Two-Sided Operators on the N-Degenerate Sector

Introduction

This article treats the two-sided operator, the Hermitian pair $\mathrm{Sd}_{\tilde Q}$, on the null cone $N=\langle\tilde Q\rangle_{\natural}=0$, the closed set of real codimension two. It is the two-sided companion of One-Sided Operators on the N-Degenerate Sector, the same polar pair read on two sides.

The two-sided operator is the Hermitian congruence

$$ \mathrm{Sd}_{\tilde Q}(\tilde X)=\tilde Q\tilde X\tilde Q^{*}. $$

What the norm decides, and what the vector form decides. The norm decides the collapse: on the whole null cone the operator has rank one, its image is a single null line, and its kernel has dimension three. Nothing about that changes inside the sector. The second condition $B^{2}=Q_x^{2}+Q_y^{2}+Q_z^{2}$ decides the spectrum and the Jordan type: on $\Sigma_{01}$ the operator has the eigenvalue $\lvert c\rvert^{2}=4a^{2}\cos^{2}\theta$ and is a projection up to scale, and on $\Sigma_{00}$ the operator is square-zero, with the single eigenvalue zero, one Jordan block of size two and two further zero directions.

The polar pair of The Polar Decomposition of an N-Degenerate Biquaternion survives on both strata,

$$ \tilde Q=\tilde Q_H\tilde Q_0, \qquad \tilde Q_H=\bigl(\tilde Q\tilde Q^{*}\bigr)^{1/2}, \qquad \tilde Q_0=\tilde Q_H^{+}\tilde Q, $$

with $\tilde Q_H$ Hermitian positive semidefinite of rank one and $\tilde Q_0$ a partial isometry. The main result is the scalar collapse of the operator,

$$ \mathrm{Sd}_{\tilde Q}=4a^{2}\,\mathrm{Sd}_{\tilde Q_0}, $$

with $4a^{2}=(2a)^{2}$ the square of the one-sided singular value. On $\Sigma_{00}$ the partial isometry is itself nilpotent, $\tilde Q_0^{2}=0$, so the operator is $4a^{2}$ times a nilpotent partial isometry. On $\Sigma_{01}$ the partial isometry is a scalar times an idempotent, and the operator is $4a^{2}$ times a projection.

The plan is the following. §The Polar Pair describes the polar pair on both strata. §The Two-Sided Operator computes the composition, the closed form, the image, the eigenvalues and the Jordan data. §The Polar Decomposition of the Two-Sided Operator proves the scalar collapse and reads the phase. §The Two Slots reads the operator through the Peirce corners and through the singular vectors. §The Exponential computes the one-parameter group. §Worked Examples treats a Hermitian idempotent, a non-Hermitian idempotent, the standard nilpotent and a complex scalar multiple.

The Polar Pair

On the null cone the spectral decomposition fails and the polar pair survives. The two strata differ in what the partial isometry is: a scalar times an idempotent on $\Sigma_{01}$, a nilpotent on $\Sigma_{00}$. The positive factor is a rank-one Hermitian projection scaled by $2a$ in both cases.

The conventions, including the two forms, are those of the companion article.

The matrix model $\mathsf M_2$ sends $e_t$ to $I_2$ and $\vec e_k$ to $-i\sigma_k$; it is multiplicative, and

$$ \det\mathsf M_2(\tilde X)=\langle\tilde X\rangle_{\natural}, \qquad \tfrac12\operatorname{tr}\bigl(\mathsf M_2(\tilde X)\mathsf M_2(\tilde X)^{\dagger}\bigr)=\langle\tilde X\rangle_{*}. $$

The non-zero null cone is the set of rank-one matrices. For $\tilde Q\neq0$ on $\mathcal N$ there are unit vectors $u,v\in\mathbb{C}^{2}$ and $a>0$ with

$$ \mathsf M_2(\tilde Q)=2a\,u v^{\dagger}, \qquad 2a=\sqrt{2\langle\tilde Q\rangle_{*}}, \qquad v^{\dagger}u=\frac{Q_t}{a}=\frac{c}{2a}. $$

The two factors are the left and right slots of the rank-one matrix. The overlap $v^{\dagger}u$ is non-zero exactly on $\Sigma_{01}$; on $\Sigma_{00}$ the slots are orthogonal. The polar pair is

$$ P=u u^{\dagger}=\frac{\tilde Q\tilde Q^{*}}{4a^{2}}, \qquad \tilde Q_H=2aP, \qquad \tilde Q_0=u v^{\dagger}=\frac{\tilde Q}{2a}, $$

with $P$ a rank-one Hermitian idempotent and $\tilde Q_0$ a partial isometry. In the closed form of the sector,

$$ a=\frac{\lvert c\rvert\sqrt{\kappa}}{2}=\lvert Q_t\rvert\sqrt{\kappa}, \qquad P=\frac{\tilde\Pi\tilde\Pi^{*}}{\kappa}, \qquad \tilde Q_0=\frac{c}{\lvert c\rvert}\frac{\tilde\Pi}{\sqrt{\kappa}}, \qquad \kappa=\frac{1+\lVert\tilde{Ζ}\rVert^{2}}{2}, $$

where $\tilde{Ζ}=2\tilde\Pi-e_t$ and $\kappa\ge1$. The overlap is $v^{\dagger}u=Q_t/a=c/(2a)$, with modulus $1/\sqrt{\kappa}=\cos\theta$, where $\theta$ is the angle between the two slots.

The two-sided operator of the pair is computed in the matrix model. For any $\tilde X$,

$$ \mathsf M_2\bigl(\mathrm{Sd}_{\tilde Q}(\tilde X)\bigr) =(2a)^{2}u v^{\dagger}\,\mathsf M_2(\tilde X)\,v u^{\dagger} =4a^{2}\bigl(v^{\dagger}\mathsf M_2(\tilde X)v\bigr)u u^{\dagger}. $$

Thus the image is the line $\mathbb{C}u u^{\dagger}=\mathbb{C}P$ and the value is the scalar $v^{\dagger}\mathsf M_2(\tilde X)v$. This formula is the source of every result below.

The canonical Hermitian pair

Let $\tilde\nu=\vec A+i\vec B$ with $\vec A,\vec B$ real, orthogonal and of equal norm $a>0$. Put

$$ \vec c=\frac{\vec A\times\vec B}{a^{2}}, \qquad \tilde\Pi_+=\frac{e_t+i\vec c}{2}, \qquad \tilde\Pi_-=\frac{e_t-i\vec c}{2}. $$

The vector $\vec c$ is real and unit; the idempotents are Hermitian and complementary. They satisfy

$$ \tilde\Pi_+\tilde\nu=\tilde\nu, \qquad \tilde\nu\tilde\Pi_+=0, \qquad \tilde\Pi_-\tilde\nu=0, \qquad \tilde\nu\tilde\Pi_-=\tilde\nu, $$

so $\tilde\nu=\tilde\Pi_+\tilde\nu\tilde\Pi_-$ lies in the mixed corner $\mathbb{B}_{+-}$. The element is annihilated by the two products that would place it in a diagonal corner. Its norms are

$$ \langle\tilde\nu\rangle_{\natural}=0, \qquad \langle\tilde\nu\rangle_{*}=2a^{2}, \qquad \tilde\nu\tilde\nu^{*}=4a^{2}\tilde\Pi_+, \qquad \tilde\nu^{*}\tilde\nu=4a^{2}\tilde\Pi_- . $$

The polar pair on the nilpotents

The polar pair is

$$ \boxed{\; P=\tilde\Pi_+, \qquad \tilde Q_H=2a\tilde\Pi_+, \qquad \tilde Q_0=\frac{\tilde\nu}{2a}, \qquad \tilde Q=\tilde Q_H\tilde Q_0, \qquad \tilde Q_0^{2}=0 . \;} $$

The Hermitian factor is positive semidefinite of rank one, with spectrum $\{0,2a\}$. The partial-isometry factor is nilpotent:

$$ \tilde Q_0\tilde Q_0^{*}=\tilde\Pi_+, \qquad \tilde Q_0^{*}\tilde Q_0=\tilde\Pi_-, \qquad \tilde Q_0\tilde Q_0^{*}\tilde Q_0=\tilde Q_0, \qquad \tilde Q_0^{2}=0 . $$

The polar pair is the rank-one singular value decomposition of the nilpotent. The "$*$-unitary" factor is a nilpotent partial isometry, not a $*$-unitary; the pair is still the general pair of The Polar Decomposition of an N-Degenerate Biquaternion, but it is as far as possible from the modulus–unit form.

The matrix model

In the matrix model $\mathsf M_2(\tilde\nu)=2a\,u v^{\dagger}$ with $u,v$ unit and orthogonal:

$$ v^{\dagger}u=\frac{Q_t}{a}=0, \qquad u^{\dagger}u=v^{\dagger}v=1, \qquad u\perp v . $$

The polar factors are

$$ P=u u^{\dagger}=\tilde\Pi_+, \qquad \tilde Q_H=2a\,u u^{\dagger}, \qquad \tilde Q_0=u v^{\dagger}, \qquad \tilde Q_0^{*}=v u^{\dagger}, \qquad \tilde Q_0^{2}=u(v^{\dagger}u)v^{\dagger}=0 . $$

The two slots are orthogonal. The overlap $v^{\dagger}u$ vanishes; this is the matrix form of the spectral degeneration. The two-sided operator is computed from the same decomposition:

$$ \mathsf M_2\bigl(\mathrm{Sd}_{\tilde Q}(\tilde X)\bigr) =(2a)^{2}u v^{\dagger}\mathsf M_2(\tilde X)v u^{\dagger} =4a^{2}\bigl(v^{\dagger}\mathsf M_2(\tilde X)v\bigr)u u^{\dagger}. $$

Because $v^{\dagger}u=0$, applying the operator twice gives zero. The single chain is visible in the formula: the scalar $v^{\dagger}\mathsf M_2(\tilde X)v$ is written on the line $u u^{\dagger}$, and the line is annihilated on the next step.

The Two-Sided Operator

The composition and the rank are fixed by the norm; the closed form, the spectrum and the Jordan data are fixed by $B^{2}$. Both strata share the rank one and the single null image line.

Elementary properties on the idempotent multiples

Proposition. Let $\tilde Q,\tilde Q'\in\mathbb{B}$. Then:

(a) $\mathrm{Sd}_{\tilde Q}$ is $\mathbb{C}$-linear;

(b) $\mathrm{Sd}_{\tilde Q\tilde Q'}=\mathrm{Sd}_{\tilde Q}\circ\mathrm{Sd}_{\tilde Q'}$;

(c) $\mathrm{Sd}_{\tilde Q}$ is invertible exactly when $\tilde Q$ is invertible, with inverse $\mathrm{Sd}_{\tilde Q^{-1}}$;

(d) $\mathrm{Sd}_{\tilde Q}$ preserves the real forms $\mathbb{B}_{\Psi}$ and $\mathbb{B}_{\Phi}$;

(e) the adjoint is $\mathrm{Sd}_{\tilde Q}^{*}=\mathrm{Sd}_{\tilde Q^{*}}$;

(f) the norm scales as

$$ \bigl\langle\mathrm{Sd}_{\tilde Q}(\tilde X)\bigr\rangle_{\natural} =\lvert\langle\tilde Q\rangle_{\natural}\rvert^{2}\langle\tilde X\rangle_{\natural}. $$

On $\Sigma_{01}$ the element is a zero divisor, so $\mathrm{Sd}_{\tilde Q}$ is not invertible. The properties are those of Operators of the Polar Word: The Common Framework; the sector changes the values.

The closed form on the idempotent multiples

Let $\tilde Q=c\tilde\Pi$. Since $\tilde\Pi^{2}=\tilde\Pi$,

$$ \mathrm{Sd}_{\tilde Q} =\lvert c\rvert^{2}\mathrm{Sd}_{\tilde\Pi}, \qquad \mathrm{Sd}_{\tilde\Pi}^{2} =\mathrm{Sd}_{\tilde\Pi}, \qquad \mathrm{Sd}_{\tilde Q}^{2} =\lvert c\rvert^{2}\mathrm{Sd}_{\tilde Q}. $$

Thus $\mathrm{Sd}_{\tilde Q}/\lvert c\rvert^{2}$ is an idempotent of the operator space, and $\mathrm{Sd}_{\tilde Q}$ is an idempotent up to scale. It is not an orthogonal projection unless $\tilde\Pi$ is Hermitian.

Theorem (rank, spectrum and Jordan form). On $\Sigma_{01}$:

(a) $\mathrm{Sd}_{\tilde Q}$ has complex rank $1$;

(b) its spectrum is $\{\lvert c\rvert^{2},0\}$, with algebraic multiplicities $1$ and $3$, and geometric multiplicities $1$ and $3$;

(c) it is diagonalizable, with minimal polynomial $X(X-\lvert c\rvert^{2})$;

(d) $\operatorname{tr}\mathrm{Sd}_{\tilde Q}=\lvert c\rvert^{2}=4a^{2}\cos^{2}\theta$ and $\det\mathrm{Sd}_{\tilde Q}=0$;

(e) for every $\tilde X$, $\langle\mathrm{Sd}_{\tilde Q}(\tilde X)\rangle_{\natural}=0$; the image lies on the null cone.

Proof. In the matrix model, the formula above gives

$$ \mathrm{Sd}_{\tilde Q}(\tilde X)=4a^{2}\bigl(v^{\dagger}\mathsf M_2(\tilde X)v\bigr)u u^{\dagger}. $$

The image is contained in the line $\mathbb{C}u u^{\dagger}$, and it is non-zero because $u u^{\dagger}$ is in the image: take $\tilde X$ with $\mathsf M_2(\tilde X)=u u^{\dagger}$; then $v^{\dagger}u u^{\dagger}v=\lvert v^{\dagger}u\rvert^{2}\neq0$. Hence the rank is one. The square is

$$ \mathrm{Sd}_{\tilde Q}^{2} =(4a^{2})^{2}\lvert v^{\dagger}u\rvert^{2} \bigl(v^{\dagger}\mathsf M_2(\tilde X)v\bigr)u u^{\dagger} =\lvert c\rvert^{2}\mathrm{Sd}_{\tilde Q}, $$

because $c=2a\,v^{\dagger}u$. The operator is a non-zero scalar multiple of an idempotent, so it is diagonalizable with eigenvalues $\lvert c\rvert^{2}$ and $0$. The trace is $\lvert c\rvert^{2}$ and the determinant is zero. The norm is $\langle\mathrm{Sd}_{\tilde Q}(\tilde X)\rangle_{\natural}=\lvert\langle\tilde Q\rangle_{\natural}\rvert^{2}\langle\tilde X\rangle_{\natural}=0$. The image is a null line because $\langle\tilde Q\tilde X\tilde Q^{*}\rangle_{\natural}=\lvert\langle\tilde Q\rangle_{\natural}\rvert^{2}\langle\tilde X\rangle_{\natural}=0$.

The image and the kernel

The image is the line

$$ \operatorname{im}\mathrm{Sd}_{\tilde Q}=\mathbb{C}P, \qquad P=\frac{\tilde Q\tilde Q^{*}}{4a^{2}}=\frac{\tilde\Pi\tilde\Pi^{*}}{\kappa}. $$

The line is null: $\langle P\rangle_{\natural}=0$ because $P$ is a non-trivial idempotent. The kernel is the hyperplane

$$ \ker\mathrm{Sd}_{\tilde Q} =\bigl\{\tilde X:\tilde\Pi\tilde X\tilde\Pi^{*}=0\bigr\}, \qquad \dim_{\mathbb{C}}\ker\mathrm{Sd}_{\tilde Q}=3. $$

In the matrix model it is $\{X:v^{\dagger}Xv=0\}$, the hyperplane orthogonal to the rank-one functional $X\mapsto v^{\dagger}Xv$. The kernel contains the right annihilator $A(\tilde Q)=\ker L_{\tilde Q}$ and the left annihilator of $\tilde Q^{*}$, because $\tilde Q\tilde X=0$ or $\tilde X\tilde Q^{*}=0$ implies $\tilde Q\tilde X\tilde Q^{*}=0$, and it is strictly larger than each of them.

When $\tilde\Pi$ is Hermitian, the Peirce form is diagonal in the eigenbasis $(\tilde\Pi,\vec\Upsilon_-,\vec\Upsilon_+,\tilde\Pi')$:

$$ \boxed{\; \mathsf M\bigl(\mathrm{Sd}_{\tilde Q}\bigr) =\operatorname{diag}\bigl(\lvert c\rvert^{2},0,0,0\bigr). \;} $$

Only the $(+,+)$ corner survives; the kernel is the sum of the other three corners. When $\tilde\Pi$ is not Hermitian, the matrix is not diagonal in that basis, but the spectrum and the rank are the same. The image is still the line $\mathbb{C}P$, and the kernel is still the hyperplane above.

Eigenvalues and singular values

The eigenvalue of $\mathrm{Sd}_{\tilde Q}$ on its image is $\lvert c\rvert^{2}$. Its non-zero singular value is $4a^{2}$: indeed

$$ \mathrm{Sd}_{\tilde Q}\mathrm{Sd}_{\tilde Q}^{*} =\mathrm{Sd}_{\tilde Q\tilde Q^{*}} =\mathrm{Sd}_{4a^{2}P} =(4a^{2})^{2}\mathrm{Sd}_P, $$

and $\mathrm{Sd}_P$ is an orthogonal projection of rank one, with singular value one. Hence the singular value is $4a^{2}$. The ratio is

$$ \frac{\lvert c\rvert^{2}}{4a^{2}} =\lvert v^{\dagger}u\rvert^{2} =\cos^{2}\theta . $$

This ratio is one exactly when $\tilde\Pi$ is Hermitian ($\kappa=1$, the slots coincide), and it lies strictly between $0$ and $1$ for a non-Hermitian idempotent. It vanishes at the wall $\Sigma_{00}$. The distinction between eigenvalue and singular value is the measure of the obliqueness of the projection.

The idempotent $\mathrm{Sd}_{\tilde\Pi}$ itself has singular value $\kappa$, because

$$ \mathrm{Sd}_{\tilde\Pi}\mathrm{Sd}_{\tilde\Pi}^{*} =\mathrm{Sd}_{\tilde\Pi\tilde\Pi^{*}} =\mathrm{Sd}_{\kappa P} =\kappa^{2}\mathrm{Sd}_P . $$

Thus $\mathrm{Sd}_{\tilde Q}=\lvert c\rvert^{2}\mathrm{Sd}_{\tilde\Pi}$ has singular value $\lvert c\rvert^{2}\kappa=4a^{2}$. The operator $\mathrm{Sd}_{\tilde Q}$ is not a partial isometry in general: its singular value is $4a^{2}$, not one. It is $4a^{2}$ times the partial isometry $\mathrm{Sd}_{\tilde Q_0}$.

The square, the rank and the spectrum

Theorem. On $\Sigma_{00}$:

(a) $\mathrm{Sd}_{\tilde Q}^{2}=0$;

(b) $\mathrm{Sd}_{\tilde Q}$ has complex rank $1$;

(c) its only eigenvalue is $0$, with algebraic multiplicity $4$ and geometric multiplicity $3$;

(d) its Jordan form is $J_2(0)\oplus0\oplus0$, with minimal polynomial $X^{2}$;

(e) $\operatorname{tr}\mathrm{Sd}_{\tilde Q}=0$ and $\det\mathrm{Sd}_{\tilde Q}=0$;

(f) for every $\tilde X$, $\langle\mathrm{Sd}_{\tilde Q}(\tilde X)\rangle_{\natural}=0$; the image lies on the null cone.

Proof. In the matrix model,

$$ \mathrm{Sd}_{\tilde Q}(\tilde X) =4a^{2}\bigl(v^{\dagger}\mathsf M_2(\tilde X)v\bigr)u u^{\dagger}. $$

The image is contained in the line $\mathbb{C}u u^{\dagger}$, and it is non-zero because $\mathrm{Sd}_{\tilde Q}(e_t)=\tilde\nu\tilde\nu^{*}=4a^{2}\tilde\Pi_+$. Hence the rank is one. Applying the formula twice,

$$ \mathrm{Sd}_{\tilde Q}^{2}(\tilde X) =(4a^{2})^{2}\bigl(v^{\dagger}\mathsf M_2(\tilde X)v\bigr) (v^{\dagger}u)(u^{\dagger}v)u u^{\dagger}=0, $$

because $v^{\dagger}u=0$. The only eigenvalue of a square-zero operator is zero; the algebraic multiplicity is four, the geometric multiplicity is the dimension of the kernel, three; the Jordan form is one block of size two and two blocks of size one. The trace is zero and the determinant is zero. The norm is $\langle\mathrm{Sd}_{\tilde Q}(\tilde X)\rangle_{\natural}=\lvert\langle\tilde Q\rangle_{\natural}\rvert^{2}\langle\tilde X\rangle_{\natural}=0$. The image is a null line because $\langle\tilde Q\tilde X\tilde Q^{*}\rangle_{\natural}=\lvert\langle\tilde Q\rangle_{\natural}\rvert^{2}\langle\tilde X\rangle_{\natural}=0$.

The image and the kernel on the nilpotents

The image is the null line

$$ \operatorname{im}\mathrm{Sd}_{\tilde Q} =\mathbb{C}\tilde\Pi_+, \qquad \tilde\Pi_+=\frac{\tilde Q\tilde Q^{*}}{4a^{2}}. $$

The kernel is the hyperplane

$$ \ker\mathrm{Sd}_{\tilde Q} =\bigl\{\tilde X:\tilde\nu\tilde X\tilde\nu^{*}=0\bigr\}, \qquad \dim_{\mathbb{C}}\ker\mathrm{Sd}_{\tilde Q}=3. $$

In the matrix model it is $\{X:v^{\dagger}Xv=0\}$. The kernel contains the right annihilator $A(\tilde\nu)=\ker L_{\tilde\nu}$ and the left annihilator of $\tilde\nu^{*}$, because $\tilde\nu\tilde X=0$ or $\tilde X\tilde\nu^{*}=0$ implies $\tilde\nu\tilde X\tilde\nu^{*}=0$. The image is contained in the kernel, since the operator is square-zero; the kernel is strictly larger, because the rank is one.

The explicit Jordan chain

The canonical Hermitian pair gives a very explicit chain. Since $\tilde\nu\tilde\Pi_-=\tilde\nu$ and $\tilde\nu\tilde\nu^{*}=4a^{2}\tilde\Pi_+$, one has

$$ \boxed{\; \mathrm{Sd}_{\tilde Q}(\tilde\Pi_-) =\tilde\nu\tilde\Pi_-\tilde\nu^{*} =\tilde\nu\tilde\nu^{*} =4a^{2}\tilde\Pi_+, \qquad \mathrm{Sd}_{\tilde Q}(\tilde\Pi_+)=0 . \;} $$

Thus

$$ \tilde\Pi_-\longmapsto 4a^{2}\tilde\Pi_+\longmapsto0 $$

is a chain of length two. The other two basis vectors are killed:

$$ \mathrm{Sd}_{\tilde Q}(\tilde\nu)=0, \qquad \mathrm{Sd}_{\tilde Q}(\vec X)=0, $$

where $\vec X$ is the corner vector spanning $\tilde\Pi_-\mathbb{B}\tilde\Pi_+$. In the basis

$$ \bigl(\tilde\Pi_+,\ \tilde\nu,\ \vec X,\ \tilde\Pi_-\bigr) $$

the two-sided operator is therefore the single nilpotent block

$$ \boxed{\; \mathsf M\bigl(\mathrm{Sd}_{\tilde Q}\bigr)= \begin{pmatrix} 0&0&0&4a^{2}\\ 0&0&0&0\\ 0&0&0&0\\ 0&0&0&0 \end{pmatrix}. \;} $$

This matrix is the cleanest form of the operator: rank one, square zero, image $\mathbb{C}\tilde\Pi_+$, kernel $\mathbb{C}\tilde\Pi_+\oplus\mathbb{C}\tilde\nu\oplus\mathbb{C}\vec X$, and the single chain from $\tilde\Pi_-$ to $\tilde\Pi_+$.

The chain crosses the Peirce diagram from the $(--)$ corner to the $(+,+)$ corner. The element $\tilde\nu$ lies in the mixed corner $(+-)$; the other mixed corner $(-+)$ is represented by $\vec X$. The two-sided maps the $(--)$ corner to the $(+,+)$ corner and kills the two mixed corners and the $(+,+)$ corner itself. This is the exact two-step collapse of the spectral degeneracy.

Eigenvalues and singular values on the nilpotents

The only eigenvalue is zero. The non-zero singular value is $4a^{2}$: indeed

$$ \mathrm{Sd}_{\tilde Q}\mathrm{Sd}_{\tilde Q}^{*} =\mathrm{Sd}_{\tilde Q\tilde Q^{*}} =\mathrm{Sd}_{4a^{2}\tilde\Pi_+} =(4a^{2})^{2}\mathrm{Sd}_{\tilde\Pi_+}, $$

and $\mathrm{Sd}_{\tilde\Pi_+}$ is an orthogonal projection of rank one, with singular value one. Hence the singular value is $4a^{2}$. The eigenvalue-to-singular-value ratio is zero. This is the maximal defect: the operator has a non-zero singular value but no non-zero eigenvalue, and its image is a null line.

The operator $\mathrm{Sd}_{\tilde Q}=4a^{2}\mathrm{Sd}_{\tilde Q_0}$ is not a partial isometry in general: its singular value is $4a^{2}$, not one. It is $4a^{2}$ times the nilpotent partial isometry $\mathrm{Sd}_{\tilde Q_0}$.

The Polar Decomposition of the Two-Sided Operator

The polar decomposition of the operator is the congruence of the polar decomposition of the element. The scalar $4a^{2}$ is the square of the element's singular value $2a$; the phase of the element never appears.

The scalar collapse on the idempotent multiples

By the polar pair theorem, $\tilde Q=2a\tilde Q_0$ and $\tilde Q_0$ is a partial isometry. The multiplicativity of $\Theta$ gives

$$ \mathrm{Sd}_{\tilde Q} =\mathrm{Sd}_{2a\tilde Q_0} =4a^{2}\mathrm{Sd}_{\tilde Q_0}. $$

This is the scalar collapse. The canonical left polar factor of $\mathrm{Sd}_{\tilde Q}$ is

$$ \lvert\mathrm{Sd}_{\tilde Q}\rvert =\bigl(\mathrm{Sd}_{\tilde Q}\mathrm{Sd}_{\tilde Q}^{*}\bigr)^{1/2} =\mathrm{Sd}_{\tilde Q_H} =4a^{2}\mathrm{Sd}_P, $$

which is positive semidefinite of rank one. On the image of $\mathrm{Sd}_{\tilde Q_0}$ it acts as the scalar $4a^{2}$, because $P\tilde Q_0=\tilde Q_0$ and therefore $\mathrm{Sd}_P\mathrm{Sd}_{\tilde Q_0}=\mathrm{Sd}_{\tilde Q_0}$. Hence

$$ \boxed{\; \mathrm{Sd}_{\tilde Q} =\mathrm{Sd}_{\tilde Q_H}\circ\mathrm{Sd}_{\tilde Q_0} =4a^{2}\mathrm{Sd}_{\tilde Q_0}, \qquad \mathrm{Sd}_{\tilde Q_0}\mathrm{Sd}_{\tilde Q_0}^{*}\mathrm{Sd}_{\tilde Q_0} =\mathrm{Sd}_{\tilde Q_0}. \;} $$

The two-sided operator is $4a^{2}$ times a partial isometry. Its non-zero singular value is $4a^{2}$, the square of the one-sided singular value $2a$. Its eigenvalue is $\lvert c\rvert^{2}=4a^{2}\cos^{2}\theta$, the squared overlap of the two slots.

The phase on the idempotent multiples

The partial-isometry factor is $\tilde Q_0=(c/\lvert c\rvert)\tilde\Pi/\sqrt{\kappa}$. Since $\Theta$ is quadratic in $\tilde Q_0$ and $\tilde Q_0^{*}$,

$$ \mathrm{Sd}_{\tilde Q_0} =\mathrm{Sd}_{\tilde\Pi}/\kappa, $$

independent of the phase of $c$. The two-sided operator therefore cancels the phase:

$$ \mathrm{Sd}_{\tilde Q} =\lvert c\rvert^{2}\mathrm{Sd}_{\tilde\Pi} =4a^{2}\mathrm{Sd}_{\tilde Q_0}. $$

The one-sided operator of the companion article keeps the phase in its eigenvalue $c$; the two-sided operator keeps only the modulus $\lvert c\rvert^{2}$ and the Hermitian projection $P$. When $\tilde\Pi$ is Hermitian and $c=\varrho e^{i\theta_{\mathrm{pha}}}$, the polar pair is $\tilde Q_H=r\tilde\Pi$ and $\tilde Q_0=e^{i\theta_{\mathrm{pha}}}\tilde\Pi$, and the two-sided operator is $r^{2}\mathrm{Sd}_{\tilde\Pi}$: the phase has cancelled between $\tilde Q_0$ and $\tilde Q_0^{*}$.

The polar pair of the element and of the operator on the idempotent multiples

The element pair is $(\tilde Q_H,\tilde Q_0)$, with $\tilde Q_H=2aP$ and $\tilde Q_0$ a partial isometry. The operator pair is $(\mathrm{Sd}_{\tilde Q_H},\mathrm{Sd}_{\tilde Q_0})$, with $\mathrm{Sd}_{\tilde Q_H}=4a^{2}\mathrm{Sd}_P$ positive and $\mathrm{Sd}_{\tilde Q_0}$ a partial isometry. The element pair is a rank-one singular value decomposition; the operator pair is its congruence. The scalar $4a^{2}$ is the square of the element's singular value $2a$, and the phase of the element is absent from the operator pair.

The scalar collapse on the nilpotents

By the polar pair theorem, $\tilde Q=2a\tilde Q_0$ with $\tilde Q_0$ a nilpotent partial isometry. The multiplicativity of $\Theta$ gives

$$ \mathrm{Sd}_{\tilde Q} =\mathrm{Sd}_{2a\tilde Q_0} =4a^{2}\mathrm{Sd}_{\tilde Q_0}. $$

This is the scalar collapse. The canonical left polar factor is

$$ \lvert\mathrm{Sd}_{\tilde Q}\rvert =\bigl(\mathrm{Sd}_{\tilde Q}\mathrm{Sd}_{\tilde Q}^{*}\bigr)^{1/2} =\mathrm{Sd}_{\tilde Q_H} =4a^{2}\mathrm{Sd}_{\tilde\Pi_+}, $$

which is positive semidefinite of rank one. On the image of $\mathrm{Sd}_{\tilde Q_0}$ it acts as the scalar $4a^{2}$, because $\tilde\Pi_+\tilde Q_0=\tilde Q_0$ and therefore $\mathrm{Sd}_{\tilde\Pi_+}\mathrm{Sd}_{\tilde Q_0}=\mathrm{Sd}_{\tilde Q_0}$. Hence

$$ \boxed{\; \mathrm{Sd}_{\tilde Q} =\mathrm{Sd}_{\tilde Q_H}\circ\mathrm{Sd}_{\tilde Q_0} =4a^{2}\mathrm{Sd}_{\tilde Q_0}, \qquad \mathrm{Sd}_{\tilde Q_0}\mathrm{Sd}_{\tilde Q_0}^{*}\mathrm{Sd}_{\tilde Q_0} =\mathrm{Sd}_{\tilde Q_0}, \qquad \mathrm{Sd}_{\tilde Q_0}^{2}=0 . \;} $$

The two-sided operator is $4a^{2}$ times a nilpotent partial isometry. Its non-zero singular value is $4a^{2}=(2a)^{2}$; its only eigenvalue is zero. The polar decomposition of the operator is therefore a scalar times a nilpotent partial isometry, the extreme case of the collapse.

The phase on the nilpotents

For a complex multiple

$$ \tilde\nu=c_0(\vec e_x+i\vec e_y), \qquad c_0=\lvert c_0\rvert e^{i\beta}, $$

one has $a=\lvert c_0\rvert$ and

$$ \tilde Q_0=e^{i\beta}\frac{\vec e_x+i\vec e_y}{2}, \qquad \mathrm{Sd}_{\tilde Q_0} =\mathrm{Sd}_{(\vec e_x+i\vec e_y)/2} =\frac14\mathrm{Sd}_{\vec e_x+i\vec e_y}, $$

independent of the phase. The two-sided operator is

$$ \mathrm{Sd}_{\tilde Q} =4\lvert c_0\rvert^{2}\mathrm{Sd}_{\tilde Q_0} =\lvert c_0\rvert^{2}\mathrm{Sd}_{\vec e_x+i\vec e_y} =\mathrm{Sd}_{\tilde\nu}, $$

so the phase has cancelled. The one-sided operator keeps the phase; the two-sided operator sees only $\lvert c_0\rvert^{2}$ and the canonical pair. The sign $\tilde\nu\mapsto-\tilde\nu$ also cancels.

The polar pair of the element and of the operator on the nilpotents

The element pair is $(\tilde Q_H,\tilde Q_0)$, with $\tilde Q_H=2a\tilde\Pi_+$ positive of rank one and $\tilde Q_0$ a nilpotent partial isometry. The operator pair is $(\mathrm{Sd}_{\tilde Q_H},\mathrm{Sd}_{\tilde Q_0})$, with $\mathrm{Sd}_{\tilde Q_H}=4a^{2}\mathrm{Sd}_{\tilde\Pi_+}$ positive of rank one and $\mathrm{Sd}_{\tilde Q_0}$ a nilpotent partial isometry. The element pair is the rank-one singular value decomposition of the nilpotent; the operator pair is its congruence. The scalar $4a^{2}$ is the square of the element's singular value $2a$.

The spectral pair of the element does not exist off the origin: the element has only the eigenvalue zero, and no non-trivial idempotent commutes with it. The Hermitian form supplies the canonical pair $(\tilde\Pi_+,\tilde\Pi_-)$, and the polar decomposition uses exactly that pair. This is the precise sense in which the polar pair survives while the spectral decomposition fails.

The Two Slots

The two slots of the two-sided operator survive the norm degeneration as the two singular vectors. On $\Sigma_{01}$ they are the column and the row of the rank-one matrix model, with overlap $\cos\theta=1/\sqrt{\kappa}$; on $\Sigma_{00}$ the overlap vanishes and the operator is square-zero.

The Peirce picture on the idempotent multiples

On $\Sigma_{01}$ the eigenbasis exists because $B^{2}\neq0$. Its spectral projectors are $\tilde\Pi$ and $\tilde\Pi'$, and the four Peirce corners are one-dimensional. For a Hermitian $\tilde\Pi$, the two-sided operator is the projection onto the $(+,+)$ corner:

$$ \mathrm{Sd}_{\tilde\Pi}(\tilde\Pi)=\tilde\Pi, \qquad \mathrm{Sd}_{\tilde\Pi}(\vec\Upsilon_-)=0, \qquad \mathrm{Sd}_{\tilde\Pi}(\vec\Upsilon_+)=0, \qquad \mathrm{Sd}_{\tilde\Pi}(\tilde\Pi')=0. $$

Thus the matrix is $\operatorname{diag}(1,0,0,0)$, and $\mathrm{Sd}_{\tilde Q}$ is $\operatorname{diag}(\lvert c\rvert^{2},0,0,0)$. The kernel is the sum of the three other corners. The two-sided operator pairs the two half-angles of the regular theory; on $\Sigma_{01}$ one half has eigenvalue zero, so only the $(+,+)$ pairing survives. The four eigenvalue products are

$$ c\bar c,\quad c\cdot0,\quad 0\cdot\bar c,\quad 0\cdot0, $$

that is, $\lvert c\rvert^{2},0,0,0$. The full rotor of the regular two-sided operator has no angle or rapidity here: the element is not invertible, and the two-sided operator is not a conformal Lorentz transformation. It is an oblique projection onto a null line.

For a non-Hermitian $\tilde\Pi$ the eigenbasis still exists, but the two-sided operator is not diagonal in it. The invariant statement is the spectrum: the four products are still $\lvert c\rvert^{2},0,0,0$, and the image is still the null line $\mathbb{C}P$, with $P=\tilde\Pi\tilde\Pi^{*}/\kappa\neq\tilde\Pi$.

The singular-vector picture on the idempotent multiples

In the matrix model the two slots are the singular vectors $u$ and $v$. The two-sided operator is

$$ \mathrm{Sd}_{\tilde Q}(\tilde X) =4a^{2}\bigl(v^{\dagger}\mathsf M_2(\tilde X)v\bigr)u u^{\dagger}. $$

It reads the $v$-slot of $\tilde X$ through the scalar $v^{\dagger}\mathsf M_2(\tilde X)v$ and writes it on the $u$-slot. The relative phase of $u$ and $v$ cancels, because the two-sided operator uses $u$ and $u^{\dagger}$ on the outside and $v$ and $v^{\dagger}$ on the inside. The eigenvalue is $4a^{2}\lvert v^{\dagger}u\rvert^{2}=\lvert c\rvert^{2}$; the singular value is $4a^{2}$. The two slots are not orthogonal on $\Sigma_{01}$: their overlap has modulus $\cos\theta=1/\sqrt{\kappa}$. At the wall $\Sigma_{00}$ the overlap vanishes, the slots are orthogonal, and the two-sided operator becomes nilpotent.

This is the precise form of the two-slot structure of the regular two-sided operator. In the regular sector the two slots are the two chiralities, and the two-sided operator carries the full angle and rapidity as the product of the two half-factors. On the null cone the two slots are the singular vectors of the rank-one element, and the two-sided operator carries only the squared overlap $\cos^{2}\theta$. The angle between the slots replaces the angle of the Lorentz transformation.

The Peirce picture on the nilpotents

In the canonical basis $(\tilde\Pi_+,\tilde\nu,\vec X,\tilde\Pi_-)$ the two-sided operator is the single block of §The Explicit Jordan Chain:

$$ \mathsf M\bigl(\mathrm{Sd}_{\tilde Q}\bigr)= \begin{pmatrix} 0&0&0&4a^{2}\\ 0&0&0&0\\ 0&0&0&0\\ 0&0&0&0 \end{pmatrix}. $$

The chain is $\tilde\Pi_-\to4a^{2}\tilde\Pi_+\to0$. The image is the $(+,+)$ corner, the chain starts in the $(--)$ corner, and the two mixed corners are killed. In the regular sector the two-sided operator is a conformal Lorentz transformation with four eigenvalue products $\lambda_i\bar\lambda_j$; here all four products vanish. In the N-regular, spectrally degenerate sector the two-sided operator is a unipotent operator of Jordan type $3+1$; here it is a nilpotent operator of Jordan type $2+1+1$. The two chiral slots of the regular theory coalesce into the single chain $\tilde\Pi_-\to\tilde\Pi_+$.

The singular-vector picture on the nilpotents

In the matrix model the two slots are the singular vectors $u$ and $v$, and they are orthogonal:

$$ v^{\dagger}u=0, \qquad \mathrm{Sd}_{\tilde Q}(\tilde X) =4a^{2}\bigl(v^{\dagger}\mathsf M_2(\tilde X)v\bigr)u u^{\dagger}. $$

The two-sided operator reads the $v$-slot of $\tilde X$ and writes it on the $u$-slot. Because $u\perp v$, the line on which it writes is annihilated on the next step. The relative phase of $u$ and $v$ cancels, and the overlap is zero. The angle $\theta$ between the slots is $\pi/2$; the eigenvalue-to-singular-value ratio is $\cos^{2}\theta=0$. The two-sided operator sees only the orthogonality.

This is the extreme case of the two-slot structure. On $\Sigma_{01}$ the slots had overlap $\cos\theta=1/\sqrt{\kappa}\in(0,1]$, and the two-sided operator was a projection up to scale. Here the overlap is zero, and the two-sided operator is nilpotent. The spectral wall is the orthogonality of the two slots.

The Exponential

On $\Sigma_{01}$ the exponential is a line of projections; on $\Sigma_{00}$ it is unipotent. In both cases it removes the norm degeneration, and on $\Sigma_{01}$ it also removes the spectral one.

The exponential on the idempotent multiples

The operator exponential. Since $\mathrm{Sd}_{\tilde Q}=\lvert c\rvert^{2}\mathrm{Sd}_{\tilde\Pi}$ and $\mathrm{Sd}_{\tilde\Pi}^{2}=\mathrm{Sd}_{\tilde\Pi}$,

$$ \boxed{\; \exp\bigl(t\mathrm{Sd}_{\tilde Q}\bigr) =I+\bigl(e^{t\lvert c\rvert^{2}}-1\bigr)\mathrm{Sd}_{\tilde\Pi} \;} $$

for every complex $t$. The eigenvalues are $e^{t\lvert c\rvert^{2}}$ and $1$ (three times); the determinant is $e^{t\lvert c\rvert^{2}}$. The group is invertible for every $t$, although the generator $\mathrm{Sd}_{\tilde Q}$ is singular. It dilates the image line and fixes the kernel hyperplane.

The element flow. The exponential $\exp(t\tilde Q)=e_t+(e^{tc}-1)\tilde\Pi$ is invertible for every $t$, and it lies in $\Sigma_{11}$ for generic $t\neq0$; at the exceptional values $e^{tc}=1$ it is $e_t$ on the wall $\Sigma_{10}$. Its two-sided operator $\mathrm{Sd}_{\exp(t\tilde Q)}$ is therefore a conformal Lorentz transformation, with the generator $\mathrm{Sd}_{\tilde Q}$:

$$ \mathrm{Sd}_{\exp(t\tilde Q)} =\exp\bigl(t\mathrm{Sd}_{\tilde Q}\bigr). $$

The generator is singular, but the one-parameter group it generates is invertible. The one-sided rate is $c$; the two-sided operator rate is $\lvert c\rvert^{2}$, the squared modulus. The generator is now an oblique projection rather than a conformal Lorentz transformation.

The exponential on the nilpotents

The operator exponential. Since $\mathrm{Sd}_{\tilde Q}^{2}=0$,

$$ \boxed{\; \exp\bigl(t\mathrm{Sd}_{\tilde Q}\bigr)=I+t\mathrm{Sd}_{\tilde Q} \;} $$

for every complex $t$. The eigenvalues are all $1$; the Jordan form is $J_2(1)\oplus1\oplus1$; the determinant is $1$. The group is unipotent: it is a shear, not a dilation. It is invertible for every $t$, although the generator is nilpotent and singular.

The element flow. Since $\tilde\nu^{2}=0$,

$$ \exp\bigl(t\tilde\nu\bigr)=e_t+t\tilde\nu . $$

The quaternionic norm is $1$, and the vector form is

$$ B^{2}\bigl(\exp(t\tilde\nu)\bigr)=t^{2}B^{2}(\tilde\nu)=0 . $$

Hence the exponential is invertible with $B^{2}=0$: for every $t$ it lies in the N-regular, spectrally degenerate sector $\Sigma_{10}$ (at $t=0$ it is $e_t$). The element flow moves one wall away: it removes the norm degeneration and preserves the spectral one. Its two-sided operator is

$$ \mathrm{Sd}_{\exp(t\tilde\nu)} =\exp\bigl(t\mathrm{Sd}_{\tilde Q}\bigr), $$

a unipotent Lorentz flow of Jordan type $2+1+1$. It fixes the kernel hyperplane and shears along the single chain. The generator $\mathrm{Sd}_{\tilde Q}$ is singular; the one-parameter group it generates is invertible.

The one-sided exponential of the companion article is a unipotent shear of the operator space, not a Lorentz transformation. The two-sided exponential is a unipotent Lorentz flow. The difference is the same as in the regular sector, at the level of the slots: the one-sided operator reads one slot and the two-sided operator pairs two. The analogy stops there. In the regular sector the one-sided slot carries the half-angle $\theta/2$; here there is no angle to halve, and the one-sided exponential is unipotent.

Worked Examples

Two examples on each stratum.

Example 1: a Hermitian idempotent

Take $\tilde Q=e_t+i\vec e_x$. Then $c=2$, $\tilde\Pi=(e_t+i\vec e_x)/2$, $\kappa=1$, $a=1$, $P=\tilde\Pi$, and $\tilde Q_H=\tilde Q$, $\tilde Q_0=\tilde\Pi$. The two-sided operator is $\mathrm{Sd}_{\tilde Q}=4\mathrm{Sd}_{\tilde\Pi}$, with rank one, eigenvalue $4$, singular value $4$, and image the line $\mathbb{C}\tilde Q$. In the basis $(e_t,\vec e_x,\vec e_y,\vec e_z)$,

$$ \mathsf M\bigl(\mathrm{Sd}_{\tilde Q}\bigr)= \begin{pmatrix} 2&-2i&0&0\\ 2i&2&0&0\\ 0&0&0&0\\ 0&0&0&0 \end{pmatrix}, $$

with trace $4$, rank one and square $\mathrm{Sd}_{\tilde Q}^{2}=4\mathrm{Sd}_{\tilde Q}$. The exponential is

$$ \exp\bigl(t\mathrm{Sd}_{\tilde Q}\bigr)=I+\bigl(e^{4t}-1\bigr)\mathrm{Sd}_{\tilde\Pi}. $$

Example 2: a non-Hermitian idempotent

Take $\tilde Q=e_t+\vec e_x+i\sqrt2\,\vec e_z$. Then $c=2$, $\kappa=2$, $a=\sqrt2$, $\tilde{Ζ}=\vec e_x+i\sqrt2\,\vec e_z$, and

$$ P=\frac12 e_t-\frac{i\sqrt2}{4}\vec e_y+\frac{i\sqrt2}{4}\vec e_z, \qquad \tilde Q_H=\sqrt2\,e_t-i\vec e_y+i\vec e_z, \qquad \tilde Q_0=\frac{\sqrt2}{4}e_t+\frac{\sqrt2}{4}\vec e_x+\frac{i}{2}\vec e_z . $$

The two-sided operator is $\mathrm{Sd}_{\tilde Q}=4\mathrm{Sd}_{\tilde\Pi}$, with rank one, eigenvalue $4$, and singular value $8$. The eigenvalue-to-singular-value ratio is $\tfrac12=\cos^{2}\theta$, because $\cos\theta=1/\sqrt2$. The image is the null line $\mathbb{C}P$, with $P\neq\tilde\Pi$; the two-sided operator is not self-adjoint, and its Peirce matrix in the spectral basis is not diagonal. The example shows the difference between the eigenvalue $\lvert c\rvert^{2}$ and the singular value $4a^{2}$ for an oblique idempotent.

Example 3: the standard nilpotent

Take

$$ \tilde\nu=\vec e_x+i\vec e_y . $$

Then $a=1$, $\vec c=\vec e_z$, and

$$ \tilde\Pi_+=\frac{e_t+i\vec e_z}{2}, \qquad \tilde\Pi_-=\frac{e_t-i\vec e_z}{2}, \qquad \tilde Q_H=e_t+i\vec e_z, \qquad \tilde Q_0=\frac{\vec e_x+i\vec e_y}{2}. $$

The matrix model is

$$ \mathsf M_2(\tilde\nu)= \begin{pmatrix}0&-2i\\0&0\end{pmatrix}, \qquad \mathsf M_2(\tilde Q_H)= \begin{pmatrix}2&0\\0&0\end{pmatrix}, \qquad \mathsf M_2(\tilde Q_0)= \begin{pmatrix}0&-i\\0&0\end{pmatrix}. $$

In the standard basis $(e_t,\vec e_x,\vec e_y,\vec e_z)$ the two-sided operator is

$$ \mathsf M\bigl(\mathrm{Sd}_{\tilde\nu}\bigr)= \begin{pmatrix} 2&0&0&2i\\ 0&0&0&0\\ 0&0&0&0\\ 2i&0&0&-2 \end{pmatrix}, $$

with trace zero, rank one and square zero. The image is the line $\mathbb{C}(e_t+i\vec e_z)=\mathbb{C}\tilde\Pi_+$; the first column is $2(e_t+i\vec e_z)=4\tilde\Pi_+$.

For the Peirce form, take

$$ \vec X=\frac{-\vec e_x+i\vec e_y}{4}, \qquad \tilde\nu\vec X=\tilde\Pi_+, \qquad \vec X\tilde\nu=\tilde\Pi_- . $$

In the basis $(\tilde\Pi_+,\tilde\nu,\vec X,\tilde\Pi_-)$,

$$ \mathsf M\bigl(\mathrm{Sd}_{\tilde\nu}\bigr)= \begin{pmatrix} 0&0&0&4\\ 0&0&0&0\\ 0&0&0&0\\ 0&0&0&0 \end{pmatrix}, $$

the single chain being $\tilde\Pi_-\to4\tilde\Pi_+\to0$. The exponential is

$$ \exp\bigl(t\mathrm{Sd}_{\tilde\nu}\bigr)=I+t\mathrm{Sd}_{\tilde\nu}. $$

Example 4: a complex scalar multiple

For

$$ \tilde\nu=c_0(\vec e_x+i\vec e_y), \qquad c_0=\lvert c_0\rvert e^{i\beta}, $$

the element has $a=\lvert c_0\rvert$, and

$$ \tilde Q_0=e^{i\beta}\frac{\vec e_x+i\vec e_y}{2}, \qquad \mathrm{Sd}_{\tilde\nu} =4\lvert c_0\rvert^{2}\mathrm{Sd}_{\tilde Q_0}. $$

The phase $e^{i\beta}$ cancels in the two-sided operator. The two-sided operator depends only on $\lvert c_0\rvert$ and the canonical pair; the one-sided operator of the companion article keeps the phase.

Summary

Main Results on the Idempotent Multiples

item value
sector $\Sigma_{01}=\{\langle\tilde Q\rangle_{\natural}=0,\ B^{2}\neq0\}$
parameter $\tilde Q=c\tilde\Pi$, $c=2Q_t\neq0$
two-sided operator $\mathrm{Sd}_{\tilde Q}(\tilde X)=\tilde Q\tilde X\tilde Q^{*}$
closed form $\mathrm{Sd}_{\tilde Q}=\lvert c\rvert^{2}\mathrm{Sd}_{\tilde\Pi}$, $\mathrm{Sd}_{\tilde\Pi}^{2}=\mathrm{Sd}_{\tilde\Pi}$
square $\mathrm{Sd}_{\tilde Q}^{2}=\lvert c\rvert^{2}\mathrm{Sd}_{\tilde Q}$
rank $1$
spectrum $\{\lvert c\rvert^{2},0\}$, multiplicities $1,3$
Jordan form diagonal, $\operatorname{diag}(\lvert c\rvert^{2},0,0,0)$; the eigenbasis of $\tilde Q$ diagonalizes it only when $\tilde\Pi$ is Hermitian
trace $\lvert c\rvert^{2}=4a^{2}\cos^{2}\theta$
the determinant of the image $0$
eigenvalue on the image $\lvert c\rvert^{2}$
singular value $4a^{2}$
image the null line $\mathbb{C}P$, $P=\tilde Q\tilde Q^{*}/(4a^{2})$
kernel $\{\tilde X:\tilde\Pi\tilde X\tilde\Pi^{*}=0\}$, a hyperplane
scalar collapse $\mathrm{Sd}_{\tilde Q}=4a^{2}\mathrm{Sd}_{\tilde Q_0}$
exponential $\exp(t\mathrm{Sd}_{\tilde Q})=I+(e^{t\lvert c\rvert^{2}}-1)\mathrm{Sd}_{\tilde\Pi}$
element flow $\exp(t\tilde Q)=e_t+(e^{tc}-1)\tilde\Pi$, a unit in $\Sigma_{11}$ for generic $t\neq0$

Main Results on the Nilpotents

item value
sector $\Sigma_{00}=\{\langle\tilde Q\rangle_{\natural}=0,\ B^{2}=0\}$
parameter $\tilde\nu=\vec A+i\vec B$, $\lVert\vec A\rVert=\lVert\vec B\rVert=a>0$, $\vec A\cdot\vec B=0$
canonical pair $\tilde\Pi_\pm=(e_t\pm i\vec c)/2$, $\vec c=\vec A\times\vec B/a^{2}$
polar pair $\tilde Q_H=2a\tilde\Pi_+$, $\tilde Q_0=\tilde\nu/(2a)$, $\tilde Q_0^{2}=0$
two-sided operator $\mathrm{Sd}_{\tilde Q}(\tilde X)=\tilde Q\tilde X\tilde Q^{*}$
square $\mathrm{Sd}_{\tilde Q}^{2}=0$
rank $1$
spectrum $\{0\}$, algebraic multiplicity $4$, geometric multiplicity $3$
Jordan form $J_2(0)\oplus0\oplus0$
trace $0$
the determinant of the image $0$
eigenvalue $0$
singular value $4a^{2}$
image the null line $\mathbb{C}\tilde\Pi_+$
kernel $\{\tilde X:\tilde\nu\tilde X\tilde\nu^{*}=0\}$, a hyperplane
chain $\tilde\Pi_-\mapsto4a^{2}\tilde\Pi_+\mapsto0$
scalar collapse $\mathrm{Sd}_{\tilde Q}=4a^{2}\mathrm{Sd}_{\tilde Q_0}$
exponential $\exp(t\mathrm{Sd}_{\tilde Q})=I+t\mathrm{Sd}_{\tilde Q}$
element flow $\exp(t\tilde\nu)=e_t+t\tilde\nu$, a unit in $\Sigma_{10}$ for every $t$ (at $t=0$ it is $e_t$)

What the Two-Sided Operator Shows

The norm fixes the rank and the image; the spectral condition fixes the eigenvalue and the Jordan type.

  • The norm degeneration fixes the rank: on the null cone the two-sided operator has rank one, its image is a single null line, and its kernel has dimension three.
  • On $\Sigma_{01}$ the operator is $\lvert c\rvert^{2}=4a^{2}\cos^{2}\theta$ times a projection, with spectrum $\{\lvert c\rvert^{2},0,0,0\}$ and no defect.
  • On $\Sigma_{00}$ the operator is square-zero, $\mathrm{Sd}_{\tilde\nu}^{2}=0$, with the single eigenvalue zero, one Jordan block of size two and two fixed zero directions: the maximal collapse.
  • The singular value $4a^{2}$ is the square of the one-sided singular value $2a$, and it is independent of the phase.
  • The polar decomposition of the operator is the congruence of the polar decomposition of the element; the positive factor is $4a^{2}\mathrm{Sd}_P$ or $4a^{2}\mathrm{Sd}_{\tilde\Pi_+}$.
  • The phase and the sign cancel: the two-sided operator sees only $\lvert c\rvert^{2}$ and the canonical pair.
  • The two slots survive as the singular vectors; their overlap $1/\sqrt{\kappa}$ vanishes exactly at the wall $\Sigma_{00}$.

Summary of Notation

Symbol Meaning
$\langle\tilde Q\rangle_{\natural}$, $\langle\tilde Q\rangle_{*}$ the quaternionic norm $Q_t^{2}+B^{2}$, zero on the null cone, and the Hermitian norm $\operatorname{Sc}(\tilde Q\tilde Q^{*})$, positive there
$\tilde Q=c\tilde\Pi$ an element of $\Sigma_{01}$
$\mathrm{Sd}_{\tilde Q}$ the two-sided operator $\tilde X\mapsto\tilde Q\tilde X\tilde Q^{*}$
$\kappa=2\langle\tilde\Pi\rangle_{*}$ the obliqueness, $\kappa\ge1$
$a=\sqrt{\langle\tilde Q\rangle_{*}/2}$ the half singular value, $2a$ the non-zero singular value
$P=\tilde Q\tilde Q^{*}/(4a^{2})$ the rank-one Hermitian idempotent and the image line
$\tilde Q_H=2aP$ the Hermitian factor
$\tilde Q_0=\tilde Q/(2a)$ the partial-isometry factor
$u,v$ the left and right singular vectors
$v^{\dagger}u=Q_t/a$ the overlap, modulus $1/\sqrt{\kappa}$
$\theta$ the angle between the slots, $\cos\theta=1/\sqrt{\kappa}$
$\tilde\nu=\vec A+i\vec B$ a non-zero element of $\Sigma_{00}$
$a=\lVert\vec A\rVert=\lVert\vec B\rVert$ the half singular value
$\vec c=\vec A\times\vec B/a^{2}$ the unit spectral direction
$\tilde\Pi_\pm=(e_t\pm i\vec c)/2$ the canonical Hermitian idempotents
$\tilde Q_H=2a\tilde\Pi_+$ the Hermitian factor
$\tilde Q_0=\tilde\nu/(2a)$ the nilpotent partial-isometry factor
$L_{\tilde\nu},R_{\tilde\nu}$ the one-sided operators of the companion article
$\vec X$ a normalized corner vector, $\tilde\nu\vec X=\tilde\Pi_+$
$J_2(0)$ the nilpotent Jordan block of size two

Further Reading