Jurkat-Richert Sieve

Let \(\mathcal{A}=\{a(n)\}_{n=1}^{\infty}\) be a sequence of non-negative integers and \[|\mathcal{A}|=\sum_{n=1}^{\infty} a(n)<\infty .\] Let \(\mathcal{P}\) be a set of primes and let \(z \in \mathbb{R}, z \geq 2\). Let \[P(z)=\prod_{\substack{p \in \mathcal{P} \\ p< z}} p .\] We define the sieving function as \[S(\mathcal{A}, \mathcal{P}, z)=\sum_{(n, P(z))=1} a(n)\] Since \[\sum_{d \mid m} \mu(d)= \begin{cases}1 & \text { if } m=1 \\ 0 & \text { otherwise }\end{cases}\] we obtain \[S(\mathcal{A}, \mathcal{P}, z)=\sum_{d \mid P(z)} \mu(d)\left|\mathcal{A}_{d}\right|\] where \[\left|\mathcal{A}_{d}\right|=\sum_{d\mid n} a(n) \quad \text { converges. }\] Assume that for every \(n \geq 1\), we have a multiplicative function \(g(d)\) such that \[0 \leqslant g(p)<1\] for all \(p \in \mathcal{P}\).
Then for \(d \mid P(z)\) square-free integers define \[r(d)=\left|\mathcal{A}_{d}\right|-\sum_{n} a(n) g_{n}(d)\] and it follows that \[\begin{aligned} S(\mathcal{A}, \mathcal{P}, z) & =\sum_{d \mid P(z)} \mu(d)\left(\sum_{n} a(n) g_{n}(d)+r(d)\right) \\ & =\sum_{n} a(n) \prod_{p \mid P(z)}\left(1-g_{n}(p)\right)+\sum_{d \mid P(z)} \mu(d) r(d)\\ &=\sum_{n} a(n) V_{n}(z)+R(z) \end{aligned}\] where \[V_{n}(z)=\prod_{p \mid P(z)}\left(1-g_{n}(p)\right) \ \text { and } \ R(z)=\sum_{d \mid P(z)} \mu(d) r(d)\] However, \(R(z)\) sometimes causes "trouble" as it is too large to estimate; in this sense, it is much larger than \(V_{n}(z)\). To reduce error, we replace the möbius function with \(\lambda^{ \pm}\) truncated function such that \[\lambda^{+}(1)=\lambda^{-}(1)=1\] and for every \(m \geq 2\), \[\sum_{d \mid m} \lambda^{+}(d) \geqslant 0 . \quad \sum_{d \mid m} \lambda^{-}(d) \leqslant 0 .\] Moreover, if \(\exists D>0\) such that \(\lambda^{+}(d)=0\) for all \(d \geq D\), then we call D a support level.
Suppose that \(\lambda^{+}\) and \(\lambda^{-}\) have \(D\) as a support level. Then, \[\sum_{(n, P(z))=1} a(n) \leqslant \sum_{n} a(n) \sum_{d\mid P(z)} \lambda^{+}(d)\] and by previous \[\sum_{n=1}^{\infty} a(n) G_{n}\left(z, \lambda^{-}\right)+R^{-} \leq S(\mathcal{A}, \mathcal{P}, z) \leq \sum_{n=1}^{\infty} a(n) G_{n}\left(z, \lambda^{+}\right)+R^{+}\] where

\[G_{n}\left(z, \lambda^{ \pm}\right)=\sum_{d \mid P(z)} \lambda^{ \pm}(d) g_{n}(d) \text { and } R^{ \pm}(z)=\sum_{\substack{d \mid P(z) \\ d<D}} \lambda^{ \pm}(d) r(d)\] We then define \(\lambda_{1}^{ \pm}(d)\) over the range \(P_{1}\), which \(P=P_1 \cup \mathcal{Q}\) and for \(d \mid P(z)\) it can be uniquely factored as \(d_1d_2\) which \(d_{1}\) is relatively prime to the product of all primes in \(\mathcal{Q}\) , denoted as \(Q\). Now we define \[\lambda^{ \pm}(d)=\lambda_{1}^{ \pm}(d) \mu\left(d_{2}\right)\] for \(d\mid P(z)\), otherwise it is 0; and it is easy to verify that it satisfies the previous properties of \(\mathrm{d}^{ \pm}\), and it has support level \(DQ\) . It follows that \[G_{n}\left(z, \lambda^{ \pm}\right)=G_{n}\left(z, \lambda_{1}^{ \pm}\right) \prod_{q\mid Q(z)}\left(1-g_{n}(q)\right)\] where \[G_{n}\left(z, \lambda_{1}^{ \pm}\right)=\sum_{d_{1} \mid P_1(z)} \lambda_{1}^{ \pm}\left(d_{1}\right) g_{n}\left(d_{1}\right).\] and we replace \(R^{ \pm}\) by \(\pm R(D Q, \mathcal{P}, z)\) for \(\left|\lambda_{1}^{ \pm}(d)\right| \leqslant 1\), \[R(D Q, \mathcal{P}, z)=\sum_{d \mid P(z)}|r(d)| .\] We then construct the truncated function based on the subset of integers from 1 to \(D\).
Let \(\beta>0, D>0\) be real numbers. Let \(D^{+}\) be the set consisting of 1 of all square-free numbers \(d\) with \[d=p_{1} p_{2} \cdots p_{k}\] such that \[p_{k}<\cdots<p_{2}<p_{1}<D\] and \[p_{m}<\left(\frac{D}{p_{1} p_{2} \cdots p_{m}}\right)^{\frac{1}{\beta}}\] for all odd integers \(m\). Let \(D^{-}\) be the set consisting of 1 and all square-free numbers \(d\) with \[d=p_{1} \cdots p_{k}\] such that \[p_{k}<\cdots<p_{2}<p_{1}<D\] and \[p_{m}<\left(\frac{D}{p_{1} \cdots p_{m}}\right)^{1 / \beta}\] for all even integers. Let \[P(D)=\prod_{\substack{p \in \mathcal{P} \\ p < D}} p\] and define

\[\lambda^{+}(d)= \begin{cases}\mu(d) & \text { if } d \in D^{+} \text {and } d \mid P(D) \\ 0 & \text { otherwise }\end{cases}\] and \[\lambda^{-}(d)= \begin{cases}\mu(d) & \text { if } d \in D^{-} \text {and } d \mid P(D) \\ 0 & \text { otherwise }\end{cases}\] It follows from the definition that these functions are truncation functions. Recall that for \(g(d)\) multiplicative function, we have \[V(z)=\prod_{\substack{p \in \mathcal{P} \\ p<z}}(1-g(p))=\sum_{p \mid P(z)} \mu(d) g(d) .\] By induction over the number of primes in an interval, we get \[\sum_{\substack{p \in \mathcal{P} \\ w \leqslant p<z}} g(p) V(p)=V(w)\] Combining with the construction of \(\lambda \pm(d)\), we obtain the following Lemma.
Lemma. Let \(\beta\) be a set of primes. For \(\beta>1\) and \(2 \leqslant z \leqslant D\). Let \[y_{m}=y_{m}\left(\beta, D, p_{1} \cdots p_{m}\right)=\left(\frac{D}{p_{1} \cdots p_{m}}\right)^{1 / \beta} .\] Let \(\lambda^{ \pm}(d)\) be the truncated functions we defined previously, let \[G\left(z, \lambda^{ \pm}\right)=\sum_{d\mid P(z)} \lambda^{ \pm}(d) g(d) .\] Let \[\operatorname{T_n}(D, z)= \sum_{\substack{p_1,...,p_n\in\mathcal{P}\\ y_n\leq p_n<...<p_1<z \\ p_m<y_m \ \forall m<n,m\equiv n(\operatorname{mod} 2)}} g(p_1...p_n)V(p_n)\] Then \[G\left(z, \lambda^{+}\right)=V(z)+\sum_{\substack{n=1 \\ n\equiv 1(\operatorname{mod} 2)}}^{\infty} \operatorname{T_n}(D, z)\] and \[G\left(z, \lambda^{-}\right)=V(z)-\sum_{\substack{n=1 \\ n \equiv 0(\bmod 2)}}^{\infty} \operatorname{T_n}(D, z)\] Moreover, \[\operatorname{T_n}(D, z) \geqslant 0\] for all \(n \geq 1\), and \[G\left(z, \lambda^{-}\right) \leqslant V(z) \leqslant G\left(z, \lambda^{+}\right) .\] If \(\beta \leqslant \frac{\log D}{\log z}=s\), then \[\operatorname{T_n}(D, z)=0 \text { for } n \leqslant s-\beta \text {. }\] Now we consider for \(\beta=2\) from now on. From the combinatorial identity of \(V(z)\), we obtain \[T_{1}(D, z)=V\left(D^{1 / 3}\right)-V(z) .\] Let \(z \geq 2\) and choose \(D \geqslant z\) if \(n\) is odd, \(D \geq z^{2}\) if \(n\) is even, so \[s=\frac{\log D}{\log z} \geqslant \begin{cases}1 & \text { if } n \text { is odd. } \\ 2 & \text { if } n \text { is even. }\end{cases}\] Let \(n \geq 2\). If \(n\) is even, or if \(n\) is odd and \(s \geqslant 3\), then \[T_{n}(D, z)=\sum_{\substack{p \in \mathcal{P} \\ p<z}} g(p) T_{n-1}\left(\frac{D}{p}, p\right) .\] If \(n\) is odd and \(1 \leqslant S \leqslant 3\), then \[\operatorname{T_n}(D, z)=\sum_{\substack{p \in \mathcal{P} \\ p < D^{\frac{1}{3}}}} g(p) T_{n-1}\left(\frac{D}{p}, p\right)\] If there exists some \(K>1\) such that \[\prod_{\substack{p \in \mathcal{P} \\ u \leqslant p<z}}(1-g(p))^{-1} \leqslant K \frac{\log z}{\log u}\] for all \(u\) such that \(1<U<Z\). Let \(\Phi\) be a continuous, increasing function on the interval \([w, z]\) which \(w\) is any real number in \((1, z)\). Then the step function \[\begin{aligned} S(u) & =\sum_{u \leqslant p(z} g(p) V(p) \\ & =V(u)-V(z) \\ & \leqslant\left(K \frac{\log z}{\log u}-1\right) V(z) . \end{aligned}\] Let \[w \leqslant p_{k}<p_{k-1}<\cdots<p_{1}<z\] which \(p_{i} \in \mathcal{P}\) for all \(i=1, \ldots, k\).
Then \[\begin{aligned} \sum_{\substack{p \in \mathcal{P} \\ w \leq p<z}} g(p) V(p) \Phi(p)= & \sum_{i=1}^{k} g\left(p_{i}\right) V\left(p_{i}\right) \Phi\left(p_{i}\right) \\ = & S\left(p_{k}\right) \Phi(w)+S\left(p_{k}\right)\left(\Phi\left(p_{w}\right)-\Phi(w)\right) \\ & +\sum_{i=1}^{k-1} S\left(p_{i}\right)\left(\Phi\left(p_{i}\right)-\Phi\left(p_{i+1}\right)\right) \\ = & S(w) \Phi(w)+\int_{w}^{z} S(u) d \Phi(u) \\ = & S(z) \Phi(z)-\int_{w}^{z} \Phi(u) d S(u) \\ \leq & (K-1) V(z) \Phi(z)-K V(z) \int_{w}^{z} \Phi(u) d\left(\frac{\log z}{\log u}\right). \end{aligned}\] For \(n=1\) and \(1 \leqslant s \leqslant 3\), we have \[\begin{aligned} \frac{T_{1}(D, z)}{V(z)} & =\frac{V\left(D^{\frac{1}{3}}\right)}{V(z)}-1 \\ & =\prod_{D^{\frac{1}{3}} \leq p<z}(1-g(p))^{-1}-1 \\ & \leqslant 3 K \frac{\log D}{\log z}-1 \\ & <\left(\frac{3}{s}-1\right)+3(K-1) \end{aligned}\] Let \(n \geq 2\). If \(n\) is even or if \(n\) is odd and \(s \geq 3\), then \[T_{n}(D, z)=\sum_{\substack{p \in \mathcal{P} \\ p<z}} g(p) T_{n-1}\left(\frac{D}{p}, p\right)\] To bound \(T_{n}(D, z)\), it is natural to assume that there is a sequence of smooth functions times \(V(z)\). Let \(\Psi_{n}\) be a sequence of smooth functions, and define \[\Phi_{n}(u)=\Psi_{n}\left(\frac{\log D}{\log n}-1\right)\] and claim that \[T_{n}(D, z)<V(z) \Psi_{n}(s) .\] then by the induction hypothesis, it follows that \[\begin{aligned} T_{n}(D, z) & <\sum_{\substack{p \in \mathcal{P} \\ p<z}} g(p) V(p) \Psi_{n-1}\left(\frac{\log D}{\log p}-1\right) \\ & =\sum_{\substack{p \in \mathcal{P} \\ p<z}} g(p) V(p) \Phi_{n}(p) \\ & =(K-1) V(z) \Psi_{n-1}(s-1)-\frac{K V(z)}{s} \int_{1}^{z} \Phi_{n-1}(u) d \frac{\log D}{\log u} \\ & =(K-1) V(z) \Psi_{n-1}(s-1)+\frac{K V(z)}{s} \int_{s}^{\infty} \Psi_{n-1}(t-1) d t \end{aligned}\] Similarly, for \(n \geq 3\) odd and \(1 \leqslant s \leqslant 3\), then \[T_{n}(D, z)<(K-1) V(z) \Psi_{n-1}(2)+K V(z) \int_{3}^{\infty} \Psi_{n-1}(t-1) d t\] then by choosing an appropiate \(K\), and constructing \(\Psi_{n}\) carefully, the induction can be proved. Let \[\Psi_{n}(s)=f_{n}(s)+h_{n}(s)\] which \(f_n \ \text{and} \ h_n\) are constructed as follows. For \(s \geqslant 1\). let \(R_{n}(s)\) be the open convex region of Euclidean space consisting of all points \(\left(t_{1}, \ldots, t_{n}\right) \in \mathbb{R}^{n}\) s.t. \[\begin{aligned} & 0<t_{n}<\cdots<t_{1}<\frac{1}{s}, \\ & t_{1}+\cdots+t_{n}+2 t_{n}>1 . \end{aligned}\] and \[t_{1}+\ldots+t_{m}+2 t_{m}<1 \ \ \ \ \text{if} \ m<n \ \text{and} \ m \equiv n(\operatorname{mod} 2).\] For \(n \geqslant 1\) and \(s \geqslant 1\), we define the function \(f_{n}(s)\) by the integral \[s f_{n}(s)=\int \cdots \int_{R_{n}(s)} \frac{d t_{1} \cdots d t_{n}}{\left(t_{1} \cdots t_{n}\right) t_{n}}\] We construct \(h(s)\) for \(s \geqslant 1\) as follows: \[h(s)=\left\{\begin{array}{lll} e^{-2} & \text { for } & 1 \leq s \leq 2 \\ e^{-s} & \text { for } & 2 \leq s \leq 3 \\ 3 s^{-1} e^{-s} & \text { for } & s \geq 3 \end{array}\right.\] For \(s \geqslant 2\), let \[H(s)=\int_{s}^{\infty} h(t-1) d t .\] Let \[\alpha=\frac{H(2)}{2 h(2)}=\frac{e^{2} H(2)}{2}=1-\frac{1}{2 e}+\frac{3 e^{2}}{2} \int_{0}^{\infty} t^{-1} e^{-t} d t\] We define the number \[\tau=\alpha+5(K-1)+11 e^{-8}\] and the function \[h_{n}(s)=(K-1) \tau^{n} e^{10} h(s)\] and in this case we let \(K\) satisfy \[1<K<1+\frac{1}{200} .\] And after some calculation, we obtain the bound for \(T_{n}(D, z)\), and for \(s \geqslant 1\) \[F(s)=1+\sum_{\substack{n=1 \\ n\equiv 1 (\operatorname{mod}2)}}^{\infty} f_{n}(s)=1+O\left(e^{-s}\right),\] and for \(s \geqslant 2\) \[f(s)=1-\sum_{\substack{n=1 \\ n \equiv 0(\operatorname{mod} 2)}}^{\infty} f_{n}(s)=1+O\left(e^{-s}\right) .\] Moreover, let \(0<\varepsilon<\frac{1}{200}, K=1+\varepsilon\), and \[G\left(z, \lambda^{ \pm}\right)=\sum_{d \mid P(z)} \lambda^{ \pm}(d) g(d) .\] Then \[G\left(z, \lambda^{+}\right)<V(z)\left(F(s)+\varepsilon e^{14-\varepsilon}\right)\] and \[G\left(z, \lambda^{-}\right)>V(z)\left(f(s)-\varepsilon e^{14-\varepsilon}\right)\] Combining with the fact that this \(\lambda^{ \pm}\) is actually \(\lambda_{1}^{ \pm}\) defined on \(\mathcal{P}_{1}\), we obtain the following theorem.
Theorem(Jurkat-Richert). Let \(\mathcal{A}=\{a(n)\}_{n=1}^{\infty}\) such that \[a(n) \geqslant 0 \quad \forall n \in \mathbb{N} \text { and }|\mathcal{A}|=\sum_{n=1}^{\infty} a(n)<\infty\] Let \(P\) be a set of primes, for \(z \geq 2\), let \[P(z)=\prod_{\substack{p \in \mathcal{P} \\ p<z}} p .\] Let \[S(\mathcal{A}, \mathcal{P}, z)=\sum_{\substack{n=1 \\(n, P(z))=1}}^{\infty} a(n)\] For every \(n \geqslant 1\), let \(g_{n}(d)\) be a multiplicative function s.t. \[0<g_{n}(p)<1 \quad \forall p \in P .\] Define \(r(d)\) by \[\left|\mathcal{A}_{d}\right|=\sum_{d=1}^{\infty} a(n)=\sum_{n=1}^{\infty} a(n) g(d)+r(d)\] Let \(\mathcal{Q}\) be a finite subset of \(\mathcal{P}\), and let \(Q\) be the product of primes in \(\mathcal{Q}\). Suppose that, for some \(\varepsilon\) satisfying \(0<\varepsilon<\frac{1}{200}\), the inequality \[\prod_{\substack{p \in \mathcal{P} \backslash \mathcal{Q} \\ u \leq p<z}}^{\pi}\left(1-g_{n}(p)\right)^{-1}<(1+\varepsilon) \frac{\log z}{\log u}\] holds for all \(1<u<z\). Then for any \(D \geq z\) there is the upper bound \[S(\mathcal{A}, \mathcal{P}, z)<\left(F(s)+\varepsilon e^{14-\varepsilon} \right)X+R\] and for \(D \geqslant z^{2}\) there is the lower bound \[S(\mathcal{A}, \mathcal{P}, z)>\left(f(s)-\varepsilon e^{14-\varepsilon}\right) X+R\] where \[\begin{aligned} s & =\frac{\log D}{\log z}, \\ X & =\sum_{n=1}^{\infty} a(n) \prod_{p\mid P(z)}(1-g(p)). \end{aligned}\] and the remainder term is \[R=\sum_{\substack{d \mid P(z) \\ d < DQ}}|r(d)| .\] If there is a multiplicative function \(g(d)\) s.t. \(g_{n}(d)=g(d)\) for all \(n\), then \[X=V(z)|\mathcal{A}|,\] where \[V(z)=\prod_{p \mid P(z)}(1-g(p)) .\]
Moreover, through the recurrence property of \(f_{n}(s)\), we could obtain and solve a differential-difference equation that allows us to compute initial values for \(F(s)\) and \(f(s)\), \[F(s)=\frac{2 e^{\gamma}}{s} \quad \text { for } \quad 1 \leqslant s \leqslant 3 \text {, }\] and \[f(s)=\frac{2 e^{\gamma} \log (s-1)}{s} \quad \text { for } \quad z \leqslant s \leqslant 4 \text {, }\] where \(\gamma\) is the Euler’s constant.