The Product of Two Unipotent Elements and Its Spectrum

Introduction

The unipotent elements of the biquaternion algebra are the elements $U=e_t+\vec\Upsilon$ with $\vec\Upsilon$ a pure nilpotent, $\vec\Upsilon^2=0$; the element is non-trivial when $\vec\Upsilon\neq0$. They are units of norm one, their spectrum is the double eigenvalue $1$, and the non-trivial ones have minimal polynomial $(X-1)^2$. The product of two non-trivial unipotents is again a unit of norm one, but it is not usually unipotent. This article computes its spectrum and shows that one complex number decides it.

Let $U_i=e_t+\vec\Upsilon_i$ and $T=U_1U_2$. Put $t=2\operatorname{Sc}(\vec\Upsilon_1\vec\Upsilon_2)$, twice the scalar part of the product of the two nilpotent parts. Then

$$ p_T(\lambda)=\lambda^2-(2+t)\lambda+1, $$

with discriminant $\Delta=t(t+4)$, scalar part $Q_t=1+t/2$ and vector square $B^2(T)=-t(t+4)/4$. The product is defective exactly at $t=0$ and $t=-4$, with the scalar exception $T=e_t$ at $t=0$; otherwise it is regular. On the regular set, the spectrum is on the unit circle exactly when $t$ is real and $-40$ or $t<-4$, and non-real and reciprocal when $t$ is not real. For real $t$, the sign of $t$ alone does not decide the type: the type is fixed by whether $t$ lies in $(-4,0)$ or outside it, equivalently by the sign of $t(t+4)$, and the sign of $t$ then separates the positive real branch from the negative one.

The two examples show two of the regular types. With $t=-1$ the product has eigenvalues $e^{\pm i\pi/3}$; with $t=1$ it has eigenvalues $(3\pm\sqrt5)/2$. The two values differ only in the pairing of the two nilpotent parts. The article also proves that every element of the norm-one group except $-e_t$ is a product of two non-trivial unipotent elements, and that every element of the group of invertibles is a central scalar times an element of the norm-one slice.

The product law on the spectral-degenerate sector, which uses the pairing $s=\vec\Upsilon_1\cdot\vec\Upsilon_2$, is The Complete Polar Decomposition of an N-Regular Biquaternion, §The Group Law; the spectral criterion, the characteristic polynomial and the cone are The Spectrum of a Biquaternion; the norm, the norm-one group and the matrix model are The N-Regular Sector and $M_2(\mathbb{C})$ Pauli Representation. The relation $t=-2s$ is the bridge between the two computations, and nothing of the cited articles is reproved here.

Conventions. The algebra is $\mathbb{B}=\mathbb{C}\otimes_{\mathbb{R}}\mathbb{H}$ on the basis $e_t,\vec e_x,\vec e_y,\vec e_z$, with $\vec e_x\vec e_y=\vec e_z$, $\vec e_y\vec e_z=\vec e_x$, $\vec e_z\vec e_x=\vec e_y$ and $\vec e_k^2=-e_t$. An element is $\tilde Q=Q_te_t+\vec Q$ with $Q_\mu\in\mathbb{C}$ and $\vec Q=Q_x\vec e_x+Q_y\vec e_y+Q_z\vec e_z$. The quaternionic norm, written $N$, is

$$ N(\tilde Q)=\langle\tilde Q\rangle_{\natural}=Q_t^2+B^2,\qquad B^2=Q_x^2+Q_y^2+Q_z^2, $$

and the matrix model is fixed by

$$ \mathsf{M}_2(e_t)=\mathrm I_2,\qquad \mathsf{M}_2(\vec e_x)=-i\sigma_1,\qquad \mathsf{M}_2(\vec e_y)=-i\sigma_2,\qquad \mathsf{M}_2(\vec e_z)=-i\sigma_3. $$

The reading of an element is twice its scalar part, $2\operatorname{Sc}\tilde Q=2Q_t=\operatorname{Tr}\mathsf{M}_2(\tilde Q)$, and the text writes $2\operatorname{Sc}\tilde Q$; the matrix trace is named only beside the matrix $\mathsf{M}_2$. The determinant is $\det\mathsf{M}_2(\tilde Q)=N(\tilde Q)$, and the corpus writes $N(\tilde Q)$ for it. The norm-one group is

$$ \mathbb{B}^{\times}_1=\{\tilde Q:N(\tilde Q)=1\}\cong SL_2(\mathbb{C}). $$

A pure nilpotent is a vector $\vec\Upsilon=\upsilon_x\vec e_x+\upsilon_y\vec e_y+\upsilon_z\vec e_z$ with $\vec\Upsilon^2=-B^2e_t=0$; it is non-zero when $\vec\Upsilon\neq0$. The dot product is $\vec\Upsilon_1\cdot\vec\Upsilon_2=\sum_k\upsilon_{1k}\upsilon_{2k}$, and $\operatorname{Sc}(\vec\Upsilon_1\vec\Upsilon_2)=-\vec\Upsilon_1\cdot\vec\Upsilon_2$. A unipotent element is $U=e_t+\vec\Upsilon$ with $\vec\Upsilon$ a pure nilpotent; it is non-trivial when $\vec\Upsilon\neq0$. An element is regular (spectrally regular) when $B^2\neq0$, that is when it is semisimple with two distinct eigenvalues, and defective when it is not semisimple, that is when $B^2=0$ and $\vec Q\neq0$. The spectral-degenerate cone is $\mathcal S=\{B^2=0\}$, and the null cone is $\mathcal N=\{N=0\}$. The algebra, the norm and the cone are Introduction to the Algebra of Biquaternions, N-Regularity and N-Degeneracy in the Biquaternion Algebra and The Spectrum of a Biquaternion.

The Norm-One Slice and the Unipotent Elements

The spectral criterion on the slice. For a general element $\tilde Q=Q_te_t+\vec Q$ the characteristic polynomial is

$$ p_{\tilde Q}(X)=X^2-2Q_tX+N(\tilde Q), $$

with discriminant $-4B^2$ and eigenvalues $Q_t\pm iB$. On the norm-one slice $N=1$ the constant term is one, so

$$ p_{\tilde Q}(X)=X^2-2Q_tX+1,\qquad \Delta=4(Q_t^2-1)=-4B^2. $$

The element is defective exactly when $B^2=0$ and $\vec Q\neq0$. On the slice this forces $Q_t^2=1$, hence $Q_t=\pm1$. The part $Q_t=1$ consists of the unipotents $e_t+\vec\Upsilon$, and the part $Q_t=-1$ consists of their negatives, $-e_t+\vec\Upsilon=-(e_t-\vec\Upsilon)$.

The table lists the $B^2=0$ part of the norm-one slice; in the last two rows $\vec\Upsilon$ is a non-zero pure nilpotent.

element $Q_t$ $B^2$ spectrum minimal polynomial
$e_t$ $1$ $0$ $\{1\}$ $X-1$
$e_t+\vec\Upsilon$ $1$ $0$ $\{1\}$ $(X-1)^2$
$-e_t$ $-1$ $0$ $\{-1\}$ $X+1$
$-e_t+\vec\Upsilon$ $-1$ $0$ $\{-1\}$ $(X+1)^2$

The two signs are exchanged by the central element $-e_t$, which multiplies the spectrum by $-1$ and leaves the defectivity unchanged; the classification of a single element is therefore the same on the two parts. The products computed below are products of two elements of the $Q_t=1$ part; the image theorem covers the rest of the group. The product of two non-trivial unipotents is again in the norm-one group, and the rest of the article computes its spectrum.

The Product and Its Characteristic Polynomial

Theorem (the product). Let $\vec\Upsilon_1,\vec\Upsilon_2$ be non-zero pure nilpotents, $U_i=e_t+\vec\Upsilon_i$, $T=U_1U_2$, and $t=2\operatorname{Sc}(\vec\Upsilon_1\vec\Upsilon_2)$. Then

(i) $T=e_t+\vec\Upsilon_1+\vec\Upsilon_2+\vec\Upsilon_1\vec\Upsilon_2$;

(ii) $2\operatorname{Sc}T=2+t$ and $\det\mathsf{M}_2(T)=1$;

(iii) the characteristic polynomial is

$$ p_T(\lambda)=\lambda^2-(2+t)\lambda+1, $$

with discriminant $\Delta=t(t+4)$;

(iv) the scalar part and the vector square are

$$ Q_t=1+\frac t2,\qquad B^2(T)=-\frac{t(t+4)}4; $$

(v) with $s=\vec\Upsilon_1\cdot\vec\Upsilon_2$, one has $t=-2s$ and

$$ B^2(T)=s(2-s). $$

Proof. The expansion is the product. Since $\vec\Upsilon_i$ is a pure vector, $2\operatorname{Sc}\vec\Upsilon_i=0$, and since $\vec\Upsilon_i^2=0$, $\det\mathsf{M}_2(U_i)=N(U_i)=1$. Hence

$$ 2\operatorname{Sc}T=2\operatorname{Sc}(e_t)+2\operatorname{Sc}(\vec\Upsilon_1)+2\operatorname{Sc}(\vec\Upsilon_2)+2\operatorname{Sc}(\vec\Upsilon_1\vec\Upsilon_2)=2+t, \qquad \det\mathsf{M}_2(T)=\det\mathsf{M}_2(U_1)\det\mathsf{M}_2(U_2)=1. $$

The characteristic polynomial of the $2\times2$ image is $\lambda^2-2\operatorname{Sc}T\,\lambda+\det T$, which gives (iii), and its discriminant is $(2+t)^2-4=t(t+4)$. The scalar part is $Q_t=1+t/2$. The product is a unit of norm one, because the norm is multiplicative and $N(U_i)=1$; hence $B^2(T)=N(T)-Q_t^2=1-Q_t^2=-t(t+4)/4$. Finally, $\operatorname{Sc}(\vec\Upsilon_1\vec\Upsilon_2)=-\vec\Upsilon_1\cdot\vec\Upsilon_2$ gives $t=2\operatorname{Sc}(\vec\Upsilon_1\vec\Upsilon_2)=-2s$, and substitution gives $B^2(T)=s(2-s)$.

Remark (the product law on the sector). For $\tilde Q_i=Q_{ti}(e_t+\vec\Upsilon_i)$, The Complete Polar Decomposition of an N-Regular Biquaternion, §The Group Law, proves

$$ B^2(\tilde Q_1\tilde Q_2)=Q_{t1}^2Q_{t2}^2\,s(2-s),\qquad s=\vec\Upsilon_1\cdot\vec\Upsilon_2. $$

On the norm-one slice $Q_{t1}=Q_{t2}=1$, and $t=-2s$; the two formulas agree. The sector's criterion that the product stays on the spectral-degenerate cone exactly when $s=0$ or $s=2$ is the present criterion $t=0$ or $t=-4$.

The Criterion

Theorem (defective or regular). With $T=U_1U_2$ and $t=2\operatorname{Sc}(\vec\Upsilon_1\vec\Upsilon_2)$:

(i) if $t\notin\{0,-4\}$, then $T$ is regular, with two distinct eigenvalues;

(ii) if $t=0$, then $T$ has the single eigenvalue $1$; it is $e_t$ exactly when $\vec\Upsilon_2=-\vec\Upsilon_1$, and otherwise it is a non-trivial unipotent, hence defective;

(iii) if $t=-4$, then $T$ has the single eigenvalue $-1$; it is $-e_t$ times a non-trivial unipotent, hence defective, and $T=-e_t$ is impossible.

Proof. (i) The discriminant $t(t+4)$ is non-zero, so the two roots of $p_T$ are distinct; a $2\times2$ matrix with distinct eigenvalues is diagonalisable, hence semisimple.

(ii) For $t=0$, $p_T(\lambda)=(\lambda-1)^2$. A semisimple element with this spectrum is $e_t$: a diagonalisable matrix with both eigenvalues $1$ is the identity. If $T=e_t$, then $U_2=U_1^{-1}=e_t-\vec\Upsilon_1$, so $\vec\Upsilon_2=-\vec\Upsilon_1$; conversely this choice gives $T=e_t$. If $T\neq e_t$, the minimal polynomial is $(X-1)^2$, so $T-e_t$ is a non-zero nilpotent and $T$ is a non-trivial unipotent.

(iii) For $t=-4$, $p_T(\lambda)=(\lambda+1)^2$. A semisimple element with this spectrum is $-e_t$. If $T=-e_t$, then $U_2=-U_1^{-1}=-e_t+\vec\Upsilon_1$, whose scalar part is $-1$, but $U_2=e_t+\vec\Upsilon_2$ has scalar part $1$; contradiction. Hence $T\neq-e_t$, the minimal polynomial is $(X+1)^2$, and $T=-e_t+M$ with $M^2=0$, $M\neq0$, that is $T=-(e_t-M)$ with $e_t-M$ unipotent.

The two roots therefore carry different exceptions.

$t$ $p_T(\lambda)$ $T$ type
$t\notin\{0,-4\}$ two distinct roots regular semisimple
$t=0$, $\vec\Upsilon_2=-\vec\Upsilon_1$ $(\lambda-1)^2$ $e_t$ scalar
$t=0$, $\vec\Upsilon_2\neq-\vec\Upsilon_1$ $(\lambda-1)^2$ non-trivial unipotent defective
$t=-4$ $(\lambda+1)^2$ $-e_t$ times a non-trivial unipotent defective

The Three Regular Types

Theorem (the regular spectrum). Let $t\notin\{0,-4\}$. The eigenvalues of $T$ are

$$ \lambda_\pm=1+\frac t2\pm\frac12\sqrt{t(t+4)},\qquad \lambda_+\lambda_-=1. $$

(i) If $t$ is real and $-4unit-circle type.

(ii) If $t$ is real and $t>0$ or $t<-4$, the eigenvalues are real and reciprocal; they are both positive for $t>0$ and both negative for $t<-4$; this is the real type.

(iii) If $t$ is not real, the eigenvalues are non-real and reciprocal, with $\lvert\lambda_\pm\rvert\neq1$; this is the non-real type.

Proof. The formula is the quadratic formula for $p_T$, and the product of the roots is the constant term, one.

(i) For real $t$ in $(-4,0)$, $\Delta=t(t+4)<0$, so the roots are non-real conjugates; a conjugate pair with product one has modulus one.

(ii) For real $t$ outside $[-4,0]$, $\Delta>0$, so the roots are real and distinct; their product is one, so they are reciprocal. For $t>0$, $Q_t=1+t/2>1$ and both roots $Q_t\pm\sqrt{Q_t^2-1}$ are positive; for $t<-4$, $Q_t<-1$ and both are negative.

(iii) If $t$ is not real, the roots cannot be real, since their sum $2+t=2Q_t$ would then be real. They are reciprocal because the product is one. If $\lvert\lambda\rvert=1$, then $1/\lambda=\bar\lambda$, so the two roots are $\lambda$ and $\bar\lambda$, and their sum $2Q_t$ is real, which forces $t$ real; contradiction. Hence $\lvert\lambda\rvert\neq1$.

$t$ $Q_t$ spectrum type
real, $-4 $-1 $Q_t\pm iB$, $\lvert\lambda\rvert=1$ unit circle
real, $t>0$ $Q_t>1$ real, reciprocal, positive real
real, $t<-4$ $Q_t<-1$ real, reciprocal, negative real
not real not real non-real, reciprocal, $\lvert\lambda\rvert\neq1$ non-real

Remark (the special line $t=-2+ib$). On this line, $Q_t=ib/2$ is purely imaginary and $B^2=1+b^2/4>0$ is real positive, but the eigenvalues are $i(b/2\pm\sqrt{1+b^2/4})$, purely imaginary and reciprocal. They are not on the unit circle unless $b=0$. Thus $B^2>0$ alone does not imply the unit-circle type; the reality of $Q_t$ is also needed. For the product, $Q_t$ is real exactly when $t$ is real.

The Sign of the Pairing

The invariant $t$ is twice the scalar part of the product of the two nilpotent parts; the invariant $s=\vec\Upsilon_1\cdot\vec\Upsilon_2=-t/2$ is their pairing, the one used in the product law on the sector. In terms of $s$, the criterion reads as follows.

$s$ $t=-2s$ spectrum type
real, $s<0$ $t>0$ real, reciprocal, positive real, positive
real, $0 $-4 unit circle unit circle
real, $s>2$ $t<-4$ real, reciprocal, negative real, negative
not real not real non-real, reciprocal non-real
$s=0$ $t=0$ $\{1\}$ trivial or non-trivial unipotent
$s=2$ $t=-4$ $\{-1\}$ $-e_t$ times a non-trivial unipotent

The sign of $s$ alone does not decide the type. It is positive on $(0,2)$ and on $(2,\infty)$, which give the unit-circle and the real types respectively; it is negative on $(-\infty,0)$, which gives the real type. The type is fixed by whether $s$ lies in $(0,2)$ or outside it, equivalently by the sign of $s(2-s)$ when $s$ is real; the sign of $s$ then separates the two real branches. Since $B^2(T)=s(2-s)$, the type is read off the sign of $B^2(T)$ on the real line away from the two roots.

In particular, the statement that a positive pairing gives a real spectrum and a negative pairing a unit-circle spectrum is false as stated: the positive interval $(0,2)$ gives the unit-circle type, and the positive interval $(2,\infty)$ gives the real type. The correct statement is that the unit-circle type is the interval $02$, and the two endpoints are the scalar and defective roots.

Two Examples

A Unit-Circle Product

Take

$$ \vec\Upsilon_1=\frac i2\vec e_x-\frac12\vec e_y,\qquad \vec\Upsilon_2=-\frac i2\vec e_x-\frac12\vec e_y. $$

Both are pure nilpotents: each has $B^2=(i/2)^2+(-1/2)^2=0$, so $\vec\Upsilon_i^2=0$. Their pairing is

$$ s=\vec\Upsilon_1\cdot\vec\Upsilon_2=\frac i2\left(-\frac i2\right)+\left(-\frac12\right)\left(-\frac12\right)=\frac12, $$

hence $t=-1$. The unipotents are

$$ U_1=e_t+\frac i2\vec e_x-\frac12\vec e_y,\qquad U_2=e_t-\frac i2\vec e_x-\frac12\vec e_y, $$

with matrix images

$$ \mathsf{M}_2(U_1)=\begin{pmatrix}1&1\\0&1\end{pmatrix},\qquad \mathsf{M}_2(U_2)=\begin{pmatrix}1&0\\-1&1\end{pmatrix}. $$

The product is

$$ T=U_1U_2=\frac12e_t-\vec e_y-\frac i2\vec e_z,\qquad \mathsf{M}_2(T)=\begin{pmatrix}0&1\\-1&1\end{pmatrix}. $$

Its invariants are $Q_t=1/2$, $B^2(T)=3/4$, $N(T)=1$, and its characteristic polynomial is

$$ p_T(\lambda)=\lambda^2-\lambda+1, $$

with eigenvalues $\lambda_\pm=(1\pm i\sqrt3)/2=e^{\pm i\pi/3}$. The product is regular and of unit-circle type.

A Real Product

Keep $\vec\Upsilon_1$ and replace the second nilpotent by

$$ \vec\Upsilon_3=\frac i2\vec e_x+\frac12\vec e_y=-\vec\Upsilon_2. $$

It is a pure nilpotent, and

$$ s=\vec\Upsilon_1\cdot\vec\Upsilon_3=\frac i2\cdot\frac i2+\left(-\frac12\right)\frac12=-\frac12, $$

hence $t=1$. The unipotent is

$$ U_3=e_t+\frac i2\vec e_x+\frac12\vec e_y,\qquad \mathsf{M}_2(U_3)=\begin{pmatrix}1&0\\1&1\end{pmatrix}, $$

and the product is

$$ T_3=U_1U_3=\frac32e_t+i\vec e_x+\frac i2\vec e_z,\qquad \mathsf{M}_2(T_3)=\begin{pmatrix}2&1\\1&1\end{pmatrix}. $$

Its invariants are $Q_t=3/2$, $B^2(T_3)=-5/4$, $N(T_3)=1$, and its characteristic polynomial is

$$ p_{T_3}(\lambda)=\lambda^2-3\lambda+1, $$

with eigenvalues $(3\pm\sqrt5)/2$. The product is regular and of real type.

data unit-circle example real example
second nilpotent $\vec\Upsilon_2=-\frac i2\vec e_x-\frac12\vec e_y$ $\vec\Upsilon_3=\frac i2\vec e_x+\frac12\vec e_y$
$s$ $1/2$ $-1/2$
$t$ $-1$ $1$
product $\frac12e_t-\vec e_y-\frac i2\vec e_z$ $\frac32e_t+i\vec e_x+\frac i2\vec e_z$
$p_T$ $\lambda^2-\lambda+1$ $\lambda^2-3\lambda+1$
spectrum $e^{\pm i\pi/3}$ $(3\pm\sqrt5)/2$
type unit circle real

The Two Roots

The Identity at $t=0$

At $t=0$ the product is $e_t$ or a non-trivial unipotent. Both occur, and the difference is the inverse:

  • with $U_2=U_1^{-1}=e_t-\vec\Upsilon_1$, the product is $T=e_t$;
  • with $U_2=U_1$, the product is $T=e_t+2\vec\Upsilon_1$, a non-trivial unipotent.

Both have $s=0$, characteristic polynomial $(\lambda-1)^2$, norm one and scalar part one; they differ in the minimal polynomial, $X-1$ against $(X-1)^2$. The root $t=0$ is therefore not a defective locus without exception: it is the coalescence of the scalar $e_t$ and the unipotent Jordan block.

The Second Root at $t=-4$

A witness is

$$ \vec\Upsilon_1=\frac i2\vec e_x-\frac12\vec e_y,\qquad \vec\Upsilon_4=-\frac32i\vec e_x-\frac52\vec e_y+2i\vec e_z. $$

The second is a pure nilpotent, because

$$ \left(-\frac32i\right)^2+\left(-\frac52\right)^2+(2i)^2=-\frac94+\frac{25}4-4=0, $$

and its pairing with $\vec\Upsilon_1$ is

$$ s=\frac i2\left(-\frac32i\right)+\left(-\frac12\right)\left(-\frac52\right)=\frac34+\frac54=2, $$

hence $t=-4$. The product is

$$ T_4=U_1U_4=-e_t-2i\vec e_x-2\vec e_y=-\bigl(e_t+2i\vec e_x+2\vec e_y\bigr), $$

and the bracket is unipotent, since the square of its vector part is zero. So $T_4$ is $-e_t$ times a non-trivial unipotent. The scalar $-e_t$ itself is not a product: if $U_1U_2=-e_t$, then $U_2=-U_1^{-1}=-e_t+\vec\Upsilon_1$, whose scalar part is $-1$, not $1$.

The Image of the Product Map

Theorem (the image). Every element of the norm-one group except $-e_t$ is a product of two non-trivial unipotent elements. The element $-e_t$ is not.

Proof. Work in $M_2(\mathbb{C})$, with $I=\mathrm I_2$. If $g=-I$, suppose $g=U_1U_2$ with $U_i=I+K_i$, $K_i^2=0$. Then $U_2=-U_1^{-1}=-I+K_1$, whose trace is $-2$, while $U_2$ has trace $2$; contradiction. If $g=I$, take any non-zero nilpotent $K$ and write $I=(I+K)(I-K)$. If $g$ is unipotent and $g\neq I$, put $K=g-I$; then $K^2=0$, $K\neq0$, and $g=(I+K/2)^2$. Otherwise $g$ is neither scalar nor unipotent, so $\operatorname{Tr}g\neq2$ and $g$ is not scalar. The linear form $\varphi(K)=\operatorname{Tr}(Kg)$ is non-zero on $\mathfrak{sl}_2$, because the trace pairing is non-degenerate and $g$ is not scalar. Since the nilpotent cone spans $\mathfrak{sl}_2$ (the remark below), $\varphi$ is non-zero at some nilpotent $K_0$; the cone is closed under scaling, so $K=\lambda K_0$ with $\lambda=(\operatorname{Tr}g-2)/\varphi(K_0)$ is a non-zero nilpotent with $\varphi(K)=\operatorname{Tr}g-2$. Put $U_1=I+K$ and $X=U_1^{-1}g=(I-K)g$. Then $\operatorname{Tr}X=\operatorname{Tr}g-\operatorname{Tr}(Kg)=2$ and $\det X=1$. If $X=I$, then $g=I+K$ is unipotent, contrary to the case; so $X\neq I$, and a non-scalar element of trace $2$ and determinant $1$ is unipotent. Hence $g=U_1X$ is a product of two unipotent elements. In each construction the two factors are non-trivial: the nilpotent $K$ is chosen non-zero, and in the last case $X\neq I$.

Remark (the spanning of $\mathfrak{sl}_2$). The matrices

$$ A_+=\begin{pmatrix}-1&1\\-1&1\end{pmatrix},\qquad A_-=\begin{pmatrix}1&1\\-1&-1\end{pmatrix} $$

are nilpotent, and $A_- - A_+ = 2\operatorname{diag}(1,-1)$. Together with $E_{12}=\begin{pmatrix}0&1\\0&0\end{pmatrix}$ and $E_{21}=\begin{pmatrix}0&0\\1&0\end{pmatrix}$, which are also nilpotent, they span $\mathfrak{sl}_2$. Hence the nilpotent cone spans $\mathfrak{sl}_2$.

Corollary. The product map from pairs of non-trivial unipotent elements to the norm-one group is surjective onto $\mathbb{B}^{\times}_1\setminus\{-e_t\}$. In particular, every regular element and every defective element of the norm-one group occurs; the only element missed is the scalar $-e_t$.

Scaling and the General Group of Invertibles

The restriction to the norm-one slice loses nothing for defective elements. Every defective element of the group of invertibles is of the form

$$ \tilde Q_2=c\,U,\qquad c\in\mathbb{C}^{\times},\quad U=e_t+\vec\Upsilon,\quad \vec\Upsilon^2=0,\quad\vec\Upsilon\neq0, $$

by The N-Regular, Spectrally Degenerate Sector, §Definition and Characterisation. For two such elements $\tilde Q_{2,i}=c_iU_i$,

$$ \tilde Q_{2,1}\tilde Q_{2,2}=c_1c_2\,T, $$

with $T=U_1U_2$ the product on the norm-one slice. The characteristic polynomial is

$$ p(\lambda)=\lambda^2-c_1c_2(2+t)\lambda+(c_1c_2)^2, $$

the discriminant is $(c_1c_2)^2t(t+4)$, and the norm is $(c_1c_2)^2$. The discriminant vanishes exactly when $t(t+4)=0$, so the product is scalar or defective exactly at the two roots; the scalar exception at $t=0$ is scaled to $c_1c_2e_t$. The spectrum is scaled by the central factor $c_1c_2$: the modulus is multiplied by $\lvert c_1c_2\rvert$ and the argument is shifted by $\arg(c_1c_2)$. The unit-circle type is preserved only when $\lvert c_1c_2\rvert=1$, and the real/non-real dichotomy is preserved by a real scale but not in general by a non-real one. The three types are properties of the norm-one slice, and that is why the classification is stated there.

Summary

The product of two non-trivial unipotent elements is a norm-one element whose spectrum is governed by the single number $t=2\operatorname{Sc}(\vec\Upsilon_1\vec\Upsilon_2)$. The characteristic polynomial is $\lambda^2-(2+t)\lambda+1$, the discriminant is $t(t+4)$, the scalar part is $1+t/2$, and the vector square is $-t(t+4)/4=s(2-s)$ with $s=\vec\Upsilon_1\cdot\vec\Upsilon_2=-t/2$.

The product is defective exactly at the two roots $t=0$ and $t=-4$, with the scalar exception $T=e_t$ at $t=0$; at $t=-4$ the product is always $-e_t$ times a non-trivial unipotent, and $-e_t$ itself never occurs. Off the roots the product is regular. The regular spectrum is on the unit circle exactly for real $t$ in $(-4,0)$, real and reciprocal for real $t$ in $(0,\infty)\cup(-\infty,-4)$, and non-real and reciprocal for non-real $t$.

The type is not fixed by the sign of $t$, nor by the sign of $s$. It is fixed by whether $t$ lies in $(-4,0)$ or outside it, equivalently by the sign of $t(t+4)$ for real $t$, equivalently by whether $s$ lies in $(0,2)$ or outside it. The sign of $t$ then separates the positive real branch from the negative one. The two examples $t=-1$ and $t=1$ illustrate two of the regular types.

Every element of the norm-one group except $-e_t$ is a product of two non-trivial unipotents. The general defective case reduces to the norm-one slice by the central scaling $\tilde Q_{2,i}=c_iU_i$, which scales the spectrum and leaves the defectivity unchanged.

Summary of Notation

symbol meaning
$\mathbb{B}$ the biquaternion algebra
$\tilde Q=Q_te_t+\vec Q$ a general element
$N(\tilde Q)=\langle\tilde Q\rangle_{\natural}=Q_t^2+B^2$ the quaternionic norm
$B^2=Q_x^2+Q_y^2+Q_z^2$ the square of the vector part
$\mathbb{B}^{\times}_1=\{N=1\}$ the norm-one group, $SL_2(\mathbb{C})$
$\vec\Upsilon_i$ a non-zero pure nilpotent, $\vec\Upsilon_i^2=0$
$U_i=e_t+\vec\Upsilon_i$ a non-trivial unipotent element
$T=U_1U_2$ the product
$t=2\operatorname{Sc}(\vec\Upsilon_1\vec\Upsilon_2)$ twice the scalar part of the product of the nilpotent parts
$s=\vec\Upsilon_1\cdot\vec\Upsilon_2=-t/2$ the pairing of the nilpotents
$p_T(\lambda)=\lambda^2-(2+t)\lambda+1$ the characteristic polynomial of $T$
$\Delta=t(t+4)=-4B^2(T)$ the discriminant
$Q_t=1+t/2$ the scalar part of $T$
$B^2(T)=-t(t+4)/4=s(2-s)$ the vector square of $T$
$2\operatorname{Sc}$ the doubled scalar part, $2\operatorname{Sc}\tilde Q=2Q_t$

Further Reading