Pietsch's variants of \(s\)-numbers for multilinear operators (Q2240439)

From MaRDI portal
scientific article
Language Label Description Also known as
English
Pietsch's variants of \(s\)-numbers for multilinear operators
scientific article

    Statements

    Pietsch's variants of \(s\)-numbers for multilinear operators (English)
    0 references
    0 references
    0 references
    0 references
    3 November 2021
    0 references
    In this paper the authors extend Pietsch's theory of $s$-numbers of operators to multilinear mappings. Given a natural number $k$ they say that an $s^{(k)}$-scale is a rule which assigns to each $k$-linear operator, $T$, from Banach spaces $X_1\times\dots\times X_k$ to $F$ a sequence of non-negative numbers $(s_n^{(k)})_n$ such that (S1) For each $k$-linear mapping $T$, $$ \|T\|=s_1^{(k)}(T)\ge s_2^{(k)}(T)\ge\dots\ge 0. $$ (S2) For every pair of $k$-linear mappings $S,T\colon X_1\times\dots\times X_k\to F$, $$ s_{m+n-1}^{(k)}(S+T)\le s_m^{(k)}(S)+s_n^{(k)}(T). $$ (S3) Given a $k$-linear mapping $T\colon X_1\times\dots\times X_k\to F$ and linear mappings $S\colon Y\to Z$, $R_j\colon W_j\to X_j$, $1\le j\le k$, $$ s_n^{(k)}(S\circ T\circ(R_1,\ldots, R_k))\le \|S\|s_n^{(k)}(T)\|R_1\|\dots \|R_k\|. $$ (S4) If $\mathrm{rank}(T)$ (defined as the dimension of the span of $T(X_1,\dots, X_k)$) is strictly less than $n$, then $S_n^{(k)}(T)=0$. (S5) If $\bigotimes_kI_n$ is the $k$-linear mapping from $\underbrace{\ell_2^n \times\dots\times \ell_2^n}_{k\text{-times}}$ to $\ell_2^{[n]^k}$ given by $$ \bigotimes_k I_n( x_1,\dots, x_k)(\mathbf{j})=x_1(j_1)\dots x_k(j_k) \ \text{for} \ \mathbf{j}=(j_1,\dots,j_k)\in [n]^k, $$ then $s_n^{(k)}(\bigotimes_k I_n)=1$. An $s^{(k)}$-scale is said to be injective if $s_n^{(k)}(T)=s_n^{(k)}(JT)$ for all $k$-linear mappings $T\colon X_1\times\dots\times X_k\to Y$ and all metric injections $J\colon Y\to Z$, and surjective if $ s_n^{(k)}(T) =s_n^{(k)}(T(Q_1,\dots,Q_k))$ for all $k$-linear mappings $T\colon X_1\times\dots\times X_k\to Y$ and all metric surjections $Q_j\colon Y_j \to X_j$, $1\le j\le k$. The authors define the $n$-th approximation number of a $k$-linear mapping $T\colon X_1\times\dots\times X_k\to Y$, $a_n^{(k)}$, by $$ a_n^{(k)}(T)=\inf\{\|T-A\|: A: X_1\times\dots\times X_k\to Y, \ k\text{-linear with rank}(A)<n\}, $$ the $n$-th Gelfand number of a $k$-linear mapping $T\colon X_1\times \dots\times X_k\to Y$, $c_n^{(k)}$, by $c_n^{(k)}=a_n^{(k)}(J_YT)$, where $J_Y\colon Y\to \ell_\infty(B_{Y^*})$ is the canonical injection given by $J_Y(y)=(y^*(y))_{y^*\in B_{ Y^*}}$, and the $n$-th Kolmogorov number of a $k$-linear mapping $T\colon X_1\times\dots\times X_k\to Y$, $d_n^{(k)}$, by $d_n^{(k)}=a_n^{(k)}(T(Q_1, \dots, Q_k))$, where $Q_j$ is the canonical metric surjection from $\ell_1 (B_{X_j})$ onto $X_j$. They show that $(a_n^{(k)})_n$ is the largest sequence of $s^{(k)}$-scales, $(c_n^{(k)})_n$ is the largest injective sequence of $s^{(k)}$-scales, and $(d_n^{(k)})_n$ is the largest surjective sequence of $s^{(k)}$-scales. Moreover, for each decreasing sequence $(\lambda_n)_n$ of strictly positive real numbers there is a diagonal bilinear mapping $D_\lambda\colon\ell_p\times\ell_p\to \ell_p$ with $a_r^{(2)}(D_\lambda) =\lambda_r$ for each $r$ in $\mathbb{N}$. In an analogous way to scales of linear operators, the $s^{(k)}$-scales of Hilbert, Weyl and Chang numbers are defined using linear operators from and to $\ell_2$. The paper concludes with the introduction of the Bernstein numbers for bilinear operators and an inspection of which of the properties of an $s^{(2)}$-scale they possess.
    0 references
    multilinear operators
    0 references
    \(s\)-numbers
    0 references
    adjoint operator
    0 references
    approximation numbers
    0 references

    Identifiers