The Harmonic Analysis: The Transform as a Change of Description
Introduction
This entry reads the harmonic-analysis subsection of the mathematical category. Eight articles, one construction. The construction is a transform, and the transform is a change of description: it rewrites a function of one variable as a function of another without adding and without deleting anything. The mathematics establishes when that rewriting is invertible, when it preserves a norm, what it does to a differential operator, and what it costs at the points where the algebra ceases to be a division algebra. Those four answers are the subject, and each of them is a physical statement before it is a mathematical one.
$$ W(t,\omega)=\exp\left(-2\pi i\,\tilde{Z}\,\omega t\right),\qquad \tilde{Z}^{2}=e_t,\qquad \tilde\Pi_\pm=\tfrac12\left(e_t\pm\tilde{Z}\right). $$
The first claim is that the transform is generated by one element of the algebra and one parameter. The element is a root of $1$; the parameter is a frequency. The kernel is the exponential of their product, and because the square of the root is the unit, the exponential closes at once into a cosine plus a sine. Nothing else enters: no axis, no ordering, no choice of coordinates. The root is the whole of the non-trivial data, and which root is chosen decides every subsequent statement, including whether the change of description preserves length.
The second claim is that the change of description is invertible unconditionally but isometric conditionally. Inversion is an identity in the algebra and holds at every root and for every sequence, including sequences whose samples are zero divisors. That is the mathematics' strongest statement about the transform and it needs no hypothesis. The preservation of a norm is a different and much narrower statement: it holds exactly at the Hermitian roots, the central pair and the purely imaginary unit vectors. At every other root the transform is still a bijection and still forgets nothing, but it rescales, and the rescaling is measured by the norm of the root.
The third claim is that the transform diagonalises exactly those operators whose coefficients do not depend on position. Differentiation becomes multiplication by $2\pi i\tilde{Z}\omega_\mu$, the d'Alembertian becomes a scalar multiplier, and a solution of an operator equation becomes a distribution supported on the zero set of the symbol. This is the licence for every wave-number transcription in the corpus, and the licence has a condition: the coefficients must be constant. Where they are not, the transform is a change of variables and not a change of basis, and the corpus says so instead of using it silently.
The fourth claim is that inversion is two contours and not one. The conjugate kernel splits into two scalar exponentials, one on each minimal ideal, so the same pole of a symbol contributes through one ideal on one side of the time axis and through the other ideal on the other side. A propagator in one variable is therefore two-sided even when every pole lies in one half-plane, and its exponential type is a pair of numbers. In more than one variable the singular set has real codimension two, no contour encircles it, and the residue must be replaced by a current or by an integral over a linking cycle. The mathematics names both and proves neither.
The fifth claim is that a placement is a convention with a price. The kernel can act on the left, on the right, or on both sides. For a scalar-valued function the three coincide and the distinction does not exist; for this algebra they are three different operators. Two of the two-sided placements fix a part of the algebra and lose the frequency of that part, carrying the surviving part at twice the frequency; every other sign fixes nothing. A two-sided transform therefore reduces to a one-sided one on no function except zero.
The sixth claim is that the modular pair of the algebra is trivial, and that the triviality is a statement about a vector and not about the algebra. The Tomita operator of the left regular image at the unit is the star, its polar decomposition has modular operator the identity, and the modular flow is constant. The reason is exact: the vector state of the unit is the trace. The indefinite version does not exist, because the left regular image is not closed under the indefinite adjoint, and its Krein envelope has no separating vector. The corpus's thermal readings therefore cannot be built on the algebra alone.
The first objection is that this is the complex Fourier transform written twice. It largely is, and the mathematics says so in the form of a theorem: every biquaternion transform is four scalar complex transforms, two on each minimal ideal, carried at opposite signs of the frequency. That is a result and not a defect, because it gives the $O(N\log N)$ algorithm and it locates the difficulty exactly. What is not in the complex theory is the root: the pair of ideals, the sign of the frequency on each, the failure of the isometry away from the Hermitian roots, and the degenerate set where the symbol stops being an invertible element. The complex transform is the case in which all of that collapses.
The second objection is that a change of description which is invertible says nothing. Invertibility says that nothing is lost; it does not say that nothing is changed. The content of this section is the second question: what the change preserves, what it rescales, and where it fails. An invertible map that multiplies lengths by a factor between $0.855$ and $1.156$ is a change of description that a physical statement cannot pass through unchanged, and the section computes that factor instead of assuming it away.
The third objection is that the physical readings are imported. They are, and every one of them is labelled and archived: the kernel as a clock, the frequency as a rate, the poles as the degenerate set, the placements as conventions, the modular flow as the thermal clock of a state. What the entry adds is the arrangement and the boundary of each reading, and the boundary is stated with the reading. Where the archive holds the reading at length, the entry links to it and does not restate it.
The letters. The algebra is $\mathbb{B}$, the root is $\tilde{Z}$ with $\tilde{Z}^{2}=e_t$, and the two idempotents of the root are $\tilde\Pi_\pm$. The kernel is $W$, its argument is the product of a frequency $\omega$ and a time $t$, and the two are conjugate. The transform of a function is written with a capital, $F$ or $\mathcal{F}[f]$, and the transform of a sequence carries a bracket, $F[\ell]$. The symbol of an operator is $\Sigma(\omega)$, its zero set is the degenerate set, and the reciprocal $\Sigma^{-1}=\bar\Sigma/N(\Sigma)$ is the adjugate over the norm. The two placements of the second root are $R'=\pm R$, and the plane of the root is $C(R)$.
Boundaries. The algebra, the conjugations and the norms are The Algebraic Vocabulary: Element, Conjugations, Six Subspaces and Two Ledgers; the derivative and the symbol are The Derivative, the Field and the Transform; the wave-particle reading of the transform is the companion entry Wave–Particle Duality and the Fourier Transform. This entry states the transform theory and its readings. It does not derive a dynamics, it does not select a root, and it does not claim that the root is fixed by anything inside the algebra.
The Source Table
| mathematical content | source article |
|---|---|
| the root of one and its pair, the discrete kernel and its closed form, the Peirce form, the classification of the roots, the discrete transform pair, invertibility, the conjugate symmetry, the factorization, the convolution theorem, the vanishing-norm issue, the two-dimensional transform, the $Z$ transform, the reduced commutative alternative | Biquaternion Discrete Harmonic Analysis |
| the continuous kernel, the transform pair, convergence, Riemann–Lebesgue, inversion, Plancherel, the factorization, the convolution theorem, the relation to the gradient and the d'Alembertian, the vanishing-norm issue, the measure transform and the failure of positivity, sampling and periodization | Biquaternion Continuous Harmonic Analysis |
| the exponential and the kernel as two readings of one object, the spectrum as two frequencies, the kernel identity, the eigenbasis as a Fourier basis, the two conjugations, normality and unitarity, the cone where no root exists, the three strata | Relations between the Biquaternion Exponential and the Fourier Kernel |
| the conjugate as adjugate, the reciprocal over the norm, the residues at simple and repeated null points, the exponential type and the tail, the two contours, the obstacle of real codimension two, inversion where the symbol is not invertible | Inversion of the Biquaternion Fourier Transform: Poles, Tails and the Degenerate Set |
| the three placements, the commutant and the complement, the one-sided placements, the two-sided placement and the two signs, the invariance groups and their intersection | The Three Kernel Placements of the Biquaternion Fourier Transform |
| the degenerate set and the rank of left multiplication, inversion in the algebra, two generalised inverses, the exponential on the degenerate set, inversion where the symbol is nowhere an invertible element | Biquaternion Fourier Inversion on the Degenerate Set |
| the Krein space and the left regular representation, the indefinite adjoint, the Krein–von Neumann envelope, the definite interpretation, the twisted fundamental symmetries | The Biquaternion Algebra and the Krein–von Neumann Property |
| the Hilbert algebra, the cyclic and separating unit, the Tomita operator, the modular conjugation and the two sides, the indefinite obstruction | The Tomita Operator and the J-Modular Pair for the Biquaternion Algebra |
The first two rows own the section. They define the kernel and prove the pair; the third identifies the kernel with the exponential of a regular element and so joins this section to the preceding one; the fourth and the sixth are the same question at the two ends of the algebra, one concerning poles and one concerning the degenerate set; the fifth is the only place where the order of multiplication is chosen; and the last two rows are the operator theory that the ordering makes possible.
The Results in Brief
The root. An element with $\tilde{Z}^{2}=e_t$ is central, $\pm e_t$, or pure with $(\vec{Z},\vec{Z})=-1$. The pure roots are a family of real dimension four, containing the purely imaginary unit vectors $i\vec u$. A root is Hermitian when $\tilde{Z}^{*}=\tilde{Z}$, and the Hermitian roots are exactly the central pair and the $i\vec u$.
The kernel. $W(t,\omega)=\exp(-2\pi i\tilde{Z}\omega t)=\cos\theta\,e_t-i\sin\theta\,\tilde{Z}$ with $\theta=2\pi\omega t$. It is invertible at every root and every phase, with inverse $W(-\omega)$, and its norm is the unit at every root. What fails at the central roots is the identification of the inverse with the conjugation, not invertibility.
The Peirce form. $W=e^{-2\pi i\omega t}\tilde\Pi_++e^{+2\pi i\omega t}\tilde\Pi_-$, a pair of conjugate unit complex numbers on the two minimal ideals. The kernel is never a zero divisor, because both coordinates are non-zero scalars.
Inversion. Unconditional: the inverse formula reproduces the function at every root, for every sequence and every integrable function, with no hypothesis on the values. A sample that is a zero divisor is recovered exactly.
Plancherel. Conditional: the transform is an isometry of $L^{2}$ exactly at the Hermitian roots. At the others it preserves $\lVert\tilde\Pi_+f\rVert^{2}+\lVert\tilde\Pi_-f\rVert^{2}$, and the excess is measured by $\lVert\tilde{Z}\rVert^{2}=1+2\lVert\vec p\rVert^{2}$.
The symbol. $\partial_\mu\mapsto2\pi i\tilde{Z}\omega_\mu$, so $\Box_\Phi\mapsto-(2\pi)^{2}(\omega_0^{2}+\omega_1^{2}+\omega_2^{2}+\omega_3^{2})$ and the hyperbolic $\partial_t^{2}-\Delta_\Phi$ has symbol $-4\pi^{2}(\omega_0^{2}-\lvert\vec\omega\rvert^{2})$, which vanishes on the null cone. The transform of a solution is supported on the zero set of the symbol.
The two contours. The conjugate kernel splits into $\tilde\Pi_+e^{2\pi i\omega t}+\tilde\Pi_-e^{-2\pi i\omega t}$, so a kernel in one variable is two-sided even when all poles lie in one half-plane. Its exponential type is $2\pi\rho$, the distance of the nearest pole from the axis, and a pole on the axis gives decay only as boundedness.
The placements. Left, right and two-sided; the two-sided at $R'=R$ fixes the plane of the root and loses its frequency, the two-sided at $R'=-R$ fixes the transverse plane and loses its frequency, and no other sign fixes anything. The surviving half is carried at twice the frequency, so the inversion carries a factor two.
The modular pair. The Tomita operator of the left regular image at the unit is the star, the modular operator is the identity and the modular flow is trivial, because the vector state of the unit is the trace. The indefinite version is unavailable: the image is not closed under the indefinite adjoint, and its envelope has no separating vector.
The Kernel Is a Root's Exponential
The Root and Its Pair
The root is the only non-trivial datum of the section. An element $\tilde{Z}$ with $\tilde{Z}^{2}=e_t$ has a vanishing vector part, and is then $\pm e_t$, or a vanishing scalar part. In the second case it is a pure root and satisfies $(\vec{Z},\vec{Z})=-1$. Writing $\vec{Z}=\vec p+i\vec p'$ over two real vectors, the pure roots are the pairs
$$ \vec p\cdot\vec p'=0,\qquad \lVert\vec p'\rVert^{2}-\lVert\vec p\rVert^{2}=1, $$
a family of real dimension four, which contains the purely imaginary unit vectors $i\vec u$ over the real unit vectors $\vec u$. The dichotomy is disjoint and exhaustive, so every statement about the kernel has two cases and no third.
The pair attached to a root is a complete family of orthogonal idempotents. With $\tilde\Pi_\pm=\tfrac12(e_t\pm\tilde{Z})$ one has $\tilde\Pi_\pm^{2}=\tilde\Pi_\pm$, $\tilde\Pi_+\tilde\Pi_-=0$ and $\tilde\Pi_++\tilde\Pi_-=e_t$, and the root is recovered as $\tilde{Z}=\tilde\Pi_+-\tilde\Pi_-$. The passage between root and pair is the halving map of the idempotent theory, and it is the reason a root is the right object for a kernel: a root is an idempotent read in another coordinate. The two members of the pair are exchanged by the sign of the root and by nothing else.
The conjugation decides what the kernel is worth. A root is Hermitian when $\tilde{Z}^{*}=\tilde{Z}$, and the Hermitian roots are the central pair and the purely imaginary unit vectors. At a Hermitian root the pair consists of two Hermitian idempotents, that is of two pure states, and it is exactly this condition under which the transform becomes an isometry. The word pure therefore enters at two levels and they coincide: a pure root and a pure state are the same hypothesis.
The Closed Form and the Unit
The exponential closes because the square of the root is the unit. With $\theta=2\pi\omega t$,
$$ W(t,\omega)=\exp\left(-2\pi i\tilde{Z}\omega t\right)=\cos\theta\,e_t-i\sin\theta\,\tilde{Z}, $$
the even part of the series being the cosine and the odd part the sine. Both factors lie in the commutative algebra $\mathbb{C}[\tilde{Z}]$, so the inverse is the closed form with the sine reversed, $W(t,\omega)^{-1}=W(t,-\omega)$, and the product is $(\cos^{2}+\sin^{2})e_t=e_t$. The inverse is therefore the kernel at the negated frequency, at every root, with no case distinction.
The kernel is invertible at every phase and every root, and the statement is stronger than invertibility. Its algebraic norm is the unit, $N(W)=1$, which is read off the closed form: writing $R=-i\tilde{Z}$, the kernel is $\cos\theta\,e_t+\sin\theta\,R$ with $R^{2}=-e_t$, and $N(a\,e_t+bR)=a^{2}+b^{2}$. The number is one for every $\theta$ and does not depend on the root, and it is the kernel's own statement that the transform never loses rank for want of a kernel: the kernel never approaches the null cone.
What the conjugation gives, and where it stops. At a pure root the quaternion conjugation reverses the vector part, so $W^{\natural}=W^{-1}$ and the natural norm of the kernel is the unit. At the central roots this fails: the kernel is the complex scalar $e^{\mp2\pi i\omega t}e_t$, which the quaternion conjugation fixes rather than inverts, while the inverse is the coefficient conjugate. The natural norm there is $W^{2}$, an invertible element and not $e_t$. The distinction affects only that pair, and every statement below that uses the conjugation of the kernel is stated for a pure root.
The Peirce form is the same statement in the coordinates that the transform uses. Substituting $\tilde{Z}=\tilde\Pi_+-\tilde\Pi_-$,
$$ W(t,\omega)=e^{-2\pi i\omega t}\,\tilde\Pi_+ + e^{+2\pi i\omega t}\,\tilde\Pi_- . $$
The kernel is a pair of conjugate unit complex numbers, one on each minimal left ideal. Three consequences follow at once. The kernel commutes with both idempotents, so it acts on each ideal by a scalar. The two scalars are conjugate, so the two ideals carry opposite phases. And neither scalar vanishes, so the kernel is never a zero divisor at any root and any phase, which is the statement of the previous paragraph in the coordinates of the ideals.
A kernel that never vanishes is a kernel that never loses information. The vanishing-norm issue of the mathematics is not a statement about the kernel: it is a statement about the values. A sequence may have samples on the null cone, and the transform of such a sequence is still exactly invertible. What a zero-divisor value obstructs is the arithmetic after the transform, that is pointwise division by the spectrum.
The Element Supplies Its Own Root
A regular element carries a kernel inside it. Let $\tilde{Q}=Q_te_t+\vec{Q}$ with $B^{2}=(\vec{Q},\vec{Q})\neq0$, let $B$ be a square root of $B^{2}$, and put $\tilde{Z}=-i\vec{Q}/B$. Then $\tilde{Z}$ is a pure root of $1$ and
$$ e^{\theta\tilde{Q}}=e^{\theta Q_t}\,W\!\left(\theta,-\frac{B}{2\pi}\right). $$
The exponential of an element is therefore the flow of that element written as a kernel, with the scalar part factoring out and the vector part supplying both the root and the rate. This is the identity that joins this section to the preceding one: the same exponential that is the flow of The Derivative, the Field and the Transform is the kernel here, read at a frequency.
Whether the element's kernel is an isometry is whether the element is normal. The root the element supplies is Hermitian exactly when $\vec{Q}/B$ is real, that is exactly when the element is normal. So the isometry of the transform, the positivity of the flow, and the normality of an element are one condition stated three ways, and the dictionary between them is the subject of Relations between the Biquaternion Exponential and the Fourier Kernel.
On the cone there is no root and no frequency. At $B=0$ the flow truncates to $e^{\theta Q_t}(e_t+\theta\vec N)$ with $\vec N^{2}=0$, and the root-free closed form extends to $B=0$; the kernel does not, because the direction $-i\vec{Q}/B$ is undefined. For $\vec Q=\epsilon\vec e_x$ and $\vec Q=\epsilon\vec e_y$ with $\epsilon\to0$ the roots tend to $-i\vec e_x$ and $-i\vec e_y$, and the two kernels differ at every $t$ where $\sin(2\pi\omega t)\neq0$. No limit exists, and the failure is not a technicality: it is the statement that a null element has no frequency.
The Transform Pair, and What Inversion Costs
The Discrete Pair
The discrete transform is the sum of the kernel against the sequence. For $f:\{0,\ldots,N-1\}\to\mathbb{B}$,
$$ F[\ell]=\sum_{n=0}^{N-1}W_N(n,\ell)f[n],\qquad W_N(n,\ell)=\exp\left(-2\pi i\tilde{Z}\frac{n\ell}{N}\right), $$
with inverse $f[n]=\frac1N\sum_{\ell}W_N(n,-\ell)F[\ell]$. The orthogonality relation is $\sum_\ell W_N(n,-\ell)W_N(m,\ell)=N\delta_{nm}e_t$, an identity in $\mathbb{C}[\tilde{Z}]$ that uses nothing else.
Inversion is unconditional, and the proof says why. Splitting $f[n]=\tilde\Pi_+f[n]+\tilde\Pi_-f[n]$, the kernel commutes with both idempotents and acts on each by a scalar, so the transform separates into a pair of ordinary complex discrete transforms, the second at the negated frequency. Each is inverted by the classical orthogonality of the $N$-th roots of unity, and the two results add to $f[n]$. Nothing in the argument uses the invertibility of a sample, so a sample on the null cone is recovered as exactly as any other. The check is immediate: a sequence of sixteen random biquaternions is recovered at $1.0\times10^{-14}$, and a sequence with a zero-divisor sample at $3.0\times10^{-15}$, with $N(f[3])=0$ measured exactly.
The symmetry of the spectrum is decided by the root. For a real scalar-valued sequence, $F[N-\ell]=F[\ell]^{\natural}$ at a pure root and $F[N-\ell]=\overline{F[\ell]}$ at the central root. For a quaternion-valued sequence the first identity fails, because the natural conjugation neither fixes the values nor preserves the order. The ordinary Hermitian symmetry of a real signal is a statement about a commutative algebra, and it survives only where the root makes the kernel a scalar or where the values are real.
What Inversion Costs
Inversion costs nothing and the transform costs a factorization. The transform is a bijection at every root; that is the whole of the invertibility statement, and it is unconditional. The cost appears in what the transform does to the arithmetic on the other side: pointwise division by the spectrum is defined exactly on the sequences whose spectrum avoids the null cone, and a convolution is undone by dividing the transform pointwise by the transform of the kernel, which exists at the frequency $\ell$ exactly when $F[\ell]$ is invertible.
The convolution theorem needs the kernel to be central on the values. With $W_N(n,\ell)a=aW_N(n,\ell)$ for every sample, $\mathcal{F}[f*g][\ell]=\mathcal{F}[f][\ell]\cdot\mathcal{F}[g][\ell]$. The hypothesis holds at the central roots, whose kernel is a complex scalar, and for every sequence valued in $\mathbb{C}[\tilde{Z}]$. It is a hypothesis and not a technicality: for general quaternion-valued sequences the products $f[m]g[n-m]$ do not reassemble, because the kernel cannot be moved past the values. A convolution theorem is a commutativity statement about the kernel, and the algebra supplies one only on a commutative subalgebra.
The continuous pair has the same shape with different hypotheses. Inversion holds for an integrable continuous function and an integrable transform, and it is again unconditional in the root. Plancherel does not: it needs the Hermitian root, and it is the next subject.
The Factorization and the Norm Deficit
Four Scalar Transforms
Every biquaternion transform is four scalar complex transforms. The two minimal ideals $\tilde\Pi_\pm\mathbb{B}$ are complementary complex planes, and each has complex dimension two. A sequence splits canonically into its two ideal components, the kernel acts on each by a scalar, and the sum separates:
$$ F[\ell]=F_+[\ell]+F_-[\ell],\qquad F_\pm[\ell]=\sum_{n}e^{\mp2\pi in\ell/N}f_\pm[n]. $$
Writing each component in a complex basis of its plane turns each of the two transforms into two scalar transforms, so the count is four, independent of the root and with no case distinction. The factorization is therefore also an algorithm: four complex fast transforms of length $N$, so $O(N\log N)$ against $O(N^{2})$ for the naive evaluation. Everything non-trivial is in the root, which supplies the pair of planes.
The check is exact and it is a check of the whole construction. For a sequence of sixteen random biquaternions, the sum of the two ideal transforms reproduces the direct transform at $5.2\times10^{-14}$, and the two idempotents satisfy $\tilde\Pi_+\tilde\Pi_-=0$ and $\tilde\Pi_++\tilde\Pi_-=e_t$ to machine precision.
The factorization is where a physical reading is possible or impossible. A transform that is four scalar transforms carries four independent phases, and a physical state that is one element of the algebra uses all four at once. The two ideals are the two signs of the frequency, and a state confined to one ideal has a definite sign; a general element has both, and the interference between them is the biquaternionic content of the description.
Isometry Only at the Hermitian Root
The kernel is unitary exactly when the root is Hermitian. $W^{*}=\cos\theta\,e_t+\sin\theta\,R=W^{-1}$ holds iff $R^{*}=-R$, which is the Hermitian condition. Then multiplication by the kernel preserves the Euclidean norm on either side, and the Plancherel identity follows by summing the two complex isometries over an orthogonal split.
Away from the Hermitian roots the transform is still invertible and no longer an isometry. The split is then oblique, and the quantity the transform preserves is $\lVert\tilde\Pi_+f\rVert^{2}+\lVert\tilde\Pi_-f\rVert^{2}$, a positive form that agrees with the Euclidean one only at the Hermitian roots. The failure is measured by the size of the root, $\lVert\tilde{Z}\rVert^{2}=1+2\lVert\vec p\rVert^{2}$, which is one exactly at the Hermitian roots.
The numbers are worth stating, because a factor is the whole physical content. At the Hermitian root $\tilde{Z}=i\vec e_x$ the kernel's action on a random element has ratio $1.000$ to $1.000$ over forty phases, and the discrete Plancherel ratio is $1.0000$. At the central root the same two figures are $1.000$ and $1.0000$. At the non-Hermitian pure root $\tilde{Z}=\vec e_x+i\sqrt2\,\vec e_y$, whose squared Euclidean norm is $3$, the kernel's action has ratio between $0.855$ and $1.156$ over the same phases, and the discrete Plancherel ratio is $1.7060$. The second number is the one that matters: the same sequence, transformed by a legitimate invertible change of description, has a spectrum whose Euclidean norm is $1.7$ times the norm of the sequence.
The obliqueness has a second face. The projections onto the plane of the root and onto its complement are orthogonal at the Hermitian root and oblique otherwise: the pairing of the two projected parts is $0$ at the Hermitian root and $-6.76$ for one random element at $\tilde{Z}=\vec e_x+i\sqrt2\,\vec e_y$, as the mathematics records. The word orthogonal in the name of the complement is then no longer earned, and the estimates that use an orthogonal splitting do not follow.
What the Deficit Is
The norm deficit is the amount by which a description fails to be a basis change. A change of basis preserves every scalar product; a change of description preserves the information and may rescale the measure. At a Hermitian root the two coincide and the transform is a change of basis. Away from it the transform is a change of description in the strict sense: complete, reversible, and not length-preserving.
The corpus needs the distinction because it writes probabilities. A probability computed on one side of the transform and read on the other is the same number only at the Hermitian root. Everywhere else the read-back requires the deficit as a factor, and a computation that omits it is wrong by a number that depends on the root. This is the sharpest limit of the section, and it is why the next reading but one is the one about unitarity belonging to the element.
The Symbol: What the Change of Description Preserves
Differentiation Becomes Multiplication
The transform diagonalises a constant-coefficient operator, and the diagonal is its symbol. With the forward kernel $\exp(-2\pi i\tilde{Z}\omega t)$ and the sign of integration by parts,
$$ \mathcal{F}\left[\frac{\partial}{\partial Q_\mu}F_{\mathrm{fn}}\right](\omega)=2\pi i\,\tilde{Z}\,\omega_\mu\,\mathcal{F}[F_{\mathrm{fn}}](\omega). $$
The derivative becomes the element $2\pi i\tilde{Z}\omega_\mu$, and a gradient in four variables becomes the biquaternion $2\pi i\tilde{Z}(\omega_0e_t+\omega_1\vec e_x+\omega_2\vec e_y+\omega_3\vec e_z)$. The structural fact is that the symbol of a gradient is an element of the algebra and not a number, so the diagonal of a first-order operator is an element and the diagonal of a second-order one is a scalar.
The d'Alembertian becomes a scalar, and the sign is carried by the square of the root. Since the symbol of $\partial_\mu$ is $2\pi i\tilde{Z}\omega_\mu$ and $\tilde{Z}^{2}=e_t$, the symbol of $\partial_\mu^{2}$ is $-(2\pi\omega_\mu)^{2}e_t$, and
$$ \mathcal{F}\left[\Box_\Phi F_{\mathrm{fn}}\right](\omega)=-(2\pi)^{2}\left(\omega_0^{2}+\omega_1^{2}+\omega_2^{2}+\omega_3^{2}\right)\mathcal{F}[F_{\mathrm{fn}}](\omega). $$
The multiplier is a negative scalar and it vanishes only at $\omega=0$: this operator is elliptic, and a solution whose transform is a function has transform supported at the origin.
The operator with a cone for a zero set is the hyperbolic one. For $\partial_t^{2}-\Delta_\Phi$ the symbol is $-4\pi^{2}(\omega_0^{2}-\omega_1^{2}-\omega_2^{2}-\omega_3^{2})$, whose zero set is the null cone $\omega_0^{2}=\lvert\vec\omega\rvert^{2}$. A solution has its transform supported there. The two operators differ by the sign of the Laplacian, and that sign is the sign that $\tilde{Z}^{2}=e_t$ supplies through the square of the imaginary scalar.
The check is a single substitution. At a frequency with $\omega_0=\lvert\vec\omega\rvert$ the hyperbolic symbol is $-4.4\times10^{-15}$ and the elliptic one is $-39.478$; off the cone, at $\omega=(1,0.3,-0.4,0.5)$, the hyperbolic symbol is $-19.739$. One of them has a cone for a zero set and the other has a point, and the difference is the inertia.
The Cone and the Degenerate Set
The zero set of the symbol is the degenerate set of the transform problem. Inversion of an operator equation by the transform is division by $\Sigma(\omega)$, and the division is available exactly where $\Sigma(\omega)$ is invertible. The set where it is not invertible is where the norm vanishes, that is the intersection of the symbol's range with the null cone; the mathematics calls it the degenerate set and computes the rank of left multiplication there.
The rank drop is by two, and the projector shows it. Left multiplication by a unit element is an invertible map of the four complex dimensions of the algebra; left multiplication by the idempotent $\tilde\Pi_+$ has rank two and determinant zero. The check is direct: $\operatorname{rank}L_{\tilde\Pi_+}=2$ with $\det L_{\tilde\Pi_+}=0$, $\operatorname{rank}L_{\tilde{Z}}=4$, and $\operatorname{rank}L_{e_t+i\vec e_y}=2$. The last element is a zero divisor and the first is a projector, and both sit at rank two.
The adjugate is the reciprocal, and it survives the degenerate set. Since $\tilde{Q}\bar{\tilde{Q}}=N(\tilde{Q})e_t$ with $N$ central, the reciprocal is $\Sigma^{-1}=\bar\Sigma/N(\Sigma)$ off the zero set, and $\det L_\Sigma=\lvert N(\Sigma)\rvert^{4}$. The adjugate $\lvert N\rvert^{2}\overline{N(\Sigma)}L_{\bar\Sigma}$ is defined everywhere, so the algebra supplies a generalised inverse in the cases where the ordinary one does not exist. That is the subject of the sixth row of the source table, and it is where a propagator on the cone has to be read.
The Licence: Constant Coefficients
The change of description is a change of basis exactly for constant coefficients, and the corpus states the condition instead of assuming it. For a constant-coefficient operator the symbol is a multiplier, the equation becomes algebraic, and the transform is exact. For a variable-coefficient operator the symbol is a function of position as well as of frequency, the equation does not become algebraic, and the transform is a change of variables that solves nothing by itself. The corpus's wave-number transcriptions are therefore transcriptions of equations whose coefficients are constant, and the licence is exactly that.
Centrality enters one step later. Even with constant coefficients, the transformed equation is an equation between elements only when the multiplier commutes with the unknown. That is the same hypothesis as the convolution theorem, met for the second time: an equation with the unknown on one side of the multiplier and a source on the other is an equation in a non-commutative algebra, and it must be solved as one.
Inversion, Poles and Tails
The Two Contours
The conjugate kernel splits into two scalar exponentials. For every root and every real $t$ and $\omega$,
$$ \overline{W(t,\omega)}=\tilde\Pi_+e^{2\pi i\omega t}+\tilde\Pi_-e^{-2\pi i\omega t}, $$
an identity that uses no commutation, since its only products are powers of the root. A contour integral of $\Sigma^{-1}\overline{W}$ is therefore a sum of two scalar integrals, each closing in the half-plane where its own exponential decays, and each picking up the residues of its own projection.
The consequence is that a kernel in one variable is two-sided even when all its poles lie in one half-plane. The same pole contributes through $\tilde\Pi_+$ on one side of the time axis and through $\tilde\Pi_-$ on the other, so there is no single contour and no single residue. The two sides are governed by different projections of the same residues, and the type of the kernel is a pair of numbers. Only in the plane case does one contribution vanish, which is why the plane case looks like a single scalar calculus. The mathematics verifies the expansion on a symbol that does not commute with the root: a direct numerical integration agrees with the residue expansion to $1.5\times10^{-7}$ relative at three times over $\lvert\omega\rvert\le60$, and to $1.4\times10^{-10}$ when the truncation and the point count are pushed out together.
The two-sidedness is not an artefact of the algebra. It is the statement that the algebra has two ideals, and an integral that resolves the kernel resolves them separately. A one-sided reading of a two-sided kernel is the error that the placements are about to make precise.
The Type Is the Distance
The exponential type of the kernel is $2\pi\rho$, where $\rho$ is the distance of the nearest pole from the real axis. With $\Sigma$ plane-valued and every pole off the axis, each ideal's contribution is bounded by $C(1+\lvert t\rvert)^{k-1}e^{-2\pi\rho\lvert t\rvert}$ for pole order $k$, and the bound is attained, so the type is not smaller. A solution of compactly supported data inherits the bound, with the radius of the support subtracted from $\lvert x\rvert$.
A pole on the axis gives no decay at all. The contour cannot be closed, the integral is a principal value, and the residue contributes half of itself. The kernel is then merely bounded, and the mathematics records that the inversion is not a bounded operator on $L^{2}$. The two facts are the same fact, and it is the fact behind every statement about a continuum.
Compact support forces a vanishing datum at every pole. A function of compact support has an entire transform, so a solution whose transform carries a pole cannot be of compact support unless the source's transform vanishes at that pole. The mathematics states it exactly: if $\hat g(\omega_0)\neq0$ at a pole of $\Sigma^{-1}$, the solution is not of compact support. A source that switches on and off cannot excite a pole without leaving a tail.
In More Than One Variable
In one variable the expansion is proved, commuting or not; in more than one it is open, for a geometric reason. The singular set $\{\zeta:N(\Sigma(\zeta))=0\}$ is, after complexification, a hypersurface of complex dimension $n-1$, that is of real dimension $2n-2$ — six in eight for $n=4$. A set of real codimension two has no point to encircle, so a residue cannot be taken at a point. What replaces it is a residue current or an integral of the Leray form over a real cycle that links the hypersurface. The mathematics names both routes — Coleff–Herrera with analytic continuation, and the Leray form over a cycle — and proves neither.
Two points remain open and neither is hidden. The kernel in $n\ge2$ as a sum of contributions, for which the residue current is the candidate and not the answer; and the points where the gradient of the norm vanishes, which the null cone meets at its vertex. On the real frequencies the count is different again: the equation is two real equations, so the set has real codimension two when they are independent, codimension one when the norm is real-valued, and is generically empty in one variable.
The Three Placements
One Integral, Three Products
All three transforms integrate the same kernel against the same value; what changes is the side on which the kernel meets it. The left placement multiplies on the left, the right placement on the right, and the two-sided placement sandwiches the value between two kernels. For a scalar-valued function the three products coincide, which is why the complex theory needs one name; for this algebra they are three operators, and the difference is visible only in a non-commutative algebra.
The commutant of a pure root is the plane of the root. For a pure root $R$ with $R^{2}=-e_t$, an element $a$ satisfies $aR=Ra$ exactly when it lies in $\mathrm{span}_{\mathbb{C}}\{e_t,R\}$, a real subspace of dimension four. The plane $C(R)$ is commutative, and the complementary subspace $V_\perp$, also of real dimension four, is spanned by a transverse frame $\nu,\eta$ with $R\nu=\eta$, $\nu R=-\eta$, $\eta R=\nu$ and $R\eta=-\nu$.
Conjugation by the kernel fixes the plane and rotates the complement. With $\theta=2\pi\omega t$,
$$ W\nu W^{-1}=\cos2\theta\,\nu-\sin2\theta\,\eta,\qquad W\eta W^{-1}=\sin2\theta\,\nu+\cos2\theta\,\eta, $$
so the conjugation is the identity on $C(R)$ and a rotation through the angle $4\pi\omega t$ on $V_\perp$. The angle is twice the phase of the kernel, and the factor two is the first appearance of the doubling that governs the two-sided placement.
The two projections onto plane and complement are orthogonal exactly at the Hermitian root. For a pure non-Hermitian root they remain idempotent and complementary and stop being perpendicular, and the pairing of the two images is then non-zero. A decomposition that is a decomposition of vector spaces is not thereby an orthogonal decomposition, and the difference is measurable.
The Two Signs That Fix a Part
There are two two-sided placements and not more, because only two signs of the second root fix a part of the algebra. With $R'=R$ the second kernel is $W^{-1}$ and the placement is conjugation by $W$, which fixes $C(R)$ pointwise. With $R'=-R$ the second kernel is $W$ itself, and the placement fixes the transverse plane pointwise while multiplying the plane by $W^{2}$. With any other second root a residue survives in every summand, so nothing is fixed.
Each sign forgets one half, and only its total integral survives. At $R'=R$ the transform vanishes at every frequency exactly for plane-valued functions of zero total integral; at $R'=-R$ it vanishes exactly for transverse-valued functions of zero total integral. Neither sign reduces the two-sided placement to a one-sided one, because a one-sided kernel never cancels. The two-sided transform is not injective, and its kernel is an infinite-dimensional space in each case.
The surviving half is carried at twice the frequency, and the inversion carries a factor two. At $R'=R$ the transverse coefficients are rotated through $4\pi\omega t$, so the pair is the complex transform of $A_1+iA_2$ at frequency $2\omega$; at $R'=-R$ the plane part is transformed at $2\omega$. The inversion must therefore integrate over $d(2\omega)$ and carries a factor two, and an inversion written without it is wrong in the argument — an error that is invisible on a single exponential, which is why it survives review.
The doubling is checkable in one kernel. For $f(t)=e^{-\pi t^{2}}\nu$ and $R'=R$ the two-sided transform is $e^{-4\pi\omega^{2}}\nu$ at every real frequency, and the exponent is four times the exponent of the one-sided transform. Measured on a grid of six thousand points: the coefficient of $\nu$ is $1.00000000$, $0.60492256$ and $0.13390572$ at $\omega=0$, $0.2$ and $0.4$, against $e^{-4\pi\omega^{2}}=1.00000000$, $0.60492256$ and $0.13390572$, with a transverse admixture of at most $2.4\times10^{-16}$.
The Conventions, Stated as a Table
| sign of the second root | part fixed by the kernel | part carried at doubled frequency | part whose frequency is lost |
|---|---|---|---|
| $R'=R$ | the plane of the root | the transverse plane | the plane of the root |
| $R'=-R$ | the transverse plane | the plane of the root | the transverse plane |
A placement is a convention because the algebra does not choose it. The left and the right placement differ by the conjugation of the values, and neither is privileged; the two-sided placements differ by a sign and neither is privileged. What is not a convention is the table: given a placement, the part it forgets is fixed. The invariance groups of the three placements intersect in the plane of the root, and that intersection is the set of operations on which the conventions agree.
Inversion on the Degenerate Set
The Adjugate Over the Norm
The reciprocal of a symbol is its adjugate over its norm. For every $\tilde{Q}$, $\tilde{Q}\bar{\tilde{Q}}=\bar{\tilde{Q}}\tilde{Q}=N(\tilde{Q})e_t$ with $N$ central and complex-valued, so off the zero set $\Sigma^{-1}=\bar\Sigma/N(\Sigma)$. In the regular representation the same identity reads $L_\Sigma^{-1}=N(\Sigma)^{-1}L_{\bar\Sigma}$, and $\det L_\Sigma=\lvert N(\Sigma)\rvert^{4}$.
The identity is the reason the degenerate set is a set and not an accident. The determinant of the left multiplication vanishes exactly where the norm of the element does, which is a complex quadric, and the rank falls by at least two at a simple null point. An element at rank two is a zero divisor; the projector $\tilde\Pi_+$ is the model, and every null element is conjugate to a projector or is nilpotent.
Two Generalised Inverses
Where the ordinary inverse does not exist, two generalised inverses do. The mathematics constructs them from the adjugate and from the Peirce coordinates respectively, and it states which one reproduces which class of solutions. The distinction matters at a pole, because a generalised inverse selects a solution and the selection is not unique.
Inversion where the symbol is nowhere an invertible element is a different problem. If the symbol is a zero divisor at every frequency, no division is available anywhere and the equation is not solved by a reciprocal at all. What remains is the algebraic inversion on the degenerate set, which is a statement about elements and not about distributions, and the exponential there, which truncates: on the cone the flow is $e^{\theta Q_t}(e_t+\theta\vec N)$ with $\vec N^{2}=0$, a polynomial in the parameter and not a periodic function.
The truncation is the algebraic form of a critical point. A flow that is a polynomial rather than a circle has no period and no frequency, which is the same statement as the absence of a root on the cone. The two facts are one fact read in the flow and in the kernel.
The Modular Pair
The Hilbert Algebra and the Tomita Operator
The algebra is a Hilbert algebra for the trace and the star. With $\langle\tilde{Q}_1,\tilde{Q}_2\rangle=\mathrm{Sc}(\tilde{Q}_1^{*}\tilde{Q}_2)$ the form is positive definite and Hermitian, and the adjoint axiom holds by associativity alone. The unit $\zeta=e_t$ is cyclic and separating for the left regular image, since $L_{\tilde{Q}}\zeta=\tilde{Q}$ is onto and injective, so the Tomita operator of the pair is defined.
The Tomita operator is the star. By definition $S(\Lambda\zeta)=\Lambda^{*}\zeta$ for $\Lambda=L_{\tilde{Q}}$, and the definite adjoint of a left multiplication is $L_{\tilde{Q}^{*}}$, so $S(\tilde{Q})=\tilde{Q}^{*}$. It is antilinear, involutive and antiunitary, with $\langle S\tilde{Q}_1,S\tilde{Q}_2\rangle=\overline{\langle\tilde{Q}_1,\tilde{Q}_2\rangle}$ and $\lVert S\tilde{Q}_1\rVert=\lVert\tilde{Q}_1\rVert$; both identities are exact in the eight real dimensions, measured at $0$ to machine precision.
The modular operator is the identity and the modular flow is trivial. Since $S$ is antiunitary, its polar decomposition is $S=\jmath$ with $\Delta_\Phi=S^{*}S=\mathrm{id}$, so $\sigma_t(\Lambda)=\Lambda$ for every $t$. The modular conjugation exchanges the sides, $\jmath L_{\tilde{Q}}\jmath=R_{\tilde{Q}^{*}}$, which is the definite shadow of the indefinite identity $(L_{\tilde{Q}})^{\dagger}=R_{\bar{\tilde{Q}}}$.
Why the Flow Is Trivial
The flow collapses because the vector state of the unit is the trace. $\omega(L_{\tilde{Q}})=\mathrm{Sc}(\tilde{Q})=Q_t$, and $\mathrm{Sc}(\tilde{Q}\tilde{Q}_1)=\mathrm{Sc}(\tilde{Q}_1\tilde{Q})$, so the functional is tracial; the modular operator is the identity exactly when the vector state is a trace. The triviality is therefore a property of the vector and not of the algebra, and that distinction is the point: $M_2(\mathbb{C})$ has vectors whose state is not the trace, and a different vector would give a non-trivial flow.
The Indefinite Obstruction
On the canonical Krein space the operator is not definable. The indefinite adjoint of a left multiplication is a right multiplication, $(L_{\tilde{Q}})^{\dagger}=R_{\bar{\tilde{Q}}}$, which lies in the image only for central $\tilde{Q}$. The left regular image is therefore not closed under the indefinite adjoint, so it is not a Krein–von Neumann algebra, and the map $\tilde{Q}\mapsto\tilde{Q}^{\dagger}$ leaves the algebra. The smallest Krein–von Neumann envelope is $\mathbb{B}\otimes\mathbb{B}_{\mathrm{opp}}\cong M_4(\mathbb{C})$, of complex dimension sixteen, and it has no separating vector: for every $\zeta\neq0$ there is a non-zero $T$ with $T\zeta=0$. So the envelope carries no modular pair either.
The twisted symmetries work and the modular conjugation does not commute with them. For $J'=R_c$ with $c^{*}=c$ and $c^{2}=e_t$ the image is a $J'$-algebra with $S'={}^{*}$ and $\Delta_\Phi'=\mathrm{id}$, but the modular conjugation must commute with the fundamental symmetry, and $\jmath'J'(\tilde{Q})=c\tilde{Q}^{*}$ against $J'\jmath'(\tilde{Q})=\tilde{Q}^{*}c$, which agree only for central $c$. The definite theory is available, the indefinite one is available only in the definite case, and the gap is a commutator.
The Interpretations
III.2.a — The Kernel Is the Clock
The object. $W(t,\omega)=\cos\theta\,e_t-i\sin\theta\,\tilde{Z}$ with $\theta=2\pi\omega t$, an invertible element of the algebra for every $t$, and its Peirce coordinates $e^{\mp2\pi i\omega t}$ on the two ideals.
The warrant. The kernel is a unit circle in the commutative plane $\mathbb{C}[\tilde{Z}]$, its algebraic norm is the unit at every phase, and its inverse is the kernel at the negated frequency. The phase advances linearly in $t$ and the rate of advance is $\omega$; nothing else in the algebra advances.
The reading. A clock needs a reference, a period and a sense. The root supplies the reference: it is the axis about which the phase turns, and a different root is a different clock. The frequency supplies the period: one turn is $t=1/\omega$. The kernel supplies neither a sense nor a rate, and it supplies them for a reason: $W(-t,\omega)=W(t,-\omega)$, so reversing the time and negating the frequency are one operation, and the algebra contains no rule that selects one of the two. The archive's Relational Time from the Sector Exchange makes the same point from the other end, for the central phase: the algebra supplies the reference and the period of a clock and neither the arrow nor the rate. What this entry adds is that the clock is indexed by a root: at a pure root there are two counters, running at opposite rates on the two ideals, and a state confined to one ideal reads a definite sign of the frequency while a general element reads both. That is the algebraic origin of the sign of the frequency, and it is where a two-component clock becomes a one-component one only by projection.
III.2.b — The Frequency Is the Rate
The object. The frequency $\omega$ conjugate to $t$, and the multiplier $2\pi i\tilde{Z}\omega_\mu$ that the transform assigns to $\partial_\mu$.
The warrant. Differentiation of the kernel multiplies it by $2\pi i\tilde{Z}\omega_\mu$, so the transform converts a rate of change into a label. The label is an element and not a number, and its four coordinates are the four rates.
The reading. A frequency is a rate of phase, and a wave number is a rate of phase per unit length; the two are the components of one element, which is why the corpus writes a four-wave-number $\tilde{K}_\Phi=i k'_te_t+\vec k_\Phi$ and not a frequency and a vector. The identification of $2\pi\omega$ with an energy requires $\hbar$ and is an input; the algebra supplies the shape of the identification and none of the number, and the entry states that boundary rather than crossing it. The Three Wavenumbers of a Particle carries the three rates that the corpus distinguishes, and Derivatives, Fields and Transforms the general dictionary.
The limit. The rate is a label of a description and not a property of an object. The same element transformed at two different roots carries two different pairs of rates, and neither is privileged by the algebra.
III.2.c — Unitarity Belongs to the Element, Not to the Transform
The object. The Plancherel identity, and the condition on the root under which it holds: $\tilde{Z}^{*}=\tilde{Z}$.
The warrant. The isometry of the kernel is $W^{*}=W^{-1}$, which is the Hermitian condition on the root; the root supplied by a regular element $\tilde{Q}$ is Hermitian exactly when $\tilde{Q}$ is normal; and the norm deficit of a non-Hermitian root is $\lVert\tilde{Z}\rVert^{2}=1+2\lVert\vec p\rVert^{2}$.
The reading. "The transform is unitary" is not a property of a transform. It is a property of the pair (transform, element), and through the element of the state. A normal element has a unitary flow and a norm-preserving description; a non-normal element has an invertible description that rescales, by the factor measured above, and a physics that reads a probability across that description without the factor is wrong. The archive states the same thing positively: Unitarity from Centrality derives norm preservation from a central unit, The Born Rule from the Hermitian Form builds the weight from the Hermitian pairing, and The Probability Band records the three species that the probability product distinguishes. What this entry adds is the negative half: the description is available far beyond the reach of the isometry, and the two must not be confused.
The limit. The condition is on the element and not on the transform, so no choice of kernel repairs a non-normal element. A root change moves the deficit; it does not remove it.
III.2.d — The Poles Are the Degenerate Set
The object. The zero set of the symbol $\Sigma(\omega)$, the set where $N(\Sigma)$ vanishes, and the poles of $\Sigma^{-1}=\bar\Sigma/N(\Sigma)$.
The warrant. Division by the symbol is available exactly where the symbol is invertible, the rank of left multiplication falls from four to two at a simple null point, and the transform of a solution of $\Sigma(\hat f)=\hat g$ is supported on the zero set of $\Sigma$.
The reading. A pole of a propagator is a zero of a symbol, and a zero of a symbol is a shell. The mass shell is the pole set of the massive symbol, the null cone is the pole set of the massless one, and the exceptional point is where two poles meet: the archive's The Zero-Divisor Cone and the Massless Spectrum, The Massless Limit as a Boundary Value of the Polar Word and The Rest Frame and the Exceptional Point are the three cases. The sharp edge that the mathematics states and the physics must not soften is this: the support of a transform is the pole set of its partner's kernel, so a coupling that enters the symbol linearly can translate the pole set and cannot reshape it. A shift moves a shell; it does not make a continuum.
The limit. The statement is about a symbol, and a symbol is the transform of a constant-coefficient operator. For a position-dependent coupling the pole set is a set of functions and the reading becomes a statement about a dispersion relation at each point, which is a different and weaker object.
III.2.e — The Three Placements Are Three Conventions
The object. The left, right and two-sided transforms, and the two signs $R'=\pm R$ of the second root.
The warrant. The three placements coincide for a scalar-valued function and differ for an element-valued one; the commutant of a pure root is its plane; the two signs fix complementary parts and lose the frequency of the part they fix; no other sign fixes anything.
The reading. A convention in this algebra is a choice of the side on which the algebra acts, and a convention has a price that can be stated before it is paid. The two-sided placement is the algebraic shape of a similarity transformation: an operation that acts on both sides of an element is blind to the phase of the part it fixes, which is exactly the blindness of a dressed observable $A\mapsto UAU^{-1}$ to the phase of the sector $U$ preserves. The factor two is the price of the surviving half, and it is a real accounting rule: the surviving part is described at twice the frequency, so an inversion written with $d\omega$ and no factor is wrong in the argument. The Pauli Correspondence and the Matrix Image is where the corpus uses a two-sided action in practice, and it is where the factor is absorbed.
The limit. The placements are three descriptions of one object, so no experiment distinguishes them; what an experiment can see is the doubling, and the doubling is a property of a two-sided kernel and not of the algebra.
III.2.f — The Transform of a Measure Is Not Positive Definite
The object. $G(\omega)=\int\chi_\omega\,d\mu$, its value $G(0)=\mu(\mathbb{R}^{n})e_t$, and the quadratic form $\sum_{k,l}\bar A_kA_lG(\omega_k-\omega_l)=\int\langle Z\rangle_{\natural}\,d\mu$.
The warrant. The identity is one line, because $\bar\chi_\omega\chi_\eta=\chi_{\eta-\omega}$ at a pure root; the right-hand side is the integral of the natural norm; and the natural norm is indefinite, vanishing on the null cone and negative on the centre. At $N=1$ with $Z=i e_t$ the integral is $-\mu$, since $\langle i e_t\rangle_{\natural}=i^{2}=-1$, while the star pairing gives $\langle i,i\rangle_{*}=+1$.
The reading. Bochner's theorem needs a positive-definite function, and positivity needs a definite form. This algebra has one and it is not the norm: the norm is a signed weight, and the sign is the sign of the sector. A density read through the norm is therefore not a probability density but a charge-like weight, and a probability is read through the star pairing, which is positive by construction and not multiplicative. The archive's probability articles are the positive half of this statement; what this entry adds is the negative half, which is that no theorem of Bochner type can be stated in the norm and that the failure needs neither a zero divisor nor an exceptional value.
The limit. The repair by the Euclidean pairing costs multiplicativity: $\lVert ZW\rVert\neq\lVert Z\rVert\lVert W\rVert$ in general, with $\lVert\tilde\Pi^{2}\rVert=\tfrac12$ against $\lVert\tilde\Pi\rVert^{2}=\tfrac14$ for the Hermitian idempotent $\tilde\Pi=\tfrac12(e_t+i\vec e_x)$. Positivity and multiplicativity are two properties and the algebra supplies them from two different pairings.
III.2.g — The Tail Is the Decay, and the Pole on the Axis Is the Stable Case
The object. The exponential type $2\pi\rho$ of the kernel, the pole on the real axis, and the compact-support criterion.
The warrant. The residue at a pole at distance $\rho$ contributes $e^{-2\pi\rho\lvert t\rvert}$ times a polynomial, the bound is attained, a pole on the axis gives only boundedness and no $L^{2}$-bounded inversion, and a compactly supported solution needs a vanishing datum at every pole of the symbol.
The reading. A resonance is a pole off the axis: its distance from the axis is its width, and the width is the decay rate of the kernel that the pole produces. A stable particle is a pole on the axis: no decay, no width, and an inversion that is not bounded. The criterion for compact support is the algebraic form of a familiar fact, that a source which switches on and off cannot excite a pole without a tail. The archive's decay and scattering articles read the poles at the level of the amplitude; this entry supplies the statement that the pole and the tail are the same object.
The limit. The exponential law is the leading behaviour of a residue, and a sum of residues is not an exponential. The exact statement is the bound and its attainment, not a decay law.
III.2.h — The Modular Clock Is Trivial, and That Is a Statement About the Vector
The object. The Tomita operator $S={}^{*}$ of the left regular image at the unit, its polar decomposition $\Delta_\Phi=\mathrm{id}$, and the modular flow $\sigma_t=\mathrm{id}$.
The warrant. The unit is cyclic and separating, the vector state is $\mathrm{Sc}$, and $\mathrm{Sc}$ is the trace; a tracial vector state gives a trivial modular operator. The indefinite version does not exist, because the image is not closed under the indefinite adjoint and its envelope has no separating vector.
The reading. The algebra alone carries no thermal time. A modular flow is a clock that a state supplies to an algebra, and the algebra's canonical state, the unit, supplies a stopped clock: the flow is the identity at every parameter. So a thermal reading needs either a different state or a larger algebra, and the corpus's thermal articles take the second route: Curved-Spacetime KMS and Modular Flow and The Bisognano–Wichmann Theorem under the Biquaternion Framework work with the modular flow of a wedge, which is not the flow of the unit, and The Unruh Effect in Biquaternionic Form reads that flow by an accelerated observer. The triviality is therefore not a defect of the algebra: it is the exact statement of which vector the algebra's own clock belongs to.
The limit. The indefinite modular theory is unavailable, so no thermal statement is claimed from the Krein structure. The twisted symmetries supply a modular pair with $\Delta_\Phi'=\mathrm{id}$, and their modular conjugation does not commute with the fundamental symmetry except for a central parameter.
The Limits
No dynamics. The section supplies a change of description and no equation. It selects no element, states no Lagrangian, and does not claim that any particular root is the physical one. The transform of an equation is exact; the choice of equation is elsewhere.
No constant is derived. $\hbar$ does not appear. The identification of a frequency with an energy, of a wave number with a momentum, of a pole with a mass and of a type with a lifetime are inputs of the corpus, and the entry states each of them as an input.
The isometry is restricted. Every norm-preserving statement requires a Hermitian root, hence a normal element. The general transform is invertible and rescaling, and the deficit is a number that must be carried.
Inversion in more than one variable is open. The residue at a point does not exist in codimension two, and the two candidate replacements are named and not used. Any statement in this entry about a propagator in four variables is a statement about the one-variable reduction or a statement about a symbol, not a contour integral in four variables.
The modular theory is the definite one. The indefinite version is unavailable, and the entry says so instead of quoting a theorem whose hypothesis fails.
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, and neither is a new physical claim.
Summary
The transform of this algebra is generated by a root of $1$ and a frequency. The kernel is $\exp(-2\pi i\tilde{Z}\omega t)$, closed at once into a cosine and a sine because the square of the root is the unit. It is invertible at every phase and every root with $N(W)=1$, its Peirce coordinates are two conjugate unit scalars, and it is never a zero divisor. The root is the only non-trivial datum, and every statement of the section is indexed by it.
Inversion is unconditional and the isometry is not. The inverse formula reproduces every sequence and every integrable function at every root, including samples on the null cone. Plancherel holds exactly at the Hermitian roots; elsewhere the transform preserves the Peirce-weighted form, and the deficit is $\lVert\tilde{Z}\rVert^{2}=1+2\lVert\vec p\rVert^{2}$, measured as a factor of $1.7060$ in the discrete case at $\tilde{Z}=\vec e_x+i\sqrt2\,\vec e_y$.
The transform is four scalar transforms. The two minimal ideals are complementary complex planes, the kernel acts on each by a scalar, and the count of four complex transforms is independent of the root. The factorization gives the $O(N\log N)$ algorithm and locates the whole difficulty in the root.
It diagonalises constant-coefficient operators and only those. Differentiation becomes multiplication by $2\pi i\tilde{Z}\omega_\mu$, the d'Alembertian becomes a negative scalar, and the hyperbolic operator's symbol vanishes on the null cone while the elliptic one's vanishes only at the origin. The licence for a wave-number transcription is the constancy of the coefficients, and centrality is needed one step later.
Inversion is two contours. The conjugate kernel splits into the two ideals, a pole contributes through one ideal on one side of the axis and the other on the other side, and the type is $2\pi\rho$. A pole on the axis gives boundedness and no decay, and compact support forces a vanishing datum at every pole. In more than one variable the singular set has real codimension two and the residue is replaced by a current or a linking cycle, both named and neither proved.
A placement is a convention with a stated price. The two-sided placements fix complementary parts and lose the frequency of the part they fix, carrying the survivor at twice the frequency, and the inversion carries a factor two. No other sign fixes anything, and no two-sided transform is injective.
The degenerate set is where the symbol stops being an invertible element. The rank of left multiplication falls from four to two, the adjugate supplies a generalised inverse, and the exponentiation truncates on the cone. The modular pair of the algebra is trivial because the unit's vector state is the trace, and the indefinite version does not exist.
Eight readings, each with a boundary. The kernel is the clock, the frequency its rate, unitarity a property of the element, the poles the degenerate set, the placements conventions, the measure transform not positive definite, the tail the decay and the axis the stable case, and the modular clock trivial for the algebra's own vector. 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 $\mathbb{C}\otimes_{\mathbb{R}}\mathbb{H}$ with basis $e_t,\vec e_x,\vec e_y,\vec e_z$ |
| the root of one | $\tilde{Z}$ | $\tilde{Z}^{2}=e_t$; central $\pm e_t$ or pure with $(\vec{Z},\vec{Z})=-1$ |
| the idempotents of the root | $\tilde\Pi_\pm$ | $\tfrac12(e_t\pm\tilde{Z})$, orthogonal, summing to $e_t$ |
| the kernel | $W(t,\omega)$ | $\exp(-2\pi i\tilde{Z}\omega t)=\cos\theta\,e_t-i\sin\theta\,\tilde{Z}$ |
| the phase | $\theta$ | $2\pi\omega t$ |
| the transform | $F,\ \mathcal{F}[f]$ | kernel against the value, left placement unless stated |
| the frequency | $\omega$ | the variable conjugate to $t$ |
| the symbol | $\Sigma(\omega)$ | multiplier of a constant-coefficient operator under the transform |
| the adjugate | $\bar\Sigma$ | $\Sigma\bar\Sigma=N(\Sigma)e_t$ |
| the norm of an element | $N(\tilde{Q})$ | $Q_t^{2}+Q_x^{2}+Q_y^{2}+Q_z^{2}$, central and complex |
| the degenerate set | $\{N(\Sigma)=0\}$ | where the symbol is not invertible |
| the plane of the root | $C(R)$ | $\mathrm{span}_{\mathbb{C}}\{e_t,R\}$, the commutant of a pure root |
| the transverse frame | $\nu,\eta$ | $R\nu=\eta$, $\nu R=-\eta$, $\eta R=\nu$ |
| the two placements | $R'=\pm R$ | the two-sided kernels that fix a part |
| the exponential type | $2\pi\rho$ | $\rho$ the distance of the nearest pole from the real axis |
| the Tomita operator | $S$ | the star, with $\Delta_\Phi=\mathrm{id}$ and $\jmath=S$ |
| the Krein envelope | $\mathbb{B}\otimes\mathbb{B}_{\mathrm{opp}}$ | $M_4(\mathbb{C})$, complex dimension sixteen |
Further Reading
- Biquaternion Discrete Harmonic Analysis, the discrete pair, the classification of the roots, the factorization and the $Z$ transform.
- Biquaternion Continuous Harmonic Analysis, the continuous pair, Plancherel, the symbol and the failure of positivity.
- Relations between the Biquaternion Exponential and the Fourier Kernel, the kernel as the flow of one element and the normality that makes it an isometry.
- Inversion of the Biquaternion Fourier Transform: Poles, Tails and the Degenerate Set, the two contours, the type and the obstacle in codimension two.
- The Three Kernel Placements of the Biquaternion Fourier Transform, the placements, the rotation of the complement and the two signs.
- Biquaternion Fourier Inversion on the Degenerate Set, the rank drop and the two generalised inverses.
- The Biquaternion Algebra and the Krein–von Neumann Property, the indefinite adjoint and the envelope.
- The Tomita Operator and the J-Modular Pair for the Biquaternion Algebra, the modular pair and its obstruction.
- Wave–Particle Duality and the Fourier Transform, the companion entry, for the transform as the switch between the two descriptions.
- The Derivative, the Field and the Transform, the preceding entry, for the operator whose symbol this entry diagonalises.
- The Derivative, the Field and the Transform, the archived dictionary of the symbol and the wave-number transcription.
- Relational Time from the Sector Exchange: The Material and Informational Clocks, the archived reading of the clock that the algebra supplies and the arrow it does not.
- The Three Wavenumbers of a Particle, the archived reading of the three rates.
- The Group of Invertibles, the Cone and the Orders, the archived reading of the massless pole set.
- The Massless Regime, the archived reading of the pole at the vertex.
- The Rest Frame and the Exceptional Point, the archived reading of the meeting of two poles.
- The Group of Invertibles, the Cone and the Orders, the archived positive half of the isometry condition.
- States, Observables and the Born Weight, the archived reading of the probability as a pairing and not a norm.
- The Partition: Two Invariants and Four Regimes, the archived reading of the three species of the probability product.
- Curved-Spacetime KMS and Modular Flow, the archived thermal reading, which needs a state other than the unit's.
- The Bisognano–Wichmann Theorem under the Biquaternion Framework, the archived identification of a modular flow with a boost.
- The Unruh Effect in Biquaternionic Form, the archived reading of that flow by an accelerated observer.