Maths WorkshopMaths problems with full solutions
Exercise 2013Topology › Topology of Mn(ℝ) › Exponential, logarithm and group topologyM1Added 6 Nov 2026

A subgroup of SO(4) that is not closed

#irrational winding#torus#Kronecker#non-closed subgroup

Statement

For t∈Rt\in\mathbb{R} let φ(t)=Rt⊕R2 t∈SO(4)\varphi(t)=R_t\oplus R_{\sqrt2\,t}\in SO(4) (block diagonal with two 2×22\times2 rotations). Show that H=φ(R)H=\varphi(\mathbb{R}) is a subgroup of SO(4)SO(4) which is not closed.

Hint

The matrices Rx⊕RyR_x\oplus R_y form a torus T≅(R/2πZ)2T\cong(\mathbb{R}/2\pi\mathbb{Z})^2. Show that HH is dense in TT but H≠TH\ne T.

Show solutionHide solution

Solution

Subgroup. φ(s+t)=φ(s)φ(t)\varphi(s+t)=\varphi(s)\varphi(t) and φ(−t)=φ(t)−1\varphi(-t)=\varphi(t)^{-1}, so HH is a subgroup of the torus T={Rx⊕Ry:x,y∈R}T=\{R_x\oplus R_y:x,y\in\mathbb{R}\}, which is a closed subgroup of SO(4)SO(4) (compact: image of [0,2π]2[0,2\pi]^2 under a continuous map).

Density in TT. *Step 1: {n2 mod 1}\{n\sqrt2\bmod1\} is dense in [0,1)[0,1).* Divide [0,1)[0,1) into NN intervals of length 1/N1/N. Among the N+1N+1 numbers {k2}\{k\sqrt2\}, k=0,…,Nk=0,\dots,N (fractional parts), two lie in the same interval (pigeonhole): ∣{k2}−{l2}∣<1/N|\{k\sqrt2\}-\{l\sqrt2\}|<1/N, so there is m=k−l≠0m=k-l\ne0 with ∥m2∥<1/N\|m\sqrt2\|<1/N (∥⋅∥\|\cdot\|: distance to the nearest integer), and ∥m2∥≠0\|m\sqrt2\|\ne0 since 2\sqrt2 is irrational. The multiples jm2jm\sqrt2, j∈Zj\in\mathbb{Z}, move along the circle R/Z\mathbb{R}/\mathbb{Z} in steps of size <1/N<1/N, so they come within 1/N1/N of every point.

*Step 2.* Let (x,y)∈R2(x,y)\in\mathbb{R}^2 and ε>0\varepsilon>0. We have φ(t+2πn)=Rt⊕R2t+2π2n\varphi(t+2\pi n)=R_t\oplus R_{\sqrt2t+2\pi\sqrt2n}. Take t=xt=x. By step 1 there is an integer nn with 2π2n2\pi\sqrt2n within ε\varepsilon of y−2xy-\sqrt2x modulo 2π2\pi (the multiples n2n\sqrt2 are dense modulo 11, so the multiples 2πn22\pi n\sqrt2 are dense modulo 2π2\pi). Then φ(x+2πn)\varphi(x+2\pi n) is within distance O(ε)O(\varepsilon) of Rx⊕RyR_x\oplus R_y. So HH is dense in TT.

H≠TH\ne T. The element R0⊕Rπ=diag(1,1,−1,−1)∈TR_0\oplus R_\pi=\mathrm{diag}(1,1,-1,-1)\in T is not in HH: φ(t)=R0⊕Rπ\varphi(t)=R_0\oplus R_\pi requires t=2πmt=2\pi m and 2t=π+2πl\sqrt2t=\pi+2\pi l for integers m,lm,l, i.e. 22m=1+2l2\sqrt2m=1+2l, impossible since the left side is irrational for m≠0m\ne0 and for m=0m=0 it gives 0=1+2l0=1+2l.

Not closed. If HH were closed in SO(4)SO(4), then, being dense in the closed subset TT, we would get T⊂H‾=HT\subset\overline H=H, contradicting H≠TH\ne T.

Remark. HH is a one-parameter subgroup that "winds" densely on the torus: it is an injective continuous image of R\mathbb{R} (the map φ\varphi is injective) but not a closed embedded curve. This is the standard example showing that Lie subgroups need not be closed.