Maths WorkshopMaths problems with full solutions
Exercise 2064Topology › Topology of Mn(ℝ) › Applications and the case n = 2M1Added 10 Nov 2026

The complex structures A² =-I on form exactly two components

#conjugacy orbits#complex structures#orientation#connected components#GL_n

Statement

Let n=2mn=2m be even and K={A∈Mn(R):A2=−I}K=\{A\in M_n(\mathbb{R}):A^2=-I\} (the set of complex structures on Rn\mathbb{R}^n). Show that KK has exactly two connected components. (You may use that GLn(R)GL_n(\mathbb{R}) has exactly two components GLn±GL_n^\pm and that GLm(C)GL_m(\mathbb{C}) is connected.) Note that K=∅K=\emptyset for nn odd.

Hint

Every A∈KA\in K is conjugate to J0=diag⁡(J,…,J)J_0=\operatorname{diag}(J,\dots,J) with J=(0−110)J=\begin{pmatrix}0&-1\\1&0\end{pmatrix}, by some P∈GLn(R)P\in GL_n(\mathbb{R}). The centraliser of J0J_0 is GLm(C)GL_m(\mathbb{C}), which has positive determinant when viewed as a real matrix. Use this to define a continuous "orientation" ε:K→{±1}\varepsilon:K\to\{\pm1\}.

Show solutionHide solution

Solution

Idea. A matrix AA with A2=−IA^2=-I makes Rn\mathbb{R}^n into a complex vector space (multiplication by ii is AA). Complex bases carry a canonical real orientation; comparing it with the standard orientation gives a sign ε(A)\varepsilon(A), which turns out to be locally constant on KK.

Step 1: normal form. If A2=−IA^2=-I then (det⁡A)2=det⁡(−I)=(−1)n(\det A)^2=\det(-I)=(-1)^n, so nn is even. Let A∈KA\in K. Make V=RnV=\mathbb{R}^n a complex vector space VAV_A by (s+it)x=sx+tAx(s+it)x=sx+tAx (this is a complex structure because A2=−IA^2=-I). Its complex dimension is m=n/2m=n/2. Choose a C\mathbb{C}-basis v1,…,vmv_1,\dots,v_m of VAV_A. Then (v1,Av1,…,vm,Avm)(v_1,Av_1,\dots,v_m,Av_m) is a real basis of Rn\mathbb{R}^n (the real span of vk,Avkv_k,Av_k is the complex span of vkv_k). Let PP be the matrix with these columns. Then Avk=Pe2kAv_k=Pe_{2k} and A(Avk)=−vkA(Av_k)=-v_k, i.e. APe2k−1=Pe2kAPe_{2k-1}=Pe_{2k} and APe2k=−Pe2k−1APe_{2k}=-Pe_{2k-1}. Thus
AP=PJ0,A=PJ0P−1.AP=PJ_0,\qquad A=PJ_0P^{-1}.

Step 2: the sign ε(A)\varepsilon(A). For A∈KA\in K and a list v=(v1,…,vm)v=(v_1,\dots,v_m) that is a C\mathbb{C}-basis of VAV_A, put
BA(v)=[ v1,Av1,…,vm,Avm ]∈GLn(R),ε(A)=sign⁡det⁡BA(v).B_A(v)=[\,v_1,Av_1,\dots,v_m,Av_m\,]\in GL_n(\mathbb{R}),\qquad\varepsilon(A)=\operatorname{sign}\det B_A(v).
*This does not depend on vv.* If v′v' is another C\mathbb{C}-basis, v′=G⋅vv'=G\cdot v for a unique G∈GLm(C)G\in GL_m(\mathbb{C}), and then BA(v′)=BA(v) ρ(G)B_A(v')=B_A(v)\,\rho(G), where ρ(G)∈GLn(R)\rho(G)\in GL_n(\mathbb{R}) is the "realification" of GG (it replaces each entry g=x+iyg=x+iy by the block (x−yyx)\begin{pmatrix}x&-y\\y&x\end{pmatrix}). The function G↦det⁡ρ(G)G\mapsto\det\rho(G) is continuous and nowhere zero on GLm(C)GL_m(\mathbb{C}), which is connected, and it equals 11 at G=IG=I. So it is positive everywhere (in fact det⁡ρ(G)=∣det⁡G∣2\det\rho(G)=|\det G|^2), and sign⁡det⁡BA(v′)=sign⁡det⁡BA(v)\operatorname{sign}\det B_A(v')=\operatorname{sign}\det B_A(v).

Step 3: ε\varepsilon is locally constant. Fix A0∈KA_0\in K and a C\mathbb{C}-basis vv of VA0V_{A_0}. The matrix BA(v)B_A(v) has entries that are linear in AA (columns vkv_k and AvkAv_k). Since det⁡BA0(v)≠0\det B_{A_0}(v)\ne0, there is a neighbourhood UU of A0A_0 in Mn(R)M_n(\mathbb{R}) on which det⁡BA(v)\det B_A(v) has the same sign and is nonzero. For A∈U∩KA\in U\cap K, the vectors vk,Avkv_k,Av_k (k≤mk\le m) form a real basis, so vv is a C\mathbb{C}-basis of VAV_A (its complex span contains all vkv_k and AvkAv_k). Hence ε(A)=sign⁡det⁡BA(v)=ε(A0)\varepsilon(A)=\operatorname{sign}\det B_A(v)=\varepsilon(A_0) on U∩KU\cap K. So ε:K→{±1}\varepsilon:K\to\{\pm1\} is continuous.

Step 4: the two pieces. In Step 1 we saw that for A=PJ0P−1A=PJ_0P^{-1} one may take vk=Pe2k−1v_k=Pe_{2k-1} and then BA(v)=PB_A(v)=P. Hence
ε(PJ0P−1)=sign⁡det⁡P.\varepsilon(PJ_0P^{-1})=\operatorname{sign}\det P.
Therefore
K+=ε−1(+1)={PJ0P−1:P∈GLn+},K−=ε−1(−1)={PJ0P−1:P∈GLn−}.K^+=\varepsilon^{-1}(+1)=\{PJ_0P^{-1}:P\in GL_n^+\},\qquad K^-=\varepsilon^{-1}(-1)=\{PJ_0P^{-1}:P\in GL_n^-\}.
Both are nonempty (J0∈K+J_0\in K^+; DJ0D−1∈K−DJ_0D^{-1}\in K^- for D=diag⁡(−1,1,…,1)D=\operatorname{diag}(-1,1,\dots,1)). Each is the image of the connected set GLn±GL_n^\pm under the continuous map P↦PJ0P−1P\mapsto PJ_0P^{-1}, hence connected. Since ε\varepsilon is continuous, K+K^+ and K−K^- are open and closed in K=K+⊔K−K=K^+\sqcup K^-. Thus KK has exactly two components.

Check (m=1m=1). For n=2n=2 the construction gives the two sheets Im⁡z=±1+∣w∣2\operatorname{Im}z=\pm\sqrt{1+|w|^2} of the problem on square roots of −I-I in M2(R)M_2(\mathbb{R}). A numerical test in n=4n=4 with random PP confirms that sign⁡det⁡BA(v)=sign⁡det⁡P\operatorname{sign}\det B_A(v)=\operatorname{sign}\det P for random complex bases vv of VAV_A.

Remark. Another way to phrase it: KK is the conjugacy orbit GLn(R)/GLm(C)GL_n(\mathbb{R})/GL_m(\mathbb{C}). The subgroup GLm(C)⊂GLn+GL_m(\mathbb{C})\subset GL_n^+ is connected, so π0\pi_0 of the orbit equals π0(GLn(R))=Z/2\pi_0(GL_n(\mathbb{R}))=\mathbb{Z}/2. Also A↦−AA\mapsto-A changes ε\varepsilon by (−1)m(-1)^m: it swaps the two components iff mm is odd.

Lesson. To prove that a set has at least two components, produce a continuous invariant with two values; to prove at most two, exhibit it as the continuous image of a space with two components.