Polynomials and identities on real Banach spaces (Q641662)

Let \(C(\mathbb{R}^n)\) be the space of all continuous functions on \(\mathbb{R}^n\). For \(z\in \mathbb{R}^n\), let \(\mathbf z\) denote the evaluation map given by \({\mathbf z}(f)=f(z)\), \(f\in C(\mathbb{R}^n)\). In the paper under review these evaluation functionals are called \textit{nodes} and occasionally \(z\) is identified with \(\mathbf z\). Let \({\mathcal F}(\mathbb{R}^n)\) be the linear space of all formal finite linear combinations of nodes, where the formal summation and multiplication by a scalar, denoted by \(\alpha {\mathbf x}\boxplus\beta {\mathbf y}\), is given by \((\alpha {\mathbf x}\boxplus\beta {\mathbf y})(f)=\alpha f(x)+\beta f(y)\). Let \(X\), \(Y\) be Banach spaces, a continuous mapping \(f:X\to Y\) is \textit{compatible} with \({\mathbf x}=a_1x_1\boxplus\cdots \boxplus a_n x_n\in {\mathcal F}({\mathbb R}^m)\) if \(\langle f\circ L,{\mathbf x}\rangle =\sum_{i=1}^n a_if(Lx_i)=0\) for all \(L\in {\mathcal L}(\mathbb{R}^m;X)\). If each \(x_k\) is written as \((x_k^1,\ldots,x_k^m)\), then \(f\) is compatible with \(\mathbf x\) if and only if \(\sum_{k=1}^n a_kf(\sum_{i=1}^mx_k^iz_i)=0\) for every \(z_1,\ldots,z_m\in X\). This last equality is called a \textit{linear identity}. The authors develop an abstract approach to the theory of linear identities. This treatment unifies and generalizes some classical results in the area. Amongst these are the classical characterization of polynomials due to Fréchet or the polarization formula for homogeneous polynomials. Among other related results, the following is proved: Theorem: Let \(X\), \(Y\) be Banach spaces, \(f:X\to Y\) be a continuous mapping and \({\mathbf x}=a_0x_0\boxplus\cdots \boxplus a_{n+1} x_{n+1}\in {\mathcal F}({\mathbb R}^m)\), \(m\geq 2\), \(n\in \mathbb{N}\cup\{0\}\). Suppose that for every \(k\neq n+1\) the vectors \(x_k\) and \(x_{n+1}\) are linearly independent, and that \(a_{n+1}\neq 0\). Let \(p+2\) be the number of distinct directions determined by the vectors \(x_0,\ldots,x_{n+1}\). If \(f\) is compatible with \(\mathbf x\), then \(f\) is a polynomial of degree at most \(p\). The linear space of all continuous mappings from \(X\) to \(Y\) which are compatible with a fixed \(\mathbf x\) is also studied and several interesting results on compatibility are proved. For instance, compatibility is used to characterize polynomials of highest possible degree. A special case corresponds to the classical Lagrange interpolation polynomial.