Age estimation algorithm based on deep learning and its application in fall detection

Jiayi Yu; Ye Tao; Huan Zhang; Zhibiao Wang; Wenhua Cui; Tianwei Shi; Jiayi Yu; Ye Tao; Huan Zhang; Zhibiao Wang; Wenhua Cui; Tianwei Shi

doi:10.3934/era.2023251

Electronic Research Archive

2023, Volume 31, Issue 8: 4907-4924. doi: 10.3934/era.2023251

Previous Article Next Article

Research article

Age estimation algorithm based on deep learning and its application in fall detection

School of Computer Science and Software Engineering, University of Science and Technology Liaoning, China

Received: 17 April 2023 Revised: 16 June 2023 Accepted: 22 June 2023 Published: 11 July 2023

With the continuous development and progress of society, age estimation based on deep learning has gradually become a key link in human-computer interaction. Widely combined with other fields of application, this paper performs a gradient division of human fall behavior according to the age estimation of the human body, a complete priority detection of the key population, and a phased single aggregation backbone network VoVNetv4 was proposed for feature extraction. At the same time, the regional single aggregation module ROSA module was constructed to encapsulate the feature module regionally. The adaptive stage module was used for feature smoothing. Consistent predictions for each task were made using the CORAL framework as a classifier and tasks were divided in binary. At the same time, a gradient two-node fall detection framework combined with age estimation was designed. The detection was divided into a primary node and a secondary node. In the first-level node, the age estimation algorithm based on VoVNetv4 was used to classify the population of different age groups. A face tracking algorithm was constructed by combining the key point matrices of humans, and the body processed by OpenPose with the central coordinates of the human face. In the secondary node, human age gradient information was used to detect human falls based on the AT-MLP model. The experimental results show that compared with Resnet-34, the MAE value of the proposed method decreased by 0.41. Compared with curriculum learning and the CORAL-CNN method, MAE value decreased by 0.17 relative to the RMSE value. Compared with other methods, the method in this paper was significantly lower, with a biggest drop of 0.51.

Keywords:

Citation: Jiayi Yu, Ye Tao, Huan Zhang, Zhibiao Wang, Wenhua Cui, Tianwei Shi. Age estimation algorithm based on deep learning and its application in fall detection[J]. Electronic Research Archive, 2023, 31(8): 4907-4924. doi: 10.3934/era.2023251

Related Papers:

[1]	Hongyan Dui, Yong Yang, Xiao Wang . Reliability analysis and recovery measure of an urban water network. Electronic Research Archive, 2023, 31(11): 6725-6745. doi: 10.3934/era.2023339
[2]	Xu Zhan, Yang Yong, Wang Xiao . Phased mission reliability analysis of unmanned ship systems. Electronic Research Archive, 2023, 31(10): 6425-6444. doi: 10.3934/era.2023325
[3]	Majed Alowaidi, Sunil Kumar Sharma, Abdullah AlEnizi, Shivam Bhardwaj . Integrating artificial intelligence in cyber security for cyber-physical systems. Electronic Research Archive, 2023, 31(4): 1876-1896. doi: 10.3934/era.2023097
[4]	Xiaoyang Xie, Shanghua Wen, Minglong Li, Yong Yang, Songru Zhang, Zhiwei Chen, Xiaoke Zhang, Hongyan Dui . Resilience evaluation and optimization for an air-ground cooperative network. Electronic Research Archive, 2024, 32(5): 3316-3333. doi: 10.3934/era.2024153
[5]	Shuang Yao, Dawei Zhang . A blockchain-based privacy-preserving transaction scheme with public verification and reliable audit. Electronic Research Archive, 2023, 31(2): 729-753. doi: 10.3934/era.2023036
[6]	Shanpu Gao, Yubo Li, Anping Wu, Hao Jiang, Feng Liu, Xinlong Feng . An intelligent optimization method for accelerating physical quantity reconstruction in computational fluid dynamics. Electronic Research Archive, 2025, 33(5): 2881-2924. doi: 10.3934/era.2025127
[7]	Wanxun Jia, Ling Li, Haoyan Zhang, Gengxiang Wang, Yang Liu . A novel nonlinear viscous contact model with a Newtonian fluid-filled dashpot applied for impact behavior in particle systems. Electronic Research Archive, 2025, 33(5): 3135-3157. doi: 10.3934/era.2025137
[8]	Sida Lin, Jinlong Yuan, Zichao Liu, Tao Zhou, An Li, Chuanye Gu, Kuikui Gao, Jun Xie . Distributionally robust parameter estimation for nonlinear fed-batch switched time-delay system with moment constraints of uncertain measured output data. Electronic Research Archive, 2024, 32(10): 5889-5913. doi: 10.3934/era.2024272
[9]	Yunying Huang, Wenlin Gui, Yixin Jiang, Fengyi Zhu . Types of systemic risk and macroeconomic forecast: Evidence from China. Electronic Research Archive, 2022, 30(12): 4469-4492. doi: 10.3934/era.2022227
[10]	Yazhou Chen, Dehua Wang, Rongfang Zhang . On mathematical analysis of complex fluids in active hydrodynamics. Electronic Research Archive, 2021, 29(6): 3817-3832. doi: 10.3934/era.2021063

Abstract

1. Introduction

Let $q$ be a power of odd prime. Several researchers have looked into a variety of properties about the primitive roots modulo $q$ . Let $g_1, g_2$ represent two primitive roots modulo $q$ , $a$ , $b$ and $c$ represent arbitrary non-zero elements in $\mathbb{F}_q$ . Is there some $q_0$ such that for all $q > q_0$ , there is always one representation

$\begin{eqnarray} a = bg_1+cg_2\ ? \end{eqnarray}$

(1.1)

For $b = 1$ and $c = -1$ , Vegh ^[1] considered a specific form of Eq (1.1), which is known as Vegh's Conjecture, (see [,§ F9] for further details). Cohen ^[3] demonstrated Vegh's Conjecture for all $q > 7$ .

For $b = 1$ and $c = 1$ , Golomb ^[4] proposed another specific form of Eq (1.1). This was proved by Sun ^[5] for $q > 2^{60}\approx 1.15\times 10^{18}$ .

Moreover, Cohen et al. ^[6] studied linear sums of primitive roots and their inverses in finite fields $\mathbb{F}_q$ and showed that if $q > 13$ , then for arbitrary non-zero $a, b \in \mathbb{F}_q$ , there is a pair of primitive elements ( $g_1$ , $g_2$ ) of $\mathbb{F}_q$ such that both $ag_1+bg_2$ and $ag^{-1}_1+bg^{-1}_2$ are primitive.

Let $p$ be an odd prime. Carlitz ^[7] relied on some results of Davenport and obtained for any $k-1$ fixed integers $c_1, c_2, \ldots, c_{k-1}$ with $c_i \geq 1 (i = 1, 2, \ldots, k-1)$ . Let $g, g_1, \ldots, g_{k-1}$ be primitive roots modulo $p$ and $N_k$ denote the number of $g\bmod p$ such that $g_1-g = c_1, \ldots, g_{k-1}-g = c_{k-1}$ . Then

$\begin{eqnarray*} N_k \sim \frac{\phi^{k}(p-1)}{p^{k-1}}\ \hfill (p\rightarrow \infty). \end{eqnarray*}$

More results of the primitive roots distribution can be found in ^[8,9,10,11].

Lehmer [,§ F12] proposed the definition of $Lehmer~~ number$ , according to which $a$ is a $Lehmer ~~number$ if and only if $a$ and $\bar{a}$ have opposite parity, i.e., $(2, a+\bar{a}) = 1$ , where $\bar{a}$ is the multiplicative inverse of $a$ modulo $p$ . It is simple to demonstrate that there are no Lehmer numbers modulo $p$ when $p = 3$ or $7$ . Zhang ^[12] established that if $M_p$ denotes the number of Lehmer numbers modulo $p$ , then

$\begin{eqnarray*} M_p = \frac{p-1}{2}+\mathrm{O}\left(p^{\frac{1}{2}}\ln^2p\right). \end{eqnarray*}$

A Lehmer number that is also a primitive root modulo $p$ will be called a $Lehmer~~ primitive~~ root$ or an $LPR$ . The inverse of an $LPR$ is also an $LPR$ . We assume that $p > 3$ because there is no Lehmer number modulo $3$ . Wang and Wang ^[13] investigated the distribution of $LPRs$ involving Golomb's conjecture. Let $G_p$ denote the number of Golomb pairs $(a, b)$ (i.e., $a+b\equiv 1(\bmod\; p)$ ) are $LPRs$ . They showed

$\begin{eqnarray*} G_p = \frac{1}{4}\frac{\phi^2(p-1)}{p-1}+O\left(\frac{\phi^2(p-1)}{p^{\frac{5}{4}}}\cdot 4^{\omega(p-1)}\cdot \ln^2p\right). \end{eqnarray*}$

Let $N_p$ denote the number of $LPRs$ modulo $p$ . For odd integers $m\geq 3$ , define the positive number $T_m$ by

$\begin{eqnarray*} T_m = \frac{2}{m\ln m}\sum\limits_{j = 1}^{(m-1)/2}\tan\left(\frac{\pi j}{m}\right). \end{eqnarray*}$

Cohen and Trudgian ^[14] improved the result of Wang and Wang ^[13] and showed

$\begin{eqnarray*} \left|N_p-\frac{\phi(p-1)}{2}\right| < T_p^2\frac{\phi(p-1)}{p-1}2^{\omega(p-1)}p^{\frac{1}{2}}\ln^2p \end{eqnarray*}$

and

$\begin{eqnarray*} \left|G_p-\frac{\phi^2(p-1)}{4(p-1)^2}(p-2)\right| < \frac{\phi^2(p-1)}{4(p-1)^2}T_p^2[2^{2\omega(p-1)}(9\ln^2p+1)-1]p^{\frac{1}{2}}, \end{eqnarray*}$

where $\frac{2}{\pi}\left(1+\frac{0.548}{\ln p}\right) < T_p < \frac{2}{\pi}\left(1+\frac{1.549}{\ln p}\right)$ .

Specifically, they obtained that for an odd prime $p(\neq 3, 7)$ , there exists an $LPR$ modulo $p$ .

Inspired by the results of Cohen and Trudgian ^[14] and Wang and Wang ^[13], we mainly studied the distribution of $LPRs$ modulo $p$ related to the Golomb's conjecture in two aspects. On the one hand, we extend Eq (1.1) to the case involving $k > 1$ variables. Let $\mathcal{R}$ be set of $LPRs$ modulo $p$ that is a subset of ${\mathbb{F}}_p$ . $a_1, a_2, \ldots, a_k, c$ are non-zero elements in ${\mathbb{F}}_p$ and $N(\mathcal{R}, p)$ denotes the number of solutions of the equation

$\begin{eqnarray*} a_1g_1+a_2g_2+\cdots+a_kg_k = c, \ g_1, g_2, \ldots, g_k \in \mathcal{R}. \end{eqnarray*}$

We consider the distribution properties of $N(\mathcal{R}, p)$ , and obtain the following:

Theorem 1. Let $p > 3$ be an odd prime. Then we have

$\begin{eqnarray*} N(\mathcal{R}, p) = \frac{\phi^k(p-1)}{2^kp}+\mathrm{O}\left(\frac{\phi^k(p-1)}{p^\frac{3}{2}}2^{k\omega(p-1)}\ln^{2k}p\right), \end{eqnarray*}$

where the symbol $\mathrm{O}$ is dependent on $k$ .

When $k = 2$ , we can obtain the number of the Golomb pairs that are $LPRs$ .

On the other hand, we consider the distribution of $k$ consecutive $LPRs$ and generalize it to a more general form.

Let $f(x)\in \mathbb{F}_p[x]$ . Define

$\begin{eqnarray*} M(f(x), \mathcal{R}, p) = \#\{x: 1\leq x\leq p-1, f(x+c_1), f(x+c_2), \cdots, f(x+c_k) \in \mathcal{R} \}. \end{eqnarray*}$

Then we have:

Theorem 2. Let $f(x)\in \mathbb{F}_p[x]$ with degree $l\geq 1$ . $c_1, c_2, \ldots, c_k$ are distinct elements in $\mathbb{F}_p$ . Suppose that one of the following conditions holds:

$\rm (i)$ $f(x)$ is irreducible,

$\rm (ii)$ $f(x)$ has no multiple zero in $\bar{\mathbb{F}}_p$ and $k = 2$ ,

$\rm (iii)$ $f(x)$ has no multiple zero in $\bar{\mathbb{F}}_p$ and $(4k)^{l} < p$ .

Then we have

$\begin{eqnarray*} &&M(f(x), \mathcal{R}, p) = \frac{1}{2^k}\frac{\phi^k(p-1)}{(p-1)^{k-1}}+\mathrm{O}\left(\frac{\phi^k(p-1)}{p^{k-\frac{1}{2}}}2^{k\omega(p-1)}\ln^{2k}p\right), \end{eqnarray*}$

where the symbol $\mathrm{O}$ is dependent on $k$ and $l$ .

Take $f(x) = x$ , $c_k = 0$ in Theorem 2. Then we can get the number of $k$ consecutive primitive roots $x, x+c_1, \ldots, x+c_{k-1}$ are Lehmer numbers, which is:

Corollary 1. Let $p$ be an odd prime. Then for any $1\leq\ x\leq(p-1)$ that is an LPR modulo $p$ , we have

$\begin{eqnarray*} &&M(x, \mathcal{R}, p) = \frac{1}{2^k}\frac{\phi^k(p-1)}{(p-1)^{k-1}}+\mathrm{O}\left(\frac{\phi^k(p-1)}{p^{k-\frac{1}{2}}}2^{k\omega(p-1)}\ln^{2k}p\right), \end{eqnarray*}$

where the symbol $\mathrm{O}$ is dependent on $k$ .

When $k = 1, 2$ , we can easily deduce the Theorem 1 and Theorem 6 in Cohen and Trudgian ^[14], respectively.

Notation: Throughout this paper, $\mathbb{F}_q$ denotes a finite field of characteristic $p$ , $\bar{\mathbb{F}}_q$ denotes the algebraic closure of $\mathbb{F}_q$ , $\phi(n)$ is reserved for the Euler function, $\mu(n)$ is the M $\ddot{o}$ bius function. We use $\omega(n)$ to denote the number of all distinct prime divisors of $n$ . Write $\sum_{\chi_d}$ to denote a sum over all $\phi(d)$ multiplicative characters $\chi_d$ of order $d$ over ${\mathbb{F}}_p$ , and denote by $\mathop{\sum_{n = 1}^{p}}^{'}$ the summation of $1\leq n\leq p$ with $(n, p) = 1$ . $\tau(\chi)$ is the classical Gauss sums associated with character $\chi$ mudulo $p$ . $f\ll g$ means $|f|\leq cg$ with some positive constant $c$ , $f = \mathrm{O}(g)$ means $f\ll g$ .

2. Some lemmas

To complete the proof of the theorems, we need following several lemmas. The proofs of these lemmas require some basic knowledge of analytic number theory, which can be found in ^[15].

Lemma 1. Let $p$ be an odd prime. Then for any integer $a$ coprime to $p$ (i.e., $(a, p) = 1$ ), we have the identity

$\frac{\phi(p-1)}{p-1}\sum\limits_{d\mid p-1}\frac{\mu(d)}{\phi(d)}\sum\limits_{\chi_d}\chi_d(a) = \left\{ \begin{array}{ll} 1, &\ { if~~ a ~~ is ~~ a~~ primitive ~~ root ~~ \text{mod}~~ p;}\\ 0, &\ {if~~ a ~~ is~~ not~~ a~~ primitive~~ root~~ \text{mod}~~ p.} \end{array}\right.$

Proof. See Proposition 2.2 of Narkiewicz ^[16]. □

Lemma 2. Let $p$ be an odd prime, $\chi$ be a nonprincipal multiplicative character modulo $p$ of order $d$ . Suppose $g(x)\in \mathbb{F}_p[x]$ has precisely $m$ distinct ones among its zeros, and suppose that $g(x)$ is not the constant multiple of a $d$ -th power over $\mathbb{F}_q$ . Then

$\left|\sum\limits_{x\in \mathbb{F}_p}\chi\left(g(x)\right)\right|\leq \left(m-1\right)\cdot p^{\frac{1}{2}}.$

Proof. See Theorem 2C in Chapter 2 of Schmidt ^[17]. □

Lemma 3. Let $\mathbb{F}_q$ be a finite field of characteristic $p$ , $\psi$ be a nontrivial additive character and $\chi$ be a nonprincipal multiplicative character on $\mathbb{F}_q$ of order $d$ . For two rational functions $f(x), g(x)\in\mathbb{F}_q[x]$ , define $K(\psi, f;\chi, g) = \sum_{x\in\mathbb{F}_q\backslash S}\chi(g(x))\psi(f(x))$ , where $S$ denotes the set of poles of $f(x)$ and $g(x)$ . Suppose the following conditions hold:

$\rm (i)$ $g(x)$ is not the constant multiple of a $d$ -th power over $\mathbb{F}_q$ .

$\rm (ii)$ $f(x)$ is not of the form $(h(x))^p-h(x)$ with a rational function $h(x)$ over $\mathbb{F}_q$ .

Then we have

$|K(\psi, f;\chi, g)|\leq(\deg(f)+m-1)\sqrt{q},$

where $m$ is the number of distinct roots and (noninfinite) poles of $g(x)$ in $\mathbb{F}_q$ .

Proof. See Theorem 2G in Chapter 2 of Schmidt ^[17]. □

Lemma 4. Let $p$ be an odd prime. Let $c_1, \cdots, c_k$ be distinct elements in $\mathbb{F}_p$ . Assume that $f(x)\in\mathbb{F}_{p}[x]$ with $\deg(f) = l$ . Define the polynomial

$h(x) = f(x+c_1)\cdots f(x+c_k).$

Suppose one of the following conditions holds:

$\rm (i)$ $f(x)$ is irreducible,

$\rm (ii)$ $f(x)$ has no multiple zero in $\bar{\mathbb{F}}_p$ and $k = 2$ ,

$\rm (iii)$ $f(x)$ has no multiple zero in $\bar{\mathbb{F}}_p$ and $(4k)^{l} < p$ .

Then $h(x)$ has at least one simple root in $\bar{\mathbb{F}}_p$ .

Proof. Suppose that $f(x)$ is irreducible. Then $f(x+c_1), \cdots, f(x+c_k)$ are distinct irreducible polynomials, and $h(x)$ has at least $k$ simple roots in $\bar{\mathbb{F}}_p$ . The cases of $\rm (ii)$ and $\rm (iii)$ can be proved by Theorem 2 and Lemma 2 of ^[18], for $k = 2$ or $(4k)^{l} < p$ , $(l, k, p)$ is "admissible triple, " then $f(x+c_1)\cdots f(x+c_k)$ has at least one simple root. □

Lemma 5. Let $p$ be an odd prime, $m_1, \ldots, m_k, n_1, \ldots, n_k$ be integers with $(m_1\cdots m_kn_1\cdots n_k, p) = 1$ , and polynomials $g(x), f_1(x), \ldots, f_k(x)\in\mathbb{F}_p[x]$ . Let $\chi$ be a Dirichlet character modulo $p$ of order $d$ . Define

$\begin{eqnarray*} &&K(\chi, g, f_1, \cdots, f_k;p)\\ && = \mathop{\sum\limits_{x = 1}^{p}}_{(f_1(x)\cdots f_k(x), p) = 1}\chi(g(x)) e\left(\frac{m_1f_1(x)+\cdots+m_kf_k(x)+n_1\overline{f_1(x)}+\cdots+n_k\overline{f_k(x)}}{p}\right). \end{eqnarray*}$

Suppose the following conditions hold:

$\rm (i)$ $g(x)$ can not be the constant multiple of a $d$ -th power over $\mathbb{F}_p$ .

$\rm (ii)$ $F(x) = f_1(x)\cdots f_k(x)$ has at least one simple root in $\bar{\mathbb{F}}_p$ .

Then we have

$\begin{eqnarray*} |K(\chi, g, f_1, \cdots, f_k;p)| \leq\left(\max(\deg(f_1), \cdots, \deg(f_k))+l\right)\sqrt{p}, \end{eqnarray*}$

where $e(x) = e^{2\pi ix}$ and $l$ is the number of distinct roots of $g(x)$ in $\bar{\mathbb{F}}_p$ .

Proof. It is clear that

$\begin{eqnarray*} &&m_1f_1(x)+\cdots+m_kf_k(x)+n_1\overline{f_1(x)}+\cdots+n_k\overline{f_k(x)}\\ && = \frac{F(x)(m_1f_1(x)+\cdots+m_kf_k(x))+n_1\frac{F(x)}{f_1(x)}+\cdots +n_k\frac{F(x)}{f_k(x)}}{F(x)} : = \frac{G(x)}{F(x)}. \end{eqnarray*}$

Condition $\rm (i)$ is the same as Lemma 3. So our goal is to prove the rational function $G(x)/F(x)$ satisfies condition $\rm (ii)$ in Lemma 3 if $F(x)$ has a simple root in $\bar{\mathbb{F}}_p$ . Assume that there are polynomials $K(x), L(x)\in\mathbb{F}_p[x]$ with $(K(x), L(x)) = 1$ such that

$\frac{G(x)}{F(x)} = \left(\frac{K(x)}{L(x)}\right)^p-\left(\frac{K(x)}{L(x)}\right).$

Then we have

$\begin{eqnarray} G(x)L(x)^p = \left(K(x)^p-K(x)L(x)^{p-1}\right)F(x). \end{eqnarray}$

(2.1)

Since $F(x) = f_1(x)\cdots f_k(x)$ has at least one simple root in $\bar{\mathbb{F}}_p$ , then there exists an irreducible polynomial $w(x)\in\mathbb{F}_p[x]$ such that $w(x)\mid F(x)$ and $w(x)^2\nmid F(x)$ . Assume that $w(x)\mid f_1(x)$ , then we have

$w(x)\nmid\frac{F(x)}{f_1(x)}, \ w(x)\mid\frac{F(x)}{f_i(x)}(i = 2, \cdots, k).$

Hence, from Eq (2.1)

$w(x)\nmid G(x)\Longrightarrow w(x)\mid L(x)^p\Longrightarrow w(x)\mid L(x)$

$w(x)^2\mid L(x)^{p-1}\Longrightarrow w(x)^2\mid K(x)^pF(x)\Longrightarrow w(x)\mid K(x),$

which contradicts to $(K(x), L(x)) = 1$ . Therefore, from Lemma 3 we get

$\begin{eqnarray*} |K(\chi, g, f_1, \cdots, f_k;p)| \leq\left(\max(\deg(f_1), \cdots, \deg(f_k))+l\right)\sqrt{p}, \end{eqnarray*}$

where $l$ is the number of distinct roots of $g(x)$ in $\bar{\mathbb{F}}_p$ . □

Lemma 6. Let $\chi$ be a primitive character modulo $p$ , $\chi_{d_i}$ be character modulo $p$ of order $d_i$ . There exist some $1\leq s_i\leq d_i$ with $(s_i, d_i) = 1$ , $i = 1, 2, \ldots, k$ . Then we have

Proof. From the definition of the Dirichlet character modulo $p$ , we can get

$\begin{eqnarray*} &&\sum\limits_{\chi_{d_1}}\cdots \sum\limits_{\chi_{d_k}}\chi_{d_1}(f(x+c_1))\cdots \chi_{d_k}(f(x+c_k))\\ & = &\mathop{{\sum\limits_{s_1 = 1}^{d_1}}\ '}\cdots \mathop{{\sum\limits_{s_k = 1}^{d_k}}\ '} e\left(\frac{s_1\cdot \text{ind}(f(x+c_1))}{d_1}\right)\cdots e\left(\frac{s_k\cdot \text{ind}(f(x+c_k))}{d_k}\right)\\ & = &\mathop{{\sum\limits_{s_1 = 1}^{d_1}}\ '}\cdots \mathop{{\sum\limits_{s_k = 1}^{d_k}}\ '}e\left(\frac{\frac{s_1(p-1)}{d_1}\cdot \text{ind}(f(x+c_1))+\cdots+\frac{s_k(p-1)}{d_k}\cdot \text{ind}(f(x+c_k))}{p-1}\right)\\ & = &\mathop{{\sum\limits_{s_1 = 1}^{d_1}}\ '}\cdots \mathop{{\sum\limits_{s_k = 1}^{d_k}}\ '}e\left(\frac{ \text{ind}(f(x+c_1))^{\frac{s_1(p-1)}{d_1}}+\cdots+ \text{ind}(f(x+c_k))^{\frac{s_k(p-1)}{d_k}}}{p-1}\right)\\ & = &\mathop{{\sum\limits_{s_1 = 1}^{d_1}}\ '}\cdots \mathop{{\sum\limits_{s_k = 1}^{d_k}}\ '}e\left(\frac{ \text{ind}\left((f(x+c_1))^{\frac{s_1(p-1)}{d_1}}\cdots(f(x+c_k))^{\frac{s_k(p-1)}{d_k}}\right)}{p-1}\right)\\ & = &\mathop{{\sum\limits_{s_1 = 1}^{d_1}}\ '}\cdots \mathop{{\sum\limits_{s_k = 1}^{d_k}}\ '}\chi\left(f(x+c_1))^{\frac{s_1(p-1)}{d_1}}\cdots(f(x+c_k))^{\frac{s_k(p-1)}{d_k}}\right), \end{eqnarray*}$

where $\text{ind}(a)$ denotes an index of $a$ with base $g$ of modulo $p$ , and $g$ is a positive primitive root of modulo $p$ . □

3. Proofs of the theorems

Firstly, we prove the Theorem 1. Let $p$ be an odd prime, $k$ be any fixed positive integer. Then for any $k$ different integers $a_1, \ a_2, \ldots, a_k \in {\mathbb{F}}_p$ , from Lemma 1 and the definition of Lehmer number we have

$\begin{eqnarray} &&N(\mathcal{R}, p) = \frac{1}{p}\sum\limits_{b = 0}^{p-1}\mathop{\sum\limits_{g_1 = 1}^{p-1}\sum\limits_{g_2 = 1}^{p-1}\cdots \sum\limits_{g_k = 1}^{p-1}}_{g_1, g_2, \ldots, g_k \in \mathcal{R}}e\left(\frac{b(a_1g_1+\cdots+a_kg_k-c)}{p}\right)\\ & = &\frac{1}{p}\frac{\phi^k(p-1)}{2^k(p-1)^k}\prod\limits_{i = 1}^{k}\left(\sum\limits_{d_i\mid p-1}\frac{\mu\left(d_i\right)}{\phi\left(d_i\right)}\sum\limits_{\chi_{d_i}}\sum\limits_{g_i = 1}^{p-1} \chi_{d_i}(g_i)\left(1-(-1)^{g_i+\overline{g_i}}\right)\right)\\ &&\cdot \sum\limits_{b = 0}^{p-1}e\left(\frac{b(a_1g_1+\cdots+a_kg_k-c)}{p}\right)\\ & = &\frac{1}{p}\frac{\phi^k(p-1)}{2^k(p-1)^k}\prod\limits_{i = 1}^{k}\left(\sum\limits_{d_i\mid p-1}\frac{\mu\left(d_i\right)}{\phi\left(d_i\right)}\sum\limits_{\chi_{d_i}}\sum\limits_{g_i = 1}^{p-1} \chi_{d_i}(g_i)\right)\sum\limits_{b = 0}^{p-1}e\left(\frac{b(a_1g_1+\cdots+a_kg_k-c)}{p}\right)\\ &&+\frac{1}{p}\frac{\phi^k(p-1)}{2^k(p-1)^k}\prod\limits_{i = 1}^{k}\left(\sum\limits_{d_i\mid p-1}\frac{\mu\left(d_i\right)}{\phi\left(d_i\right)}\sum\limits_{\chi_{d_i}}\sum\limits_{g_i = 1}^{p-1} \chi_{d_i}(g_i)\right)\sum\limits_{t = 1}^{k}(-1)^{t}\mathop{\sum\limits_{i_1 = 1}^{k}\sum\limits_{i_2 = 1}^{k}\cdots \sum\limits_{i_t = 1}^{k}}_{i_1 < i_2 < \cdots < i_t}l_{i_1}l_{i_2}\cdots l_{i_t}\\ &&\cdot \sum\limits_{b = 0}^{p-1}e\left(\frac{b(a_1g_1+\cdots+a_kg_k-c)}{p}\right)\\ & = &A_1+A_2, \end{eqnarray}$

(3.1)

where $l_i = (-1)^{g_i+\overline{g_i}}, i = 1, 2, \cdots, k$ .

$\begin{eqnarray} &&A_1 = \frac{1}{p}\frac{\phi^k(p-1)}{2^k(p-1)^k}\prod\limits_{i = 1}^{k}\left(\sum\limits_{d_i\mid p-1}\frac{\mu\left(d_i\right)}{\phi\left(d_i\right)}\sum\limits_{\chi_{d_i}}\sum\limits_{g_i = 1}^{p-1} \chi_{d_i}(g_i)\right)\sum\limits_{b = 0}^{p-1}e\left(\frac{b(a_1g_1+\cdots+a_kg_k-c)}{p}\right)\\ & = &\frac{1}{p}\frac{\phi^k(p-1)}{2^k(p-1)^k}\left[\sum\limits_{g_1 = 1}^{p-1}\cdots \sum\limits_{g_k = 1}^{p-1}\sum\limits_{b = 0}^{p-1}e\left(\frac{b(a_1g_1+\cdots+a_kg_k-c)}{p}\right)\right.\\ &&+\left.\mathop{\sum\limits_{d_1\mid p-1}\cdots \sum\limits_{d_k\mid p-1}}_{d_1\cdots d_k > 1}\frac{\mu\left(d_1\right)}{\phi\left(d_1\right)}\cdots\frac{\mu\left(d_k\right)}{\phi\left(d_k\right)}\sum\limits_{\chi_{d_1}}\cdots\sum\limits_{\chi_{d_k}}\sum\limits_{g_1 = 1}^{p-1}\cdots \sum\limits_{g_k = 1}^{p-1} \chi_{d_1}(g_1)\cdots \chi_{d_k}(g_k)\right.\\ &&\cdot\left.\sum\limits_{b = 0}^{p-1}e\left(\frac{b(a_1g_1+\cdots+a_kg_k-c)}{p}\right)\right]\\ & = &\frac{1}{p}\frac{\phi^k(p-1)}{2^k(p-1)^k}\left[(p-1)^k+(-1)^{k+1}+\mathop{\sum\limits_{d_1\mid p-1}\cdots \sum\limits_{d_k\mid p-1}}_{d_1\cdots d_k > 1}\frac{\mu\left(d_1\right)}{\phi\left(d_1\right)}\cdots\frac{\mu\left(d_k\right)}{\phi\left(d_k\right)}\right.\\ &&\cdot\left.\sum\limits_{\chi_{d_1}}\cdots\sum\limits_{\chi_{d_k}}\sum\limits_{g_1 = 1}^{p-1}\cdots \sum\limits_{g_k = 1}^{p-1} \chi_{d_1}(g_1)\cdots \chi_{d_k}(g_k)\sum\limits_{b = 0}^{p-1}e\left(\frac{b(a_1g_1+\cdots+a_kg_k-c)}{p}\right)\right]. \end{eqnarray}$

(3.2)

From Eq (3.2), let

$\begin{eqnarray*} A_{11}& = &\mathop{\sum\limits_{d_1\mid p-1}\cdots \sum\limits_{d_k\mid p-1}}_{d_1\cdots d_k > 1}\frac{\mu\left(d_1\right)}{\phi\left(d_1\right)}\cdots\frac{\mu\left(d_k\right)}{\phi\left(d_k\right)}\sum\limits_{\chi_{d_1}}\cdots\sum\limits_{\chi_{d_k}}\sum\limits_{g_1 = 1}^{p-1}\cdots \sum\limits_{g_k = 1}^{p-1} \chi_{d_1}(g_1)\cdots \chi_{d_k}(g_k)\\ &&\cdot\sum\limits_{b = 0}^{p-1}e\left(\frac{b(a_1g_1+a_2g_2+\cdots+a_kg_k-c)}{p}\right)\\ & = &\mathop{\sum\limits_{d_1\mid p-1}\cdots \sum\limits_{d_k\mid p-1}}_{d_1\cdots d_k > 1}\frac{\mu\left(d_1\right)}{\phi\left(d_1\right)}\cdots\frac{\mu\left(d_k\right)}{\phi\left(d_k\right)}\sum\limits_{\chi_{d_1}}\cdots\sum\limits_{\chi_{d_k}}\sum\limits_{g_1 = 1}^{p-1}\cdots \sum\limits_{g_k = 1}^{p-1} \chi_{d_1}(g_1)\cdots \chi_{d_k}(g_k)\\ &&+\mathop{\sum\limits_{d_1\mid p-1}\cdots \sum\limits_{d_k\mid p-1}}_{d_1\cdots d_k > 1}\frac{\mu\left(d_1\right)}{\phi\left(d_1\right)}\cdots\frac{\mu\left(d_k\right)}{\phi\left(d_k\right)}\sum\limits_{\chi_{d_1}}\cdots\sum\limits_{\chi_{d_k}}\sum\limits_{b = 1}^{p-1}\sum\limits_{g_1 = 1}^{p-1}\chi_{d_1}(g_1)e\left(\frac{ba_1g_1}{p}\right)\\ &&\cdots \sum\limits_{g_k = 1}^{p-1}\chi_{d_k}(g_k)e\left(\frac{ba_kg_k}{p}\right)e\left(\frac{-bc}{p}\right)\\ & = &\mathop{\sum\limits_{d_1\mid p-1}\cdots \sum\limits_{d_k\mid p-1}}_{d_1\cdots d_k > 1}\frac{\mu\left(d_1\right)}{\phi\left(d_1\right)}\cdots\frac{\mu\left(d_k\right)}{\phi\left(d_k\right)}\sum\limits_{\chi_{d_1}}\cdots\sum\limits_{\chi_{d_k}}\sum\limits_{b = 1}^{p-1}\sum\limits_{g_1 = 1}^{p-1}\chi_{d_1}(g_1)e\left(\frac{ba_1g_1}{p}\right)\\ &&\cdots \sum\limits_{g_k = 1}^{p-1}\chi_{d_k}(g_k)e\left(\frac{ba_kg_k}{p}\right)e\left(\frac{-bc}{p}\right). \end{eqnarray*}$

Using the properties of Gauss sums we can get

$\begin{eqnarray*} \left|A_{11}\right|& = &\left|\mathop{\sum\limits_{d_1\mid p-1}\cdots \sum\limits_{d_k\mid p-1}}_{d_1\cdots d_k > 1}\frac{\mu\left(d_1\right)}{\phi\left(d_1\right)}\cdots\frac{\mu\left(d_k\right)}{\phi\left(d_k\right)}\sum\limits_{\chi_{d_1}}\cdots\sum\limits_{\chi_{d_k}}\sum\limits_{b = 1}^{p-1}\sum\limits_{g_1 = 1}^{p-1}\chi_{d_1}(g_1)e\left(\frac{ba_1g_1}{p}\right)\right.\\ &&\left.\cdots \sum\limits_{g_k = 1}^{p-1}\chi_{d_k}(g_k)e\left(\frac{ba_kg_k}{p}\right)e\left(\frac{-bc}{p}\right)\right|\\ & = &\left|\mathop{\sum\limits_{d_1\mid p-1}}_{d_1 > 1}\cdots \mathop{\sum\limits_{d_k\mid p-1}}_{d_k > 1}\frac{\mu\left(d_1\right)}{\phi\left(d_1\right)}\cdots\frac{\mu\left(d_k\right)}{\phi\left(d_k\right)}\sum\limits_{\chi_{d_1}}\cdots\sum\limits_{\chi_{d_k}}\sum\limits_{b = 1}^{p-1}\sum\limits_{g_1 = 1}^{p-1}\chi_{d_1}(g_1)e\left(\frac{ba_1g_1}{p}\right)\right.\\ &&\cdots \sum\limits_{g_k = 1}^{p-1}\chi_{d_k}(g_k)e\left(\frac{ba_kg_k}{p}\right)e\left(\frac{-bc}{p}\right)\\ &&+\mathop{\sum\limits_{d_1\mid p-1}}_{d_1 > 1}\cdots \mathop{\sum\limits_{d_{k-1}\mid p-1}}_{d_{k-1} > 1}\frac{\mu\left(d_1\right)}{\phi\left(d_1\right)}\cdots\frac{\mu\left(d_{k-1}\right)}{\phi\left(d_{k-1}\right)}\sum\limits_{\chi_{d_1}}\cdots\sum\limits_{\chi_{d_{k-1}}}\sum\limits_{b = 1}^{p-1}\sum\limits_{g_1 = 1}^{p-1}\chi_{d_1}(g_1)e\left(\frac{ba_1g_1}{p}\right)\\ &&\cdots \sum\limits_{g_{k-1} = 1}^{p-1}\chi_{d_{k-1}}(g_{k-1})e\left(\frac{ba_{k-1}g_{k-1}}{p}\right)\sum\limits_{g_k = 1}^{p-1}e\left(\frac{ba_kg_k}{p}\right)e\left(\frac{-bc}{p}\right)\\ &&+\cdots+\mathop{\sum\limits_{d_1\mid p-1}}_{d_1 > 1}\frac{\mu\left(d_1\right)}{\phi\left(d_1\right)}\sum\limits_{\chi_{d_1}}\sum\limits_{b = 1}^{p-1}\sum\limits_{g_1 = 1}^{p-1}\chi_{d_1}(g_1)e\left(\frac{ba_1g_1}{p}\right)\sum\limits_{g_2 = 1}^{p-1}e\left(\frac{ba_2g_2}{p}\right)\\ &&\left.\cdots \sum\limits_{g_k = 1}^{p-1}e\left(\frac{ba_kg_k}{p}\right)e\left(\frac{-bc}{p}\right)\right|\\ &\ll&2^{k\omega(p-1)}p^{\frac{k+1}{2}}, \end{eqnarray*}$

where we have used the fact that $\sum_{d|n}|\mu(d)| = 2^{\omega(n)}$ .

Hence, Eq (3.2) and the above formulae yield that

$\begin{eqnarray} A_1 = \frac{\phi^k(p-1)}{2^kp}+\mathrm{O}\left(\frac{\phi^k(p-1)}{p^{\frac{k+1}{2}}}2^{k\omega(p-1)}\right). \end{eqnarray}$

(3.3)

Then we compute $A_2$ in Eq (3.1). For simplicity, let

$U_m(u) = \sum\limits_{u = 1}^{p-1}(-1)^ue\left(\frac{-mu}{p}\right),$

noting that

$\sum\limits_{u = 1}^{p-1}(-1)^ue\left(\frac{-mu}{p}\right) = \frac{1-e(\frac{m}{p})}{1+e(\frac{m}{p})} = \frac{i\sin (\pi m/p)}{\cos(\pi m/p)},$

$\sum\limits_{m = 1}^{p-1}\left|\frac{\sin (\pi m/p)}{\cos(\pi m/p)}\right| = T_pp\ln p.$

Hence,

$\begin{eqnarray} \left|\sum\limits_{m = 1}^{p-1}U_m(u)\right|\leq\sum\limits_{m = 1}^{p-1}\left|\sum\limits_{u = 1}^{p-1}(-1)^ue\left(\frac{-mu}{p}\right)\right| = T_pp\ln p. \end{eqnarray}$

(3.4)

Noting that, if $m = 0$ , then $\sum_{u = 1}^{p-1}(-1)^ue\left(\frac{-mu}{p}\right) = \sum_{u = 1}^{p-1}(-1)^u = 0$ , since $p$ is odd. Hence,

$\begin{eqnarray} &&l_i = (-1)^{g_i+\overline{g_i}}\\ & = &\frac{1}{p}\sum\limits_{m_i = 0}^{p-1}\sum\limits_{u_i = 1}^{p-1}(-1)^{u_i}e\left(\frac{m_i(g_i-u_i)}{p}\right)\cdot \frac{1}{p}\sum\limits_{n_i = 0}^{p-1}\sum\limits_{v_i = 1}^{p-1}(-1)^{v_i}e\left(\frac{n_i(\overline{g_i}-v_i)}{p}\right)\\ & = &\frac{1}{p^2}\sum\limits_{m_i, n_i = 0}^{p-1}e\left(\frac{m_ig_i+n_i\overline{g_i}}{p}\right)\sum\limits_{u_i = 1}^{p-1}(-1)^{u_i}e\left(\frac{-m_iu_i}{p}\right)\sum\limits_{v_i = 1}^{p-1}(-1)^{v_i}e\left(\frac{-n_iv_i}{p}\right)\\ & = &\frac{1}{p^2}\sum\limits_{m_i, n_i = 1}^{p-1}e\left(\frac{m_ig_i+n_i\overline{g_i}}{p}\right)U_{m_i}(u_i)U_{n_i}(v_i). \end{eqnarray}$

(3.5)

From the above discussion and Eq (3.1), we can obtain

$\begin{eqnarray} \left|A_2\right|& = &\left|\frac{1}{p}\frac{\phi^k(p-1)}{2^k(p-1)^k}\prod\limits_{i = 1}^{k}\left(\sum\limits_{d_i\mid p-1}\frac{\mu\left(d_i\right)}{\phi\left(d_i\right)}\sum\limits_{\chi_{d_i}}\sum\limits_{g_i = 1}^{p-1} \chi_{d_i}(g_i)\right)\sum\limits_{t = 1}^{k}(-1)^{t}\mathop{\sum\limits_{i_1 = 1}^{k}\cdots \sum\limits_{i_t = 1}^{k}}_{i_1 < \cdots < i_t}l_{i_1}\cdots l_{i_t}\right.\\ &&\left.\cdot \sum\limits_{b = 0}^{p-1}e\left(\frac{b(a_1g_1+a_2g_2+\cdots+a_kg_k-c)}{p}\right)\right|\\ &\leq&\frac{1}{p}\frac{\phi^k(p-1)}{2^k(p-1)^k}\sum\limits_{t = 1}^{k}\binom{k}{t}T^{2t}_p\ln^{2t}p\sum\limits_{d_1\mid p-1}\cdots \sum\limits_{d_k\mid p-1}\frac{|\mu\left(d_1\right)|}{\phi\left(d_1\right)}\cdots\frac{|\mu\left(d_k\right)|}{\phi\left(d_k\right)}\\ &&\sum\limits_{\chi_{d_1}}\cdots\sum\limits_{\chi_{d_k}}\left|\sum\limits_{b = 0}^{p-1}\sum\limits_{g_1 = 1}^{p-1}\cdots\sum\limits_{g_k = 1}^{p-1}\chi_{d_1}(g_1)\cdots \chi_{d_k}(g_k)\right.\\ &&\left.\cdot e\left(\frac{m_1g_1+n_1\overline{g_1}+\cdots+m_tg_t+n_t\overline{g_t}}{p}\right)e\left(\frac{b(a_1g_1+\cdots+a_kg_k-c)}{p}\right)\right|\\ & = &\frac{1}{p}\frac{\phi^k(p-1)}{2^k(p-1)^k}\sum\limits_{t = 1}^{k}\binom{k}{t}T^{2t}_p\ln^{2t}p\left[\mathop{\sum\limits_{d_1\mid p-1}}_{d_1 > 1}\cdots \mathop{\sum\limits_{d_k\mid p-1}}_{d_k > 1}\frac{|\mu\left(d_1\right)|}{\phi\left(d_1\right)}\cdots\frac{|\mu\left(d_k\right)|}{\phi\left(d_k\right)}\right.\\ &&\sum\limits_{\chi_{d_1}}\cdots\sum\limits_{\chi_{d_k}}\left|\sum\limits_{b = 0}^{p-1}\sum\limits_{g_1 = 1}^{p-1}\cdots\sum\limits_{g_k = 1}^{p-1}\chi_{d_1}(g_1)\cdots \chi_{d_k}(g_k)\right.\\ &&\left.\cdot e\left(\frac{m_1g_1+n_1\overline{g_1}+\cdots+m_tg_t+n_t\overline{g_t}}{p}\right)e\left(\frac{b(a_1g_1+\cdots+a_kg_k-c)}{p}\right)\right|\\ &&+\mathop{\sum\limits_{d_1\mid p-1}}_{d_1 > 1}\cdots \mathop{\sum\limits_{d_{k-1}\mid p-1}}_{d_{k-1} > 1}\frac{|\mu\left(d_1\right)|}{\phi\left(d_1\right)}\cdots\frac{|\mu\left(d_{k-1}\right)|}{\phi\left(d_{k-1}\right)}\\ &&\sum\limits_{\chi_{d_1}}\cdots\sum\limits_{\chi_{d_{k-1}}}\left|\sum\limits_{b = 0}^{p-1}\sum\limits_{g_1 = 1}^{p-1}\cdots\sum\limits_{g_k = 1}^{p-1}\chi_{d_1}(g_1)\cdots \chi_{d_{k-1}}(g_{k-1})\right.\\ &&\left.\cdot e\left(\frac{m_1g_1+n_1\overline{g_1}+\cdots+m_tg_t+n_t\overline{g_t}}{p}\right)e\left(\frac{b(a_1g_1+\cdots+a_kg_k-c)}{p}\right)\right|\\ &&+\cdots+\mathop{\sum\limits_{d_1\mid p-1}}_{d_1 > 1}\frac{|\mu\left(d_1\right)|}{\phi\left(d_1\right)}\sum\limits_{\chi_{d_1}}\left|\sum\limits_{b = 0}^{p-1}\sum\limits_{g_1 = 1}^{p-1}\cdots\sum\limits_{g_k = 1}^{p-1}\chi_{d_1}(g_1)\right.\\ &&\left.\cdot e\left(\frac{m_1g_1+n_1\overline{g_1}+\cdots+m_tg_t+n_t\overline{g_t}}{p}\right)e\left(\frac{b(a_1g_1+\cdots+a_kg_k-c)}{p}\right)\right|\\ &&+\left|\sum\limits_{b = 0}^{p-1}\sum\limits_{g_1 = 1}^{p-1}\cdots\sum\limits_{g_k = 1}^{p-1}e\left(\frac{m_1g_1+n_1\overline{g_1}+\cdots+m_tg_t+n_t\overline{g_t}}{p}\right)\right.\\ &&\left.\left.\cdot e\left(\frac{b(a_1g_1+\cdots+a_kg_k-c)}{p}\right)\right|\right]. \end{eqnarray}$

(3.6)

Summing the above formula for $t$ from 1 to $k$ , then the last term of Eq (3.6) is

$\begin{eqnarray*} &&\frac{1}{p}\frac{\phi^k(p-1)}{2^k(p-1)^k}\sum\limits_{t = 1}^{k}\binom{k}{t}T^{2t}_p\ln^{2t}p\left|\sum\limits_{b = 0}^{p-1}\sum\limits_{g_1 = 1}^{p-1}\cdots\sum\limits_{g_k = 1}^{p-1}e\left(\frac{m_1g_1+n_1\overline{g_1}+\cdots+m_tg_t+n_t\overline{g_t}}{p}\right)\right.\\ &&\left.\cdot e\left(\frac{b(a_1g_1+\cdots+a_kg_k-c)}{p}\right)\right|\\ & = &\frac{1}{p}\frac{\phi^k(p-1)}{2^k(p-1)^k}\left[kT^2_p\ln^{2}p\left|\sum\limits_{b = 0}^{p-1}\sum\limits_{g_1 = 1}^{p-1}\cdots\sum\limits_{g_k = 1}^{p-1}e\left(\frac{m_1g_1+n_1\overline{g_1}}{p}\right)e\left(\frac{b(a_1g_1+\cdots+a_kg_k-c)}{p}\right)\right|\right.\\ &&+\cdots+\binom{k}{k-1}T^{2(k-1)}_p\ln^{2(k-1)}p\cdot\left|\sum\limits_{b = 0}^{p-1}\sum\limits_{g_1 = 1}^{p-1}\cdots\sum\limits_{g_k = 1}^{p-1}\right.\\ &&\left.\cdot e\left(\frac{m_1g_1+n_1\overline{g_1}+\cdots+m_{k-1}g_{k-1}+n_{k-1}\overline{g_{k-1}}}{p}\right)e\left(\frac{b(a_1g_1+\cdots+a_kg_k-c)}{p}\right)\right|\\ &&+T^{2k}_p\ln^{2k}p\left|\sum\limits_{b = 0}^{p-1}\sum\limits_{g_1 = 1}^{p-1}\cdots\sum\limits_{g_k = 1}^{p-1}e\left(\frac{m_1g_1+n_1\overline{g_1}+\cdots+m_kg_k+n_k\overline{g_k}}{p}\right)\right.\\ &&\left.\left.\cdot e\left(\frac{b(a_1g_1+\cdots+a_kg_k-c)}{p}\right)\right|\right]\\ &\ll&\frac{\phi^k(p-1)}{p^{k+1}}\ln^{2k}p\left(p^{k-\frac{1}{2}}+\cdots+p^{\frac{k+1}{2}}\right)\\ &\ll&\frac{\phi^k(p-1)}{p^{k+1}}\ln^{2k}p\cdot p^{k-\frac{1}{2}} = \frac{\phi^k(p-1)}{p^{\frac{3}{2}}}\ln^{2k}p, \end{eqnarray*}$

here we have utilized $T^2_p < \frac{4}{\pi^2}\left(1+\frac{1.549}{\ln p}\right)^2 < 2.4$ and the results in Wang and Wang (see Lemma 2.2 of ^[13]) that

$\begin{eqnarray*} \left|\sum\limits_{a = 1}^{p-1}\chi_d(a)e\left(\frac{ma+n\overline{a}}{p}\right)\right|\ll p^{\frac{1}{2}}. \end{eqnarray*}$

Similarly, note that $\sum_{d|n}|\mu(d)| = 2^{\omega(n)}$ and we can get the estimate of the other terms of Eq (3.6). Then we have

$\begin{eqnarray} A_2&\ll&\frac{\phi^k(p-1)}{p^\frac{3}{2}}2^{k\omega(p-1)}\ln^{2k}p. \end{eqnarray}$

(3.7)

Inserting Eqs (3.3) and (3.7) into (3.1), we can deduce that

$\begin{eqnarray*} N(\mathcal{R}, p)& = &\frac{\phi^k(p-1)}{2^kp}+\mathrm{O}\left(\frac{\phi^k(p-1)}{p^{\frac{k+1}{2}}}2^{k\omega(p-1)}\right)+\mathrm{O}\left(\frac{\phi^k(p-1)}{p^\frac{3}{2}}2^{k\omega(p-1)}\ln^{2k}p\right)\\ & = &\frac{\phi^k(p-1)}{2^kp}+\mathrm{O}\left(\frac{\phi^k(p-1)}{p^\frac{3}{2}}2^{k\omega(p-1)}\ln^{2k}p\right). \end{eqnarray*}$

This proves the Theorem 1.

Now we prove the Theorem 2. Let $\mathcal{A}$ denote the set of integers $1\leq x\leq p$ such that

$\begin{eqnarray*} \prod\limits_{i = 1}^{k}f(x+c_i)\equiv 0 (\bmod p). \end{eqnarray*}$

By the definition of primitive roots and Lehmer number, it follows that

$\begin{eqnarray} &&M(f(x), \mathcal{R}, p)\\ & = &\frac{1}{2^k}\frac{\phi^k(p-1)}{(p-1)^k}\prod\limits_{i = 1}^{k}\left(\sum\limits_{d_i\mid p-1}\frac{\mu\left(d_i\right)}{\phi\left(d_i\right)}\sum\limits_{\chi_{d_i}}\mathop{\sum\limits_{x = 1}^{p}}_{x\notin \mathcal{A}} \chi_{d_i}(f(x+c_i))\left(1-(-1)^{f(x+c_i)+\overline{f(x+c_i)}}\right)\right)\\ & = &\frac{1}{2^k}\frac{\phi^k(p-1)}{(p-1)^k}\prod\limits_{i = 1}^{k}\left(\sum\limits_{d_i\mid p-1}\frac{\mu\left(d_i\right)}{\phi\left(d_i\right)}\sum\limits_{\chi_{d_i}}\mathop{\sum\limits_{x = 1}^{p}}_{x\notin \mathcal{A}} \chi_{d_i}(f(x+c_i))\right)\\ &&+\frac{1}{2^k}\frac{\phi^k(p-1)}{(p-1)^k}\prod\limits_{i = 1}^{k}\left(\sum\limits_{d_i\mid p-1}\frac{\mu\left(d_i\right)}{\phi\left(d_i\right)}\sum\limits_{\chi_{d_i}}\mathop{\sum\limits_{x = 1}^{p}}_{x\notin \mathcal{A}} \chi_{d_i}(f(x+c_i))\right) \sum\limits_{t = 1}^{k}(-1)^{t}\mathop{\sum\limits_{i_1 = 1}^{k}\cdots \sum\limits_{i_t = 1}^{k}}_{i_1 < \cdots < i_t}g_{i_1}\cdots g_{i_t}\\ & = &\frac{1}{2^k}\frac{\phi^k(p-1)}{(p-1)^k}\left(B_1+B_2\right), \end{eqnarray}$

(3.8)

where $g_i = (-1)^{f(x+c_i)+\overline{f(x+c_i)}}, i = 1, 2, \ldots, k$ .

$\begin{eqnarray*} &&B_1 = \prod\limits_{i = 1}^{k}\left(\sum\limits_{d_i\mid p-1}\frac{\mu\left(d_i\right)}{\phi\left(d_i\right)}\sum\limits_{\chi_{d_i}}\mathop{\sum\limits_{x = 1}^{p}}_{x\notin \mathcal{A}} \chi_{d_i}(f(x+c_i))\right)\\ & = &\mathop{\sum\limits_{x = 1}^{p}}_{x\notin \mathcal{A}}1 +\prod\limits_{i = 1}^{k}\left(\mathop{\sum\limits_{d_i\mid p-1}}_{\prod\limits_{i = 1}^{k}d_i > 1}\frac{\mu\left(d_i\right)}{\phi\left(d_i\right)}\sum\limits_{\chi_{d_i}}\mathop{\sum\limits_{x = 1}^{p}}_{x\notin \mathcal{A}} \chi_{d_i}(f(x+c_i))\right). \end{eqnarray*}$

Obviously,

$\begin{eqnarray*} \left|\mathop{\sum\limits_{x = 1}^{p}}_{x\notin \mathcal{A}}1-p\right|\leq kl. \end{eqnarray*}$

From Lemma 6 we have

$\begin{eqnarray*} &&\sum\limits_{\chi_{d_1}}\sum\limits_{\chi_{d_2}}\cdots \sum\limits_{\chi_{d_k}}\mathop{\sum\limits_{x = 1}^{p}}_{x\notin \mathcal{A}} \chi_{d_1}(f(x+c_1)) \chi_{d_2}(f(x+c_2))\cdots \chi_{d_k}(f(x+c_k))\\ & = &\mathop{{\sum\limits_{s_1 = 1}^{d_1}}\ '}\cdots \mathop{{\sum\limits_{s_k = 1}^{d_k}}\ '}\mathop{\sum\limits_{x = 1}^{p}}_{x\notin \mathcal{A}} \chi\left((f(x+c_1))^{\frac{s_1(p-1)}{d_1}}\cdots(f(x+c_k))^{\frac{s_k(p-1)}{d_k}}\right). \end{eqnarray*}$

Due to $d_1d_2\cdots d_k > 1$ , and

$\frac{s_i(p-1)}{d_i} < p-1\ \text{for}\ d_i > 1 (i = 1, 2, \ldots, k),$

from Lemma 4 we can get that the polynomial

$\begin{eqnarray*} (f(x+c_1))^{\frac{s_1(p-1)}{d_1}}\cdots(f(x+c_k))^{\frac{s_k(p-1)}{d_k}} \end{eqnarray*}$

has a root in $\bar{\mathbb{F}}_p$ with multiples less than $p-1$ , thus it can not be multiple of a $(p-1)$ -th power of polynomial over $\mathbb{F}_p$ . Take $g(x) = (f(x+c_1))^{\frac{s_1(p-1)}{d_1}}\cdots(f(x+c_k))^{\frac{s_k(p-1)}{d_k}}$ , in Lemma 2 we have

$\begin{eqnarray*} \left|\mathop{\sum\limits_{x = 1}^{p}}_{x\notin \mathcal{A}}\chi\left(f(x+c_1)^{\frac{s_1(p-1)}{d_1}}\cdots f(x+c_k)^{\frac{s_k(p-1)}{d_k}}\right)\right| < (kl-1)p^{\frac{1}{2}}. \end{eqnarray*}$

Hence, we have

$\begin{eqnarray} &&\left|B_1-(p-kl)\right| < (2^{k\omega(p-1)}-1)(kl-1)p^{\frac{1}{2}}\leq 2^{k\omega(p-1)}(kl-1)p^{\frac{1}{2}}. \end{eqnarray}$

(3.9)

Using the methods in the proof of Theorem 1 we have

$\begin{eqnarray*} &&g_i = \frac{1}{p^2}\sum\limits_{m_i, n_i = 1}^{p-1}e\left(\frac{m_i(f(x+c_i))+n_i\overline{f(x+c_i)}}{p}\right)U_{m_i}(u_i)U_{n_i}(v_i). \end{eqnarray*}$

From the above discussion and Lemma 5, we can obtain

$\begin{eqnarray} \left|B_2\right|& < &\left|\prod\limits_{i = 1}^{k}\left(\sum\limits_{d_i\mid p-1}\frac{\mu\left(d_i\right)}{\phi\left(d_i\right)}\sum\limits_{\chi_{d_i}}\mathop{\sum\limits_{x = 1}^{p}}_{x\notin \mathcal{A}} \chi_{d_i}(f(x+c_i))\right) \sum\limits_{t = 1}^{k}(-1)^{t}\mathop{\sum\limits_{i_1 = 1}^{k}\sum\limits_{i_2 = 1}^{k}\cdots \sum\limits_{i_t = 1}^{k}}_{i_1 < i_2 < \cdots < i_t}g_{i_1}g_{i_2}\cdots g_{i_t}\right|\\ & < &\prod\limits_{i = 1}^{k}\left(\sum\limits_{d_i\mid p-1}\frac{\left|\mu\left(d_i\right)\right|}{\phi\left(d_i\right)}\sum\limits_{\chi_{d_i}}\right)\sum\limits_{t = 1}^{k}\binom{k}{t}T^{2t}_p\ln^{2t}p\cdot\left|\mathop{\sum\limits_{x = 1}^{p}}_{x\notin \mathcal{A}} \chi_{d_i}(f(x+c_i))\right.\\ &&\left.\cdot e\left(\frac{m_1(f(x+c_1))+n_1\overline{(f(x+c_1))}+\cdots+m_t(f(x+c_t))+n_t\overline{(f(x+c_t))}}{p}\right)\right|\\ & < & 2^{k\omega(p-1)}\cdot\sum\limits_{t = 1}^{k}\binom{k}{t}T^{2t}_p\ln^{2t}p(kl+l)p^{\frac{1}{2}}. \end{eqnarray}$

(3.10)

Combing Eqs (3.8), (3.9) and (3.10) we have

$\begin{eqnarray} &&\left|M(f(x), \mathcal{R}, p)-\frac{1}{2^k}\frac{\phi^k(p-1)}{(p-1)^k}(p-kl)\right|\\ & < &\frac{1}{2^k}\frac{\phi^k(p-1)}{(p-1)^k}\left[2^{k\omega(p-1)}(kl-1)p^{\frac{1}{2}}+2^{k\omega(p-1)}\cdot\sum\limits_{t = 1}^{k}\binom{k}{t}T^{2t}_p\ln^{2t}p(kl+l)p^{\frac{1}{2}}\right]\\ & = &\frac{1}{2^k}\frac{\phi^k(p-1)}{(p-1)^k}2^{k\omega(p-1)}p^{\frac{1}{2}}\cdot\left[(kl-1)+((k+1)l)\sum\limits_{t = 1}^{k}\binom{k}{t}T^{2t}_p\ln^{2t}p\right]. \end{eqnarray}$

(3.11)

Then we have

$\begin{eqnarray*} &&M(f(x), \mathcal{R}, p) = \frac{1}{2^k}\frac{\phi^k(p-1)}{(p-1)^{k-1}}+\mathrm{O}\left(\frac{\phi^k(p-1)}{p^{k-\frac{1}{2}}}2^{k\omega(p-1)}\ln^{2k}p\right). \end{eqnarray*}$

This complete the proof of Theorem 2.

4. Conclusions

From two perspectives, this paper consider the distribution of $LPRs$ that are related to the generalized Golomb's conjecture. Theorem 1 extends the binary linear equation $ag_1+bg_2 = c$ to the multivariate linear equation $a_1g_1+a_2g_2+\cdots+a_kg_k = c$ , and uses the properties of Gauss sums to derive an asymptotic formula for the number of its solutions $g_1, g_2, \ldots, g_k$ that are $LPRs$ . Theorem 2 considers $k$ consecutive $LPRs$ and employs the upper bound estimation of the generalized Kloosterman sums to provide a more general result that for $f(x)\in {\mathbb{F}}_p[x]$ , $k$ polynomials $f(x+c_1), f(x+c_2), \ldots, f(x+c_k)$ are Lehmer primitive roots modulo $p$ .

Use of AI tools declaration

The authors declare they have not used Artificial Intelligence (AI) tools in the creation of this article.

Acknowledgments

The author gratefully appreciates the referees and academic editor for their helpful and detailed comments.

This work is supported by the N. S. F. (12126357) of P. R. China and the Natural Science Basic Research Plan in Shaanxi Province of China (2023-JC-QN-0050).

Conflict of interest

The author declare there are no conflicts of interest.

References

[1]	A. F. Bekhit, Introduction to computer vision, in Computer Vision and Augmented Reality in iOS, 1 (2022), 1−20. https://doi.org/10.1007/978-1-4842-7462-0_1
[2]	J. Han, L. Shao, D. Xu, J. Shotton, Enhanced computer vision with microsoft kinect sensor: a review, IEEE Trans. Cybern., 43 (2013), 1318−1334. https://doi.org/10.1109/TCYB.2013.2265378 doi: 10.1109/TCYB.2013.2265378
[3]	Y. O. Sharrab, I. Alsmadi, N. J. Sarhan, Towards the availability of video communication in artificial intelligence-based computer vision systems utilizing a multi-objective function, Cluster Comput., 25 (2022), 231−247. https://doi.org/10.1007/s10586-021-03391-4 doi: 10.1007/s10586-021-03391-4
[4]	N. Haering, P. L. Venetianer, A. Lipton, The evolution of video surveillance:an overview, Mach. Vision Appl., 19 (2008), 279−290. https://doi.org/10.1007/s00138-008-0152-0 doi: 10.1007/s00138-008-0152-0
[5]	Y. Zhang, H. Liu, Constraints and countermeasures of the new situation of population on the future development of higher vocational education−based on the analysis of the seventh national population survey (in Chinese), Educ. Vocation, 6 (2022), 12−20. https://doi.org/10.13615/j.cnki.1004-3985.2022.06.016 doi: 10.13615/j.cnki.1004-3985.2022.06.016
[6]	P. Li, Y. Hu, X. Wu, R. He, Z. Sun, Deep label refinement forage estimation, Pattern Recognit., 100 (2020), 107178. https://doi.org/10.1016/j.patcog.2019.107178 doi: 10.1016/j.patcog.2019.107178
[7]	Y. Yu, K. Tang, Y. Liu, A fine-tuning based approach for daily activity recognition between smart homes, Appl. Sci., 13 (2023). https://doi.org/10.3390/app13095706 doi: 10.3390/app13095706
[8]	Z. Li, F. Liu, W. Yang, S. Peng, J. Zhou, A survey of convolutional neural networks: analysis, applications, and prospects, IEEE Trans. Neural Networks Learn. Syst., 33 (2022), 6999−7019. https://doi.org/10.1109/TNNLS.2021.3084827 doi: 10.1109/TNNLS.2021.3084827
[9]	O. Guehairia, A. Ouamane, F. Dornaika, A. Taleb-Ahmed, Deep random forest for facial age estimation based on face images, in 2020 1st International Conference on Communications, Control Systems and Signal Processing (CCSSP), IEEE, (2020), 305−309. https://doi.org/10.1109/CCSSP49278.2020.9151621
[10]	M. M. Badr, A. M. Sarhan, R. M. Elbasiony, ICRL: using landmark ratios with cascade model for an accurate age estimation system using deep neural networks, J. Intell. Fuzzy Syst., 43 (2022), 72−79. https://doi.org/10.3233/JIFS-211267 doi: 10.3233/JIFS-211267
[11]	B. Zhang, Y. Bao, Age estimation of faces in videos using head pose estimation and convolutional neural networks, Sensors, 22 (2022), 4171. https://doi.org/10.3390/s22114171 doi: 10.3390/s22114171
[12]	S. Pramanik, H. A. B. Dahlan, Face age estimation using shortcut identity connection of convolutional neural network, Int. J. Adv. Comput. Sci. Appl., 13 (2022), 515−521. https://doi.org/10.14569/IJACSA.2022.0130459 doi: 10.14569/IJACSA.2022.0130459
[13]	K. Y. Chang, C. S. Chen, Y. P. Hung, Ordinal hyperplanes ranker with cost sensitivities for age estimation, in CVPR 2011, IEEE, (2011), 585−592. https://doi.org/10.1109/CVPR.2011.5995437
[14]	W. Wang, T. Ishikawa, H. Watanabe, Facial age estimation by curriculum learning, in 2020 IEEE 9th Global Conference on Consumer Electronics (GCCE), IEEE, (2020), 138−139. https://doi.org/10.1109/GCCE50665.2020.9291929
[15]	G. L. Santos, P. T. Endo, K. H. de Carvalho Monteiro, E. da Silva Rocha, I. Silva, T. Lynn, Accelerometer-based human fall detection using convolutional neural networks, Sensors, 19 (2019), 1644. https://doi.org/10.3390/s19071644 doi: 10.3390/s19071644
[16]	Z. Niu, M. Zhou, L. Wang, X. Gao, G. Hua, Ordinal regression with multiple output cnn for age estimation, in 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), (2016), 4920−4928. https://doi.org/10.1109/CVPR.2016.532
[17]	A. Schmeling, G. Geserick, W. Reisinger, A. Olze, Age estimation, Forensic Sci. Int., 165 (2007), 178−181. https://doi.org/10.1016/j.forsciint.2006.05.016 doi: 10.1016/j.forsciint.2006.05.016
[18]	K. He, X. Zhang, S. Ren, J. Sun, Deep residual learning for image recognition, in 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), (2016), 770−778. https://doi.org/10.1109/CVPR.2016.90
[19]	G. Huang, Z. Liu, L. van der Maaten, K. Q. Weinberger, Densely connected convolutional networks, in 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), (2017), 4700−4708. https://doi.org/10.1109/CVPR.2017.243
[20]	O. Agbo-Ajala, S. Viriri, Deep learning approach for facial age classification:a survey of the state-of-the-art, Artif. Intell. Rev., 54 (2021), 179−213. https://doi.org/10.1007/s10462-020-09855-0 doi: 10.1007/s10462-020-09855-0
[21]	Y. Ma, Y. Tao, Y. Gong, W. Cui, B. Wang, Driver identification and fatigue detection algorithm based on deep learning, Math. Biosci. Eng., 20 (2023), 8162−8189. https://doi.org/10.3934/mbe.2023355 doi: 10.3934/mbe.2023355

Reader Comments

Your name:*

Email:*
© 2023 the Author(s), licensee AIMS Press. This is an open access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/4.0)