13. Lecture 13  PDF

Ternary Goldbach problem

The counting function for the number of representations of an odd integer \(N\) as the sum of three primes is \[r(N)=\sum_{p_{1}+p_{2}+p_{3}=N} 1.\]

The following is Vinogradov’s asymptotic formula for \(r(N)\).

There exists an arithmetic function \(\mathfrak{S}(N)\) and \(c_{1}, c_{2}\in\mathbb R_+\) such that \[c_{1}<\mathfrak{S}(N)<c_{2}\] for all sufficiently large odd integers \(N\), and \[r(N)=\mathfrak{G}(N) \frac{N^{2}}{2(\log N)^{3}}\left(1+O\left(\frac{\log \log N}{\log N}\right)\right).\]

The arithmetic function \(\mathfrak{S}(N)\) is called the singular series for the ternary Goldbach problem.

The singular series

  • We begin by studying the arithmetic function

    \[\mathfrak{S}(N)=\sum_{q=1}^{\infty} \frac{\mu(q) c_{q}(N)}{\varphi(q)^{3}},\] where \[c_{q}(N)=\sum_{\substack{a=1 \\(q, a)=1}}^{q} e(a N / q)\] is Ramanujan’s sum.

  • The function \(\mathfrak{S}(N)\) is called the singular series for the ternary Goldbach problem.

  • The Ramanujan sum \(c_q(n)\) is a multiplicative function of \(q\); that is, if \((q, q') = 1\), then \[c_{qq'}(n) = c_q(n) \cdot c_{q'}(n).\]

  • The Ramanujan sum can be expressed in the form

    \[c_q(n) = \sum_{d \mid (q, n)} \mu\left(\frac{q}{d}\right) d,\]

  • In particular, if \((q, n) = 1\), then \(c_q(n) =\mu(q)\).

The singular series \(\mathfrak{S}(N)\) converges absolutely and uniformly in \(N\) and has the Euler product representation \[\mathfrak{S}(N)=\prod_{p}\left(1+\frac{1}{(p-1)^{3}}\right) \prod_{p \mid N}\left(1-\frac{1}{p^{2}-3 p+3}\right).\]

There exist positive constants \(c_{1}\) and \(c_{2}\) such that \[c_{1}<\mathfrak{S}(N)<c_{2},\] for all positive integers \(N\). Moreover, for any \(\varepsilon>0\),

\[\mathfrak{S}(N, Q)=\sum_{q \leq Q} \frac{\mu(q) c_{q}(N)}{\varphi(q)^{3}}=\mathfrak{S}(N)+O\left(Q^{-(1-\varepsilon)}\right),\] where the implied constant depends only on \(\varepsilon\in(0, 1)\).

Decomposition into major and minor arcs

  • We decompose the unit interval \([0,1]\) into two disjoint sets: the major arcs \(\mathfrak{M}\) and the minor arcs \(\mathfrak{m}\).

  • Let \(B>0\) and set \[Q=(\log N)^{B}\]

  • For \(1 \leq q \leq Q\) and \(0 \leq a \leq q\) satisfying \((a, q)=1\), the major arc \(\mathfrak{M}(q, a)\) is the interval consisting of all real numbers \(\alpha \in[0,1]\) so that \[\left|\alpha-\frac{a}{q}\right| \leq \frac{Q}{N}.\]

  • If \(\alpha \in \mathfrak{M}(q, a) \cap \mathfrak{M}\left(q^{\prime}, a^{\prime}\right)\) and \(a / q \neq a^{\prime} / q^{\prime}\), then \(\left|a q^{\prime}-a^{\prime} q\right| \geq 1\) and \[\begin{aligned} \frac{1}{Q^{2}} & \leq \frac{1}{q q^{\prime}} \leq \frac{\left|a q^{\prime}-a^{\prime} q\right|}{q q^{\prime}}=\left|\frac{a}{q}-\frac{a^{\prime}}{q^{\prime}}\right| \\ & \leq\left|\frac{a}{q}-\alpha\right|+\left|\alpha-\frac{a^{\prime}}{q^{\prime}}\right| \leq \frac{2 Q}{N} \end{aligned}\] or, equivalently, \[N \leq 2 Q^{3}=2(\log N)^{3 B},\]

  • This is impossible for \(N\) sufficiently large.

  • Therefore, the major arcs \(\mathfrak{M}(q, a)\) are pairwise disjoint for large \(N\). The set of major arcs is \[\mathfrak{M}=\bigcup_{q=1}^{Q} \bigcup_{\substack{a=0 \\(a, q)=1}}^{q} \mathfrak{M}(q, a) \subseteq[0,1],\] and the set of minor arcs is \[\mathfrak{m}=[0,1] \backslash \mathfrak{M},\]

  • We consider a weighted sum over the representations of \(N\) as a sum of three primes: \[R(N)=\sum_{p_{1}+p_{2}+p_{1}=N} \log p_{1} \log p_{2} \log p_{3}\]

  • Vinogradov obtained an asymptotic formula for \(R(N)\), from which the ternary Goldbach problem will follow by an elementary argument.

Circle method

  • We can use the circle method to express the representation function \(R(N)\) as the integral of a trigonometric polynomial over the major and minor arcs. Let \[F(\alpha)=\sum_{p \leq N}(\log p) e(p \alpha).\]

  • This exponential sum over primes is the generating function for \(R(N)\): \[\begin{aligned} R(N) & =\sum_{p_{1}+p_{2}+p_{3}=N} \log p_{1} \log p_{2} \log p_{3}\\ &=\int_{0}^{1} F(\alpha)^{3} e(-N \alpha) d \alpha \\ & =\int_{\mathfrak{M}} F(\alpha)^{3} e(-N \alpha) d \alpha+\int_{\mathfrak{m}} F(\alpha)^{3} e(-N \alpha) d \alpha. \end{aligned}\]

  • The main term in Vinogradov’s theorem will come from the integral over the major arcs, and the integral over the minor arcs will be negligible.

  • Just as in the Hardy–Littlewood asymptotic formula, the integral over the major arcs in Vinogradov’s theorem is (except for a small error term) the product of the singular series \(\mathfrak{S}(N)\) and an integral \(J(N)\). In this case, the integral \(J(N)\) is very easy to evaluate.

The integral over the major arcs

Let \[u(\beta)=\sum_{m=1}^{N} e(m \beta).\]

Then \[J(N)=\int_{-1 / 2}^{1 / 2} u(\beta)^{3} e(-N \beta) d \beta=\frac{N^{2}}{2}+O(N).\]

Proof. The number of representations of \(N\) as the sum of three positive integers is \[\begin{aligned} J(N) & =\int_{-1 / 2}^{1 / 2} u(\beta)^{3} e(-N \beta) d \beta \\ & =\int_{-1 / 2}^{1 / 2} \sum_{m_{1}=1}^{N} \sum_{m_{2}=1}^{N} \sum_{m_{3}=1}^{N} e\left(\left(m_{1}+m_{2}+m_{3}-N\right) \beta\right) d \beta \\ & =\binom{N-1}{2} =\frac{N^{2}}{2}+O(N). \end{aligned}\]$$\tag*{$\blacksquare$}$$

Siegel–Walfisz and approximations

If \(q \in \mathbb Z_+\) and \((q, a)=1\), then, for any \(C>0\), one has \[\vartheta(x ; q, a)=\sum_{\substack{p \leq x \\ p=a(\bmod q)}} \log p=\frac{x}{\varphi(q)}+O\left(\frac{x}{(\log x)^{C}}\right)\] for all \(x \geq 2\), where the implied constant depends only on \(C\).

Let \[F_{x}(\alpha)=\sum_{p=1}(\log p) e(p \alpha).\]

Let \(B, C\in\mathbb R_+\). If \(1 \leq q \leq Q=(\log N)^{B}\) and \((q, a)=1\), then \[F_{x}(a / q)=\frac{\mu(q)}{\varphi(q)} x+O\left(\frac{Q N}{(\log N)^{C}}\right)\] for \(1 \leq x \leq N\), where the implied constant depends only on \(B\) and \(C\).

Further approximations

Remember that \[u(\beta)=\sum_{m=1}^{N} e(m \beta).\]

Let \(B, C\in\mathbb R_+\) be such that \(C>2 B\). If \(\alpha \in \mathfrak{M}(q, a)\) and \(\beta=\alpha-a / q\), then \[F(\alpha)=\frac{\mu(q)}{\varphi(q)} u(\beta)+O\left(\frac{Q^{2} N}{(\log N)^{C}}\right),\] and \[F(\alpha)^{3}=\frac{\mu(q)}{\varphi(q)^{3}} u(\beta)^{3}+O\left(\frac{Q^{2} N^{3}}{(\log N)^{C}}\right),\] where the implied constants depend only on \(B\) and \(C\).

Contribution on major arcs

For any \(B, C, \varepsilon\in\mathbb R_+\) with \(C>2 B\), the integral over the major arcs is \[\begin{aligned} \int_{\mathfrak{M}} F(\alpha)^{3} e(-N \alpha) d \alpha &=\mathfrak{S}(N) \frac{N^{2}}{2}\\ &+O\left(\frac{N^{2}}{(\log N)^{(1-\varepsilon) B}}\right)+O\left(\frac{N^{2}}{(\log N)^{C-5 B}}\right), \end{aligned}\] where the implied constants depend only on \(B, C\), and \(\varepsilon\).

An exponential sum over primes

Recall that \[F(\alpha)=\sum_{p \leq N}(\log p) e(p \alpha).\]

If \[\left|\alpha-\frac{a}{q}\right| \leq \frac{1}{q^{2}},\] where \(a, q\in\mathbb Z\) are such that \(1 \leq q \leq N\) and \((a, q)=1\), then \[|F(\alpha)| =O\bigg(\left(\frac{N}{q^{1 / 2}}+N^{4 / 5}+N^{1 / 2} q^{1 / 2}\right)(\log N)^{4}\bigg).\]

Vaughan’s identity

For \(u \geq 1\), let \[M_{u}(k)=\sum_{\substack{d, k \\ d \leq u}} \mu(d) .\]

Let \(\Phi(k, l)\) be an arithmetic function of two variables. Then \[\sum_{u<l \leq N} \Phi(1, l)+\sum_{u<k \leq N} \sum_{u<l \leq \frac{N}{k}} M_{u}(k) \Phi(k, l)=\sum_{d \leq u} \sum_{u<l \leq \frac{N}{d}} \sum_{m \leq \frac{N}{ld}} \mu(d) \Phi(d m, l).\]

Proof. Hint: evaluate the sum \[S=\sum_{k=1}^{N} \sum_{u<l \leq \frac{N}{k}} M_{u}(k) \Phi(k, l)\] in two different ways.$$\tag*{$\blacksquare$}$$

Vaughan’s identity in estimates of exponential sums

Let \(\Lambda(l)\) be the von Mangoldt function. For every \(\alpha\in\mathbb R\), one has \[F(\alpha)=\sum_{p \leq N}(\log p) e(p \alpha)=S_{1}-S_{2}-S_{3}+O\big(N^{1 / 2}\big),\] where \[\begin{aligned} & S_{1}=\sum_{d \leq N^{2 / 5}} \sum_{l \leq \frac{N}{d}} \sum_{m \leq \frac{N}{l d}} \mu(d) \Lambda(l) e(\alpha d l m), \\ & S_{2}=\sum_{d \leq N^{2 / 5}} \sum_{l \leq N^{2 / 5}} \sum_{m \leq \frac{N}{ld}} \mu(d) \Lambda(l) e(\alpha d l m), \end{aligned}\] and \[S_{3}=\sum_{k>N^{2 / S}} \sum_{N^{2 / 5}<l \leq N / k} M_{N^{2 / 5}}(k) \Lambda(l) e(\alpha k l).\]

Proof. Apply Vaughan’s identity with \(u=N^{2/5}\) and \(\Phi(k, l)=\Lambda(l)e(\alpha kl)\), and note \[\sum_{u<l \leq N} \Phi(1, l)=F(\alpha)+O\big(N^{1 / 2}\big). \]$$\tag*{$\blacksquare$}$$

If \[\left|\alpha-\frac{a}{q}\right| \leq \frac{1}{q^{2}},\] where \(1 \leq q \leq N\) and \((a, q)=1\), then \[\left|S_{1}\right| =O\bigg(\left(\frac{N}{q}+N^{2 / 5}+q\right)(\log N)^{2} \bigg).\]

If \[\left|\alpha-\frac{a}{q}\right| \leq \frac{1}{q^{2}},\] where \(1 \leq q \leq N\) and \((a, q)=1\), then \[\left|S_{2}\right| =O\bigg(\left(\frac{N}{q}+N^{4 / 5}+q\right)(\log N)^{2}\bigg).\]

If \[\left|\alpha-\frac{a}{q}\right| \leq \frac{1}{q^{2}},\] where \(1 \leq q \leq N\) and \((a, q)=1\), then \[\left|S_{3}\right| =O\bigg(\left(\frac{N}{q^{1 / 2}}+N^{4 / 5}+N^{1 / 2} q^{1 / 2}\right)(\log N)^{4}\bigg).\]

For any \(B>0\), we have \[\Big|\int_{\mathfrak{m}} F(\alpha)^{3} e(-\alpha N) d \alpha\Big| =O \bigg(\frac{N^{2}}{(\log N)^{(B / 2)-5}}\bigg).\] where the implied constant depends only on \(B\).

  • Recall that \(Q=(\log N)^B\). Let \(\alpha \in \mathfrak{m}=[0,1] \backslash \mathfrak{M}\). By Dirichlet’s theoremfor any real number \(\alpha\) there exists a fraction \(a / q \in[0,1]\) with \(1 \leq q \leq N / Q\) and \((a, q)=1\) such that \[\left|\alpha-\frac{a}{q}\right| \leq \frac{Q}{q N} \leq \min \left(\frac{Q}{N}, \frac{1}{q^{2}}\right).\]

  • If \(q \leq Q\), then \(\alpha \in \mathfrak{M}(q, a) \subseteq \mathfrak{M}\), which is false. Therefore, \[Q<q \leq \frac{N}{Q}.\]

  • Then by the previous theorem we obtain \[\begin{aligned} |F(\alpha)| & =O\bigg(\left(\frac{N}{q^{1 / 2}}+N^{4 / 5}+N^{1 / 2} q^{1 / 2}\right)(\log N)^{4}\bigg) \\ & =O\bigg(\left(\frac{N}{(\log N)^{B / 2}}+N^{4 / 5}+N^{1 / 2}\left(\frac{N}{(\log N)^{B}}\right)^{1 / 2}\right)(\log N)^{4}\bigg) \\ & =O\bigg(\frac{N}{(\log N)^{(B / 2)-4}}\bigg). \end{aligned}\]

  • Since \[\vartheta(N)=\sum_{p \leq N} \log p = O(N),\] we have \[\int_{0}^{1}|F(\alpha)|^{2} d \alpha=\sum_{p \leq N}(\log p)^{2} \leq \log N \sum_{p \leq N} \log p = O(N \log N).\]

  • Thus \[\begin{aligned} \int_{\mathfrak{m}}|F(\alpha)|^{3} d \alpha & =O \bigg(\sup \{|F(\alpha)|: \alpha \in \mathfrak{m}\} \int_{\mathfrak{m}}|F(\alpha)|^{2} d \alpha \bigg)\\ & =O\bigg( \frac{N}{(\log N)^{(B / 2)-4}} \int_{0}^{1}|F(\alpha)|^{2} d \alpha \bigg)\\ & =O \bigg(\frac{N^{2}}{(\log N)^{(B / 2)-5}}\bigg). \end{aligned}\] This completes the proof. $$\tag*{$\blacksquare$}$$

Vinogradov’s theorem

Let \(\mathfrak{S}(N)\) be the singular series for the ternary Goldbach problem, i.e. \[\mathfrak{S}(N)=\prod_{p}\left(1+\frac{1}{(p-1)^{3}}\right) \prod_{p \mid N}\left(1-\frac{1}{p^{2}-3 p+3}\right).\]

  • For all sufficiently large odd integers \(N\) and for every \(A>0\), we have \[R(N)=\sum_{p_{1}+p_{2}+p_{1}=N} \log p_{1} \log p_{2} \log p_{3}=\mathfrak{S}(N) \frac{N^{2}}{2}+O\left(\frac{N^{2}}{(\log N)^{A}}\right),\] where the implied constant depends only on \(A\).

  • In particular, we have \[r(N)=\sum_{p_{1}+p_{2}+p_{3}=N} 1=\mathfrak{S}(N) \frac{N^{2}}{2(\log N)^{3}}\left(1+O\left(\frac{\log \log N}{\log N}\right)\right).\]

Top