The Complete Polar Decomposition of an N-Regular Biquaternion
Introduction
The general polar decomposition writes every biquaternion as a Hermitian positive factor times a partial isometry, and it is the subject of The Polar Decomposition of a Biquaternion:
$$ \tilde Q=\tilde Q_H\,\tilde Q_0 $$
On the $\natural$-regular elements, those with nonzero quaternionic norm, the pair sharpens: the Hermitian factor is positive definite and the isometry is $*$-unitary. The pair has two factors, and each factor splits again into two. This article takes the four apart.
The first split is the modulus against the unit. A $\natural$-regular element is a modulus times a unit of norm one,
$$ \tilde Q=\varrho\,e^{i\theta_{\mathrm{pha}}}\,\tilde U,\qquad \varrho^{2}e^{2i\theta_{\mathrm{pha}}}=\langle\tilde Q\rangle_{\natural},\qquad \langle\tilde U\rangle_{\natural}=1, $$
where $\varrho>0$ and $\varrho\,e^{i\theta_{\mathrm{pha}}}$ is the square root of the quaternionic norm taken on the branch $\theta_{\mathrm{pha}}\in(-\pi/2,\pi/2]$ fixed in §The Two Factors.
The second split separates the modulus into a scale and a phase, and the unit into a boost and a rotor:
$$ \varrho\,e^{i\theta_{\mathrm{pha}}}=\varrho\cdot e^{i\theta_{\mathrm{pha}}},\qquad \tilde U=\tilde U_{\mathrm b}\,\tilde u_{\mathrm r} $$
The four factors are the scale $\varrho>0$, the central phase $e^{i\theta_{\mathrm{pha}}}$ of modulus one on the branch $\theta_{\mathrm{pha}}\in(-\pi/2,\pi/2]$, the boost $\tilde U_{\mathrm b}$, Hermitian positive definite of norm one, and the rotor $\tilde u_{\mathrm r}$, a unit real quaternion. Read as a word,
$$ \boxed{\;\tilde Q=\varrho\,e^{i\theta_{\mathrm{pha}}}\,\tilde U_{\mathrm b}\,\tilde u_{\mathrm r}\;} $$
The name. This four-factor word is the complete polar decomposition of the biquaternion. Complete names the refinement and not the domain: the word carries all four factors at once, one per slot, $1+1+3+3$ real parameters, and each of the three pairings is a regrouping of it rather than a rival reading of the element. The domain is the one of the pair it refines, the $\natural$-regular set, where the boost is positive definite and the central phase lies in its branch; the word does not extend to the null cone, where the general decomposition keeps two factors and the $*$-unitary factor becomes a partial isometry (The Polar Decomposition of an N-Degenerate Biquaternion). The corpus cites the word under this name.
The word has a second face, in which the boost and the rotor are written as exponentials. Each of the two unit factors is the exponential of one real parameter along one real unit axis, and the two factors differ in the nature of the axis.
The boost $\tilde U_{\mathrm b}$ is the Hermitian positive definite factor of norm one. Its axis $\vec u_{\mathrm b}$ is a real unit vector and its parameter is the rapidity $\theta_{\mathrm{rpd}}\ge0$, and the imaginary unit in front of the axis makes it a Lorentz boost,
$$ \tilde U_{\mathrm b}=\exp\Bigl(i\frac{\theta_{\mathrm{rpd}}}{2}\vec u_{\mathrm b}\Bigr). $$
The rotor $\tilde u_{\mathrm r}$ is a unit real quaternion, an element of $SU(2)$. Its axis $\vec u_{\mathrm r}$ is a real unit vector and its parameter is the rotor angle $\theta_{\mathrm{axr}}\in[0,2\pi]$, and the axis enters without an imaginary unit, so the exponential is an ordinary cosine and sine and the factor is a spatial rotation,
$$ \tilde u_{\mathrm r}=\exp\Bigl(\frac{\theta_{\mathrm{axr}}}{2}\vec u_{\mathrm r}\Bigr). $$
The two exponentials are the two unit factors of the word, and the word reads
$$ \boxed{\;\tilde Q=\varrho\,e^{i\theta_{\mathrm{pha}}}\,\exp\Bigl(i\frac{\theta_{\mathrm{rpd}}}{2}\vec u_{\mathrm b}\Bigr)\exp\Bigl(\frac{\theta_{\mathrm{axr}}}{2}\vec u_{\mathrm r}\Bigr)\;} $$
The boxed equality is the complete polar decomposition of the biquaternion in its exponential face. It holds on the $\natural$-regular elements, those with nonzero quaternionic norm $\langle\tilde Q\rangle_{\natural}\neq0$, equivalently on the invertible biquaternions. The scale and the phase are the logarithm and the argument of the modulus. The four factors regroup in three ways, one per pairing, and the table below gives the three usual decompositions.
| reading | form | pairing | factors | real dimensions | available when |
|---|---|---|---|---|---|
| Cartan pair | $\tilde Q=\tilde Q_H\tilde Q_0=(\varrho\tilde U_{\mathrm b})(e^{i\theta_{\mathrm{pha}}}\tilde u_{\mathrm r})$ | Cartan Hermitian | $2$ | $4+4$ | $\langle\tilde Q\rangle_{\natural}\neq0$ |
| modulus–unit | $\tilde Q=\varrho\,e^{i\theta_{\mathrm{pha}}}\tilde U=(\varrho e^{i\theta_{\mathrm{pha}}})(\tilde U_{\mathrm b}\tilde u_{\mathrm r})$ | Hamilton root | $2$ | $2+6$ | $\langle\tilde Q\rangle_{\natural}\neq0$ and $B^{2}\neq0$ |
| quaternion | $\tilde Q=(\varrho\tilde u_{\mathrm r})(e^{i\theta_{\mathrm{pha}}}\,\tilde u_{\mathrm r}^{-1}\tilde U_{\mathrm b}\tilde u_{\mathrm r})$ | quaternion | $2$ | $4+4$ | $\langle\tilde Q\rangle_{\natural}\neq0$ and the real-coefficient part is nonzero |
The closed forms of §The Four Factors compute the same four factors; the exponents display their ranges, and §The Result in Terms of the Coordinates gives the seven data that determine them. The four exponents do not add. Their sum is a formal exponent whose exponential misses the element off the normal set, and §The Logarithm computes the defect.
One theory, two strata. The existence of the word is a property of the group of invertibles alone. The second invariant, the vector form $B^{2}=Q_x^{2}+Q_y^{2}+Q_z^{2}$, decides only its shape. Off the wall $B^{2}\neq0$ the element has two distinct eigenvalues and an eigenbasis, and the spectral reading is available. On the wall $B^{2}=0$ the two eigenvalues coalesce, the eigenbasis disappears, and the unit factor is $\pm$unipotent. The word survives the passage to the wall, because it is a decomposition of the group and not of the element. The spectrum does not survive. That contrast is the spine of the article, and it is the reason the two cases are one theory and not two.
The article proves the sharpening and computes the modulus and the unit factor (§The Two Factors), splits them into the four factors with their exponentials and ranges (§The Four Factors), works the component form and the parameter count (§The Component Form), proves existence and uniqueness twice, geometrically and algebraically (§Existence and Uniqueness), fixes the domain and the normal form at the wall (§The Domain), computes the seven data in the coordinates (§The Result in Terms of the Coordinates), reads the word against the spectral decomposition (§The Spectral Reading), proves the rigidity of the unipotent factor and the closed form of the wall (§The Unipotent Unit Factor), separates the two strata of that wall (§The Two Strata), reads the group of invertibles with its coordinates (§The Group of Invertibles), separates the two halves of the algebra (§The Two Halves), gives the three pairings (§The Three Pairings), treats the logarithm and its defect (§The Logarithm), the group law (§The Group Law), the matrix picture (§The Matrix Picture) and the spinor module (§The Operator Reading: the Twist), closes with the algorithm, the worked example and the two walls of the sector (§The Algorithm, §Worked Example, §The Two Walls).
The conventions and the symbols are those of Algebra and Element Conventions, N-Regularity and N-Degeneracy in the Biquaternion Algebra and The Four Forms and the Four Algebraic Norms over the Biquaternion Algebra; the symbols are collected in §Summary of Notation. The product is the plain product, the conjugation is ${}^{*}=\bar{\cdot}\circ{}^{\natural}$, and the norm is the quaternionic norm $\langle\tilde Q\rangle_{\natural}$. The exponential and the logarithm are those of Biquaternion Elementary Functions, the eigenbasis is that of The Eigenbasis of a Spectrally Regular Biquaternion, and the numerical cases are in Worked Examples of the Complete Polar Decomposition of an N-Regular Biquaternion.
The Two Factors
Let $\tilde Q$ be $\natural$-regular, $\langle\tilde Q\rangle_{\natural}\neq0$. Then $\tilde Q$ is invertible (N-Regularity and N-Degeneracy in the Biquaternion Algebra, §The $\natural$-regular case: The Group of Invertibles), and the two factors of the general decomposition are
$$ \tilde Q_H=\bigl(\tilde Q\tilde Q^{*}\bigr)^{1/2},\qquad \tilde Q_0=\tilde Q_H^{-1}\tilde Q . $$
Proposition (the sharpened factors). If $\langle\tilde Q\rangle_{\natural}\neq0$, then $\tilde Q_H$ is positive definite and $\tilde Q_0$ is $*$-unitary.
Proof. The element $\tilde Q\tilde Q^{*}$ is Hermitian and positive definite, being the product of an invertible element and its Hermitian conjugate; its unique square root in the cone is positive definite, so $\tilde Q_H$ is positive definite and invertible. Then
$$ \tilde Q_0\tilde Q_0^{*}=\tilde Q_H^{-1}\tilde Q\tilde Q^{*}\tilde Q_H^{-1}=\tilde Q_H^{-1}\tilde Q_H^{2}\tilde Q_H^{-1}=e_t, $$
and symmetrically $\tilde Q_0^{*}\tilde Q_0=e_t$.
The unitarity is the sharpening: the partial isometry of the general case is now invertible, and its two projections are both $e_t$.
The modulus. The norm is a complex number, and it vanishes exactly on the zero-divisor cone $\mathcal N$ (N-Regularity and N-Degeneracy in the Biquaternion Algebra). A modulus is a square root of the norm,
$$ \varrho^{2}e^{2i\theta_{\mathrm{pha}}}=\langle\tilde Q\rangle_{\natural}, $$
and the two roots differ by a sign, so the root is not defined by the norm alone. The branch is fixed once and for all. With $a=\operatorname{Re}\langle\tilde Q\rangle_{\natural}$ and $b=\operatorname{Im}\langle\tilde Q\rangle_{\natural}$, put
$$ \varrho\,e^{i\theta_{\mathrm{pha}}}=\sqrt{\frac{\lvert\langle\tilde Q\rangle_{\natural}\rvert+a}{2}}+i\,\operatorname{sgn}(b)\sqrt{\frac{\lvert\langle\tilde Q\rangle_{\natural}\rvert-a}{2}},\qquad \operatorname{sgn}(0)=+1 . $$
The root lies in the principal half-plane
$$ \mathcal H=\{\varrho\,e^{i\theta_{\mathrm{pha}}}\in\mathbb C:\operatorname{Re}(\varrho\,e^{i\theta_{\mathrm{pha}}})>0\}\cup\{it:t>0\}, $$
and the other root $-\varrho\,e^{i\theta_{\mathrm{pha}}}$ lies outside it. On $\mathcal H$ the root is a function of the norm and is unique. The sign convention at $b=0$ places a negative real norm on the positive imaginary axis, and the root is continuous away from the negative real axis, where the branch cut lies. The branch is the only convention of the decomposition.
The scale and the phase. The polar coordinates of the modulus on $\mathcal H$ are
$$ \varrho=\sqrt{\lvert\langle\tilde Q\rangle_{\natural}\rvert}>0,\qquad \theta_{\mathrm{pha}}=\arg(\varrho\,e^{i\theta_{\mathrm{pha}}}),\qquad \lvert e^{i\theta_{\mathrm{pha}}}\rvert=1 . $$
The phase factor lies in the half of the unit circle cut by the branch, and the angle lies in the strip that parametrises it,
$$ e^{i\theta_{\mathrm{pha}}}\in\{z\in\mathbb C:\lvert z\rvert=1,\ \operatorname{Re}z>0\}\cup\{i\},\qquad \theta_{\mathrm{pha}}\in(-\pi/2,\pi/2] . $$
The angle vanishes on the positive real axis, takes the value $\pi/2$ on the positive imaginary axis, and never takes the values $\pi$ or $-\pi/2$.
The unit factor. Put
$$ \tilde U=\frac{\tilde Q}{\varrho\,e^{i\theta_{\mathrm{pha}}}}. $$
Since $\varrho\,e^{i\theta_{\mathrm{pha}}}$ is central and $\varrho^{2}e^{2i\theta_{\mathrm{pha}}}=\langle\tilde Q\rangle_{\natural}$, the norm is multiplicative and
$$ \langle\tilde U\rangle_{\natural}=\frac{\langle\tilde Q\rangle_{\natural}}{\varrho^{2}e^{2i\theta_{\mathrm{pha}}}}=1 . $$
The unit factor lies on the norm-one set
$$ \mathbb B^{\times}_{1}=\{\tilde U\in\mathbb B:\langle\tilde U\rangle_{\natural}=1\}, $$
which is a group: the product of two norm-one elements has norm one, and the inverse of a norm-one element is its conjugate, $\tilde U^{-1}=\tilde U^{\natural}$. Under the matrix model the norm is the determinant, so the group is the special linear group,
$$ \boxed{\;\mathbb B^{\times}_{1}\cong SL_2(\mathbb C)\;} $$
of complex dimension three and real dimension six (N-Regularity and N-Degeneracy in the Biquaternion Algebra, §The $\natural$-regular case: The Group of Invertibles).
The group is not compact. It contains the Hermitian positive elements of norm one, and those are unbounded: the element $c\,e_t+i\sqrt{c^{2}-1}\,\vec e_x$ has norm one for every $c\ge1$. This is the sharpest difference from the quaternion case, where the norm-one set is the sphere $S^{3}$. The biquaternion unit factor has three compact directions and three non-compact ones, and the split into boost and rotor separates them.
The phrase unit factor names the norm-one elements. It is not the group of invertibles $\mathbb B^{\times}$, the wider set of invertible elements; the word unit in the phrase means norm one and never invertibility, as in Norm and Unitary Conventions, §The Unitary Convention. The group of invertibles is $\mathbb B\setminus\mathcal N$, of real dimension eight, and the norm-one group is a hypersurface inside it.
The count. A biquaternion carries eight real parameters. The modulus carries two, the scale and the phase. The unit factor is constrained by the single complex equation $\langle\tilde U\rangle_{\natural}=1$, so it retains six real degrees of freedom. Thus
$$ 8=2+6 . $$
No coordinate has been chosen on the unit factor: the equation is one complex condition, and the six parameters are the dimension of $SL_2(\mathbb C)$, not a chart.
The Four Factors
The two factors of the previous section are each the product of two.
The split of the modulus
The modulus splits into the scale and the phase,
$$ \varrho\,e^{i\theta_{\mathrm{pha}}},\qquad \varrho>0,\qquad \theta_{\mathrm{pha}}\in(-\pi/2,\pi/2] . $$
The two are central and commute with every element. They carry the norm between them,
$$ \varrho^{2}=\lvert\langle\tilde Q\rangle_{\natural}\rvert,\qquad e^{2i\theta_{\mathrm{pha}}}=\frac{\langle\tilde Q\rangle_{\natural}}{\lvert\langle\tilde Q\rangle_{\natural}\rvert}, $$
so the scale is the modulus of the norm and the phase is its unimodular direction, read through the branch.
The split of the unit factor
The boost. The boost of $\tilde Q$ is
$$ \tilde U_{\mathrm b}=\bigl(\tilde U\tilde U^{*}\bigr)^{1/2}. $$
It is Hermitian, $\tilde U_{\mathrm b}=\tilde U_{\mathrm b}^{*}$, positive definite, and of norm one. Indeed $\tilde U\tilde U^{*}$ is Hermitian positive definite, because $\tilde U$ is invertible, and its norm is $\langle\tilde U\rangle_{\natural}\overline{\langle\tilde U\rangle_{\natural}}=1$; the square root in the cone is therefore Hermitian positive definite of norm one. The boost is the Hermitian positive factor of the polar decomposition of the unit factor on the left,
$$ \tilde U=\tilde U_{\mathrm b}\,\tilde u_{\mathrm r}. $$
Closed form. Put $s=\operatorname{Sc}(\tilde U\tilde U^{*})$. The Cayley–Hamilton identity of $\tilde U\tilde U^{*}$ is $X^{2}-2sX+e_t=0$, because its determinant is one, and the square root is
$$ \boxed{\;\tilde U_{\mathrm b}=\frac{\tilde U\tilde U^{*}+e_t}{\sqrt{2(1+s)}}\;} $$
with inverse
$$ \tilde U_{\mathrm b}^{-1}=\frac{(2s+1)e_t-\tilde U\tilde U^{*}}{\sqrt{2(1+s)}} . $$
The scalar $s$ is real and at least one: in the matrix model the two eigenvalues of $\tilde U\tilde U^{*}$ are positive and of product one, so their sum is at least two. The denominators are therefore real and positive. The boost is the non-compact half of the unit factor; the set of Hermitian positive elements of norm one is the hyperbolic space $H^{3}$, of real dimension three.
The rotor. The rotor of $\tilde Q$ is
$$ \tilde u_{\mathrm r}=\tilde U_{\mathrm b}^{-1}\tilde U . $$
It is $*$-unitary, because
$$ \tilde u_{\mathrm r}\tilde u_{\mathrm r}^{*}=\tilde U_{\mathrm b}^{-1}\tilde U\tilde U^{*}\tilde U_{\mathrm b}^{-1}=\tilde U_{\mathrm b}^{-1}\tilde U_{\mathrm b}^{2}\tilde U_{\mathrm b}^{-1}=e_t, $$
and it has norm one, by multiplicativity. Its coefficients are real. Write $\tilde u_{\mathrm r}=\tilde q_1+i\tilde q_2$ with $\tilde q_1,\tilde q_2$ real quaternions. The identity $\tilde u_{\mathrm r}\tilde u_{\mathrm r}^{*}=e_t$ gives $\lvert\tilde q_1\rvert^{2}+\lvert\tilde q_2\rvert^{2}=1$ in scalar parts, and the identity $\langle\tilde u_{\mathrm r}\rangle_{\natural}=1$ gives $\lvert\tilde q_1\rvert^{2}-\lvert\tilde q_2\rvert^{2}=1$. Hence $\lvert\tilde q_2\rvert=0$ and $\tilde q_2=0$, so the rotor is a real quaternion. It is therefore a unit real quaternion, an element of
$$ SU(2)=\{\tilde u\in\mathbb H_{\mathbb B}:\langle\tilde u\rangle_{\natural}=1\}=S^{3}, $$
the compact half of the unit factor. Its closed form is
$$ \tilde u_{\mathrm r}=\frac{(2s+1)e_t-\tilde U\tilde U^{*}}{\sqrt{2(1+s)}}\;\tilde U . $$
Both conditions are needed in the reality argument: unitarity alone does not force the coefficients to be real, and the norm-one condition alone does not force the element to be $*$-unitary. Together they do.
The word
Combining the two splits, with each unit factor written as an exponential of one real parameter,
$$ \boxed{\; \tilde Q=\varrho\,e^{i\theta_{\mathrm{pha}}}\,\tilde U_{\mathrm b}\,\tilde u_{\mathrm r} =\varrho\,e^{i\theta_{\mathrm{pha}}}\exp\Bigl(i\frac{\theta_{\mathrm{rpd}}}{2}\vec u_{\mathrm b}\Bigr)\exp\Bigl(\frac{\theta_{\mathrm{axr}}}{2}\vec u_{\mathrm r}\Bigr) \;} $$
with
$$ \varrho>0,\qquad \theta_{\mathrm{pha}}\in\Bigl(-\frac{\pi}{2},\frac{\pi}{2}\Bigr],\qquad \theta_{\mathrm{rpd}}\ge0,\qquad \theta_{\mathrm{axr}}\in[0,2\pi], $$
and $\vec u_{\mathrm b},\vec u_{\mathrm r}$ real unit vectors, the boost axis and the rotor axis.
The closed forms and the exponentials are the same four factors, and the table gives both.
| factor | closed form | exponent | range | real dimension | type |
|---|---|---|---|---|---|
| $\varrho$ | $\varrho=\lvert\langle\tilde Q\rangle_{\natural}\rvert^{1/2}$ | $\ln \varrho$ | $\varrho>0$ | $1$ | positive real, central |
| $e^{i\theta_{\mathrm{pha}}}$ | $e^{i\theta_{\mathrm{pha}}}=\varrho\,e^{i\theta_{\mathrm{pha}}}/\varrho$ | $i\theta_{\mathrm{pha}}$ | $\theta_{\mathrm{pha}}\in(-\frac{\pi}{2},\frac{\pi}{2}]$ | $1$ | unit complex scalar, central |
| $\tilde U_{\mathrm b}$ | $(\tilde U\tilde U^{*})^{1/2}$, or $\cosh\frac{\theta_{\mathrm{rpd}}}{2}e_t+i\sinh\frac{\theta_{\mathrm{rpd}}}{2}\vec u_{\mathrm b}$ | $i\frac{\theta_{\mathrm{rpd}}}{2}\vec u_{\mathrm b}$ | $\theta_{\mathrm{rpd}}\ge0$, $\vec u_{\mathrm b}\in S^{2}$ | $3$ | Hermitian positive definite, norm one |
| $\tilde u_{\mathrm r}$ | $\tilde U_{\mathrm b}^{-1}\tilde U$, or $\cos\frac{\theta_{\mathrm{axr}}}{2}e_t+\sin\frac{\theta_{\mathrm{axr}}}{2}\vec u_{\mathrm r}$ | $\frac{\theta_{\mathrm{axr}}}{2}\vec u_{\mathrm r}$ | $\theta_{\mathrm{axr}}\in[0,2\pi]$, $\vec u_{\mathrm r}\in S^{2}$ | $3$ | unit real quaternion, $SU(2)$ |
The four dimensions add to eight, $8=1+1+3+3$, the real dimension of the algebra. The scale and the phase are central and commute with the other two; the boost and the rotor do not commute, and their order is part of the definition: the boost is the left Hermitian factor, so the word is boost then rotor. The other order gives a different pair, $\tilde u_{\mathrm r}\tilde U_{\mathrm b}=(\tilde u_{\mathrm r}\tilde U_{\mathrm b}\tilde u_{\mathrm r}^{-1})\tilde u_{\mathrm r}$, equal to the first exactly when the boost and the rotor commute.
The two exponents of the unit factors lie in the two summands of the Cartan decomposition of the algebra, $\mathfrak{p}=i\{\text{real pure vectors}\}$ and $\mathfrak{k}=\{\text{real pure vectors}\}$, and the logarithm of the modulus lies on the two real lines $\mathbb{R}e_t$ and $i\mathbb{R}e_t$. §The Logarithm names the four summands.
The two readings. The four factors regroup into the Cartan pair and into the modulus–unit pair:
$$ \boxed{\;\tilde Q_H=\varrho\,\tilde U_{\mathrm b},\qquad \tilde Q_0=e^{i\theta_{\mathrm{pha}}}\,\tilde u_{\mathrm r},\qquad \varrho\,e^{i\theta_{\mathrm{pha}}}=\varrho e^{i\theta_{\mathrm{pha}}},\qquad \tilde U=\tilde U_{\mathrm b}\tilde u_{\mathrm r}.\;} $$
Indeed $\tilde Q\tilde Q^{*}=\varrho^{2}\tilde U_{\mathrm b}^{2}$, since $e^{i\theta_{\mathrm{pha}}}$ and $\tilde u_{\mathrm r}$ are $*$-unitary, so the Hermitian square root is $\varrho\tilde U_{\mathrm b}$; and $\tilde Q_H^{-1}\tilde Q=e^{i\theta_{\mathrm{pha}}}\tilde u_{\mathrm r}$. The Hermitian factor carries the scale and the boost, and the $*$-unitary factor the phase and the rotor. The first is Hermitian positive definite, the second $*$-unitary, and the two are the Cartan Hermitian pairing of §The Three Pairings. The modulus–unit pair brackets the same word the other way, and the two bracketings are the two two-factor readings of the four factors.
The branch cut
The four separate factors are not continuous across the negative real axis of the norm, although the pair $(\tilde Q_H,\tilde Q_0)$ is. At a norm that crosses the cut, $\varrho\,e^{i\theta_{\mathrm{pha}}}$ jumps from $i\sqrt{\lvert N\rvert}$ to $-i\sqrt{\lvert N\rvert}$, so $\theta_{\mathrm{pha}}$ advances by $\pi$ and $\tilde u_{\mathrm r}$ changes sign while their product does not. The scale, the boost, the Hermitian factor and the $*$-unitary factor are continuous there; the phase and the rotor are not. The four-factor word is therefore a decomposition of the element into four continuous functions off the cut, and into four functions that jump together on it.
The Component Form
Let
$$ c_{\mathrm b}=\cosh\frac{\theta_{\mathrm{rpd}}}{2}, \qquad s_{\mathrm b}=\sinh\frac{\theta_{\mathrm{rpd}}}{2}, \qquad c_{\mathrm r}=\cos\frac{\theta_{\mathrm{axr}}}{2}, \qquad s_{\mathrm r}=\sin\frac{\theta_{\mathrm{axr}}}{2}, \qquad c=\vec u_{\mathrm b}\cdot\vec u_{\mathrm r}. $$
Multiplying the two exponentials in the order of the word gives
$$ \boxed{\; \tilde U=\bigl(c_{\mathrm b}c_{\mathrm r}-i s_{\mathrm b}s_{\mathrm r}c\bigr)e_t +i s_{\mathrm b}c_{\mathrm r}\,\vec u_{\mathrm b} +c_{\mathrm b}s_{\mathrm r}\,\vec u_{\mathrm r} +i s_{\mathrm b}s_{\mathrm r}\,\vec u_{\mathrm b}\times\vec u_{\mathrm r} \;} $$
The four terms are the scalar term, corrected by the angle between the axes, the boost axis, the rotor axis and the cross term. The cross term is present exactly when both angles are nonzero and the axes are neither parallel nor antiparallel. It is the term that a sum of the two exponents cannot produce, because the sum contains only the two axes and their linear combinations.
The scalar part is
$$ \operatorname{Sc}\tilde U=c_{\mathrm b}c_{\mathrm r}-i s_{\mathrm b}s_{\mathrm r}c, $$
and the vector part is the remaining three terms. The norm of $\tilde U$ is one, so
$$ \bigl(\operatorname{Sc}\tilde U\bigr)^2+\vec U\cdot\vec U=1, $$
and the eigenvalues of $\tilde U$ are
$$ \mu_\pm=\operatorname{Sc}\tilde U\pm i\sqrt{1-\bigl(\operatorname{Sc}\tilde U\bigr)^2}. $$
The Parameter Count
The data are
$$ \underbrace{1}_{\ln \varrho} +\underbrace{1}_{\theta_{\mathrm{pha}}} +\underbrace{1+2}_{\theta_{\mathrm{rpd}},\vec u_{\mathrm b}} +\underbrace{1+2}_{\theta_{\mathrm{axr}},\vec u_{\mathrm r}} =8, $$
the real dimension of the group of invertibles. The word is a coordinate system on that group, cut by the ranges of the two angles and of the phase. The two axes are points of $S^2$.
Existence and Uniqueness
Theorem (the complete polar decomposition). Every $\tilde Q\in\Sigma_{11}$ has a unique writing
$$ \tilde Q=\varrho\,e^{i\theta_{\mathrm{pha}}}\,\tilde U_{\mathrm b}\,\tilde u_{\mathrm r}, $$
with $\varrho>0$, $\theta_{\mathrm{pha}}\in(-\frac{\pi}{2},\frac{\pi}{2}]$, $\tilde U_{\mathrm b}$ Hermitian positive of norm one, and $\tilde u_{\mathrm r}$ a unit real quaternion. Consequently it has the exponential word above; the axes are unique when $\theta_{\mathrm{rpd}}>0$ and $\theta_{\mathrm{axr}}\notin\{0,2\pi\}$.
Proof. The modulus $\varrho\,e^{i\theta_{\mathrm{pha}}}$ is the principal square root of $N$, and $\tilde U=\tilde Q/(\varrho\,e^{i\theta_{\mathrm{pha}}})$ has norm one. The matrix image of $\tilde U$ lies in $SL(2,\mathbb{C})$. The global Cartan decomposition
$$ SL(2,\mathbb{C})=P\cdot K, \qquad P=\{\text{positive Hermitian of determinant }1\}, \qquad K=SU(2), $$
gives a unique factorization $\tilde U=\tilde U_{\mathrm b}\tilde u_{\mathrm r}$ with $\tilde U_{\mathrm b}\in P$ and $\tilde u_{\mathrm r}\in K$. A positive Hermitian determinant-one matrix is the exponential of a unique Hermitian traceless matrix, which is $i\frac{\theta_{\mathrm{rpd}}}{2}\vec u_{\mathrm b}$ with $\theta_{\mathrm{rpd}}\ge0$ and a unit axis. A $*$-unitary determinant-one matrix is the exponential of an anti-Hermitian traceless matrix, which is $\frac{\theta_{\mathrm{axr}}}{2}\vec u_{\mathrm r}$ with $\theta_{\mathrm{axr}}\in[0,2\pi]$. The endpoint $\theta_{\mathrm{axr}}=2\pi$ is the central rotor $-e_t$, where the axis is undefined. The scale and the phase are the polar coordinates of $\varrho\,e^{i\theta_{\mathrm{pha}}}$, unique in the branch.
The proof uses only the invertibility of $\tilde Q$; the spectral condition $B^2\neq0$ enters only through the domain, not through the existence. This is why the theorem holds on the wall as it stands, with the same proof and the same ranges.
The Algebraic Proof
The same statement follows from the modulus–unit theorem without the matrix model.
Proposition (the modulus–unit form). Every $\natural$-regular biquaternion has exactly one modulus–unit decomposition with modulus in $\mathcal H$, and exactly one four-factor word with that modulus.
Proof (existence). Let $\varrho\,e^{i\theta_{\mathrm{pha}}}$ be the principal square root of $\langle\tilde Q\rangle_{\natural}$ in $\mathcal H$, and put $\tilde U=\tilde Q/(\varrho\,e^{i\theta_{\mathrm{pha}}})$. Then $\langle\tilde U\rangle_{\natural}=1$, so $\tilde U$ is invertible and the boost and rotor of §The Four Factors are defined. With $\varrho=\lvert\langle\tilde Q\rangle_{\natural}\rvert^{1/2}$ and $e^{i\theta_{\mathrm{pha}}}$ the unit factor of the norm, the word follows by multiplication.
Proof (uniqueness). Suppose $\tilde Q=\varrho\,e^{i\theta_{\mathrm{pha}}}\tilde U=\varrho\,e^{i\theta_{\mathrm{pha}}}{}'\tilde U'$ with $\varrho\,e^{i\theta_{\mathrm{pha}}},\varrho\,e^{i\theta_{\mathrm{pha}}}{}'\in\mathcal H$ and $\langle\tilde U\rangle_{\natural}=\langle\tilde U'\rangle_{\natural}=1$. Taking norms gives $\varrho^{2}e^{2i\theta_{\mathrm{pha}}}=\varrho\,e^{i\theta_{\mathrm{pha}}}{}'{}^{2}$, so $\varrho\,e^{i\theta_{\mathrm{pha}}}{}'=\pm\varrho\,e^{i\theta_{\mathrm{pha}}}$; both lie in $\mathcal H$, which contains no pair of opposite points, so $\varrho\,e^{i\theta_{\mathrm{pha}}}{}'=\varrho\,e^{i\theta_{\mathrm{pha}}}$. Dividing, $\tilde U'=\tilde U$. The polar decomposition of $\tilde U$ is unique: $\tilde U\tilde U^{*}$ has exactly one Hermitian positive square root, hence $\tilde U_{\mathrm b}$ is unique, and then $\tilde u_{\mathrm r}=\tilde U_{\mathrm b}^{-1}\tilde U$ is unique.
The word exists exactly on the $\natural$-regular elements. On the null cone the norm vanishes, the modulus is zero, and no modulus–unit form exists; the two-factor decomposition survives and is the subject of The Polar Decomposition of an N-Degenerate Biquaternion.
The uniqueness is a statement about the factors, not about a coordinate on the boost or on the rotor. Two different charts of $H^{3}$ assign different parameters to the same boost, and two different charts of $S^{3}$ assign different parameters to the same rotor; the factors themselves are fixed.
The Degeneracies
The boost axis is undefined at $\theta_{\mathrm{rpd}}=0$, where the boost is the identity. The rotor axis is undefined at $\theta_{\mathrm{axr}}=0$, where the rotor is the identity, and at $\theta_{\mathrm{axr}}=2\pi$, where the rotor is $-e_t$ for every axis. These are the usual coordinate singularities of polar coordinates; the element is smooth there and the word is unique.
| Parameter | Range | Unique? |
|---|---|---|
| $\varrho$ | $(0,\infty)$ | yes |
| $\theta_{\mathrm{pha}}$ | $(-\frac{\pi}{2},\frac{\pi}{2}]$ | yes: the branch |
| $\theta_{\mathrm{rpd}}$ | $[0,\infty)$ | yes |
| $\vec u_{\mathrm b}$ | $S^2$ | no: undefined at $\theta_{\mathrm{rpd}}=0$ |
| $\theta_{\mathrm{axr}}$ | $[0,2\pi]$ | yes |
| $\vec u_{\mathrm r}$ | $S^2$ | no: undefined at $\theta_{\mathrm{axr}}\in\{0,2\pi\}$ |
At a degeneracy the word is still unique as an element of the group, but the coordinate chart is singular. The singularity is not a failure of the decomposition; it is the price of the polar coordinates.
The Domain
The domain is cut by two invariants. One decides whether the element is invertible, the other decides the shape of the word. The first fixes the sector and the second fixes its two strata.
The Two Invariants
Let
$$ \tilde Q=Q_te_t+Q_x\vec e_x+Q_y\vec e_y+Q_z\vec e_z, \qquad N=\langle\tilde Q\rangle_{\natural}=Q_t^2+Q_x^2+Q_y^2+Q_z^2, \qquad B^2=Q_x^2+Q_y^2+Q_z^2. $$
The number $N$ is the reduced product $\tilde Q\tilde Q^{\natural}$, and $B^2$ is the quadratic form of the vector part. The two are complex numbers. The spectrally regular sector is
$$ \Sigma_{11}=\{\tilde Q\in\mathbb{B}:\ N\neq0,\ B^2\neq0\}. $$
The name is the one of The Spectrum of a Biquaternion: the element is N-regular when $N\neq0$, and spectrally regular when $B^2\neq0$. The two conditions are independent, and their four combinations are the four sectors of the algebra. This article treats the two combinations with $N\neq0$: the compact notation below writes $11$ for the regular case and $10$ for the wall. The combinations with $N=0$ are the boundary, treated in §The Two Walls.
Five Descriptions
The sector has five equivalent descriptions.
Proposition (the descriptions of $\Sigma_{11}$). For $\tilde Q\in\mathbb{B}$ the following are equivalent.
- $N\neq0$ and $B^2\neq0$.
- The two eigenvalues $\lambda_\pm=Q_t\pm iB$ are distinct and nonzero.
- The minimal polynomial $X^2-2Q_tX+N$ is squarefree with nonzero constant term.
- The element is invertible and its spectrum has two distinct points.
- With $\varrho\,e^{i\theta_{\mathrm{pha}}}$ the principal square root of $N$ and $\tilde U=\tilde Q/(\varrho\,e^{i\theta_{\mathrm{pha}}})$, the scalar part of $\tilde U$ is not $\pm1$.
Proof. The characteristic polynomial is $X^2-2Q_tX+N$. Its discriminant is $4(Q_t^2-N)=-4B^2$, so it is squarefree exactly when $B^2\neq0$, and its constant term is nonzero exactly when $N\neq0$. The eigenvalues are $Q_t\pm iB$, distinct and nonzero under the same two conditions. The element is invertible because $N\neq0$. A scalar has $B^2=0$, but $B^2=0$ also allows the non-scalar Jordan elements; the spectral condition is the distinctness, not the non-scalarity. For the last item, $\operatorname{Sc}\tilde U=Q_t/(\varrho\,e^{i\theta_{\mathrm{pha}}})$, hence
$$ \bigl(\operatorname{Sc}\tilde U\bigr)^2=\frac{Q_t^2}{N}, \qquad B^2=N-Q_t^2=N\Bigl(1-\bigl(\operatorname{Sc}\tilde U\bigr)^2\Bigr), $$
so $B^2\neq0$ is equivalent to $\operatorname{Sc}\tilde U\notin\{1,-1\}$.
The last description separates the two nonzero-$N$ sectors: they are the two sides of the pair of hyperplanes $\operatorname{Sc}\tilde U=\pm1$ in the group of invertibles, and the wall is the union of the two hyperplanes.
Geometry
The sector is the complement of the union $NB^2=0$ of the null cone and the spectral cone. It is open and dense in $\mathbb{B}$, of complex dimension $4$ and real dimension $8$. It is not a subgroup: $e_t$ has $B^2=0$ and lies outside, and the scalar line $\mathbb{C}^{\times}e_t$ lies entirely outside. It contains the pure vectors with $B^2\neq0$, such as $\vec e_x+\vec e_y$, for which $Q_t=0$ and the two eigenvalues are opposite. It also contains every Hermitian element with $B^2<0$ and $N\neq0$, such as $\sqrt2\,e_t+i\vec e_x$.
The sector is connected, and it is not simply connected; the loop $t\mapsto e_t+\varepsilon e^{2\pi it}\vec e_x$ winds around the spectral cone. The topology is not used below beyond the parameter count.
The Normal Form at the Wall
Proposition (the normal form). Every $\tilde Q\in\Sigma_{10}$ has a unique writing
$$ \boxed{\tilde Q=Q_t\bigl(e_t+\vec\Upsilon\bigr)}, \qquad \vec\Upsilon^2=0, \qquad Q_t\neq0, $$
with $Q_t$ the scalar part and $\vec\Upsilon$ the vector part divided by $Q_t$.
Proof. Since $B^2=0$ the vector part is null, $\vec Q^2=-B^2=0$. Since $N=Q_t^2+B^2=Q_t^2$ and $N\neq0$, the scalar part is nonzero. Put $\vec\Upsilon=\vec Q/Q_t$; then $\vec\Upsilon^2=\vec Q^2/Q_t^2=0$, and $\tilde Q=Q_te_t+\vec Q=Q_t(e_t+\vec\Upsilon)$. The writing is unique because $Q_t$ is the scalar part.
The element is a scalar multiple of a unipotent element. The unipotent factor $e_t+\vec\Upsilon$ has norm one, because
$$ \langle e_t+\vec\Upsilon\rangle_{\natural}=1+\vec\Upsilon^2=1, $$
and it is invertible, with inverse $e_t-\vec\Upsilon$.
The Unit Factor at the Wall
The modulus is the principal square root of $N=Q_t^2$. Since $Q_t^2$ has the two square roots $\pm Q_t$, the modulus is $\varrho\,e^{i\theta_{\mathrm{pha}}}=\pm Q_t$, and the unit factor is
$$ \tilde U=\frac{\tilde Q}{\varrho\,e^{i\theta_{\mathrm{pha}}}} =\frac{Q_t}{\varrho\,e^{i\theta_{\mathrm{pha}}}}\bigl(e_t+\vec\Upsilon\bigr) =\varepsilon\bigl(e_t+\vec\Upsilon\bigr), \qquad \varepsilon=\frac{Q_t}{\varrho\,e^{i\theta_{\mathrm{pha}}}}=\pm1. $$
The sign $\varepsilon$ is fixed by the branch of the modulus. It is $+1$ when $\varrho\,e^{i\theta_{\mathrm{pha}}}=Q_t$ and $-1$ when $\varrho\,e^{i\theta_{\mathrm{pha}}}=-Q_t$. The unit factor is therefore $\pm$unipotent, and the sign is the only trace of the scalar part that survives in it.
The Result in Terms of the Coordinates
Setting. Let $\tilde Q=Q_te_t+Q_x\vec e_x+Q_y\vec e_y+Q_z\vec e_z$ with complex coordinates, and put
$$ N=Q_t^2+Q_x^2+Q_y^2+Q_z^2,\qquad B^2=Q_x^2+Q_y^2+Q_z^2,\qquad \lVert\tilde Q\rVert^2=|Q_t|^2+|Q_x|^2+|Q_y|^2+|Q_z|^2. $$
On the sector $N\neq0$ and $B^2\neq0$. Here $\operatorname{Arg}$ is the principal argument in $(-\pi,\pi]$, $\lVert\cdot\rVert$ is the Euclidean norm of a real three-vector, $\operatorname{Im}$ acts componentwise, and $(j,k,l)$ is a cyclic permutation of $(x,y,z)$.
Theorem (the word and its seven data). Every $\tilde Q$ of the sector is
$$ \boxed{\;\tilde Q=\varrho e^{i\theta_{\mathrm{pha}}}\,\tilde U_{\mathrm b}\,\tilde u_{\mathrm r} =\varrho e^{i\theta_{\mathrm{pha}}}\exp\Bigl(i\frac{\theta_{\mathrm{rpd}}}{2}\vec u_{\mathrm b}\Bigr)\exp\Bigl(\frac{\theta_{\mathrm{axr}}}{2}\vec u_{\mathrm r}\Bigr)\;} $$
with the two unit factors
$$ \boxed{\;\tilde U_{\mathrm b}=\cosh\frac{\theta_{\mathrm{rpd}}}{2}\,e_t+i\sinh\frac{\theta_{\mathrm{rpd}}}{2}\,\vec u_{\mathrm b},\qquad \tilde u_{\mathrm r}=\cos\frac{\theta_{\mathrm{axr}}}{2}\,e_t+\sin\frac{\theta_{\mathrm{axr}}}{2}\,\vec u_{\mathrm r}\;} $$
and with the seven data determined by $Q_t,Q_x,Q_y,Q_z$. Write $\vec Q=(Q_x,Q_y,Q_z)$.
$$ \boxed{\;\varrho=\sqrt{|N|}>0\;} $$
$$ \boxed{\;\theta_{\mathrm{pha}}=\frac12\operatorname{Arg}N\;} $$
$$ \boxed{\;\cosh\theta_{\mathrm{rpd}}=\frac{\lVert\tilde Q\rVert^2}{|N|},\qquad \theta_{\mathrm{rpd}}=\operatorname{arcosh}\frac{\lVert\tilde Q\rVert^2}{|N|}\;} $$
$$ \boxed{\;C=\sqrt{\frac{\lVert\tilde Q\rVert^2+|N|}{2}},\qquad S=\sqrt{\frac{\lVert\tilde Q\rVert^2-|N|}{2}},\qquad \cosh\frac{\theta_{\mathrm{rpd}}}{2}=\frac{C}{\sqrt{|N|}},\qquad \sinh\frac{\theta_{\mathrm{rpd}}}{2}=\frac{S}{\sqrt{|N|}}\;} $$
$$ \boxed{\;a_j=\frac{2}{|N|}\Bigl(\operatorname{Im}\bigl(\bar Q_tQ_j\bigr)-\operatorname{Im}\bigl(Q_k\bar Q_l\bigr)\Bigr),\qquad \vec a=(a_x,a_y,a_z),\qquad \vec u_{\mathrm b}=\frac{\vec a}{\lVert\vec a\rVert}\;} $$
$$ \boxed{\;\tilde U_{\mathrm b}=\frac{C\,e_t+iS\,\vec u_{\mathrm b}}{\sqrt{|N|}} =\frac{\sqrt{\lVert\tilde Q\rVert^2+|N|}\;e_t+i\sqrt{\lVert\tilde Q\rVert^2-|N|}\;\vec u_{\mathrm b}}{\sqrt{2|N|}}\;} $$
$$ \boxed{\;\tilde u_{\mathrm r}=\frac{e^{-i\theta_{\mathrm{pha}}}}{|N|}\bigl(C-iS\,\vec u_{\mathrm b}\bigr)\tilde Q\;} $$
$$ \boxed{\;\theta_{\mathrm{axr}}=2\arccos\frac{C\operatorname{Re}\bigl(e^{-i\theta_{\mathrm{pha}}}Q_t\bigr)-S\operatorname{Im}\bigl(e^{-i\theta_{\mathrm{pha}}}D\bigr)}{|N|},\qquad D=\vec u_{\mathrm b}\cdot\vec Q\;} $$
$$ \boxed{\;\vec u_{\mathrm r}=\frac{e^{-i\theta_{\mathrm{pha}}}\bigl(C\vec Q-iS(Q_t\vec u_{\mathrm b}+\vec u_{\mathrm b}\times\vec Q)\bigr)}{\bigl\lVert C\vec Q-iS(Q_t\vec u_{\mathrm b}+\vec u_{\mathrm b}\times\vec Q)\bigr\rVert}\;} $$
The boost axis is undefined when $\theta_{\mathrm{rpd}}=0$; the rotor axis is undefined when $\theta_{\mathrm{axr}}\in\{0,2\pi\}$. The vector $\mathrm{Vect}\tilde u_{\mathrm r}$ is the vector part of $\tilde u_{\mathrm r}$. At $\theta_{\mathrm{rpd}}=0$ the boost is $e_t$, and the box for $\tilde u_{\mathrm r}$ reduces to $\tilde u_{\mathrm r}=\tilde Q/(\varrho\,e^{i\theta_{\mathrm{pha}}})$.
Proof.
The modulus. By the modulus–unit theorem, $\varrho\,e^{i\theta_{\mathrm{pha}}}$ is the principal square root of $N$. Write $\varrho\,e^{i\theta_{\mathrm{pha}}}=\varrho e^{i\theta_{\mathrm{pha}}}$ with $\varrho>0$ and $\theta_{\mathrm{pha}}\in(-\pi/2,\pi/2]$. Then $\varrho^2=|\varrho\,e^{i\theta_{\mathrm{pha}}}|^2=|N|$ and $2\theta_{\mathrm{pha}}\equiv\operatorname{Arg}N\pmod{2\pi}$. The range of $\theta_{\mathrm{pha}}$ selects the representative, so $\varrho=\sqrt{|N|}$ and $\theta_{\mathrm{pha}}=\frac12\operatorname{Arg}N$.
The product $\tilde Q\tilde Q^*$. With $\tilde Q^*=\bar Q_t-\bar{\vec Q}$,
$$ \tilde Q\tilde Q^* =(Q_t+\vec Q)(\bar Q_t-\bar{\vec Q}) =\bigl(|Q_t|^2+|\vec Q|^2\bigr)e_t+\bar Q_t\vec Q-Q_t\bar{\vec Q}-\vec Q\times\bar{\vec Q}. $$
The scalar part is $\lVert\tilde Q\rVert^2=|Q_t|^2+|\vec Q|^2$. For pure quaternions $u,v$ one has $uv=-u\cdot v+u\times v$, with the bilinear dot product. Hence, componentwise,
$$ \bar Q_t\vec Q-Q_t\bar{\vec Q}=2i\operatorname{Im}\bigl(\bar Q_t\vec Q\bigr), \qquad \bigl(\vec Q\times\bar{\vec Q}\bigr)_j=2i\operatorname{Im}\bigl(Q_k\bar Q_l\bigr), $$
and therefore
$$ \tilde Q\tilde Q^*=\lVert\tilde Q\rVert^2\,e_t+i|N|\vec a, $$
with $\vec a$ the vector of the box.
The boost. The unit factor is $\tilde U=\tilde Q/(\varrho\,e^{i\theta_{\mathrm{pha}}})$, so
$$ \tilde U\tilde U^*=\frac{\tilde Q\tilde Q^*}{|\varrho\,e^{i\theta_{\mathrm{pha}}}|^2} =\frac{\lVert\tilde Q\rVert^2}{|N|}\,e_t+i\vec a. $$
The word gives $\tilde U=\tilde U_{\mathrm b}\tilde u_{\mathrm r}$ with $\tilde u_{\mathrm r}$ $*$-unitary, hence
$$ \tilde U\tilde U^*=\tilde U_{\mathrm b}\tilde u_{\mathrm r}\tilde u_{\mathrm r}^*\tilde U_{\mathrm b}^*=\tilde U_{\mathrm b}^2. $$
With $\tilde U_{\mathrm b}=\cosh\frac{\theta_{\mathrm{rpd}}}{2}e_t+i\sinh\frac{\theta_{\mathrm{rpd}}}{2}\vec u_{\mathrm b}$ and $\vec u_{\mathrm b}^2=-1$,
$$ \tilde U_{\mathrm b}^2=\cosh\theta_{\mathrm{rpd}}\,e_t+i\sinh\theta_{\mathrm{rpd}}\,\vec u_{\mathrm b}. $$
Comparing the scalar parts gives $\cosh\theta_{\mathrm{rpd}}=\lVert\tilde Q\rVert^2/|N|$ and $\theta_{\mathrm{rpd}}=\operatorname{arcosh}(\lVert\tilde Q\rVert^2/|N|)$; the argument is real because $\lVert\tilde Q\rVert^2\ge|N|$ by Cauchy–Schwarz. Comparing the vector parts gives $\sinh\theta_{\mathrm{rpd}}\,\vec u_{\mathrm b}=\vec a$, hence $\vec u_{\mathrm b}=\vec a/\lVert\vec a\rVert$ when $\theta_{\mathrm{rpd}}>0$. The box for $\tilde U_{\mathrm b}$ follows from $\cosh\frac{\theta_{\mathrm{rpd}}}{2}=\sqrt{(\lVert\tilde Q\rVert^2+|N|)/(2|N|)}$ and $\sinh\frac{\theta_{\mathrm{rpd}}}{2}=\sqrt{(\lVert\tilde Q\rVert^2-|N|)/(2|N|)}$.
The rotor. The boost is Hermitian of norm one, so $\tilde U_{\mathrm b}^{-1}=\tilde U_{\mathrm b}^{\natural}=\cosh\frac{\theta_{\mathrm{rpd}}}{2}e_t-i\sinh\frac{\theta_{\mathrm{rpd}}}{2}\vec u_{\mathrm b}$. Therefore
$$ \tilde u_{\mathrm r}=\tilde U_{\mathrm b}^{-1}\tilde U =\frac{1}{\varrho\,e^{i\theta_{\mathrm{pha}}}}\Bigl(\cosh\frac{\theta_{\mathrm{rpd}}}{2}e_t-i\sinh\frac{\theta_{\mathrm{rpd}}}{2}\vec u_{\mathrm b}\Bigr)\tilde Q. $$
Substituting the half-angle values, and using $\varrho\,e^{i\theta_{\mathrm{pha}}}=\varrho e^{i\theta_{\mathrm{pha}}}$ and $\varrho^2=|N|$, gives the box for $\tilde u_{\mathrm r}$.
The angles. By the existence theorem, $\tilde u_{\mathrm r}$ is a unit real quaternion. Write $\tilde u_{\mathrm r}=\cos\frac{\theta_{\mathrm{axr}}}{2}e_t+\sin\frac{\theta_{\mathrm{axr}}}{2}\vec u_{\mathrm r}$ with $\theta_{\mathrm{axr}}\in[0,2\pi]$, so $\operatorname{Sc}\tilde u_{\mathrm r}=\cos\frac{\theta_{\mathrm{axr}}}{2}$ and $\mathrm{Vect}\tilde u_{\mathrm r}=\sin\frac{\theta_{\mathrm{axr}}}{2}\vec u_{\mathrm r}$ with $\sin\frac{\theta_{\mathrm{axr}}}{2}\ge0$. For a real unit vector $u$ and a complex vector $\tilde V$ with scalar part $\tilde V_t$, one has $u\tilde V=\tilde V_tu-u\cdot\tilde V+u\times\tilde V$. Applying this to the box for $\tilde u_{\mathrm r}$ gives, with $D=\vec u_{\mathrm b}\cdot\vec Q$,
$$ \operatorname{Sc}\tilde u_{\mathrm r}=\frac{e^{-i\theta_{\mathrm{pha}}}}{|N|}\bigl(CQ_t+iSD\bigr), \qquad \mathrm{Vect}\tilde u_{\mathrm r}=\frac{e^{-i\theta_{\mathrm{pha}}}}{|N|}\bigl(C\vec Q-iS(Q_t\vec u_{\mathrm b}+\vec u_{\mathrm b}\times\vec Q)\bigr). $$
Taking the real part of the scalar part gives the box for $\theta_{\mathrm{axr}}$. Taking the norm of the vector part gives $\sin\frac{\theta_{\mathrm{axr}}}{2}=\lVert C\vec Q-iS(Q_t\vec u_{\mathrm b}+\vec u_{\mathrm b}\times\vec Q)\rVert/|N|$, and $\vec u_{\mathrm r}=\mathrm{Vect}\tilde u_{\mathrm r}/\sin\frac{\theta_{\mathrm{axr}}}{2}$ is the box for $\vec u_{\mathrm r}$.
The assembly. The boost box gives $\tilde U_{\mathrm b}=\cosh\frac{\theta_{\mathrm{rpd}}}{2}e_t+i\sinh\frac{\theta_{\mathrm{rpd}}}{2}\vec u_{\mathrm b}$ and the rotor box gives $\tilde u_{\mathrm r}=\cos\frac{\theta_{\mathrm{axr}}}{2}e_t+\sin\frac{\theta_{\mathrm{axr}}}{2}\vec u_{\mathrm r}$; their product is $\tilde U_{\mathrm b}\tilde u_{\mathrm r}=\tilde U=\tilde Q/(\varrho\,e^{i\theta_{\mathrm{pha}}})$, which is the word.
Check. For
$$ \tilde Q=\Bigl(1-\frac{i}{2}\Bigr)e_t+\frac{1+i}{\sqrt2}\vec e_x+\frac{1}{\sqrt2}\vec e_y+\frac{i}{2}\vec e_z, $$
one has $N=1$ and $\lVert\tilde Q\rVert^2=3$, hence $\varrho=1$, $\theta_{\mathrm{pha}}=0$, $\theta_{\mathrm{rpd}}=1.762747174$, $\vec a=2\sqrt2\,\vec e_x$ and $\vec u_{\mathrm b}=\vec e_x$. The boxes give
$$ \tilde U_{\mathrm b}=\sqrt2\,e_t+i\vec e_x, \qquad \tilde u_{\mathrm r}=\frac{1}{\sqrt2}e_t+\frac12(\vec e_x+\vec e_y), \qquad \theta_{\mathrm{axr}}=\frac{\pi}{2}, \qquad \vec u_{\mathrm r}=\frac{\vec e_x+\vec e_y}{\sqrt2}. $$
The Boxes at the Wall
The general boxes are stated for every invertible element. On the wall the normal form $\tilde Q=Q_t(e_t+\vec\Upsilon)$ evaluates them in closed shape, and the theorem below is that evaluation.
Setting. Use the normal form of §The Domain, $\tilde Q=Q_t(e_t+\vec\Upsilon)$, and write
$$ \vec\Upsilon=\vec A+i\vec B,\qquad \vec A,\vec B\in\mathbb R^3,\qquad a=\lVert\vec A\rVert. $$
Since $\vec\Upsilon^2=0$, one has $\vec A\cdot\vec B=0$ and $\lVert\vec A\rVert=\lVert\vec B\rVert=a$. In terms of the coordinates, write $Q_t=q_t+iq'_t$ and $Q_j=q_j+iq'_j$ for $j=x,y,z$, with the eight coefficients real, and put $\vec q=(q_x,q_y,q_z)$ and $\vec q'=(q'_x,q'_y,q'_z)$. Then
$$ \vec A=\frac{q_t\,\vec q+q'_t\,\vec q'}{q_t^2+q_t'^2},\qquad \vec B=\frac{q_t\,\vec q'-q'_t\,\vec q}{q_t^2+q_t'^2}, $$
since $Q_t^{-1}=(q_t-iq'_t)/(q_t^2+q_t'^2)$ and $\vec\Upsilon=\vec Q\,Q_t^{-1}$.
Theorem (the word and its seven data). On the Jordan part $a>0$, every $\tilde Q$ of the sector is
$$ \boxed{\;\tilde Q=Q_t(e_t+\vec\Upsilon)=\varrho e^{i\theta_{\mathrm{pha}}}\,\tilde U_{\mathrm b}\,\tilde u_{\mathrm r} =\varrho e^{i\theta_{\mathrm{pha}}}\exp\Bigl(i\frac{\theta_{\mathrm{rpd}}}{2}\vec u_{\mathrm b}\Bigr)\exp\Bigl(\frac{\theta_{\mathrm{axr}}}{2}\vec u_{\mathrm r}\Bigr)\;} $$
with the two unit factors
$$ \boxed{\;\tilde U_{\mathrm b}=\cosh\frac{\theta_{\mathrm{rpd}}}{2}\,e_t+i\sinh\frac{\theta_{\mathrm{rpd}}}{2}\,\vec u_{\mathrm b},\qquad \tilde u_{\mathrm r}=\cos\frac{\theta_{\mathrm{axr}}}{2}\,e_t+\sin\frac{\theta_{\mathrm{axr}}}{2}\,\vec u_{\mathrm r}\;} $$
and with the seven data determined by $Q_t,Q_x,Q_y,Q_z$. Here $\varrho\,e^{i\theta_{\mathrm{pha}}}$ is the principal square root of $Q_t^2$, that is, the root with non-negative real part and, on the imaginary axis, non-negative imaginary part.
$$ \boxed{\;\varrho=|Q_t|,\qquad \varrho\,e^{i\theta_{\mathrm{pha}}}=\begin{cases}Q_t,&\operatorname{Re}Q_t>0\ \text{or}\ \bigl(\operatorname{Re}Q_t=0,\ \operatorname{Im}Q_t>0\bigr),\\[2pt]-Q_t,&\text{otherwise},\end{cases}\qquad \varepsilon=\frac{Q_t}{\varrho\,e^{i\theta_{\mathrm{pha}}}}=\pm1\;} $$
$$ \boxed{\;\theta_{\mathrm{pha}}=\arg(\varrho\,e^{i\theta_{\mathrm{pha}}})=\begin{cases}\arg Q_t,&\varepsilon=+1,\\[2pt]\arg(-Q_t),&\varepsilon=-1,\end{cases}\;} $$
$$ \boxed{\;\theta_{\mathrm{rpd}}=2\operatorname{arsinh}a\;} $$
$$ \boxed{\;\theta_{\mathrm{axr}}=\begin{cases}2\arctan a,&\varepsilon=+1,\\[2pt]2\pi-2\arctan a,&\varepsilon=-1.\end{cases}\;} $$
$$ \boxed{\;\vec u_{\mathrm b}=\frac{\vec B+\vec A\times\vec B}{a\sqrt{1+a^2}},\qquad \vec u_{\mathrm r}=\varepsilon\,\frac{\vec A}{a}\;} $$
$$ \boxed{\;\tilde u_{\mathrm r}=\varepsilon\,\frac{e_t+\vec A}{\sqrt{1+a^2}},\qquad \tilde U_{\mathrm b}=\sqrt{1+a^2}\,e_t+\frac{i}{\sqrt{1+a^2}}\bigl(\vec B+\vec A\times\vec B\bigr)\;} $$
On the scalar line $a=0$ one has $\vec\Upsilon=0$, $\theta_{\mathrm{rpd}}=0$, and $\theta_{\mathrm{axr}}=0$ if $\varepsilon=+1$, $\theta_{\mathrm{axr}}=2\pi$ if $\varepsilon=-1$; both axes are arbitrary.
Proof.
The modulus and the sign. Since $N=Q_t^2$, the modulus satisfies $\varrho^{2}e^{2i\theta_{\mathrm{pha}}}=Q_t^2$, hence $\varrho\,e^{i\theta_{\mathrm{pha}}}=\pm Q_t$. The principal branch selects the sign displayed, and $\varrho=|\varrho\,e^{i\theta_{\mathrm{pha}}}|=|Q_t|$, $\varepsilon=Q_t/(\varrho\,e^{i\theta_{\mathrm{pha}}})=\pm1$. Writing $\varrho\,e^{i\theta_{\mathrm{pha}}}=\varrho e^{i\theta_{\mathrm{pha}}}$ gives $\theta_{\mathrm{pha}}=\arg(\varrho\,e^{i\theta_{\mathrm{pha}}})$, with the principal argument in $(-\pi,\pi]$; since $\varrho\,e^{i\theta_{\mathrm{pha}}}=\varepsilon Q_t$, this is the box for $\theta_{\mathrm{pha}}$.
The structure of $\vec\Upsilon$. With $\vec\Upsilon=\vec A+i\vec B$,
$$ \vec\Upsilon^2=\vec A^2-\vec B^2+2i\vec A\cdot\vec B =\bigl(\lVert\vec B\rVert^2-\lVert\vec A\rVert^2\bigr)+2i\vec A\cdot\vec B, $$
because $\vec A^2=-\lVert\vec A\rVert^2$ and $\vec B^2=-\lVert\vec B\rVert^2$ for real vectors. The vanishing of $\vec\Upsilon^2$ gives $\lVert\vec A\rVert=\lVert\vec B\rVert=a$ and $\vec A\cdot\vec B=0$.
The product $\tilde U\tilde U^*$. The unit factor is $\tilde U=\varepsilon(e_t+\vec\Upsilon)$ and its star conjugate is $\tilde U^*=\varepsilon(e_t-\bar{\vec\Upsilon})$. Hence
$$ \tilde U\tilde U^*=(e_t+\vec\Upsilon)(e_t-\bar{\vec\Upsilon}) =e_t+\vec\Upsilon-\bar{\vec\Upsilon}-\vec\Upsilon\bar{\vec\Upsilon}. $$
Now $\vec\Upsilon-\bar{\vec\Upsilon}=2i\vec B$, and, using $\vec A\cdot\vec B=0$ and $\lVert\vec A\rVert=\lVert\vec B\rVert=a$,
$$ \vec\Upsilon\bar{\vec\Upsilon} =(\vec A+i\vec B)(\vec A-i\vec B) =-2a^2-2i\,\vec A\times\vec B. $$
Therefore
$$ \tilde U\tilde U^*=(1+2a^2)e_t+2i\bigl(\vec B+\vec A\times\vec B\bigr). $$
The boost. Since $\tilde U=\tilde U_{\mathrm b}\tilde u_{\mathrm r}$ with $\tilde u_{\mathrm r}$ $*$-unitary, $\tilde U\tilde U^*=\tilde U_{\mathrm b}^2$. With $\tilde U_{\mathrm b}=\cosh\frac{\theta_{\mathrm{rpd}}}{2}e_t+i\sinh\frac{\theta_{\mathrm{rpd}}}{2}\vec u_{\mathrm b}$,
$$ \tilde U_{\mathrm b}^2=\cosh\theta_{\mathrm{rpd}}\,e_t+i\sinh\theta_{\mathrm{rpd}}\,\vec u_{\mathrm b}. $$
Comparing the scalar parts gives $\cosh\theta_{\mathrm{rpd}}=1+2a^2=1+2\sinh^2\frac{\theta_{\mathrm{rpd}}}{2}$, so $\sinh\frac{\theta_{\mathrm{rpd}}}{2}=a$, $\cosh\frac{\theta_{\mathrm{rpd}}}{2}=\sqrt{1+a^2}$ and $\theta_{\mathrm{rpd}}=2\operatorname{arsinh}a$. Comparing the vector parts gives $\sinh\theta_{\mathrm{rpd}}\,\vec u_{\mathrm b}=2(\vec B+\vec A\times\vec B)$. Since $\vec A\cdot\vec B=0$,
$$ \lVert\vec B+\vec A\times\vec B\rVert^2=\lVert\vec B\rVert^2+\lVert\vec A\times\vec B\rVert^2=a^2+a^4=a^2(1+a^2), $$
so $2\lVert\vec B+\vec A\times\vec B\rVert=2a\sqrt{1+a^2}=\sinh\theta_{\mathrm{rpd}}$. This gives the boxes for $\vec u_{\mathrm b}$ and $\tilde U_{\mathrm b}$.
The rotor. Put $c=\sqrt{1+a^2}$ and $w=\sinh\frac{\theta_{\mathrm{rpd}}}{2}\vec u_{\mathrm b}=(\vec B+\vec A\times\vec B)/c$, so that $\tilde U_{\mathrm b}=c\,e_t+iw$. The candidate
$$ \tilde u_{\mathrm r}=\varepsilon\,\frac{e_t+\vec A}{c} $$
is a unit real quaternion, because $(e_t+\vec A)(e_t-\vec A)=e_t-\vec A^2=(1+a^2)e_t=c^2e_t$. Moreover, using $\vec B\vec A=-\vec A\times\vec B$ and $(\vec A\times\vec B)\vec A=a^2\vec B$,
$$ w(e_t+\vec A)=\frac{(\vec B+\vec A\times\vec B)(e_t+\vec A)}{c}=c\,\vec B, $$
hence
$$ \tilde U_{\mathrm b}\tilde u_{\mathrm r} =(c\,e_t+iw)\,\varepsilon\,\frac{e_t+\vec A}{c} =\varepsilon\bigl(e_t+\vec A+i\vec B\bigr) =\varepsilon(e_t+\vec\Upsilon)=\tilde U. $$
By the uniqueness of the complete polar decomposition, this is the rotor, and the box for $\tilde u_{\mathrm r}$ follows.
The angles. From the box, $\operatorname{Sc}\tilde u_{\mathrm r}=\varepsilon/c$ and $\mathrm{Vect}\tilde u_{\mathrm r}=\varepsilon\vec A/c$. With $\theta_{\mathrm{axr}}\in[0,2\pi]$, $\cos\frac{\theta_{\mathrm{axr}}}{2}=\varepsilon/c$ and $\sin\frac{\theta_{\mathrm{axr}}}{2}=a/c\ge0$. For $\varepsilon=+1$ this gives $\frac{\theta_{\mathrm{axr}}}{2}=\arctan a$; for $\varepsilon=-1$ it gives $\frac{\theta_{\mathrm{axr}}}{2}=\pi-\arctan a$. Hence the box for $\theta_{\mathrm{axr}}$, and
$$ \vec u_{\mathrm r}=\frac{\mathrm{Vect}\tilde u_{\mathrm r}}{\sin\frac{\theta_{\mathrm{axr}}}{2}} =\varepsilon\,\frac{\vec A}{a}. $$
The scalar line. If $a=0$ then $\vec B=0$ and $\vec A=0$, so $\vec\Upsilon=0$ and $\tilde Q=Q_te_t$. Then $\theta_{\mathrm{rpd}}=0$, and the rotor is $\varepsilon e_t$, that is, $\theta_{\mathrm{axr}}=0$ for $\varepsilon=+1$ and $\theta_{\mathrm{axr}}=2\pi$ for $\varepsilon=-1$. Both axes are arbitrary.
Check. For
$$ \tilde Q=e_t+\frac{i}{\sqrt2}\vec e_x+\vec e_y+\frac{i}{\sqrt2}\vec e_z, $$
one has $\vec A=\vec e_y$, $\vec B=(\vec e_x+\vec e_z)/\sqrt2$, $a=1$, $Q_t=1$, $\varepsilon=1$. The boxes give
$$ \theta_{\mathrm{rpd}}=1.762747174,\qquad \theta_{\mathrm{axr}}=\frac{\pi}{2},\qquad \vec u_{\mathrm b}=\vec e_x,\qquad \vec u_{\mathrm r}=\vec e_y,\qquad \tilde u_{\mathrm r}=\frac{e_t+\vec e_y}{\sqrt2},\qquad \tilde U_{\mathrm b}=\sqrt2\,e_t+i\vec e_x. $$
The Spectral Reading
The word is read here against the spectral decomposition, factor by factor. The two agree on the eigenvalues of the unit factor and disagree everywhere else.
The Eigenbasis
Let $\tilde Q\in\Sigma_{11}$, and fix a square root $B$ of $B^2$. Define
$$ \vec{Ζ}=-i\frac{\vec Q}{B}, \qquad \tilde\Pi_\pm=\frac{e_t\pm\vec{Ζ}}{2}, \qquad \lambda_\pm=Q_t\pm iB. $$
Proposition (the eigenbasis). The element $\vec{Ζ}$ is a root of one,
$$ \vec{Ζ}^2=e_t, $$
the two elements $\tilde\Pi_\pm$ are complementary idempotents,
$$ \tilde\Pi_\pm^2=\tilde\Pi_\pm, \qquad \tilde\Pi_+\tilde\Pi_-=\tilde\Pi_-\tilde\Pi_+=0, \qquad \tilde\Pi_++\tilde\Pi_-=e_t, $$
and
$$ \tilde Q=\lambda_+\tilde\Pi_++\lambda_-\tilde\Pi_-. $$
The eigenvalues are distinct and nonzero, and the idempotents are the spectral projections.
Proof. For a pure vector, $\vec Q^2=-\vec Q\cdot\vec Q=-B^2$. Hence
$$ \vec{Ζ}^2=\Bigl(-i\frac{\vec Q}{B}\Bigr)^2=(-i)^2\frac{\vec Q^2}{B^2}=(-1)\frac{-B^2}{B^2}=e_t. $$
The idempotent identities follow from $\vec{Ζ}^2=e_t$ and the commutation of $\vec{Ζ}$ with itself. The spectral decomposition is checked by expanding the right-hand side:
$$ \lambda_+\tilde\Pi_++\lambda_-\tilde\Pi_- =\frac{\lambda_++\lambda_-}{2}e_t+\frac{\lambda_+-\lambda_-}{2}\vec{Ζ} =Q_te_t+iB\vec{Ζ} =Q_te_t+\vec Q. $$
The two signs of $B$ exchange $\vec{Ζ}$ with $-\vec{Ζ}$, $\tilde\Pi_+$ with $\tilde\Pi_-$, and $\lambda_+$ with $\lambda_-$; the unordered pair is unchanged. The element is semisimple, its minimal polynomial is $X^2-2Q_tX+N$, and its centraliser is $\mathbb{C}[\tilde Q]\cong\mathbb{C}\oplus\mathbb{C}$.
The Unit Factor on the Eigenlines
Since $\tilde U=\tilde Q/(\varrho\,e^{i\theta_{\mathrm{pha}}})$ and $\varrho\,e^{i\theta_{\mathrm{pha}}}$ is central,
$$ \tilde U=\mu_+\tilde\Pi_++\mu_-\tilde\Pi_-, \qquad \mu_\pm=\frac{\lambda_\pm}{\varrho\,e^{i\theta_{\mathrm{pha}}}}. $$
The two eigenvalues of the unit factor satisfy
$$ \mu_+\mu_-=\frac{N}{\varrho^{2}e^{2i\theta_{\mathrm{pha}}}}=1, \qquad \mu_++\mu_-=\frac{2Q_t}{\varrho\,e^{i\theta_{\mathrm{pha}}}}=2\operatorname{Sc}\tilde U, \qquad \mu_+\neq\mu_-. $$
They are the roots of $X^2-2U_tX+1$, with $U_t=\operatorname{Sc}\tilde U$, hence
$$ \boxed{\mu_\pm=U_t\pm i\sqrt{1-U_t^2}}. $$
The spectrum of $\tilde U$ therefore depends on $\tilde Q$ only through the single complex number $U_t=Q_t/(\varrho\,e^{i\theta_{\mathrm{pha}}})$. The two eigenvalues are distinct because $B^2\neq0$, and their product is one.
The spectral data of the element and of the unit factor are related by
$$ \lambda_\pm=\varrho\,e^{i\theta_{\mathrm{pha}}}\mu_\pm, \qquad N=\varrho^{2}e^{2i\theta_{\mathrm{pha}}}, \qquad Q_t=\varrho\,e^{i\theta_{\mathrm{pha}}} U_t, \qquad B^2=\varrho^{2}e^{2i\theta_{\mathrm{pha}}}(1-U_t^2). $$
The last identity is the spectral form of the defining condition: the sector $B^2\neq0$ is the locus $U_t^2\neq1$, and the spectral cone $B^2=0$ is the locus $U_t=\pm1$.
The Normal Case
The eigenbasis is a polar basis when the element commutes with its Hermitian conjugate.
Lemma. For every $\tilde Q=Q_te_t+\vec Q$,
$$ [\tilde Q,\tilde Q^*]=-2\,\vec Q\times\overline{\vec Q}. $$
Proof. The scalar parts are central and commute with everything. For pure vectors,
$$ \vec Q\,\overline{\vec Q}=-\vec Q\cdot\overline{\vec Q}+\vec Q\times\overline{\vec Q}, \qquad \overline{\vec Q}\,\vec Q=-\vec Q\cdot\overline{\vec Q}-\vec Q\times\overline{\vec Q}, $$
so the commutator is $-2\vec Q\times\overline{\vec Q}$.
Theorem (the normal dichotomy). For $\tilde Q\in\Sigma_{11}$ with word as above, the following are equivalent.
- $\tilde Q$ is normal, $[\tilde Q,\tilde Q^*]=0$.
- $\vec Q\times\overline{\vec Q}=0$, that is, $\vec Q=\omega\vec u$ with $\omega\in\mathbb{C}$ and $\vec u$ a real unit vector.
- The two polar factors $\tilde U_{\mathrm b}$ and $\tilde u_{\mathrm r}$ commute with $\tilde Q$.
- The two axes are parallel or antiparallel, or one of the two angles vanishes.
- The word is a single exponential.
In this case the common axis is $\pm i\vec{Ζ}$ whenever it is defined, and
$$ \boxed{\; \tilde Q=\varrho e^{i\theta_{\mathrm{pha}}}\exp\Bigl(\frac{\theta_{\mathrm{axr}}+i\theta_{\mathrm{rpd}}}{2}\vec u\Bigr), \qquad \vec u=\pm i\vec{Ζ}, \;} $$
with the sign chosen so that $\theta_{\mathrm{rpd}}\ge0$. The eigenvalues of the unit factor are
$$ \boxed{\mu_\pm=e^{\mp\frac{\theta_{\mathrm{rpd}}}{2}}e^{\pm i\frac{\theta_{\mathrm{axr}}}{2}}}, $$
and the spectrum of $\tilde U$ determines the two angles,
$$ \theta_{\mathrm{rpd}}=\ln\Bigl\lvert\frac{\mu_-}{\mu_+}\Bigr\rvert, \qquad \theta_{\mathrm{axr}}=2\arg\mu_+\pmod{2\pi}. $$
Proof. The equivalence of 1 and 2 is the lemma, with the rank-one condition on the complex vector $\vec Q$. If $\tilde Q$ is normal then $\tilde U$ is normal, because $\tilde U=\tilde Q/(\varrho\,e^{i\theta_{\mathrm{pha}}})$ and $\varrho\,e^{i\theta_{\mathrm{pha}}}$ is central. The Hermitian factor $\tilde U_{\mathrm b}=(\tilde U\tilde U^*)^{1/2}$ is a function of $\tilde U\tilde U^*$, hence commutes with $\tilde U$, and $\tilde u_{\mathrm r}=\tilde U_{\mathrm b}^{-1}\tilde U$ commutes with $\tilde U$ as well. Both commute with $\tilde Q=\varrho\,e^{i\theta_{\mathrm{pha}}}\tilde U$, which gives 3. Conversely, if the two factors commute with $\tilde Q$, they commute with $\tilde U$ and with each other, so
$$ \tilde Q\tilde Q^*=|\varrho\,e^{i\theta_{\mathrm{pha}}}|^2\tilde U_{\mathrm b}\tilde u_{\mathrm r}\tilde u_{\mathrm r}^*\tilde U_{\mathrm b} =|\varrho\,e^{i\theta_{\mathrm{pha}}}|^2\tilde U_{\mathrm b}^2 =\tilde Q^*\tilde Q, $$
which gives 1. If 3 holds, both factors are functions of $\tilde U$, hence diagonal in its eigenbasis, so their axes are the two eigenlines, which are $\pm i\vec{Ζ}$; this gives 4. If 4 holds, the two exponents commute, so the product is the exponential of their sum, which gives 5. If 5 holds, $\tilde Q$ is a scalar multiple of the exponential of a complex multiple of a real pure vector, and such an exponential is normal; this gives 1.
For the formula, write $\tilde U_{\mathrm b}=\exp(i\frac{\theta_{\mathrm{rpd}}}{2}\vec u)$ and $\tilde u_{\mathrm r}=\exp(\frac{\theta_{\mathrm{axr}}}{2}\vec u)$ on the common axis. Their product is $\exp(\frac{\theta_{\mathrm{axr}}+i\theta_{\mathrm{rpd}}}{2}\vec u)$ because the exponents commute. On the two eigenlines the product acts by $e^{-\theta_{\mathrm{rpd}}/2}e^{i\theta_{\mathrm{axr}}/2}$ and $e^{\theta_{\mathrm{rpd}}/2}e^{-i\theta_{\mathrm{axr}}/2}$; with the labelling of $\mu_\pm$ fixed by $\lambda_\pm$, and the sign of the common axis chosen accordingly, this is the displayed pair $e^{\mp\theta_{\mathrm{rpd}}/2}e^{\pm i\theta_{\mathrm{axr}}/2}$. The extraction of the angles follows from the moduli and the arguments.
The display is the amalgamation of the four factors into one exponential. The complex angle is $\frac{\theta_{\mathrm{axr}}+i\theta_{\mathrm{rpd}}}{2}$; the real part is the rotor angle and the imaginary part is the rapidity. The sign of the imaginary part is the sign convention of the boost: the other choice reverses the common axis.
Remark (the axes in the normal case). The common axis is $\pm i\vec{Ζ}$. Since $\vec{Ζ}=-i\vec Q/B$ and $\vec Q=\omega\vec u$, the axis is $\pm\vec u$: the real direction of the vector part. The eigenbasis and the polar frame are the same frame. The element is normal exactly when its vector part has a real direction; a complex vector part with two independent real directions is non-normal.
The Non-Normal Case
In the general case the polar factors are not functions of $\tilde Q$, and the eigenbasis is not a polar basis.
Proposition (the failure of the spectral reading). Let $\tilde Q\in\Sigma_{11}$ be non-normal. Then
$$ [\tilde U_{\mathrm b},\tilde Q]\neq0, \qquad [\tilde u_{\mathrm r},\tilde Q]\neq0, $$
and the axes are not the eigenlines. The eigenvalues $\mu_\pm$ still determine the unordered pair $\{\mu_+,\mu_-\}$, but they do not determine $\theta_{\mathrm{rpd}}$, $\theta_{\mathrm{axr}}$, $\vec u_{\mathrm b}$ or $\vec u_{\mathrm r}$.
Proof. The factors commute with $\tilde Q$ exactly when the element is normal, by the theorem. If they do not commute, they are not diagonal in the eigenbasis, so their axes cannot be the eigenlines. The eigenvalues of $\tilde U$ are functions of $U_t$ alone, by $\mu_\pm=U_t\pm i\sqrt{1-U_t^2}$; the word parameters also depend on the relative angle of the axes, through $\operatorname{Sc}\tilde U=c_{\mathrm b}c_{\mathrm r}-i s_{\mathrm b}s_{\mathrm r}c$. There are different triples $(\theta_{\mathrm{rpd}},\theta_{\mathrm{axr}},c)$ with the same $U_t$, hence the same spectrum, and with different words.
The last sentence is the precise statement of the gap. In the normal case $c=\pm1$, and the pair $(\theta_{\mathrm{rpd}},\theta_{\mathrm{axr}})$ is recovered from the spectrum. In the non-normal case $c$ is a third unknown, and the spectrum is one complex equation for three real unknowns.
Example. Let $\tilde Q=\bigl(1-\frac{i}{2}\bigr)e_t+\frac{1+i}{\sqrt2}\vec e_x+\frac{1}{\sqrt2}\vec e_y+\frac{i}{2}\vec e_z$. It is N-regular and spectrally regular, and non-normal, because
$$ \vec Q\times\overline{\vec Q} =\Bigl(-\frac{i}{\sqrt2},\frac{i}{\sqrt2},i\Bigr)\neq0. $$
Its word has
$$ \theta_{\mathrm{rpd}}=1.7627471740, \qquad \vec u_{\mathrm b}=\vec e_x, \qquad \theta_{\mathrm{axr}}=\frac{\pi}{2}, \qquad \vec u_{\mathrm r}=\frac{\vec e_x+\vec e_y}{\sqrt2}, \qquad c=\frac{1}{\sqrt2}. $$
The eigenvalues of the unit factor have moduli $0.480533816$ and $2.081019$, and arguments $\pm0.674888846$. The normal-case formulas would give $\theta_{\mathrm{rpd}}=1.466$ and $\theta_{\mathrm{axr}}=1.350$, not the true $1.763$ and $1.571$. The spectrum is the same; the reading is not.
The example is case 5 of the worked examples. The gap is not a rounding effect: the ratio of the two moduli is $4.331$, whose logarithm is $1.466$, while the rapidity is $1.763$.
What the Spectrum Determines
The comparison is a count of real data.
| Data | Real dimension | Determines |
|---|---|---|
| $N$, $B^2$ | $4$ | $\varrho,\theta_{\mathrm{pha}},U_t$ |
| spectrum $\{\lambda_+,\lambda_-\}$ | $4$ | the same, in spectral form |
| eigenprojectors $\tilde\Pi_\pm$ | $4$ | the eigenlines; the polar axes only in the normal case |
| word $(\varrho,\theta_{\mathrm{pha}},\theta_{\mathrm{rpd}},\vec u_{\mathrm b},\theta_{\mathrm{axr}},\vec u_{\mathrm r})$ | $8$ | the element |
The spectrum and the eigenprojectors together determine the element. The polar word is another eight data. The two parametrizations share the four data $(\varrho,\theta_{\mathrm{pha}},U_t)$; the spectrum determines the remaining data only in the normal case, where the eigenprojectors are the polar axes. In the non-normal case the eigenprojectors do not determine the polar axes, and the spectrum does not determine the two angles.
The spectral decomposition is a diagonalization over $\mathbb{C}$: it writes $\tilde Q$ as two complex multiples of two idempotents. The polar decomposition is a Cartan decomposition over $\mathbb{R}$: it writes $\tilde Q$ as a positive Hermitian factor times a $*$-unitary factor. The two agree on the scalar factor and on the eigenvalues of $\tilde Q$, and on nothing else in the non-normal case.
Remark (the two decompositions are transverse). The eigenbasis depends on the complex structure of the vector part; the polar frame depends on the Hermitian structure of the unit factor. The normal elements are the intersection: the elements for which the two frames coincide. The intersection is the normal set, $Q_t+\omega\vec u$ with $Q_t,\omega\in\mathbb{C}$ and $\vec u\in S^2$, of real dimension $2+2+2=6$; it is not open in the sector.
The Unipotent Unit Factor
On the wall the unit factor is $\pm$unipotent. One complex equation fixes it, and that equation forces the two angles and the two axes.
The Scalar Constraint
Theorem (the rigidity of the word). Let $\tilde Q\in\Sigma_{10}$ with word as above. Then
$$ \operatorname{Sc}\tilde U=\varepsilon=\pm1, \qquad \bigl(\tilde U-\varepsilon e_t\bigr)^2=0, $$
and the parameters satisfy
$$ \boxed{\cosh\frac{\theta_{\mathrm{rpd}}}{2}\cos\frac{\theta_{\mathrm{axr}}}{2}=\varepsilon}, \qquad \boxed{\sinh\frac{\theta_{\mathrm{rpd}}}{2}\sin\frac{\theta_{\mathrm{axr}}}{2}\,\vec u_{\mathrm b}\cdot\vec u_{\mathrm r}=0}. $$
Proof. By the normal form, $\tilde U=\varepsilon(e_t+\vec\Upsilon)$ with $\vec\Upsilon^2=0$. Its scalar part is $\varepsilon$, and
$$ \bigl(\tilde U-\varepsilon e_t\bigr)^2 =\varepsilon^2\vec\Upsilon^2=0. $$
The scalar part of the component form is $c_{\mathrm b}c_{\mathrm r}-i s_{\mathrm b}s_{\mathrm r}c$. It equals the real number $\varepsilon$, so its imaginary part vanishes, which gives the second equation, and its real part gives the first.
The two equations are the whole constraint. The scalar part of the unit factor is a complex equation, and the unitarity of $\tilde U$ has already been used; the constraint is one complex equation on the six parameters of the unit factor, which is why the unipotent set has real dimension four.
The Orthogonal Frame
Theorem (the orthogonal frame). On the Jordan part of $\Sigma_{10}$, the word has
$$ \theta_{\mathrm{rpd}}>0, \qquad \theta_{\mathrm{axr}}\notin\{0,\pi,2\pi\}, \qquad \vec u_{\mathrm b}\perp\vec u_{\mathrm r}, $$
and the two angles are tied by
$$ \boxed{\Bigl\lvert\cos\frac{\theta_{\mathrm{axr}}}{2}\Bigr\rvert=\frac{1}{\cosh\frac{\theta_{\mathrm{rpd}}}{2}}}, \qquad \varepsilon=\operatorname{sgn}\cos\frac{\theta_{\mathrm{axr}}}{2}. $$
Conversely, every choice of $\theta_{\mathrm{axr}}\in(0,\pi)\cup(\pi,2\pi)$, of $\vec u_{\mathrm b}\in S^2$ and of $\vec u_{\mathrm r}$ in the great circle orthogonal to $\vec u_{\mathrm b}$, with $\theta_{\mathrm{rpd}}$ defined by the displayed relation, gives a word of an element of the Jordan part.
Proof. On the Jordan part $\vec\Upsilon\neq0$, so $\tilde U\neq\varepsilon e_t$. If $\theta_{\mathrm{rpd}}=0$ then $c_{\mathrm b}=1$ and the first constraint gives $c_{\mathrm r}=\pm1$, so $\theta_{\mathrm{axr}}\in\{0,2\pi\}$ and the component form is $\tilde U=\pm e_t$, a contradiction. Hence $\theta_{\mathrm{rpd}}>0$ and $s_{\mathrm b}>0$. If $\theta_{\mathrm{axr}}=0$ or $2\pi$ then $s_{\mathrm r}=0$ and the component form is again scalar. If $\theta_{\mathrm{axr}}=\pi$ then $c_{\mathrm r}=0$, and the first constraint gives $0=\pm1$, impossible. Hence $\theta_{\mathrm{axr}}\notin\{0,\pi,2\pi\}$ and $s_{\mathrm r}>0$. The second constraint then forces $c=0$. The first gives $|c_{\mathrm b}c_{\mathrm r}|=1$, which is the displayed relation, and the sign of $c_{\mathrm r}$ is the sign of $\varepsilon$. The converse is a direct substitution: with orthogonal axes and the relation, the scalar part is $\varepsilon$ and the vector part is null, so the unit factor is unipotent and the element lies in the Jordan part.
The frame is therefore orthogonal, and one of the two angles is a function of the other. The independent data of the unipotent factor are one angle and a frame: $\theta_{\mathrm{axr}}$ (one real number), $\vec u_{\mathrm b}$ (two), and $\vec u_{\mathrm r}$ in the orthogonal circle (one), which is four real dimensions, the dimension of the punctured nilpotent cone.
The Closed Form
Substituting the constraints into the component form gives the shape of the whole sector.
Theorem (the closed form). Every $\tilde Q\in\Sigma_{10}$ is
$$ \boxed{\; \tilde Q=Q_t\Bigl(e_t+ i\tanh\frac{\theta_{\mathrm{rpd}}}{2}\,\vec u_{\mathrm b} +\tan\frac{\theta_{\mathrm{axr}}}{2}\,\vec u_{\mathrm r} +i\tanh\frac{\theta_{\mathrm{rpd}}}{2}\tan\frac{\theta_{\mathrm{axr}}}{2}\;\vec u_{\mathrm b}\times\vec u_{\mathrm r}\Bigr) \;} $$
with $\cosh\frac{\theta_{\mathrm{rpd}}}{2}\cos\frac{\theta_{\mathrm{axr}}}{2}=\varepsilon$; on the Jordan part the two axes are orthogonal. Equivalently,
$$ \boxed{\; \tilde Q=Q_t\Bigl(e_t+\vec\Upsilon\Bigr), \qquad \vec\Upsilon=i\tanh\frac{\theta_{\mathrm{rpd}}}{2}\,\vec u_{\mathrm b} +\tan\frac{\theta_{\mathrm{axr}}}{2}\,\vec u_{\mathrm r} +i\tanh\frac{\theta_{\mathrm{rpd}}}{2}\tan\frac{\theta_{\mathrm{axr}}}{2}\;\vec u_{\mathrm b}\times\vec u_{\mathrm r}. \;} $$
Proof. The scalar part of the unit factor is $\varepsilon$, so $c_{\mathrm r}=\varepsilon/c_{\mathrm b}$. On the Jordan part $s_{\mathrm r}>0$, and $\tan\frac{\theta_{\mathrm{axr}}}{2}=s_{\mathrm r}/c_{\mathrm r}=\varepsilon c_{\mathrm b}s_{\mathrm r}$, hence $s_{\mathrm r}=\varepsilon\tan\frac{\theta_{\mathrm{axr}}}{2}/c_{\mathrm b}$. The component form then gives
$$ i s_{\mathrm b}c_{\mathrm r} =i\varepsilon\tanh\frac{\theta_{\mathrm{rpd}}}{2}, \qquad c_{\mathrm b}s_{\mathrm r} =\varepsilon\tan\frac{\theta_{\mathrm{axr}}}{2}, \qquad i s_{\mathrm b}s_{\mathrm r} =i\varepsilon\tanh\frac{\theta_{\mathrm{rpd}}}{2}\tan\frac{\theta_{\mathrm{axr}}}{2}. $$
The vector part of the unit factor is $\varepsilon$ times the bracket, and $\tilde U=\varepsilon(e_t+\vec\Upsilon)$, so the bracket is $\vec\Upsilon$; hence $\tilde Q=Q_t(e_t+\vec\Upsilon)$. On the scalar line $\tan\frac{\theta_{\mathrm{axr}}}{2}=\tanh\frac{\theta_{\mathrm{rpd}}}{2}=0$, and the same display reduces to $Q_te_t$.
The formula is the exact shape of the sector. The complex number $Q_t$ carries the scale and the phase; the rest is a unipotent written in the frame. The vector part is automatically null: with the three orthogonal directions, its square is
$$ \vec\Upsilon^2=\tanh^2\frac{\theta_{\mathrm{rpd}}}{2}-\tan^2\frac{\theta_{\mathrm{axr}}}{2} +\tanh^2\frac{\theta_{\mathrm{rpd}}}{2}\tan^2\frac{\theta_{\mathrm{axr}}}{2}, $$
which vanishes by $|\cos\frac{\theta_{\mathrm{axr}}}{2}|=1/\cosh\frac{\theta_{\mathrm{rpd}}}{2}$. The cross term carries the plus sign because its coefficient is imaginary.
Remark (the scalar factor). The unit factor depends only on $\theta_{\mathrm{axr}}$ and the frame, not on $Q_t$. Changing $Q_t$ by a nonzero complex number changes the scale and the phase, and possibly the sign $\varepsilon$, but leaves $\vec\Upsilon$ unchanged. The sector is the product of the scalar line $\mathbb{C}^{\times}$ with the punctured nilpotent cone; the frame and the two angles are functions of the second factor alone, and the sign is carried by the first.
The Two Strata
The wall splits into two strata. The scalar line is the set $\vec\Upsilon=0$; the Jordan part is its complement in $\Sigma_{10}$, where $\vec\Upsilon\neq0$ and the element is defective. The word takes a different shape on each, and it degenerates on neither.
The Scalar Line
On the scalar line $\vec\Upsilon=0$, the element is $Q_te_t$, the unit factor is $\varepsilon e_t$, and
$$ \theta_{\mathrm{rpd}}=0, \qquad \theta_{\mathrm{axr}}\in\{0,2\pi\}, \qquad \vec u_{\mathrm b},\vec u_{\mathrm r}\ \text{free}. $$
The word is
$$ \tilde Q=\varrho e^{i\theta_{\mathrm{pha}}}\,\varepsilon e_t =Q_te_t. $$
The sign $\varepsilon$ is fixed by the branch: if $\varrho\,e^{i\theta_{\mathrm{pha}}}=Q_t$ then $\varepsilon=+1$ and $\theta_{\mathrm{axr}}=0$; if $\varrho\,e^{i\theta_{\mathrm{pha}}}=-Q_t$ then $\varepsilon=-1$ and $\theta_{\mathrm{axr}}=2\pi$. The axes are undefined, because the boost and the rotor are central. The scalar line is the normal locus of the sector: a normal element with a single eigenvalue is a scalar, and every scalar is normal.
The Jordan Part
On the Jordan part $\vec\Upsilon\neq0$, the element is defective: its minimal polynomial is $(X-Q_t)^2$, its geometric multiplicity is one, and it has no eigenbasis. The word has
$$ \theta_{\mathrm{rpd}}>0, \qquad \theta_{\mathrm{axr}}\in(0,\pi)\cup(\pi,2\pi), \qquad \vec u_{\mathrm b}\perp\vec u_{\mathrm r}, $$
and the two possible signs of $\cos\frac{\theta_{\mathrm{axr}}}{2}$ are the two values of $\varepsilon$. The word is
$$ \tilde Q=Q_t\Bigl(e_t+\vec\Upsilon(\theta_{\mathrm{axr}},\vec u_{\mathrm b},\vec u_{\mathrm r})\Bigr), $$
with $\vec\Upsilon$ as in the closed form and $\theta_{\mathrm{rpd}}$ determined by $\theta_{\mathrm{axr}}$. The data are
$$ \underbrace{2}_{Q_t} +\underbrace{1}_{\theta_{\mathrm{axr}}} +\underbrace{2}_{\vec u_{\mathrm b}} +\underbrace{1}_{\vec u_{\mathrm r}\perp\vec u_{\mathrm b}} =6, $$
the real dimension of the sector. The rapidity and the sign are functions of the angle.
The Jordan part is unbounded: as $\theta_{\mathrm{axr}}\to\pi$, $\theta_{\mathrm{rpd}}\to\infty$ and $|\vec\Upsilon|\to\infty$. The value $\theta_{\mathrm{axr}}=\pi$ is not attained, because it would force $c_{\mathrm r}=0$ and violate the scalar constraint.
The Chart of the Wall
The data of the sector are
$$ \underbrace{2}_{Q_t} +\underbrace{1}_{\theta_{\mathrm{axr}}} +\underbrace{2}_{\vec u_{\mathrm b}} +\underbrace{1}_{\vec u_{\mathrm r}\perp\vec u_{\mathrm b}} =6, $$
the real dimension of the wall. The rapidity and the sign are determined by
$$ \cosh\frac{\theta_{\mathrm{rpd}}}{2}=\frac{1}{\lvert\cos\frac{\theta_{\mathrm{axr}}}{2}\rvert}, \qquad \varepsilon=\operatorname{sgn}\cos\frac{\theta_{\mathrm{axr}}}{2}. $$
The coordinate map is
$$ (Q_t,\theta_{\mathrm{axr}},\vec u_{\mathrm b},\vec u_{\mathrm r}) \longmapsto Q_t\Bigl(e_t+\vec\Upsilon(\theta_{\mathrm{axr}},\vec u_{\mathrm b},\vec u_{\mathrm r})\Bigr), $$
from $\mathbb{C}^{\times}\times\bigl((0,\pi)\cup(\pi,2\pi)\bigr)\times US^2$ to the Jordan part, where $US^2=\{(\vec u_{\mathrm b},\vec u_{\mathrm r}):\lvert\vec u_{\mathrm b}\rvert=\lvert\vec u_{\mathrm r}\rvert=1,\ \vec u_{\mathrm b}\perp\vec u_{\mathrm r}\}$ is the unit tangent bundle of the sphere. It is a bijection onto the Jordan part; the scalar line is the common limit $\theta_{\mathrm{axr}}\to0$ and $\theta_{\mathrm{axr}}\to2\pi$, where the frame is free.
The coordinate lines are three families:
- the scalar line $Q_t\mapsto Q_t(e_t+\vec\Upsilon)$, a complex plane with the origin removed;
- the angle line $\theta_{\mathrm{axr}}\mapsto Q_t(e_t+\vec\Upsilon(\theta_{\mathrm{axr}},\ldots))$, a non-compact curve from the scalar line to infinity, with $\theta_{\mathrm{rpd}}$ growing as $\theta_{\mathrm{axr}}\to\pi$;
- the frame sphere $(\vec u_{\mathrm b},\vec u_{\mathrm r})\mapsto Q_t(e_t+\vec\Upsilon)$, a circle bundle over $S^2$.
The sector retracts onto the scalar line, because the nilpotent part can be contracted to zero along $\vec\Upsilon\mapsto(1-t)\vec\Upsilon$, which stays null. Hence $\Sigma_{10}$ is connected and $\pi_1(\Sigma_{10})=\mathbb{Z}$, carried by the phase of $Q_t$.
The Group of Invertibles
The group of invertibles is the $\natural$-regular set, $\mathbb B^{\times}=\mathbb B\setminus\mathcal N$, the complement of the zero-divisor cone. The decomposition read as a map
$$ \Phi:\mathbb B^{\times}\longrightarrow\mathcal H\times SL_2(\mathbb C),\qquad \Phi(\tilde Q)=(\varrho\,e^{i\theta_{\mathrm{pha}}},\tilde U), $$
is a bijection: it is injective by the uniqueness theorem, and it is surjective because for $(\varrho\,e^{i\theta_{\mathrm{pha}}},\tilde U)\in\mathcal H\times SL_2(\mathbb C)$ the element $\varrho\,e^{i\theta_{\mathrm{pha}}}\tilde U$ has norm $\varrho^{2}e^{2i\theta_{\mathrm{pha}}}$, whose principal root is $\varrho\,e^{i\theta_{\mathrm{pha}}}$. Under the matrix model it is the factorisation of an invertible matrix into a complex scalar and a matrix of determinant one.
The map is not a group homomorphism. The scale is multiplicative, and the phase is multiplicative only up to a sign,
$$ \varrho(\tilde Q_1\tilde Q_2)=\varrho(\tilde Q_1)\varrho(\tilde Q_2),\qquad \theta_{\mathrm{pha}}(\tilde Q_1\tilde Q_2)\equiv\theta_{\mathrm{pha}}(\tilde Q_1)+\theta_{\mathrm{pha}}(\tilde Q_2)\pmod{\pi}, $$
The moduli and the phases multiply. The unit factors do not: after the multiplication the middle product $\tilde u_{\mathrm r,1}\tilde U_{\mathrm b,2}$ has to be split again, and §The Group Law gives the correction.
Proposition (the maximal compact subgroup). The maximal compact subgroup of $\mathbb B^{\times}$ is the $*$-unitary group $U(2)$, the set of words with $\varrho=1$ and $\tilde U_{\mathrm b}=e_t$,
$$ U(2)=\{e^{i\theta}\,\tilde u_{\mathrm r}:\theta\in\mathbb R,\ \tilde u_{\mathrm r}\in SU(2)\}. $$
Proof. The compact part of the word is the phase times the rotor, and both are $*$-unitary, so their product is $*$-unitary; the scale and the boost are positive and non-compact. Conversely, if $V$ is $*$-unitary, then $\det V$ is a unit complex number; choose $e^{i\theta_{\mathrm{pha}}}$ in the branch with $e^{2i\theta_{\mathrm{pha}}}=\det V$, and $K=e^{-i\theta_{\mathrm{pha}}}V$ is $*$-unitary of determinant one, hence a unit real quaternion. Any compact subgroup has a bounded scale, hence scale one, and a bounded boost; the boost is trivial on it, since the powers of an eigenvalue $\lambda\neq1$ are unbounded or tend to zero. Hence a compact subgroup lies in $U(2)$, which is compact and therefore maximal.
The group retracts onto $U(2)$ by shrinking the scale to one and pushing the boost to $e_t$ along the Hermitian positive elements of norm one. In particular $\mathbb B^{\times}$ is connected, and its fundamental group is that of $U(2)$, namely $\mathbb Z$, generated by the phase loop around the cone (The Biquaternion Group of Invertibles as a Topological Group). The rotor contributes the simply connected factor $SU(2)=S^{3}$, so the fundamental group is carried by the phase alone.
The compact part has two circular directions, and they are not interchangeable. The central phase is central and acts on the whole algebra by multiplication; the circle inside the rotor is a one-parameter subgroup of $SU(2)$ and acts by conjugation, fixing the scalar line and acting on the vector part. The first carries the fundamental group of the group of invertibles; the second lies in the simply connected sphere and contributes no loop. The extra compact direction is the centre of the algebra, not a factor of the sphere. This is the biquaternion counterpart of the absence of a second continuous factor in the quaternion case: there the sphere is simple modulo its centre, and the only factor beyond it is the scale line.
The Homotopy Type
Forgetting the branch, the map $\tilde Q\mapsto(\varrho\,e^{i\theta_{\mathrm{pha}}},\tilde U)$ is a bijection onto $\mathbb{C}^{\times}\times\mathrm{SL}_2(\mathbb{C})$, and it is not a homeomorphism: on the cut the phase jumps by $-1$ and the rotor changes sign. The positive cone is homeomorphic to $\mathbb{R}^{3}$ and the rotor group is $S^{3}$, so $\mathrm{SL}_2(\mathbb{C})\simeq\mathbb{R}^{3}\times S^{3}$, and $U(2)\cong U(1)\times SU(2)$ as a space gives $\mathbb{B}^{\times}\simeq S^{1}\times S^{3}$.
The Coordinate Lines
The parameters are coordinates on the group, and their ranges make the word map a bijection up to the axis degeneracies. The coordinate lines are four one-parameter families:
- the scale line $\varrho\mapsto \varrho e^{i\theta_{\mathrm{pha}}}\tilde U_{\mathrm b}\tilde u_{\mathrm r}$, a half-line;
- the phase circle $\theta_{\mathrm{pha}}\mapsto \varrho e^{i\theta_{\mathrm{pha}}}\tilde U_{\mathrm b}\tilde u_{\mathrm r}$, a circle cut to a half-open interval by the branch;
- the boost hyperbola $\theta_{\mathrm{rpd}}\mapsto \varrho e^{i\theta_{\mathrm{pha}}}\exp(i\frac{\theta_{\mathrm{rpd}}}{2}\vec u_{\mathrm b})\tilde u_{\mathrm r}$, a non-compact ray;
- the rotor arc $\theta_{\mathrm{axr}}\mapsto \varrho e^{i\theta_{\mathrm{pha}}}\tilde U_{\mathrm b}\exp(\frac{\theta_{\mathrm{axr}}}{2}\vec u_{\mathrm r})$, a great semicircle from $e_t$ to $-e_t$.
The Conformal Factor
Under the two-sided operator $\tilde QX\tilde Q^{*}$, the quaternionic norm of $X$ is multiplied by
$$ \varrho^{4}=\exp(4\ln \varrho)=\bigl\lvert\langle\tilde Q\rangle_{\natural}\bigr\rvert^{2}, $$
so lengths are multiplied by $\varrho^{2}=\lvert\langle\tilde Q\rangle_{\natural}\rvert$. The scale exponent is a quarter of the logarithm of the conformal factor; the phase, the boost and the rotor are isometries of the form.
The Two Halves
Every biquaternion is written uniquely as a real quaternion plus $i$ times a real quaternion,
$$ \tilde Q=\tilde q+i\tilde p,\qquad \tilde q,\tilde p\in\mathbb H_{\mathbb B},\qquad \mathbb B=\mathbb H_{\mathbb B}\oplus i\mathbb H_{\mathbb B}, $$
where $\mathbb H_{\mathbb B}$ is the real span of $e_t,\vec e_x,\vec e_y,\vec e_z$ and $i$ is the central scalar imaginary. On the two halves the norm reads
$$ \langle\tilde Q\rangle_{\natural}=\langle\tilde q\rangle_{\natural}-\langle\tilde p\rangle_{\natural}+2i\,\langle\tilde q,\tilde p\rangle, $$
with $\langle\tilde q\rangle_{\natural}=\sum_\mu q_\mu^{2}$ the real quaternion norm and $\langle\tilde q,\tilde p\rangle=\operatorname{Sc}(\tilde q^{\natural}\tilde p)$ the Euclidean inner product of $\mathbb R^{4}$. The real part is the difference of the two squared lengths and the imaginary part is twice their inner product, so the norm vanishes exactly when the halves have equal length and are orthogonal.
The four factors sit in the two halves as follows. The scale is a real scalar and lies in the first half. The phase is a complex scalar, so its imaginary part is central. The rotor is a real quaternion and lies in the first half. The boost has its scalar part in the first half and its vector part in the second,
$$ \tilde U_{\mathrm b}=a\,e_t+i\vec w,\qquad a\in\mathbb R,\quad \vec w\in\mathbb R^{3},\qquad a^{2}-\lvert\vec w\rvert^{2}=1 . $$
The boost is the only factor whose vector part lies in the second half, and it is the factor that carries the indefiniteness: its scalar part is at least one and its vector part is tied to it by the norm-one condition. The boost is trivial exactly when the unit factor is $*$-unitary, that is, on the elements $\varrho e^{i\theta_{\mathrm{pha}}}\tilde u_{\mathrm r}$; in particular it is trivial on both halves.
This is the precise sense in which the biquaternion case is the complexification of the quaternion one. The scale and the rotor are the quaternion polar decomposition; the complex scalar adds the phase, one compact direction, and the second half adds the boost, three non-compact ones.
The Three Pairings
The four factors can be bracketed in three ways. Each bracketing is a pairing: the factors are regrouped, and the two composite factors are recognised as elements of a smaller structure. Two pairings are partial, their construction dividing by a part of the element that can vanish; the third is total on the $\natural$-regular set.
The quaternion pairing
$$ \tilde Q=\bigl(\varrho\tilde u_{\mathrm r}\bigr)\bigl(e^{i\theta_{\mathrm{pha}}}\,\tilde u_{\mathrm r}^{-1}\tilde U_{\mathrm b}\tilde u_{\mathrm r}\bigr). $$
The first factor is the real quaternion $\varrho\tilde u_{\mathrm r}$: a positive scale times a unit real quaternion. The second is the phase times the boost carried into the rotor frame. The pairing is partial: its construction divides by the part of $\tilde Q$ whose four coefficients are real, so it is available only where that part is nonzero.
The Hamilton root pairing
$$ \tilde Q=(\varrho e^{i\theta_{\mathrm{pha}}})\bigl(\tilde U_{\mathrm b}\tilde u_{\mathrm r}\bigr)=\varrho\,e^{i\theta_{\mathrm{pha}}}\,\tilde U . $$
The first factor is the complex modulus. The second is the unit factor, read on the root of $-e_t$ parallel to the vector part. With $\tilde Q=Q_te_t+\vec Q$, $\vec Q=Q_x\vec e_x+Q_y\vec e_y+Q_z\vec e_z$, $B^{2}=Q_x^{2}+Q_y^{2}+Q_z^{2}$ and $B$ a chosen square root of $B^{2}$,
$$ \tilde U_{\mathrm b}\tilde u_{\mathrm r}=\cos\Theta+\vec U\sin\Theta,\qquad \vec U=\frac{\vec Q}{B},\qquad \cos\Theta=\frac{Q_t}{\varrho\,e^{i\theta_{\mathrm{pha}}}},\qquad \sin\Theta=\frac{B}{\varrho\,e^{i\theta_{\mathrm{pha}}}}. $$
The pairing is partial: the normalisation divides by $B$, so the vector part must be non-null, $B^{2}\neq0$.
The Cartan Hermitian pairing
$$ \tilde Q=\bigl(\varrho\tilde U_{\mathrm b}\bigr)\bigl(e^{i\theta_{\mathrm{pha}}}\tilde u_{\mathrm r}\bigr)=\tilde Q_H\,\tilde Q_0 . $$
The first factor is Hermitian positive definite, the second $*$-unitary. The pairing solves $\tilde Q\tilde Q^{*}=\tilde H^{2}$, so both factors are functions of the element alone, and it is total on the $\natural$-regular set. It is the algebra form of the matrix polar decomposition.
The three pairings, with the form and the domain of each, are the three lines of the table of the Introduction. The two partial domains are not nested, and the Cartan domain contains both. The elements of the $\natural$-regular set outside both partial domains are the nonzero purely imaginary complex scalars.
The Logarithm
Three logarithms meet in this section. The Cartan logarithm is the Hermitian exponent of the unit factor, a single logarithm solves $\exp X=\tilde Q$, and the formal exponent is the sum of the four, which is a logarithm only in the normal case.
The Cartan Logarithm
Write $\tilde U=\tilde U_{\mathrm b}\tilde u_{\mathrm r}$ and put
$$ \tilde H=\operatorname{Log}\tilde U_{\mathrm b}=i\frac{\theta_{\mathrm{rpd}}}{2}\vec u_{\mathrm b}, \qquad \tilde K=\operatorname{Log}\tilde u_{\mathrm r}=\frac{\theta_{\mathrm{axr}}}{2}\vec u_{\mathrm r}. $$
The pair $(\tilde H,\tilde K)$ is the Cartan logarithm of $\tilde U$. It is unique with the ranges of the theorem, except at the central rotor, and it satisfies
$$ \tilde U=\exp(\tilde H)\exp(\tilde K). $$
The two exponents lie in the two summands of the Cartan decomposition of the Lie algebra,
$$ \mathbb{B}=\mathbb{R}e_t\ \oplus\ i\mathbb{R}e_t\ \oplus\ \mathfrak{p}\ \oplus\ \mathfrak{k}, \qquad \mathfrak{p}=i\{\text{real pure vectors}\}, \qquad \mathfrak{k}=\{\text{real pure vectors}\}, $$
of real dimensions $1+1+3+3=8$. The boost exponent is in $\mathfrak{p}$ and the rotor exponent in $\mathfrak{k}$.
The Single Logarithm
A single logarithm of $\tilde Q$ is an element $\operatorname{Log}\tilde Q$ with $\exp(\operatorname{Log}\tilde Q)=\tilde Q$. Since $\mathbb{B}\cong M_2(\mathbb{C})$ and every invertible complex matrix has a logarithm, the exponential is surjective onto the invertibles. A single logarithm is related to the Cartan logarithm by
$$ \operatorname{Log}\tilde Q=\ln \varrho+i\theta_{\mathrm{pha}}+\operatorname{Log}\bigl(\tilde U_{\mathrm b}\tilde u_{\mathrm r}\bigr), $$
where the last term is a logarithm inside $\mathfrak{gl}(2,\mathbb{C})$, not inside $\mathfrak{sl}(2,\mathbb{C})$; its trace may be a nonzero multiple of $2\pi i$.
The formal exponent is the sum of the four exponents,
$$ \tilde Q_{\log}=\ln \varrho+i\theta_{\mathrm{pha}}+i\frac{\theta_{\mathrm{rpd}}}{2}\vec u_{\mathrm b}+\frac{\theta_{\mathrm{axr}}}{2}\vec u_{\mathrm r}. $$
It is not the single logarithm in general. The log defect is the Baker–Campbell–Hausdorff correction, whose first term is
$$ \frac12[\tilde H,\tilde K] =\frac{i\theta_{\mathrm{rpd}}\theta_{\mathrm{axr}}}{4}\,\vec u_{\mathrm b}\times\vec u_{\mathrm r} \in\mathfrak{p}. $$
The first correction is again a boost. The formal exponent is a logarithm exactly when this correction and all higher ones vanish. In the normal case the axes are parallel, the commutator vanishes, and the formal exponent is a logarithm. In the non-normal case the correction is nonzero, and the exponential of the formal exponent misses the element; the worked examples give defects $0.5$ and $0.4222$.
The Fibre
The exponential is not injective. Its kernel contains the central periods $2\pi ik\,e_t$, the pure periods $2\pi\vec U$ for real unit vectors, and the mixed elements $\pi i\,e_t+\pi\vec U$. The fibre over an element is not a coset of the kernel, because the exponential is not a homomorphism.
The polar parameters do not cut the fibre of the exponential to one point: the four exponents sum to the formal exponent, which is not a logarithm. What they cut to one point, up to the axis degeneracies, is the fibre of the word map
$$ (\varrho,\theta_{\mathrm{pha}},\theta_{\mathrm{rpd}},\vec u_{\mathrm b},\theta_{\mathrm{axr}},\vec u_{\mathrm r}) \longmapsto \varrho e^{i\theta_{\mathrm{pha}}}\exp\Bigl(i\frac{\theta_{\mathrm{rpd}}}{2}\vec u_{\mathrm b}\Bigr) \exp\Bigl(\frac{\theta_{\mathrm{axr}}}{2}\vec u_{\mathrm r}\Bigr). $$
The word is a section of that map, not of the exponential. It is a global coordinate system on the group of invertibles, and its domain is the complement of the null cone.
The Exact Logarithm at the Wall
The special feature of the sector is that the true logarithm is not a transcendental object: the unipotent factor has a truncated logarithm.
Theorem (the exact logarithm). For $\tilde Q=Q_t(e_t+\vec\Upsilon)\in\Sigma_{10}$,
$$ \boxed{\operatorname{Log}\tilde Q=\log Q_t+\vec\Upsilon}, $$
where $\log Q_t$ is any logarithm of the nonzero complex number $Q_t$ and the principal choice is $\ln|Q_t|+i\arg Q_t$. The right-hand side is exact, with no series and no correction.
Proof. The exponential of a nilpotent element truncates:
$$ \exp(\vec\Upsilon)=e_t+\vec\Upsilon, $$
because $\vec\Upsilon^2=0$. The scalar $Q_t$ commutes with $\vec\Upsilon$, so
$$ \exp(\log Q_t+\vec\Upsilon)=Q_t(e_t+\vec\Upsilon)=\tilde Q. $$
The logarithm is the sum of the scalar logarithm and the nilpotent vector part. The period $2\pi ik\,e_t$ can be added to the scalar logarithm, and no other ambiguity is present on the Jordan part: the logarithm is unique modulo the central periods, because the nilpotent part is fixed by $\tilde Q$.
The operator form of this single exponent is in §The Exponential at the Wall of One-Sided Operators on the N-Regular Sector and of Two-Sided Operators on the N-Regular Sector: the exponential of the left multiplication is the shear $\exp(L_{\tilde Q})=e^{Q_t}(I+Q_tL_{\vec\Upsilon})$, and the two-sided operator is the null rotation $\mathrm{Sd}_{\tilde Q}=|Q_t|^2e^{A+B}$.
The Formal Exponent and the Defect at the Wall
The formal exponent is the sum of the four exponents,
$$ \tilde Q_{\log}=\ln \varrho+i\theta_{\mathrm{pha}}+i\frac{\theta_{\mathrm{rpd}}}{2}\vec u_{\mathrm b}+\frac{\theta_{\mathrm{axr}}}{2}\vec u_{\mathrm r}. $$
It is not the logarithm. The true logarithm is
$$ \operatorname{Log}\tilde Q=\ln \varrho+i\theta_{\mathrm{pha}}+i\pi\frac{1-\varepsilon}{2}+\vec\Upsilon, $$
because $Q=\varrho\,e^{i\theta_{\mathrm{pha}}}\tilde U$ with $\operatorname{Log}\tilde U=\vec\Upsilon+i\pi\frac{1-\varepsilon}{2}$, so $\operatorname{Log}\tilde Q=\log\varrho\,e^{i\theta_{\mathrm{pha}}}+\operatorname{Log}\tilde U$; equivalently, $\log Q_t=\log\varrho\,e^{i\theta_{\mathrm{pha}}}+i\pi\frac{1-\varepsilon}{2}$ and the unipotent part adds $\vec\Upsilon$. The log defect is therefore explicit:
$$ \boxed{\; \operatorname{Log}\tilde Q-\tilde Q_{\log} =i\pi\frac{1-\varepsilon}{2} +i\Bigl(\varepsilon\tanh\frac{\theta_{\mathrm{rpd}}}{2}-\frac{\theta_{\mathrm{rpd}}}{2}\Bigr)\vec u_{\mathrm b} +\Bigl(\varepsilon\tan\frac{\theta_{\mathrm{axr}}}{2}-\frac{\theta_{\mathrm{axr}}}{2}\Bigr)\vec u_{\mathrm r} +i\varepsilon\tanh\frac{\theta_{\mathrm{rpd}}}{2}\tan\frac{\theta_{\mathrm{axr}}}{2}\;\vec u_{\mathrm b}\times\vec u_{\mathrm r}. \;} $$
The defect is the norm of $\exp(\tilde Q_{\log})-\tilde Q$, not the norm of the log defect, because the exponential is not linear. The log defect shows the three corrections: the scalar period, the nonlinear terms $\varepsilon\tanh\frac{\theta_{\mathrm{rpd}}}{2}-\frac{\theta_{\mathrm{rpd}}}{2}$ and $\varepsilon\tan\frac{\theta_{\mathrm{axr}}}{2}-\frac{\theta_{\mathrm{axr}}}{2}$, and the cross term. The numerical defect is computed in the worked examples.
On the Jordan part the cross term is nonzero, because $\theta_{\mathrm{rpd}}>0$, $\theta_{\mathrm{axr}}\notin\{0,\pi,2\pi\}$ and the axes are orthogonal. The formal exponent is therefore not the logarithm. The cross term has the shape of the first correction of the Baker–Campbell–Hausdorff series,
$$ \frac12[\tilde H,\tilde K] =\frac{i\theta_{\mathrm{rpd}}\theta_{\mathrm{axr}}}{4}\,\vec u_{\mathrm b}\times\vec u_{\mathrm r}, $$
with the sign of $\varepsilon$; the exact log defect also contains the nonlinear terms $\varepsilon\tanh\frac{\theta_{\mathrm{rpd}}}{2}-\frac{\theta_{\mathrm{rpd}}}{2}$ and $\varepsilon\tan\frac{\theta_{\mathrm{axr}}}{2}-\frac{\theta_{\mathrm{axr}}}{2}$.
On the scalar line the cross term is absent, and the formal exponent is a logarithm: for $\varepsilon=+1$ it is $\ln \varrho+i\theta_{\mathrm{pha}}$, and for $\varepsilon=-1$ it is $\ln \varrho+i\theta_{\mathrm{pha}}+\pi\vec u_{\mathrm r}$, whose exponential is $-\varrho\,e^{i\theta_{\mathrm{pha}}}$ for every invertible $\vec u_{\mathrm r}$.
The Group Law
The product of two words is
$$ \bigl(r_1e^{i\theta_{\mathrm{pha},1}}e^{\tilde H_1}e^{\tilde K_1}\bigr) \bigl(r_2e^{i\theta_{\mathrm{pha},2}}e^{\tilde H_2}e^{\tilde K_2}\bigr) =r_1r_2e^{i(\theta_{\mathrm{pha},1}+\theta_{\mathrm{pha},2})}e^{\tilde H_1}e^{\tilde K_1}e^{\tilde H_2}e^{\tilde K_2}. $$
The central factors add. The non-central ones do not: the middle product $e^{\tilde K_1}e^{\tilde H_2}$ must be re-split as a boost times a rotor. The first correction is a commutator, so the rotor of the first word conjugates the boost of the second; the parameters of the product are not the sums of the parameters.
The brackets of the two summands close as a symmetric pair.
| Bracket | Lies in | Reading |
|---|---|---|
| $[\mathfrak{k},\mathfrak{k}]$ | $\mathfrak{k}$ | two rotors compose to a rotor |
| $[\mathfrak{p},\mathfrak{p}]$ | $\mathfrak{k}$ | two boosts compose to a boost times a rotor |
| $[\mathfrak{k},\mathfrak{p}]$ | $\mathfrak{p}$ | a rotor conjugates a boost to a boost |
The second row is the Thomas–Wigner rotation. The third is the statement that the boost set is a symmetric space and not a group: the commutator of an infinitesimal rotor and an infinitesimal boost is a boost, so the boost directions are stable under the adjoint action of the rotor, while the boost set itself is not closed under multiplication. The first row is the subgroup $SU(2)$ of the rotor.
| Data | Law |
|---|---|
| scale and phase | multiply |
| boost | conjugated by the rotor, not added |
| rotor | multiplies |
| exponent | does not add: the Baker–Campbell–Hausdorff correction |
| group | $\mathbb{B}^{\times}\cong\mathrm{GL}_2(\mathbb{C})$ |
The Product at the Wall
The sector is closed under inversion:
$$ \tilde Q^{-1}=Q_t^{-1}\bigl(e_t-\vec\Upsilon\bigr). $$
It is not closed under multiplication. The product of two elements is
$$ \tilde Q_1\tilde Q_2 =Q_{t1}Q_{t2}\bigl(e_t+\vec\Upsilon_1\bigr)\bigl(e_t+\vec\Upsilon_2\bigr) =Q_{t1}Q_{t2}\bigl(e_t+\vec\Upsilon_1+\vec\Upsilon_2+\vec\Upsilon_1\vec\Upsilon_2\bigr). $$
Let
$$ \sigma=\vec\Upsilon_1\cdot\vec\Upsilon_2. $$
The scalar part of the product is $Q_{t1}Q_{t2}(1-\sigma)$, and the product has norm one, so its vector form is
$$ \boxed{B^2(\tilde Q_1\tilde Q_2)=Q_{t1}^2Q_{t2}^2\,\sigma(2-\sigma)}. $$
The product lies in $\Sigma_{10}$ exactly when $\sigma=0$ or $\sigma=2$; otherwise it lies in $\Sigma_{11}$.
Proof. Write the product as $Q_{t1}Q_{t2}W$, with $W=(1+\vec\Upsilon_1)(1+\vec\Upsilon_2)$. The scalar part of $W$ is $1-\vec\Upsilon_1\cdot\vec\Upsilon_2=1-\sigma$, because the dot part of $\vec\Upsilon_1\vec\Upsilon_2$ is $-\vec\Upsilon_1\cdot\vec\Upsilon_2$ and the cross part is a pure vector. The norm of $W$ is one, because it is the product of two unipotent elements of norm one. An element of norm one with scalar part $1-\sigma$ has vector form $1-(1-\sigma)^2=\sigma(2-\sigma)$; the central scalar $Q_{t1}Q_{t2}$ multiplies the vector form by $(Q_{t1}Q_{t2})^2$.
Example. Take $u_1=e_t+\vec e_x+i\vec e_y$ and $u_2=e_t+\vec e_y+i\vec e_z$. Then $\sigma=i$, and
$$ u_1u_2=(1-i)e_t+\vec e_y+(1+i)\vec e_z, $$
with $B^2=1+2i\neq0$. The product has left the sector. Take instead $u_1$ and $u_2=e_t+\vec e_x-i\vec e_y$; then $\sigma=2$, and
$$ u_1u_2=-e_t+2\vec e_x-2i\vec e_z, $$
with $B^2=0$; the product stays. The two cases are the two roots of $\sigma(2-\sigma)=0$.
The Matrix Picture
The algebra is isomorphic to the complex matrix algebra,
$$ \mathsf M:\mathbb B\longrightarrow M_2(\mathbb C),\qquad \mathsf M(e_t)=I,\qquad \mathsf M(\vec e_k)=-i\sigma_k, $$
with $\sigma_k$ the Pauli matrices, and the isomorphism is one of complex algebras, not only of real ones ($M_2(\mathbb{C})$ Pauli Representation). The determinant is the biquaternion norm,
$$ \det\mathsf M(\tilde Q)=\langle\tilde Q\rangle_{\natural}, $$
and the norm-one group is the special linear group.
The word becomes
$$ M=\varrho\,e^{i\theta_{\mathrm{pha}}}\,M_{\mathrm b}\,M_{\mathrm r},\qquad M_{\mathrm b}=M_{\mathrm b}^{\dagger},\quad M_{\mathrm b}>0,\quad \det M_{\mathrm b}=1,\quad M_{\mathrm r}\in SU(2), $$
and the matrix polar decomposition groups it into two:
$$ M=P\,V,\qquad P=(MM^{\dagger})^{1/2}=\varrho\,M_{\mathrm b},\qquad V=e^{i\theta_{\mathrm{pha}}}\,M_{\mathrm r}. $$
The positive factor is the scale times the boost, of determinant $\varrho^{2}$; the $*$-unitary factor is the phase times the rotor, of determinant $e^{2i\theta_{\mathrm{pha}}}$. The matrix decomposition does not see the split of $P$ into scale and boost, nor that of $V$ into phase and rotor: those splits use the fact that $P$ is a positive real multiple of an element of the algebra of determinant one, and that $V$ is a unit complex multiple of a real quaternion.
| factor | matrix | the determinant of the image | action |
|---|---|---|---|
| scale | $rI$, $\varrho>0$ | $\varrho^{2}$ | dilation of ratio $\varrho$ |
| central phase | $e^{i\theta_{\mathrm{pha}}} I$ | $e^{2i\theta_{\mathrm{pha}}}$ | central multiplication |
| boost | $M_{\mathrm b}$, Hermitian positive | $1$ | congruence $X\mapsto M_{\mathrm b}XM_{\mathrm b}^{\dagger}$ |
| rotor | $M_{\mathrm r}\in SU(2)$ | $1$ | conjugation $X\mapsto M_{\mathrm r}XM_{\mathrm r}^{-1}$ |
The spectral reading is the diagonalization of the matrix. The matrix $\mathsf M(\tilde Q)$ has the two distinct eigenvalues $\lambda_\pm$ and is diagonalizable; the polar factors are diagonal in an eigenbasis exactly when the matrix is normal, which is the matrix form of §The Normal Case.
The Two Summands
The four exponents sit in the two summands of
$$ \mathfrak{gl}_2(\mathbb{C})=\mathfrak{u}(2)\oplus\mathfrak{H}, \qquad \mathfrak{u}(2)=\{X:X^\dagger=-X\}, \qquad \mathfrak{H}=\{X:X^\dagger=X\}, $$
of real dimensions $4$ and $4$. The scale and the boost lie in $\mathfrak H$; the phase and the rotor lie in $\mathfrak u(2)$. The central real line is the scale, the central imaginary line is the phase, and the traceless parts are the boost and the rotor.
The Singular Values
The two scales of the matrix are $\varrho e^{\theta_{\mathrm{rpd}}/2}$ and $\varrho e^{-\theta_{\mathrm{rpd}}/2}$, each with multiplicity one in the regular sector. Their product is $\varrho^2=|\varrho\,e^{i\theta_{\mathrm{pha}}}|^2$, and their ratio is $e^{\theta_{\mathrm{rpd}}}$. The determinant is $N=\varrho^{2}e^{2i\theta_{\mathrm{pha}}}$, of modulus $\varrho^2$.
The Picture at the Wall
On the Jordan part, $\mathsf M(\vec\Upsilon)$ is a nonzero nilpotent matrix, so
$$ \mathsf M(\tilde Q)=Q_t\bigl(I+\mathcal N\bigr), \qquad \mathcal N=\mathsf M(\vec\Upsilon), \qquad \mathcal N^2=0, \qquad \mathcal N\neq0. $$
The matrix is $Q_t$ times a single unipotent Jordan block. Its eigenvalue is $Q_t$, with algebraic multiplicity two and geometric multiplicity one; its determinant is $Q_t^2$; its two singular values are $\varrho e^{\theta_{\mathrm{rpd}}/2}$ and $\varrho e^{-\theta_{\mathrm{rpd}}/2}$. The polar factors are the positive and $*$-unitary factors of the block, and the $*$-unitary factor is the rotor. The sign $\varepsilon$ is the sign of the scalar part of the unit block.
The spectral reading is degenerate in the matrix picture: the matrix has no eigenbasis, the two eigenvalues coincide, and the diagonal form does not exist. The polar decomposition still exists, because it is the Cartan decomposition of $SL(2,\mathbb{C})$, which does not require diagonalizability. This is the matrix form of the contrast between the two cases.
The Operator Reading: the Twist
The word also acts. On the module $S=\mathbb{C}^2$ left multiplication is
$$ \psi\longmapsto\mathsf M(\tilde Q)\,\psi=\varrho\,e^{i\theta_{\mathrm{pha}}}\,\mathsf M(\tilde U)\,\psi , $$
with $\mathsf M(\tilde U)\in SL_2(\mathbb{C})$. The scalar $\varrho\,e^{i\theta_{\mathrm{pha}}}$ needs no module of its own: by Schur's lemma the centre acts on a simple module by a scalar, so the scale and the phase multiply the module instead of enlarging it. They produce a one-dimensional representation of the scaling group,
$$ \chi(\varrho\,e^{i\theta_{\mathrm{pha}}})=\varrho\,e^{i\theta_{\mathrm{pha}}} , $$
and the module on which the general element acts is the spinor module twisted by $\chi$, written $S\otimes\chi$. The norm-one case $\varrho\,e^{i\theta_{\mathrm{pha}}}=1$ is the untwisted case; the untwisted operators are the norm-one elements, the double cover of the Lorentz group.
A spinor is a vector of the module, not the operator and not the rotor. The word writes the operator as $\varrho\,e^{i\theta_{\mathrm{pha}}}\tilde U$; the spinor is what that operator acts on. The distinction matters because the operator has more factors than the module can distinguish: the module sees the scale and the phase only through the single scalar $\varrho\,e^{i\theta_{\mathrm{pha}}}$.
The Labels of the Twisted Representations
A character of the scaling group is
$$ \chi_{a,b}(A)=A^{a}\bar A^{b},\qquad A=\varrho e^{i\theta_{\mathrm{pha}}},\quad a,b\in\mathbb{C},\quad a-b\in\mathbb{Z}, $$
the last condition making the character single-valued on $\mathbb{C}^{\times}$. The irreducible smooth finite-dimensional modules of the group of invertibles are
$$ (j,j')_{(a,b)}=V_j\boxtimes V_{j'}\otimes\chi_{a,b}, $$
where $V_j$ is the $SL_2(\mathbb{C})$-module of spin $j$ of Biquaternion Rotations and Lorentz Transformations, §Finite-Dimensional Representations of the Lorentz Group. The central element $-I$ acts on $(j,j')$ by $(-1)^{2j+2j'}$ and through the character by $(-1)^{a-b}$, so a twisted module exists exactly when
$$ a-b\equiv 2j+2j'\pmod 2 . $$
The defining module of $GL_2(\mathbb{C})$ is $(\tfrac12,0)_{(1,0)}$: a spinor is multiplied by $A$, which is a square root of the determinant character, since $\det(AI)=A^{2}$.
Reality and unitarity read off the character. Complex conjugation exchanges the two factors and conjugates the character, $\overline{(j,j')_{(a,b)}}=(j',j)_{(b,a)}$, so a module is self-conjugate exactly when $j=j'$ and $b=a$. And $\lvert\chi_{a,b}(A)\rvert=\varrho^{\mathrm{Re}(a+b)}$, so the twist is $*$-unitary exactly when $\mathrm{Re}(a+b)=0$; even then the norm-one part contributes no nontrivial finite-dimensional $*$-unitary representation, which is the non-unitarity recorded for $SL_2(\mathbb{C})$ in the rotations article.
What the Algebra Carries
The algebra is four-dimensional over $\mathbb{C}$, and its natural self-actions carry a short list:
- Left multiplication. $\mathbb{B}\cong S\oplus S$ as a left module, so $L$ carries two copies of $(\tfrac12,0)_{(1,0)}$.
- Right multiplication. $\mathbb{B}\cong S^{*}\oplus S^{*}$ with $S^{*}\cong S$, so $R$ carries two copies of the same chirality, twisted the same way. Neither one-sided action produces the opposite handedness.
- The two-sided operator. It is irreducible on the algebra and carries $(\tfrac12,\tfrac12)_{(1,1)}$, the four-vector scaled by $\varrho^{2}$.
- Nothing higher. The modules of $M_2(\mathbb{C})$ are direct sums of copies of $S$, so every irreducible summand is the spinor and the dimension is even. The modules $(1,0)$, $(1,1)$ and $(\tfrac32,0)$ are not of that form. The tower is generated by the symmetric powers of $S$ and of $\bar S$, built in the tensor category.
The regular modules themselves are The Category of Modules in the Biquaternion Algebra; the two-sided operator that carries the four-vector is Two-Sided Operators on the N-Regular Sector.
The Algorithm
The four factors are computed in five steps.
-
The norm. Compute $\langle\tilde Q\rangle_{\natural}=\sum_\mu Q_\mu^{2}$. If it is zero, stop: the element is $\natural$-degenerate and has no four-factor word.
-
The modulus. Compute $\varrho\,e^{i\theta_{\mathrm{pha}}}$ from the branch formula of §The Two Factors. The result satisfies $\varrho^{2}e^{2i\theta_{\mathrm{pha}}}=\langle\tilde Q\rangle_{\natural}$ and lies in $\mathcal H$.
-
The scale and the phase. Put $\varrho=\lvert\langle\tilde Q\rangle_{\natural}\rvert^{1/2}$ and $\theta_{\mathrm{pha}}=\tfrac12\arg\langle\tilde Q\rangle_{\natural}$, so that $\varrho\,e^{i\theta_{\mathrm{pha}}}$ is the principal square root of the norm.
-
The unit factor. Put $\tilde U=\tilde Q/(\varrho\,e^{i\theta_{\mathrm{pha}}})$ and check $\langle\tilde U\rangle_{\natural}=1$.
-
The boost and the rotor. Compute $\tilde U\tilde U^{*}$ and $s=\operatorname{Sc}(\tilde U\tilde U^{*})$, then
$$ \tilde U_{\mathrm b}=\frac{\tilde U\tilde U^{*}+e_t}{\sqrt{2(1+s)}},\qquad \tilde u_{\mathrm r}=\frac{(2s+1)e_t-\tilde U\tilde U^{*}}{\sqrt{2(1+s)}}\;\tilde U . $$
The checks are
$$ \tilde U_{\mathrm b}^{2}=\tilde U\tilde U^{*},\qquad \tilde u_{\mathrm r}\tilde u_{\mathrm r}^{*}=e_t,\qquad \varrho\,e^{i\theta_{\mathrm{pha}}}\,\tilde U_{\mathrm b}\,\tilde u_{\mathrm r}=\tilde Q . $$
The five steps use one square root of a complex number and one of a real number, and no spectral decomposition.
Worked Example
Take
$$ \tilde Q=(1+i)e_t+\vec e_x+i\vec e_y . $$
The norm is
$$ \langle\tilde Q\rangle_{\natural}=(1+i)^{2}+1+i^{2}=2i+1-1=2i, $$
so $\lvert N\rvert=2$, $a=0$, $b=2$, and the principal root is $\varrho\,e^{i\theta_{\mathrm{pha}}}=1+i$. The scale and the phase are
$$ \varrho=\sqrt2,\qquad \theta_{\mathrm{pha}}=\frac{\pi}{4},\qquad e^{i\theta_{\mathrm{pha}}}=\frac{1+i}{\sqrt2}. $$
The unit factor is
$$ \tilde U=\frac{\tilde Q}{1+i}=e_t+\frac{1-i}{2}\vec e_x+\frac{1+i}{2}\vec e_y, $$
with norm one. The Hermitian square is
$$ \tilde U\tilde U^{*}=2e_t-i\vec e_x+i\vec e_y+i\vec e_z,\qquad s=2,\qquad 2(1+s)=6, $$
and the boost and the rotor are
$$ \tilde U_{\mathrm b}=\frac{3e_t-i\vec e_x+i\vec e_y+i\vec e_z}{\sqrt6},\qquad \tilde u_{\mathrm r}=\sqrt{\frac23}\,e_t+\frac{1}{\sqrt6}\vec e_x+\frac{1}{\sqrt6}\vec e_y . $$
The four factors are collected in the table.
| factor | value | quaternionic norm |
|---|---|---|
| $\varrho$ | $\sqrt2$ | $\varrho^{2}=2$ |
| $\theta_{\mathrm{pha}}$ | $\pi/4$ | $e^{2i\theta_{\mathrm{pha}}}=i$ |
| $\tilde U_{\mathrm b}$ | $(3e_t-i\vec e_x+i\vec e_y+i\vec e_z)/\sqrt6$ | $1$ |
| $\tilde u_{\mathrm r}$ | $\sqrt{2/3}\,e_t+\vec e_x/\sqrt6+\vec e_y/\sqrt6$ | $1$ |
The checks hold: $\tilde U_{\mathrm b}^{2}=\tilde U\tilde U^{*}$, $\tilde u_{\mathrm r}\tilde u_{\mathrm r}^{*}=e_t$, and $\varrho e^{i\theta_{\mathrm{pha}}}\tilde U_{\mathrm b}\tilde u_{\mathrm r}=\tilde Q$. The Hermitian factor is $\tilde Q_H=\varrho\tilde U_{\mathrm b}$, of norm $\varrho^{2}=2$, and the $*$-unitary factor is $\tilde Q_0=e^{i\theta_{\mathrm{pha}}}\tilde u_{\mathrm r}$, of norm $e^{2i\theta_{\mathrm{pha}}}=i$; their product has norm $2i$, the norm of the element.
The example is on the wall, $B^{2}=1+i^{2}=0$: it is an element of $\Sigma_{10}$, and its word is the unipotent case of §The Unipotent Unit Factor. An off-wall case is checked in §The Result in Terms of the Coordinates, and sixteen further cases, off the wall and on it, are in Worked Examples of the Complete Polar Decomposition of an N-Regular Biquaternion.
The Two Walls
The sector is the complement of the union of two cones.
| Wall | Equation | Stratum | What happens to the word |
|---|---|---|---|
| null cone | $N=0$ | $\Sigma_{01}\cup\Sigma_{00}$ | $\varrho\to0$, $\theta_{\mathrm{rpd}}\to\infty$ |
| spectral cone | $B^2=0$ | $\Sigma_{10}\cup\Sigma_{00}$ | word finite, eigenbasis degenerates |
The two cones meet at $\Sigma_{00}$, the nilpotent stratum. The two walls behave differently, and the difference is the subject of this section.
The Null Cone
On the null cone the modulus vanishes and no unit factor exists. A nonzero element of the cone is a zero divisor of rank one. As an element of the sector approaches the cone with $B^2$ bounded away from zero, the rapidity diverges and the scale collapses. The standard family is
$$ \tilde Q(\theta_{\mathrm{rpd}})=R e^{-\theta_{\mathrm{rpd}}/2}\,\omega\exp\Bigl(i\frac{\theta_{\mathrm{rpd}}}{2}\vec u_{\mathrm b}\Bigr)\tilde u_{\mathrm r}, \qquad \theta_{\mathrm{rpd}}>0, $$
with $R>0$, $|\omega|=1$ and unit axes. Every member is invertible, with modulus $R e^{-\theta_{\mathrm{rpd}}/2}$, and its word has rapidity $\theta_{\mathrm{rpd}}$. The limit is
$$ \tilde Q(\theta_{\mathrm{rpd}})\longrightarrow \frac{R\omega}{2}\bigl(e_t+i\vec u_{\mathrm b}\bigr)\tilde u_{\mathrm r}, \qquad \theta_{\mathrm{rpd}}\to\infty, $$
a nonzero null element of rank one. It is nilpotent exactly when $\tilde u_{\mathrm r}$ anticommutes with $\vec u_{\mathrm b}$, and it has nonzero scalar part otherwise. The scale vanishes as $e^{-\theta_{\mathrm{rpd}}/2}$ and the boost diverges as $e^{\theta_{\mathrm{rpd}}/2}$; the product stays finite.
The Spectral Cone
On the spectral cone the word does not diverge. Let $\tilde Q_k\to\tilde Q_0$ with $N_k\to N_0\neq0$ and $B_k^2\to0$. The modulus converges, the unit factor converges, and the Cartan factors converge, because the Cartan decomposition is continuous. Hence the polar data have a limit: the limiting word is the word of $\tilde Q_0$, which lies in $\Sigma_{10}$ and is computed in §The Unipotent Unit Factor.
The eigenbasis does not survive. The root is
$$ \vec{Ζ}=-i\frac{\vec Q}{B}, $$
and $B\to0$ while $\vec Q\to\vec Q_0\neq0$ on the spectral cone, so $|\vec{Ζ}|\to\infty$. The idempotents
$$ \tilde\Pi_\pm=\frac{e_t\pm\vec{Ζ}}{2} $$
diverge, and the spectral decomposition has no limit. The eigenvalues $\lambda_\pm=Q_t\pm iB$ coalesce, the minimal polynomial becomes a square, and the element becomes defective.
Proposition (the word survives the spectral wall, the eigenbasis does not). Let $\tilde Q_k\in\Sigma_{11}$ converge to $\tilde Q_0\in\Sigma_{10}$ with $N_0\neq0$ and $\vec Q_0\neq0$, so that the limit lies in the Jordan part. Then $\tilde U_k\to\tilde U_0$, $\tilde U_{\mathrm b,k}\to\tilde U_{\mathrm b,0}$ and $\tilde u_{\mathrm r,k}\to\tilde u_{\mathrm r,0}$, so the word converges to the word of $\tilde Q_0$. The root $\vec{Ζ}_k$ diverges in norm, and no spectral idempotent of $\tilde Q_k$ converges to a nonzero idempotent.
Proof. The modulus is continuous in $N$ away from zero, and the Cartan factors are continuous functions of $\tilde U$; this gives the convergence of the word. The root satisfies $|\vec{Ζ}_k|=|\vec Q_k|/|B_k|\to\infty$ because $\vec Q_k\to\vec Q_0\neq0$ and $B_k\to0$. An idempotent of the form $(e_t\pm\vec{Ζ}_k)/2$ is unbounded, so it has no finite limit.
At the scalar line the limit is different. If $\tilde Q_0=Q_0e_t$ is a nonzero scalar, then $\vec Q_0=0$, the root is $0/0$, and its limit depends on the path: the family $Q_k=1+\epsilon_k\vec e_x$ has $\vec{Ζ}_k=-i\vec e_x$, so the idempotents converge to nonzero idempotents of the ambient algebra even though the limit is scalar. The eigenvalues coalesce, but the eigenbasis need not diverge. The proposition is therefore stated for the Jordan part, which is the part of the wall that the sector meets transversally.
The contrast is the exact content of the difference between the two readings. The polar decomposition is a decomposition of the group of invertibles, and it is continuous across the spectral wall. The spectral decomposition is a decomposition of the element, and it degenerates there. The sector $\Sigma_{11}$ is the open set on which both exist; its wall $B^2=0$ is where only the polar one survives.
The Wall as a Space
The wall $\Sigma_{10}$ is closed under inversion but not under multiplication. It is a complex hypersurface of $\mathbb{B}$, not an open set: its closure is the spectral cone $B^2=0$, and the points of that closure with $N=0$ form the nilpotent stratum $\Sigma_{00}$.
The Nilpotent Stratum
As $Q_t\to0$ with $\vec\Upsilon$ fixed, the element tends to zero and the modulus tends to zero. The unit factor does not degenerate:
$$ \tilde U=\varepsilon\bigl(e_t+\vec\Upsilon\bigr) $$
is independent of $Q_t$. The scale collapses, the phase follows the argument of $Q_t$, and the unipotent shape survives. This is the structural difference from the regular sector, where the rapidity diverges as the modulus vanishes. In $\Sigma_{10}$ the modulus is the scalar part, and the only thing that vanishes is the scalar.
A nonzero nilpotent of $\Sigma_{00}$ is reached by letting $Q_t\to0$ and $\vec\Upsilon\to\infty$ at the rate $Q_t\vec\Upsilon\to\vec N$ with $\vec N^2=0$. The unit factor diverges, the word parameters leave every bounded set, and the limit is the nilpotent. The sector is not compact, and the angle $\theta_{\mathrm{axr}}\to\pi$ is the direction in which it is unbounded.
The Spectral Wall From Inside
Seen from $\Sigma_{11}$, the locus $B^2=0$ is a wall: the word converges to the word of the limit, as §The Spectral Cone proves, while the eigenbasis diverges and the spectral decomposition disappears. Seen from $\Sigma_{10}$, the same locus is the ambient space. The two descriptions are the two sides of one hypersurface: defective elements, no eigenbasis, and a polar word that is still defined.
| Approach | Locus | What happens |
|---|---|---|
| nilpotent stratum | $N=0$, $B^2=0$ | $Q_t\to0$; the unit factor survives if $\vec\Upsilon$ is fixed; the scale vanishes |
| spectral wall from $\Sigma_{11}$ | $B^2=0$, $N\neq0$ | the word converges; the eigenbasis diverges |
Summary
The complete polar decomposition of an $\natural$-regular biquaternion is the modulus–unit form refined factor by factor into four, each factor an exponential of one real parameter; it is read against the spectral decomposition and, on the spinor module, as a twisted representation.
| item | result |
|---|---|
| domain | $\langle\tilde Q\rangle_{\natural}\neq0$ |
| Cartan pair | $\tilde Q=\tilde Q_H\tilde Q_0$, $\tilde Q_H$ positive definite, $\tilde Q_0$ $*$-unitary |
| modulus–unit | $\tilde Q=\varrho\,e^{i\theta_{\mathrm{pha}}}\tilde U$, $\varrho\,e^{i\theta_{\mathrm{pha}}}\in\mathcal H$, $\langle\tilde U\rangle_{\natural}=1$ |
| modulus | $\varrho^{2}e^{2i\theta_{\mathrm{pha}}}=\langle\tilde Q\rangle_{\natural}$, $\varrho\,e^{i\theta_{\mathrm{pha}}}\in\mathcal H$ |
| scale and phase | $\varrho=\lvert\langle\tilde Q\rangle_{\natural}\rvert^{1/2}$, $\theta_{\mathrm{pha}}=\tfrac12\arg\langle\tilde Q\rangle_{\natural}$ |
| boost | $\tilde U_{\mathrm b}=(\tilde U\tilde U^{*})^{1/2}$, Hermitian positive definite, norm one |
| rotor | $\tilde u_{\mathrm r}=\tilde U_{\mathrm b}^{-1}\tilde U$, unit real quaternion |
| word | $\tilde Q=\varrho e^{i\theta_{\mathrm{pha}}}\tilde U_{\mathrm b}\tilde u_{\mathrm r}$ |
| dimensions | $8=2+6=1+1+3+3$ |
| two readings | $\tilde Q_H=\varrho\tilde U_{\mathrm b}$, $\tilde Q_0=e^{i\theta_{\mathrm{pha}}}\tilde u_{\mathrm r}$ |
| spinor module | $S\otimes\chi$ with $\chi(\varrho\,e^{i\theta_{\mathrm{pha}}})=\varrho\,e^{i\theta_{\mathrm{pha}}}$; the irreducible modules $(j,j')_{(a,b)}$, $a-b\equiv2j+2j'\pmod2$ |
| group of invertibles | $SL_2(\mathbb C)\cong H^{3}\times S^{3}$ |
| group of invertibles | $\mathbb B^{\times}\cong GL_2(\mathbb C)$, maximal compact $U(2)$, fundamental group $\mathbb Z$ |
| pairings | quaternion, Hamilton root, Cartan Hermitian; the first two partial, the last total |
| closed forms | $\tilde U_{\mathrm b}=(\tilde U\tilde U^{*}+e_t)/\sqrt{2(1+s)}$, $\tilde u_{\mathrm r}=\bigl((2s+1)e_t-\tilde U\tilde U^{*}\bigr)\tilde U/\sqrt{2(1+s)}$ |
| component form | $\tilde U=(c_{\mathrm b}c_{\mathrm r}-is_{\mathrm b}s_{\mathrm r}c)e_t+is_{\mathrm b}c_{\mathrm r}\vec u_{\mathrm b}+c_{\mathrm b}s_{\mathrm r}\vec u_{\mathrm r}+is_{\mathrm b}s_{\mathrm r}\vec u_{\mathrm b}\times\vec u_{\mathrm r}$ |
| eigenvalues | $\lambda_\pm=Q_t\pm iB$ |
| eigenbasis | $\vec{Ζ}=-i\vec Q/B$, $\tilde\Pi_\pm=\frac12(e_t\pm\vec{Ζ})$ |
| unit-factor eigenvalues | $\mu_\pm=\lambda_\pm/(\varrho\,e^{i\theta_{\mathrm{pha}}})=U_t\pm i\sqrt{1-U_t^2}$ |
| normal case | $\mu_\pm=e^{\mp\theta_{\mathrm{rpd}}/2}e^{\pm i\theta_{\mathrm{axr}}/2}$, axes $\pm i\vec{Ζ}$ |
| normal criterion | $\vec Q\times\overline{\vec Q}=0$ |
| spectral wall | $B^2=0\iff U_t=\pm1$ |
| explicit data | §The Result in Terms of the Coordinates: boxes for $\varrho,\theta_{\mathrm{pha}},\theta_{\mathrm{rpd}},\vec u_{\mathrm b},\tilde U_{\mathrm b},\tilde u_{\mathrm r},\theta_{\mathrm{axr}},\vec u_{\mathrm r}$ |
| domain, wall | $N\neq0$, $B^2=0$ |
| normal form | $\tilde Q=Q_t(e_t+\vec\Upsilon)$, $\vec\Upsilon^2=0$ |
| unit factor, wall | $\tilde U=\varepsilon(e_t+\vec\Upsilon)$, $\varepsilon=Q_t/(\varrho\,e^{i\theta_{\mathrm{pha}}})=\pm1$ |
| scalar constraint | $\cosh\frac{\theta_{\mathrm{rpd}}}{2}\cos\frac{\theta_{\mathrm{axr}}}{2}=\varepsilon$ |
| axis constraint | $\sinh\frac{\theta_{\mathrm{rpd}}}{2}\sin\frac{\theta_{\mathrm{axr}}}{2}\,\vec u_{\mathrm b}\cdot\vec u_{\mathrm r}=0$ |
| Jordan part | $\theta_{\mathrm{rpd}}>0$, $\theta_{\mathrm{axr}}\notin\{0,\pi,2\pi\}$, $\vec u_{\mathrm b}\perp\vec u_{\mathrm r}$ |
| angle relation | $\bigl|\cos\frac{\theta_{\mathrm{axr}}}{2}\bigr|=1/\cosh\frac{\theta_{\mathrm{rpd}}}{2}$ |
| closed form, wall | $\tilde Q=Q_t\bigl(e_t+i\tanh\frac{\theta_{\mathrm{rpd}}}{2}\vec u_{\mathrm b}+\tan\frac{\theta_{\mathrm{axr}}}{2}\vec u_{\mathrm r}+i\tanh\frac{\theta_{\mathrm{rpd}}}{2}\tan\frac{\theta_{\mathrm{axr}}}{2}\vec u_{\mathrm b}\times\vec u_{\mathrm r}\bigr)$ |
| exact logarithm | $\operatorname{Log}\tilde Q=\log Q_t+\vec\Upsilon$ |
| product criterion | $B^2(\tilde Q_1\tilde Q_2)=Q_{t1}^2Q_{t2}^2\sigma(2-\sigma)$, $s=\vec\Upsilon_1\cdot\vec\Upsilon_2$ |
| dimension, wall | $6$: $Q_t$ (2), angles (2), frame (3), angle relation (1) |
The Comparison
| Property | Spectral reading | Polar reading |
|---|---|---|
| decomposition of | the element $\tilde Q$ | the unit factor $\tilde U$ |
| factors | two idempotents | positive Hermitian times $*$-unitary |
| coefficients | $\lambda_\pm$ | exponentials of $\theta_{\mathrm{rpd}},\theta_{\mathrm{axr}}$ |
| diagonal in the eigenbasis | always | exactly when normal |
| survives $B^2\to0$ | no on the Jordan part | yes |
| real dimension | $8$ | $8$ |
| shared data | $\varrho,\theta_{\mathrm{pha}},U_t$ | $\varrho,\theta_{\mathrm{pha}},U_t$ |
| The last row is the statement of the article: the spectral reading is defined on the open sector, the polar reading is defined on the whole group of invertibles. |
The four statements to carry away are these. On the $\natural$-regular set the general pair sharpens, the Hermitian factor becoming positive definite and the isometry factor $*$-unitary. The pair then splits twice, the modulus into a scale and a phase and the unit factor into a boost and a rotor, and the four factors fill the eight dimensions of the algebra. The regroupings of the four factors give the two-factor readings and the three pairings, of which the Cartan Hermitian pairing is the general polar decomposition itself. Read on the spinor module $S=\mathbb{C}^{2}$, left multiplication acts by the scalar $\varrho\,e^{i\theta_{\mathrm{pha}}}$ and the norm-one operator $\tilde U$; the scale and the phase enter through the character $\chi(\varrho\,e^{i\theta_{\mathrm{pha}}})=\varrho\,e^{i\theta_{\mathrm{pha}}}$, and the irreducible modules are the $(j,j')_{(a,b)}$ with $a-b\equiv2j+2j'\pmod2$.
What the Sector Shows
- The word exists for every element of the N-regular sector, off the wall and on it, and it is unique up to the axis degeneracies.
- The existence theorem uses only the invertibility of $\tilde Q$. The invariant $B^2$ enters through the domain and through the shape of the word, not through the existence.
- Off the wall the eigenbasis exists because the two eigenvalues are distinct. It is a spectral basis, not a polar basis. The polar factors are diagonal in it exactly for the normal elements, and those are exactly the elements whose two axes are parallel and whose word is a single exponential.
- In the non-normal case the spectrum determines the eigenvalues of the unit factor but not the rapidity, the rotor angle or the axes. The polar word carries four real data that the spectrum does not.
- On the wall the unit factor is $\pm$unipotent: its two eigenvalues coincide, its minimal polynomial is a square, and it has no eigenbasis. The word still exists, with four real data: the scalar part, one angle and a frame.
- On the wall the scalar part of the unit factor is $\pm1$. This one complex equation ties the two angles, $\cosh\frac{\theta_{\mathrm{rpd}}}{2}\cos\frac{\theta_{\mathrm{axr}}}{2}=\pm1$, and on the Jordan part it forces the two axes to be orthogonal.
- The true logarithm on the wall is exact and truncated, $\log Q_t+\vec\Upsilon$. The formal exponent is not a logarithm: the log defect is the cross term plus the nonlinear terms of the two angles.
- The word is continuous across the spectral wall; the eigenbasis is not. This is the precise sense in which the polar decomposition is a decomposition of the group and the spectral decomposition is a decomposition of the element.
Summary of Notation
| symbol | meaning |
|---|---|
| $\tilde Q$ | an $\natural$-regular biquaternion |
| $\langle\tilde Q\rangle_{\natural}=\tilde Q\tilde Q^{\natural}$ | the quaternionic norm, complex and nonzero |
| $\varrho\,e^{i\theta_{\mathrm{pha}}}$ | the modulus, the principal root of $\langle\tilde Q\rangle_{\natural}$ in $\mathcal H$ |
| $\varrho$ | the scale, $\varrho$ |
| $\theta_{\mathrm{pha}}$ | the central phase angle, $\arg(\varrho\,e^{i\theta_{\mathrm{pha}}})$, with $e^{i\theta_{\mathrm{pha}}}=\varrho\,e^{i\theta_{\mathrm{pha}}}/\varrho$ |
| $\tilde U$ | the unit factor, $\tilde Q/(\varrho\,e^{i\theta_{\mathrm{pha}}})$, of norm one |
| $\tilde U_{\mathrm b}$ | the boost, $(\tilde U\tilde U^{*})^{1/2}$ |
| $\tilde u_{\mathrm r}$ | the rotor, $\tilde U_{\mathrm b}^{-1}\tilde U$ |
| $\tilde Q_H$ | the Hermitian factor, $\varrho\tilde U_{\mathrm b}$ |
| $\tilde Q_0$ | the $*$-unitary factor, $e^{i\theta_{\mathrm{pha}}}\tilde u_{\mathrm r}$ |
| $s$ | $\operatorname{Sc}(\tilde U\tilde U^{*})$ |
| $B^{2}$ | $Q_x^{2}+Q_y^{2}+Q_z^{2}$ |
| $H^{3}$ | the Hermitian positive elements of norm one |
| $S$ | the spinor module $\mathbb{C}^{2}$ |
| $\chi$, $\chi_{a,b}$ | the characters $\chi(\varrho\,e^{i\theta_{\mathrm{pha}}})=\varrho\,e^{i\theta_{\mathrm{pha}}}$ and $\chi_{a,b}(A)=A^{a}\bar A^{b}$ |
| $(j,j')_{(a,b)}$ | the irreducible module $V_j\boxtimes V_{j'}\otimes\chi_{a,b}$ |
| $\mathcal H$ | the principal half-plane of the modulus |
| Symbol | Meaning |
| $\mathbb{B}$ | the biquaternion algebra |
| $\tilde Q=Q_te_t+\vec Q$ | a biquaternion |
| $Q_t,Q_x,Q_y,Q_z$ | the four complex coordinates |
| $N=\langle\tilde Q\rangle_{\natural}$ | the reduced product $Q_t^2+Q_x^2+Q_y^2+Q_z^2$ |
| $B^2$ | the vector form $Q_x^2+Q_y^2+Q_z^2$ |
| $\Sigma_{11}$ | the N-regular, spectrally regular sector |
| $\varrho\,e^{i\theta_{\mathrm{pha}}}=\varrho e^{i\theta_{\mathrm{pha}}}$ | the modulus, principal square root of $N$ |
| $\tilde U=\tilde Q/(\varrho\,e^{i\theta_{\mathrm{pha}}})$ | the unit factor |
| $\tilde U_{\mathrm b},\tilde u_{\mathrm r}$ | the Hermitian positive factor and the rotor, $\tilde U=\tilde U_{\mathrm b}\tilde u_{\mathrm r}$ |
| $\theta_{\mathrm{rpd}}$ | the rapidity |
| $\theta_{\mathrm{axr}}$ | the rotor angle |
| $\vec u_{\mathrm b},\vec u_{\mathrm r}$ | the two axes |
| $c=\vec u_{\mathrm b}\cdot\vec u_{\mathrm r}$ | the cosine between the axes |
| $\vec{Ζ}=-i\vec Q/B$ | the root of one carried by the element |
| $\tilde\Pi_\pm=\frac12(e_t\pm\vec{Ζ})$ | the spectral idempotents |
| $\lambda_\pm=Q_t\pm iB$ | the eigenvalues |
| $\mu_\pm=\lambda_\pm/(\varrho\,e^{i\theta_{\mathrm{pha}}})$ | the eigenvalues of the unit factor |
| $\tilde H,\tilde K$ | the Cartan pair |
| $\tilde Q_{\log}$ | the formal exponent |
| $\mathsf M(\tilde Q)$ | the matrix $Q_tI-iQ_x\sigma_x-iQ_y\sigma_y-iQ_z\sigma_z$ |
| $c_{\mathrm b},s_{\mathrm b},c_{\mathrm r},s_{\mathrm r}$ | the half-angle cosines and sines |
| $\lVert\tilde Q\rVert^2=|Q_t|^2+|\vec Q|^2$ | the Euclidean norm squared of the coordinates |
| $\vec a$ | the vector $\sinh\theta_{\mathrm{rpd}}\,\vec u_{\mathrm b}$ |
| $C,S$ | the half-angle numerators, $\cosh\frac{\theta_{\mathrm{rpd}}}{2}=C/\sqrt{|N|}$, $\sinh\frac{\theta_{\mathrm{rpd}}}{2}=S/\sqrt{|N|}$ |
| $\mathrm{Vect}\tilde X$ | the vector part of $\tilde X$ |
| $\Sigma_{10}$ | the N-regular, spectrally degenerate sector |
| $\vec\Upsilon=\vec Q/Q_t$ | the nilpotent direction on the wall, $\vec\Upsilon^2=0$ |
| $\vec A=\operatorname{Re}\vec\Upsilon$, $\vec B=\operatorname{Im}\vec\Upsilon$ | the real and imaginary parts, $\vec A\cdot\vec B=0$ |
| $a=\lVert\vec A\rVert=\lVert\vec B\rVert$ | the common length on the wall |
| $\varepsilon=Q_t/(\varrho\,e^{i\theta_{\mathrm{pha}}})=\pm1$ | the sign of the unit factor on the wall |
| $Q_\mu=q_\mu+iq'_\mu$ | the eight real coordinates |
| $\vec q=(q_x,q_y,q_z)$, $\vec q'=(q'_x,q'_y,q'_z)$ | the real and imaginary vector coordinates |
| $\sigma=\vec\Upsilon_1\cdot\vec\Upsilon_2$ | the invariant of a product on the wall |
| $a,b$ | the real and imaginary parts of $\langle\tilde Q\rangle_{\natural}$ in the branch formula of §The Two Factors |
Further Reading
- Roger Penrose, "A generalized inverse for matrices", Proceedings of the Cambridge Philosophical Society 51 (1955) 406–413. The four equations and the matrix polar decomposition.
- Gene H. Golub and Charles F. Van Loan, Matrix Computations, 4th edition (Johns Hopkins University Press, 2013). The polar decomposition, its Hermitian and $*$-unitary factors, and the numerical algorithms.
- Nicholas J. Higham, Functions of Matrices: Theory and Computation (SIAM, 2008). The matrix square root, the polar factor and their derivatives.
- Max-Albert Knus, Alexander Merkurjev, Markus Rost and Jean-Pierre Tignol, The Book of Involutions, Colloquium Publications 44 (American Mathematical Society, 1998). Algebras with involution and the Hermitian structures the factors carry.
- The Quaternion Pairing of the Complete Polar Decomposition of an N-Regular Biquaternion, The Hamilton Root Pairing of the Complete Polar Decomposition of an N-Regular Biquaternion and The Cartan Hermitian Pairing of the Complete Polar Decomposition of an N-Regular Biquaternion, for the three regroupings of the four factors.
- Biquaternion Rotations and Lorentz Transformations, §Finite-Dimensional Representations of the Lorentz Group, for the highest-weight-$2j$ modules and the non-unitarity, and §The Double Covers and the Lorentz Group, for the double cover.
- The Category of Modules, for the regular modules and the bi-module structure used in §What the Algebra Carries.
- The Polar Decomposition of a Biquaternion and The Polar Decomposition of an N-Degenerate Biquaternion, for the two-factor pair that the word refines and its behaviour on the cone.
- Relations between the Complete Polar Decomposition and Eigenbasis of an N-Regular Biquaternion, for the dictionary between the word and the eigenvalue data, and its limit on the Jordan part.
- Worked Examples of the Complete Polar Decomposition of an N-Regular Biquaternion, for the numerical cases off the wall and on it, the normal and non-normal comparison, and the defect tables.
- The Eigenbasis of a Spectrally Regular Biquaternion, for the eigenbasis, the Peirce corners and the spinors.
- The N-Regular, Spectrally Degenerate Sector, for the normal forms and the two strata of the wall.
- The Spectrum, for the four sectors and the discriminants.
- The One-Sided Operator Exponential on the N-Regular Sector, for the operator form of the exact logarithm on the wall.
- The Two-Sided Exponential on the N-Regular Sector, for the two-sided operator form.
- Biquaternion Elementary Functions, for the exponential, the logarithm, the kernel and the fibre.
- Biquaternion Lie Group and Exponential Structure, for the Cartan decomposition and the Baker–Campbell–Hausdorff series.
- Conventions in Mathematics, for the notation.