A wonderful embedding of the loop group (Q529238)

\textit{C. De Concini} and \textit{C. Procesi} [Lect. Notes Math. 996, 1--44 (1983; Zbl 0581.14041)] invented the \textit{Wonderful Compactification} of a complex semisimple adjoint group \(G\). This compactificaton is a smooth projective variety containing \(G\) as a dense open subvariety and where the boundary is a normal crossing divisor with structure determined by the root datum of \(G\). This compactification is claimed to have many applications, particularly in spherical geometry. In the present article, the author consider the loop group \(LG\) of the group \(G\) and construct an analogue of the wonderful compactification. The loop group can be defined as the group of maps from a punctured formal disc to the group \(G\). The points are the \(\mathbb C((z))\)-points in \(G\), that is \(G(\mathbb C((z))\), where \(C((z))\) is the field of formal Laurent series. An embedding \(X^{\mathrm{aff}}\) is constructed which is not compact, but which have nice properties justifying it as a loop group analogue of the wonderful compactification. The main application of the theory developed in this work concerns the moduli stack \(\mathcal M_G(C)\) of \(G\)-bundles on a family of nodal curves \(C\). This is not a compact stack, and so \(X^{\mathrm{aff}}\) is used as a kind of compactification. The compactification of the coarse moduli space of \(G\)-bundles on \(C\), is the original goal of the wonderful compactification. There exists compactification for semistable groups, but the author claims that non of these are leading to a satisfactory construction of a compact moduli space for bundles over families of nodal curves. This article shows that a compactification \(\overline G\) of \(G\) is insufficient to compactify \(\mathcal M_G(C)\) in general, but that the embedding \(X^{\mathrm{aff}}\) gives enough additional data to compactify \(\mathcal M_G(C)\). \textit{E. Frenkel} et al. [Adv. Math. 288, 201--239 (2016; Zbl 1342.14111)] use the compactification of the moduli of \(\mathbb C^\times\)-bundles to define Gromov-Witten invariants for \([\mathrm{pt}/\mathbb C^\times]\). Their index theorem suggest that similar invariants can be defined on a completion of \(\mathcal M_G(C)\). Then the parallels between \(X^{\mathrm{aff}}\) and the wonderful compactification \(\overline{G_{\mathrm{ad}}}\) suggest that also other constructions related to \(\overline{G_{\mathrm{ad}}}\) have loop analogues. Vinberg has given an alternative construction of \(\overline{G_{\mathrm{ad}}}\) using a monoid \(S_G\) called the Vinberg monoid, the construction was generalized by Thaddeus and Martens to provide stacky compactifications for any split reductive group. The main result of this article makes it relevant to ask if there is a loop group analogue of the Vinberg monoid. In geometric representation theory, the wonderful compactification is used e.g. to define the Harish-Chandra transform defined on the category of \(D\)-modules on \(G\), and the results of this article indicates that similar constructions exists for \(D\)-modules on the loop group LG. The wonderful compactification generalizes to elliptic Springer theory by giving an analogue of Lustig's character sheaves for loop groups. This is originally used to give a geometric construction of character sheaves for \(p\)-adic groups. In the present work, the intention is to study sheaves on conjugacy classes in \(LG\). Applying results from \textit{V. Baranovsky} and \textit{V. Ginzburg} [Int. Math. Res. Not. 1996, No. 15, 733--751 (1996; Zbl 0992.20034)], this can be studied by sheaves on the moduli space \(M_{G,E}\) of semistable bundles on an elliptic curve. Because conjugacy classes in \(LG\) are \(\Delta(LG)\) orbits in \(LG\), the author investigates if \(\Delta(\mathbb C^\times\ltimes LG)\) orbits in \(X^{\mathrm{aff}}\) have a modular interpretation like the interpretation given by Baranovsky and Ginsburg. If this was true, then it can be used to formulate a theory of character sheaves for loop groups. The author points out that \(LG\) is an ind-scheme rather than a scheme. In this article, the objects under study is thus the category of \textit{ind-schemes}, and the reader is supposed to have good knowledge of such, together with their stack theory. Then \(LG\) is studied through its representations, and it is proved that \(LG\) has a class of projective representations behaving in many ways like the finite dimensional representations of a semisimple group. These are the \textit{honest representations} of a central extension \(\widetilde{LG}\) of \(LG\) by \(\mathbb C^\times\). Such representations are infinite dimensional, and one introduce another \(\mathbb C^\times\) and puts \(G^{\mathrm{aff}}=\mathbb C^\times\ltimes\widetilde{LG}\) to decompose the representation into a direct sum of finite dimensional weight spaces for a maximal torus in \(G^{\mathrm{aff}}(\mathbb C)\). The author uses the representation theory of \(G^{\mathrm{aff}}(\mathbb C)=\mathbb C^\times\ltimes\widetilde{LG}(\mathbb C)\) to replace representation theory of a finite dimensional semisimple group. \(G^{\mathrm{aff}}(\mathbb C)\) is called a Kac-Moody group and is associated to an affine Dynkin diagram in the same way a semisimple group is associated to a Dynkin Diagram. The wonderful compactification of \(G\) is the compactification of \(G_{\mathrm{ad}}=G/Z(G)\) given by choosing a regular dominant weight \(\lambda\), letting \(V(\lambda)\) be the associated highest weight representation of \(G\), and defining \(\overline{G_{\mathrm{ad}}}=\overline{G\times G[\mathrm{id}]}\subset\mathbb P \mathrm{End}(V(\lambda)).\) The analogue construction is then given by letting \(\lambda\) be a dominant weight \(\underline{\lambda}\) of \(G^{\mathrm{aff}}\) and using the associated representation \(V(\underline{\lambda})\) to construct an ind-scheme \(\mathbb P \mathrm{End}^{\mathrm{ind}}(V(\underline{\lambda}))\) to obtain \(X^{\mathrm{aff}}=\overline{G^{\mathrm{aff}}\times G^{\mathrm{aff}}[\mathrm{id}]}\subset\mathbb P \mathrm{End}^{\mathrm{ind}}(V(\underline{\lambda})).\) The main result in the article states that when \(G\) is a simple, connected and and simply connected group over \(\mathbb C\) with maximal torus \(T\), the ind-scheme \(X^{\mathrm{aff}}\) contains \(G^{\mathrm{aff}}_{\mathrm{ad}}\) as a dense open sub-ind scheme. The main theorem also contains statements of the properties of \(X^{\mathrm{aff}}\), the boundary of the wonderful compactification of \(LG\), and the properties of the maximal torus. Also, the maximal tori are studied by toric varieties, leading to an explicit description of the orbits of the group-action. The article contains a very nice study of compactifications, and introduce new tools to study such. It is not very self-contained, as stacks, ind-schemes, and Lie-algebra actions is a basis for the article, making it much deeper than it seems. However, taking this basis into account, the article is very well written and give a framework for wonderful compactifications.