Homeomorphism theorem for sums of translates on the real axis

Tatiana M. Nikiforova · 2025

Read Homeomorphism theorem for sums of translates on the real axis by Tatiana M. Nikiforova (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2509-07776.txt.

Introduction

Tatiana M. Nikiforova, writing alone in 2025, proves a homeomorphism theorem for sums of translates on the real line. Sums of translates build a function by adding shifted copies of kernel functions to a background field, generalizing logarithms of weighted algebraic polynomials. The kernels are concave on each side of a singularity at zero, and the translate positions can be chosen freely along the line.

For each placement, local maxima of the sum are taken over the intervals between consecutive translates and the two outer rays. The differences between successive local maxima form a vector-valued function of the placement. Under monotonicity assumptions on the kernels, Nikiforova proves that this difference map is a homeomorphism, a continuous bijection with continuous inverse, onto its range. Such rigidity results underpin equilibrium and approximation problems where translate positions must be solved from prescribed peak data.

Catalog record

CREATORTatiana M. Nikiforova
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE48 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2509-07776

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Homeomorphism theorem for sums of translates on the real axis Thanks: The work was performed as part of research conducted in the Ural Mathematical Center with the financial support of the Ministry of Science and Higher Education of the Russian Federation (Agreement number 075-02-2025-1549). Tatiana M. Nikiforova Abstract . In this paper, we study sums of translates on the real axis. These functions generalize logarithms of weighted algebraic polynomials. Namely, we are dealing with the following functions F ⁡ ( 𝐲 , t ) := J ⁡ ( t ) + ∑ j = 1 n K j ​ ( t − y j ) , 𝐲 := ( y 1 , … , y n ) , y 1 ⩽ … ⩽ y n , F(\mathbf{y},t):=J(t)+\sum\limits_{j=1}^{n}K_{j}(t-y_{j}),\quad\mathbf{y}:=(y_{1},\ldots,y_{n}),\ y_{1}\leqslant\ldots\leqslant y_{n}, where the field function J J is a function defined on ℝ \mathbb{R} , which is "admissible" for the kernels K 1 , … , K n K_{1},\ldots,K_{n} concave on ( − ∞ , 0 ) (-\infty,0) and on ( 0 , ∞ ) (0,\infty) and having a singularity at 0 . 0. We consider "local maxima" m 0 ​ ( 𝐲 ) : = sup t ∈ ( − ∞ , y 1 ] F ( 𝐲 , t ) , m n ( 𝐲 ) := sup t ∈ [ y n , ∞ ) F ( 𝐲 , t ) , m j ​ ( 𝐲 ) : = sup t ∈ [ y j , y j + 1 ] F ( 𝐲 , t ) , j = 1 , … , n − 1 , \displaystyle\begin{aligned} m_{0}(\mathbf{y})&:=\sup\limits_{t\in(-\infty,y_{1}]}F(\mathbf{y},t),\quad m_{n}(\mathbf{y}):=\sup\limits_{t\in[y_{n},\infty)}F(\mathbf{y},t),\\ m_{j}(\mathbf{y})&:=\sup\limits_{t\in[y_{j},y_{j+1}]}F(\mathbf{y},t),\quad j=1,\ldots,n-1,\end{aligned} and the difference function D ⁡ ( 𝐲 ) := ( m 1 ​ ( 𝐲 ) − m 0 ​ ( 𝐲 ) , m 2 ​ ( 𝐲 ) − m 1 ​ ( 𝐲 ) , … , m n ​ ( 𝐲 ) − m n − 1 ​ ( 𝐲 ) ) . D(\mathbf{y}):=(m_{1}(\mathbf{y})-m_{0}(\mathbf{y}),m_{2}(\mathbf{y})-m_{1}(\mathbf{y}),\ldots,m_{n}(\mathbf{y})-m_{n-1}(\mathbf{y})). We prove that, under certain assumptions on monotonicity of the kernels, D D is a homeomorphism between its domain and ℝ n . \mathbb{R}^{n}. 2010 Mathematics Subject Classification 41A50, 41A52, 42A15, 26A51 1. Introduction In this paper, we study sums of translates on the real axis. These functions generalize logarithms of weighted algebraic polynomials. Namely, we are dealing with the following functions F(\mathbf{y},t):=J(t)+\sum\limits_{j=1}^{n}K_{j}(t-y_{j}),\quad\mathbf{y}:=(y_{1},\ldots,y_{n}),\ y_{1}\leqslant\ldots\leqslant y_{n}, where the field function J is a function defined on \mathbb{R} , which is "admissible" for the kernels K_{1},\ldots,K_{n} concave on (-\infty,0) and on (0,\infty) and having a singularity at 0. The sums of translates and the minimax problem for such functions were first considered by P. C. Fenton in 2000 [ 1 ] . He considered one kernel with assumptions of monotonicity, smoothness, singularity of its derivative at 0 , and a concave field J continuous at the ends of the segment. Fenton’s original goal was to prove P. D. Barry’s conjecture from 1962 on the growth of entire functions, which he succeeded in 1981 [ 2 ] . And although A. A. Goldberg proved this conjecture a little earlier [ 3 ] , Fenton obtained other interesting results in the theory of entire functions using his approach [ 4 ] , [ 5 ] . In our previous paper [ 6 ] , we studied the minimax problem for the sums of translates on the real axis. The main method was a reduction to the minimax theorem proved by B. Farkas, B. Nagy and Sz. Gy. Révész for a segment (see [ 7 ] , [ 8 ] ). The uniqueness of the minimax point followed immediately from this reduction. To prove the uniqueness of the minimax point on the segment, the authors used the so-called homeomorphism theorem [ 9 , Th. 7.1] , which was also proved by them. Now, we prove a similar homeomorphism theorem for the real axis. As for the segment, this result provides the uniqueness of the minimax point on the real axis, independently of the specific reduction technique. We consider "local maxima" (1) \displaystyle m_{0}(\mathbf{y}) \displaystyle:=\sup\limits_{t\in(-\infty,y_{1}]}F(\mathbf{y},t),\quad m_{n}(\mathbf{y}):=\sup\limits_{t\in[y_{n},\infty)}F(\mathbf{y},t), \displaystyle m_{j}(\mathbf{y}) \displaystyle:=\sup\limits_{t\in[y_{j},y_{j+1}]}F(\mathbf{y},t),\quad j=1,\ldots,n-1, and the difference function D(\mathbf{y}):=(m_{1}(\mathbf{y})-m_{0}(\mathbf{y}),m_{2}(\mathbf{y})-m_{1}(\mathbf{y}),\ldots,m_{n}(\mathbf{y})-m_{n-1}(\mathbf{y})). We prove that, under certain assumptions on monotonicity of the kernels, D is a homeomorphism between its domain and \mathbb{R}^{n}. Results of this kind are inspired by the problem of optimizing the Lagrange interpolation of a continuous function. Let us give an overview of the results known to us. Let f be a function continuous on [0,1] and y_{0}=0<y_{1}<\ldots<y_{n}<1=y_{n+1} be interpolation nodes. Denote by \pi_{n+1} the space of polynomials of degree at most n+1 and by P_{\mathbf{y}}:C_{[0,1]}\to\pi_{n+1} the Lagrange interpolation operator P_{\mathbf{y}}f(t):=\sum\limits_{j=0}^{n+1}f(y_{j})\ell_{j}(t),\quad\ell_{j}(t):=\prod\limits_{i\neq j}\dfrac{t-y_{j}}{y_{i}-y_{j}}. It is easy to show that \|P_{\mathbf{y}}f-f\|_{C_{[0,1]}}\leqslant\dist(f,\pi_{n+1})(1+\|P_{\mathbf{y}}\|), where \|P_{\mathbf{y}}\| is the operator norm. Therefore, it is natural to minimize \|P_{\mathbf{y}}\| by \mathbf{y} to optimize the interpolation. It is known that \|P_{\mathbf{y}}\|=\|\Lambda_{\mathbf{y}}\|_{C_{[0,1]}}, where \Lambda_{\mathbf{y}}(t):=\sum\limits_{j=0}^{n+1}|\ell_{j}(t)| is the Lebesgue function. Thus, this optimization problem is exactly a minimax problem for the functions \Lambda_{\mathbf{y}}. Denote \lambda_{j}(\mathbf{y}):=\max\limits_{t\in[y_{j},y_{j+1}]}\Lambda_{\mathbf{y}}(t),\ j=0,\ldots,n. In 1931, S. N. Bernstein [ 10 ] conjectured that the minimum of \|\Lambda_{\mathbf{y}}\|_{C_{[0,1]}} is attained when \mathbf{y} is an equioscillation point , i. e., \lambda_{0}(\mathbf{y})=\ldots=\lambda_{n}(\mathbf{y}). In 1977, T. A. Kilgore [ 11 ] proved Bernstein’s conjecture. Moreover, he showed the uniqueness of the equioscillation point. More precisely, Kilgore’s note [ 12 ] describing the proof of the statement “the minimax point is an equioscillation point” was first published, and a few months later, the complete proof was presented in [ 11 ] . Subsequently, C. R. de Boor and A. Pinkus [ 13 ] also proved Bernstein’s conjecture. In fact, they obtained the following general result. Theorem A . Let S^{[0,1]}:=\{\mathbf{y}=(y_{1},\ldots,y_{n})\in\mathbb{R}^{n}:\ 0<y_{1}<\ldots<y_{n}<1\}. Then the difference function D_{\lambda}:S^{[0,1]}\to\mathbb{R}^{n},\quad\mathbf{y}\mapsto(\lambda_{1}(\mathbf{y})-\lambda_{0}(\mathbf{y}),\ldots,\lambda_{n}(\mathbf{y})-\lambda_{n-1}(\mathbf{y})) is a homeomorphism between S^{[0,1]} and \mathbb{R}^{n}. It immediately implies that there is exactly one equioscillation point. De Boor and Pinkus then refer to Kilgore’s earlier note with the proof that the minimax point is an equioscillation point, and together all this proves Bernstein’s conjecture. In 1996, Y. G. Shi [ 14 ] considered more general functions \varphi_{j}(\cdot),\ j=0,\ldots,n, instead of \lambda_{j}(\cdot). Shi supposed that all the functions \varphi_{j} are continuously differentiable on S^{[0,1]} and satisfy conditions \min\limits_{0\leqslant j\leqslant n}\lim\limits_{y_{j+1}-y_{j}\to 0}\ \max\limits_{0\leqslant i\leqslant n-1}|\varphi_{i+1}(\mathbf{y})-\varphi_{i}(\mathbf{y})|=\infty and \Phi_{k}(\mathbf{y}):=\det\left(\dfrac{\partial\varphi_{i}(\mathbf{y})}{\partial y_{j}}\right)^{n\ n}_{j=1,\ i=0,\ i\neq k}\neq 0,\quad\mathbf{y}\in S^{[0,1]},\ k=0,\ldots,n. Shi dealt with the following minimax problem: find a vector \mathbf{y}=(y_{1},\ldots,y_{n}),\ 0<y_{1}<\ldots<y_{n}<1, that minimizes \max\limits_{j=0,\ldots,n}\varphi_{j}(\cdot). Under these assumptions, Shi proved that there exists a unique extremal point and it has the equioscillation property. Moreover, Shi obtained the homeomorphism theorem for the difference function D_{\varphi}:S^{[0,1]}\to\mathbb{R}^{n},\quad\mathbf{y}\mapsto(\varphi_{1}(\mathbf{y})-\varphi_{0}(\mathbf{y}),\ldots,\varphi_{n}(\mathbf{y})-\varphi_{n-1}(\mathbf{y})). In particular, this theorem implies the uniqueness of the minimax point. In 2018, B. Farkas, B. Nagy and Sz. Gy. Révész presented a solution of the minimax problem and a homeomorphism result for sums of translates F(\mathbf{y},t):=K_{0}(t)+\sum\limits_{j=1}^{n}K_{j}(t-y_{j}) on a torus [ 15 ] . Here K_{0},\ldots,K_{n}:\ \mathbb{R}\to[-\infty,0) are 2\pi -periodic functions, strictly concave on (0,2\pi) . Assuming that for each j=0,\ldots,n the function K_{j} belongs to C^{2}(0,2\pi) with K^{\prime\prime}_{j}<0 and K(0)=K(2\pi)=-\infty, they proved that the difference function of local maxima is a homeomorphism between its domain and \mathbb{R}^{n} . On the one hand, the sums of translates approach is more specific than Shi’s. On the other hand, the authors provided an example [ 15 ,

Read the complete text