betaFeedback

WorksMathematics

Holomorphic flexibility properties of strongly commuting varieties

Abstract

We study holomorphic flexibility properties of the strongly commuting variety of a complex semisimple algebraic group GG. We prove that the smooth locus of the strongly commuting variety of GG is Oka-11. We also prove that the smooth locus of the commuting variety of a semisimple complex Lie algebra is Oka-11, and that it contains a Zariski dense open subset which is algebraically elliptic.

Introduction

Main results. This paper studies holomorphic flexibility properties of two commuting varieties: the commuting variety of a complex algebraic group GG, defined as

R(G):={(x,y)∈G2∣xy=yx},R(G) := \{(x,y) \in G^2 \mid xy = yx\},

and the commuting variety of a complex Lie algebra g\mathfrak{g}, defined as

C(g):={(x,y)∈g2∣[x,y]=0}.C(\mathfrak{g}) := \{(x,y) \in\mathfrak{g}^2 \mid[x,y] = 0\}.

The Oka-1 property is a holomorphic flexibility property introduced by Alarcón and Forstnerič [1]. Roughly speaking, a complex manifold is Oka-1 if every continuous map from an open Riemann surface into it is homotopic to a holomorphic map, with approximation and interpolation. The precise definition is recalled in § 2.1.

For a semisimple algebraic group GG, let q:G~→Gq:\widetilde{G}\to G be its universal covering. Define

Rstr(G):=(q×q)(R(G~))R_{\mathrm{str}}(G):=(q\times q)(R(\widetilde{G}))

and call it the strongly commuting variety of GG. Thus, a pair belongs to Rstr(G)R_{\mathrm{str}}(G) precisely when its entries lift to commuting elements in G~\widetilde{G}. Observe that if GG is simply connected then R(G)=Rstr(G)R(G)=R_{\mathrm{str}}(G). Our main results are the following:

Theorem 1. Let GG be a semisimple complex algebraic group. The smooth locus of the strongly commuting variety Rstr(G)R_{\mathrm{str}}(G) of GG is Oka-1.

In particular, Theorem 1 implies that any continuous family of commuting pairs in the smooth locus of Rstr(G)R_{\mathrm{str}}(G), parameterised by an open Riemann surface, is homotopic to a holomorphic family, with approximation and interpolation. We prove an analogous result in the Lie algebra case:

Theorem 2. Let g\mathfrak{g} be a semisimple complex Lie algebra. The smooth locus of the commuting variety C(g)C(\mathfrak{g}) of g\mathfrak{g} is Oka-1. Moreover, there is a dense open subset of C(g)C(\mathfrak{g}) which is algebraically elliptic.

Historical background. Commuting varieties were first studied in [MT55] for the Lie algebra gln\mathfrak{gl}_n. The authors showed that the commuting variety is irreducible of dimension n2+nn^2+n. This result was later generalised by Richardson [46] who proved that the commuting varieties of reductive Lie algebras (resp. semisimple simply connected groups) over algebraically closed fields of characteristic zero are irreducible. The Lie algebra result was extended in [36] in the case of a field of positive characteristic and under certain mild assumptions on GG. Other versions of the commuting variety have also been studied. For instance, in [7] and [44], motivated by the study of moduli of vector bundles on algebraic surfaces, the authors investigated commuting varieties of nilpotent endomorphisms of a vector space, and related them to certain punctual Hilbert schemes. For other variants, in [11] and [31] the authors studied and classified varieties of almost commuting pairs of elements in compact Lie groups KK which appear naturally when one studies flat connections on principal KK-bundles over a torus. One may also consider the variety of commuting nn-tuples of a reductive group GG and of its maximal compact subgroup KK. The topology and cohomology of these varieties have been studied in [Bai07, GPS12, PS13]. In particular, it was shown that the variety of commuting nn-tuples in KK is a strong deformation retract of the variety of commuting nn-tuples in GG. Commuting varieties of algebraic groups are also special cases of representation varieties.

Representation varieties Hom⁡(π1(Σ),G)\operatorname{Hom}(\pi_1(\Sigma),G) are varieties of representations of the fundamental group of a surface Σ\Sigma into a group GG. Thus, the commuting variety of GG is the representation variety associated with a torus and the group GG. The character variety associated with Σ\Sigma and GG is the quotient of the representation variety by the conjugation action of GG, parameterising representations up to equivalence. Character varieties, together with their related moduli spaces of flat connections and Higgs bundles, have been studied using a wide range of methods, appearing in many areas of mathematics and physics including non-abelian Hodge theory [49, 50], mirror symmetry [25], geometric Langlands [BD96, BZN18], and gauge theory [5, 26]. In this paper we study commuting varieties from the point of view of Oka theory.

Hyperbolic manifolds and Oka manifolds sit at opposite ends in the landscape of complex manifolds. The former are holomorphically rigid and the latter are holomorphically flexible. For instance, a theorem of Liouville implies that holomorphic maps C→D\mathbb{C} \to\mathbb{D} are constant, illustrating the rigidity of such maps. Oka theory makes precise the property of a complex manifold of admitting ‘many’ holomorphic maps into it. Oka theory originates in the classical work of Oka and Grauert concerning analytic problems on Stein manifolds. This work gave rise to the Oka principle which says that analytic problems on Stein spaces have analytic solutions whenever there are no topological obstructions. In his seminal 1989 paper [22], Gromov introduced the notion of a dominating spray, a geometric condition which implies strong holomorphic flexibility properties. More precisely, he proved that if XX admits a dominating spray, then XX satisfies an Oka principle for holomorphic maps from Stein spaces SS: every continuous map S→XS \to X is homotopic to a holomorphic one with approximation and interpolation. Building on Gromov’s work, Forstnerič and others developed the modern theory of Oka manifolds and established the equivalence of several approximation, interpolation, and parametric forms of the Oka property [16, 17].

Oka theory has now seen a diverse range of applications. These include the classification of holomorphic vector bundles on Stein spaces, the existence of holomorphic sections with prescribed local properties, problems concerning holomorphic embeddings, submersions, and matrix factorisations [18, 29]. An application closer to the setting of this paper appears in the work of Guralnick [23] and Leiterer [35] where the authors prove that holomorphic maps A,B:S→Mat⁡n(C)A,B:S \to\operatorname{Mat}_n(\mathbb{C}) from 1-dimensional Stein spaces SS which are locally holomorphically similar are globally holomorphically similar. Guralnick proved this when SS is an open Riemann surface and Leiterer extended it to arbitrary 1-dimensional Stein spaces. In [35], Leiterer exploited an Oka principle for Oka pairs (in the sense of Forster and Ramsptt [FR66]) to reduce the assertion that locally holomorphically similar matrices are globally holomorphically similar to a certain topological statement and then proved the topological statement. In [17 Theorem 6.14.9], the results of Guralnick and Leiterer are obtained via an application of an Oka principle for sections of branched holomorphic maps (cf. [17 Theorem 6.14.6]).

In [1], Alarcón and Forstnerič introduced the Oka-1 property which restricts the source manifolds to open Riemann surfaces. Thus, every Oka manifold is Oka-1. Roughly speaking, Oka-1 manifolds are characterised by the property that they admit many holomorphic maps from open Riemann surfaces. They proved that a complex manifold dominable by Cn\mathbb{C}^{n} at most points is Oka-1. Furthermore, they introduced a local spray approximation property which implies the Oka-1 property, and is one of the main tools used in this paper.

Idea of proof. We outline the key ideas of the proof of Theorem 1; the proof of Theorem 2 is analogous. Suppose first that GG is simply connected, and put r=rank⁡(G)r=\operatorname{rank}(G). The commuting variety R(G)R(G) is singular, so flexibility cannot be applied directly to the whole variety. Instead, we consider the open subset of pairs whose first entry is regular, and show that it lies in the smooth locus. The fibre of the first projection map to GG over a regular element xx of GG is its centraliser ZG(x)Z_G(x). Although the centraliser is abelian, it is not generally connected, so the exponential map from its Lie algebra only hits the identity component. We prove a structural result for ZG(x)Z_G(x) (Lemma 7) which says the remaining components can be reached by multiplication by powers of xx and central elements. This leads us to a family of local biholomorphisms

Φ(n,ε):Greg×Cr⟶Rreg(G),(n,ε)∈Z×Z(G),\Phi_{(n,\varepsilon)}:G_{\mathrm{reg}}\times\mathbb{C}^{r}\longrightarrow R_{\mathrm{reg}}(G),\qquad(n,\varepsilon)\in\mathbb{Z}\times Z(G),

whose images cover Rreg(G)R_{\mathrm{reg}}(G). These maps are constructed using differentials of the fundamental characters of GG [52 §8]. The source Greg×CrG_{\mathrm{reg}}\times\mathbb{C}^{r} is flexible and hence has the local spray approximation property (LSAP), recalled in § 2.1. The local biholomorphisms transfer this property to Rreg(G)R_{\mathrm{reg}}(G). The complement of Rreg(G)R_{\mathrm{reg}}(G) in the smooth locus of R(G)R(G) is an analytic subset of positive codimension, so a criterion [1 Proposition 7.13(b)] implies the whole smooth locus is Oka-1.

For a general semisimple group GG, we pass to the universal covering q:G~→Gq:\widetilde{G}\to G. Since R(G~)R(\widetilde{G}) is the inverse image of Rstr(G)R_{\mathrm{str}}(G) under q×qq\times q, restricting this covering to Rreg(G~)R_{\mathrm{reg}}(\widetilde{G}) allows us to transfer LSAP from Rreg(G~)R_{\mathrm{reg}}(\widetilde{G}) to the regular strongly commuting variety of GG. Again, using the criterion in [1 Proposition 7.13(b)] gives us the Oka-1 property for Rstr(G)smR_{\mathrm{str}}(G)_{\mathrm{sm}}.

The full Oka property. Results in the very recent preprint [2], which originated in [41], show that our proofs of Theorem 1 and Theorem 2 give a much stronger conclusion, namely the full Oka property. Indeed, the proof of Theorem 1 shows that we can find a family of local biholomorphisms Φ(n,ε):Greg×Cr→Rreg\Phi_{(n,\varepsilon)}:G_{\mathrm{reg}}\times\mathbb{C}^{r}\to R_{\mathrm{reg}} whose images cover Rreg(G)R_{\mathrm{reg}}(G) and where Greg×CrG_{\mathrm{reg}}\times\mathbb{C}^{r} is flexible and hence strongly dominable. It follows that Rreg(G)R_{\mathrm{reg}}(G) is strongly dominable and hence Oka by [2 Theorem 1.4]. Moreover, by [2 Theorem 1.8], the Oka property extends across analytic subsets of positive codimension. Similarly, for the strongly commuting variety Rstr(G)R_{\mathrm{str}}(G) of a semisimple group GG, there is a surjective local biholomorphism Rreg(G~)→Rstr,reg(G)R_{\mathrm{reg}}(\widetilde{G}) \to R_{\mathrm{str,reg}}(G) which makes Rstr,reg(G)R_{\mathrm{str,reg}}(G) strongly dominable and hence Oka by [2 Theorem 1.4]. Thus, the whole smooth locus Rstr(G)smR_{\mathrm{str}}(G)_{\mathrm{sm}} is Oka by [2 Theorem 1.8]. Similarly, the conclusions that the smooth loci of the commuting variety of a complex semisimple Lie algebra g\mathfrak{g} and the commuting variety of PSL⁡p(C)\operatorname{PSL}_{p}(\mathbb{C}) for prime pp are Oka-1 can be upgraded to the full Oka property by the same reasoning.

Outline of the paper. In § 2 we briefly discuss some preliminaries regarding the Oka-1 property and the local spray approximation property which is a key tool in the proofs of the main results. We briefly recall the definition of regular elements in a simply connected semisimple group GG and prove a structural result for the centralisers of such elements. Section 3 is devoted to the study of the regular commuting variety, a Zariski open subset of the commuting variety of GG that plays a central role in the proof. In § 3.1, we prove that the regular commuting variety is contained in the smooth locus of the commuting variety. We then define the local biholomorphisms in § 3.2 and use these to prove Theorem 1 in § 3.3 for the case of simply connected semisimple groups. In § 4, we define the strongly commuting variety and give four equivalent descriptions of which two are topological and two are algebraic. In § 4.1, we prove Theorem 1. In § 4.2, we drop the simple connectedness assumption by proving the Oka-1 property for the smooth locus of the commuting variety of PSL⁡p(C)\operatorname{PSL}_{p}(\mathbb{C}) when pp is prime. Finally, in § 5, we move to the Lie algebra case and prove Theorem 2.

Acknowledgements. The author would like to thank Finnur Lárusson for the many helpful discussions, for carefully reading the manuscript and providing feedback, and for pointing out the upgrade from the Oka-1 property to the full Oka property. The author is also grateful to David Baraglia for reading and providing helpful comments on this paper. This research was supported by the Australian Government Research Training Program.

Use of artificial intelligence. This paper was written entirely by the author. ChatGPT was used for finding literature, and GPT-5.6 was used during the preparation of the manuscript for proof-reading and checking mathematical arguments. In particular, it helped identify minor errors in Lemma 7 and Theorem 17.

Conventions. Throughout this paper, all algebraic groups, varieties and Lie algebras are defined over the field of complex numbers C\mathbb{C}. We use the convention that semisimple algebraic groups are connected.

Preliminaries

In this section, we recall the notions of holomorphic flexibility used in the paper and establish some of the key ingredients used in the proof of Theorem 1. We give a structural result for the centralisers of regular elements in a simply connected almost simple algebraic group, and recall how differentials of the fundamental characters of GG can provide a basis for their Lie algebras.

Holomorphic flexibility and Oka-1 manifolds

There are several precise notions of holomorphic flexibility. Here, we define the notions relevant to the paper. These notions pertain to smooth varieties and they include flexibility, (algebraic) ellipticity, the local spray approximation property, and the Oka-1 property. The notion of a flexible variety was introduced by Arzhantsev et al. [4] and further developed in [3].

Definition 3. An irreducible variety XX is flexible if for all smooth points x∈Xx \in X, the tangent space TxXT_xX is spanned by the tangent vectors to the orbits HxH_x of one-parameter unipotent subgroups H⊂Aut⁡(X)H \subset\operatorname{Aut}(X).

For an irreducible affine variety XX of dimension at least two, flexibility is equivalent to the subgroup SAut⁡(X)⊂Aut⁡(X)\operatorname{SAut}(X) \subset\operatorname{Aut}(X), generated by one-parameter unipotent subgroups, acting infinitely transitively on the smooth locus of XX [3]. Furthermore, it is shown in [3 Proposition A.3] that if a smooth irreducible variety is flexible then it admits a dominating spray, a geometric condition defined by Gromov [22] which implies the Oka principle for holomorphic sections of holomorphic fibre bundles over Stein manifolds.

Definition 4. Let XX be a complex manifold. A dominating spray on XX is a holomorphic vector bundle p:E→Xp : E \to X together with a holomorphic map s:E→Xs : E \to X such that ss restricts to the identity on the zero section ZZ, and for each x∈Z≅Xx \in Z \cong X the tangent map dxsd_xs sends the fibre Ex=p−1(x)E_x = p^{-1}(x) surjectively onto TxXT_xX. Furthermore, XX is called elliptic if it admits a dominating spray. We say that XX is algebraically elliptic if XX is a smooth algebraic variety and the holomorphic vector bundle, together with the dominating spray, are algebraic.

Every elliptic manifold is Oka (cf. [17 Corollary 5.6.14]) and hence Oka-1. The Oka-1 property defines a class of complex manifolds characterised by the property that they admit many holomorphic curves parameterised by open Riemann surfaces. This class of complex manifolds was introduced in [1] and further studied in [FL25]. Let us recall the definition.

Definition 5. A connected complex manifold XX is an Oka-1 manifold if for any open Riemann surface Σ\Sigma, compact Runge set KK in Σ\Sigma, a discrete sequence (ai)i∈I(a_i)_{i\in I} in Σ\Sigma without repetitions, continuous map f:Σ→Xf : \Sigma\to X which is holomorphic on a neighbourhood of K∪⋃i∈I{ai}K \cup\bigcup_{i\in I}\{a_i\}, distance function dd inducing the topology of XX, number ϵ>0\epsilon> 0, and integers ki≥1k_i \ge1 there is a holomorphic map F:Σ→XF : \Sigma\to X which is homotopic to ff and satisfies

  1. sup⁡p∈Kd(F(p),f(p))<ϵ\sup_{p\in K} d(F(p), f(p)) < \epsilon, and

  1. FF agrees with ff to order kik_i at the point aia_i for every ii.

We now introduce the local spray approximation property (LSAP). This property, introduced in [1], implies the Oka-1 condition [1]. The LSAP property plays an important role when we study the holomorphic flexibility properties of commuting varieties.

A pair of compact topological discs D⊂D′⊂CD \subset D' \subset\mathbb{C} is called a special pair if both have piecewise smooth boundaries and D′∖D˚D' \setminus\mathring{D} is a disc attached to DD along a boundary arc. A holomorphic map F:D×BN→XF : D \times\mathbb{B}^{N} \to X to a complex manifold XX is called a holomorphic spray of maps D→XD \to X with core f=F(⋅,0)f = F(\cdot,0), where BN\mathbb{B}^{N} denotes the unit ball in CN\mathbb{C}^{N}. Such a spray is said to be dominating if the partial derivative ∂∂t∣t=0F(z,t):CN→Tf(z)X\left.\frac{\partial}{\partial t}\right|_{t=0}F(z,t) : \mathbb{C}^{N} \to T_{f(z)}X is surjective for all z∈Dz \in D.

Definition 6. A complex manifold XX has the local spray approximation property (LSAP) at x∈Xx \in X if there is an open neighbourhood V⊂XV \subset X of xx satisfying the following condition. Given a special pair of compact discs D⊂D′⊂CD \subset D' \subset\mathbb{C} and dominating holomorphic spray F:D×BN→VF : D \times\mathbb{B}^{N} \to V, there is a number r∈(0,1)r \in(0,1) such that FF can be approximated uniformly on D×rBND \times r\mathbb{B}^{N} by holomorphic maps G:D′×rBN→XG : D' \times r\mathbb{B}^{N} \to X. A manifold XX has LSAP if the above condition holds at every x∈Xx \in X.

By [1 Proposition 7.13 (b)], every manifold with LSAP is Oka-1. It is not known whether every Oka-1 manifold has LSAP.

Regular elements and centralisers

Let GG be a semisimple simply connected linear algebraic group. Following [52], we say an element x∈Gx \in G is regular if its centraliser ZG(x)Z_G(x) in GG has minimal dimension equal to r:=rank⁡(G)r := \operatorname{rank}(G). In [52], it is shown that the subset GregG_{\mathrm{reg}} of regular elements is Zariski open and dense in GG. Moreover, since GG is assumed to be simply connected, ZG(x)Z_G(x) is abelian [34 Theorem D].

Centralisers of regular elements

Here we give a decomposition of the centraliser in GG of a regular element in a simply connected almost simple algebraic group GG, which is vital for the proof of Theorem 1. The corresponding result for simply connected semisimple groups follows by applying the lemma to each almost simple factor. The proof is a consequence of [51 Remark 3]; see also [38 Lemma 34]. For an algebraic group HH, we denote by H∘H^{\circ} the connected component of the identity of HH.

Lemma 7. Let GG be a simply connected, almost simple algebraic group and let x∈Gregx \in G_{\mathrm{reg}} be a regular element. Then

ZG(x)=⋃n∈ZxnZG(x)∘Z(G).Z_G(x) = \bigcup_{n \in\mathbb{Z}} x^n Z_G(x)^{\circ} Z(G).

Equivalently, the quotient

ZG(x)/(ZG(x)∘Z(G))Z_G(x)/(Z_G(x)^{\circ}Z(G))

is finite cyclic generated by the class of xx.

Proof. Write x=xsxu=xuxsx=x_sx_u=x_ux_s for the Jordan decomposition of xx, and let H:=ZG(xs)H:=Z_G(x_s) be the centraliser of xsx_s in GG. Using the uniqueness of the Jordan decomposition, it is readily checked that

ZG(x)=ZG(xs)∩ZG(xu)=ZH(xu).Z_G(x)=Z_G(x_s)\cap Z_G(x_u)=Z_H(x_u).

Further, since x∈Gregx\in G_{\mathrm{reg}}, the unipotent part xux_u is a regular unipotent element of HH [13 Section 1.14], and since GG is simply connected, [28 Theorem 2.11] shows that HH is a connected reductive subgroup of maximal rank. Since xu∈Hx_u\in H is regular, it lies in a unique Borel subgroup B=TUB=TU of HH, where TT is the maximal torus and UU the unipotent radical of BB. It follows from [28 Proposition 4.1] that ZH(xu)=ZB(xu)Z_H(x_u)=Z_B(x_u) and ZB(xu)=Z(H)ZU(xu)Z_B(x_u)=Z(H)Z_U(x_u). Thus,

ZG(x)=ZH(xu)=Z(H)ZU(xu).Z_G(x)=Z_H(x_u)=Z(H)Z_U(x_u).

Consider the quotient map π:G→Gad:=G/Z(G)\pi:G\to G_{\mathrm{ad}}:=G/Z(G) and put H‾=ZGad(xs‾)∘\overline{H}=Z_{G_{\mathrm{ad}}}(\overline{x_s})^\circ. Since H=ZG(xs)H=Z_G(x_s) is connected and π\pi is a central isogeny, we have π(H)=H‾\pi(H)=\overline{H}. Moreover,

πZ:=π∣Z(H):Z(H)⟶Z(H‾)\pi_Z:=\left.\pi\right|_{Z(H)}:Z(H)\longrightarrow Z(\overline{H})

is surjective. Indeed, if h∈Hh\in H and π(h)∈Z(H‾)\pi(h)\in Z(\overline{H}) then the homomorphism

H⟶Z(G),a⟼[h,a]H\longrightarrow Z(G),\qquad a\longmapsto[h,a]

is constant, because HH is connected and Z(G)Z(G) is finite, and its value at the identity is 11. Hence [h,a]=1[h,a]=1 for every a∈Ha\in H, so h∈Z(H)h\in Z(H). By [51 Remark 3], the quotient

Z(H‾)/Z(H‾)∘Z(\overline{H})/Z(\overline{H})^\circ

is a finite cyclic group generated by the class xs‾⋅Z(H‾)∘\overline{x_s}\cdot Z(\overline{H})^\circ. We claim that

Z(H)/Z(H)∘Z(G)≅Z(H‾)/Z(H‾)∘.Z(H)/Z(H)^\circ Z(G)\cong Z(\overline{H})/Z(\overline{H})^\circ.

Indeed, consider the map

Z(H)→ πZ Z(H‾)⟶Z(H‾)/Z(H‾)∘.Z(H)\xrightarrow{\ \pi_Z\ }Z(\overline{H})\longrightarrow Z(\overline{H})/Z(\overline{H})^\circ.

It is surjective with kernel given by πZ−1(Z(H‾)∘)=Z(H)∘Z(G)\pi_Z^{-1}(Z(\overline{H})^\circ)=Z(H)^\circ Z(G). Thus,

Z(H)=⋃n∈ZxsnZ(H)∘Z(G).Z(H)=\bigcup_{n\in\mathbb{Z}}x_s^n Z(H)^\circ Z(G).

Substituting this decomposition into ZG(x)=Z(H)ZU(xu)Z_G(x)=Z(H)Z_U(x_u), we obtain

ZG(x)=Z(H)ZU(xu)=⋃n∈ZxsnZ(H)∘ZU(xu)Z(G).\begin{aligned} Z_G(x)&=Z(H)Z_U(x_u)\\ &=\bigcup_{n\in\mathbb{Z}}x_s^nZ(H)^\circ Z_U(x_u)Z(G). \end{aligned}

Since Z(H)∘Z(H)^\circ and ZU(xu)Z_U(x_u) are connected subgroups of ZG(x)∘Z_G(x)^\circ, their product is contained in ZG(x)∘Z_G(x)^\circ. Therefore

Z(H)∘ZU(xu)Z(G)⊂ZG(x)∘Z(G).Z(H)^\circ Z_U(x_u)Z(G) \subset Z_G(x)^\circ Z(G).

It follows that

ZG(x)⊂⋃n∈ZxsnZG(x)∘Z(G).Z_G(x) \subset\bigcup_{n\in\mathbb{Z}} x_s^n Z_G(x)^\circ Z(G).

The reverse inclusion is clear. We conclude that

ZG(x)=⋃n∈ZxsnZG(x)∘Z(G).Z_G(x) = \bigcup_{n\in\mathbb{Z}} x_s^n Z_G(x)^\circ Z(G).

Finally, we may replace xsx_s by xx in this decomposition. Indeed, since xxs−1=xux x_s^{-1}=x_u and

xu∈ZU(xu)⊂ZG(x)∘,x_u \in Z_U(x_u) \subset Z_G(x)^\circ,

the images of xx and xsx_s in the quotient

ZG(x)/(ZG(x)∘Z(G))Z_G(x)/(Z_G(x)^\circ Z(G))

coincide. Hence

ZG(x)=⋃n∈ZxnZG(x)∘Z(G).Z_G(x) = \bigcup_{n\in\mathbb{Z}} x^n Z_G(x)^\circ Z(G).

Equivalently, the quotient

ZG(x)/ZG(x)∘Z(G)Z_G(x)/Z_G(x)^\circ Z(G)

is generated by the class of xx. □

Remark 8. When G=SLn(C)G=\mathrm{SL}_n(\mathbb{C}) and x∈Gx\in G is regular, the quotient

ZG(x)/ZG(x)∘Z(G)Z_G(x)/Z_G(x)^\circ Z(G)

is trivial. Indeed, in the notation of the proof of the lemma, we saw above that

ZG(x)=Z(H)ZU(xu).Z_G(x)=Z(H)Z_U(x_u).

As GG is of type AA, the pseudo-Levi subgroup H=ZG(xs)H=Z_G(x_s) is actually a Levi subgroup. By [12 Proposition 3.5] this is equivalent to Z(H)=Z(H)∘Z(G)Z(H)=Z(H)^\circ Z(G). We saw above that

ZG(x)=Z(H)ZU(xu)=Z(H)∘Z(G)ZU(xu)Z_G(x)=Z(H)Z_U(x_u)=Z(H)^\circ Z(G)Z_U(x_u)

and that Z(H)∘ZU(xu)⊂ZG(x)∘Z(H)^\circ Z_U(x_u)\subset Z_G(x)^\circ, hence ZG(x)⊂ZG(x)∘Z(G)Z_G(x)\subset Z_G(x)^\circ Z(G) and the quotient

ZG(x)/ZG(x)∘Z(G)Z_G(x)/Z_G(x)^\circ Z(G)

is trivial.

We provide an example in which this quotient is non-trivial.

Example 9. Let G=Sp⁡4(C)={g∈SL⁡4(C)∣gtAg=A}G=\operatorname{Sp}_{4}(\mathbb{C})=\{g\in\operatorname{SL}_{4}(\mathbb{C})\mid g^{t}Ag=A\}, where

A=(0100−1000000100−10).A=\begin{pmatrix}0&1&0&0\\-1&0&0&0\\0&0&0&1\\0&0&-1&0\end{pmatrix}.

Consider the semisimple element

s=(1000010000−10000−1)∈T={(t10000t1−10000t20000t2−1):t1,t2∈C×}.s=\begin{pmatrix}1&0&0&0\\0&1&0&0\\0&0&-1&0\\0&0&0&-1\end{pmatrix}\in T=\left\{\begin{pmatrix}t_{1}&0&0&0\\0&t_{1}^{-1}&0&0\\0&0&t_{2}&0\\0&0&0&t_{2}^{-1}\end{pmatrix}:t_{1},t_{2}\in\mathbb{C}^{\times}\right\}.

The centraliser of ss is given by H=ZG(s)≃SL⁡2(C)×SL⁡2(C)H=Z_{G}(s)\simeq\operatorname{SL}_{2}(\mathbb{C})\times\operatorname{SL}_{2}(\mathbb{C}). Thus, the centre of HH is

Z(H)≃μ2×μ2,Z(H)\simeq\mu_{2}\times\mu_{2},

where μn\mu_{n} denotes the group of nthn^{\text{th}}-roots of unity. Thus, the identity component Z(H)∘Z(H)^{\circ} is trivial. However, the centre of GG is

Z(G)={I2⊕I2,−I2⊕−I2}≃μ2,Z(G)=\{I_{2}\oplus I_{2},-I_{2}\oplus-I_{2}\}\simeq\mu_{2},

where I2I_{2} denotes the 22-by-2 identity matrix. Thus,

Z(H)/(Z(H)∘Z(G))=Z(H)/Z(G)≃μ2,Z(H)/(Z(H)^{\circ}Z(G))=Z(H)/Z(G)\simeq\mu_{2},

and the class of ss generates this quotient. Choose the regular unipotent element

u=(1100010000110001)u=\begin{pmatrix}1&1&0&0\\0&1&0&0\\0&0&1&1\\0&0&0&1\end{pmatrix}

of HH and let a=sua=su. If UU is the unipotent radical of the unique Borel subgroup of HH containing uu, then

ZG(a)=Z(H)ZU(u)Z_{G}(a)=Z(H)Z_{U}(u)

and

ZU(u)={(1c01)⊕(1d01)∈U∣c,d∈C}.Z_{U}(u)=\left\{\begin{pmatrix}1&c\\0&1\end{pmatrix}\oplus\begin{pmatrix}1&d\\0&1\end{pmatrix}\in U\mid c,d\in\mathbb{C}\right\}.

In particular, aa is regular in GG since dim⁡(ZG(a))=dim⁡(ZU(u))=2\dim(Z_{G}(a))=\dim(Z_{U}(u))=2. Moreover, ZG(a)Z_{G}(a) has four connected components corresponding to the elements in the centre of HH,

Z(H)={(I2⊕I2),(I2⊕−I2),(−I2⊕I2),(−I2⊕−I2)}.Z(H)=\{(I_{2}\oplus I_{2}),(I_{2}\oplus-I_{2}),(-I_{2}\oplus I_{2}),(-I_{2}\oplus-I_{2})\}.

To see that ZG(x)≠ZG(x)∘Z(G)Z_{G}(x)\ne Z_{G}(x)^{\circ}Z(G) in general, let I2⊕I2I_{2}\oplus I_{2} and −I2⊕−I2-I_{2}\oplus-I_{2} be the central elements of GG. Then the union

(I2⊕I2)ZU(u)∪(−I2⊕−I2)ZU(u)=ZG(a)∘Z(G)(I_{2}\oplus I_{2})Z_{U}(u)\cup(-I_{2}\oplus-I_{2})Z_{U}(u)=Z_{G}(a)^{\circ}Z(G)

does not cover all of ZG(a)Z_G(a), as it does not contain the connected components represented by (I2⊕−I2)(I_2 \oplus-I_2) and (−I2⊕I2)(-I_2 \oplus I_2). Since a=sua=su and u∈ZG(a)∘u\in Z_G(a)^\circ, the classes of aa and ss in

ZG(a)/(ZG(a)∘Z(G))Z_G(a)/(Z_G(a)^\circ Z(G))

coincide. Thus, multiplication by aa gives the remaining coset

ZG(a)=ZG(a)∘Z(G)∪a⋅ZG(a)∘Z(G).Z_G(a)=Z_G(a)^\circ Z(G)\cup a\cdot Z_G(a)^\circ Z(G).

Differentials of invariant functions

Recall that the action of GG on itself by conjugation induces an action of GG on its ring of functions. We denote by C[G]G\mathbb{C}[G]^G the corresponding ring of invariants. For f∈C[G]Gf\in\mathbb{C}[G]^G, the differential dfg:TgG→Cdf_g:T_gG\to\mathbb{C}, after a translation, defines an element in the dual space g∗\mathfrak{g}^* given by

dfg(X)=ddt∣t=0f(gexp⁡(tX)).df_g(X)=\left.\frac{d}{dt}\right|_{t=0}f(g\exp(tX)).

Since we assume GG is semisimple, the Killing form κ\kappa gives an isomorphism g→g∗\mathfrak{g}\to\mathfrak{g}^*. Let Fg:=κ−1(dfg)∈gF_g:=\kappa^{-1}(df_g)\in\mathfrak{g} be the element in g\mathfrak{g} corresponding to dfgdf_g under this isomorphism. The element FgF_g is also defined by the formula

κ(Fg,X)=ddt∣t=0f(gexp⁡(tX)),for all X∈g.\kappa(F_g,X)=\left.\frac{d}{dt}\right|_{t=0}f(g\exp(tX)),\qquad\text{for all }X\in\mathfrak{g}.

The following is stated in [21 page 264]; we prove it for completeness:

Proposition 10. Let GG be a semisimple algebraic group. Given g∈Gg\in G and an invariant function f∈C[G]Gf\in\mathbb{C}[G]^G, the corresponding element FgF_g lies in the Lie algebra Lie⁡(ZG(g))\operatorname{Lie}(Z_G(g)) of the centraliser ZG(g)Z_G(g).

Proof. Recall that

Lie⁡(ZG(g))={α∈g∣Ad⁡g(α)=α}=ker⁡(Ad⁡g−I:g→g).\operatorname{Lie}(Z_G(g))=\{\alpha\in\mathfrak{g}\mid\operatorname{Ad}_g(\alpha)=\alpha\}=\ker(\operatorname{Ad}_g-I:\mathfrak{g}\to\mathfrak{g}).

It suffices to show that

Fg∈(im⁡(Ad⁡g−I))⊥=ker⁡(Ad⁡g−I),F_g\in\bigl(\operatorname{im}(\operatorname{Ad}_g-I)\bigr)^\perp=\ker(\operatorname{Ad}_g-I),

where (−)⊥(-)^\perp denotes the orthogonal complement with respect to the form κ\kappa. To this end, we compute

κ(Fg,Ad⁡gα−α)=κ(Fg,Ad⁡gα)−κ(Fg,α).\kappa(F_g,\operatorname{Ad}_g\alpha-\alpha)=\kappa(F_g,\operatorname{Ad}_g\alpha)-\kappa(F_g,\alpha).

We are done if we show the right hand side is zero. We have

κ(Fg,Ad⁡gα)=ddt∣t=0f(gexp⁡(tAd⁡g(α)))=ddt∣t=0f(ggexp⁡(tα)g−1)\kappa(F_g,\operatorname{Ad}_g\alpha)=\left.\frac{d}{dt}\right|_{t=0}f(g\exp(t\operatorname{Ad}_g(\alpha)))=\left.\frac{d}{dt}\right|_{t=0}f(gg\exp(t\alpha)g^{-1})
=ddt∣t=0f(gexp⁡(tα))=κ(Fg,α),=\left.\frac{d}{dt}\right|_{t=0}f(g\exp(t\alpha))=\kappa(F_g,\alpha),

and the statement follows. □\square

Example 11. Let G=SLn(C)G = \mathrm{SL}_n(\mathbb{C}), f=tr⁡f = \operatorname{tr}, and g∈Gg \in G. We will show that Fg=g−tr⁡(g)nI∈Lie⁡(ZG(g))F_g = g - \frac{\operatorname{tr}(g)}{n}I \in\operatorname{Lie}(Z_G(g)). Instead of using the Killing form, in this example we use the non-degenerate invariant bilinear form κ(X,Y)=tr⁡(XY)\kappa(X,Y) = \operatorname{tr}(XY). We have κ(Fg,X)=tr⁡(FgX)\kappa(F_g,X) = \operatorname{tr}(F_gX) and, on the other hand, we have

κ(Fg,X)=ddt∣t=0f(gexp⁡(tX))=tr⁡(gX),for all X∈sln(C).\kappa(F_g,X) = \left.\frac{d}{dt}\right|_{t=0} f(g\exp(tX)) = \operatorname{tr}(gX), \qquad\text{for all } X \in\mathfrak{sl}_n(\mathbb{C}).

Thus,

0=tr⁡(FgX)−tr⁡(gX)=tr⁡((Fg−g)X).0 = \operatorname{tr}(F_gX) - \operatorname{tr}(gX) = \operatorname{tr}((F_g-g)X).

Hence we conclude that Fg−gF_g-g is orthogonal to sln(C)\mathfrak{sl}_n(\mathbb{C}). Using the standard decomposition gln(C)=sln(C)⊕CIn\mathfrak{gl}_n(\mathbb{C}) = \mathfrak{sl}_n(\mathbb{C}) \oplus\mathbb{C}I_n, the orthogonal complement of sln(C)\mathfrak{sl}_n(\mathbb{C}) with respect to the trace form is CIn\mathbb{C}I_n. Thus,

Fg−g=λInF_g-g = \lambda I_n

for some λ∈C\lambda\in\mathbb{C}. Taking traces on both sides, one finds that λ=−tr⁡(g)n\lambda= -\frac{\operatorname{tr}(g)}{n}. Thus,

Fg=g−tr⁡(g)nIn.F_g = g - \frac{\operatorname{tr}(g)}{n}I_n.

Since GG is semisimple, its algebra of invariant functions is a polynomial algebra

C[G]G=C[χ1,…,χr]\mathbb{C}[G]^G = \mathbb{C}[\chi_1,\ldots,\chi_r]

where r=rank⁡(G)r = \operatorname{rank}(G) and the χi\chi_i are the characters of the fundamental representations (cf. [52]). For instance, when G=SLn(C)G = \mathrm{SL}_n(\mathbb{C}), the generators are the coefficients of the characteristic polynomial. We end this section with a proposition due to Steinberg which will be useful for defining a basis for the Lie algebras of the centralisers of regular elements:

Theorem 12. [[52 Theorem 8.1]] Let GG be a semisimple simply connected algebraic group. An element x∈Gx \in G is regular if and only if the differentials of the fundamental characters dχ1,…,dχrd\chi_1,\ldots,d\chi_r are linearly independent at xx.

Regular commuting variety

In this section, we prove that the regular commuting variety lies inside the smooth locus of the commuting variety and that it has LSAP, and deduce the Oka-1 property of the smooth locus.

Smoothness of the regular commuting variety

Let GG be a simply connected semisimple algebraic group and let R=R(G)R = R(G) denote the commuting variety defined as

R={(x,y)∈G×G∣xy=yx}.R = \{(x,y) \in G \times G \mid xy = yx\}.

There is a projection map from RR to GG:

π:R⟶G,(x,y)⟼x.\pi: R \longrightarrow G, \qquad(x,y) \longmapsto x.

Consider the inverse image under π\pi of the dense open subset of regular elements Greg⊂GG_{\mathrm{reg}} \subset G

Rreg:=π−1(Greg)={(x,y)∈R∣x∈Greg}.R_{\mathrm{reg}} := \pi^{-1}(G_{\mathrm{reg}}) = \{(x,y) \in R \mid x \in G_{\mathrm{reg}}\}.

We call the open subset Rreg⊂RR_{\mathrm{reg}} \subset R the regular commuting variety. Furthermore, RregR_{\mathrm{reg}} is irreducible since it is an open subset of the irreducible variety RR [46]. Now we have a surjective map π∣Rreg:Rreg→Greg\left.\pi\right|_{R_{\mathrm{reg}}} : R_{\mathrm{reg}} \to G_{\mathrm{reg}} between irreducible varieties whose fibre over each point x∈Gregx \in G_{\mathrm{reg}} has dimension equal to rank⁡(G)\operatorname{rank}(G). By [48 §6.3, Theorem 1.25] it follows that

dim⁡R=dim⁡Rreg=dim⁡(G)+rank⁡(G).\dim R = \dim R_{\mathrm{reg}} = \dim(G) + \operatorname{rank}(G).

We identify TxGT_xG and TyGT_yG with g=Lie⁡(G)\mathfrak{g} = \operatorname{Lie}(G) by right translation, and hence T(x,y)(G×G)T_{(x,y)}(G \times G) with g2\mathfrak{g}^2. Consider the set

K(x,y)={(α,β)∈g2∣Ad⁡x(β)−β=Ad⁡y(α)−α}.K_{(x,y)} = \{(\alpha,\beta) \in\mathfrak{g}^2 \mid\operatorname{Ad}_x(\beta) - \beta= \operatorname{Ad}_y(\alpha) - \alpha\}.

Then differentiating the commuting equation xy=yxxy = yx gives T(x,y)R⊆K(x,y)T_{(x,y)}R \subseteq K_{(x,y)}. We have the following proposition:

Proposition 13. For every (x,y)∈Rreg(x,y) \in R_{\mathrm{reg}} the projection

p:K(x,y)⟶TxGp : K_{(x,y)} \longrightarrow T_xG

onto the first coordinate is surjective. Moreover, the point (x,y)(x,y) is smooth.

Proof. To show that pp is surjective, we need, for a given α∈g\alpha\in\mathfrak{g}, to show that there is β∈g\beta\in\mathfrak{g} with

Ad⁡x(β)−β=Ad⁡y(α)−α.\operatorname{Ad}_x(\beta) - \beta= \operatorname{Ad}_y(\alpha) - \alpha.

Consider the linear map

Tx:g⟶g,β⟼Ad⁡x(β)−β.T_x : \mathfrak{g} \longrightarrow\mathfrak{g}, \qquad\beta\longmapsto\operatorname{Ad}_x(\beta) - \beta.

It suffices to show that Ad⁡y(α)−α\operatorname{Ad}_y(\alpha) - \alpha lies in im⁡(Tx)\operatorname{im}(T_x). Let κ\kappa denote the Killing form on g\mathfrak{g}, which is non-degenerate because g\mathfrak{g} is semisimple. By basic linear algebra,

(im⁡Tx)⊥=ker⁡(Tx∗),(\operatorname{im} T_x)^\perp= \ker(T_x^*),

where (im⁡Tx)⊥(\operatorname{im} T_x)^\perp denotes the orthogonal complement of im⁡(Tx)\operatorname{im}(T_x) and Tx∗T_x^* the adjoint with respect to κ\kappa.

We claim that Tx∗=Tx−1T_x^* = T_{x^{-1}}. Indeed, using the Ad-invariance of κ\kappa, for β,γ∈g\beta,\gamma\in\mathfrak{g} we compute

κ(Tx(β),γ)=κ(Ad⁡x(β)−β,γ)=κ(β,Ad⁡x−1(γ)−γ)=κ(β,Tx−1(γ)),\kappa(T_x(\beta),\gamma) = \kappa(\operatorname{Ad}_x(\beta) - \beta,\gamma) = \kappa(\beta,\operatorname{Ad}_{x^{-1}}(\gamma) - \gamma) = \kappa(\beta,T_{x^{-1}}(\gamma)),

so Tx∗=Tx−1T_x^* = T_{x^{-1}}. Hence,

im⁡(Tx)=(ker⁡Tx−1)⊥=(ker⁡Tx)⊥,\operatorname{im}(T_x) = (\ker T_{x^{-1}})^\perp= (\ker T_x)^\perp,

where the last equality holds because Ad⁡x(β)=β\operatorname{Ad}_x(\beta) = \beta if and only if $\operatorname{Ad}_{x^{-1}}(\beta) = \beta. Therefore, it suffices to show that

κ(z,Ad⁡y(α)−α)=0for all z∈ker⁡(Tx).\kappa(z,\operatorname{Ad}_{y}(\alpha)-\alpha)=0 \quad\text{for all } z\in\ker(T_x).

But for z∈ker⁡(Tx)z\in\ker(T_x), using Ad⁡\operatorname{Ad}-invariance of κ\kappa,

κ(z,Ad⁡y(α)−α)=κ(z,Ad⁡y(α))−κ(z,α)=κ(Ad⁡y−1(z)−z,α).\kappa(z,\operatorname{Ad}_{y}(\alpha)-\alpha)=\kappa(z,\operatorname{Ad}_{y}(\alpha))-\kappa(z,\alpha)=\kappa(\operatorname{Ad}_{y^{-1}}(z)-z,\alpha).

Finally, since x∈Gregx\in G_{\mathrm{reg}}, its centraliser ZG(x)Z_G(x) in GG is abelian [34 Theorem D]. As yy commutes with xx, we have y∈ZG(x)y\in Z_G(x), so yy acts trivially on Lie⁡(ZG(x))=ker⁡(Tx)\operatorname{Lie}(Z_G(x))=\ker(T_x). Hence, Ad⁡y−1(z)=z\operatorname{Ad}_{y^{-1}}(z)=z for all z∈ker⁡(Tx)z\in\ker(T_x), giving

κ(Ad⁡y−1(z)−z,α)=0.\kappa(\operatorname{Ad}_{y^{-1}}(z)-z,\alpha)=0.

This proves that Ad⁡y(α)−α∈im⁡(Tx)\operatorname{Ad}_{y}(\alpha)-\alpha\in\operatorname{im}(T_x), so pp is surjective. Consequently,

dim⁡(K(x,y))=dim⁡G+dim⁡(ker⁡(p))=dim⁡G+rank⁡(G).\dim(K_{(x,y)})=\dim G+\dim(\ker(p))=\dim G+\operatorname{rank}(G).

Moreover, for (x,y)∈Rreg(x,y)\in R_{\mathrm{reg}} we have

dim⁡R≤dim⁡(T(x,y)R)≤dim⁡(K(x,y))=dim⁡R,\dim R\leq\dim(T_{(x,y)}R)\leq\dim(K_{(x,y)})=\dim R,

so (x,y)(x,y) is a smooth point in RR and thus RregR_{\mathrm{reg}} is contained in the smooth locus of RR. □

Local parameterisation of the regular commuting variety

In this section, we give local parameterisations for the open subset RregR_{\mathrm{reg}} of the commuting variety of a simply connected almost simple algebraic group GG. The more general simply connected semisimple case will follow by taking products over the almost simple factors. For x∈Gx\in G, we write z(x):=Lie⁡(ZG(x))\mathfrak{z}(x):=\operatorname{Lie}(Z_G(x)) for the Lie algebra of its centraliser.

Let F1,…,Fr:Greg→gF_1,\ldots,F_r:G_{\mathrm{reg}}\to\mathfrak{g} be holomorphic maps associated to a set of generators for the invariant functions on GG, so that Fx,i:=Fi(x)F_{x,i}:=F_i(x), for i=1,…,ri=1,\ldots,r, form a basis for z(x)\mathfrak{z}(x) (cf. § 2.2.2).

Lemma 14. Let GG be a simply connected almost simple algebraic group of rank rr. For a fixed integer nn and ε∈Z(G)\varepsilon\in Z(G), the map

Φ(n,ε):Greg×Cr⟶Rreg,(x,λ1,…,λr)⟼(x,εxnexp⁡(∑i=1rλiFx,i))\Phi_{(n,\varepsilon)}:G_{\mathrm{reg}}\times\mathbb{C}^{r}\longrightarrow R_{\mathrm{reg}},\qquad(x,\lambda_1,\ldots,\lambda_r)\longmapsto\left(x,\varepsilon x^n\exp\left(\sum_{i=1}^{r}\lambda_iF_{x,i}\right)\right)

is a local biholomorphism onto its image.

Proof. Since

dim⁡Rreg=dim⁡G+rank⁡(G)=dim⁡(Greg×Cr),\dim R_{\mathrm{reg}}=\dim G+\operatorname{rank}(G)=\dim(G_{\mathrm{reg}}\times\mathbb{C}^{r}),

by the holomorphic inverse function theorem, it suffices to show that the differential of Φ(n,ε)\Phi_{(n,\varepsilon)} is injective at every point of Greg×CrG_{\mathrm{reg}}\times\mathbb{C}^{r}.

Fix ε∈Z(G)\varepsilon\in Z(G) and n∈Zn\in\mathbb{Z}. For ease of notation, write

Φ=Φ(n,ε)and(x,λ):=(x,λ1,…,λr).\Phi=\Phi_{(n,\varepsilon)}\quad\text{and}\quad(x,\lambda):=(x,\lambda_1,\ldots,\lambda_r).

Let (α,β1,…,βr)∈TxGreg⊕Cr(\alpha,\beta_1,\ldots,\beta_r)\in T_xG_{\mathrm{reg}}\oplus\mathbb{C}^r and suppose that the differential at (x,λ)∈Greg×Cr(x,\lambda)\in G_{\mathrm{reg}}\times\mathbb{C}^r satisfies

dΦ(x,λ)(α,β1,…,βr)=0.d\Phi_{(x,\lambda)}(\alpha,\beta_1,\ldots,\beta_r)=0.

The restriction of Φ\Phi to the first coordinate is the identity map on GregG_{\mathrm{reg}}, and hence so is its differential. Then, by assumption, we must have α=0\alpha=0. In the second coordinate, all terms involving d(xn)x(α)d(x^n)_x(\alpha) and (dFi)x(α)(dF_i)_x(\alpha) vanish. Thus the second component of dΦ(x,λ)(α,β1,…,βr)d\Phi_{(x,\lambda)}(\alpha,\beta_1,\ldots,\beta_r) reduces to

εxndexp⁡A(x,λ)(∑i=1rβiFx,i)=0.\varepsilon x^n d\exp_{A(x,\lambda)}\left(\sum_{i=1}^{r}\beta_iF_{x,i}\right)=0.

We now compute the differential dexp⁡A(x,λ)(∑i=1rβiFx,i)d\exp_{A(x,\lambda)}\left(\sum_{i=1}^{r}\beta_iF_{x,i}\right). Write

A(x,λ):=∑i=1rλiFx,i∈z(x)andB:=∑i=1rβiFx,iA(x,\lambda):=\sum_{i=1}^{r}\lambda_iF_{x,i}\in\mathfrak{z}(x) \qquad\text{and}\qquad B:=\sum_{i=1}^{r}\beta_iF_{x,i}

and observe that A(x,λ)A(x,\lambda) and BB both lie in z(x)\mathfrak{z}(x) (cf. Proposition 10). Since x∈Gregx\in G_{\mathrm{reg}} and GG is simply connected, the Lie algebra z(x)\mathfrak{z}(x) is abelian ([52 Remark 3.13] or [34 Theorem D]). In particular, A(x,λ)A(x,\lambda) and BB commute, so we have

exp⁡(A(x,λ)+tB)=exp⁡(A(x,λ))exp⁡(tB).\exp(A(x,\lambda)+tB)=\exp(A(x,\lambda))\exp(tB).

So

dexp⁡A(x,λ)(B)=ddt∣t=0exp⁡(A(x,λ)+tB)=exp⁡(A(x,λ))B.d\exp_{A(x,\lambda)}(B)=\left.\frac{d}{dt}\right|_{t=0}\exp(A(x,\lambda)+tB)=\exp(A(x,\lambda))B.

In particular, we get

εxnexp⁡(A(x,λ))B=0.\varepsilon x^n\exp(A(x,\lambda))B=0.

Since εxnexp⁡(A(x,λ))\varepsilon x^n\exp(A(x,\lambda)) is invertible, it follows that B=0B=0, that is,

∑i=1rβiFx,i=0.\sum_{i=1}^{r}\beta_iF_{x,i}=0.

By Theorem 12, Fx,1,…,Fx,rF_{x,1},\ldots,F_{x,r} are linearly independent in z(x)\mathfrak{z}(x), so βi=0\beta_i=0 for all i=1,…,ri=1,\ldots,r. Thus

(α,β1,…,βr)=0,(\alpha,\beta_1,\ldots,\beta_r)=0,

and hence dΦ(x,λ)d\Phi_{(x,\lambda)} is injective. □

Lemma 15. Let GG be a simply connected almost simple algebraic group. Then

Rreg=⋃(n,ε)∈Z×Z(G)im⁡(Φ(n,ε)).R_{\mathrm{reg}}=\bigcup_{(n,\varepsilon)\in\mathbb{Z}\times Z(G)}\operatorname{im}(\Phi_{(n,\varepsilon)}).

Proof. Let (x,y)∈Rreg(x,y)\in R_{\mathrm{reg}}. Then x∈Gregx\in G_{\mathrm{reg}} and y∈ZG(x)y\in Z_G(x). By Lemma 7, there exist n∈Zn\in\mathbb{Z}, ε∈Z(G)\varepsilon\in Z(G), and h∈ZG(x)∘h\in Z_G(x)^\circ such that y=xnεhy=x^n\varepsilon h. We will show that hh lies in the image of exp⁡:z(x)→ZG(x)∘\exp:\mathfrak{z}(x)\to Z_G(x)^\circ. Since ZG(x)∘Z_G(x)^\circ is a connected commutative linear algebraic group, it is isomorphic to a product (C×)s×U(\mathbb{C}^{\times})^s\times U, where UU is a connected commutative unipotent group [13 Section 1.6]. The exponential maps from the Lie algebras Lie⁡(C×)\operatorname{Lie}(\mathbb{C}^{\times}) and Lie⁡(U)\operatorname{Lie}(U) to the respective groups C×\mathbb{C}^{\times} and UU are surjective. Hence, the exponential map

exp⁡:z(x)⟶ZG(x)∘\exp: \mathfrak{z}(x) \longrightarrow Z_G(x)^\circ

is surjective. Therefore there exists an α∈z(x)\alpha\in\mathfrak{z}(x) with h=exp⁡(α)h=\exp(\alpha). Since Fx,1,…,Fx,rF_{x,1},\ldots,F_{x,r} form a basis of z(x)\mathfrak{z}(x), we can write

α=∑i=1rλiFx,i\alpha=\sum_{i=1}^{r}\lambda_iF_{x,i}

for some λ1,…,λr∈C\lambda_1,\ldots,\lambda_r\in\mathbb{C} and, in particular,

y=xnεexp⁡(∑i=1rλiFx,i).y=x^n\varepsilon\exp\left(\sum_{i=1}^{r}\lambda_iF_{x,i}\right).

Hence

(x,y)=Φ(n,ε)(x,λ1,…,λr).(x,y)=\Phi_{(n,\varepsilon)}(x,\lambda_1,\ldots,\lambda_r).

Thus, the images of the maps Φ(n,ε)\Phi_{(n,\varepsilon)} cover RregR_{\mathrm{reg}}.

We will require a special case of the following lemma which says that a complex manifold covered by images of local biholomorphisms from LSAP manifolds has LSAP.

Lemma 16. Let (Xα)α∈I(X_\alpha)_{\alpha\in I} be a family of complex manifolds, each having LSAP, and let YY be another complex manifold. Suppose further for each α\alpha we have a collection {Φα,β:Xα→Y}β∈J\{\Phi_{\alpha,\beta}:X_\alpha\to Y\}_{\beta\in J} of local biholomorphisms indexed by a set JJ such that

⋃(α,β)∈I×Jim⁡(Φα,β)=Y.\bigcup_{(\alpha,\beta)\in I\times J}\operatorname{im}(\Phi_{\alpha,\beta})=Y.

Then YY has the LSAP property.

Proof. Let y0∈Yy_0\in Y. By the assumption in the lemma, there are α∈I\alpha\in I, β∈J\beta\in J, x0∈Xαx_0\in X_\alpha and neighbourhoods U⊂XαU\subset X_\alpha of x0x_0 and W⊂YW\subset Y of y0y_0 such that

Φα,β:U⟶W\Phi_{\alpha,\beta}:U\longrightarrow W

is a biholomorphism. Since XαX_\alpha has LSAP, after possibly shrinking UU, we may assume that UU is an LSAP neighbourhood of x0x_0. We claim that WW is an LSAP neighbourhood of y0y_0. Let D⊂D′⊂CD\subset D'\subset\mathbb{C} be a special pair of compact discs, and let

H:D×BN⟶WH:D\times\mathbb{B}^{N}\longrightarrow W

be a dominating holomorphic spray. Composing with Φα,β−1\Phi_{\alpha,\beta}^{-1}, we obtain a dominating holomorphic spray

H~:=Φα,β−1∘H:D×BN⟶U⊂Xα.\widetilde{H}:=\Phi_{\alpha,\beta}^{-1}\circ H:D\times\mathbb{B}^{N}\longrightarrow U\subset X_\alpha.

By the LSAP property for X~α\widetilde{X}_{\alpha}, there exists r=r(H~)∈(0,1)r=r(\widetilde{H})\in(0,1) such that H~\widetilde{H} can be approximated uniformly on D×rB‾ND\times r\overline{B}^{N} by holomorphic maps

G~:D′×rB‾N⟶Xα.\widetilde{G}:D'\times r\overline{B}^{N}\longrightarrow X_{\alpha}.

Composing with Φα,β\Phi_{\alpha,\beta}, we obtain holomorphic maps

G:=Φα,β∘G~:D′×rB‾N⟶Y.G:=\Phi_{\alpha,\beta}\circ\widetilde{G}:D'\times r\overline{B}^{N}\longrightarrow Y.

Since Φα,β\Phi_{\alpha,\beta} is holomorphic, these maps approximate HH uniformly on D×rB‾ND\times r\overline{B}^{N}. Therefore WW is an LSAP neighbourhood of y0y_{0}, so YY has the LSAP property at y0y_{0}. Since y0∈Yy_{0}\in Y was arbitrary, YY has the LSAP property. □\square

Oka-1 theorem

Theorem 17. Let GG be a connected simply connected semisimple algebraic group. Then RregR_{\mathrm{reg}} has the LSAP property.

Proof. Write G=G1⋯GmG=G_{1}\cdots G_{m} as the product of its almost simple factors. Since GG is simply connected, the multiplication map

G1×⋯×Gm⟶GG_{1}\times\cdots\times G_{m}\longrightarrow G

is an isomorphism [39 §24a]. Under this isomorphism,

Greg≅∏j=1m(Gj)regandRreg(G)≅∏j=1mRreg(Gj).G_{\mathrm{reg}}\cong\prod_{j=1}^{m}(G_{j})_{\mathrm{reg}}\qquad\text{and}\qquad R_{\mathrm{reg}}(G)\cong\prod_{j=1}^{m}R_{\mathrm{reg}}(G_{j}).

Applying Lemmas 14 and 15 to each almost simple factor and taking products, we obtain a family of local biholomorphisms from Greg×CrG_{\mathrm{reg}}\times\mathbb{C}^{r} whose images cover Rreg(G)R_{\mathrm{reg}}(G).

By Lemma 16, it remains to show that Greg×CrG_{\mathrm{reg}}\times\mathbb{C}^{r} has LSAP. Since finite products of LSAP manifolds have LSAP and Cr\mathbb{C}^{r} has LSAP, it is enough to show that GregG_{\mathrm{reg}} does. Indeed, in [4 Proposition 1.1] it is shown that semisimple algebraic groups are flexible and in [14 Theorem 1.1] that complements of flexible affine varieties of a closed subvariety of codimension ≥2\geq2 remain flexible. Finally, Steinberg [52 Theorem 1.3 (a) and (b)] showed that GregG_{\mathrm{reg}} is the complement of a closed subvariety of GG of codimension ≥3\geq3. Thus, GregG_{\mathrm{reg}} is flexible, hence has the LSAP property. □\square

:::

Remark 18. Recall that we defined RregR_{\mathrm{reg}} as the preimage of GregG_{\mathrm{reg}} by the projection map π:G×G⟶G\pi:G\times G\longrightarrow G on the first component. However, the choice of projection map to the first component was arbitrary. If we denote the two projection maps πi\pi_i for i=1,2i=1,2, then Rreg=π1−1∣R(G)(Greg)R_{\mathrm{reg}}=\pi_1^{-1}|_{R(G)}(G_{\mathrm{reg}}). Define

Rreg(i):=πi−1∣R(G)(Greg).R_{\mathrm{reg}}^{(i)}:=\pi_i^{-1}|_{R(G)}(G_{\mathrm{reg}}).

By symmetry the conclusion of Theorem 17 also holds for

Rreg(2)={(x,y)∈R∣y∈Greg}.R_{\mathrm{reg}}^{(2)}=\{(x,y)\in R\mid y\in G_{\mathrm{reg}}\}.

Since the LSAP property is a Hausdorff local property [FL25, Proposition 5.5], we have that the union Rreg(1)∪Rreg(2)R_{\mathrm{reg}}^{(1)} \cup R_{\mathrm{reg}}^{(2)} has LSAP.

Corollary 19. Let GG be a simply connected semisimple algebraic group. The smooth locus R(G)smR(G)_{\mathrm{sm}} of the commuting variety of GG is an Oka-1 manifold.

Proof. Set n=dim⁡C(Rsm)n = \dim_{\mathbb{C}}(R_{\mathrm{sm}}). The complement E=Rsm∖RregE = R_{\mathrm{sm}} \setminus R_{\mathrm{reg}} is a proper analytic subset of RsmR_{\mathrm{sm}} and thus has complex dimension dim⁡C(E)≤n−1\dim_{\mathbb{C}}(E) \le n - 1. Thus, the Hausdorff measure HRsm,g2n−1(E)\mathcal{H}^{2n-1}_{R_{\mathrm{sm}},g}(E) of EE is zero. By Theorem 17, RregR_{\mathrm{reg}} has LSAP and hence RsmR_{\mathrm{sm}} is Oka-1 by [1 Proposition 7.13 (b)]. □

Remark 20. (1) There are many instances where the assumption of simple connectedness of GG was used. One of the main reasons we require simple connectedness is Steinberg’s result that, for GG simply connected, Fx,1,…,Fx,r∈z(x)F_{x,1}, \ldots, F_{x,r} \in\mathfrak{z}(x) are linearly independent if and only if x∈Gregx \in G_{\mathrm{reg}}. Other applications of simple connectedness are that the centralisers of semisimple elements are connected and that the centralisers of regular elements are abelian.

(2) Within the classification of reductive algebraic groups, the simply connected almost simple algebraic groups are SLn+1(C)\mathrm{SL}_{n+1}(\mathbb{C}), Spin⁡2n+1(C)\operatorname{Spin}_{2n+1}(\mathbb{C}), Sp⁡2n(C)\operatorname{Sp}_{2n}(\mathbb{C}), Spin⁡2n(C)\operatorname{Spin}_{2n}(\mathbb{C}) together with the simply connected forms of the exceptional types G2G_2, F4F_4, E6E_6, E7E_7, E8E_8 [37 Table 9.2, p. 72]. Every simply connected semisimple algebraic group is a finite direct product of groups from this list [39 §24a].

Strongly commuting varieties

The motivation for defining strongly commuting varieties is the following. Suppose that we are interested in commuting elements in PSLn(C)\mathrm{PSL}_n(\mathbb{C}). Two elements [x],[y]∈PSLn(C)[x], [y] \in\mathrm{PSL}_n(\mathbb{C}) commute if and only if their representatives x,y∈SLn(C)x, y \in\mathrm{SL}_n(\mathbb{C}) commute up to an element of the centre ω∈Z(SLn(C))\omega\in Z(\mathrm{SL}_n(\mathbb{C})), i.e., xy=ωyxxy = \omega yx. Hence, we may impose a stronger condition on the elements [x],[y]∈PSLn(C)[x], [y] \in\mathrm{PSL}_n(\mathbb{C}) to commute by requiring that the representatives x,y∈SLn(C)x, y \in\mathrm{SL}_n(\mathbb{C}) commute in SLn(C)\mathrm{SL}_n(\mathbb{C}). In this case, we will say that [x][x] and [y][y] strongly commute. We give four equivalent descriptions of which two are topological and two are algebraic in nature. The algebraic formulations are inspired by [46 Section 3].

We begin with some notation. Let GG be a connected semisimple algebraic group and let q:G~→Gq : \widetilde{G} \to G be its universal covering, and let Γ:=ker⁡q\Gamma:= \ker q. Since G~\widetilde{G} is connected and Γ\Gamma is a finite normal subgroup, it follows that Γ⊂Z(G~)\Gamma\subset Z(\widetilde{G}). Let (x,y)∈R(G)(x, y) \in R(G) and (X,Y)∈G~2(X, Y) \in\widetilde{G}^{2} be any lift. Define the set-theoretic map

κ:R(G)⟶Γ,(x,y)⟼[X,Y].\kappa: R(G) \longrightarrow\Gamma,\qquad(x,y) \longmapsto[X,Y].

Proposition 21. The map κ\kappa is well-defined and is a morphism of varieties.

Before proving this proposition, for each γ∈Γ\gamma\in\Gamma, we introduce

Rγ(G~)={(X,Y)∈G~2∣[X,Y]=γ}.R_{\gamma}(\widetilde{G}) = \{(X,Y) \in\widetilde{G}^{2} \mid[X,Y] = \gamma\}.

Then, we have

(q×q)−1(R(G))=⨆γ∈ΓRγ(G~).(q \times q)^{-1}(R(G)) = \bigsqcup_{\gamma\in\Gamma} R_{\gamma}(\widetilde{G}).

Indeed, for any (X,Y)∈Rγ(G~)(X,Y) \in R_{\gamma}(\widetilde{G}) the images q(X)q(X) and q(Y)q(Y) commute. Conversely, for any (X,Y)∈(q×q)−1(x,y)(X,Y) \in(q \times q)^{-1}(x,y) with (x,y)∈R(G)(x,y) \in R(G), we have that [X,Y]=γ[X,Y] = \gamma. Applying qq to both sides shows that γ∈Γ\gamma\in\Gamma. Now define the following subset of R(G)R(G)

Rγ(G):=κ−1(γ).R_{\gamma}(G) := \kappa^{-1}(\gamma).

We think of Rγ(G)R_{\gamma}(G) as the subset of R(G)R(G) whose lifts to G~\widetilde{G} commute up to the element γ∈Γ\gamma\in\Gamma. It follows immediately that Rγ(G)=(q×q)(Rγ(G~))R_{\gamma}(G) = (q \times q)(R_{\gamma}(\widetilde{G})) and Rγ(G)∩Rδ(G)=∅R_{\gamma}(G) \cap R_{\delta}(G) = \varnothing if γ≠δ\gamma\ne\delta.

Proof of Proposition 21. Since q×q:G~×G~→G×Gq \times q : \widetilde{G} \times\widetilde{G} \to G \times G is finite and hence closed, the Rγ(G)=(q×q)(Rγ(G~))R_{\gamma}(G) = (q \times q)(R_{\gamma}(\widetilde{G})) are closed in R(G)R(G), as the Rγ(G~)R_{\gamma}(\widetilde{G}) are closed. Since there are finitely many γ∈Γ\gamma\in\Gamma, the Rγ(G)R_{\gamma}(G) are pairwise disjoint and their union covers R(G)R(G). It follows that the Rγ(G)R_{\gamma}(G) are also open. The result of the proposition follows from the following claim: A set-theoretic map f:X→Γf : X \to\Gamma between an affine variety X⊂AnX \subset A^n and a finite variety Γ⊂Am\Gamma\subset A^m is induced by a morphism of varieties if and only if the fibres are open and closed. The forward direction is obvious. The converse is an application of [24 II, Ex 2.19]. □

Definition 22. Two commuting elements a,b∈Ga,b \in G strongly commute if

(a,b)∈(q×q)(R(G~)).(a,b) \in(q \times q)(R(\widetilde{G})).

We define the strongly commuting variety of GG as

Rstr(G):=(q×q)(R(G~))=R1(G).R_{\mathrm{str}}(G) := (q \times q)(R(\widetilde{G})) = R_1(G).

By Richardson’s theorem, R(G~)R(\widetilde{G}) is irreducible because G~\widetilde{G} is semisimple and simply connected. Therefore, the strongly commuting variety is the image of the irreducible variety, hence is irreducible and connected and contains (1,1)(1,1). Thus, it is the identity component of R(G)R(G). The topology of the identity component of R(G)R(G) has been studied in [Bai07,GPS12,PS13].

We now state another topological description.

Definition 23. Two commuting elements a,b∈Ga,b \in G topologically strongly commute if for some paths α,β:[0,1]→G\alpha,\beta: [0,1] \to G, with

α(0)=β(0)=1andα(1)=aandβ(1)=b(1)\alpha(0) = \beta(0) = 1 \quad\text{and} \quad\alpha(1) = a \quad\text{and} \quad\beta(1) = b \tag*{(1)}

the product paths α(t)β(t)\alpha(t)\beta(t) and β(t)α(t)\beta(t)\alpha(t) in GG are homotopic relative to the endpoints. Denote by Rtop(G)R_{\mathrm{top}}(G) the collection of pairs of elements which topologically strongly commute.

The next definition is algebraic and comes from [46 §3].

Definition 24. Let E′(G)\mathcal{E}'(G) denote the pairs of elements (x,y)∈G2(x,y) \in G^2 such that there exists a maximal torus TT containing both xx and yy. Let E(G)\mathcal{E}(G) denote the Zariski closure of E′(G)\mathcal{E}'(G).

We now introduce a fourth definition which is also inspired by [46 §3]:

Definition 25. Consider the subset

S′(G)={(x,y)∈R(G)∣∃z∈Z(G) such that zy∈ZG(x)}\mathcal{S}'(G)=\{(x,y)\in R(G)\mid\exists z\in Z(G)\text{ such that }zy\in Z_G(x)\}

of R(G)R(G). Denote the Zariski closure of S′(G)\mathcal{S}'(G) by S(G)\mathcal{S}(G).

Theorem 26. Let GG be a semisimple algebraic group. We have

Rstr(G)=Rtop(G)=E(G)=S(G).R_{\mathrm{str}}(G)=R_{\mathrm{top}}(G)=\mathcal{E}(G)=\mathcal{S}(G).

We prove this in the following order: Rstr(G)=Rtop(G)R_{\mathrm{str}}(G)=R_{\mathrm{top}}(G) then Rstr(G)=E(G)R_{\mathrm{str}}(G)=\mathcal{E}(G), and lastly E(G)=S(G)\mathcal{E}(G)=\mathcal{S}(G).

Proposition 27. The equality Rstr(G)=Rtop(G)R_{\mathrm{str}}(G)=R_{\mathrm{top}}(G) holds.

Proof. Let (a,b)∈Rstr(G)(a,b)\in R_{\mathrm{str}}(G), i.e., a,b∈Ga,b\in G strongly commute. We will show that (a,b)∈Rtop(G)(a,b)\in R_{\mathrm{top}}(G), i.e., aa and bb topologically strongly commute. Let α,β:[0,1]→G\alpha,\beta:[0,1]\to G be paths with

α(0)=β(0)=1andα(1)=aandβ(1)=b.\alpha(0)=\beta(0)=1\quad\text{and}\quad\alpha(1)=a\quad\text{and}\quad\beta(1)=b.

We must show that the product paths αβ\alpha\beta and βα\beta\alpha are homotopic. By assumption, there are lifts a~,b~∈G~\widetilde{a},\widetilde{b}\in\widetilde{G} such that

a~b~=b~a~.\widetilde{a}\widetilde{b}=\widetilde{b}\widetilde{a}.

Let α~,β~:[0,1]→G~\widetilde{\alpha},\widetilde{\beta}:[0,1]\to\widetilde{G} be the unique lifts of α\alpha and β\beta such that

α~(0)=1G~=β~(0).\widetilde{\alpha}(0)=1_{\widetilde{G}}=\widetilde{\beta}(0).

Then α~(1)\widetilde{\alpha}(1) and a~\widetilde{a} differ by an element u^\widehat{u} in the kernel ker⁡(q)\ker(q). Specifically, if α~(1)=u~\widetilde{\alpha}(1)=\widetilde{u} then q(a~u~−1)=1Gq(\widetilde{a}\widetilde{u}^{-1})=1_G. Similarly, for β~(1)=v~\widetilde{\beta}(1)=\widetilde{v}, we have b~v~−1∈ker⁡q\widetilde{b}\widetilde{v}^{-1}\in\ker q. Define u^:=a~u~−1\widehat{u}:=\widetilde{a}\widetilde{u}^{-1} and v^:=b~v~−1\widehat{v}:=\widetilde{b}\widetilde{v}^{-1}. Since finite normal subgroups of connected algebraic groups are central, it follows that u^,v^∈ker⁡(q)\widehat{u},\widehat{v}\in\ker(q) are central elements in G~\widetilde{G}. Moreover, from the assumption that a~b~=b~a~\widetilde{a}\widetilde{b}=\widetilde{b}\widetilde{a} it follows that

α~(1)β~(1)=β~(1)α~(1).\widetilde{\alpha}(1)\widetilde{\beta}(1)=\widetilde{\beta}(1)\widetilde{\alpha}(1).

Thus, the endpoints of the lifts α~\widetilde{\alpha} and β~\widetilde{\beta} commute so the products of the lifts α~β~\widetilde{\alpha}\widetilde{\beta} and β~α~\widetilde{\beta}\widetilde{\alpha} have the same endpoints and therefore are homotopic by some homotopy H~\widetilde{H} as G~\widetilde{G} is simply connected. Then q∘H~q\circ\widetilde{H} gives a homotopy between αβ\alpha\beta and βα\beta\alpha. Thus, aa and bb topologically strongly commute.

Conversely, suppose aa and bb topologically strongly commute. We are required to show that there exist a~,b~∈G~\widetilde{a},\widetilde{b}\in\widetilde{G} such that a~b~=b~a~\widetilde{a}\widetilde{b}=\widetilde{b}\widetilde{a} and q(a~)=aq(\widetilde{a})=a and q(b~)=bq(\widetilde{b})=b. The paths α,β:[0,1]→G\alpha,\beta:[0,1]\to G have unique lifts α~,β~:[0,1]→G~\widetilde{\alpha},\widetilde{\beta}:[0,1]\to\widetilde{G} with start point 1G~1_{\widetilde{G}}. The product paths α~β~\widetilde{\alpha}\widetilde{\beta} and β~α~\widetilde{\beta}\widetilde{\alpha} satisfy

q(α~β~)=αβandq(β~α~)=βαq(\widetilde{\alpha}\widetilde{\beta})=\alpha\beta\quad\text{and}\quad q(\widetilde{\beta}\widetilde{\alpha})=\beta\alpha

because qq is a homomorphism of groups so α~β~\widetilde{\alpha}\widetilde{\beta} and β~α~\widetilde{\beta}\widetilde{\alpha} are the unique lifts for the paths αβ\alpha\beta and βα\beta\alpha. Since α~(0)β~(0)=1G~=β~(0)α~(0)\widetilde{\alpha}(0)\widetilde{\beta}(0)=1_{\widetilde{G}}=\widetilde{\beta}(0)\widetilde{\alpha}(0) by the covering homotopy lemma,

α~(t)β~(t)andβ~(t)α~(t)\widetilde{\alpha}(t)\widetilde{\beta}(t)\quad\text{and}\quad\widetilde{\beta}(t)\widetilde{\alpha}(t)

have the same endpoints and are homotopic. In particular,

α~(1)β~(1)=β~(1)α~(1).\widetilde{\alpha}(1)\widetilde{\beta}(1)=\widetilde{\beta}(1)\widetilde{\alpha}(1).

Thus, (α~(1),β~(1))∈R(G~)(\widetilde{\alpha}(1),\widetilde{\beta}(1))\in R(\widetilde{G}) and

(q×q)(α~(1),β~(1))=(a,b)∈R(G).(q\times q)(\widetilde{\alpha}(1),\widetilde{\beta}(1))=(a,b)\in R(G).

□

Proposition 28. The equality Rstr(G)=E(G)R_{\mathrm{str}}(G)=E(G) holds.

Proof. If two elements a~,b~\widetilde{a},\widetilde{b} are contained in a maximal torus T~⊂G~\widetilde{T}\subset\widetilde{G} then their images q(a~)q(\widetilde{a}) and q(b~)q(\widetilde{b}) are contained in the maximal torus T:=q(T~)T:=q(\widetilde{T}) of GG [10 Proposition 11.4]. Conversely, if T⊂GT\subset G is a maximal torus, then T~=q−1(T)∘\widetilde{T}=q^{-1}(T)^\circ is a maximal torus of G~\widetilde{G} with q(T~)=Tq(\widetilde{T})=T. Hence any pair in T2T^2 has lifts in T~2\widetilde{T}^2. Since the map q×qq\times q is a finite map, it is closed and, in particular, the map commutes with taking Zariski closures. With these facts at hand we have

E(G)=E′(G)‾=(q×q)(E′(G~)‾)=(q×q)(E(G~))=(q×q)(R(G~))=Rstr(G).E(G)=\overline{E'(G)}=(q\times q)(\overline{E'(\widetilde{G})})=(q\times q)(E(\widetilde{G}))=(q\times q)(R(\widetilde{G}))=R_{\mathrm{str}}(G).

Note that the fourth equality is [46 Theorem C]. □

Proposition 29. The equality E(G)=S(G)E(G)=S(G) holds.

Proof. The inclusion S(G)⊂E(G)S(G)\subset E(G) follows from [46 §3, Theorem B]. For the other inclusion, suppose xx and yy are contained in the maximal torus TT so that (x,y)∈E′(G)(x,y)\in E'(G). Then y∈ZG(x)y\in Z_G(x) and T⊂ZG(x)∘T\subset Z_G(x)^\circ, thus, y∈ZG(x)∘y\in Z_G(x)^\circ. In particular, (x,y)∈S′(G)(x,y)\in S'(G). Thus, E(G)⊂S(G)E(G)\subset S(G). □

Oka-1 for strongly commuting varieties

In this section we prove that the smooth locus of the strongly commuting variety of a semisimple algebraic group is Oka-1. Let GG be a semisimple algebraic group and Rstr(G)R_{\mathrm{str}}(G) be its strongly commuting variety. The open subset of Rstr(G)R_{\mathrm{str}}(G) defined by

Rstr,reg(G):={(x,y)∈Rstr(G)∣x∈Greg}R_{\mathrm{str,reg}}(G):=\{(x,y)\in R_{\mathrm{str}}(G)\mid x\in G_{\mathrm{reg}}\}

is called the regular strongly commuting variety.

Theorem 30. Let GG be a connected semisimple algebraic group. Then the regular strongly commuting variety has LSAP. Consequently, the smooth locus Rstr(G)smR_{\mathrm{str}}(G)_{\mathrm{sm}} is an Oka-1 manifold.

Proof. Since q:G~→Gq : \widetilde{G} \to G is a central isogeny, it is a finite holomorphic covering map hence a local biholomorphism. Moreover, regular elements in G~\widetilde{G} descend to regular elements in GG via qq [34 Proposition 2.3 (2)]. It follows that the map

q×q:Rreg(G~)⟶Rstr,reg(G)q \times q : R_{\mathrm{reg}}(\widetilde{G}) \longrightarrow R_{\mathrm{str,reg}}(G)

is a surjective local biholomorphism. Since Rreg(G~)R_{\mathrm{reg}}(\widetilde{G}) is smooth and has LSAP by Theorem 17, it follows that Rstr,reg(G)R_{\mathrm{str,reg}}(G) is smooth and also has LSAP. Set n=dim⁡C(Rstr(G)sm)n = \dim_{\mathbb{C}}(R_{\mathrm{str}}(G)_{\mathrm{sm}}). The complement E=Rstr(G)sm∖Rstr,reg(G)E = R_{\mathrm{str}}(G)_{\mathrm{sm}} \setminus R_{\mathrm{str,reg}}(G) is a proper analytic subset of complex dimension dim⁡C(E)≤n−1\dim_{\mathbb{C}}(E) \le n - 1. Thus the Hausdorff measure HRstr(G)sm,g2n−1(E)\mathcal{H}^{2n-1}_{R_{\mathrm{str}}(G)_{\mathrm{sm}},g}(E) of EE is zero. Thus, Rstr(G)smR_{\mathrm{str}}(G)_{\mathrm{sm}} is Oka-1 by [1 Proposition 7.13 (b)]. □

Oka-1 for PSLp_p(C)

In Theorem 30 we showed that the smooth locus of the strongly commuting variety of a semisimple algebraic group is Oka-1. In this section, we prove an Oka-1 result for the full commuting variety of the semisimple (but not simply connected) group PSL⁡p(C)\operatorname{PSL}_p(\mathbb{C}) when pp is prime. This will involve the study of almost commuting elements which have previously been studied in [11] and [31] in the case of compact Lie groups. Here, we adapt some of their results to the setting of semisimple algebraic groups.

We return to the case where GG is a semisimple algebraic group and let q:G~→Gq : \widetilde{G} \to G be its universal covering and let Γ=ker⁡(q)\Gamma= \ker(q). Then

(q×q)−1(R(G))=∐γ∈ΓRγ(G~)(q \times q)^{-1}(R(G)) = \coprod_{\gamma\in\Gamma} R_{\gamma}(\widetilde{G})

where

Rγ(G~)={(X,Y)∈G~2∣[X,Y]=γ}R_{\gamma}(\widetilde{G}) = \{(X,Y) \in\widetilde{G}^{2} \mid[X,Y] = \gamma\}

and we have identified the strongly commuting variety with the image (q×q)(R1(G~))(q \times q)(R_{1}(\widetilde{G})). It is an interesting problem to describe the components Rγ(G~)R_{\gamma}(\widetilde{G}), i.e., collections of elements commuting up to a central element γ∈Γ\gamma\in\Gamma. Following Serre [47], we define the notion of irreducible pairs:

Definition 31. A subgroup HH of GG is said to be irreducible if it is not contained in any proper parabolic subgroup of GG. We say that a pair (x,y)∈G×G(x,y) \in G \times G is irreducible if ⟨x,y⟩\langle x,y \rangle is an irreducible subgroup of GG. Here ⟨x,y⟩⊂G\langle x,y \rangle\subset G denotes the Zariski closure of the group generated by elements x,y∈Gx,y \in G.

Note that irreducibility of (x,y)(x,y) implies that H:=⟨x,y⟩H := \langle x,y \rangle is reductive. Indeed, the unipotent radical Ru(H)R_{\mathrm{u}}(H) of HH is normalised by HH, and hence H⊂NG(Ru(H))H \subset N_G(R_{\mathrm{u}}(H)). By [27 Cor 30.3], NG(Ru(H))N_G(R_{\mathrm{u}}(H)) is contained in a parabolic subgroup of GG, so, by irreducibility of HH, it follows that Ru(H)={1}R_{\mathrm{u}}(H) = \{1\}. Consequently, HH is reductive and irreducibility of a pair (x,y)∈G2(x,y) \in G^2 implies that ZG(H)Z_G(H) is finite (cf. [45 Lemma 7.6]). Note that, in [11], irreducible pairs are called rank zero pairs.

Lemma 32. Let GG be a semisimple algebraic group and let x,y∈Gx,y \in G be an irreducible pair satisfying

xy=ωyx,ω∈Z(G).xy=\omega yx,\qquad\omega\in Z(G).

Then xx and yy are semisimple of finite order. Moreover, if GG is simply connected then xx and yy are regular elements.

Proof. Since Z(G)Z(G) is finite, we may assume that ω\omega has finite order mm. Using the relation in the statement, we have

xm=yωmxmy−1=yxmy−1.x^{m}=y\omega^{m}x^{m}y^{-1}=yx^{m}y^{-1}.

Thus, xm∈ZG(y)x^{m} \in Z_G(y) and similarly, ym∈ZG(x)y^{m} \in Z_G(x). Thus, xmx^{m} and ymy^{m} are contained in the finite group ZG(H)Z_G(H), where H=⟨x,y⟩H=\langle x,y\rangle, and thus have finite order by [45 Lemma 7.6]. Furthermore, since xx and yy have finite order they must be semisimple. Moreover, if GG is simply connected then ZG(x)Z_G(x) and ZG(y)Z_G(y) are connected [28 Theorem 2.11] and reductive. Furthermore, yy normalises ZG(x)Z_G(x) and thus conjugation by yy restricted to ZG(x)Z_G(x) defines an automorphism of ZG(x)Z_G(x) whose fixed-point set is the finite group ZG(x)∩ZG(y)=ZG(x,y)Z_G(x) \cap Z_G(y)=Z_G(x,y). By [53 Corollary 10.10], ZG(x)Z_G(x) is necessarily solvable. Thus, ZG(x)Z_G(x) is a connected reductive solvable group and therefore a torus [27 Theorem 19.3 (b)]. By symmetry the same argument holds for ZG(y)Z_G(y). Thus, xx and yy are regular elements. □

The following lemma is the algebraic group version of [11 Proposition 4.1.1]:

Lemma 33. Let GG be a semisimple simply connected algebraic group and let ω∈Z(G)\omega\in Z(G). Suppose that

(x1,y1),(x2,y2)∈G×G(x_1,y_1),(x_2,y_2) \in G \times G

are irreducible pairs satisfying

xiyi=ωyixi,i=1,2.x_i y_i=\omega y_i x_i,\qquad i=1,2.

Assume that x1x_1 and x2x_2 are regular semisimple. Then the two pairs are simultaneously in GG if and only if x1x_1 and x2x_2 are conjugate in GG.

Proof. The forward direction is trivial. For the converse, we may assume that x1=x2=x∈Tx_1=x_2=x \in T. The goal is to find a t∈ZG(x)t \in Z_G(x) so that ty1t−1=y2ty_1t^{-1}=y_2. Since yiy_i normalises TT, the commutator map

φyi:T⟶T,t↦[t,yi]\varphi_{y_i}:T \longrightarrow T,\qquad t \mapsto[t,y_i]

is well-defined. Its kernel is

ker⁡φyi=ZT(yi)=ZG(yi)∩T=ZG(yi)∩ZG(x)=ZG(yi,x).\ker\varphi_{y_i}=Z_T(y_i)=Z_G(y_i) \cap T=Z_G(y_i) \cap Z_G(x)=Z_G(y_i,x).

Since the pair (x,yi)(x,y_i) is irreducible, ker⁡φyi\ker\varphi_{y_i} is finite, hence dim⁡im⁡φyi=dim⁡T\dim\operatorname{im}\varphi_{y_i}=\dim T and, since the image is closed, we may conclude that φyi\varphi_{y_i} is surjective. Furthermore, one

can check that y2y1−1∈ZG(x)=Ty_{2}y_{1}^{-1} \in Z_{G}(x)=T, hence there is a t∈Tt \in T such that φy1(t)=y2y1−1\varphi_{y_{1}}(t)=y_{2}y_{1}^{-1} which is equivalent to ty1t−1=y2ty_{1}t^{-1}=y_{2}. Since t∈ZG(x)t \in Z_{G}(x), we have

t(x,y1)t−1=(txt−1,ty1t−1)=(x,y2).t(x,y_{1})t^{-1}=(txt^{-1},ty_{1}t^{-1})=(x,y_{2}).

Lemma 34. Let G=∏i=1rSLni(C)G=\prod_{i=1}^{r}\mathrm{SL}_{n_i}(\mathbb{C}) and let ω=ω1⋯ωr∈Z(G)\omega=\omega_{1}\cdots\omega_{r}\in Z(G) be a central element where each ωi\omega_i generates Z(SLni(C))Z(\mathrm{SL}_{n_i}(\mathbb{C})). Then all pairs (x,y)∈G×G(x,y)\in G\times G satisfying

xy=ωyxxy=\omega yx

are irreducible.

Proof. Let U⊂CniU\subset\mathbb{C}^{n_i} be a nonzero invariant subspace for the action of Hi=⟨xi,yi⟩H_i=\langle x_i,y_i\rangle and v∈Uv\in U an eigenvector. Since ωi\omega_i has order nin_i, using the relation xiyi=ωiyixix_iy_i=\omega_i y_ix_i, it is easy to see that xix_i has eigenvectors {v,yiv,…,yini−1v}\{v,y_iv,\ldots,y_i^{n_i-1}v\} which are all linearly independent, hence form a basis for Cni\mathbb{C}^{n_i}. Thus, U=CniU=\mathbb{C}^{n_i} and it follows that each HiH_i is irreducible and hence so is the pair (x,y)∈G×G(x,y)\in G\times G.

Corollary 35. Let pp be a prime number. Every pair (x,y)∈SLp(C)×SLp(C)(x,y)\in\mathrm{SL}_{p}(\mathbb{C})\times\mathrm{SL}_{p}(\mathbb{C}) satisfying

xy=ωyx,ω≠1∈Z(SLp(C))xy=\omega yx,\qquad\omega\ne1\in Z(\mathrm{SL}_{p}(\mathbb{C}))

is irreducible and all such pairs are conjugate. In particular, Rω(SLp(C))R_{\omega}(\mathrm{SL}_{p}(\mathbb{C})) is a homogeneous space isomorphic to PSLp(C)\mathrm{PSL}_{p}(\mathbb{C}).

Proof. Let pp be a prime number and fix a nontrivial central element ω∈Z(SLp(C))\omega\in Z(\mathrm{SL}_{p}(\mathbb{C})). Let (xi,yi)(x_i,y_i) for i=1,2i=1,2 be to pairs in SLp(C)×SLp(C)\mathrm{SL}_{p}(\mathbb{C})\times\mathrm{SL}_{p}(\mathbb{C}). Using the defining relation, one sees that the eigenvalues of x1x_1 and x2x_2 are equal, thus x1x_1 and x2x_2 are conjugate and by Lemma 33 so are the pairs. If g∈Stab⁡SLp(C)((x,y))g\in\operatorname{Stab}_{\mathrm{SL}_{p}(\mathbb{C})}((x,y)), then gg is an intertwiner for the restriction to ⟨x,y⟩\langle x,y\rangle of the standard representation of SLp(C)\mathrm{SL}_{p}(\mathbb{C}) on Cp\mathbb{C}^{p}. Since HH is irreducible, Schur’s lemma implies that gg is a scalar multiple of the identity, hence g∈Z(SLp(C))g\in Z(\mathrm{SL}_{p}(\mathbb{C})). Clearly, Z(SLp(C))⊂Stab⁡SLp(C)((x,y))Z(\mathrm{SL}_{p}(\mathbb{C}))\subset\operatorname{Stab}_{\mathrm{SL}_{p}(\mathbb{C})}((x,y)), hence Rω(SLp(C))≃SLp(C)/Z(SLp(C))R_{\omega}(\mathrm{SL}_{p}(\mathbb{C}))\simeq\mathrm{SL}_{p}(\mathbb{C})/Z(\mathrm{SL}_{p}(\mathbb{C})).

Theorem 36. Let G=PSLp(C)G=\mathrm{PSL}_{p}(\mathbb{C}), where pp is a prime. Then the smooth locus R(PSLp(C))smR(\mathrm{PSL}_{p}(\mathbb{C}))_{\mathrm{sm}} of the commuting variety of PSLp(C)\mathrm{PSL}_{p}(\mathbb{C}) is Oka-1.

Proof. Let q:SLp(C)→PSLp(C)q:\mathrm{SL}_{p}(\mathbb{C})\to\mathrm{PSL}_{p}(\mathbb{C}) be the quotient map. Then

(q×q)−1(R(PSLp(C)))=⋃ω∈Z(SLp(C))Rω(SLp(C)).(q\times q)^{-1}(R(\mathrm{PSL}_{p}(\mathbb{C})))=\bigcup_{\omega\in Z(\mathrm{SL}_{p}(\mathbb{C}))}R_{\omega}(\mathrm{SL}_{p}(\mathbb{C})).

Let G~=SLp(C)\widetilde{G}=\mathrm{SL}_{p}(\mathbb{C}) and G=PSLp(C)G=\mathrm{PSL}_{p}(\mathbb{C}) and let Γ=ker⁡(q:G~→G)\Gamma=\ker(q:\widetilde{G}\to G). We have already seen that

R(G)=⋃ω∈ΓRω(G).R(G)=\bigcup_{\omega\in\Gamma}R_{\omega}(G).

By Corollary 35, the Rω(G~)R_{\omega}(\widetilde{G}), and hence also Rω(G)R_{\omega}(G), are homogeneous for ω≠1\omega\ne1. Consequently, the Rω(G)R_{\omega}(G) with ω≠1\omega\ne1 are Oka-1. The smooth locus Rstr(PSLp(C))smR_{\mathrm{str}}(\mathrm{PSL}_{p}(\mathbb{C}))_{\mathrm{sm}} is Oka-1 by Theorem 30. Therefore all connected components of the full smooth locus are Oka-1. The statement of the theorem follows since the Riemann surface in the definition of Oka-1 is connected. □

Remark 37. The assumption that pp is prime in Theorem 36 is used to ensure that every non-identity component Rω(PSLp(C))R_{\omega}(\mathrm{PSL}_{p}(\mathbb{C})) is a homogeneous space. If pp were not prime, a non-identity element ω∈Z(SLp(C))\omega\in Z(\mathrm{SL}_{p}(\mathbb{C})) need not generate the whole centre. The varieties Rω(SLp(C))R_{\omega}(\mathrm{SL}_{p}(\mathbb{C})) corresponding to such elements need not be homogeneous spaces. We refer the reader to [11 §4] for the classification of general ω\omega-pairs in the case of compact Lie groups.

The Lie algebra case

In this section we study the holomorphic flexibility properties of the smooth locus of the commuting variety of a semisimple Lie algebra.

Let g\mathfrak{g} be a rank rr complex semisimple Lie algebra with adjoint group GG. Recall that an element x∈gx \in\mathfrak{g} is regular if the dimension of its centraliser z(x)\mathfrak{z}(x) in g\mathfrak{g} is minimal equal to the rank of g\mathfrak{g}. The commuting variety of the Lie algebra g\mathfrak{g} is defined as

C=C(g):={(x,y)∈g×g∣[x,y]=0}.\mathcal{C}=\mathcal{C}(\mathfrak{g}):=\{(x,y)\in\mathfrak{g}\times\mathfrak{g}\mid[x,y]=0\}.

We define two dense open subsets

Ci:={(x1,x2)∈C∣xi∈greg},i=1,2.\mathcal{C}_{i}:=\{(x_{1},x_{2})\in\mathcal{C}\mid x_{i}\in\mathfrak{g}_{\mathrm{reg}}\},\qquad i=1,2.

Furthermore, we denote by Creg\mathcal{C}_{\mathrm{reg}} their union

Creg:=C1∪C2.\mathcal{C}_{\mathrm{reg}}:=\mathcal{C}_{1}\cup\mathcal{C}_{2}.

Remark 38. In [43 §2], the author defines the subset

C∘={(x,y)∈C∣dim⁡(z(x)∩z(y))=r},\mathcal{C}^{\circ}=\{(x,y)\in\mathcal{C}\mid\dim(\mathfrak{z}(x)\cap\mathfrak{z}(y))=r\},

also called the regular subset of C\mathcal{C}, which is an open subset of the smooth locus Csm\mathcal{C}_{\mathrm{sm}} [[43 Theorem 1.3 (ii)]]. We remark that the open subset Creg=C1∪C2\mathcal{C}_{\mathrm{reg}}=\mathcal{C}_{1}\cup\mathcal{C}_{2} we defined above is contained in C∘\mathcal{C}^{\circ}. The distinction already appears in sl4(C)\mathfrak{sl}_{4}(\mathbb{C}) for x=e1,2+e3,4x=e_{1,2}+e_{3,4} and y=e1,3+e2,4y=e_{1,3}+e_{2,4} where ei,j∈sl4(C)e_{i,j}\in\mathfrak{sl}_{4}(\mathbb{C}) denotes the standard basis vector whose iith row and jjth column is 1 and zero elsewhere. One verifies that (x,y)∈C∘(x,y)\in\mathcal{C}^{\circ} but neither xx nor yy is regular elements. Hence, in general there is a proper containment Creg⊂Csm\mathcal{C}_{\mathrm{reg}}\subset\mathcal{C}_{\mathrm{sm}}.

By Chevalley’s theorem, we may choose homogeneous generators p1,…,pr∈C[g]Gp_{1},\ldots,p_{r}\in\mathbb{C}[\mathfrak{g}]^{G}. For x∈gx\in\mathfrak{g}, the differentials (dpi)x(dp_{i})_{x} may be identified with elements of the dual space g∗\mathfrak{g}^{*}. Using the non-degeneracy of the Killing form κ\kappa, we have an isomorphism B:g→g∗B:\mathfrak{g}\to\mathfrak{g}^{*}. Define Fi(x)∈gF_i(x)\in\mathfrak{g} as the element Fi(x):=B−1((dpi)x)F_i(x):=B^{-1}((dp_i)_x). Equivalently, Fi(x)F_i(x) is defined by the formula

(dpi)x(v)=κ(Fi(x),v).(dp_i)_x(v)=\kappa(F_i(x),v).

Lemma 39. For every x∈gregx\in\mathfrak{g}_{\mathrm{reg}}, the F1(x),…,Fr(x)∈z(x)F_1(x),\ldots,F_r(x)\in\mathfrak{z}(x) form a basis.

Proof. See [33 Proposition 1.3] and [32 Theorem 0.1]. □

Proposition 40. The map

Ψ:greg×Cr⟶C1,(x,λ1,…,λr)⟼(x,∑i=1rλiFi(x))\Psi:\mathfrak{g}_{\mathrm{reg}}\times\mathbb{C}^{r}\longrightarrow\mathbb{C}_1,\qquad(x,\lambda_1,\ldots,\lambda_r)\longmapsto\left(x,\sum_{i=1}^{r}\lambda_iF_i(x)\right)

is an algebraic isomorphism.

Proof. The map Ψ\Psi is algebraic so we need to show that it has an algebraic inverse. To do so, we construct local inverses on a suitable open cover of greg\mathfrak{g}_{\mathrm{reg}} and then show they agree on overlaps.

Fix a basis e1,…,ene_1,\ldots,e_n for g\mathfrak{g} and identify g≃Cn\mathfrak{g}\simeq\mathbb{C}^{n}. For each rr-subset I={i1<⋯<ir}I=\{i_1<\cdots<i_r\} of the nn-set {1,…,n}\{1,\ldots,n\} define the coordinate projection πI:Cn→Cr\pi_I:\mathbb{C}^{n}\to\mathbb{C}^{r}. For each y∈gy\in\mathfrak{g} denote by yIy_I the image of yy under the projection map πI\pi_I under the above identification. Explicitly, if y=∑i=1nλieiy=\sum_{i=1}^{n}\lambda_i e_i then yIy_I corresponds to

πI(∑i=1nλiei)=(λi1,…,λir).\pi_I\left(\sum_{i=1}^{n}\lambda_i e_i\right)=(\lambda_{i_1},\ldots,\lambda_{i_r}).

Now, for each x∈gregx\in\mathfrak{g}_{\mathrm{reg}}, let M(x)M(x) be the n×rn\times r matrix whose columns are the Fi(x)F_i(x) for i=1,…,ri=1,\ldots,r. That is,

M(x)=[F1(x) ⋯ Fr(x)].M(x)=[F_1(x)\ \cdots\ F_r(x)].

Thus, we may write the map Ψ:greg×Cr→C1\Psi:\mathfrak{g}_{\mathrm{reg}}\times\mathbb{C}^{r}\to\mathbb{C}_1 as

(x,λ1,…,λr)⟼(x,M(x)(λ1,…,λr)t).(x,\lambda_1,\ldots,\lambda_r)\longmapsto\left(x,M(x)(\lambda_1,\ldots,\lambda_r)^{t}\right).

By Lemma 39, for x∈gregx\in\mathfrak{g}_{\mathrm{reg}} there exists a subset I={i1<⋯<ir}⊂{1,…,n}I=\{i_1<\cdots<i_r\}\subset\{1,\ldots,n\} of rr rows of M(x)M(x) such that the rr-by-rr sub-matrix

MI(x):=πI∘M(x):Cr⟶CrM_I(x):=\pi_I\circ M(x):\mathbb{C}^{r}\longrightarrow\mathbb{C}^{r}

of M(x)M(x) is invertible. For such a choice of II, the restriction πI∣z(x):z(x)→Cr\left.\pi_I\right|_{\mathfrak{z}(x)}:\mathfrak{z}(x)\to\mathbb{C}^{r} is an isomorphism. For each rr-subset II as above, define the open set of greg\mathfrak{g}_{\mathrm{reg}}

UI={x∈greg∣det⁡(MI(x))≠0}.U_I=\{x\in\mathfrak{g}_{\mathrm{reg}}\mid\det(M_I(x))\ne0\}.

Hence, we have shown that, for each x∈gregx\in\mathfrak{g}_{\mathrm{reg}} there exists a subset II for which MI(x)M_I(x) is invertible, and thus the collection of all {UI}I\{U_I\}_I forms an open cover of greg\mathfrak{g}_{\mathrm{reg}}. Corresponding to the open subsets UI⊂gregU_I\subset\mathfrak{g}_{\mathrm{reg}}, we define

C1UI:=(UI×g)∩C1={(x,y)∈C1∣x∈UI}⊂C1.\mathbb{C}_1^{U_I}:=(U_I\times\mathfrak{g})\cap\mathbb{C}_1=\{(x,y)\in\mathbb{C}_1\mid x\in U_I\}\subset\mathbb{C}_1.

Then, as II varies over all rr-subsets of {1,…,n}\{1,\ldots,n\}, the open sets C1UI\mathcal{C}_1^{U_I} cover C1\mathcal{C}_1. For each C1UI\mathcal{C}_1^{U_I} define

ΨUI−1 ⁣:C1UI→UI×Cr,(x,y)↦(x,(MI(x)−1∘πI)(y)).\Psi_{U_I}^{-1}\colon\mathcal{C}_1^{U_I}\to U_I\times\mathbb{C}^r,\qquad(x,y)\mapsto\left(x,(M_I(x)^{-1}\circ\pi_I)(y)\right).

If (x,y)∈C1UI∩UJ(x,y)\in\mathcal{C}_1^{U_I\cap U_J} for two subsets II and JJ of {1,…,n}\{1,\ldots,n\}, the restrictions of ΨUI−1\Psi_{U_I}^{-1} and ΨUJ−1\Psi_{U_J}^{-1} to the intersection C1UI∩UJ\mathcal{C}_1^{U_I\cap U_J} agree. Indeed, since (x,y)∈C1(x,y)\in\mathcal{C}_1, by Lemma 39 we have y∈z(x)=Span⁡C{F1(x),…,Fr(x)}y\in\mathfrak{z}(x)=\operatorname{Span}_{\mathbb{C}}\{F_1(x),\ldots,F_r(x)\}. Therefore, there are unique constants λ1,…,λr\lambda_1,\ldots,\lambda_r such that y=∑i=1rλiFi(x)y=\sum_{i=1}^r\lambda_iF_i(x) which under the identification g≃Cn\mathfrak{g}\simeq\mathbb{C}^n we write as

y=M(x)⋅(λ1,…,λr)t.y=M(x)\cdot(\lambda_1,\ldots,\lambda_r)^{t}.

Then, in particular,

(MI(x)−1∘πI)(y)=(MI(x)−1∘πI)(M(x)⋅(λ1,…,λr)t)=(λ1,…,λr)t,(M_I(x)^{-1}\circ\pi_I)(y)=(M_I(x)^{-1}\circ\pi_I)\left(M(x)\cdot(\lambda_1,\ldots,\lambda_r)^{t}\right)=(\lambda_1,\ldots,\lambda_r)^{t},

and, similarly, (MJ(x)−1∘πJ)(y)=(λ1,…,λr)t(M_J(x)^{-1}\circ\pi_J)(y)=(\lambda_1,\ldots,\lambda_r)^{t}. Thus, the ΨUI−1\Psi_{U_I}^{-1} agree on the overlaps. The ΨUI−1\Psi_{U_I}^{-1} define local inverses to Ψ\Psi since the restrictions ΨUI:=Ψ∣UI×Cr:UI×Cr→C1UI\Psi_{U_I}:=\left.\Psi\right|_{U_I\times\mathbb{C}^r}:U_I\times\mathbb{C}^r\to\mathcal{C}_1^{U_I} are given by

(x,λ1,…,λr)⟼(x,M(x)(λ1,…,λr)t).(x,\lambda_1,\ldots,\lambda_r)\longmapsto\left(x,M(x)(\lambda_1,\ldots,\lambda_r)^{t}\right).

Furthermore, since matrix inversion is algebraic, the ΨUI−1\Psi_{U_I}^{-1} are algebraic and glue to a global algebraic inverse. □

Remark 41. In [43 Theorem 1.12] the author proves the variety C\mathcal{C} is rational. We give an alternative proof of this fact: The composition C1≃greg×Cr↪g×Cr≃Adim⁡(g)+r\mathcal{C}_1\simeq\mathfrak{g}_{\mathrm{reg}}\times\mathbb{C}^r\hookrightarrow\mathfrak{g}\times\mathbb{C}^r\simeq\mathbb{A}^{\dim(\mathfrak{g})+r} identifies the dense open subset C1⊂C\mathcal{C}_1\subset\mathcal{C} with a dense open subset of Adim⁡(g)+r\mathbb{A}^{\dim(\mathfrak{g})+r}, hence C\mathcal{C} is rational.

Corollary 42. The regular commuting variety Creg\mathcal{C}_{\mathrm{reg}} is algebraically elliptic.

Proof. The Zariski open sets Ci\mathcal{C}_i, i=1,2i=1,2 in Creg\mathcal{C}_{\mathrm{reg}} are algebraically elliptic since they are each isomorphic to the elliptic manifold greg×Cr\mathfrak{g}_{\mathrm{reg}}\times\mathbb{C}^r. Indeed, by [Vel72, Theorem 4.12], g∖greg\mathfrak{g}\setminus\mathfrak{g}_{\mathrm{reg}} is a closed subvariety of codimension 3 in the flexible variety g≃Adim⁡(g)\mathfrak{g}\simeq\mathbb{A}^{\dim(\mathfrak{g})}. By [14 Theorem 1.1], greg\mathfrak{g}_{\mathrm{reg}} is flexible and hence the product greg×Ar\mathfrak{g}_{\mathrm{reg}}\times\mathbb{A}^r is algebraically elliptic. Furthermore, algebraic ellipticity is Zariski local, so the union Creg=C1∪C2\mathcal{C}_{\mathrm{reg}}=\mathcal{C}_1\cup\mathcal{C}_2 is algebraically elliptic [30 Theorem 0.1]. □

Remark 43. Applying Corollary 42 in the case of g=sl2(C)\mathfrak{g}=\mathfrak{sl}_2(\mathbb{C}), we see that the smooth locus

Csm=Creg=C∖{(0,0)}\mathcal{C}_{\mathrm{sm}}=\mathcal{C}_{\mathrm{reg}}=\mathcal{C}\setminus\{(0,0)\}

is algebraically elliptic (see also [43 Example 1.10]).

Corollary 44. The smooth locus Csm\mathcal{C}_{\mathrm{sm}} is Oka-1.

Proof. Set N=dim⁡C(Csm)N=\dim_{\mathbb{C}}(\mathcal{C}_{\mathrm{sm}}) and let E=Csm∖C1E=\mathcal{C}_{\mathrm{sm}}\setminus\mathcal{C}_1. Since C1\mathcal{C}_1 is Zariski dense, EE is a proper analytic subvariety with dim⁡C(E)≤dim⁡Csm−1\dim_{\mathbb{C}}(E)\leq\dim\mathcal{C}_{\mathrm{sm}}-1. Consequently, the Hausdorff measure HCsm,g2N−1(E)\mathcal{H}_{\mathcal{C}_{\mathrm{sm},g}}^{2N-1}(E) of EE is zero. Since C1\mathcal{C}_1 is algebraically elliptic, it has the LSAP property and thus Csm\mathcal{C}_{\mathrm{sm}} is Oka-1 [1 Proposition 7.13 (b)]. □

References

School of Mathematical Sciences, College of Science, Adelaide University, SA 5005, Australia

Email address: [email protected]

References

  1. [1]A. Alarcón and F. Forstnerič, Oka-1 manifolds, Math. Z. 311 (2025), no. 2, Paper No. 33, 34 pp.
  2. [2]S. An, Y.-H. Du, B. Guo, P.-C. Wang, and S.-Y. Xie, A local approximation characterization of Oka manifolds (2026), available at arXiv:2610.10475v3[math.CV]. Version 3, 19 October 2026.
  3. [3]I. Arzhantsev, H. Flenner, S. Kaliman, F. Kutzschebauch, and M. Zaidenberg, Flexible varieties and automorphism groups, Duke Math. J. 162 (2013), no. 4, 767–823.
  4. [4]I. V. Arzhantsev, M. G. Zaidenberg, and K. G. Kuyumzhiyan, Flag varieties, toric varieties, and suspensions: three examples of infinite transitivity, Mat. Sb. 203 (2012), no. 7, 3–30 (Russian, with Russian summary); English transl., Sb. Math. 203 (2012), no. 7, 923–949.
  5. [5]M. F. Atiyah and R. Bott, The Yang–Mills equations over Riemann surfaces, Philos. Trans. Roy. Soc. London Ser. A 308 (1983), no. 1505, 523–615.
  6. [7]V. Baranovsky, The variety of pairs of commuting nilpotent matrices is irreducible, Transform. Groups 6 (2001), no. 1, 3–8, DOI 10.1007/BF01236059. MR1825165DOI
  7. [10]A. Borel, Linear algebraic groups, 2nd ed., Graduate Texts in Mathematics, vol. 126, Springer-Verlag, New York, 1991.
  8. [11]A. Borel, R. Friedman, and J. W. Morgan, Almost commuting elements in compact Lie groups, Mem. Amer. Math. Soc. 157 (2002), no. 747, x+136.DOI
  9. [12]G. Carnovale and F. Esposito, On sheets of conjugacy classes in good characteristic, Int. Math. Res. Not. IMRN 4 (2012), 810–828.DOI
  10. [13]R. W. Carter, Finite groups of Lie type, Wiley Classics Library, John Wiley & Sons, Ltd., Chichester, 1993. Conjugacy classes and complex characters; Reprint of the 1985 original; A Wiley-Interscience Publication.
  11. [14]H. Flenner, S. Kaliman, and M. Zaidenberg, A Gromov-Winkelmann type theorem for flexible varieties, J. Eur. Math. Soc. (JEMS) 18 (2016), no. 11, 2483–2510.DOI
  12. [16]F. Forstnerič, Runge approximation on convex sets implies the Oka property, Ann. of Math. (2) 163 (2006), no. 2, 689–707.DOI
  13. [17]F. Forstnerič, Stein manifolds and holomorphic mappings, 2nd ed., Ergebnisse der Mathematik und ihrer Grenzgebiete. 3. Folge. A Series of Modern Surveys in Mathematics [Results in Mathematics and Related Areas. 3rd Series. A Series of Modern Surveys in Mathematics], vol. 56, Springer, Cham, 2017. The homotopy principle in complex analysis.
  14. [18]F. Forstnerič, Holomorphic embeddings and immersions of Stein manifolds: a survey, Geometric complex analysis, Springer Proc. Math. Stat., vol. 246, Springer, Singapore, 2018, pp. 145–169.arxiv.org/abs/1709.05630
  15. [21]W. Goldman, Invariant functions on Lie groups and Hamiltonian flows of surface group representations, Invent. Math. 85 (1986), no. 2, 263–302.DOI
  16. [22]M. Gromov, Oka’s principle for holomorphic sections of elliptic bundles, J. Amer. Math. Soc. 2 (1989), no. 4, 851–897.DOI
  17. [23]R. Guralnick, Similarity of holomorphic matrices, Linear Algebra Appl. 99 (1988), 85–96.DOI
  18. [24]R. Hartshorne, Algebraic geometry, Graduate Texts in Mathematics, vol. 52, Springer-Verlag, New York-Heidelberg, 1977.
  19. [25]T. Hausel and M. Thaddeus, Mirror symmetry, Langlands duality, and the Hitchin system, Invent. Math. 153 (2003), no. 1, 197–229.DOI
  20. [26]N. J. Hitchin, The self-duality equations on a Riemann surface, Proc. London Math. Soc. (3) 55 (1987), no. 1, 59–126.DOI
  21. [27]J. E. Humphreys, Linear algebraic groups, Graduate Texts in Mathematics, vol. 21, Springer-Verlag, New York-Heidelberg, 1975.DOI
  22. [28]J. E. Humphreys, Conjugacy classes in semisimple algebraic groups, Mathematical Surveys and Monographs, vol. 43, American Mathematical Society, Providence, RI, 1995.
  23. [29]B. Ivarsson and F. Kutzschebauch, Holomorphic factorization of mappings into SL_n(C), Ann. of Math. (2) 175 (2012), no. 1, 45–69.arxiv.org/abs/0812.0312
  24. [30]S. Kaliman and M. Zaidenberg, Gromov ellipticity and subellipticity, Forum Math. 36 (2024), no. 2, 373–376.arxiv.org/abs/2301.03058
  25. [31]V. G. Kac and A. V. Smilga, Vacuum structure in supersymmetric Yang-Mills theories with any gauge group, The many faces of the superworld, World Sci. Publ., River Edge, NJ, 2000, pp. 185–234.
  26. [32]B. Kostant, Lie group representations on polynomial rings, Amer. J. Math. 85 (1963), 327–404.
  27. [33]B. Kostant, The solution to a generalized Toda lattice and representation theory, Adv. in Math. 34 (1979), no. 3, 195–338.DOI
  28. [34]J. F. Kurtzke Jr., Centralizers of irregular elements in reductive algebraic groups, Pacific J. Math. 104 (1983), no. 1, 133–154.DOI
  29. [35]J. Leiterer, Similarity of holomorphic matrices on 1-dimensional Stein spaces, Functional analysis and geometry: Selim Grigorievich Krein centennial, Contemp. Math., vol. 733, Amer. Math. Soc., [Providence], RI, 2019, pp. 209–231.arxiv.org/abs/1710.10969
  30. [36]P. Levy, Commuting varieties of Lie algebras over fields of prime characteristic, J. Algebra 250 (2002), no. 2, 473–484.DOI
  31. [37]G. Malle and D. Testerman, Linear algebraic groups and finite groups of Lie type, Cambridge Studies in Advanced Mathematics, vol. 133, Cambridge University Press, Cambridge, 2011.DOI
  32. [38]G. J. McNinch and E. Sommers, Component groups of unipotent centralizers in good characteristic, J. Algebra 260 (2003), no. 1, 323–337. Special issue celebrating the 80th birthday of Robert Steinberg.
  33. [39]J. S. Milne, Algebraic groups, Cambridge Studies in Advanced Mathematics, vol. 170, Cambridge University Press, Cambridge, 2017. The theory of group schemes of finite type over a field.
  34. [41]OpenAI, Every complex K3 surface is Oka, 2026. OpenAI Math Release preprint OAI:Every-complex-K3-surface-is-Oka-September-23-2026.
  35. [43]V. Popov, Irregular and singular loci of commuting varieties, Transform. Groups 13 (2008), no. 3–4, 819–837.arxiv.org/abs/0801.3074
  36. [44]A. Premet, Nilpotent commuting varieties of reductive Lie algebras, Invent. Math. 154 (2003), no. 3, 653–683.
  37. [45]Z. Reichstein and B. Youssin, Essential dimensions of algebraic groups and a resolution theorem for G-varieties, Canad. J. Math. 52 (2000), no. 5, 1018–1056. With an appendix by János Kollár and Endre Szabó.DOI
  38. [46]R. Richardson, Commuting varieties of semisimple Lie algebras and algebraic groups, Compositio Math. 38 (1979), no. 3, 311–327.
  39. [47]J-P. Serre, Complète réductibilité, Astérisque 299 (2005), Exp. No. 932, viii, 195–217 (French, with French summary). Séminaire Bourbaki. Vol. 2003/2004.
  40. [48]I. Shafarevich, Basic algebraic geometry 1: Varieties in projective space, 3rd ed., Springer, Heidelberg, 2013.DOI
  41. [49]C. T. Simpson, Higgs bundles and local systems, Inst. Hautes Études Sci. Publ. Math. 75 (1992), 5–95.DOI
  42. [50]C. T. Simpson, Moduli of representations of the fundamental group of a smooth projective variety. I, Inst. Hautes Études Sci. Publ. Math. 79 (1994), 47–129.DOI
  43. [51]E. Sommers, A generalization of the Bala–Carter theorem for nilpotent orbits, Internat. Math. Res. Notices 11 (1998), 539–562.
  44. [52]R. Steinberg, Regular elements of semisimple algebraic groups, Inst. Hautes Études Sci. Publ. Math. 25 (1965), 49–80.DOI
  45. [53]R. Steinberg, Endomorphisms of linear algebraic groups, Memoirs of the American Mathematical Society, vol. 80, American Mathematical Society, Providence, RI, 1968.DOI

Paper details

vlnc:W_m2ddtsw9
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 0

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Mathematics

ChatGPT, GPT-5.6

View statement
Looking for related papers…
Contents