The Minimal Polynomial and the Two Jordan Regimes of $M_2(\mathbb{C})$

Introduction

At order two the minimal polynomial has degree one or two, and the degree decides the whole Jordan theory.

The article proves that the minimal polynomial of $A$ is $\lambda-\lambda_0$ exactly when $A$ is scalar and equals the characteristic polynomial otherwise, describes the three Jordan forms of order two, separates the semisimple regime from the single Jordan block, computes the Jordan–Chevalley decomposition explicitly through the traceless part, and reads the algebra $\mathbb{C}[A]$ generated by one element. It assumes The Element: Rank, Trace, Determinant, Characteristic Polynomial at $n=2$ for the traceless part and its relation $A_0^{2}=-\det(A_0)I$, and The Elementary Symmetric Functions: $\operatorname{tr}$ and $\det$ in Dimension Two for the discriminant.

It defers the classification of the conjugacy classes to The Spectrum of a $2\times2$ Matrix, and the idempotent built on a simple spectrum to Lagrange Idempotents of a $2\times2$ Matrix with Simple Spectrum. The general theory of the corner recursion is The Peirce Decomposition and the Corner Recursion of $M_n(\mathbb{C})$; at order two the recursion has depth one and produces nothing new.

The Minimal Polynomial

Definition. The minimal polynomial $m_A$ is the monic polynomial of least degree with $m_A(A)=0$.

Proposition. The minimal polynomial exists, is unique, divides every polynomial that vanishes at $A$, and divides the characteristic polynomial.

Proof. The powers $I,A,A^{2},\dots$ lie in a four-dimensional space, so some nontrivial linear combination vanishes and a monic polynomial of least degree exists; the set of vanishing polynomials is an ideal of $\mathbb{C}[\lambda]$, which is principal, and its monic generator is unique. Since $\chi_A(A)=0$ by Cayley–Hamilton, $\chi_A$ lies in that ideal, so $m_A$ divides it.

Corollary. The degree of $m_A$ is one or two.

Proof. The characteristic polynomial has degree two, and the minimal polynomial is a monic divisor of it of positive degree.

Proposition. $\deg m_A=1$ exactly when $A$ is a scalar matrix.

Proof. If $m_A=\lambda-\lambda_0$ then $A-\lambda_0I=0$, so $A=\lambda_0I$; conversely a scalar matrix is annihilated by $\lambda-\lambda_0$.

Corollary. For a nonscalar $A$ the minimal polynomial is the characteristic polynomial, that is $m_A=\chi_A=\lambda^{2}-(\operatorname{tr}A)\lambda+\det A$.

The number of distinct roots is the number of eigenspaces. A matrix of order two is diagonalisable exactly when its minimal polynomial has simple roots, because the Chinese remainder theorem then splits the space into the kernels of $\lambda-\lambda_i$. That criterion is the one used throughout, and it needs no Jordan form.

Three Cases, Two Regimes

Theorem. A nonscalar element of $M_2(\mathbb{C})$ satisfies exactly one of the two following conditions:

$$ \Delta\neq0\ \text{(two distinct eigenvalues, diagonalisable)},\qquad \Delta=0\ \text{(one eigenvalue, one Jordan block)} . $$

Proof. By the previous section $m_A=\chi_A$, so the roots of $m_A$ are the eigenvalues. If $\Delta\neq0$ the two roots are distinct and $A$ is diagonalisable. If $\Delta=0$ then $\chi_A=(\lambda-\lambda_0)^{2}$ and $m_A=\chi_A$ has the double root $\lambda_0$; a matrix whose minimal polynomial has a repeated root is not diagonalisable, because a diagonal matrix with the single eigenvalue $\lambda_0$ would be $\lambda_0I$, which is excluded.

Corollary (the two regimes). The semisimple regime is the union of the scalars and the set $\{\Delta\neq0\}$; the nilpotent regime is the set $\{\Delta=0\}\setminus\{\lambda I\}$, on which $A=\lambda I+N$ with $N\neq0$ nilpotent.

The regimes are separated by one polynomial. The discriminant is a quadratic form in the entries, so the boundary between the regimes is a cone of complex dimension $3$, and the nilpotent regime is that cone minus the scalars. The regular semisimple set is the complement, and it is open and dense.

Everything is decided by the discriminant. Diagonalisable, one eigenline, two eigenlines: each of these is a property of $\Delta$ and of the rank of $A-\lambda I$. At order two there is only one possible degeneracy, the merging of the two eigenvalues, and it cannot be partial.

The Jordan Forms at Order Two

Theorem. Every element of $M_2(\mathbb{C})$ is conjugate to exactly one of

$$ \begin{pmatrix}\lambda_1&0\\ 0&\lambda_2\end{pmatrix}\ (\lambda_1\neq\lambda_2),\qquad \begin{pmatrix}\lambda&0\\ 0&\lambda\end{pmatrix},\qquad \begin{pmatrix}\lambda&1\\ 0&\lambda\end{pmatrix}. $$

Proof. If $\Delta\neq0$ the two eigenlines are independent and the matrix is diagonal in their basis. If $\Delta=0$ and $A=\lambda I$ the matrix is the second form. Otherwise $N=A-\lambda I$ is nonzero and nilpotent, hence of rank one, and a basis $(v,Nv)$ with $v\notin\ker N$ puts it in the third form: $Nv$ is a nonzero vector killed by $N$, so the matrix of $N$ is $E_{12}$.

Corollary (the single nilpotent orbit). All nonzero nilpotents of order two are conjugate.

Proof. Each is conjugate to $E_{12}$ by the theorem, and conjugacy is transitive. This is where the dimension pays: at order three the nilpotent orbits are two, indexed by the partitions $3$ and $2+1$, while at order two only the partition $2$ survives.

The companion pairs. The two matrices $\begin{pmatrix}0&1\\ 0&0\end{pmatrix}$ and $\begin{pmatrix}0&0\\ 1&0\end{pmatrix}$ are conjugate by the swap of the two basis vectors, which is $\begin{pmatrix}0&1\\ 1&0\end{pmatrix}$ of square $I$ and determinant $-1$. In the biquaternion language these are $\vec\Upsilon_{+-}$ and $\vec\Upsilon_{-+}$, and the conjugating swap is one point of the gauge of The Companion Nilpotents of the Biquaternion Algebra.

The Jordan–Chevalley Decomposition

Theorem. Every $A$ has a unique decomposition $A=A_{s}+A_{n}$ with $A_{s}$ semisimple, $A_{n}$ nilpotent and $A_{s}A_{n}=A_{n}A_{s}$.

Proof. If $A$ is scalar, take $A_{s}=A$ and $A_{n}=0$. If $\Delta\neq0$, $A$ is semisimple, so $A_{s}=A$ and $A_{n}=0$. If $\Delta=0$ and $A$ is not scalar, put $A_{s}=\tfrac12(\operatorname{tr}A)I$ and $A_{n}=A_{0}$; the traceless part has $\det A_{0}=-\tfrac14\Delta=0$, hence $A_{0}^{2}=0$ by the identity $A_{0}^{2}=-\det(A_{0})I$, so $A_{n}$ is nilpotent and it commutes with the scalar $A_{s}$. Uniqueness follows because the semisimple part must have the same spectrum as $A$ and commute with it.

Corollary (the two regimes in coordinates). The decomposition is

$$ A=\tfrac12(\operatorname{tr}A)I+A_{0}, $$

and the traceless part is nilpotent exactly when $\Delta=0$. The nilpotent regime is the regime in which the traceless part is the nilpotent part.

Proof. The formula is the split of The Element: Rank, Trace, Determinant, Characteristic Polynomial at $n=2$. The traceless part is nilpotent exactly when $\det A_{0}=0$, and $\det A_{0}=-\tfrac14\Delta$; when it is not nilpotent it is semisimple, because a traceless element with nonzero determinant has the two distinct eigenvalues $\pm\sqrt{-\det A_{0}}$.

The scalar part is the spectral average. At order two the semisimple part in the degenerate case is the average of the two equal eigenvalues, that is $\tfrac12(\operatorname{tr}A)I$; at order $n$ it would be the Lagrange interpolant of the spectrum on the distinct eigenvalues, which is Roots of Unity and the Lagrange Idempotents of $M_n(\mathbb{C})$ at general order.

The Algebra Generated by One Element

Definition. $\mathbb{C}[A]$ is the subalgebra generated by $A$ and $I$, that is the image of the map $P\mapsto P(A)$ from polynomials to the algebra.

Proposition. If $A$ is scalar then $\mathbb{C}[A]=\mathbb{C}I$ is one-dimensional; if $A$ is not scalar then $\mathbb{C}[A]$ is two-dimensional, with basis $I,A$, and it is isomorphic to $\mathbb{C}[\lambda]/(\chi_A)$.

Proof. The kernel of $P\mapsto P(A)$ is the ideal generated by $m_A$, of degree $1$ or $2$; the image has dimension equal to that degree. For a nonscalar $A$ the polynomials $I,A$ are independent, and every higher power reduces to a combination of them by Cayley–Hamilton.

Corollary. A nonscalar element generates a commutative algebra of dimension $2$ inside $M_2(\mathbb{C})$, and that algebra is its own centraliser.

Proof. The centraliser of $A$ contains $\mathbb{C}[A]$; its dimension is $2$ when the eigenvalues are distinct, since the centraliser is then the set of matrices diagonal in the eigenbasis, and $2$ also in the Jordan case, where the centraliser is the set of upper triangular matrices in the Jordan basis. So the two spaces coincide.

The two-dimensional algebra is a plane in a four-dimensional space. Its projectivisation is a line, and the line is the pencil spanned by $I$ and $A$. That line is the first case of the two-sided operators of The Two-Sided Operators and the Segre Cone at $n=2$, where the pencils through a point are classified.

The Discriminant Strata

The discriminant stratifies the algebra in three pieces. The pieces are $\{\Delta\neq0\}$, the open dense set; $\{\Delta=0\}\setminus\mathbb{C}I$, of complex dimension $3$; and $\mathbb{C}I$, of dimension $1$.

Proposition. On the open set the map $A\mapsto(\operatorname{tr}A,\det A)$ is a submersion and its fibres are the conjugacy classes; on the discriminant cone the fibres are unions of classes, degenerate along the scalars.

Proof. The Jacobian statement is The Elementary Symmetric Functions: $\operatorname{tr}$ and $\det$ in Dimension Two. A fibre over a point with $\Delta\neq0$ consists of matrices with two prescribed distinct eigenvalues, and all of them are conjugate; a fibre over a point with $\Delta=0$ contains the scalar $\tfrac12\tau I$ together with all the Jordan matrices of that eigenvalue, which form a single class of dimension $2$ plus the point.

Corollary. The nilpotent regime is a single conjugacy orbit for each eigenvalue $\lambda$, namely $\lambda I$ plus the nonzero nilpotents.

Proof. Two matrices $\lambda I+N$ and $\lambda I+N'$ with $N,N'$ nonzero nilpotent are conjugate by the conjugacy of the nilpotents, and the scalar part is unchanged by conjugation.

The Real Reading and the Irreducible Case

Over $\mathbb{R}$ the characteristic polynomial can be irreducible, and then the minimal polynomial is irreducible too.

Proposition. For $X\in M_2(\mathbb{R})$ with $\Delta<0$, the polynomial $\chi_X$ is irreducible over $\mathbb{R}$, and $X$ is not diagonalisable over $\mathbb{R}$ but is diagonalisable over $\mathbb{C}$.

Proof. A real quadratic with negative discriminant has no real root, hence no linear factor. Over $\mathbb{C}$ the two roots are distinct, so the minimal polynomial has simple roots and the matrix is diagonalisable; the eigenvectors are conjugates of one another and not real.

Example. The rotation of angle $\theta$ has $\chi=\lambda^{2}-2\cos\theta\,\lambda+1$ and $\Delta=4(\cos^{2}\theta-1)<0$ for $\theta\notin\pi\mathbb{Z}$. The real algebra $\mathbb{R}[X]$ is then isomorphic to $\mathbb{C}$, generated by $X$ with $X^{2}-2\cos\theta\,X+I=0$; the isomorphism sends $X$ to $e^{i\theta}$.

Corollary. Over $\mathbb{R}$ the three regimes are: two real eigenvalues, one real eigenvalue with a Jordan block, and a conjugate pair. The third has no analogue over $\mathbb{C}$, where the eigenvalues always exist.

Summary

The minimal polynomial of $A\in M_2(\mathbb{C})$ is $\lambda-\lambda_0$ exactly for the scalars and equals $\chi_A$ otherwise, so its degree is one or two and it has a repeated root exactly on the discriminant cone $\Delta=0$. A nonscalar matrix is therefore either semisimple with two distinct eigenvalues or conjugate to a single Jordan block $\begin{pmatrix}\lambda&1\\0&\lambda\end{pmatrix}$, and all nonzero nilpotents are conjugate.

The Jordan–Chevalley decomposition is $A=\tfrac12(\operatorname{tr}A)I+A_{0}$, with the traceless part nilpotent exactly when $\Delta=0$; otherwise the traceless part is semisimple and the decomposition is trivial. The algebra $\mathbb{C}[A]$ has dimension $1$ for scalars and $2$ otherwise, with basis $I,A$, and it equals the centraliser. Over $\mathbb{R}$ a negative discriminant gives an irreducible characteristic polynomial and a copy of $\mathbb{C}$ inside $M_2(\mathbb{R})$.

Summary of Notation

Symbol Meaning
$m_A$ the minimal polynomial of $A$
$\chi_A$ the characteristic polynomial
$\Delta$ the discriminant $(\operatorname{tr}A)^{2}-4\det A$
$A_{s},A_{n}$ the semisimple and nilpotent parts of $A$
$A_{0}$ the traceless part, the nilpotent part in the degenerate case
$N$ a nilpotent element, $N^{2}=0$
$\mathbb{C}[A]$ the algebra generated by $A$, of dimension $\deg m_A$
$\lambda$ the single eigenvalue in the Jordan regime

Further Reading