Polynomials on Banach spaces

(1) Definition (Homogeneous polynomials).

  • A map \(p: E \to F\) between Banach spaces is a \(k\)-homogeneous polynomial if there exists a symmetric bounded multilinear map \(A \in \KL^s({}^k E, F)\) with

    \[ p(x) = \frac{1}{k!}\, A[x^{\otimes k}]. \]
  • \(\KP^k(E, F)\) is the space of \(k\)-homogeneous polynomials.

  • A polynomial map is a finite sum \(p = \sum_{k=0}^n p_k\) with \(p_k \in \KP^k(E, F)\).
  • \(\KP_{\leq n}(E, F)\) is the space of polynomials of degree \(\leq n\).

(2) Lemma (Polarization). The top-degree multilinear form of a polynomial \(p = \sum_{k=0}^n \frac{1}{k!} A_k[x^{\otimes k}]\) is recovered by the forward difference E0004:

\[ A_n[v_1, \dots, v_n] = \Delta(p; 0; v_1, \dots, v_n). \]

Every polynomial has a unique representation. The degree, the degree-\(k\) part \(\pi_k(p)\), and the truncation \(\pi_{\leq k}(p)\) are therefore well defined.

Proof. By multilinearity, \(p(\sum_{i \in I} v_i) = \sum_{J \subseteq I} A_{|J|}[v_J]\). The alternating sum \(\sum_I (-1)^{n-|I|}\) inverts this by Boolean Möbius inversion E0002: all terms with \(J \subsetneq [n]\) cancel by \((1-1)^{n-|J|} = 0\), leaving \(A_n[v_1, \dots, v_n]\). Uniqueness follows by induction: recover \(A_n\) via polarization, subtract, and repeat.

(3) Lemma (Polynomial properties).

  • Composition: \(\deg(q \circ p) \leq \deg(q) \cdot \deg(p)\).
  • Lipschitz: \(\|\Delta(p; 0; x)\| \leq C \|x\|\) for small \(\|x\|\).
  • Upper vanishing: \(\|\pi_{> k}(p)(x)\| / \|x\|^k \to 0\) as \(x \to 0\).
  • Lower vanishing: If \(\|p(x)\| / \|x\|^k \to 0\) and \(k \geq \deg(p)\), then \(p = 0\).

Proof. Ad composition) Expanding \(q(p(x))\) by multilinearity: each term \(B[A_1[x^{\otimes j_1}], \dots, A_\ell[x^{\otimes j_\ell}]]\) is a bounded multilinear composition, hence a homogeneous polynomial of degree \(j_1 + \cdots + j_\ell \leq \ell \cdot \deg(p)\).

Ad Lipschitz) \(\|\Delta(p; 0; x)\| = \|\sum_{j=1}^n \frac{1}{j!} A_j[x^{\otimes j}]\| \leq \sum_j \frac{\|A_j\|}{j!} \|x\|^j \leq C\|x\|\) for \(\|x\| < 1\).

Ad upper vanishing) Each \(p_j\) with \(j > k\) satisfies \(\|p_j(x)\| \leq C_j \|x\|^j\), so \(\|p_j(x)\| / \|x\|^k = C_j \|x\|^{j-k} \to 0\).

Ad lower vanishing) If \(p \neq 0\), let \(j_0\) be the minimal nonzero degree and choose \(v\) with \(p_{j_0}(v) \neq 0\). Then \(p(tv) = t^{j_0} p_{j_0}(v) + O(t^{j_0+1})\): for \(j_0 < k\) the ratio \(\|p(tv)\| / t^k\) diverges; for \(j_0 = k\) it converges to \(\|p_{j_0}(v)\| \neq 0\). Either contradicts \(p = o(\|x\|^k)\).