Lagrange Idempotents of a $2\times2$ Matrix with Simple Spectrum

Introduction

A two-by-two matrix with two distinct eigenvalues has exactly two Lagrange idempotents $\rho_{\pm}$, each of rank one and trace one, and they reconstruct the matrix as $\lambda_{+}\rho_{+}+\lambda_{-}\rho_{-}$.

The article defines the idempotents by interpolation at the two eigenvalues, proves the four identities that characterise them, exhibits the involution $\rho_{+}-\rho_{-}$ as the normalised shift of the matrix, writes the idempotents in the trace and the determinant through the discriminant, and identifies them with the spectral projectors of the resolvent. It assumes The Elementary Symmetric Functions: $\operatorname{tr}$ and $\det$ in Dimension Two and The Minimal Polynomial and the Two Jordan Regimes of $M_2(\mathbb{C})$, and it defers the general spectral theory to The Spectrum of a $2\times2$ Matrix.

On the biquaternion side the same pair is $\tilde\Pi_{\pm}=\tfrac12(e_t\pm\tilde{Ζ})$ in The Square Roots of One and the Idempotents of the Biquaternion Algebra and The Peirce Basis of the Biquaternion Algebra, and the translation is the one of $M_2(\mathbb{C})$ Pauli Representation.

The Interpolation Problem

The Lagrange problem asks for a polynomial of degree one taking prescribed values at two points. For two distinct numbers $\lambda_{+}$ and $\lambda_{-}$ there are exactly two such polynomials for the two values $1$ and $0$.

Proposition. The polynomials are $\ell_{+}(z)=\dfrac{z-\lambda_{-}}{\lambda_{+}-\lambda_{-}}$ and $\ell_{-}(z)=\dfrac{z-\lambda_{+}}{\lambda_{-}-\lambda_{+}}$.

Proof. Each is of degree one, $\ell_{+}(\lambda_{+})=1$ and $\ell_{+}(\lambda_{-})=0$, and the two conditions determine a polynomial of degree one uniquely.

Corollary. For every $z$ one has $\ell_{+}(z)+\ell_{-}(z)=1$ and $\ell_{+}(z)\ell_{-}(z)=0$ exactly at $z=\lambda_{\pm}$.

Proof. The two numerators add to $\lambda_{+}-\lambda_{-}$ up to the denominator, and the product vanishes at the two nodes.

Definition. The Lagrange idempotents of a matrix $X$ with simple spectrum are $\rho_{\pm}=\ell_{\pm}(X)$.

Corollary. The idempotents are polynomials of degree one in $X$, so they commute with $X$ and with every matrix that commutes with $X$.

Proof. A polynomial in $X$ commutes with $X$, and it commutes with everything in the centraliser of $X$.

The Four Identities

Theorem. For a matrix with two distinct eigenvalues the Lagrange idempotents satisfy

$$ \rho_{+}+\rho_{-}=I,\qquad \rho_{+}\rho_{-}=0,\qquad \rho_{\pm}^{2}=\rho_{\pm},\qquad X=\lambda_{+}\rho_{+}+\lambda_{-}\rho_{-}. $$

Proof. The first two follow from the two scalar identities of interpolation applied to $X$; the third follows from the second and the first; the fourth is the interpolation of the identity function $z\mapsto z$, which is of degree one and takes the values $\lambda_{\pm}$ at the two nodes.

Corollary. Each idempotent has rank one, trace one and determinant zero.

Proof. The trace is additive and $\operatorname{tr}(I)=2$, so $\operatorname{tr}\rho_{+}+\operatorname{tr}\rho_{-}=2$; the products $X\rho_{\pm}=\lambda_{\pm}\rho_{\pm}$ show that the image of $\rho_{\pm}$ lies in the eigenspace of $\lambda_{\pm}$, which is one-dimensional; hence the rank is one and the trace is the rank.

Corollary. $X\rho_{\pm}=\lambda_{\pm}\rho_{\pm}$ and $\rho_{\pm}X=\lambda_{\pm}\rho_{\pm}$, so the columns of $\rho_{+}$ are eigenvectors of $\lambda_{+}$ and its rows are left eigenvectors.

Proof. Multiply the reconstruction identity by $\rho_{\pm}$ and use orthogonality and idempotence.

Corollary. The pair is unique: any two idempotents summing to the identity and satisfying the reconstruction identity are the Lagrange ones.

Proof. The reconstruction identity forces $X\rho_{\pm}=\lambda_{\pm}\rho_{\pm}$, and an idempotent with a one-dimensional image spanned by an eigenvector is determined by that eigenvector.

The Involution of the Pair

Theorem. The difference $\zeta=\rho_{+}-\rho_{-}$ is an involution: $\zeta^{2}=I$, and $\rho_{\pm}=\tfrac12(I\pm\zeta)$.

Proof. Expand $\zeta^{2}=\rho_{+}+\rho_{-}$ using orthogonality and idempotence, which gives $I$; the second formula is the sum and the difference of the two.

Corollary. In terms of the matrix, $\zeta=\dfrac{2X-(\operatorname{tr}X)I}{\sqrt{\Delta}}$ with $\Delta=(\operatorname{tr}X)^{2}-4\det X$ and the sign of the square root fixed by $\lambda_{+}-\lambda_{-}$.

Proof. Adding the two formulas $\rho_{\pm}=\pm\dfrac{X-\lambda_{\mp}I}{\lambda_{+}-\lambda_{-}}$ gives $\zeta=\dfrac{2X-(\lambda_{+}+\lambda_{-})I}{\lambda_{+}-\lambda_{-}}$, and the denominator is the square root of the discriminant.

Corollary. The involution $\zeta$ has trace zero and determinant $-1$, and it is the normalised traceless part of the matrix.

Proof. $\operatorname{tr}\zeta=\operatorname{tr}\rho_{+}-\operatorname{tr}\rho_{-}=1-1=0$; an involution with trace zero has the eigenvalues $\pm1$, so its determinant is $-1$.

The involution separates the two eigenvalues without diagonalising. The matrix is recovered as $X=\tfrac12(\operatorname{tr}X)I+\tfrac12\sqrt{\Delta}\,\zeta$, which is the spectral decomposition written in the basis of the identity and the involution.

Proof. Substitute $\rho_{\pm}=\tfrac12(I\pm\zeta)$ into the reconstruction identity and collect the two terms.

The Coordinates in Trace and Determinant

Proposition. With $t=\operatorname{tr}X$ and $\delta=\det X$ the eigenvalues are the roots of $z^{2}-tz+\delta$, namely $\lambda_{\pm}=\tfrac12\bigl(t\pm\sqrt{t^{2}-4\delta}\bigr)$.

Proof. The characteristic polynomial is $z^{2}-tz+\delta$ by The Elementary Symmetric Functions: $\operatorname{tr}$ and $\det$ in Dimension Two, and the quadratic formula gives the two roots.

Corollary. The idempotents are

$$ \rho_{\pm}=\frac{X-\lambda_{\mp}I}{\lambda_{+}-\lambda_{-}}=\pm\frac{2X-tI\mp\sqrt{\Delta}\,I}{2\sqrt{\Delta}}. $$

Proof. Substitute the roots into the definition and clear the denominators.

Corollary. The idempotents are rational functions of $X$ with the discriminant in the denominator, so they are defined exactly where the spectrum is simple.

Proof. The square root appears only through $\sqrt{\Delta}$, which divides, and the denominator vanishes exactly when the two eigenvalues coincide.

Simple Spectrum and the Discriminant

Theorem. The spectrum is simple exactly when the discriminant $\Delta=t^{2}-4\delta$ is nonzero, and then the two idempotents exist and are distinct.

Proof. The eigenvalues coincide exactly when the square root vanishes, which is the vanishing of the discriminant; and the two idempotents differ by the involution, which is nonzero when the eigenvalues differ.

Corollary. At $\Delta=0$ the two formulas degenerate, and the pair is replaced by the Jordan regime of The Minimal Polynomial and the Two Jordan Regimes of $M_2(\mathbb{C})$.

Proof. The denominator of both idempotents is $\sqrt{\Delta}$, which vanishes at the double eigenvalue, while the matrix still has a single eigenvalue and possibly a nontrivial Jordan block.

Corollary. The discriminant is the square of the difference of the eigenvalues, so it is the invariant that measures the distance from the diagonalisable case.

Proof. $(\lambda_{+}-\lambda_{-})^{2}=(\lambda_{+}+\lambda_{-})^{2}-4\lambda_{+}\lambda_{-}=t^{2}-4\delta$.

The Spectral Projectors

Proposition. The idempotents are the residues of the resolvent: $\rho_{\pm}=\dfrac{1}{2\pi i}\oint_{\gamma_{\pm}}(zI-X)^{-1}\,dz$ around a small circle enclosing $\lambda_{\pm}$ alone.

Proof. The spectral decomposition writes $zI-X=(z-\lambda_{+})\rho_{+}+(z-\lambda_{-})\rho_{-}$ with orthogonal idempotents, so the inverse is $\dfrac{\rho_{+}}{z-\lambda_{+}}+\dfrac{\rho_{-}}{z-\lambda_{-}}$; the residue at $\lambda_{+}$ is $\rho_{+}$ and the residue at $\lambda_{-}$ is $\rho_{-}$.

Corollary. The inverse is a sum of two simple fractions, so the resolvent has two simple poles and no other singularity.

Corollary. The projection is the derivative of the reconstruction: $\rho_{+}=\dfrac{X-\lambda_{-}I}{\lambda_{+}-\lambda_{-}}$ is the Lagrange polynomial evaluated at the matrix, and the two readings agree.

Proof. Both are polynomials in $X$ taking the values $1$ and $0$ at the two eigenvalues, and such a polynomial of degree one is unique.

Corollary. The trace of $\rho_{+}X$ is $\lambda_{+}$, so the eigenvalue is read as a trace against the projector.

Proof. $\operatorname{tr}(\rho_{+}X)=\lambda_{+}\operatorname{tr}(\rho_{+})=\lambda_{+}$, since $\operatorname{tr}\rho_{+}=1$.

The Corpus Translation

Proposition. Under the matrix model the pair $\rho_{\pm}$ is the pair $\tilde\Pi_{\pm}$, and the involution $\zeta$ is the root of one $\tilde{Ζ}$.

Proof. The corpus defines $\tilde\Pi_{\pm}=\tfrac12(e_t\pm\tilde{Ζ})$ with $\tilde{Ζ}^{2}=e_t$, which is the pair of identities proved above for $\rho_{\pm}$ and $\zeta$.

Corollary. The reconstruction $X=\tfrac12(\operatorname{tr}X)I+\tfrac12\sqrt{\Delta}\,\zeta$ is the corpus's polar form of the element in its root and its modulus.

Proof. The coefficient of the identity is the scalar part and the coefficient of the involution is the modulus, both read on the matrix side.

Summary

A two-by-two matrix with two distinct eigenvalues has the Lagrange idempotents $\rho_{\pm}=\ell_{\pm}(X)$, the unique polynomials of degree one in $X$ taking the values $1$ and $0$ at the two eigenvalues. They satisfy $\rho_{+}+\rho_{-}=I$, $\rho_{+}\rho_{-}=0$, $\rho_{\pm}^{2}=\rho_{\pm}$ and $X=\lambda_{+}\rho_{+}+\lambda_{-}\rho_{-}$, each has rank one and trace one, and their difference is the involution $\zeta=(2X-tI)/\sqrt{\Delta}$ with $\rho_{\pm}=\tfrac12(I\pm\zeta)$.

The pair exists exactly where the discriminant $\Delta=t^{2}-4\delta$ is nonzero, and the same objects are the spectral projectors of the resolvent, the residues at the two poles. On the biquaternion side the pair is $\tilde\Pi_{\pm}$ and the involution is the root of one $\tilde{Ζ}$.

Summary of Notation

Symbol Meaning
$\lambda_{\pm}$ the two eigenvalues
$\rho_{\pm}$ the Lagrange idempotents
$\zeta$ $\rho_{+}-\rho_{-}$, the involution
$t,\delta$ the trace and the determinant
$\Delta$ $t^{2}-4\delta$, the discriminant
$\ell_{\pm}$ the interpolating polynomials of degree one

Further Reading