ge n(n+1)/4+n\tau \]
the inequality holds

\[ J_{k,n}(0,\ldots,0)=I_{k,n}\ll P^{2k-n(n+1)/2+\delta}, \tag{2} \]

where
\[ \delta=\frac{1}{2}n(n+1)/(1-1/n)^\tau . \]

It follows from this theorem that for \(k \ge n(n+1)/4+n\tau\), for \(N_{k,n}\) the estimate

\[ N_{k,n}\ll P^{2k-r(r+1)/2+\delta}(q_{r+1}\ldots q_n)^{-1} \tag{3} \]

holds.

For any \(k\) we trivially have

\[ N_{k,n}\gg P^{2k-n(n+1)/2}(q_{r+1}\ldots q_n)^{-1}. \]

Thus, estimate (3) is sharp.

The problem arises of obtaining estimates of the form (3) for smaller values of \(k\). Let \(1\le r\le n\); \(P^r\ll q\le P^r\); \(q\le q_\nu<2q\); \(\nu=r+1,\ldots,n\); \(Q=q^{\,n-r}\). In \((^2)\), for a certain special \(q_0\) and \(q_{r+1}=\ldots=q_n=q_0\), the estimate

\[ N_{k,n}\ll P^{2k-rn+r(r-1)/2+\varepsilon_1}, \tag{4} \]

if \(k \geqslant 6rn\ln n\). In the present article it is proved that, for \(k \geqslant 6rn\ln n\), the estimate (4) holds for almost all sets \((q_{r+1},\ldots,q_n)\). Moreover, a lower estimate is given for the number of sets \((q_{r+1},\ldots,q_n)\) for which, when \(k \leqslant 6rn\ln n\), the estimate (4) cannot be obtained. In the last two theorems, estimates are obtained for \(I_{k,n}\) and inequalities between the quantities \(I_{k,n}\) and \(I_{k,m}\), which refine the corresponding theorems of \((^2)\).

Theorem 1. Among the \(Q\) sets of moduli \((q_{r+1},\ldots,q_n)\) there exist \(Q(1-Q^{-\varepsilon_0})\) such that, for these moduli, when \(k \geqslant 6rn\ln n\), the estimate
\[ N_{k,n}\ll P^{2k-rn+r(r-1)/2+\varepsilon}Q^{\varepsilon_0}, \]
holds, where \(\varepsilon\) is arbitrarily small and \(\varepsilon_0\) is an arbitrary positive quantity, \(0<\varepsilon_0<1\).

Proof. Arrange all \(N_{k,n}\) in increasing order. We shall use the fact that \(N_{k,n}\), corresponding to the set of moduli \((q_{r+1},\ldots,q_n)\), is equal to
\[ \sum_{\lambda_{r+1},\ldots,\lambda_n} J_{k,n}(0,\ldots,0,q_{r+1}\lambda_{r+1},\ldots,q_k\lambda_n). \]
Summing the last \(Q^{1-\varepsilon_0}\) terms of our nondecreasing sequence and applying the estimates of Theorem 3 of article \((^2)\), we obtain the assertion of the theorem.

Theorem 2. Let the set \((q_{r+1},\ldots,q_n)\) be such that
\[ q_i=p^{\alpha_i}q_i',\quad 1\leqslant \alpha_i\leqslant i,\quad (q_i',p)=1,\quad i=r+1,\ldots,n;\quad 2p<P. \]
Then for the corresponding \(N_{k,n}\) the following lower estimate holds:
\[ N_{k,n}\gg P^{2k-rn+r(r-1)/2}p^{-2k+\alpha_{r+1}+\cdots+\alpha_n}. \]

Proof. Estimating from below the number of those solutions of system \((1')\) with the given \((q_{r+1},\ldots,q_n)\) which are multiples of \(p\), we obtain the assertion of the theorem.

Corollary. Let \(\alpha_{r+1}+\cdots+\alpha_n>Crn\ln n\) and \(p\to\infty\) as \(P\to\infty\). Then, for \(k\leqslant Crn\ln n\), one cannot obtain an estimate for \(N_{k,n}\) of the form (4).

In \((^2)\) the estimate
\[ I_{k,n}\ll P^{2k-r} \tag{5} \]
was obtained for \(k \geqslant 6rn\ln n\).

Theorem 3. The estimate (5) holds for \(k\geqslant c_1r^2\), if \(1\leqslant r\leqslant \frac13 n\), and for \(k\geqslant c_2n^2\ln n\), if \(\frac13 n<r\leqslant n\).

Proof. The second case follows trivially from the asymptotic formula in Waring’s problem obtained by I. M. Vinogradov (see \((^1)\), p. 300). Consider the case \(1\leqslant r\leqslant \frac13 n\). Let \(s=2r\) and let \(p\) be a prime number satisfying the inequalities
\[ P^{s/(s+1)}<p<2P^{s/(s+1)}. \]
If \(\overline N_{k,n}\) is the number of solutions of the congruence
\[ x_1^n+\cdots-y_k^n\equiv 0 \pmod {p^{s+1}}, \]
\[ 1\leqslant x_i,y_i\leqslant p^{1+1/s},\quad i=1,2,\ldots,k, \]
then it is clear that
\[ J_{k,n}\leqslant \overline N_{k,n}, \]

Take \(x=py+z,\ 1\le z\le p,\ y\le p^{1/s}\). After the corresponding transformations we obtain

\[ N_{k,n}\ll p^{2k/s}+p^{2k-s-1}\sum_{a=1}^{p^{s+1}} \left|\sum_{y\le p^{1/s}}\exp \frac{2\pi ia}{p^s} \left(b_1y+\ldots+b_sp^{s-1}y^s\right)\right|^{2k}\ll \]

\[ \ll p^{2k/s}+p^{2k}N'_{k,n}, \]

where \(N'_{k,n}\) is the number of solutions of the congruence

\[ b_1(y_1+\cdots-y_k)+pb_2(y_1^2+\cdots-y_k^2)+\cdots+ p^{s-1}b_s(y_1^s+\cdots-y_k^s)\equiv \]

\[ \equiv 0\pmod {p^s}, \]

\[ (b_1,p)=\cdots=(b_s,p)=1;\qquad 0\le y_i;\quad y_i\le p^{1/s},\quad i=1,2,\ldots,k. \]

The last congruence is equivalent to the system of equations (1) for \(\lambda_1=\ldots=\lambda_n=0\). Applying estimate (2) and choosing \(\tau\) in a suitable way, we obtain the assertion of the theorem.

Theorem 4. If \(k=c_3m^2\ln m,\ 2\le m\le n\), then the inequality

\[ I_{k,n}\ll I_{k,m} \]

holds.

The proof follows from Theorem 4 (see (2), Theorem 2).

Received
5 IV 1965

CITED LITERATURE

¹ I. M. Vinogradov, Selected Works, Publishing House of the Academy of Sciences of the USSR, 1952. ² A. A. Karatsuba, DAN, 165, No. 1 (1965).

Submission history

SYSTEMS OF CONGRUENCES AND EQUATIONS OF WARING TYPE