Maths WorkshopMaths problems with full solutions
Exercise 2043Topology › Topology of Mn(ℝ) › Open, closed, interior, boundaryM1Added 7 Nov 2026

The closure of the real-diagonalisable matrices

#diagonalisable#closure#triangularisation#continuity of roots

Statement

Show that the closure of the set of R\mathbb{R}-diagonalisable matrices is {A:χA splits over R}\{A:\chi_A\text{ splits over }\mathbb{R}\}, i.e. the set of matrices all of whose eigenvalues are real.

Hint

(⊂\subset) the roots of a limit of real-rooted polynomials are real. (⊃\supset) triangularise over R\mathbb{R} and perturb the diagonal to make the eigenvalues distinct.

Show solutionHide solution

Solution

Idea. The only obstruction to being a limit of diagonalisable matrices is having non-real eigenvalues. Real-rooted matrices can always be nudged to have distinct real eigenvalues.

Let T={A:χA splits over R}\mathcal T=\{A:\chi_A\text{ splits over }\mathbb{R}\} (all roots real, with multiplicity).

Step 1: DR‾⊂T\overline{\mathcal D_{\mathbb{R}}}\subset\mathcal T. If AA is diagonalisable over R\mathbb{R} then A∈TA\in\mathcal T, so DR⊂T\mathcal D_{\mathbb{R}}\subset\mathcal T. It suffices to show that T\mathcal T is closed. Let Ak∈TA_k\in\mathcal T, Ak→AA_k\to A, and let zz be a root of χA\chi_A. By continuity of roots (Part 3), for kk large there is a root zkz_k of χAk\chi_{A_k} with ∣z−zk∣→0|z-z_k|\to0. The roots zkz_k are real, and the limit of real numbers is real, so z∈Rz\in\mathbb{R}. Hence all roots of χA\chi_A are real: A∈TA\in\mathcal T.

Step 2: T⊂DR‾\mathcal T\subset\overline{\mathcal D_{\mathbb{R}}}. Let A∈TA\in\mathcal T. Since all eigenvalues are real, AA is triangularisable over R\mathbb{R} (Schur: A=PTP−1A=PTP^{-1} with PP invertible, TT upper triangular with the eigenvalues on the diagonal; one can take PP orthogonal). Let Δε=εdiag⁡(1,2,…,n)\Delta_\varepsilon=\varepsilon\operatorname{diag}(1,2,\dots,n). The matrix Tε=T+ΔεT_\varepsilon=T+\Delta_\varepsilon is upper triangular with diagonal entries tii+iεt_{ii}+i\varepsilon. These are pairwise distinct for all small ε≠0\varepsilon\ne0: tii+iε=tjj+jεt_{ii}+i\varepsilon=t_{jj}+j\varepsilon forces ε=tii−tjjj−i\varepsilon=\frac{t_{ii}-t_{jj}}{j-i} for some pair i≠ji\neq j, which excludes only finitely many values of ε\varepsilon. For such ε\varepsilon the matrix TεT_\varepsilon has nn distinct real eigenvalues, so is diagonalisable over R\mathbb{R}, and so is Aε=PTεP−1A_\varepsilon=PT_\varepsilon P^{-1}. Finally ∥Aε−A∥=∥PΔεP−1∥≤εn∥P∥∥P−1∥→0\|A_\varepsilon-A\|=\|P\Delta_\varepsilon P^{-1}\|\le\varepsilon n\|P\|\|P^{-1}\|\to0.

Remark.
- Example: (1101)∈T\begin{pmatrix}1&1\\0&1\end{pmatrix}\in\mathcal T is the limit of (1101+ε)\begin{pmatrix}1&1\\0&1+\varepsilon\end{pmatrix}; but the rotation (0−110)∉T\begin{pmatrix}0&-1\\1&0\end{pmatrix}\notin\mathcal T and is at positive distance from DR\mathcal D_{\mathbb{R}}.
- Thus the three sets satisfy S=int⁡DR⊂DR⊂DR‾=T\mathcal S=\operatorname{int}\mathcal D_{\mathbb{R}}\subset\mathcal D_{\mathbb{R}}\subset\overline{\mathcal D_{\mathbb{R}}}=\mathcal T, with both inclusions strict for n≥2n\ge2.