The Exponential and Its Three Pairings

Introduction

This entry reads the polar-decomposition subsection of the mathematical category. Eighteen articles, one object and one split. The object is a biquaternion written as a scale, a phase and two exponentials of a unit axis. The split is the grouping of those four factors into two, and there are exactly three ways to make it. The mathematics carries out the split twice, once on the sector where the element has two distinct eigenvalues and once on the sector where it has one.

$$ \boxed{\;\tilde Q=\varrho\,e^{i\theta_{\mathrm{pha}}}\,\tilde U_{\mathrm b}\,\tilde u_{\mathrm r},\qquad \tilde U_{\mathrm b}=\exp\Bigl(i\frac{\theta_{\mathrm{rpd}}}{2}\vec u_{\mathrm b}\Bigr),\qquad \tilde u_{\mathrm r}=\exp\Bigl(\frac{\theta_{\mathrm{axr}}}{2}\vec u_{\mathrm r}\Bigr)\;} $$

The first claim is that the complete polar decomposition is a coordinate chart on the group of invertibles and nothing more. Eight real parameters, matching the real dimension of the algebra: one for the scale, one for the phase, and two for each of the two exponentials, which are a rapidity with an axis and an angle with an axis. Every element off the null cone has exactly one word up to the degenerate directions where an axis is undefined. This is a statement about a group of transformations and not about a process: the word says what an element is at one instant and not how it got there.

The second claim is that the word survives both walls and only one of them is fatal. At the null cone $N=0$ the scale collapses and the rapidity diverges in the same breath, and the product has a finite limit that is a null element. At the spectral cone $B^{2}=0$ the word has a finite limit and the eigenbasis does not: the two eigenvalues coalesce, the minimal polynomial becomes a square, and the idempotents diverge while the Cartan factors converge. The polar decomposition is continuous across the spectral wall and the spectral decomposition is not, and that asymmetry is the sharpest mathematical statement in the section.

The third claim is that there are exactly three pairings because there are exactly three ways to bracket four factors. The quaternion pairing groups the scale with the rotor; the Hamilton pairing groups the scale with the phase; the Cartan pairing groups the scale with the boost. Each produces a first factor of a different kind — a real quaternion, a complex scalar, a Hermitian positive element — and each carries the norm differently. The three are not equivalent: two of them are partial and one is total, and the witnesses that separate their domains are single elements that can be written down.

The fourth claim is that normality is the condition under which the split into rate and frame is itself splittable. In general the two polar factors do not commute with the element, so they are not diagonal in its eigenbasis, and the eigenbasis is not a polar basis. The spectrum still determines the eigenvalues of the unit factor and stops there: it gives one complex equation for three real unknowns. Exactly on the normal set the two frames coincide and the word collapses to a single exponential. The gap is measurable and the mathematics measures it: for one worked element the normal-case formula would give a rapidity of $1.466$ where the true rapidity is $1.763$.

The fifth claim is that the degenerate sector is not a limit case but a rigid structure. On $B^{2}=0$ with $N\neq0$ the unit factor is $\pm$unipotent, its two angles are tied by one equation, its two axes are orthogonal, and the closed form reduces the four word parameters to one complex number, one angle and a frame. The logarithm is exact and truncated: $\operatorname{Log}\tilde Q=\log Q_t+\vec\Upsilon$ with $\vec\Upsilon^{2}=0$. This is the flow that cannot be split, and the impossibility is algebraic rather than numerical.

The sixth claim is that one slot carries half the angle and two slots carry the full angle. A one-sided operator is a representation, so its exponential is the exponential of the element, and its eigenvalues are $\lambda_\pm=Q_t\pm iB$ twice each. The two-sided operator is quadratic, its exponential is not the exponential of an element, and its eigenvalues are the four products $\lambda_i\bar\lambda_j$. The spinor therefore sees $e^{\pm\theta/2}$ and the vector sees $e^{\pm\theta}$, and the factor two between them is the whole of the bookkeeping that separates the two actions. The mathematics states it as a doubling rule with three cases, one for each kind of generator.

The first objection is that this is a coordinate change dressed as a decomposition. For the word, it is. The content is in the three pairings and in the two walls: which factor is invariant, which is frame, which domain each split requires, and what survives where the element stops being semisimple. A coordinate chart that is a chart on the whole group but a chart with two different shapes on the two sectors is telling you where the algebra changes kind.

The second objection is that three pairings of one word cannot all be physical. They cannot, and the entry does not claim it. Each pairing is a dictionary with a different reach: the Cartan pairing is defined on the largest set, the Hamilton pairing carries the invariant, the quaternion pairing carries the frame. A physical statement is attached to a pairing and inherits that pairing's domain, and the entry states the domain with the statement.

The third objection is that the degenerate sector is empty of content because it has no eigenbasis. It has no eigenbasis and it has a lock. On $\Sigma_{10}$ the rapidity is a function of the spin angle, the two axes are orthogonal, and the element is the scalar part times a unipotent written in an orthogonal frame. A sector with one angle of freedom instead of two is a sector with a prediction, and the corpus reads that prediction as the rest frame and the exceptional point.

The letters. The algebra is $\mathbb{B}$; the element is $\tilde Q=Q_te_t+\vec Q$; the norms are $N=\langle\tilde Q\rangle_{\natural}=Q_t^{2}+B^{2}$ and $B^{2}=\langle\vec Q,\vec Q\rangle_{\natural}$; the modulus is $\varrho\,e^{i\theta_{\mathrm{pha}}}=\sqrt N=\varrho e^{i\theta_{\mathrm{pha}}}$; the unit factor is $\tilde U=\tilde Q/(\varrho\,e^{i\theta_{\mathrm{pha}}})$. The two unit factors are the boost $\tilde U_{\mathrm b}$ of rapidity $\theta_{\mathrm{rpd}}$ and the rotor $\tilde u_{\mathrm r}$ of angle $\theta_{\mathrm{axr}}$. The eigenvalues are $\lambda_\pm=Q_t\pm iB$. The sectors are $\Sigma_{11}$ for $N\neq0$, $B^{2}\neq0$ and $\Sigma_{10}$ for $N\neq0$, $B^{2}=0$. The three pairings are named by their first factor: quaternion, Hamilton, Cartan.

Boundaries. The complete polar decomposition, with its four factors, is the subject of The Complete Polar Decomposition of an N-Regular Biquaternion; the exponential and the logarithm are Biquaternion Elementary Functions; the spectrum is Spectral Regularity and Degeneracy in the Biquaternion Algebra; the sectors are the four sector articles. The physical dictionary of the pairings is the companion entry The Polar Decomposition and Its Three Pairings, and the readings below are carried at length in the archived articles named with each. This entry states the decomposition, its two sectors, its three pairings and its two actions. It does not select a pairing, it does not derive a dynamics, and it does not claim that the word is conserved.

The Source Table

The nine articles cover one sector, $N\neq0$, in both of its spectral halves: each article proves its own existence, uniqueness and logarithm, and the spectral half is stated inside the article rather than split off into a companion.

the block: $N\neq0$ what it owns
The Complete Polar Decomposition of an N-Regular Biquaternion the word, the seven data, existence and uniqueness, the logarithm, the group law, the matrix picture, the two walls; on the wall the normal form, the rigidity, the closed form and the truncated logarithm
Worked Examples of the Complete Polar Decomposition of an N-Regular Biquaternion the numerical cases of the word, eight off the wall and eight on it
The Quaternion Pairing of the Complete Polar Decomposition of an N-Regular Biquaternion the scale-with-rotor bracketing, its construction, its domain, and the orthogonal modulus at the wall
The Hamilton Root Pairing of the Complete Polar Decomposition of an N-Regular Biquaternion the scale-with-phase bracketing, the root of $-1$, the eigenbasis and the two spinor spaces; and the failure of the root at the wall, where the axis is defined and the angle is pinned
The Cartan Hermitian Pairing of the Complete Polar Decomposition of an N-Regular Biquaternion the scale-with-boost bracketing, the two involutive kinds, the total domain, and the unipotent product at the wall
Relations between the Complete Polar Decomposition and Eigenbasis of an N-Regular Biquaternion the comparison of the polar frame with the eigenbasis, the normal case, the non-normal gap; and the Jordan chain where the eigenbasis fails
Worked Examples of the Complete Polar Decomposition in the Eigenbasis of an N-Regular Biquaternion the numerical cases of the comparison and of the failure of the eigenbasis
The One-Sided Operator Exponential on the N-Regular Sector the representation identity, the spectrum on one slot, the half-angle, and the defective case
The Two-Sided Exponential on the N-Regular Sector the quadratic operator, the four products, the doubling rule, and the null rotation without a common eigenbasis

The shape of the table is the shape of the section. Nine articles for the sector with two eigenvalues and nine for the sector with one, and the two sets are parallel line by line: the word, its relation to the eigenbasis, its numbers, its numbers with the eigenbasis, the three pairings, and the two actions. The parallelism is the section's claim: the polar decomposition does not care whether the element is semisimple, and every other decomposition in the algebra does.

The two articles that carry the most weight are the first of each block, which prove the word and its uniqueness. Everything else in the section is either a reading of the word (the three pairings), a comparison of the word with another structure (the eigenbasis relations), or an analytic consequence of it (the two exponentials). The worked examples exist because the gap between the polar frame and the eigenframe is numerical before it is conceptual.

The Results in Brief

The word. Every element with $N\neq0$ is $\varrho e^{i\theta_{\mathrm{pha}}}\tilde U_{\mathrm b}\tilde u_{\mathrm r}$ with $\varrho>0$, $\theta_{\mathrm{pha}}\in(-\pi/2,\pi/2]$, $\theta_{\mathrm{rpd}}\ge0$, $\theta_{\mathrm{axr}}\in[0,2\pi]$ and two real unit axes. The word is unique up to the directions where an axis is undefined.

The seven data. $\varrho=\sqrt{|N|}$, $\theta_{\mathrm{pha}}=\tfrac12\operatorname{Arg}N$, $\cosh\theta_{\mathrm{rpd}}=\lVert\tilde Q\rVert^{2}/|N|$, the boost axis from the imaginary parts of the cross products, and the rotor recovered from $\tilde Q$, the boost and the phase. Verified: on three random elements of $\Sigma_{11}$ the reconstructed element differs from the original by at most $1.5\times10^{-16}$, and the two unit factors have norm one to $1\times10^{-16}$.

The parameter count. $1+1+(1+2)+(1+2)=8$, the real dimension of the group of invertibles. The word is a chart on that group, not on the algebra.

The eigenvalues. $\lambda_\pm=Q_t\pm iB$, distinct and nonzero exactly on $\Sigma_{11}$, coalescent on $\Sigma_{10}$. Their product is the norm and their sum is twice the trace.

The normal case. The two polar factors commute with $\tilde Q$ exactly when $\tilde Q$ is normal; then the eigenbasis is a polar basis, the two axes are parallel, and the word collapses to a single exponential. Verified: for a normal element both commutators vanish at $4.5\times10^{-16}$; for the non-normal example they are $1.7$ and $1.0$ in Euclidean norm.

The non-normal gap. The spectrum determines $\{\varrho,\theta_{\mathrm{pha}},U_t\}$ and stops. The rapidity, the angle and the axes are three further unknowns for one complex equation, and the worked example exhibits two different words with the same spectrum.

The rigid sector. On $\Sigma_{10}$ the normal form is $\tilde Q=Q_t(e_t+\vec\Upsilon)$ with $\vec\Upsilon^{2}=0$, the unit factor is $\varepsilon(e_t+\vec\Upsilon)$, the angles satisfy $\cosh(\theta_{\mathrm{rpd}}/2)\cos(\theta_{\mathrm{axr}}/2)=\varepsilon$, the axes are orthogonal on the Jordan part, and the closed form reduces the four parameters to one complex number, one angle and a frame.

The three pairings. Quaternion: $\tilde Q=\tilde q\exp(i\tilde\Theta)$ with $\tilde q=\varrho\tilde u_{\mathrm r}$ a real quaternion. Hamilton: $\tilde Q=\varrho\,e^{i\theta_{\mathrm{pha}}}\exp(\vec U\Theta)$ with $\varrho\,e^{i\theta_{\mathrm{pha}}}$ central and $\vec U$ a root of $-e_t$. Cartan: $\tilde Q=\tilde Q_H\tilde Q_0$ with $\tilde Q_H^{2}=\tilde Q\tilde Q^{*}$ Hermitian positive and $\tilde Q_0$ $*$-unitary.

The domains. The Cartan pairing and the word exist on the whole complement of the null cone. Hamilton needs $B^{2}\neq0$ as well. The quaternion factorisation needs $N\neq0$ and its construction needs the real quaternion part of $\tilde Q$ invertible.

One slot and two. $\exp(tL_{\tilde Q})=L_{\exp(t\tilde Q)}$ with eigenvalues $e^{t\lambda_\pm}$ twice each, determinant $e^{4tQ_t}$ and trace $4e^{tQ_t}\cos(tB)$. The two-sided operator has eigenvalues $\lambda_i\bar\lambda_j$, determinant $e^{4t|Q_t|^{2}}$, and no element identity.

The doubling. The scale doubles its exponent, the phase produces nothing, the boost and the rotor double theirs. A boost of rapidity $\theta$ has one-slot eigenvalues $e^{\pm\theta/2}$ and two-slot eigenvalues $e^{\pm\theta}$.

The logarithm. The exponential is onto the group of invertibles and its image excludes exactly the null cone; a logarithm exists on the invertibles and only there, and the fibre is the kernel, a lattice on the diagonal part and continuous off it.

The Word and Its Domain

The Two Invariants and the Four Sectors

Two complex numbers decide everything.

$$ N=Q_t^{2}+Q_x^{2}+Q_y^{2}+Q_z^{2},\qquad B^{2}=Q_x^{2}+Q_y^{2}+Q_z^{2}. $$

The first is the reduced product $\tilde Q\tilde Q^{\natural}$ and the second is the quadratic form of the vector part. Each carries one complex condition, so the pair cuts the algebra into four sectors, and the word behaves differently in each. The two conditions are independent: $e_t+i\vec e_x$ has $N=0$ and $B^{2}=-1$, and $e_t+\vec e_x+i\vec e_y$ has $N=2$ and $B^{2}=0$.

The sector of this entry is the union of two of the four. On $\Sigma_{11}$ and $\Sigma_{10}$ the norm does not vanish, so the element is invertible and a modulus exists; the two sectors differ in whether the vector part is null. The other two sectors have $N=0$: the element is a zero divisor, there is no modulus, and no word exists. The entry's boundary is therefore the null cone, and the spectral cone $B^{2}=0$ is an internal wall.

Five descriptions of the regular sector. $N\neq0$ with $B^{2}\neq0$ is equivalent to each of: the two eigenvalues $Q_t\pm iB$ distinct and nonzero; the minimal polynomial $X^{2}-2Q_tX+N$ squarefree with nonzero constant term; the element an invertible element with two spectral points; and, with $\tilde U=\tilde Q/(\varrho\,e^{i\theta_{\mathrm{pha}}})$, the scalar part of $\tilde U$ not equal to $\pm1$. The last is the one that separates the two sectors most clearly: they are the two sides of the pair of hyperplanes $\operatorname{Sc}\tilde U=\pm1$ in the group of invertibles.

The parameter count is the chart's dimension. The data are $\ln \varrho$, $\theta_{\mathrm{pha}}$, the rapidity with a point of $S^{2}$, and the angle with a point of $S^{2}$, which is $1+1+3+3=8$, the real dimension of the algebra. The word is therefore a coordinate system on the group of invertibles and not a constraint on the element.

The Word

The two unit factors are exponentials of pure elements with opposite signs of the square. The axis of the boost satisfies $\vec u_{\mathrm b}^{2}=-e_t$ and its exponent carries an $i$, so the square of the exponent is positive and the exponential is a hyperbolic cosine with an imaginary sine:

$$ \tilde U_{\mathrm b}=\cosh\frac{\theta_{\mathrm{rpd}}}{2}\,e_t+i\sinh\frac{\theta_{\mathrm{rpd}}}{2}\,\vec u_{\mathrm b}, \qquad \tilde u_{\mathrm r}=\cos\frac{\theta_{\mathrm{axr}}}{2}\,e_t+\sin\frac{\theta_{\mathrm{axr}}}{2}\,\vec u_{\mathrm r}. $$

The rotor has no $i$ in front, so its exponent squares negative and the exponential is a circular cosine with a real sine. One factor is Hermitian and has scalar part at least one; the other is a real quaternion and has scalar part in $[-1,1]$. This is the whole content of the word: a dilation, a central turn, a hyperbolic turn and a circular turn, in that order.

The order is forced and the uniqueness is up to the axes. The four factors are read off in the order written, and the only freedom is the direction of an axis where the corresponding angle vanishes. At $\theta_{\mathrm{rpd}}=0$ the boost is $e_t$ and its axis is undefined; at $\theta_{\mathrm{axr}}\in\{0,2\pi\}$ the rotor is $\pm e_t$ and its axis is undefined. Nothing else is free.

The word is a chart, so it says nothing about time. A chart of a group describes positions, not trajectories. The corpus's flow language attaches to the exponential of an element, which is a one-parameter subgroup and therefore a trajectory; the word attaches to an element, which is a point. The two are related and not identical, and the entry keeps them apart: the logarithm recovers the word from the element, and the exponential generates the flow from a generator.

The Two Walls

wall equation stratum what happens to the word what happens to the eigenbasis
null cone $N=0$ $\Sigma_{01}\cup\Sigma_{00}$ $\varrho\to0$, $\theta_{\mathrm{rpd}}\to\infty$ none exists
spectral cone $B^{2}=0$ $\Sigma_{10}\cup\Sigma_{00}$ finite limit, the word of the limit idempotents diverge

The two walls fail in opposite directions. Approaching the null cone along the standard family

$$ \tilde Q(\theta_{\mathrm{rpd}})=R e^{-\theta_{\mathrm{rpd}}/2}\,\omega\exp\Bigl(i\frac{\theta_{\mathrm{rpd}}}{2}\vec u_{\mathrm b}\Bigr)\tilde u_{\mathrm r}, $$

the modulus falls as $e^{-\theta_{\mathrm{rpd}}/2}$ and the boost grows as $e^{+\theta_{\mathrm{rpd}}/2}$, and the product converges to $\frac{R\omega}{2}(e_t+i\vec u_{\mathrm b})\tilde u_{\mathrm r}$, a null element of rank one. The word diverges in both of its parameters and the element has a limit: the chart breaks and the group does not.

Which stratum the limit lands on is decided by the rotor and not by the rate. Write $\tilde u_{\mathrm r}=a\,e_t+b\,\vec u_{\mathrm b}+\tilde u_\perp$ with $a,b\in\mathbb{R}$ and $\tilde u_\perp$ orthogonal to $\vec u_{\mathrm b}$, and put $L$ for the limit. Then

$$ 2\,\mathrm{Sc}L=R\omega(a-ib), \qquad L^{2}=R\omega(a-ib)\,L , $$

so $L$ is $R\omega(a-ib)$ times a minimal idempotent when $a-ib\neq0$, and a nilpotent when $a=b=0$. The second case is the rotor that anticommutes with the boost axis, and twice the scalar part is the whole of the test. Verified: both identities hold exactly on four rotor orientations, and at $\tilde u_{\mathrm r}=\vec e_x$ with $\vec u_{\mathrm b}=\vec e_z$ the limit is $\tfrac{1}{2}(\vec e_x+i\vec e_y)$ with $L^{2}=0$ identically.

The two strata are therefore two boundary values of one family. The same divergence, at the same rate, along the same path, reaches an idempotent or a nilpotent according to the relative orientation of the rotor and the boost axis. A statement about a boundary point of the word is a statement about a family and about that orientation, and not about the point alone.

The spectral wall is the reverse. The modulus is continuous in $N$ away from zero, the Cartan factors are continuous functions of the unit factor, so the word converges to the word of the limit point. The root $\vec Z=-i\vec Q/B$ diverges, because $\vec Q\to\vec Q_0\neq0$ while $B\to0$, so the idempotents $(e_t\pm\vec Z)/2$ are unbounded and no spectral idempotent has a finite limit. The polar decomposition crosses the wall; the spectral decomposition does not.

The exception is the scalar line. At $\tilde Q_0=Q_0e_t$ the root is $0/0$ and its limit depends on the path: the family $Q_k=1+\epsilon_k\vec e_x$ has $\vec Z_k=-i\vec e_x$ for every $k$, so the idempotents converge to nonzero idempotents of the ambient algebra although the limit is scalar. The divergence of the eigenbasis is therefore a statement about the Jordan part of the wall and not about the whole of it.

The Regular Case: The Two-Mode Flow

The Two Eigenvalues and the Two Modes

The element has two spectral points and the flow has two rates. $\lambda_\pm=Q_t\pm iB$, and the unit factor's eigenvalues are $\mu_\pm=\lambda_\pm/(\varrho\,e^{i\theta_{\mathrm{pha}}})$ with $\mu_+\mu_-=1$. The exponents of the two modes are opposite, because the product of the eigenvalues is the norm and the norm of the unit factor is one: whatever one mode gains in modulus the other loses.

A flow with two opposite rates is a boost, a rotation, or a mixture. When $Q_t/(\varrho\,e^{i\theta_{\mathrm{pha}}})$ is real the two eigenvalues are complex conjugates of unit modulus and the flow is a rotation; when $Q_t/(\varrho\,e^{i\theta_{\mathrm{pha}}})$ is real and greater than one the eigenvalues are real reciprocals and the flow is a boost; in general the two eigenvalues have both a modulus ratio and an argument difference, and the flow is a rotation composed with a dilation. The three cases are the three types of the regular sector, and the type is fixed by a single complex number.

The pair of exponents is the two-mode description. One mode advances and one retreats, at rates that are equal in magnitude exactly when the element is normal. The difference of the rates is the rapidity and the sum of the argument changes is the rotation; the equality of the magnitudes is the Hermitian condition on the unit factor. This is why a single exponential can carry only a boost or only a rotation and not both: a single exponential has one axis, and two independent rates need two.

The Normal Case: The Frame Is the Eigenframe

The commutator with the Hermitian conjugate is one cross product.

$$ [\tilde Q,\tilde Q^{*}]=-2\,\vec Q\times\overline{\vec Q}. $$

The scalar parts are central and cancel; the vector part gives the cross product of $\vec Q$ with its complex conjugate. Normality is therefore the vanishing of that cross product, which is the condition that $\vec Q$ and $\overline{\vec Q}$ be parallel over $\mathbb{C}$.

Normality is exactly the collapse of the word. The two polar factors commute with $\tilde Q$ if and only if $\tilde Q$ is normal, and then they are diagonal in the eigenbasis, their axes are the eigenlines, and the two exponents combine into one. The word degenerates from four factors to two: a modulus and a single exponential of an axis times a complex angle.

The check is sharp in both directions. For a normal element of the form $Q_t+\omega\vec u$ the two commutators vanish at $4.5\times10^{-16}$ and $4.3\times10^{-16}$ in Euclidean norm. For the non-normal element of the next subsection they are $1.7$ and $1.0$. There is no middle ground: the polar frame is either the eigenframe or it is not, and the numerical gap between the two cases is two orders of magnitude above any rounding.

The Non-Normal Case: What the Spectrum Does Not Determine

The eigenvalues of the unit factor are functions of its scalar part alone. $\mu_\pm=U_t\pm i\sqrt{1-U_t^{2}}$, so the spectrum of $\tilde U$ depends on $U_t$ and on nothing else. But the word's parameters also depend on the relative angle of the two axes, through $\operatorname{Sc}\tilde U=c_{\mathrm b}c_{\mathrm r}-i s_{\mathrm b}s_{\mathrm r}c$ with $c=\vec u_{\mathrm b}\cdot\vec u_{\mathrm r}$. The spectrum is one complex equation for three real unknowns, and it does not solve it.

The worked example is exact. Let

$$ \tilde Q=\Bigl(1-\frac{i}{2}\Bigr)e_t+\frac{1+i}{\sqrt2}\vec e_x+\frac{1}{\sqrt2}\vec e_y+\frac{i}{2}\vec e_z . $$

Its norm is $1$, its vector form is $\frac14$, and it is non-normal because $\vec Q\times\overline{\vec Q}=(-i/\sqrt2,\,i/\sqrt2,\,i)\neq0$. Its word has $\theta_{\mathrm{rpd}}=1.7627471740$, $\vec u_{\mathrm b}=\vec e_x$, $\theta_{\mathrm{axr}}=\pi/2$, $\vec u_{\mathrm r}=(\vec e_x+\vec e_y)/\sqrt2$ and $c=1/\sqrt2$. The eigenvalues of the unit factor have moduli $0.480533816$ and $2.081019$ and arguments $\pm0.674888846$.

The gap is the whole point of the comparison. The normal-case formula, which reads the rapidity from the ratio of the moduli, gives $\log(2.081019/0.480534)=1.465715$ and an angle of $1.350$. The true values are $1.763$ and $1.571$. Both words have the same spectrum, both are legitimate, and only one belongs to the element. The eigenbasis is not a polar basis and the spectrum does not know the difference.

The count of real data makes the gap structural. The spectrum and the eigenprojectors together determine the element, and they are eight real data. The word is another eight. They share four — $\varrho$, $\theta_{\mathrm{pha}}$, and the complex $U_t$ — and the remaining four are the same four only in the normal case, where the eigenprojectors are the polar axes. Off that set the two parametrizations are transverse, and the normal set is the intersection: $Q_t+\omega\vec u$ with two complex scalars and a unit vector, of real dimension $6$, not open in the sector.

The Degenerate Case: The Flow That Cannot Be Split

The Normal Form

On $\Sigma_{10}$ the element is a scalar times a unipotent. Since $B^{2}=0$ the vector part is null, and since $N=Q_t^{2}+B^{2}\neq0$ the scalar part is nonzero. Hence

$$ \tilde Q=Q_t\bigl(e_t+\vec\Upsilon\bigr),\qquad \vec\Upsilon=\vec Q/Q_t,\qquad \vec\Upsilon^{2}=0, $$

and the writing is unique because $Q_t$ is the scalar part. The unipotent factor has norm one and inverse $e_t-\vec\Upsilon$, so every element of the sector is invertible.

The sector splits into two strata. The scalar line $\vec\Upsilon=0$ is the set of nonzero complex scalars, closed in the sector and of measure zero. The Jordan part $\vec\Upsilon\neq0$ is open and dense, with minimal polynomial $(X-Q_t)^{2}$ and geometric multiplicity one. The two strata are the two possible Jordan forms of an element with a single eigenvalue.

The unit factor is $\pm$unipotent. With $\varrho\,e^{i\theta_{\mathrm{pha}}}=\pm Q_t$ the principal square root,

$$ \tilde U=\tilde Q/(\varrho\,e^{i\theta_{\mathrm{pha}}})=\varepsilon\bigl(e_t+\vec\Upsilon\bigr),\qquad \varepsilon=Q_t/(\varrho\,e^{i\theta_{\mathrm{pha}}})=\pm1, $$

so the unit factor is unipotent up to a sign and the sign is the only trace of the scalar part that survives in it. Its scalar part is exactly $\varepsilon$.

The Rigidity: One Complex Equation on Six Parameters

The word is constrained, not free. On $\Sigma_{11}$ the two angles are independent; here they are tied. The scalar part of the unit factor is $c_{\mathrm b}c_{\mathrm r}-i s_{\mathrm b}s_{\mathrm r}c$, and it must equal the real number $\varepsilon$. Its imaginary part vanishes and its real part is $\varepsilon$:

$$ \boxed{\;\cosh\frac{\theta_{\mathrm{rpd}}}{2}\cos\frac{\theta_{\mathrm{axr}}}{2}=\varepsilon\;}, \qquad \sinh\frac{\theta_{\mathrm{rpd}}}{2}\sin\frac{\theta_{\mathrm{axr}}}{2}\,\vec u_{\mathrm b}\cdot\vec u_{\mathrm r}=0 . $$

One complex equation on the six parameters of the unit factor, which is why the unipotent set has real dimension four.

The frame is orthogonal on the Jordan part. There $\theta_{\mathrm{rpd}}>0$, $\theta_{\mathrm{axr}}\notin\{0,\pi,2\pi\}$ and $\vec u_{\mathrm b}\perp\vec u_{\mathrm r}$, and the two angles are tied by

$$ \Bigl\lvert\cos\frac{\theta_{\mathrm{axr}}}{2}\Bigr\rvert=\frac{1}{\cosh\frac{\theta_{\mathrm{rpd}}}{2}},\qquad \varepsilon=\operatorname{sgn}\cos\frac{\theta_{\mathrm{axr}}}{2}. $$

The rapidity is a function of the spin angle and the sign is a function of the quadrant. The independent data are the angle, the boost axis and the rotor axis in the great circle orthogonal to the boost axis: $1+2+1=4$, the dimension of the punctured nilpotent cone.

The lock is checkable in one line. At $\theta_{\mathrm{axr}}=0.5,1,2,3$ the relation gives $\theta_{\mathrm{rpd}}=0.505291$, $1.044476$, $2.452382$, $6.681355$, and the product $\cosh(\theta_{\mathrm{rpd}}/2)\cos(\theta_{\mathrm{axr}}/2)$ is $1.000000000000$ in all four cases. As $\theta_{\mathrm{axr}}\to\pi$ the rapidity diverges; $\theta_{\mathrm{axr}}=\pi$ itself is not attained, because it would force $\cos(\theta_{\mathrm{axr}}/2)=0$.

The Closed Form

The whole sector has one shape.

$$ \tilde Q=Q_t\Bigl(e_t+i\tanh\frac{\theta_{\mathrm{rpd}}}{2}\,\vec u_{\mathrm b}+\tan\frac{\theta_{\mathrm{axr}}}{2}\,\vec u_{\mathrm r} +i\tanh\frac{\theta_{\mathrm{rpd}}}{2}\tan\frac{\theta_{\mathrm{axr}}}{2}\;\vec u_{\mathrm b}\times\vec u_{\mathrm r}\Bigr). $$

The complex number $Q_t$ carries the scale and the phase; the bracket is a unipotent written in an orthogonal frame, and it is automatically null because the three coefficients satisfy $\tanh^{2}-\tan^{2}+\tanh^{2}\tan^{2}=0$ under the locking relation. The cross term carries the plus sign because its coefficient is imaginary.

The scalar factor and the frame are independent. Changing $Q_t$ changes the scale, the phase and possibly the sign, and leaves $\vec\Upsilon$ untouched. The sector is therefore the product of the scalar line with the punctured nilpotent cone, and the frame and the angles are functions of the second factor alone.

The logarithm is exact and truncated. Because $\vec\Upsilon^{2}=0$,

$$ \operatorname{Log}\tilde Q=\log Q_t+\vec\Upsilon, $$

and the exponential inverts it: $\exp(Q_t(e_t+\vec\Upsilon))=e^{Q_t}(e_t+Q_t\vec\Upsilon)$, verified on nilpotent elements to $2.8\times10^{-15}$. The sum of the four word exponents is not a logarithm, because the boost and the rotor do not commute; the log defect is the cross term plus the nonlinear terms.

What Is Lost

There is no eigenbasis. The minimal polynomial is a square, the geometric multiplicity is one, and the two eigenvalues coincide. The root $\vec Z=-i\vec Q/B$ does not exist because $B=0$, and the idempotents do not exist either. What replaces them is the pair $(Q_t,\vec\Upsilon)$, which is a Jordan datum and not a spectral one.

The flow is a polynomial and not a periodic function. On the cone the exponentiation truncates at first order, so a one-parameter family generated by a nilpotent element grows linearly in the parameter and never returns. A flow that never returns has no period and no frequency, which is the same statement as the absence of a root. The degenerate sector is therefore not a small perturbation of the regular one: the two regimes differ in kind, and the difference is visible in the series.

The rigidity is the sector's content. A generic element has two angles; here the rapidity is a function of the angle. A field that fixes its spin angle has no freedom left in its rapidity, and the velocity is locked to the spin. That lock is not imposed: it is the condition $B^{2}=0$ read in the word.

The Three Pairings

Three Bracketings, Three Moduli

Four factors can be bracketed in exactly three ways, and each bracketing gives a first factor of a different kind.

pairing first factor second factor norm of the first norm of the second
Hamilton $\varrho\,e^{i\theta_{\mathrm{pha}}}=\varrho e^{i\theta_{\mathrm{pha}}}$, a complex scalar $\exp(\vec U\Theta)=\tilde U_{\mathrm b}\tilde u_{\mathrm r}$ $\varrho^{2}e^{2i\theta_{\mathrm{pha}}}$ $e_t$
quaternion $\tilde q=\varrho\tilde u_{\mathrm r}$, a real quaternion $\exp(i\tilde\Theta)=e^{i\theta_{\mathrm{pha}}}\tilde u_{\mathrm r}^{-1}\tilde U_{\mathrm b}\tilde u_{\mathrm r}$ $\varrho^{2}$ $e^{2i\theta_{\mathrm{pha}}}$
Cartan $\tilde Q_H=\varrho\tilde U_{\mathrm b}$, Hermitian positive $\tilde Q_0=e^{i\theta_{\mathrm{pha}}}\tilde u_{\mathrm r}$, $*$-unitary $\varrho^{2}$ $e^{2i\theta_{\mathrm{pha}}}$

The three rows split the same norm $\varrho^{2}e^{2i\theta_{\mathrm{pha}}}=\varrho^{2}e^{2i\theta_{\mathrm{pha}}}$ in three ways. The Hamilton row puts the whole norm in a central first factor and leaves a unit second factor. The other two put the scale in the first factor and the phase in the second, and differ in the kind of the first factor.

The first factor's kind is the pairing's contribution. A complex scalar is invariant under every conjugation; a real quaternion is a spatial frame; a Hermitian positive element is an amplitude. The three pairings are therefore three answers to the question what of an element survives a motion of the algebra, and the answers differ because the motions differ.

The three first factors are three readings of one element and not three summands of it. Their real dimensions are $4$, $2$ and $4$, which add to $10$ against the algebra's $8$, and the redundancy is exactly the scale $\varrho$, which all three carry. The three identities $\tilde Q=\tilde q\exp(i\tilde\Theta)=\varrho\,e^{i\theta_{\mathrm{pha}}}\exp(\vec U\Theta)=\tilde Q_H\tilde Q_0$ are each the whole element written three times, so no statement of the form $\tilde Q=\tilde q+\varrho\,e^{i\theta_{\mathrm{pha}}}+\tilde Q_H$ exists and no conservation law attaches to a factor.

The Quaternion Pairing

The scale joins the rotor and the phase joins the boost.

$$ \tilde Q=\tilde q\exp(i\tilde\Theta),\qquad \tilde q=\varrho\tilde u_{\mathrm r},\qquad \exp(i\tilde\Theta)=e^{i\theta_{\mathrm{pha}}}\,\tilde u_{\mathrm r}^{-1}\tilde U_{\mathrm b}\tilde u_{\mathrm r}. $$

The rotor is moved to the left, past the central phase, next to the scale; the boost is conjugated by the rotor so that the two non-central factors can stand together on the right as one exponential of a real quaternion. The modulus is an ordinary quaternion in its polar form and the exponent has scalar part $\theta_{\mathrm{pha}}$ and pure part $\tfrac12\theta_{\mathrm{rpd}}\vec u_{\mathrm b}'$ with the boost axis rotated by the rotor.

Verified. On four random elements of $\Sigma_{11}$ the product $\tilde q\exp(i\tilde\Theta)$ reproduces $\tilde Q$ to $1.8\times10^{-15}$, the modulus is a real quaternion to $3\times10^{-16}$ in its imaginary part, and $N(\tilde q)=\varrho^{2}$ to $2\times10^{-15}$.

The pairing is partial in its construction only. The factorisation inherits the word and exists wherever the word does. The construction, which reads $\tan\tilde\Theta=\tilde q_r^{-1}\tilde q_i$ from the real and imaginary quaternion parts, needs $\tilde q_r$ invertible, and $\tilde q_r$ is the real part of the element, not of the modulus. The witness is $i\vec e_x+i\vec e_y$: its norm is $-2$, so the factorisation exists, and its real part is $0$, so the construction does not.

The Hamilton Pairing

The scale joins the phase and the boost joins the rotor.

$$ \tilde Q=\varrho\,e^{i\theta_{\mathrm{pha}}}\exp(\vec U\Theta),\qquad \varrho\,e^{i\theta_{\mathrm{pha}}}=\sqrt N,\qquad \vec U=\vec Q/B,\qquad \vec U^{2}=-e_t,\qquad \cos\Theta=Q_t/(\varrho\,e^{i\theta_{\mathrm{pha}}}),\quad \sin\Theta=B/(\varrho\,e^{i\theta_{\mathrm{pha}}}) . $$

This is the representation of the classical biquaternion literature, and its axis is a root of $-e_t$; the pure root of $+1$ that the kernel of the Fourier section uses is $\vec Z=-i\vec U$, and the two are the same axis read at two squares. The angle is complex, and its real part is half the rotation while its imaginary part is half the rapidity only when the two axes are aligned.

Verified. On four random elements the product reproduces $\tilde Q$ to $1.3\times10^{-15}$ and the second factor has norm one to $2.4\times10^{-15}$.

The pairing is partial in both senses. The axis $\vec U=\vec Q/B$ requires $B^{2}\neq0$, so both the factorisation and the construction fail on the spectral cone. What it contributes is the only central first factor and the only unit second factor in the series: the modulus is the one datum that no conjugation moves, and the second factor is a Lorentz transformation in its own right.

The Cartan Pairing

The scale joins the boost and the phase joins the rotor.

$$ \tilde Q=\tilde Q_H\tilde Q_0,\qquad \tilde Q_H=\sqrt{\tilde Q\tilde Q^{*}}=\varrho\tilde U_{\mathrm b},\qquad \tilde Q_0=\tilde Q_H^{-1}\tilde Q=e^{i\theta_{\mathrm{pha}}}\tilde u_{\mathrm r}. $$

The first factor is Hermitian and positive, the second is $*$-unitary, and the two exponents lie in the two summands of the Cartan decomposition of the algebra as a real Lie algebra: $\ln\tilde Q_H=\ln \varrho\,e_t+i\tfrac12\theta_{\mathrm{rpd}}\vec u_{\mathrm b}$ in the Hermitian sector, $\ln\tilde Q_0=i\theta_{\mathrm{pha}}e_t+\tfrac12\theta_{\mathrm{axr}}\vec u_{\mathrm r}$ in the anti-Hermitian one.

Verified. On random elements $\tilde Q_H$ is Hermitian to $0$, $\tilde Q_H^{2}=\tilde Q\tilde Q^{*}$ to $8.9\times10^{-16}$, $\tilde Q_0\tilde Q_0^{*}=e_t$ to $6.6\times10^{-16}$ and the product reproduces $\tilde Q$ to $4.0\times10^{-16}$.

The pairing is total on the complement of the null cone. Both factors are functions of the element alone, both are defined wherever the modulus is, and neither requires a non-null vector part. The second factor is a partial isometry on the cone itself and a $*$-unitary element off it, which is why the pairing extends where the word does not.

The two exponents are in the two sectors and the two factors are not. $\tilde Q_H$ is Hermitian and positive, which places it in the informational sector only in the sense that its logarithm is there; $\tilde Q_0$ is $*$-unitary and its logarithm is anti-Hermitian, which places it in the material sector. The factors themselves are typed by their involutions and not by their logarithms, and conflating the two typings is the error that the sector articles warn against.

The Domains and the Witnesses

element $N$ $B^{2}$ real part of $\tilde Q$ invertible Hamilton quaternion construction Cartan
$(1+i)e_t+\vec e_x+2\vec e_y$ $5+2i$ $5$ yes yes yes yes
$e_t+\vec e_x+i\vec e_y$ $1$ $0$ yes no yes yes
$i\vec e_x+i\vec e_y$ $-2$ $-2$ no yes no yes
$i$ $-1$ $0$ no no no yes
$e_t+i\vec e_x$ $0$ $-1$ yes no word no word no word

Five elements separate the cases and no theorem is needed to see it. The first line is generic and all three pairings exist. The second separates Hamilton: the vector part is null so the root is undefined. The third separates the quaternion construction: the norm is nonzero so the factorisation exists and the real part is zero so the construction does not. The fourth is missed by both partial pairings and caught by Cartan. The fifth lies on the null cone and no pairing exists.

The domains are not nested. The Hamilton domain and the quaternion-construction domain are disjoint on the third and fourth lines in opposite directions, and the Cartan domain contains both. The complement of the null cone is exactly covered by Cartan plus the two partial constructions, and the cover is what makes the Cartan pairing the reference rather than a favourite.

The Three Splits Are Not the Three Pairings

A pairing brackets a product and a split decomposes the algebra. The algebra has three natural two-way splits: real against imaginary coefficients, Hermitian against anti-Hermitian, scalar against vector. Each belongs to a different reading, and none is the same structure as a bracketing, even though both lists have three entries.

split pieces used by reading
real vs imaginary coefficients $\mathbb{B}_{H}\oplus\mathbb{B}_{iH}$ the quaternion construction the real quaternion frame and the imaginary-coefficient part
Hermitian vs anti-Hermitian $\mathbb{B}_{\Psi}\oplus\mathbb{B}_{\Phi}$ the physical sectors, the Cartan exponents the informational and material generators
scalar vs vector $\mathbb{B}_{C}\oplus\mathbb{B}_{V}$ the four-vector reading, the eigenvalue decomposition time and space, the spectral axis

The identification of the first split with the second is the standard error. The material sector has an imaginary scalar and a real vector, so it mixes the real and imaginary coefficient splits; a table that identifies the real part with the informational sector and the imaginary part with the material sector is wrong at four of the eight coordinates. The quaternion pairing uses the first split and the physical sectors are the second, and the two are different classifications of the same nine objects.

The Cartan exponents are the one place where a split and a pairing meet. The logarithms lie one in each sector, and the factors do not: $\tilde Q_H$ is Hermitian and positive, $\tilde Q_0$ is $*$-unitary, and neither is a general element of a sector. A pairing groups factors and a split types generators, and the Cartan pairing is the case where the grouping makes the typing possible.

One Slot and Two Slots

The One-Sided Exponential

A one-sided operator is a representation, so its exponential is the exponential of the element.

$$ \exp\bigl(tL_{\tilde Q}\bigr)=L_{\exp(t\tilde Q)},\qquad \exp\bigl(tR_{\tilde Q}\bigr)=R_{\exp(t\tilde Q)} . $$

The proof is one line: $L_{\tilde Q}^{n}=L_{\tilde Q^{n}}$, so the series collects term by term. This is the reason the one-sided reading is the natural home of the spinor exponent, and it is the reason the one-sided operator has a group and the two-sided one does not.

The spectrum is the eigenvalue repeated twice. On $\Sigma_{11}$ the operator is diagonalisable with eigenvalues $\lambda_+,\lambda_+,\lambda_-,\lambda_-$, so

$$ \det\exp\bigl(tL_{\tilde Q}\bigr)=e^{4tQ_t},\qquad \operatorname{tr}\exp\bigl(tL_{\tilde Q}\bigr)=4e^{tQ_t}\cos(tB). $$

Verified at $t=0.7$ on three random elements: the eigenvalues come in the pairs $\lambda_\pm$ twice each, and the determinant and the trace agree with the formulas to $1.2\times10^{-14}$.

The rate is the eigenvalue and not its square. For a central element $c\,e_t$ the group is $e^{tc}\,\mathrm{id}$, while the two-sided companion gives $e^{t|c|^{2}}\mathrm{id}$. The one-sided rate is $c$ and the two-sided rate is $|c|^{2}$, and the two are equal exactly when $|c|^{2}=2\operatorname{Re}c$ — that is, for a single point and not on a set.

The Two-Sided Exponential

The two-sided operator is quadratic, so there is no element identity.

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

satisfies $\mathrm{Sd}_{\tilde Q}^{n}=\mathrm{Sd}_{\tilde Q^{n}}$ only in the sense that the modulus is powered twice. What replaces the identity is the spectral form: on $\Sigma_{11}$ the operator has eigenvalues $\lambda_i\bar\lambda_j$ and

$$ \det\exp\bigl(t\,\mathrm{Sd}_{\tilde Q}\bigr)=\exp\bigl(4t|Q_t|^{2}\bigr). $$

Verified: the determinant agrees with the formula to a relative $1\times10^{-11}$ on random elements.

The absence of an element identity is itself the doubling. The two-sided operator exponentiates its image, not its parameter: for a central element the group is $e^{t|c|^{2}}\mathrm{id}$ while $\mathrm{Sd}_{\exp(tc)}=e^{2t\operatorname{Re}c}\mathrm{id}$, and with $c=0.8+0.3i$ the two numbers are $2.075$ and $4.953$. A quadratic operator has a linear group, and the factor two between the generator and its image is the whole of the difference.

The Doubling Rule

Write each factor as $F=\exp(tA)$ and the two-sided action is a congruence by the two exponents.

$$ \mathrm{Sd}_F(\tilde X)=\exp(tA)\,\tilde X\,\exp(tA^{*}). $$

Three cases exhaust the word. If $A^{*}=A$ the two slots carry the same exponent and the outputs add: the scale, with $A=e_t$ central, becomes multiplication by $e^{2tA}$, and the boost, with $A=i\vec u_{\mathrm b}$ non-central, becomes a congruence whose parameter is $2t$. If $A^{*}=-A$ the two slots carry opposite exponents and the congruence is a conjugation: the rotor's parameter doubles into an angle and the phase's cancels.

factor exponent kind one-slot output two-slot output
scale $\varrho$ $\ln \varrho$ real scalar $\ln \varrho$ $2\ln \varrho$, a dilation $\varrho^{2}$
phase $e^{i\theta_{\mathrm{pha}}}$ $i\theta_{\mathrm{pha}}$ imaginary scalar $\theta_{\mathrm{pha}}$ $0$
boost $\tilde U_{\mathrm b}$ $i\tfrac12\theta_{\mathrm{rpd}}\vec u_{\mathrm b}$ imaginary vector $\tfrac12\theta_{\mathrm{rpd}}$ $\theta_{\mathrm{rpd}}$, a rapidity
rotor $\tilde u_{\mathrm r}$ $\tfrac12\theta_{\mathrm{axr}}\vec u_{\mathrm r}$ real vector $\tfrac12\theta_{\mathrm{axr}}$ $\theta_{\mathrm{axr}}$, a rotation

The numbers are exact and elementary. For a boost of rapidity $0.9$ the one-slot eigenvalues are $e^{\pm0.45}=1.568312$ and $0.637628$; the two-slot eigenvalues are $e^{\pm0.9}=2.459603$ and $0.406570$ together with $1$ twice. The scale and the phase are central, so their rows are read as scalar multipliers; for a purely central element the spectrum collapses and the table is a table of scalars.

The doubling is a statement about representation and not about algebra. The two slots are the two tensor factors of the matrix model, $\mathbb{B}\cong S\otimes S^{*}$ with $S=\mathbb{C}^{2}$, and the two-sided operator is $\rho_{\mathrm{rep}}(\tilde Q)\otimes\bar\rho_{\mathrm{rep}}(\tilde Q)$. The tensor product of the defining representation and its conjugate is the vector, $2\otimes\bar2=4$, so the two-sided operator is the vector and the one-sided operator on a single slot is the spinor. The weights add, and adding two half-angles gives the full angle.

Spinor Linear and Vector Bilinear

The two actions differ in degree. A four-vector transforms by the congruence $X\mapsto\tilde Q_0X\tilde Q_0^{*}$, which is bilinear in the spinor when $X=\xi\xi^{*}$; a spinor transforms by left multiplication $\xi\mapsto\tilde Q_0\xi$, which is linear. The first law is quadratic in the spinor and the second is linear, and the two are not two conventions but two different representations.

The sign of a full turn is visible only in the linear one. A rotation through $2\pi$ gives $-\tilde Q_0$ on the spinor and $+\mathrm{id}$ on the four-vector, because the congruence by $-\tilde Q_0$ is the congruence by $\tilde Q_0$. The spinor therefore has a double cover and the vector does not, and the algebraic origin of the double cover is the half-angle in the exponent.

The two Peirce copies are not the two spinor chiralities. The two-column decomposition of the matrix model gives two copies of the same Weyl representation, and a spinor chirality is a value of the chirality operator and not a column of that display. The two slots are the two sides of an action, and a side is not a value. Confusing the two is the commonest error of the reading, and the section separates them by naming the slots: the Peirce columns are the two ideals, and the slots are the two factors of the tensor product.

The Logarithm and the Fibre

The Image Is the Group of Invertibles

The exponential is entire, never zero, and onto the invertibles.

$$ \exp:\mathbb{B}\longrightarrow\mathbb{B}^{\times},\qquad \langle\exp\tilde Q\rangle_{\natural}=\exp(2Q_t)\neq0, $$

the norm identity being the multiplicativity of $N$ applied to $\exp(Q_t e_t)\exp(\vec Q)$. The image excludes exactly the null cone, and the exclusion is sharp: every invertible element has a logarithm, because $\mathbb{B}\cong M_2(\mathbb{C})$ and every invertible complex matrix has a logarithm.

The two regimes of the closed form are the two sectors.

regime condition $\exp\tilde Q$
oscillatory $B\neq0$ $e^{Q_t}(\cos B\,e_t+\sin B\,\vec U)$
nilpotent or scalar $B=0$ $e^{Q_t}(e_t+\vec Q)$

The second row is the truncation, and it is the reason the degenerate sector has an exact logarithm: the exponential is affine there and its inverse is elementary.

The Fibre

A logarithm is a generator, and a generator is not unique. The kernel of the exponential is the set of elements whose exponential is the unit, and a logarithm is fixed only up to the kernel. The mathematics computes the kernel, the principal branch and the fibre, and the three statements belong together: the image says which processes have a generator, the kernel says how many, and the cone says where the answer stops.

The kernel is not the lattice, and the difference is not pedantic. On the diagonal part it is the parity-matched lattice $\pi i m\,e_t+\pi n\,\vec U$ with $m\equiv n\pmod 2$, and off it the kernel carries continuous families. It is therefore neither discrete nor a subgroup: $2\pi\vec e_x$ and $2\pi\vec e_y$ both exponentiate to $e_t$, and $2\pi(\vec e_x+\vec e_y)$ does not. It also meets the cone, at $\pi i e_t+\pi\vec e_z$, which has norm zero and exponentiates to $e_t$. So the fibre of a logarithm is a lattice coset only for a non-scalar unit, and the kernel is a union of conjugacy classes rather than a group. Verified: the three witnesses exponentiate as stated, and the fourth has $N=0$.

The non-uniqueness is a physical statement and not a defect. Two generators that differ by a kernel element generate the same flow, so a flow does not determine its generator. A description that reports a generator has therefore reported more than the flow contains, and the extra is exactly the kernel.

The Interpretations

III.3.a — The Word Is a Flow Split into Rate and Frame

The object. $\tilde Q=\varrho e^{i\theta_{\mathrm{pha}}}\tilde U_{\mathrm b}\tilde u_{\mathrm r}$, and the three bracketings of its four factors.

The warrant. The four factors are unique off the null cone, the parameter count is the real dimension of the group, and the three bracketings exhaust the ways of pairing four ordered factors.

The reading. A flow has a rate and a frame: the rate says how fast the parameter advances and the frame says along which axis and in which orientation. The word separates them, and the three pairings are three ways of drawing the line between them. The Hamilton pairing puts the rate entirely in the first factor and leaves the frame as the second; the Cartan pairing puts the scale in the first factor and the frame data in both; the quaternion pairing puts the spatial frame in the first factor and the rate in the second. No pairing is canonical because the algebra has no canonical line between rate and frame, which is why three of them exist. The archive's The Polar Decomposition and Its Three Pairings carries this reading at length, and The Exponential, the Logarithm and the Generator adds the third term: the generator is the element whose exponential is the flow, and the word is what the generator produces.

The limit. The word says nothing about dynamics. It is a chart of a group, so it describes a flow at one parameter value, and a statement about how the factors evolve requires an equation that the word does not supply.

III.3.b — The Two Modes Are the Two Directions of the Same Flow

The object. $\lambda_\pm=Q_t\pm iB$ and the exponents $\mu_\pm=\lambda_\pm/(\varrho\,e^{i\theta_{\mathrm{pha}}})$ with $\mu_+\mu_-=1$.

The warrant. The two eigenvalues are distinct on $\Sigma_{11}$, their product is the norm, and the unit factor's eigenvalues are their ratios to the modulus, so the two modes are reciprocal.

The reading. A flow with two reciprocal modes is a flow that gains in one direction what it loses in the other. That is the algebraic shape of a boost, of a rotation, and of a mixture of the two: a boost is two real reciprocal modes, a rotation is two conjugate modes of unit modulus, and a general element is two modes with both a modulus ratio and an argument difference. The corpus's polar-variable articles read a field this way — an amplitude and a phase per mode — and the two-mode structure is what makes the reading possible: the coupled equations in The Coupled Dirac Equation in Biquaternionic Polar Variables are two equations because there are two modes, and they are coupled because the two modes are not independent.

The limit. The two modes are the element's own, not the field's. A field that is a function of position has a two-mode structure at every point, and a global statement about modes requires the modes to be continuous, which is not automatic across the spectral wall.

III.3.c — The Mass Is the Modulus, and the Modulus Is the Only Invariant

The object. $\varrho\,e^{i\theta_{\mathrm{pha}}}=\sqrt N=\varrho e^{i\theta_{\mathrm{pha}}}$ in the Hamilton pairing.

The warrant. $\varrho\,e^{i\theta_{\mathrm{pha}}}$ is central, so it commutes with every element and is untouched by a conjugation; the quaternion and Cartan first factors are not central and do change under conjugation; and $\lambda_+\lambda_-=\varrho^{2}e^{2i\theta_{\mathrm{pha}}}=N$.

The reading. The pairing that isolates the invariant is the one whose first factor is central, and it is the only one. For a four-momentum the modulus is the mass: the product of the two light-cone combinations, the one number that a change of frame cannot move. The energy is not that number — it is a component, it is the trace, and the trace is frame-dependent. The phase of the modulus carries the causal character when the norm is real: $0$ for a spacelike element, $\pi/2$ for a timelike one, undefined for a null one; and $\tfrac12\arg N$ in general. The Hamilton Pairing: Mass as Modulus and the Complex Angle carries the reading and its boundaries, and The N Norm: Mass, Angle and the Cone supplies the norm's own dictionary.

The limit. The identification of a modulus with a mass is an input of the corpus, not a theorem of the algebra. The algebra supplies a central invariant and its phase; the name mass, the factor $mc$ and the assignment of units are physical choices, and the entry states them as choices.

III.3.d — The Amplitude Is the Hermitian Factor, and Its Exponent Lies in One Sector

The object. $\tilde Q=\tilde Q_H\tilde Q_0$ with $\tilde Q_H=\sqrt{\tilde Q\tilde Q^{*}}$ Hermitian positive and $\tilde Q_0$ $*$-unitary.

The warrant. $\tilde Q_H$ is the unique Hermitian positive square root, both factors are functions of the element, and the pairing is defined on the whole complement of the null cone.

The reading. The Hermitian positive factor is the part that changes lengths and not phases; the $*$-unitary factor is the part that preserves the inner product. That is the matrix polar decomposition read in the algebra, and it is the only pairing that survives the spectral wall, which makes it the pairing of choice whenever the element may be defective. The corpus reads the first factor as an amplitude and the pair as the amplitude–frame split: The H Norm: Amplitude and Frame derives the amplitude as the measure that survives the largest group of motions, and The General Polar Decomposition, the Amplitude and the Operator-to-State Map reads the positive factor as the unnormalised amplitude of a state.

The limit. The amplitude is an operator and not a number. Reading it as a probability requires the trace normalisation, and reading it as a wave function requires a choice of slot, which the Cartan pairing does not make.

III.3.e — The Quaternion Modulus Is the Spatial Frame

The object. $\tilde q=\varrho\tilde u_{\mathrm r}$, a real quaternion, and $\exp(i\tilde\Theta)$ with a real quaternion exponent.

The warrant. The rotor is a real quaternion of norm one, $\varrho$ is real and positive, so the first factor lies in the real quaternion algebra, which is the algebra of spatial rotations; and the boost is conjugated by the rotor into the exponent.

The reading. The first factor is the frame: a scale and an orientation, with no phase and no boost. The second factor carries the phase and the boost, with the boost axis expressed in the rotated frame. The pairing is therefore the one that answers in which frame is this element described before answering anything else, and it is the natural pairing for a polarization or a triad. It is also the only pairing whose modulus is an ordinary quaternion, which is why it is the one that reduces, for a real quaternion, to the classical polar form.

The limit. The construction needs the real part of the element invertible, so it is unavailable exactly where the element is purely imaginary — a set of real codimension four that is large in the algebra and includes physically natural elements such as $i\vec e_x+i\vec e_y$.

III.3.f — The Degenerate Case Is the Flow That Cannot Be Split

The object. $\Sigma_{10}$ with the normal form $Q_t(e_t+\vec\Upsilon)$, the locking identity and the orthogonal frame.

The warrant. The minimal polynomial is a square, the eigenvalues coincide, the root does not exist, and the two angles of the word satisfy one equation instead of being free.

The reading. A flow that cannot be split into rate and frame is a flow with one freedom instead of two. On this sector the rapidity is a function of the spin angle, the two axes are orthogonal, and the element is a scalar times a unipotent written in an orthogonal frame. The physical content is a lock: fixing the spin angle fixes the speed, and the lock is not a choice but the condition $B^{2}=0$. At one end of the range the field is at rest; at the other the rapidity diverges and the velocity tends to the speed of light. Spin-Velocity Locking on the Degenerate Sector reads the lock for a fluid, and The N-Regular, Spectrally Degenerate Sector: The Rest Frame and the Exceptional Point reads the two strata as a rest frame and an exceptional point.

The limit. The lock holds exactly on $B^{2}=0$ and on no neighbourhood of it, because the two conditions that produce it are equalities. A statement about an approximate lock is a statement about a perturbation and requires an estimate that the sector does not supply.

III.3.g — The Half-Angle Is the Spinor and the Full Angle Is the Vector

The object. The one-sided exponentials with eigenvalues $e^{t\lambda_\pm}$ and the two-sided operator with eigenvalues $\lambda_i\bar\lambda_j$.

The warrant. The one-sided operator is a representation and its exponential is the exponential of the element; the two-sided operator is $\rho_{\mathrm{rep}}\otimes\bar\rho_{\mathrm{rep}}$; the weights of the tensor product are the sums of the weights.

The reading. The half-angle is what a spinor sees and the full angle is what a vector sees, and the factor two is the statement $2\otimes\bar2=4$ read in the exponents. A rotation of $\theta$ on a vector is a rotation of $\theta/2$ on a spinor, and the algebraic origin is the tensor product and not a property of rotations. The corpus uses the half-angle wherever a spinor is the object: The Eigenspinor: The Lorentz Rotor as a Function of Proper Time attaches a rotor to each event of a worldline and reads the half-angle as the spinor's proper-time phase, and Similitudes Between Biquaternion Rotors and Hamiltonian Flow compares the rotor's conjugation with a canonical flow.

The limit. The half-angle belongs to a slot and not to an element. A statement that halves an angle must name the slot it is acting on, and a statement about an element's own angle is a statement about the two-sided action.

III.3.h — The Factor of Two Is Where the Errors Live

The object. The doubling rule, and the four objects that the corpus calls a boost.

The warrant. The scale doubles its exponent, the phase produces nothing, the boost and the rotor double theirs; and for a four-vector element the word's rapidity is twice the kinematic rapidity.

The reading. A factor that appears in three of four rows and in both actions is a factor that will be dropped. The corpus states the three characteristic errors and their checks: reading a four-vector's word rapidity as its kinematic rapidity doubles it; using the frame rotor where the four-vector factor belongs halves it; and identifying the Lorentz factor with $\cosh\theta_{\mathrm{rpd},w}$ instead of $\cosh(\theta_{\mathrm{rpd},w}/2)$ overstates the energy. Each error is a factor two and each is invisible in a single exponential. The Polar Decomposition and Its Three Pairings, §The Factor of Two, and Norm and Pairing Conventions, §The Four Objects Called a Boost, are the corpus's bookkeeping; Exercise: Boosting a Four-Velocity and Rapidity Composition is the worked case.

The limit. The bookkeeping fixes the relation between the word's parameters and the kinematic ones; it does not fix which object a given article means. That is a matter of convention, and the convention must be read off the article rather than inferred.

The Limits

No dynamics. The entry decomposes an element and states nothing about its evolution. The word is a chart of the group of invertibles, and a chart does not supply a vector field.

No preferred pairing. The three pairings are three dictionaries with three domains. The entry states the domains and refuses to name a fundamental one; where the corpus names one for a purpose, the entry records the purpose.

The gap is not removable. Off the normal set the spectrum does not determine the word, and no additional spectral data repairs it, because the missing data are the relative orientation of two frames and not a spectral quantity.

The degenerate sector is rigid, not approximate. The locking relation is an equality. Nothing in the entry is a limit of the regular case, and the closed form on $\Sigma_{10}$ is exact.

The exponentials are two objects. The one-sided exponential is the exponential of the element; the two-sided one is not, and its determinant involves $|Q_t|^{2}$ rather than $Q_t$. Any statement that transfers a formula from one to the other must be checked, and the central element is where the difference is smallest and still visible.

The readings are the archive's. Each of the eight is labelled, bounded, and carried at length in an archived article. The entry's contribution is the arrangement and the boundary.

Summary

The word is a coordinate chart on the group of invertibles. $\varrho e^{i\theta_{\mathrm{pha}}}\tilde U_{\mathrm b}\tilde u_{\mathrm r}$ with eight real parameters, unique up to the directions where an axis vanishes, verified to $1.5\times10^{-16}$ on random elements of the regular sector.

Two sectors, two shapes. On $\Sigma_{11}$ the element has two distinct eigenvalues and a two-mode flow with reciprocal exponents. On $\Sigma_{10}$ it is a scalar times a unipotent, the eigenvalues coincide, the angles are tied by $\cosh(\theta_{\mathrm{rpd}}/2)\cos(\theta_{\mathrm{axr}}/2)=\varepsilon$, the axes are orthogonal, and the closed form reduces four parameters to one complex number, one angle and a frame.

The two walls fail in opposite directions. At the null cone the word diverges and the element has a limit; at the spectral cone the word has a limit and the eigenbasis diverges. The polar decomposition crosses the spectral wall and the spectral decomposition does not.

Normality is the collapse of the word. The polar factors commute with the element exactly when it is normal, and then the eigenbasis is a polar basis. Off that set the spectrum is one complex equation for three real unknowns, and the worked example shows a rapidity of $1.763$ where the normal-case formula would give $1.466$.

Three pairings, three moduli, three domains. Quaternion: a real quaternion modulus and a real quaternion exponent, partial in its construction. Hamilton: a central modulus and a unit second factor, partial on the spectral cone. Cartan: a Hermitian positive amplitude and a $*$-unitary frame, total off the null cone.

The three splits are not the three pairings. A pairing brackets a product, a split decomposes the algebra. The real–imaginary split is not the sector split, and identifying them is the standard error.

One slot carries half the angle. $\exp(tL_{\tilde Q})=L_{\exp(t\tilde Q)}$ with eigenvalues $e^{t\lambda_\pm}$ twice each, determinant $e^{4tQ_t}$, trace $4e^{tQ_t}\cos(tB)$. The two-sided operator has eigenvalues $\lambda_i\bar\lambda_j$, determinant $e^{4t|Q_t|^{2}}$, and no element identity.

The doubling rule governs every factor. The scale doubles, the phase cancels, the boost and the rotor double. A boost of rapidity $0.9$ has one-slot eigenvalues $1.568312$ and $0.637628$ and two-slot eigenvalues $2.459603$, $0.406570$, $1$, $1$.

The logarithm exists on the invertibles and nowhere else. The exponential is onto the group of invertibles with the null cone excluded, the fibre is the kernel, a lattice only on the diagonal part, and on the degenerate sector the logarithm is exact and truncated.

Eight readings, each with a boundary. The word is a flow split into rate and frame; the two modes are the two directions of one flow; the mass is the modulus and the only invariant; the amplitude is the Hermitian factor; the quaternion modulus is the spatial frame; the degenerate case is the flow that cannot be split; the half-angle is the spinor and the full angle is the vector; the factor of two is where the errors live. Each is the archive's, each is bounded, and none is a new physical claim.

Summary of Notation

object notation content
the algebra $\mathbb{B}$ biquaternions with basis $e_t,\vec e_x,\vec e_y,\vec e_z$
the element $\tilde Q$ $Q_te_t+Q_x\vec e_x+Q_y\vec e_y+Q_z\vec e_z$
the quaternionic norm $N$ $Q_t^{2}+Q_x^{2}+Q_y^{2}+Q_z^{2}$, complex and multiplicative
the vector form $B^{2}$ $Q_x^{2}+Q_y^{2}+Q_z^{2}$
the Euclidean norm $\lVert\tilde Q\rVert^{2}$ $\lvert Q_t\rvert^{2}+\lvert Q_x\rvert^{2}+\lvert Q_y\rvert^{2}+\lvert Q_z\rvert^{2}$
the modulus $\varrho\,e^{i\theta_{\mathrm{pha}}}$ $\sqrt N=\varrho e^{i\theta_{\mathrm{pha}}}$, $\varrho>0$, $\theta_{\mathrm{pha}}\in(-\pi/2,\pi/2]$
the unit factor $\tilde U$ $\tilde Q/(\varrho\,e^{i\theta_{\mathrm{pha}}})$
the boost $\tilde U_{\mathrm b}$ $\exp(i\tfrac12\theta_{\mathrm{rpd}}\vec u_{\mathrm b})$, Hermitian, positive
the rotor $\tilde u_{\mathrm r}$ $\exp(\tfrac12\theta_{\mathrm{axr}}\vec u_{\mathrm r})$, a real quaternion
the rapidity $\theta_{\mathrm{rpd}}$ real, non-negative, with axis $\vec u_{\mathrm b}\in S^{2}$
the rotor angle $\theta_{\mathrm{axr}}$ in $[0,2\pi]$, with axis $\vec u_{\mathrm r}\in S^{2}$
the eigenvalues $\lambda_\pm$ $Q_t\pm iB$, distinct and nonzero on $\Sigma_{11}$
the regular sector $\Sigma_{11}$ $N\neq0$ and $B^{2}\neq0$
the degenerate sector $\Sigma_{10}$ $N\neq0$ and $B^{2}=0$
the null vector $\vec\Upsilon$ $\vec Q/Q_t$ on $\Sigma_{10}$, with $\vec\Upsilon^{2}=0$
the sign $\varepsilon$ $Q_t/(\varrho\,e^{i\theta_{\mathrm{pha}}})=\pm1$ on $\Sigma_{10}$
the quaternion modulus $\tilde q$ $\varrho\tilde u_{\mathrm r}$, a real quaternion
the Hamilton axis $\vec U$ $\vec Q/B$, a root of $-e_t$
the Hamilton angle $\Theta$ complex, with $\cos\Theta=Q_t/(\varrho\,e^{i\theta_{\mathrm{pha}}})$ and $\sin\Theta=B/(\varrho\,e^{i\theta_{\mathrm{pha}}})$
the amplitude $\tilde Q_H$ $\sqrt{\tilde Q\tilde Q^{*}}$, Hermitian positive
the frame $\tilde Q_0$ $\tilde Q_H^{-1}\tilde Q$, $*$-unitary
the one-sided operator $L_{\tilde Q},R_{\tilde Q}$ left and right multiplication, a representation
the two-sided operator $\mathrm{Sd}_{\tilde Q}$ $\tilde X\mapsto\tilde Q\tilde X\tilde Q^{*}$, quadratic

Further Reading