Whitney's extension problem for \(C^m\) (Q855260)

This important and impressive paper provides an answer to the following old problem of Whitney: being given a real-valued function \( \varphi \) defined on a compact subset \( E \) of \( {\mathbb R}^n \), how can we tell whether there exists \( F\in C^m({\mathbb R}^n)\) such that \( F_{| E}=\varphi \)? In the particular case \(m=1 \), this problem was solved by \textit{G. Glaeser} [J. Anal. Math. 6, 1--124 (1958; Zbl 0091.28103)] using a geometrical object called the ``iterated paratangent space''. A decisive breakthrough in the general case was made by \textit{E. Bierstone, P. D. Milman}, and \textit{F. Pawłucki} in [Invent. Math. 151, No.~2, 329--352 (2003; Zbl 1031.58002)]. They constructed a ``paratangent bundle'', which is an analogue of Glaeser's iterated paratangent space relevant to \( C^m({\mathbb R}^m) \), and they conjectured a solution of Whitney's problem based on their construction. In the present paper, the paratangent bundle is replaced by a natural variant whose definition goes along the following scheme: with each point \( y\in E \) is associated a ``holding space for \( \varphi \)'', that is, an affine subspace \( H(y) \) of the vector space \( \mathcal P \) of real \( m^{ th} \) degree polynomials on \( {\mathbb R}^n \), such that, for all \( F \in C^m({\mathbb R}^n) \) satisfying \( F_{| E}=\varphi \), the \( m\)-jet of \( F \) at \( y \) belongs to \( H(y) \) (here, we make the convention that the empty set is allowed as an affine subspace of \( \mathcal P \)). Starting with the trivial choice \( H(y)=H_0(y)=\{P\in {\mathcal P} : P(y)=\varphi(y)\} \), the author produces a sequence \( H_0(y)\supseteq H_1(y)\supseteq \cdots\supseteq H_{l-1}(y)\supseteq H_l(y)\supseteq \cdots \) of holding spaces for \( \varphi \) by iterating an explicit construction that he calls a ``Glaeser refinement''. Following a crucial argument of Glaeser and Bierstone-Milman-Pawłucki, he shows that the process stabilizes at stage \( l=2\dim{\mathcal P}+1 \). The corresponding \( H_l(y) \) is called the ``stable holding space for \( \varphi \)'' and denoted by \( H_*(y) \). In principle, \( H_*(y) \) is computable from the data \( \varphi \) and \( E \). The answer to Whitney's problem then can be stated as follows: \( \varphi\) extends to a \(C^m \) function on \({\mathbb R}^n \) if and only if \( H_*(y) \) is nonempty for all \( y\in E \). In this case, given \( y_0 \) in \( E \) and \( P_0\in {\mathcal P} \), we have \( P_0\in H_*(y_0) \) if and only if there exists \( F\in C^m({\mathbb R}^n) \) with \( F_{| E}=\varphi \) whose \( m \)-jet at \( y_0 \) coincides with \( P_0 \). The paper includes several other results, among which the heart of the matter is a technical statement about the control of the \( C^m \) norm of functions interpolating \( m \)-jets on \( E \) in terms of bounds on finite subsets of \( E \) of suitable cardinality \( k^\sharp \) depending only on \( m \) and \( n \). The proofs are quite difficult; techniques of modern real analysis and convexity arguments are combined into an intricate induction process.