Abstract:Newton's inequalities $c_n^2 \ge c_{n-1}c_{n+1}$ are shown to hold for the normalized coefficients $c_n$ of the characteristic polynomial of any $M$- or inverse $M$-matrix. They are derived by establishing first an auxiliary set of inequalities also valid for both of these classes. They are also used to derive some new necessary conditions on the eigenvalues of nonnegative matrices.
What problem does this paper attempt to address?
The main problem that this paper attempts to solve is to prove a set of inequalities satisfied by the coefficients of the characteristic polynomials of M - matrices and inverse M - matrices, namely Newton's inequalities. Specifically, the paper aims to verify the following conjecture:
For any M - matrix or inverse M - matrix \(A\), the normalized coefficients \(c_j(A)\) of its characteristic polynomial satisfy Newton's inequalities:
\[c_j^2(A)\geq c_{j - 1}(A)c_{j + 1}(A),\quad j = 1,\ldots,n - 1.\]
### Main contributions
1. **Prove Newton's inequalities**: By introducing auxiliary inequalities and using induction, the author finally proves that the above Newton's inequalities hold for M - matrices and inverse M - matrices.
2. **New necessary conditions**: Based on the proof of Newton's inequalities, the author derives some new necessary conditions for the eigenvalues of non - negative matrices. These conditions can be used for further research on the inverse eigenvalue problem (NIEP) of non - negative matrices.
### Specific steps
1. **Definitions and notations**
- \(A\in\mathbb{C}^{n\times n}\) represents a complex matrix.
- \(A[\alpha]\) represents the determinant of the principal sub - matrix determined by the index set \(\alpha\).
- \(c_j(A)\) represents the normalized coefficient of the characteristic polynomial of matrix \(A\):
\[c_j(A):=\sum_{\substack{\alpha\in\langle n\rangle\\\#\alpha = j}}\frac{A[\alpha]}{\binom{n}{j}},\quad j = 0,\ldots,n.\]
2. **Introduction of auxiliary inequalities**
- Define the function \(S_{m_1,m_2,k}(A)\):
\[S_{m_1,m_2,k}(A):=\sum_{\substack{\alpha\in\langle n\rangle,\#\alpha = m_1\\\beta\in\langle n\rangle,\#\beta = m_2\\\#(\alpha\cap\beta)=k}}A[\alpha]A[\beta].\]
- Prove the following inequality by induction:
\[\frac{S_{m,m,k}(A)}{S_{m,m,k}(I_n)}\geq\frac{S_{m + 1,m - 1,k}(A)}{S_{m + 1,m - 1,k}(I_n)},\]
where \(I_n\) is the \(n\times n\) identity matrix.
3. **Proof of Newton's inequalities**
- Using the above auxiliary inequalities and combining with the non - negativity of the Hermitian form \(\Psi_m\), finally prove Newton's inequalities.
4. **Application to the inverse eigenvalue problem of non - negative matrices**
- Derive some new necessary conditions for judging whether a given \(n\)-tuple can be the spectrum of a certain non - negative matrix.
### Conclusion
This paper not only proves that the coefficients of the characteristic polynomials of M - matrices and inverse M - matrices satisfy Newton's inequalities, but also provides new tools and methods for the inverse eigenvalue problem of non - negative matrices. This helps to understand the structure and properties of non - negative matrices more deeply and provides a theoretical basis for further research.