Maths WorkshopMaths problems with full solutions
Exercise 1345Topology › Topology of Mn(ℝ) › Applications and the case n = 2L3Added 6 Nov 2026

The quotient of M_n(R) by conjugation is not Hausdorff

#quotient topology#conjugation#Jordan block#separation axioms

Statement

Let GLn(R)GL_n(\mathbb{R}) act on Mn(R)M_n(\mathbb{R}) by conjugation and give the orbit space Mn(R)/GLn(R)M_n(\mathbb{R})/GL_n(\mathbb{R}) the quotient topology. Show that for n≥2n\ge2 it is not Hausdorff, and not even T1T_1.

Hint

Compare the orbit of the identity with the orbit of the Jordan block J=(1101)⊕In−2J=\begin{pmatrix}1&1\\0&1\end{pmatrix}\oplus I_{n-2}. The matrices Jt=(1t01)⊕In−2J_t=\begin{pmatrix}1&t\\0&1\end{pmatrix}\oplus I_{n-2} tend to II as t→0t\to0 but are all conjugate to JJ for t≠0t\ne0.

Show solutionHide solution

Solution

Idea. The orbit of the Jordan block is not closed: its closure contains the orbit {I}\{I\} of the identity. In the quotient, any neighbourhood of [I][I] then contains [J][J].

Step 1: JtJ_t is conjugate to JJ. For t≠0t\ne0, let Dt=diag⁡(t,1,…,1)D_t=\operatorname{diag}(t,1,\dots,1). Then
DtJDt−1=(1t01)⊕In−2=Jt.D_tJD_t^{-1}=\begin{pmatrix}1&t\\0&1\end{pmatrix}\oplus I_{n-2}=J_t.
(Check in the 2×22\times2 block: (t001)(1101)(1/t001)=(tt01)(1/t001)=(1t01)\begin{pmatrix}t&0\\0&1\end{pmatrix}\begin{pmatrix}1&1\\0&1\end{pmatrix}\begin{pmatrix}1/t&0\\0&1\end{pmatrix}=\begin{pmatrix}t&t\\0&1\end{pmatrix}\begin{pmatrix}1/t&0\\0&1\end{pmatrix}=\begin{pmatrix}1&t\\0&1\end{pmatrix}.) Also Jt→IJ_t\to I as t→0t\to0, and J≠IJ\ne I (and JJ is not conjugate to II, since PIP−1=IPIP^{-1}=I).

Step 2: neighbourhoods of [I][I] contain [J][J]. Let π:Mn→Mn/GLn\pi:M_n\to M_n/GL_n be the quotient map and WW an open set of the quotient containing [I][I]. Then π−1(W)\pi^{-1}(W) is open, contains II, and is a union of orbits. As Jt→IJ_t\to I, Jt∈π−1(W)J_t\in\pi^{-1}(W) for small t≠0t\ne0; since JtJ_t is conjugate to JJ and π−1(W)\pi^{-1}(W) is a union of orbits, J∈π−1(W)J\in\pi^{-1}(W), i.e. [J]∈W[J]\in W.

Step 3: conclusion. Every open neighbourhood of [I][I] contains the different point [J][J]. So no open sets separate [I][I] and [J][J] (not Hausdorff), and there is no neighbourhood of [I][I] missing [J][J] (not T1T_1: the point [J][J] is not closed, because every neighbourhood of [I][I] meets {[J]}\{[J]\}).

Remark. The quotient is well behaved on the open set of matrices with nn distinct real eigenvalues, where conjugacy classes are classified by the eigenvalues (and are closed). The "bad" behaviour comes from non-diagonalisable matrices: the closure of a conjugacy class contains matrices with the same characteristic polynomial but "more diagonal" Jordan form.

Lesson. Closures of orbits that are not closed destroy separation axioms in the quotient.