Extensions of isomorphisms of subvarieties in flexible varieties (Q2183725)

This paper deals with the following interesting question (which can be asked in both algebraic and analytic settings): Given a variety X, and two subvarieties \(Y_1\) and \(Y_2\). Assume that \(\phi :Y_1\rightarrow Y_2\) is an isomorphism. Is there an automorphism \(\Phi :X\rightarrow X\) so that \(\Phi |_{Y_1}=\phi\)? Most of the results below can be extended to a general ground field different from \(\mathbb{C}\). A famous result concerning this question is the Abhyankar-Moh-Suzuki theorem, asserting that in the algebraic setting, if \(X=\mathbb{C}^2\), \(Y_1,Y_2\subset X\) are isomorphic to a line, then such a \(\Phi\) exists. The main theme in the algebraic setting is that the answer is Yes if one considers \(X=\mathbb{C}^n\), and \(Y_1,Y_2\) have large enough codimensions. One such result was proven by Jelonek when \(Y_1,Y_2\) are smooth (codimension \(\geq (3n+2)/4\)). This result has been generalised (and with a better bound on the codimension) to the case where \(Y_1,Y_2\) are non-smooth by the author [Proc. Am. Math. Soc. 113, No. 2, 325--334 (1991; Zbl 0743.14011)] and \textit{V. Srinivas} [Math. Ann. 289, No. 1, 125--132 (1991; Zbl 0725.14003)] (codimension \(\geq (n+1)/2\), and when \(Y_1,Y_2\) are not smooth then the conditions also involve \(TY_1, TY_2\)). In general, the assumption of high codimension is needed. For example, \textit{J. Blanc} and \textit{I. van Santen} [Trans. Am. Math. Soc. 371, No. 12, 8429--8465 (2019; Zbl 1471.14124)] showed that there are closed curves algebraically isomorphic to \(\mathbb{C}\) in \(SL(2,\mathbb{C})\) which cannot be transferred to each other by an algebraic automorphism of \(SL(2,\mathbb{C})\) (the same is true for closed surfaces algebraically isomorphic to \(\mathbb{C}^2\)). In this paper, the author extends the previous results, giving confirmative answers in the following cases: \(X\)= \(\mathbb{C}^n\setminus Z\) (where Z is an algebraic subvariety of codimension at least \(2\)), these varieties have been first considered by \textit{M. Gromov} [Partial differential relations. Springer, Cham (1986; Zbl 0651.53001)] and \textit{J. Winkelmann} [Math. Z. 204, No. 1, 117--127 (1990; Zbl 0701.32014)] in relation to flexibility (i.e. the existence of large special algebraic automorphisms), see Theorem 0.2; \(X=\) a hypersurface in \(\mathbb{C}^m\), see Theorem 0.3; and \(X=SL(n,\mathbb{C})\), see Theorems 0.4 and 0.5. Precise assumptions are needed, see the mentioned theorems. Theorem 0.4 has some unexpected aspects in that it assumes the varieties involved to be algebraic and its automorphism \(\Phi\) is to be found in the holomorphic category. Note that for the holomorphic setting, the answer to the question at the beginning of this review is usually No (see works by \textit{J.-P. Rosay} and \textit{W. Rudin} [Trans. Am. Math. Soc. 310, No. 1, 47--86 (1988; Zbl 0708.58003)], [\textit{F. Forstnerič} et al., Ark. Mat. 34, No. 1, 97--101 (1996; Zbl 0853.58015)], \textit{H. Derksen} and \textit{F. Kutzschebauch} [Math. Ann. 311, No. 1, 41--53 (1998; Zbl 0911.32042)], and many more). For the techniques for the proofs of the results in this paper, we present for example the main steps for proving part a of Theorem 7.1 (= Theorem 0.2). This theorem asserts that if \(X=\mathbb{C}^n\setminus Z\) (where \(Z\) is a closed algebraic subvariety of \(\mathbb{C}^n\) codimension \(\geq 2\)), \(Y_1,Y_2\) are closed algebraic subvarieties of \(\mathbb{C}^n\) so that \(Y_1\cap Z=Y_2\cap Z=\emptyset\), \(\varphi :Y_1\rightarrow Y_2\) is an isomorphism, and some conditions on the codimensions of \(Y_1.Y_2\) and \(Z\) are satisfied, then there is \(\Phi :X\rightarrow X\) an automorphism extending \(\varphi\). The proof consists of showing that maps \(\rho _{k,h}: \mathbb{C}^n\rightarrow \mathbb{C}^k\), where \(k\leq n\), \(\rho _{k,h}=\rho _k\circ h\), with \(h\in SL(n,\mathbb{C})\) a general automorphism of \(\mathbb{C}^n\) and \(\rho _k:\mathbb{C}^n\rightarrow \mathbb{C}^k\) is the natural projection [Based on the observation that indeed \(\rho _{k,h}\) are, when \(h\) varies, general linear projections from \(\mathbb{C}^n\) to \(\mathbb{C}^k\), in previous sections the author developed a generalisation for the case \(X\rightarrow P\), where \(\rho _k\) is replaced by a fixed map \(\rho :X\rightarrow P\) and \(h_k\) is replaced by general automorphisms \(h:X\rightarrow X\).], has some good properties, including: \begin{itemize} \item[(3)] There is an automorphism \(\alpha :X \rightarrow X\) so that \(\rho _{k,h}\circ \varphi \circ \alpha ^{-1}|_{\alpha (Y_1)}=\rho _{k,h}|_{\alpha (Y_1)}\) for \(k\leq \dim (Y_1)\), \item[(4)] if \(k\geq n-2\) then \(\rho _{k,h}|_{Y_1}\) is a closed embedding, and \item[(5)] if \(k\geq n-2\) then \(\rho _{k,h}(Y_1)\cap \rho _{k,h}(Z)=\emptyset\). \end{itemize} For (4) the author needs to make sure that the map is a proper map. This is discussed in detail in previous sections. If one applies these properties for \(k=n\), one obtains in particular an automorphism \(\alpha\) of \(X\) which extends \(\varphi\) as wanted. The proofs of these properties (3)--(5) are by induction on \(k\), starting with the case \(k=0\). It is done in 2 stages: Stage 1, increasing \(k\) from \(0\) to \(\dim (Y_1)\), and Stage 2, increasing \(k\) from \(\dim (Y_1)\) to \(n\). The case when \(k\) is small (i.e. \(\leq \dim (Y_1)\)) is dealt with directly and easier. The case when \(k\) is large (i.e. \(\geq \dim (Y_1)+1\)) is similar, but need some extra cautions given in Remark 4.3. Note that in the paper the statements of properties (3)--(5) above are kind of streamed line of the actual statements in the paper. For (3), in the paper no \(\alpha\) is mentioned, and it is only stated for \(k\leq \dim (Y_1)\) (but then it makes readers confused when later the author writes that he will prove (3) also for \(k\geq \dim (Y_1)\), and writes that Theorem 7.1 a is obtained by taking \(k=n\)). For (4) and (5): in the paper only the case \(k=n-2\) is stated and again it causes confusions.