1. Introduction
In this paper, our primary focus is on Newton–Maclaurin-type inequalities pertaining to linear combinations of elementary symmetric functions. First, define
\begin{equation*} \sigma_k({\boldsymbol x})={\displaystyle\sum_{1\leqslant i_{1} \lt \cdots \lt i_{k}\leqslant {n}}}x_ {i_{1}}\cdots x_ {i_{k}}\end{equation*} which is a k-th primitive symmetric function and the variables
${\boldsymbol
x}=(x_1,x_2,\cdots,x_{n})\in \mathbb{R}^n$. For example, when n = 3,
It is convenient to define
$\sigma_0({\boldsymbol x})=1$, and define
$\sigma_k({\boldsymbol x})=0$ if k < 0 or k > n. Furthermore, define a k-th primitive symmetric mean as
\begin{equation*}E_k({\boldsymbol x})=\displaystyle\frac{\sigma_k({\boldsymbol x})}{C_{n}^k},\quad k=1,2,\cdots,{n},\end{equation*} where
$C_{n}^k=\displaystyle\frac{n!}{k!(n-k)!}$.
The classical Newton [Reference Newton17]–Maclaurin [Reference Maclaurin15] inequalities are
where
$\Gamma_{k}$ is the Gårding’s cone
All of the above inequalities are strict unless
$x_1=\cdots=x_n$, or both sides are 0 at the same time. Over the years, numerous scholars have contributed numerous generalizations of Newton–Maclaurin-type inequalities, including those presented in [Reference Ellard and Smigoc3, Reference Menon16, Reference Niculescu18, Reference Rosset21].
The Newton–Maclaurin inequalities play important roles in deriving theoretical results for fully nonlinear partial differential equations and geometric analysis. There are many important results that need to use the Newton–Maclaurin inequalities, such as [Reference Guan and Guan5–Reference Guan, Ma and Zhou7, Reference Huisken11], etc. This is because the following k-Hessian equations and curvature equations
are central studies in the field of fully nonlinear partial differential equations and geometric analysis. Their left-hand side k-Hessian operator σk is the primary symmetric function about the eigenvalues
$\lambda=(\lambda_1,\cdots,\lambda_n)$ of the Hessian matrix
$(u_{ij})$ or the principal curvatures
$\kappa=(\kappa_1,\cdots,\kappa_n)$ of the hypersurface.
In recent years, the fully non-linear equations derived from linear combinations of primary symmetric functions have received increasing attention. For example, the following special Lagrangian equation
\begin{equation*}{\rm Imdet}(\delta_{ij}+{\rm i}u_{ij})=\displaystyle\sum_{k=0}^{[(n-1)/2]}(-1)^{k}\sigma_{2k+1}
(\lambda(u_{ij}))=0\end{equation*}was derived by Harvey and Lawson in their study of the minimal submanifold problem [Reference Harvey and Lawson10]. The fully nonlinear partial differential equations
\begin{equation*}P_m(u_{ij})=\sum_{k=0}^{m-1}(l_k^+)^{m-k}(x)\sigma_k(u_{ij})=g^{m-1}(x)
\end{equation*}were studied by Krylov [Reference Krylov12] and Dong [Reference Dong1]. The following curvature equations and complex Hessian equations
\begin{equation*}\displaystyle\sum_{s=0}^k\alpha_s\sigma_k(\kappa(\chi))=f(\chi,\upsilon(\chi)),\quad
\chi\in M,\end{equation*}
\begin{equation*}\displaystyle\sum_{s=0}^k\alpha_s\sigma_k(\lambda)=f(z,u,Du)\end{equation*}were studied by Li–Ren–Wang [Reference Li, Ren and Wang13] and Dong [Reference Dong2]. For the following curvature equations
\begin{equation*}\sigma_k(W_u(x))+\alpha
\sigma_{k-1}(W_u(x))=\displaystyle\sum_{l=0}^{k-2}\alpha_l(x)\sigma_l(W_u(x)),\quad
x\in\mathbb{S}^n,\end{equation*} Guan–Zhang [Reference Guan and Zhang8] and Zhou [Reference Zhou23] studied the curvature estimates and the interior gradient estimates respectively. Recently, Liu and Ren [Reference Liu and Ren14] discussed the Pogorelov-type C 2 estimation for
$(k-1)$-convex and k-convex solutions of the following Sum Hessian equations
The Newton–Maclaurin inequalities can be perceived as an intrinsic characteristic of the k-Hessian operator. If the operators situated on the left-hand side of the aforementioned equations can satisfy a corresponding Newton–Maclaurin style inequality, it will naturally facilitate further research on the equation. A natural question is whether the Newton–Maclaurin-type inequalities still hold for the operators of linear combinations of these primary symmetric functions? In a recent paper, we obtained the following Newton–Maclaurin-type inequalities for operators
$E_k+\alpha E_{k-1}$ in [Reference Ren19]:
For any real number
$\alpha\in\mathbb{R}$ and any
$\boldsymbol{x}=(x_{1},x_{2},\cdots,x_{n})\in\mathbb{R}^{n},$
Further, if
$\alpha\geqslant0$ and
then
\begin{equation*}[E_{1}(\boldsymbol x)+\alpha ]\geqslant [E_{2}(\boldsymbol x)+\alpha E_{1}(\boldsymbol x)]^{\frac{1}{2}}\geqslant\cdots\geqslant[E_{k}(\boldsymbol x)+\alpha E_{k-1}(\boldsymbol x)]^{\frac{1}{k}}.\end{equation*} A natural question is that if the Newton–Maclaurin-type inequality still holds for the functions which obtained by linear combination of general primary symmestric functions. That is, for
$\alpha=(\alpha_1,\cdots,\alpha_n)\in \mathbb{R}^n, x=(x_1,\cdots,x_{n+1})\in
\mathbb{R}^{n+1}$, whether the following inequalities hold?
\begin{equation}
\left(\displaystyle\sum_{k=1}^n \alpha_kE_k(x)\right)^2\geqslant\left(\displaystyle\sum_{k=1}^n
\alpha_k E_{k-1}(x)\right)\left(\displaystyle\sum_{k=1}^n\alpha_k E_{k+1}(x)\right).
\end{equation}A counterexample was given in [Reference Ren19] to illustrate that inequality (1.5) generally does not hold. So can we add some conditions to α to ensure inequality (1.5) holds?
In this paper, we consider the Newton–Maclaurin-type inequalities for operators of three elementary symmetric functions
$E_k(\boldsymbol x)+aE_{k-1}(\boldsymbol x)+bE_{k-2}(\boldsymbol x)$. For convenience, we define
where
$\alpha, \beta\in\mathbb{R}$ and
$\boldsymbol x=(x_{1},x_{2},\cdots,x_{n})\in\mathbb{R}^{n}$.
Theorem 1. For any real
$\alpha, \beta\in\mathbb{R}$ and any
$\boldsymbol
x\in\mathbb{R}^{n}$, we have
The inequality is strict unless any
${\it n}$ elements among
$x_1,x_2,\cdots, x_n,-\alpha,-\beta$ are equal, or unless both sides of the inequality are 0.
Further, if
$\alpha,\beta \geqslant 0, E_1(\boldsymbol x)\geqslant 0, E_2(\boldsymbol x)\geqslant 0$, and
Then
\begin{equation}
S_1(\boldsymbol x) \geqslant S_{2}^{1/2}(\boldsymbol x) \geqslant \cdots\geqslant S_{k}^{1/k}(\boldsymbol
x),\quad k=2,3,\cdots,n.
\end{equation} Obviously, theorem
$1$ is a generalization of classical Newton–Maclaurin inequality and Newton–Maclaurin type inequality in [Reference Ren19]. The structural condition of constants a and b in operators
$[E_{k}(\boldsymbol {x})+a E_{k-1}(\boldsymbol x)+b E_{k-2}(\boldsymbol x)]$ can be seen as the following polynomial of t
only has real roots. And its real roots are
$-\alpha$ and
$-\beta$. The polynomial (1.8) which only has real roots is a sufficient condition of theorem 1. Probably, this condition is also necessary. In Section 2, we will provide a counterexample for the case of k = 3. For the case of k > 3, we can only provide partial counterexamples.
Similar as [Reference Hardy, Littlewood and Polya9] for
$E_{k}(\boldsymbol x)$, from (1.6) we have
Corollary 2. Let
$3\leqslant l \lt k\leqslant n$, assume
then
For the more terms case, we can establish the Newton–Maclaurin-type inequalities for operators
\begin{equation}
S_{k;s}(\boldsymbol x):=C^{0}_{s}E_{k}(\boldsymbol x)+C^{1}_{s}\alpha E_{k-1}(\boldsymbol
x)+\cdots+C^{s}_{s}\alpha^s E_{k-s}(\boldsymbol x),
\end{equation} where
$\alpha\in\mathbb{R}$ and
$\boldsymbol
x=(x_{1},x_{2},\cdots,x_{n})\in\mathbb{R}^{n}$.
Theorem 3. For general
$s\geqslant1$, any real
$\alpha\in\mathbb{R}$ and any
$\boldsymbol
x\in\mathbb{R}^{n}$, we have
\begin{align}
S_{k;s}^{2}(\boldsymbol x)\geqslant S_{k-1;s}(\boldsymbol x)S_{k+1;s}(\boldsymbol x), \quad k=s+1,\cdots, n-1.
\end{align} The inequality is strict unless
$x_{1}=x_{2}=\cdots=x_{n}$, or unless there are
$n-s$ elements in
$x_1,x_2,\cdots,x_n$ take the value
$-\alpha$, or unless both sides of the inequality are 0.
Further, if
$\alpha \geqslant 0, E_1(\boldsymbol x)\geqslant 0, E_2(\boldsymbol x)\geqslant 0,\cdots,E_s(\boldsymbol x)\geqslant 0$, and
Then (1.10) hold for all
$k=1,\cdots,n-1$, and
\begin{equation}
S_{1;s}(\boldsymbol x) \geqslant S_{2;s}^{1/2}(\boldsymbol x)\geqslant\cdots \geqslant
S_{k;s}^{1/k}(\boldsymbol x),\quad k=2,3,\cdots,n.
\end{equation} Similar as [Reference Hardy, Littlewood and Polya9] for
$E_{k}(\boldsymbol x)$, from (1.10) we have
Corollary 4. Let
$s+1 \leqslant l \lt k\leqslant n$, assume
then
The organization of this paper is as follows. In section 2, we will prove theorem 1 and in section 3 we will prove theorem 3.
2. The proof of theorem 1
First, we consider the following inequality in Euclidean space
$\mathbb{R}^{4}$.
Lemma 5. For any real numbers
$\alpha, \beta\in\mathbb{R}$ and
$\boldsymbol
z=(z_{1},z_{2},z_{3},z_{4})\in\mathbb{R}^{4}$, we have
The inequality is strict unless any 4 elements among
$z_1,z_2,z_3, z_4,-\alpha,-\beta$ are equal, or unless both sides of the inequality are 0.
Proof. The following is one of the most straightforward and simplest calculations
\begin{align}
&576[S_{3}^{2}(\boldsymbol z)- S_{2}(\boldsymbol z)S_{4}(\boldsymbol z)] \nonumber\\
=&576\Big\{[E_{3}(\boldsymbol z)+(\alpha+\beta )E_{2}(\boldsymbol z)+\alpha\beta E_{1}(\boldsymbol z)]^{2} \nonumber\\
&-[E_{2}(\boldsymbol z)+(\alpha+\beta )E_{1}(\boldsymbol z)+\alpha\beta][E_{4}(\boldsymbol z)+(\alpha+\beta )E_{3}(\boldsymbol z)+\alpha\beta E_{2}(\boldsymbol z)]\Big\} \nonumber\\
=&3\Big\{[(z_{1}-z_{2})(z_{3}+\alpha)(z_{4}+\beta)+(z_{1}-z_{2})(z_{4}+\alpha)(z_{3}+\beta)]^{2}\\
&+[(z_{1}-z_{3})(z_{2}+\alpha)(z_{4}+\beta)+(z_{1}-z_{3})(z_{4}+\alpha)(z_{2}+\beta)]^{2} \nonumber\\
&+[(z_{1}-z_{4})(z_{2}+\alpha)(z_{3}+\beta)+(z_{1}-z_{4})(z_{3}+\alpha)(z_{2}+\beta)]^{2} \nonumber\\
&+[(z_{2}-z_{3})(z_{1}+\alpha)(z_{4}+\beta)+(z_{2}-z_{3})(z_{4}+\alpha)(z_{1}+\beta)]^{2} \nonumber\\
&+[(z_{2}-z_{4})(z_{1}+\alpha)(z_{3}+\beta)+(z_{2}-z_{4})(z_{3}+\alpha)(z_{1}+\beta)]^{2} \nonumber\\
&+[(z_{3}-z_{4})(z_{1}+\alpha)(z_{2}+\beta)+(z_{3}-z_{4})(z_{2}+\alpha)(z_{1}+\beta)]^{2}\Big\} \nonumber\\
&+2\Big\{[(z_{1}-z_{2})(z_{3}-z_{4})(\alpha-\beta)]^2 \nonumber\\
&+[(z_{1}-z_{3})(z_{2}-z_{4})(\alpha-\beta)]^2 \nonumber\\
&+[(z_{1}-z_{4})(z_{2}-z_{3})(\alpha-\beta)]^2\Big\}\geqslant0. \nonumber
\end{align} It is obvious to see that the equality sign of (2.2) holds when any four elements among
$z_1,z_2,z_3, z_4,-\alpha,-\beta$ are equal. Thus we complete the proof of lemma 5.
Remark 6. The calculation of equation (2.2) is trivial but tedious. The same goes for some equations on page 9. Therefore, we suggest that readers use mathematical software (For example, Mathematica) to assist in the checking.
The following lemma is a useful tool to prove Newton’s inequalities from [Reference Hardy, Littlewood and Polya9, Reference Niculescu18, Reference Sylvester22].
Lemma 7. If
is a homogeneous function of the n-th degree in
$x$ and
$y$ whose roots are all real for
$x$ and
$\textstyle y$, then the same is true for all non-identical 0 equations
\begin{equation*}
\frac{\partial^{i+j}F}{\partial x^{i}\partial y^{j}}=0,
\end{equation*}obtained from it by partial differentiation with respect to
${\it x}$ and
$y$. Further, if Q is one of these equations, and it has a multiple root γ, then γ is also a higher multiplicity root of the equation which is derived from the differentiation of Q.
Now we prove theorem 1.
Proof of theorem 1
Assuming that P(t) is an n-degree polynomial with real roots
$x_{1},x_{2},\cdots,x_{n}$. Then P(t) can be represented as
\begin{equation*}
P(t)=\prod_{i=1}^{n}(t-x_{i})=E_{0}(\boldsymbol x)t^{n}-C_{n}^{1} E_{1}(\boldsymbol x) t^{n-1}+C_{n}^{2} E_{2}(\boldsymbol x) t^{n-2}-\cdots+(-1)^{n} E_{n}(\boldsymbol x),
\end{equation*}We apply lemma 7 to the following related homogeneous polynomial
Consider the case of the derivative
$\displaystyle\frac{\partial^{n-4}
F}{\partial t^{n-1-k} \partial s^{k-3}}$, where
$k=3, \cdots, n-1$. We obtain the following family of quartic polynomial
where
$k=3,\cdots,n-1$. The above polynomials only have real roots by lemma 7.
Next we will divide it into two cases to deal with (1.6):
Case A: When
$E_{k+1}(\boldsymbol x)\not=0$, from (2.3), we get the polynomial
\begin{equation*}
s^{4}-\frac{4 E_{k}({\boldsymbol x})}{E_{k+1}({\boldsymbol x})} s^{3}+\frac{6
E_{k-1}({\boldsymbol x})}{E_{k+1}({\boldsymbol x})}s^{2}-\frac{4E_{k-2}({\boldsymbol
x})}{E_{k+1}({\boldsymbol x})}s+\frac{E_{k-3}({\boldsymbol x})}{E_{k+1}({\boldsymbol x})},
\end{equation*} which has four real roots. We denote the roots by
${z}_{1},{z}_{2},{z}_{3},{z}_{4}$, and denote
${\boldsymbol z}=({z}_{1},{z}_{2},{z}_{3},{z}_{4})$, then we have
\begin{equation*}
E_{1}({\boldsymbol z})=\frac{E_{k}({\boldsymbol x})}{E_{k+1}({\boldsymbol x})}, ~~E_{2}({\boldsymbol
z})=\frac{E_{k-1}({\boldsymbol x})}{E_{k+1}({\boldsymbol x})}, ~~ E_{3}({\boldsymbol z}
)=\frac{E_{k-2}({\boldsymbol x})}{E_{k+1}({\boldsymbol x})}, ~~ E_{4}({\boldsymbol
z})=\frac{E_{k-3}({\boldsymbol x})}{E_{k+1}({\boldsymbol x})}.
\end{equation*} Replacing α and β with
$\displaystyle\frac{1}{\alpha}$ and
$\displaystyle\frac{1}{\beta}$ respectively, and using lemma 5 we obtain (1.6).
Case B: When
$E_{k+1}(\boldsymbol x)=0$, we first consider the case of the derivative
$\displaystyle\frac{\partial^{n-1-k}
F}{\partial t^{n-1-k}}$. We obtain the following k + 1-order polynomial
\begin{equation}
E_{0}(\boldsymbol x)t^{k+1}-C_{k+1}^{1} E_{1}(\boldsymbol x) t^{k}+C_{k+2}^{2}
E_{2}(\boldsymbol x) t^{k-1}-\cdots+(-1)^{k+1} E_{k+1}(\boldsymbol x),
\end{equation} and the polynomial only has real roots by lemma 7. We denote the roots by
${y}_{1},{y}_{2},\cdots,{y}_{k+1}$, and denote
${\boldsymbol y}=({y}_{1},{y}_{2},\cdots,{y}_{k+1})$. Since
this shows that the polynomial (2.4) has roots of value 0, and we denote these 0 value roots as
${y}_{1},{y}_{2},\cdots,{y}_{r}$ respectively. Let
$\mathbf{e}=(1,\cdots,1,0,\cdots,0)$ represent a vector where the first r elements are 1 and the last
$k+1-r$ elements are 0. Let
$\widetilde{\boldsymbol y}=\boldsymbol y+\varepsilon\mathbf{e}=(\varepsilon,\cdots,\varepsilon,y_{r+1},\cdots,y_{k+1})$, then for any ɛ > 0, we have
$E_{k+1}(\widetilde{\boldsymbol y})\not= 0,$ and
\begin{equation*}
\lim_{\varepsilon\rightarrow0}E_{m}(\widetilde{\boldsymbol y})=E_{m}( \boldsymbol
y)=E_{m}( \boldsymbol x),\quad {\rm for~~all}~~ m=1,\cdots,k+1.
\end{equation*}In the following we consider the k + 1-order polynomial
\begin{align*}
E_{0}(\widetilde{\boldsymbol y})t^{k+1}-C_{k+1}^{1}E_{1}({\widetilde{\boldsymbol y}})t^{k}s+C_{k+2}^{2}E_{2}({\widetilde{\boldsymbol y}})t^{k-1}s^{2}-\cdots+(-1)^{k+1}E_{k+1}({\widetilde{\boldsymbol y}})s^{k+1}.
\end{align*} By the definition of
$\widetilde{\boldsymbol y}$, we know that this polynomial only has real roots. Differentiating the polynomial above with respect to s k − 3 times, we obtain
where
$k=3,\cdots,n-1$.
Since
$E_{k+1}(\widetilde{\boldsymbol y})\not= 0,$ by (2.5), we have the polynomial
\begin{equation*}
s^{4}-\frac{4E_{k}(\widetilde{\boldsymbol y})}{E_{k+1}(\widetilde{\boldsymbol y})} s^{3}+\frac{6
E_{k-1}(\widetilde{\boldsymbol y})}{E_{k+1}(\widetilde{\boldsymbol y})}
s^{2}-\frac{4E_{k-2}(\widetilde{\boldsymbol y})}{E_{k+1}(\widetilde{\boldsymbol
y})}s+\frac{E_{k-3}(\widetilde{\boldsymbol y})}{E_{k+1}(\widetilde{\boldsymbol y})},
\end{equation*}which only has real roots.
We denote the roots by
$\widetilde{z}_{1},\widetilde{z}_{2},\widetilde{z}_{3},\widetilde{z}_{4}$ and denote
$\widetilde{\boldsymbol z}=(\widetilde{z}_{1},\widetilde{z}
_{2},\widetilde{z}_{3},\widetilde{z}_{4})$, then
\begin{equation*}
E_{1}(\widetilde{\boldsymbol z})=\frac{E_{k}(\widetilde{\boldsymbol
y})}{E_{k+1}(\widetilde{\boldsymbol y})}, \quad E_{2}(\widetilde{\boldsymbol
z})=\frac{E_{k-1}(\widetilde{\boldsymbol y})}{E_{k+1}(\widetilde{\boldsymbol y})},
\quad E_{3}(\widetilde{\boldsymbol z})=\frac{E_{k-2}(\widetilde{\boldsymbol
y})}{E_{k+1}(\widetilde{\boldsymbol y})}, \quad E_{4}(\widetilde{\boldsymbol
z})=\frac{E_{k-3}(\widetilde{\boldsymbol y})}{E_{k+1}(\widetilde{\boldsymbol y})}.
\end{equation*}Using lemma 5 we get
Taking
$\varepsilon\rightarrow0$, we have
$\widetilde{\boldsymbol
y}\rightarrow \boldsymbol y$, which gives
then we obtain
When any n elements among
$x_1,x_2,\cdots, x_n,-\alpha,-\beta$ are equal, by lemma 7, it can be known that four elements among
$z_1,z_2,z_3,z_4,-\alpha,-\beta$ are equal. From lemma 5 and the above proof process, we know that the equality in inequality (1.6) holds at this time.
Now, we prove (1.7). Notice that
$E_0(\boldsymbol x)=1, E_k(\boldsymbol x)=0$ for k < 0. When
$\alpha,\beta\geqslant 0,E_1(\boldsymbol x)\geqslant 0$ and
$E_2(\boldsymbol
x)\geqslant 0$, we have
\begin{align}
&S_{1}^2(\boldsymbol x)-S_{0}(\boldsymbol x)S_{2}(\boldsymbol x)\\
=&[E_1+(\alpha+\beta)]^2-[E_2+(\alpha+\beta)E_1+\alpha\beta] \nonumber\\
=&E_1^2-E_2+(\alpha+\beta)E_1+(\alpha+\beta)^2-\alpha\beta\geqslant 0, \nonumber
\end{align}and
\begin{align*}
&S^2_{2}(\boldsymbol x)-S_{1}(\boldsymbol x)S_{3}(\boldsymbol x)\\
=&E_2^2+(\alpha^2+\beta^2+\alpha\beta)E_1^2+\alpha^2\beta^2+(\alpha+\beta)E_1E_2+\alpha\beta(\alpha+\beta)E_1\\
&-[E_1E_3+(\alpha+\beta)E_3+(\alpha^2+\beta^2)E_2]\\
\geqslant&\alpha\beta E_1^2+\alpha^2\beta^2+\alpha\beta(\alpha+\beta)E_1\geqslant 0.
\end{align*}So the proof of (1.7) is similar to that of [Reference Hardy, Littlewood and Polya9, Reference Ren19]. (2.6) imply that
\begin{equation*}S_{1}(\boldsymbol x)\geqslant S_{2}^{1/2}(\boldsymbol x).\end{equation*} For
$2\leqslant k\leqslant n-1$, we assume
\begin{equation*}S_{k-1}^{1/{(k-1)}}(\boldsymbol x) \geqslant S_{k}^{1/k}(\boldsymbol x).\end{equation*}Combining the inequality (1.6) and the inequalities above, we get
\begin{align*}
S_{k}^2(\boldsymbol x) \geqslant S_{k-1}(\boldsymbol x)S_{k+1}(\boldsymbol x) \geqslant
S_{k}^{(k-1)/k}(\boldsymbol x)S_{k+1}(\boldsymbol x).
\end{align*}Thus,
\begin{equation*}
S_{k}^{1/k}(\boldsymbol x)\geqslant S_{k+1}^{1/(k+1)}(\boldsymbol x).
\end{equation*}The inequality (1.7) is proved.
$\bullet$ Counter example
Assuming polynomial
$t^2+at+b$ has two complex roots
$-c\pm {\rm
i}d$ with d ≠ 0, then
can be rewritten as
Next, we will provide a counterexample for the case of k = 3. That is to say, we found a
$\boldsymbol z_0\in \mathbb{R}^n$ which satisfies
Denote
$\boldsymbol z'=(z_3,\cdots,z_n)$, by the well-known equation (see [Reference Li, Ren and Wang13, Reference Ren and Wang20])
we have
We divide it into the following cases.
Case 1.
$|c|\geqslant |d|$. We choose
$\boldsymbol
z_0=((n-3)c,(n-3)c,-c,-c,\cdots,-c)$, using (2.7), we have
\begin{equation*}E_1(\boldsymbol z_0)=\displaystyle\frac{2(n-3)c-(n-2)c}{n}=\displaystyle\frac{(n-4)c}{n},\end{equation*}
\begin{align*}
E_2(\boldsymbol
z_0)=&\displaystyle\frac{c^2}{C_n^2}[(n-3)^2-2(n-3)(n-2)+C_{n-2}^2]=\displaystyle\frac{-(n-3)c^2}{n-1},\\
E_3(\boldsymbol
z_0)=&\displaystyle\frac{c^3}{C_n^3}[-(n-k)^2(n-2)+2(n-k)C_{n-2}^2-C_{n-2}^3]=\displaystyle\frac{-(n-3)(n-4)c^3}{n(n-1)},\\
E_4(\boldsymbol
z_0)=&\displaystyle\frac{c^4}{C_n^4}[(n-k)^2C_{n-2}^2-2(n-k)C_{n-2}^3+C_{n-2}^4]=\displaystyle\frac{(5n^2-25n+32)c^4}{n(n-1)},
\end{align*}and
\begin{align*}
S_2'(\boldsymbol z_0)=&E_2(\boldsymbol z_0)+2c E_1(\boldsymbol
z_0)+(c^2+d^2)=\displaystyle\frac{2(n-2)^2c^2+n(n-1)d^2}{n(n-1)},\\
S_3'(\boldsymbol z_0)=&\displaystyle\frac{-2(n-2)^2c^3+(n^2-5n+4)cd^2}{n(n-1)},\\
S_4'(\boldsymbol z_0)=&\displaystyle\frac{2(n-2)^2c^4-n(n-3)c^2d^2}{n(n-1)}.
\end{align*}So we have
\begin{equation*}
[S_3'(\boldsymbol z_0)]^2-S_{2}'(\boldsymbol z_0)S_{4}'(\boldsymbol
z_0)=\displaystyle\frac{2(n-2)^3c^2d^2[-2(n-2)c^2+(n-1)d^2]}{n^2(n-1)^2} \lt 0.
\end{equation*} Case 2.
$|c| \lt |d|$. We choose
$\boldsymbol
z_0=((n-3)d,(n-3)d,-d,-d,\cdots,-d)$, using (2.7), we have
\begin{equation*}
\begin{array}{ll} E_1(\boldsymbol z_0)=\displaystyle\frac{(n-4)d}{n}, &
E_2(\boldsymbol
z_0)=\displaystyle\frac{-(n-3)d^2}{n-1},\\
E_3(\boldsymbol z_0)=\displaystyle\frac{-(n-3)(n-4)d^3}{n(n-1)},\quad & E_4(\boldsymbol
z_0)=\displaystyle\frac{(5n^2-25n+32)d^4}{n(n-1)},
\end{array}
\end{equation*}and
\begin{align*}
S_2'(\boldsymbol z_0)=&\displaystyle\frac{n(n-1)c^2+2(n^2-5n+4)cd+2nd^2}{n(n-1)},\\
S_3'(\boldsymbol z_0)=&\displaystyle\frac{(n^2-5n+4)c^2d-2n(n-3)cd^2+2(n-4)d^3}{n(n-1)},\\
S_4'(\boldsymbol
z_0)=&\displaystyle\frac{-n(n-3)c^2d^2-2(n^2-7n+12)cd^3+2(2n^2-11n+16)d^4}{n(n-1)}.
\end{align*}So we have
\begin{align*}
&[S_3'(\boldsymbol z_0)]^2-S_{2}'(\boldsymbol z_0)S_{4}'(\boldsymbol
z_0)\\
=&\displaystyle\frac{2(n-2)^3d^2[(n-1)c^4+2(n-4)c^2d^2-4(n-4)cd^3-4d^4]}{n^2(n-1)^2}\\
\leqslant&\displaystyle\frac{2(n-2)^3d^2[(n-1)c^4-2(n-4)cd^3-4d^4]}{n^2(n-1)^2} \lt 0.
\end{align*}Remark 8. For the case of k = 4, when
$|c|\geqslant |d|$, we can choose
which satisfies
\begin{equation}
[S_k'(\boldsymbol z_0)]^2-S_{k-1}'(\boldsymbol z_0)S_{k+1}'(\boldsymbol z_0) \lt 0.
\end{equation} For the case of k = 5, when
$|c|\geqslant |d|$ and
$n\leqslant 9$, the
$\boldsymbol z_0$ in (2.8) also satisfies inequalities (2.9). Since the calculations similar as k = 3, here we omitted the calculation process. We suggest that readers verify it by using mathematical software.
3. The proof of theorem 3
To prove theorem
$3$, we need to prove the following lemma.
Lemma 9. For general
$k\geqslant2,~s=k-1,~n=k+1$, any real
$\alpha\in\mathbb{R}$ and any
$\boldsymbol z\in\mathbb{R}^{n}$, we have
\begin{equation}
S_{k;s}^{2}(\boldsymbol z)\geqslant S_{k-1;s}(\boldsymbol z)S_{k+1;s}(\boldsymbol z).
\end{equation} The inequality is strict unless both sides of the inequality sign are 0, or unless
$z_{1}=z_{2}=\cdots=z_{n}$, or unless there are two elements in
$z_1,z_2,\cdots,z_n$ take the value
$-\alpha$.
Proof. Let’s first prove two equations.
Equation 1. For
$n=k+1$, we have
\begin{equation}
kn^2(E_k^2-E_{k-1}E_{k+1})(\boldsymbol z)=\sum_{1\leqslant i \lt j\leqslant
n}(z_i-z_j)^2\prod_{l\neq i,j}z_l^2.
\end{equation}We can obtain this equation by directly calculation. For convenience, we can always rewrite
\begin{equation*}\sigma_{n-1}=\displaystyle\sum_{i=1}^n\frac{\sigma_n}{z_i},\quad \sigma_{n-2}=\displaystyle\sum_{1\leqslant i \lt j\leqslant n}\frac{\sigma_n}{z_iz_j},\end{equation*} regardless of whether zi and zj are 0 or not. Note that
$k=n-1$, the left-hand side of (3.2) can be represented as
\begin{align*}
kn^2(E_k^2-E_{k-1}E_{k+1})(\boldsymbol z)
=&(n-1)\sigma_{n-1}^2-2n\sigma_{n-2}\sigma_n\\
=&(n-1)\Big(\sum_{i=1}^n\frac{\sigma_n}{z_i}\Big)^2-2n\sum_{1\leqslant i \lt j\leqslant n}\frac{\sigma_n}{z_iz_j}\sigma_n\\
=&\sigma_n^2\Big[(n-1)\Big(\sum_{i=1}^n\frac{1}{z_i}\Big)^2-2n\sum_{1\leqslant i \lt j\leqslant n}\frac{1}{z_iz_j}\Big]\\
=&\sigma_n^2\Big[(n-1)\Big(\sum_{i=1}^n\frac{1}{z_i^2}\Big)-2\sum_{1\leqslant
i \lt j\leqslant n}\frac{1}{z_iz_j}\Big].
\end{align*}Meanwhile, the right-hand side of (3.2)
\begin{align*}
\sum_{1\leqslant i \lt j\leqslant n}(z_i-z_j)^2\prod_{l\neq i,j}z_l^2
=&\sum_{1\leqslant i \lt j\leqslant n}(z_i-z_j)^2\frac{\sigma_n^2}{z_i^2z_j^2}\\
=&\sigma_n^2\sum_{1\leqslant i \lt j\leqslant n}(\frac{1}{z_i}-\frac{1}{z_j})^2\\
=&\sigma_n^2\Big[(n-1)\Big(\sum_{i=1}^n\frac{1}{z_i^2}\Big)-2\sum_{1\leqslant
i \lt j\leqslant n}\frac{1}{z_iz_j}\Big].
\end{align*}Comparing the left-hand side with the right-hand side, we obtain (3.2).
Equation 2. For
$n=k+1, s=k-1$,
\begin{equation}
(S_{k;s}^2-S_{k-1;s}S_{k+1;s})(\boldsymbol z)=(E_k^2-E_{k-1}E_{k+1})(\boldsymbol
z+\alpha\boldsymbol e).
\end{equation} We give the expansions of each term in the equation above. By the definition of
$S_{k;s}$ in (1.9),
\begin{align*}
S_{k;s}(\boldsymbol z)&=\sum_{i=k-s}^kC^{i-1}_{s}\alpha^{k-i}E_i(\boldsymbol z)=\sum_{i=1}^k\frac{i(k+1-i)}{(k+1)k}\alpha^{k-i}\sigma_i,\\
S_{k-1;s}(\boldsymbol z)&=\sum_{i=k-1-s}^{k-1}C^{i}_{s}\alpha^{k-1-i}E_i(\boldsymbol z)=\sum_{i=0}^{k-1}\frac{(k-i)(k+1-i)}{(k+1)k}\alpha^{k-1-i}\sigma_i,\\
S_{k+1;s}(\boldsymbol
z)&=\sum_{i=k+1-s}^{k+1}C^{i-2}_{s}\alpha^{k+1-i}E_i(\boldsymbol
z)=\sum_{i=2}^{k+1}\frac{i(i-1)}{(k+1)k}\alpha^{k+1-i}\sigma_i.
\end{align*}Using the well-known result
\begin{equation*}
E_k(\boldsymbol z+\alpha\boldsymbol e)=\sum_{i=0}^{k} C^{i}_{k}\alpha^{k-i}E_i(\boldsymbol
z),
\end{equation*} where
$\boldsymbol e=(1,1,\cdots,1)\in \mathbb{R}^n$(see [Reference Garding4]), we have
\begin{align*}
E_k(\boldsymbol z+\alpha\boldsymbol e)&=\sum_{i=0}^{k} C^{i}_{k}\alpha^{k-i}E_i(\boldsymbol z)=\sum_{i=0}^k\frac{k+1-i}{k+1}\alpha^{k-i}\sigma_i,\\
E_{k-1}(\boldsymbol z+\alpha\boldsymbol
e)&=\sum_{i=0}^{k-1}C^{i}_{k-1}\alpha^{k-1-i}E_i(\boldsymbol
z)=\sum_{i=0}^{k-1}\frac{(k-i)(k+1-i)}{(k+1)k}\alpha^{k-1-i}\sigma_i,\\
E_{k+1}(\boldsymbol z+\alpha\boldsymbol e)&=\sum_{i=0}^{k+1}
C^{i}_{k+1}\alpha^{k+1-i}E_i(\boldsymbol
z)=\sum_{i=0}^{k+1}\alpha^{k+1-i}\sigma_i.
\end{align*}From the expansion above, we only need to prove that all the terms containing α at both hand sides of (3.3) are equal.
On the left-hand side of (3.3), the coefficient of the term
$\alpha^{2k-b}(b\geqslant2)$ is
\begin{align*}
L_{2k-b}&=\sum_{\substack{i+j=b\\1\leqslant i\leqslant k\\1\leqslant j\leqslant
k}}\frac{(k+1-i)(k+1-j)ij}{(k+1)^2k^2}\sigma_i\sigma_j-\sum_{\substack{p+q=b\\0\leqslant
p\leqslant k-1\\2\leqslant q\leqslant
k+1}}\frac{(k+1-p)(k-p)q(q-1)}{(k+1)^2k^2}\sigma_p\sigma_q\\
&:=L_{2k-b}^1-L_{2k-b}^2.
\end{align*}Since
\begin{align*}
L_{2k-b}^2=\sum_{\substack{p+q=b\\1\leqslant p\leqslant k\\1\leqslant q\leqslant
k}}\frac{(k+1-p)(k-p)q(q-1)}{(k+1)^2k^2}\sigma_p\sigma_q+\frac{b(b-1)}{(k+1)k}\sigma_b
\end{align*} when
$2\leqslant b\leqslant k+1$, and
\begin{align*}
L_{2k-b}^2=\sum_{\substack{p+q=b\\1\leqslant p\leqslant k\\1\leqslant q\leqslant
k}}\frac{(k+1-p)(k-p)q(q-1)}{(k+1)^2k^2}\sigma_p\sigma_q+\frac{(2k+2-b)(2k+1-b)}{(k+1)k}\sigma_{b-k-1}\sigma_{k+1}
\end{align*} when
$k+2\leqslant b\leqslant 2k$, then
\begin{equation}
L_{2k-b}=\left\{\begin{array}{ll}\displaystyle\sum_{\substack{i+j=b\\1\leqslant i\leqslant
k\\1\leqslant j\leqslant
k}}\frac{(i-j+1)(k+1-i)j}{(k+1)^2k}\sigma_i\sigma_j&\\[12mm]
-\displaystyle\frac{b(b-1)}{(k+1)k}\sigma_b,&2\leqslant
b\leqslant k+1,\\[5mm]
\displaystyle\sum_{\substack{i+j=b\\1\leqslant i\leqslant k\\1\leqslant j\leqslant
k}}\displaystyle\frac{(i-j+1)(k+1-i)j}{(k+1)^2k}\sigma_i\sigma_j&\\[12mm]
-\displaystyle\frac{(2k+2-b)(2k+1-b)}{(k+1)k}\sigma_{b-k-1}\sigma_{k+1},&
k+2\leqslant b\leqslant 2k.
\end{array}\right.
\end{equation} On the right-hand side of (3.3), the coefficient of the term
$\alpha^{2k-b}(b\geqslant2)$ is
\begin{align*}
R_{2k-b}&=\sum_{\substack{i+j=b\\0\leqslant i\leqslant k\\0\leqslant j\leqslant
k}}\frac{(k+1-i)(k+1-j)}{(k+1)^2}\sigma_i\sigma_j-\sum_{\substack{p+q=b\\0\leqslant
p\leqslant k-1\\0\leqslant q\leqslant
k+1}}\frac{(k+1-p)(k-p)}{(k+1)k}\sigma_p\sigma_q\\
&:=R_{2k-b}^1-R_{2k-b}^2.
\end{align*} Obviously
$R_{2k-b}=0$ when
$b=0 (i=0,j=0,p=0,q=0)$. It is easy to get
$R_{2k-b}=0$ when
$b=1 (i=1,j=0;i=0,j=1;p=1,q=0;p=0,q=1)$.
When
$2\leqslant b\leqslant k+1$,
\begin{align*}
R_{2k-b}^1=&\sum_{\substack{i+j=b\\1\leqslant i\leqslant k\\1\leqslant j\leqslant
k}}\frac{(k+1-i)(k+1-j)}{(k+1)^2}\sigma_i\sigma_j+\frac{2(k+1-b)}{k+1}\sigma_b,
\end{align*}
\begin{align*}
R_{2k-b}^2=&\sum_{\substack{p+q=b\\1\leqslant p\leqslant k\\1\leqslant q\leqslant
k}}\frac{(k+1-p)(k-p)}{(k+1)k}\sigma_p\sigma_q+\sigma_b+\frac{(k+1-b)(k-b)}{(k+1)k}\sigma_b,
\end{align*}so
\begin{equation}
R_{2k-b}=\sum_{\substack{i+j=b\\1\leqslant i\leqslant k\\1\leqslant j\leqslant
k}}\frac{(k+1-i)(ik-jk+i)}{(k+1)^2k}\sigma_i\sigma_j-\frac{b(b-1)}{(k+1)k}\sigma_b,~~2\leqslant
b\leqslant k+1.
\end{equation} When
$k+2\leqslant b\leqslant 2k$, by
\begin{align*}
R_{2k-b}^1=&\sum_{\substack{i+j=b\\1\leqslant i\leqslant k\\1\leqslant j\leqslant
k}}\frac{(k+1-i)(k+1-j)}{(k+1)^2}\sigma_i\sigma_j,
\end{align*}
\begin{align*}
R_{2k-b}^2=&\sum_{\substack{p+q=b\\1\leqslant p\leqslant k\\1\leqslant q\leqslant
k}}\frac{(k+1-p)(k-p)}{(k+1)k}\sigma_p\sigma_q+\frac{(2k+2-b)(2k+1-b)}{(k+1)k}\sigma_{b-k-1}\sigma_{k+1},
\end{align*}we get
\begin{align}
R_{2k-b}=&\sum_{\substack{i+j=b\\1\leqslant i\leqslant k\\1\leqslant j\leqslant
k}}\frac{(k+1-i)(ik-jk+i)}{(k+1)^2k}\sigma_i\sigma_j \\
&-\frac{(2k+2-b)(2k+1-b)}{(k+1)k}\sigma_{b-k-1}\sigma_{k+1},\quad
k+2\leqslant b\leqslant 2k. \nonumber
\end{align}Since
\begin{align*}
&(i-j+1)(k+1-i)j+(j-i+1)(k+1-j)i\\
=&(k+1-i)(ik-jk+i)+(k+1-j)(jk-ik+j),
\end{align*}we have
\begin{equation}
\sum_{\substack{i+j=b\\1\leqslant i\leqslant k\\1\leqslant j\leqslant
k}}\frac{(i-j+1)(k+1-i)j}{(k+1)^2k}\sigma_i\sigma_j=\sum_{\substack{i+j=b\\1\leqslant
i\leqslant k\\1\leqslant j\leqslant
k}}\frac{(k+1-i)(ik-jk+i)}{(k+1)^2k}\sigma_i\sigma_j.
\end{equation}Using (3.4),(3.5),(3.6) and (3.7), we get equation (3.3).
Now we prove (3.1). Using the equation(3.2) and (3.3), we obtain
\begin{align*}
kn^2(S_{k;s}^2-S_{k-1;s}S_{k+1;s})(\boldsymbol
z)=&kn^2(E_k^2-E_{k-1}E_{k+1})(\boldsymbol z+\alpha\boldsymbol e)\\
=&\sum_{1\leqslant i \lt j\leqslant n}(z_i-z_j)^2\prod_{l\neq
i,j}(z_l+\alpha)^2\geqslant0.
\end{align*} Specifically, when
$s=3, k=4, n=5$, we have
\begin{align*}
&100[S_{4;3}^{2}(\boldsymbol z)- S_{3;3}(\boldsymbol z)S_{5;3}(\boldsymbol z)]\\
=&(z_1-z_2)^2(z_3+\alpha)^2(z_4+\alpha)^2(z_5+\alpha)^2+(z_1-z_3)^2(z_2+\alpha)^2(z_4+\alpha)^2(z_5+\alpha)^2\\
&+(z_1-z_4)^2(z_2+\alpha)^2(z_3+\alpha)^2(z_5+\alpha)^2+(z_1-z_5)^2(z_2+\alpha)^2(z_3+\alpha)^2(z_4+\alpha)^2\\
&+(z_2-z_3)^2(z_1+\alpha)^2(z_4+\alpha)^2(z_5+\alpha)^2+(z_2-z_4)^2(z_1+\alpha)^2(z_3+\alpha)^2(z_5+\alpha)^2\\
&+(z_2-z_5)^2(z_1+\alpha)^2(z_3+\alpha)^2(z_4+\alpha)^2+(z_3-z_4)^2(z_1+\alpha)^2(z_2+\alpha)^2(z_5+\alpha)^2\\
&+(z_3-z_5)^2(z_1+\alpha)^2(z_2+\alpha)^2(z_4+\alpha)^2+(z_4-z_5)^2(z_1+\alpha)^2(z_2+\alpha)^2(z_3+\alpha)^2\geqslant0.
\end{align*} It is obvious to see that the equality holds when
$z_{1}=z_{2}=\cdots=z_{n}$, or any two elements of
$z_{1},z_{2},\cdots,z_{n}$ are equal to
$-\alpha$. We completed the proof of lemma 9.
Proof of theorem 3
By combining lemma 7 with lemma 9 and using the method in the proof of (1.6) in theorem 1, we can similarly prove (1.10) in theorem 3.
The proof of (1.11) is similar to (1.7), where the key step is to prove
\begin{equation}
S_{k;s}^{2}(\boldsymbol x)- S_{k-1;s}(\boldsymbol x)S_{k+1;s}(\boldsymbol x)\geqslant 0, \quad
k=1,\cdots, s,
\end{equation} with the condition
$\alpha \geqslant 0, E_1(\boldsymbol x)\geqslant 0, E_2(\boldsymbol x)\geqslant
0,\cdots,E_s(\boldsymbol x)\geqslant 0$.
Now we prove (3.8). Let
$1\leqslant k \lt s$. By the definition of
$S_{k;s}$ in (1.9), we have
\begin{align*}
S_{k;s}(\boldsymbol z)&=\sum_{i=0}^kC^{i}_{s}\alpha^{i}E_{k-i}(\boldsymbol z),\\
S_{k-1;s}(\boldsymbol z)&=\sum_{i=0}^{k-1}C^{i}_{s}\alpha^{i}E_{k-1-i}(\boldsymbol z),\\
S_{k+1;s}(\boldsymbol z)&=\sum_{i=0}^{k+1}C^{i}_{s}\alpha^{i}E_{k+1-i}(\boldsymbol
z).
\end{align*}From the expansion above, we only need to show that all the terms containing α at left-hand sides of (3.8) are nonnegative. The coefficient of the term α b is
\begin{align*}
L_{b}&=\sum_{\substack{i+j=b\\0\leqslant i\leqslant k\\0\leqslant j\leqslant
k}}C_s^{i}C_s^{j}E_{k-i}E_{k-j}-\sum_{\substack{p+q=b\\0\leqslant p\leqslant
k-1\\0\leqslant q\leqslant k+1}}C_s^{p}C_s^{q}E_{k-1-p}E_{k+1-q}\\
&:=L_{b}^1-L_{b}^2.
\end{align*}Obviously, from the expression of Lb above, we have
and
\begin{align*}
L_k=&\sum_{i=0}^kC_s^{i}C_s^{k-i}E_{k-i}E_{i}-\sum_{i=0}^{k-1}C_s^{i}C_s^{k-i}E_{k-i-1}E_{i+1}\\
\geqslant&\sum_{i=0}^{k-1}(C_s^{i}C_s^{k-i}E_{k-i}E_{i}-C_s^{i}C_s^{k-i}E_{k-i-1}E_{i+1})\\
=&\sum_{i=0}^{[k]/2}(C_s^{i}C_s^{k-i}E_{k-i}E_{i}+C_s^{k-1-i}C_s^{i+1}E_{i+1}E_{k-1-i}\\
&-C_s^{i}C_s^{k-i}E_{k-i-1}E_{i+1}-C_s^{k-1-i}C_s^{i+1}E_{i}E_{k-i})\\
:=&\sum_{i=0}^{[k]/2}R_i.
\end{align*} In the second equality above, we used a summation method from both ends to the middle for i. Since
$i\leqslant [k]/2$, it implies that
$k-i\geqslant i$. If
$k-i \gt i$, from
we get
$R_i\geqslant 0$. If
$k-i=i$,
So we proved
$L_k\geqslant 0$.
When
$0 \lt b \lt k$, we see that
\begin{align*}
L_{b}^1&=\sum_{i=0}^bC_s^{i}C_s^{b-i}E_{k-i}E_{k-b+i},\\
L_{b}^2&=\sum_{i=0}^bC_s^{i}C_s^{b-i}E_{k-i-1}E_{k-b+i+1}.
\end{align*} Using a similar method to the proof of
$L_k\geqslant 0$, we can get
When
$k \lt b \lt 2k$, we have
\begin{align*}
L_{b}^1&=\sum_{i=b-k}^k C_s^{i}C_s^{b-i}E_{k-i}E_{k-b+i},\\
L_{b}^2&=\sum_{p=b-(k+1)}^{k-1}
C_s^{p}C_s^{b-p}E_{k-p-1}E_{k-b+p+1}=\sum_{i=b-k}^{k}
C_s^{i-1}C_s^{b-i+1}E_{k-i}E_{k-b+i}.
\end{align*} The proof of
$L_b\geqslant 0$ is also similar to that of
$L_k\geqslant 0$.
Acknowledgments
Research of the second author is supported by NSFC Grant No. 11871243.
