An External Construction of Space from Quaternionic Spinors

Introduction

The corpus meets the quaternion algebra from inside its own programme: the algebra is chosen, the physical interpretations are made afterwards, and every claim is a claim about what follows from the choice. This article records a source that arrives from the opposite direction. Its author is not working in the biquaternion programme and does not use the corpus's conventions; he takes a bi-quaternionic field as the primary object and reports that the tetrad, the torsion and the contorsion are all bilinear composites of it, so that the spacetime metric is precipitated by the field rather than assumed. The paper is Julian G. B. Northey, "An External Construction of Space: Unifying Dirac and Einstein–Cartan through Quaternionic Spinors", International Journal of Quantum Foundations 11 (2025) 444–469.

The source is ambitious and its ambitions overlap the corpus's gaps: a single action from which the Dirac, Einstein–Cartan and spin–torsion equations are said to follow; an intrinsic arrow of time from the field's phase; a torsion-induced quantum potential; and a conserved torsion entropy current. It also reaches into material the corpus has no equivalent of at all — Kozyrev's causal time, a torsion Bohm potential, torsion waves, and three proposed experiments with numerical sensitivities.

The article has one job, and it is deliberately narrow and unsympathetic. It records the construction, and it tests every display that can be tested by recomputation. The quoted displays keep the source's spelling, including the indices it raises; every display of this article's own is written with all indices low, because the corpus has one metric, the unit metric, and a raised coefficient index would name a metric the corpus does not have. The result is a mixed one. Several of the source's displays do not reproduce, and the worst of them go to the construction itself rather than to a typo. The composite tetrad vanishes identically on every plane wave, which is to say on the free Dirac solutions it is meant to describe. The torsion defined by the source's own Eq. (5) is nonzero on three of its twenty-four independent components, where the Einstein–Cartan spin density the same section equates it to is totally antisymmetric and carries four. The third is different in kind: the Fierz identity of Eq. (3) does reproduce, in every reading, and it is empty. Three of its four terms vanish identically for every field, so all it states is that the quaternion norm is multiplicative, whatever the value of its coefficient.

Nothing here endorses the source's programme, and nothing here uses it as evidence for the corpus's. One paper that finds the algebra natural shows that the algebra can be found, not that it must be. The corpus's own position is stated in Curved Spacetime and the Biquaternion Framework and in Torsion, Contorsion and the Spin Connection in Biquaternionic Form: the algebra can carry the frame and the connection, and it generates neither. The source's programme is exactly the step the corpus declines to take — geometry generated by matter — and this article's job is to say precisely how far the step can be taken before it fails.

Frame, gradient and pairing. Displays are read with $\varepsilon=(1,-1,-1,-1)$, the weight of the $\mathbb{B}_{\Phi}$ frame of the source's own matrices; the frame names the coordinate and the metric, and the generators carry no frame label. The source's $q_\mu$ keep the source's notation: $q_0=1$ and $q_\eta$ the three imaginary quaternion units, the corpus's $e_t$ and $\vec e_\eta$, and $q_\mu$ acting on a field is the left multiplication $\tilde{Ξ}\mapsto e_\mu\tilde{Ξ}$.

The derivative of this article is the corpus's biquaternionic gradient, and it is the only first-order operator on the corpus's side of the comparison: $$ \tilde\nabla_\Phi=\sum_\mu e_\mu\frac{\partial}{\partial R_\mu}, \qquad (\tilde\nabla_\Phi)^{\natural}=\sum_\mu\varepsilon_\mu e_\mu\frac{\partial}{\partial R_\mu}, \qquad \Box_\Phi=\mathrm{Sc}\bigl(\tilde\nabla_\Phi(\tilde\nabla_\Phi)^{\natural}\bigr)=\frac{\partial^2}{\partial R_t^2}+\Delta_\Phi , $$ which on the material sector reads $\Box_\Phi=\Delta_\Phi-\partial^2/\partial {r'_{t}}^2$. The first-order dynamics is the one-sided pair $\tilde\nabla_\Phi\tilde{Ξ}_{R}=k_m\tilde{Ξ}_{L}$, $(\tilde\nabla_\Phi)^{\natural}\tilde{Ξ}_{L}=k_m\tilde{Ξ}_{R}$, and $\tilde\nabla_\Phi^2$ is not $\Box_\Phi$: the two differ because $e_t$ commutes with the three spatial units and the mixed derivatives then add, so no single field carries the mass at first order. Here $\hbar/(mc)$ is the mass length.

The source's four generators act on the field, and they are written here as the actions they are: $$ C_t:\tilde{Ξ}\longmapsto i\tilde{Ξ}\vec e_x,\qquad C_\eta:\tilde{Ξ}\longmapsto i\vec e_\eta\tilde{Ξ}\vec e_y,\qquad \mathsf{V}_{\mathrm{vol}}:\tilde{Ξ}\longmapsto -i\tilde{Ξ}\vec e_z . $$ The time generator acts from the right alone, because its left multiplier is the identity of the algebra; the three spatial ones act from both sides; the chirality acts from the right again. Each is $\pm i$ times one left multiplier and at most one right multiplier, all of them units, so its square follows from $\vec e_\eta^2=-e_t$ and $i^2=-1$ in two steps; these are derived values, computed here from the article's own actions, and not a rule taken from elsewhere. They are $C_t^2=+1$, $C_x^2=C_y^2=C_z^2=-1$, ${\mathsf{V}_{\mathrm{vol}}}^{2}=+1$, two generators on distinct indices anticommute, and $\mathsf{V}_{\mathrm{vol}}$ anticommutes with all four. In one line, with $\varepsilon=(1,-1,-1,-1)$, $\{C_\mu,C_\nu\}=2\varepsilon_\mu\delta_{\mu\nu}$. The bivector slot the source writes $q_{\mu\nu}=\tfrac12[q_\mu,q_\nu]$ is the commutator of two of the actions, $$ C_{\mu\nu}=\tfrac12\bigl[C_\mu,C_\nu\bigr]=\tfrac12\bigl(C_\mu C_\nu-C_\nu C_\mu\bigr), $$ which for two distinct indices is the product $C_\mu C_\nu$ itself, and which for two spatial indices is left multiplication by the bivector $\tfrac12[q_\mu,q_\nu]$. For one time and one spatial index neither description holds: the commutator is the two-sided action $\tilde{Ξ}\mapsto\vec e_\eta\tilde{Ξ}\vec e_z$, and the three bivectors with a time index are the only ones of the six that carry a right multiplier.

The source's Dirac adjoint is written as a pairing of two states, $$ \bigl\langle\tilde{Ξ},Ξ\bigr\rangle:=\tilde{Ξ}^{\dagger}i\,Ξ\,\vec e_x , $$ the dagger and the fixed unit $i\vec e_x$ closing the second state on the right. It is a pairing and not an action of the algebra: it is not an element and it multiplies no field from either side. No other right multiplier in this article stands outside an action: $C_t$, $C_\eta$ and $\mathsf{V}_{\mathrm{vol}}$ each carry one as part of a declared action, and this one carries one inside a bilinear. That is why it is declared here, so that it is not read as an action.

Two rules are used silently throughout. The first is $i^2=-1$: the unit belongs to the action, so the product $iC_t$ is the action $\tilde{Ξ}\mapsto-\tilde{Ξ}\vec e_x$, carrying no second unit, and a product of two of the actions brings one sign flip and no second unit either. The second is that $\mathrm{ReTr}$ keeps the central part of its argument. Only the scalar direction has a trace among the sixteen — every non-scalar direction of the algebra is traceless — so the trace of an element is its scalar coefficient times the dimension, and a display in $\mathrm{ReTr}$ is a display in the scalar part up to that factor. Two traces are in play, and the factor differs. The paper's $\mathrm{Tr}_{\mathbb H}$ is the two-dimensional one, with $\mathrm{Tr}_{\mathbb H}(q_\mu q_\nu)=2\delta_{\mu\nu}$ and hence factor $2$, and it is the trace the Fierz derivation cites. The trace of Eqs. (4)–(6) is the four-dimensional one, with factor $4$. The factor is carried wherever a numerical coefficient is at stake and dropped wherever only the vanishing of a display is.

The bilinear rule. It follows from the second rule and is the workhorse of every support count below. For an element $a$ with real coefficients, the pairing sees only the coefficient of $e_t$, $$ \mathrm{ReTr}\bigl\langle\tilde{Ξ},\,a\,\tilde{Ξ}\bigr\rangle=a_0\,\mathrm{ReTr}\bigl\langle\tilde{Ξ},\tilde{Ξ}\bigr\rangle , $$ because the vector part of $a$ contributes only to the imaginary part of $\langle\tilde{Ξ},a\tilde{Ξ}\rangle$. It is verified on random pairs to machine precision, and it is why so many of the source's bilinears are either a fixed multiple of $\langle\tilde{Ξ},\tilde{Ξ}\rangle$ or zero. Real coefficients matter: a biquaternion $a$ with complex ones does contribute a real part through its vector part, and none of the source's displays has that feature.

The Source's Aims

The paper's plan is stated in its Section I: Dirac fields are taken as bi-quaternionic fields, spacetime geometry is taken as emergent from them, and both are to follow from a single variational principle. Its announced results are a composite tetrad and contorsion built from bilinears in $\tilde{Ξ}$, a torsion that is the "spin-squared of quaternionic commutators", a constrained action with Lagrange multipliers, an intrinsic time from the phase of $\tilde{Ξ}$, torsion-wave solutions carrying an entropy current, and three experimental predictions.

The corpus is not the intended audience, and the source does not engage with it. The overlap is structural: the source's "$q_\mu$" are the corpus's algebra units, its bi-quaternionic field is the corpus's biquaternion, and its programme is the corpus's open question — whether the algebra or the matter field can generate geometry — answered in the affirmative. Whether the answer survives recomputation is what this article establishes, and in the places that matter most it does not.

The Source's Algebra, and Four Displays Under Test

The source builds a four-generator representation out of quaternions in its Section 2. Its Eq. (1) prints four generators as $2\times2$ blocks over the quaternion units, $q_0=1$ and $q_\eta$ the three imaginary ones: the time generator carries $1$ on the two off-diagonal blocks, and each spatial generator carries $q_\eta$ on one off-diagonal block and $-q_\eta$ on the other, which the source writes with the letter $\gamma$ and a frame bracket of its own. They are written $G_\mu$ here because the source's arrays are not the corpus's frame actions. Read as actions on the field, each is one left multiplication and at most one right multiplication, $$ G_0:\tilde{Ξ}\mapsto i\vec e_x\tilde{Ξ},\qquad G_x:\tilde{Ξ}\mapsto-\vec e_y\tilde{Ξ}\vec e_x,\qquad G_y:\tilde{Ξ}\mapsto \vec e_y\tilde{Ξ}\vec e_y,\qquad G_z:\tilde{Ξ}\mapsto-\vec e_y\tilde{Ξ}\vec e_z , $$ each verified as an operator identity at residual zero. The three spatial ones share their left factor up to the joint sign the decomposition leaves free, and none of the four is $C_\mu$: the corpus's spatial actions put $\vec e_\eta$ on the left and a fixed $\vec e_y$ on the right, and its time action is a right multiplication alone where the source's is a left one. That is also where the printed square comes from: the block of $G_\eta$ carries $q_\eta$ twice, once with each sign, so its square carries $(-q_\eta)(q_\eta)=-q_\eta^2$ on each of the two halves, and no step of this kind carries an invertible element $i$. Two factors that each square to $-1$ then square to $+1$, and the actions confirm it — $G_x^2$, $G_y^2$ and $G_z^2$ each return the field unchanged, at residual zero. Section 2.2 asserts the anticommutation relations $\{G_\mu,G_\nu\}=2\varepsilon_\mu\delta_{\mu\nu}$ from them, with inertia $(+,-,-,-)$. Four displays in that section and the next are tested here; three of them fail outright, and the fourth holds and is empty.

The block map is off by the factor $i$. The declared units have $q_\eta^2=-1$, so the printed square $G_\eta^2=-q_\eta^2$ is $+1$ on every generator, and $\{G_\eta,G_\eta\}=+2$ where the signs demand $-2$. The substitution that repairs it is $q_\eta\mapsto iq_\eta$, which gives $q_\eta^2=+1$ and hence $G_\eta^2=-1$; it is the substitution the paper never makes. The block map is therefore not a map of the required square for the $q_\eta$ the same section defines, and the consistent choice for that $q_\eta$ is the sign the source does not print.

The trace normalisation contradicts the embedding. The source fixes the normalisation by $\mathrm{Tr}_{\mathbb H}(q_\mu q_\nu)=2\delta_{\mu\nu}$, twice over — Section 2.3 prints the four entries one by one and Appendix A.5 prints them in one line — and Appendix A.5 states the embedding explicitly, sending $q_0$ to the identity and $q_\eta$ to $-i$ times the corresponding Pauli matrix. With that embedding the computed values are $+2$ for the time–time entry and $-2$ for the three spatial ones, so $$ \mathrm{Tr}_{\mathbb H}(q_\mu q_\nu)=2\varepsilon_\mu\delta_{\mu\nu},\qquad \varepsilon=(1,-1,-1,-1), $$ and not $2\delta_{\mu\nu}$. The printed $2\delta_{\mu\nu}$ is the normalisation of the embedding with $+i$ in place of $-i$, the same substitution that repairs the block map: one change of convention repairs both displays, which is how the two defects are known to be one. The price is that $q_\eta^2=+1$, which Appendix A.1 denies with $i^2=j^2=k^2=-1$; the three statements — A.1's units, the block map and the trace rule — cannot all hold, and the paper prints all three. As printed, the trace normalisation and the block map cannot be used together, and this is the seed of the ambiguity in the Fierz identity below, whose coefficient is fixed only once the reading of $\mathrm{Re}$ is.

The completeness relation is over a four-element basis, not sixteen. The source's Eq. (2) expands $$ \sum_A \check q_A\otimes q_A=1\otimes1+\tfrac12\sum_\mu q_\mu\otimes q_\mu+\tfrac{1}{2!}\sum_{\mu<\nu}q_{\mu\nu}\otimes q_{\mu\nu}+\cdots , $$ with $\check q_A$ the dual basis, and the comment that the coefficient $\tfrac12$ ensures $\mathrm{Tr}_{\mathbb H}(q_\mu q_\nu)=2\delta_{\mu\nu}$. The listed elements, however, are the four quaternion units and their antisymmetrized products $q_{\mu\nu}=\tfrac12[q_\mu,q_\nu]$, and the products add nothing: for the invertibles $[q_0,q_i]=0$ for all three spatial $i$, and $[q_i,q_j]=2\epsilon_{ij\eta}q_\eta$ is parallel to an invertible element. So $q_{0i}=0$ and each $q_{ij}$ is a sign times one of the invertibles, and the list spans four independent elements of $\mathbb{H}_{\mathbb{C}}$, not the sixteen graded products of Frame and Metric Conventions, $$ \bigl\{e_t,\ \mathsf{V}_{\mathrm{vol}},\ C_\mu,\ \mathsf{V}_{\mathrm{vol}} C_\mu,\ C_{\mu\nu}\bigr\},\qquad \mu<\nu , $$ that its own Appendix B.1 uses, with $1+4+6+4+1=16$ elements. The ellipsis in Eq. (2) is therefore doing the work the four-element list cannot: it has to supply the twelve graded products that the listed terms never reach.

The Fierz identity's coefficient depends on a notation the display does not fix, and the identity is empty either way. The source's Eq. (3) reads $$ \bigl\langle\tilde{Ξ},Ξ\bigr\rangle\bigl\langleΞ,\tilde{Ξ}\bigr\rangle=\tfrac14\sum_A \mathrm{Re}\bigl\langle\tilde{Ξ},\check q_A\tilde{Ξ}\bigr\rangle\ \mathrm{Re}\bigl\langleΞ,q_AΞ\bigr\rangle ,\qquad q_A \in \{1, q_\mu, q_{\mu\nu}, \dots\} , $$ "where the coefficient $\tfrac14$ derives from $\mathrm{Tr}_{\mathbb H}(q^Aq_A)=2$". The display is ambiguous, because $\mathrm{Re}$ alone is not the $\mathrm{ReTr}$ of Eqs. (4)–(6), and the coefficient depends on which is meant. Read literally — $\mathrm{Re}$ the real part of a quaternion, no trace — the left side is the real number $\lvert\langle\tilde{Ξ},Ξ\rangle\rvert^2=\lvert\tilde{Ξ}\rvert^2\lvertΞ\rvert^2$ and the coefficient is one: $$ \boxed{\;\bigl\langle\tilde{Ξ},Ξ\bigr\rangle\bigl\langleΞ,\tilde{Ξ}\bigr\rangle=\sum_{A\in\{1,q_1,q_2,q_3\}} \mathrm{Re}\bigl\langle\tilde{Ξ},q_A\tilde{Ξ}\bigr\rangle\ \mathrm{Re}\bigl\langleΞ,q_AΞ\bigr\rangle\;} $$ Read in the two-dimensional quaternion trace the derivation sentence itself cites, $\mathrm{Tr}_{\mathbb H}(q_Aq_A)=2$, each bilinear is twice the quaternion real part, the sum is four times as large, and the coefficient is $\tfrac14$ — which is what the display prints. That is one of three normalisations available in the same paper and the only one for which the printed number is right: with the plain $\mathrm{Re}$ of the display the coefficient is $1$, and with the four-dimensional trace of Eqs. (4)–(6), which counts a central element four times rather than twice, it is $\tfrac1{16}$. The identity itself is exact in all three readings on random pairs, at residual zero; only the coefficient changes, and it changes with the square of the trace dimension. So the printed number is not an arithmetic slip: it is the right number for the trace its own derivation sentence cites and the wrong number for the trace the composites use, and the symbol $\mathrm{Re}$ is what leaves the reader unable to tell which of the two is meant. The stated derivation then fits only in the form $\tfrac{1}{2\mathrm{Tr}_{\mathbb H}(q_Aq_A)}=\tfrac14$; taken as $1/\mathrm{Tr}_{\mathbb H}(q_Aq_A)=\tfrac12$ it gives the wrong coefficient for every channel, and gives the same one for all of them, which is why it cannot be checked by inspection.

What is wrong with the identity is that it is empty. In either reading the three non-scalar terms vanish identically, because $$ \mathrm{Re}\bigl\langle\tilde{Ξ},\,q_\eta\tilde{Ξ}\bigr\rangle=0 $$ for the three imaginary units and every field, real or complex — the scalar part is $i$ times a real number. Both sides therefore reduce to multiplicativity of the quaternion norm, $\langle\tilde{Ξ},Ξ\rangle\langleΞ,\tilde{Ξ}\rangle=\lvert\tilde{Ξ}\rvert^2\lvertΞ\rvert^2$. An identity that holds because three of its four terms are zero carries no information about the field, and the claim it is offered to support — that it "enables closed-form expression of all geometric quantities as bilinears in $\tilde{Ξ}$" — is not supported by it. A Fierz rearrangement earns that description when it trades a product of bilinears for a sum, not when everything but the scalar channel has already vanished.

One restriction is independent of the normalisation. Over complex coefficients the left side is the complex number $\lvert\langle\tilde{Ξ},Ξ\rangle\rvert^2=\lvert\tilde{Ξ}\rvert^2\lvertΞ\rvert^2$, whose real and imaginary parts are both nonzero for a generic pair — the quaternionic norm of a biquaternion is not real — while the right side is real. No value of the coefficient restores the identity there: it is an identity of the real quaternion algebra, and the source's $\tilde{Ξ}\in\mathbb{H}_{\mathbb{C}}$ is outside its range.

The Composite Geometry and the Constrained Action

The heart of the construction is the source's Section 4. With $\kappa\equiv8\pi G$ the source defines three bilinear composites, $$ e^a{}_\mu=\kappa^{-1}\,\mathrm{ReTr}\bigl\langle\tilde{Ξ},\,C_a\,\partial \tilde{Ξ}/\partial x_\mu\bigr\rangle,\qquad T_\rho{}_{\mu\nu}=\kappa\,\mathrm{ReTr}\bigl\langle\tilde{Ξ},\,\tfrac12[q_\mu,q_\nu]\,q_\rho\,\tilde{Ξ}\bigr\rangle,\qquad N_{\mathrm{cont},\mu ab}=\kappa\,\mathrm{ReTr}\bigl\langle\tilde{Ξ},\,C_{ab}\,\partial \tilde{Ξ}/\partial x_\mu\bigr\rangle, $$ and identifies them with the tetrad, the torsion tensor and the contorsion. The frame action $C_a$ carries the tetrad's frame slot, and $C_{ab}=\tfrac12[C_a,C_b]$ is the bivector slot, so the contorsion needs no new letter. In Eq. (5) the multipliers are the invertibles and not the actions: $q_\mu$ acts by left multiplication, $\tilde{Ξ}\mapsto e_\mu\tilde{Ξ}$, and the product $\tfrac12[q_\mu,q_\nu]q_\rho$ is an element of the algebra. The whole element then multiplies $\tilde{Ξ}$ from the left, and within it the rightmost factor is the one nearest the field, which is why the free index is the last one written in $[q_\lambda,q_\mu]q_\nu$. The letters $q$ name elements here and never a side: only the position of the element relative to $\tilde{Ξ}$ does. Throughout, the tetrad of Eq. (4) is written $\exp(a){}_\mu$ for the source's $e^a{}_\mu$, the frame index in parentheses so that it does not collide with the basis unit $e_\mu$, and $e$ is $\det\exp(a){}_\mu$.

The source then matches the composites to the standard Einstein–Cartan–Sciama–Kibble objects by $T_\rho{}_{\mu\nu}=\kappa\,\Sigma_{\rho\mu\nu}$, with $\Sigma_{\rho\mu\nu}$ the totally antisymmetric spin density of three frame actions, $$ \Sigma_{\rho\mu\nu}=\bigl\langle\tilde{Ξ},\operatorname{Alt}\bigl[C_\rho C_\mu C_\nu\bigr]\tilde{Ξ}\bigr\rangle , $$ where $\operatorname{Alt}$ is the normalised antisymmetrisation, the average over the six permutations weighted by their signs. It states that the contorsion–torsion relation is "automatically satisfied via Fierz identities".

The consistency checks are tautologies or false. The three checks offered for Eqs. (4)–(6) do not test the construction. Check 1, that $T_\rho{}_{\mu\nu}$ is antisymmetric in $\mu,\nu$, follows from the antisymmetry of the commutator and uses neither the field nor the trace. Check 2 displays $$ \mathrm{ReTr}\bigl\langle\tilde{Ξ},\,\tfrac12(q_\mu q_\nu+q_\nu q_\mu)\,q_\rho\,\tilde{Ξ}\bigr\rangle=0 \quad\Longrightarrow\quad T_{\rho\,(\mu\nu)}=0 , $$ and the left side is not zero. The halved anticommutator is $\varepsilon_\mu$ at $\mu=\nu$; the spatial unit itself when one index is the time index; and zero only when the two indices are distinct and both spatial. So the display vanishes except where the product $\tfrac12(q_\mu q_\nu+q_\nu q_\mu)q_\rho$ collapses to a scalar. On those ten index triples it is $\pm\mathrm{ReTr}\langle\tilde{Ξ},\,\tilde{Ξ}\rangle$ exactly, with the coefficient $+1$ at $(\rho,\mu,\nu)=(0,0,0)$ and $-1$ on the other nine. A scalar density of the field is not a torsion at any index; and $T_{\rho(\mu\nu)}$ is not defined in any case, since $T$ is built from the commutator alone. Check 3 is a single component with a proportionality sign. That component is $T^3{}_{12}$, one of the three that Eq. (5) leaves nonzero, and the source reads it as a confirmation: it is a confirmation of Eq. (5), and it is also the first sight of the three-component support established below. A fourth check, "guaranteed by Section 2", is not a check of Eqs. (4)–(6) at all; it defers to the block map, and the block map is off by the factor $i$.

The vanishing of the three imaginary bilinears, established above and used repeatedly below, is exact: with $\tilde{Ξ}=u+iv$ in real quaternions $u,v$ the scalar part of $\langle\tilde{Ξ},q_\eta\tilde{Ξ}\rangle$ is $i\,[\mathrm{Sc}\langle u,q_kv\rangle-\mathrm{Sc}\langle v,q_ku\rangle]$, a pure imaginary number whose real part is zero. The spatial components of the source's "vector current" are therefore zero for every field, and only the time component, $\mathrm{ReTr}\langle\tilde{Ξ},\,\tilde{Ξ}\rangle$, survives.

The printed contorsion–torsion relation carries the last index in the wrong order. The source prints $N_{\mathrm{cont},ab\mu}=\tfrac12(T_{\mu ab}-T_{ab\mu}+T_{ba\mu})$, with the frame pair first. The relation that is exact for any torsion, and antisymmetric in its first and third slots, is $$ N_{\mathrm{cont},\mu ab}=\tfrac12\bigl(T_{\mu ab}-T_{ab\mu}+T_{b\mu a}\bigr), $$ whose third term carries the order $b\mu a$. On a totally antisymmetric torsion the three cyclic orders coincide, $T_{\mu ab}=T_{ab\mu}=T_{b\mu a}$, so the standard bracket is $T_{\mu ab}-T_{\mu ab}+T_{\mu ab}=T_{\mu ab}$ and the contorsion is $\tfrac12T_{\mu ab}$; the printed third term $T_{ba\mu}$ is the odd permutation and equals $-T_{\mu ab}$, so the printed bracket is $-T_{\mu ab}$ and the printed contorsion is $-\tfrac12T_{\mu ab}$. On $T_{012}=T_{120}=T_{201}=1$ the printed form gives $-\tfrac12$ where the standard one gives $+\tfrac12$: the coefficient $\tfrac12$ is right and the sign is reversed.

That is a sign convention rather than an arithmetic slip, and it is the more awkward defect for being one. A contorsion defined by $\Gamma=\{\,\}-N_{\mathrm{cont}}$ instead of $\Gamma=\{\,\}+K$ makes the printed form correct, and the paper never states which convention it uses; its two displays, $N_{\mathrm{cont},ab\mu}$ in Section 5.1 and $N_{\mathrm{cont},\mu ab}$ in Section 5.7, agree with each other to $-\tfrac12T$ and never fix it. Since the same $N_{\mathrm{cont}}$ is then built as the composite $\kappa\,\mathrm{ReTr}\langle\tilde{Ξ},\,C_{ab}\partial \tilde{Ξ}/\partial x_\mu\rangle$ and fed into the torsion–spin relation of Appendix B, the sign of the matter coupling is left undetermined by the text.

The composite tetrad vanishes on the free Dirac solutions. This is the decisive failure of the section. For a plane wave $\tilde{Ξ}=\tilde{Ξ}\exp(ik\cdot x)$ with constant $\tilde{Ξ}$, the derivative gives $\partial \tilde{Ξ}/\partial x_\mu=ik_{\Phi,\mu}\tilde{Ξ}$, so, with $n$ the dimension of the trace, $$ e^a{}_\mu=\kappa^{-1}\,\mathrm{ReTr}\bigl\langle\tilde{Ξ},\,C_a\,ik_{\Phi,\mu}\tilde{Ξ}\bigr\rangle =n\,\kappa^{-1}k_{\Phi,\mu}\,\mathrm{Re}\bigl(i\,\langle\tilde{Ξ},C_a\tilde{Ξ}\rangle\bigr)=0 , $$ and the vanishing is independent of whether $n$ is $2$ or $4$. The bilinear $\langle\tilde{Ξ},C_a\tilde{Ξ}\rangle=\tilde{Ξ}^{\dagger} C_tC_a\tilde{Ξ}$ is real for any field, because $C_tC_a$ is Hermitian for all four actions — and the pairing is exactly $C_t$ in the adjoint, $\langle\tilde{Ξ},Ξ\rangle=\tilde{Ξ}^{\dagger} C_tΞ$ — so $\langle\tilde{Ξ},C_a\tilde{Ξ}\rangle$ times the unit $i$ is pure imaginary and its real part vanishes. The composite tetrad is therefore zero for every plane wave, verified to $4\times10^{-16}$ on random fields, and the induced metric $g_{\mu\nu}=\delta_{ab}\exp(a){}_\mu\exp(b){}_\nu$ is the zero metric. The construction does not merely fail to be a tetrad — it assigns no geometry at all to the free Dirac field, which is the only field the paper's action contains before the constraints are imposed. No check in the paper detects this, and Section 4.2's "Example verification" does not evaluate Eq. (4).

Three of the twenty-four torsion components survive. The object of Eq. (5) is not a torsion tensor with a minority of its components missing; it is a three-component object. Two facts account for all of it, and the bilinear rule turns both into one line. Every component with a $0$ in the antisymmetric pair vanishes, because $\tfrac12[q_0,q_\eta]=0$. For the six spatial pairs $q_iq_j=\epsilon_{ij\eta}q_\eta$, so the element whose $e_t$-coefficient the bilinear reads is $\epsilon_{ij\eta}q_kq_\rho$. That product is $\pm e_t$ only where $\rho$ is the unit $q_\eta$ of the pair — that is, on the three cyclic triples — and there $q_\eta^2=-1$, so the coefficient is $-1$. The three reversed triples are their antisymmetric partners and carry no fourth component. Hence $$ T_{3\,12}=T_{1\,23}=T_{2\,31}=-\kappa\,\mathrm{ReTr}\bigl\langle\tilde{Ξ},\,\tilde{Ξ}\bigr\rangle, \qquad\text{all other independent components zero}, $$ verified exactly on random fields: three of the twenty-four independent components are nonzero, and the only other entries are their antisymmetric partners $T_{\rho\,\nu\mu}=-T_{\rho\,\mu\nu}$.

The output index moves between displays. Eq. (5) prints the output slot of $T$ first and in the raised position of the source's convention, with the unit multiplying in its raised form, and the display above is the same object with both low. Two signs meet there and cancel: writing the tensor's output index low costs one $\varepsilon_\rho$, and writing the unit low costs another, because in the source's convention each raised slot carries a factor $\varepsilon_\rho$. So $\varepsilon_\rho^2=+1$ and the all-low form is the same tensor with the same sign, not a convention change. Section 11 step 2 prints the same object with the output slot at the other end, $T_{\lambda\mu\nu}=\kappa^{-1}\mathrm{ReTr}\langle\tilde{Ξ},\,[q_\lambda,q_\mu]\,q_\nu\tilde{Ξ}\rangle$, where the multiplier is $q_\nu$ and the free slot is therefore the last one — and the same step prints the Fierz relation it feeds from as $\langle\tilde{Ξ},C_\lambda C_{\mu\nu}\tilde{Ξ}\rangle=2\kappa T_{\lambda\mu\nu}$, with the free slot first. The two displays of one paragraph therefore disagree with each other, and not only with Eq. (5). For an object antisymmetric in one pair only, which slot carries the output is not bookkeeping: the entropy current contracts all three indices against $\epsilon_{\mu\nu\rho\sigma}$, and the contraction is blind to the displacement only if the tensor is totally antisymmetric, which Eq. (5) is not. The paper's own Section 4.2 example settles which placement it intends and then mis-signs it: with the source's raised unit, $q_3$ taken with a minus sign, $\tfrac12[q_1,q_2]\,(-q_3)=q_3(-q_3)=+1$ and $T_{312}=+\kappa\mathrm{ReTr}\langle\tilde{Ξ},\,\tilde{Ξ}\rangle$, while the example prints $-\mathrm{ReTr}\langle\tilde{Ξ},\,\tilde{Ξ}\rangle$ — the value obtained with the unit low, $q_3$, multiplying a commutator whose own indices the source raises.

The object the same section matches it to, $\Sigma_{\rho\mu\nu}$ of the display above, is totally antisymmetric and carries four independent components, nonzero on the time–space triples where Eq. (5) is zero. The two therefore have disjoint support, and it is the matching that carries the claim that the source reproduces Einstein–Cartan.

The couplings do not close dimensionally. The source gives the tetrad a factor $\kappa^{-1}$ and the contorsion a factor $\kappa$ while setting both couplings to the same $\kappa=8\pi G$; its Section 6 "Coupling Constant Verification" prints $\kappa_e=\kappa_K=8\pi G=8\pi G$, a chain whose last two entries are the same number. The two bullets that follow restate which coupling multiplies which field rather than checking either. With the canonical dimension $[\tilde{Ξ}]=M^{3/2}$ and $[G]=M^{-2}$, the tetrad composite has dimension $$ [\kappa^{-1}\langle\tilde{Ξ},\partial_\mu\tilde{Ξ}\rangle]=M^{+2}\,M^{3/2}\,M\,M^{3/2}=M^{6}, $$ where a tetrad must be dimensionless, and the contorsion composite has dimension $$ [\kappa\,\langle\tilde{Ξ},\partial_\mu\tilde{Ξ}\rangle]=M^{-2}\,M^{3/2}\,M\,M^{3/2}=M^{2}, $$ where a contorsion must have dimension $M$, the dimension of an inverse length. No single dimension for $\tilde{Ξ}$ repairs this: a dimensionless tetrad needs $[\tilde{Ξ}]=M^{-3/2}$ and a contorsion of dimension $M$ needs $[\tilde{Ξ}]=M^{+1}$, and the paper assigns one $\tilde{Ξ}$. The torsion fares the same way against itself. Eq. (5) defines it with $\kappa$ in the numerator. Section 11 step 2 defines the same quantity from the same bilinear with $\kappa^{-1}$ in the denominator and without the factor $\tfrac12$, so the paper's own two definitions differ by $2\kappa^{-2}=2/(8\pi G)^2$. Appendix B.2 defines it a third time from the axial bilinear $\langle\tilde{Ξ},\mathsf{V}_{\mathrm{vol}} C_\rho\tilde{Ξ}\rangle$, again with $\kappa^{-1}$.

The constrained action does not constrain. The master action of Section 5 is $$ S=\int d^4r\, e\Bigl[\frac{1}{2\kappa}\exp(a){}_\mu \exp(b){}_\nu \mathrm{Riem}_{\mu\nu ab}(e,K) +\bigl\langle\tilde{Ξ},\,iC^a\sum_\mu e_{a,\mu}D_\mu(e,K)\,\tilde{Ξ}\bigr\rangle-m\,\bigl\langle\tilde{Ξ},\tilde{Ξ}\bigr\rangle+\mathcal{L}_{\text{constr}}\Bigr], $$ with $C^a=g^{ab}C_b$ raised through the metric and the curvature pair written world indices first, in the corpus's order, and it adds the constraint Lagrangian $$ \mathcal{L}_{\text{constr}} = \lambda_{\mu a}\bigl(\exp(a){}_\mu - \exp(a){}_\mu[\tilde{Ξ}]\bigr) + \Lambda_{\mu ab}\bigl(N_{\mathrm{cont},\mu ab} - N_{\mathrm{cont},\mu ab}[\tilde{Ξ}]\bigr) , $$ with the composite expressions of Section 4 on the right, and with the coefficient $c_A$ renamed from the source's $C_A$ so that it does not collide with the frame actions $C_\mu$. One remark on the printed form belongs before the variation. The index on the action is raised by the metric, $iC^a=i g^{ab}C_b$, so the kinetic term carries a minus sign on the three spatial derivatives and none on the time derivative.

Variation with respect to the multipliers returns $e=e[\tilde{Ξ}]$ and $M_{\mathrm{map}}=M_{\mathrm{map}}[\tilde{Ξ}]$, which the source describes as enforcing "our composite ansätze once and for all". But those equations are the definitions of $e[\tilde{Ξ}]$ and $K[\tilde{Ξ}]$; a "constraint" that repeats a definition removes no degree of freedom, and substituting it back into the action replaces the tetrad by a first derivative of the field, making the gravitational term higher order in the field.

The Dirac equation is not derived from the action either. Section 5.5 varies with respect to the adjoint, $\tilde{Ξ}^{\dagger}$, writes the constraint's contribution as a sum over the sixteen graded products, $\Gamma_A$ built from the $G_\mu$ and read as arrays, $$ Ζ[\lambda,\Lambda](\tilde{Ξ})=\sum_A c_A\,\mathrm{Tr}_{\mathbb H}\bigl\langle\tilde{Ξ},\,\Gamma_A\partial \tilde{Ξ}/\partial x_\mu\bigr\rangle\,\Gamma_A\partial \tilde{Ξ}/\partial x_\mu , $$ and declares it zero "by orthogonality". Orthogonality does not deliver that: a sum of orthogonal basis elements vanishes only when every coefficient vanishes, and each coefficient vanishing is the statement to be proved. The basis invoked is the sixteen-element one, whose four-element counterpart failed at Eq. (2). Since the Dirac kinetic term is already written into the action by hand, the equation that comes out is the equation that went in; the step's only real task is to show that the constraint term does not spoil it, and that is what the non sequitur leaves undone.

The Two Dirac Equations

The kinetic term of that action is one Dirac operator, the corpus's Dirac equation is another, and the two are not the same operator written twice, so they are displayed separately. The source's operator, written as the action it is, is $$ \tilde{Ξ}\longmapsto iC^0\frac{\partial\tilde{Ξ}}{\partial h_t}-i\sum_\eta C_\eta\frac{\partial\tilde{Ξ}}{\partial x_\eta} =-\frac{\partial\tilde{Ξ}}{\partial h_t}\vec e_x+\sum_\eta\vec e_\eta\frac{\partial\tilde{Ξ}}{\partial x_\eta}\vec e_y , $$ where the source raises the index with its metric, so that the part along $\eta$ is $-C_\eta$ in the all-low reading, and the unit $i$ of the operator cancels the unit of the action once in each term. It is two-sided: the time derivative acts by right multiplication alone, because the left multiplier of $C_t$ is the identity of the algebra, the three spatial derivatives act by a left and a right multiplication, and the mass acts by left multiplication. Its square is minus the flat d'Alembertian of $g$, because the four actions anticommute on distinct indices and square to $\pm1$, $$ \bigl(iC^a\frac{\partial}{\partial x_a}\bigr)^2 =-\varepsilon_\mu\frac{\partial^2}{\partial x_\mu^2} =-\partial^2/\partial h_t^2+\Delta_\Phi , \qquad \Delta_\Phi=\sum_\eta\partial^2/\partial x_\eta^2 , $$ and on the solutions of the equation it is $m^2$, which is the Klein–Gordon shell the signs of $g$ require. The square is scalar, and it is scalar for the source's operator and not for the gradient: on a generic wavevector the source's square departs from a multiple of the identity by $9\times10^{-16}$, while $\tilde\nabla_\Phi^2$ departs from one by $0.9$. That $\tilde\nabla_\Phi^2$ is not $\Box_\Phi$ is why the corpus carries no one-field first-order massive equation and why its mass sits in the off-diagonal pair of the next display. The source's equation is $$ iC^a\frac{\partial\tilde{Ξ}}{\partial x_a}=k_m\tilde{Ξ}. $$

The corpus's equation is the one-sided pair $$ \tilde\nabla_\Phi\tilde{Ξ}_{R}=k_m\tilde{Ξ}_{L}, \qquad (\tilde\nabla_\Phi)^{\natural}\tilde{Ξ}_{L}=k_m\tilde{Ξ}_{R}, \qquad \tilde\nabla_\Phi=\sum_\mu e_\mu\,\partial/\partial R_\mu , $$ in which the gradient is an element acting by left multiplication on the algebra, and the mass is the off-diagonal pairing, $L$ with $R$. The gradient is primary there and the source's operator is a frame-dependent contraction, as The Dirac Equation in Biquaternionic Form states; it is not obtainable by rewriting the gradient.

The difference is one line, and it is a difference in side. The letter of an element is the value of the chirality action on it, written $L$ for the minus value and $R$ for the plus value, and two facts carry every display below: a product takes the letter of its right factor, and left multiplication keeps the letter. Right multiplication by $e_t$ and by $\vec e_z$ therefore keeps the letter, right multiplication by $\vec e_x$ and by $\vec e_y$ exchanges the two, and a left multiplier never changes it. Every term of the source's kinetic operator carries a right multiplier — $C_t$ by $\vec e_x$, $C_\eta$ by $\vec e_y$ — so the whole kinetic term exchanges $L$ and $R$, while the mass term $k_m\tilde{Ξ}$ is left multiplication by a constant and keeps the letter. In the corpus's pair the gradient $\tilde\nabla_\Phi$ is left multiplication and keeps the letter, while the mass is the off-diagonal pairing, which is the exchange: the gradient preserves and the mass is the flip. Neither display is a rewriting of the other: they differ in the operator, in the field it acts on and in the letter, and the two forms are kept side by side in the corpus's own Dirac article for exactly this reason.

The Einstein equation does not follow. Variation with respect to the tetrad gives, in the source's own notation, $$ G_{\mu a}=2\kappa\bigl(T_{\mu a}-\lambda_{\mu a}\bigr) , $$ a coefficient $2\kappa=16\pi G$ where the printed target $G_{\mu\nu}=8\pi G T_{\mu\nu}$ needs $\kappa=8\pi G$. The two agree only if $\lambda_{\mu a}=\tfrac12T_{\mu a}$, a specific value and not a vanishing one — or if the stress tensor is defined here with one factor of two fewer than the target uses. The source says that "when constraints hold, $\lambda_{\mu a}$ becomes auxiliary and we recover $G_{\mu\nu}=8\pi G T_{\mu\nu}$", but variation with respect to the multiplier returns the definition of the composite and fixes nothing about $\lambda$. Einstein's equation therefore does not follow from the action as printed.

The Teleparallel Sector

Section 9.1 adds a Lagrange multiplier $\mathrm{Sd}_{\mu\nu ab}$ enforcing $\mathrm{Riem}_{\mu\nu ab}(\omega)=0$, so that the geometry is teleparallel, and identifies the Weitzenböck gauge $\omega_{\mu ab}=0$ with the contorsion given by minus the Levi-Civita spin connection of the tetrad, $N_{\mathrm{cont},\mu ab}=-\omega_{\mu ab}(e)$. Both statements of geometry are correct and standard: the Weitzenböck connection $$ \Gamma_{\lambda\mu\nu}=\sum_a e_{a\lambda}\,\frac{\partial \exp(a){}_\nu}{\partial x_\mu} $$ is curvature-free by construction, and the teleparallel gauge is the one in which the whole connection is contorsion. Both were verified independently for this corpus in Torsion, Contorsion and the Spin Connection in Biquaternionic Form, where the vanishing of the curvature and the non-vanishing of the torsion were checked exactly.

One defect in that paragraph is checkable on the page and needs no computation. The display for the contorsion reads $N_{\mathrm{cont},\mu ab}=-\omega_{\mu ab}(e)=e_{a\nu}\frac{\partial e_{b]\nu}}{\partial r_{x[\mu}}-e_{b\nu}\frac{\partial e_{a]\nu}}{\partial r_{x[\mu}}$, and the line immediately below, which the same paragraph equates to it, reads $\kappa\,\mathrm{ReTr}\langle\tilde{Ξ},\,C_{ab}\,\partial \tilde{Ξ}/\partial x_\mu\rangle=-e_{a\nu}\frac{\partial e_{b]\nu}}{\partial r_{x[\mu}}-e_{b\nu}\frac{\partial e_{a]\nu}}{\partial r_{x[\mu}}$. The second term changes sign between two consecutive displays of the same object, so at most one of them is right.

What is not correct is what the source puts into that identity. The teleparallel gauge condition, read as an equation for the contorsion — hence, through Eq. (6), for the bilinear $$ N_{\mathrm{cont},\mu ab}=\kappa\,\mathrm{ReTr}\bigl\langle\tilde{Ξ},\,C_{ab}\,\partial \tilde{Ξ}/\partial x_\mu\bigr\rangle $$ — must equal $-\omega_{\mu ab}(e)$ of the composite tetrad, the Levi-Civita connection of $\exp(a){}_\nu[\tilde{Ξ}]$. That is an over-determined differential equation on $\tilde{Ξ}$, not a consequence of anything, and the composite tetrad it is asked to reproduce vanishes on the free solutions. The teleparallel section therefore rests on the same composite that fails above.

The display that states the constraint is also the clearest single measure of the dimensional failure, because it puts three different powers of mass in one equation. In the article's own letters it reads $$ \mathrm{Riem}_{\mu\nu ab}\bigl(\omega(e)\bigr) =-2\kappa\,D_{[\mu}\mathrm{ReTr}\bigl\langle\tilde{Ξ},\,C_{ab}\,\frac{\partial\tilde{Ξ}}{\partial r_{x\nu]}}\bigr\rangle -2\kappa^{2}\,\mathrm{ReTr}\bigl\langle\tilde{Ξ},\,C_{ac}\,\frac{\partial\tilde{Ξ}}{\partial r_{x[\mu}}\bigr\rangle\mathrm{ReTr}\bigl\langle\tilde{Ξ},\,C_{c|b]}\,\frac{\partial\tilde{Ξ}}{\partial r_{x\nu]}}\bigr\rangle , $$ the bivector $C_{ab}$ of the composites, and with the counting used throughout the article — $[\tilde{Ξ}]=M^{3/2}$, $[\partial]=M$, $[\kappa]=M^{-2}$ — the three members scale as $$ \bigl[\mathrm{Riem}\bigr]=M^{2},\qquad \bigl[2\kappa D\,\mathrm{ReTr}\langle\tilde{Ξ},\,\partial_\mu\tilde{Ξ}\rangle\bigr]=M^{-2}\,M\,M^{4}=M^{3},\qquad \bigl[2\kappa^{2}\bigl(\mathrm{ReTr}\langle\tilde{Ξ},\,\partial_\mu\tilde{Ξ}\rangle\bigr)^{2}\bigr]=M^{-4}\,M^{8}=M^{4} . $$ A single equation cannot hold at three different orders, so the constraint is dimensionally impossible before any of its content is examined. The reason is not a stray exponent. Both terms on the right are pieces of the contorsion, since $\kappa\,\mathrm{ReTr}\langle\tilde{Ξ},\,C_{ab}\partial \tilde{Ξ}/\partial x_\mu\rangle=N_{\mathrm{cont},\mu ab}$ makes the first $-2D_{[\mu}N_{\mathrm{cont},\nu]ab}$ and the second a product of two contorsions: they are the $DN_{\mathrm{cont}}$ and the $N_{\mathrm{cont}}^{2}$ of $\mathrm{Riem}(\omega(e)+K)$. For a contorsion of dimension $M$, which is the dimension a contorsion has, both terms would be $M^2$ and the display would close. The source's composite has dimension $M^2$ instead, and one false power in the composite becomes $M^3$ in $DN_{\mathrm{cont}}$ and $M^4$ in $N_{\mathrm{cont}}^{2}$ — the mismatch grows as the construction proceeds, which is why it cannot be repaired downstream.

Kozyrev's Causal Time

Section 9.2 embeds Kozyrev's "time current" as the quaternionic phase of the field, defining $\tau(x)=\mathrm{Arg}_{\mathbb{H}}\tilde{Ξ}(x)$ and giving its gradient as $$ \partial/\partial x_\mu\,\tau=\frac{\mathrm{Re}\bigl\langle\tilde{Ξ},\,q_5q_\mu\tilde{Ξ}\bigr\rangle}{\mathrm{Re}\bigl\langle\tilde{Ξ},\,\tilde{Ξ}\bigr\rangle} , \qquad q_5=q_1q_2q_3 , $$ with $q_5$ described as "the 'imaginary' unit in $\mathbb{H}$ that anticommutes with $q_\mu$". That description cannot be right, and the failure is complete rather than partial. In the quaternion units the product $q_5=q_1q_2q_3$ is the real scalar $$ q_5=q_1q_2q_3=-e_t , $$ verified directly: it is central, commuting with every $q_\mu$ and with everything else, and it is not invertible imaginary at all. There is no element of $\mathbb{H}$ that anticommutes with all four $q_\mu$, because every quaternion commutes with $q_0=1$. Substituting $q_5=-1$ into the displayed gradient gives $$ \partial/\partial x_\mu\,\tau=-\,\frac{\mathrm{Re}\bigl\langle\tilde{Ξ},\,q_\mu\tilde{Ξ}\bigr\rangle}{\mathrm{Re}\bigl\langle\tilde{Ξ},\,\tilde{Ξ}\bigr\rangle} , $$ whose time component is $$ \partial/\partial x_0\,\tau=-\,\frac{\mathrm{Re}\bigl\langle\tilde{Ξ},\,q_0\tilde{Ξ}\bigr\rangle}{\mathrm{Re}\bigl\langle\tilde{Ξ},\,\tilde{Ξ}\bigr\rangle}=-1 . $$ The three spatial components vanish identically, because $\mathrm{Re}\langle\tilde{Ξ},\,q_\eta\tilde{Ξ}\rangle=0$ for the imaginary units, so the time component is the whole of it and the gradient is the constant covector $$ \partial \tau/\partial x_\mu=(-1,0,0,0) $$ for every $\tilde{Ξ}$. It is the gradient of $-t$ and of nothing else, it carries no information about the field, and it cannot distinguish one field from another. Nor does any other reading of $q_5$ rescue the display: $q_5=q_1$ gives $(0,-1,0,0)$, $q_5=q_2$ gives $(0,0,-1,0)$, $q_5=q_3$ gives $(0,0,0,-1)$, and the reversed product $q_1q_3q_2=+1$ gives $(+1,0,0,0)$ — a constant covector in every case. The bilinear rule of the introduction is the reason, and it is exact: $qq_\mu$ is an element with real coefficients, so $\mathrm{ReTr}\langle\tilde{Ξ},\,q q_\mu\tilde{Ξ}\rangle$ equals the coefficient of $e_t$ in $qq_\mu$ times $\mathrm{ReTr}\langle\tilde{Ξ},\,\tilde{Ξ}\rangle$, and the ratio the display takes is that coefficient alone. The gradient is therefore the fixed covector of scalar parts of $qq_\mu$, for every $\tilde{Ξ}$, and no choice of $q_5$ can make it depend on the field. The type mismatch is visible before any evaluation: $\tau=\mathrm{Arg}_{\mathbb{H}}\tilde{Ξ}$ is invertible quaternion, so $\partial \tau/\partial x_\mu$ is quaternion-valued, whereas the printed formula returns the real number $\mathrm{Re}\langle\tilde{Ξ},\,q_5q_\mu\tilde{Ξ}\rangle/\mathrm{Re}\langle\tilde{Ξ},\,\tilde{Ξ}\rangle$; a four-component derivative cannot be given by one real ratio. The Madelung-type relation the source needs — a phase gradient given by an axial-vector-to-density ratio — is not what its formula produces, so the intrinsic time field $\tau=\mathrm{Arg}_{\mathbb{H}}\tilde{Ξ}$ is not recovered, and the claim that causal asymmetry is realised "intrinsically, without new fields" has no support.

The rest of that section is the retarded kernel and the bilocal action $$ G_{\mathrm{prop}}(x-y)=\ell_c^{-2}\,\mathrm{H}_{\mathrm{hvs}}(x_0-y_0)\,\mathcal{G}_m(x-y), \qquad \delta S_K=\tfrac12\int d^4r\,d^4y\;\frac{\partial\tau}{\partial x_\mu}(x)\,G_{\mathrm{prop}}(x-y)\,\frac{\partial\tau}{\partial x_\mu}(y), \qquad \frac{\partial}{\partial r_\mu}=\varepsilon_\mu\,\frac{\partial}{\partial x_\mu}\quad(\text{no sum}) . $$ The kernel is the standard retarded Green's function combination; the source itself notes that its Lorentz invariance holds "modulo frame choice for $\Theta$", which is the standard qualification, and since $\tau$ is not the field claimed, the construction adds nothing testable. The source's own "Caveat" names Kozyrev's claims as "controversial and largely unreplicated". The corpus's route to causal asymmetry, in Causality and the Light Cone as an Information Barrier in Biquaternionic Form and Thermal Time and the Modular Flow in the Biquaternion Framework, does not use Kozyrev and does not depend on this section.

The Torsion Bohm Potential

Section 10 proposes a torsion contribution to the quantum potential. It first re-reads the quaternion commutator as a torsion, $$ \tfrac12\bigl[q_\mu q_\nu-q_\nu q_\mu\bigr]=2\,T_{\mu\nu}{}_\rho\,q_\rho , $$ and then, from a Madelung decomposition, writes $$ Q_T=\beta\,T_{\mu\nu\rho}T_{\mu\nu\rho}, \qquad \beta=\frac{\hbar c\hbar/(mc)}{8} , $$ while the "Quantum Potential Comparison" box at the end of the same section writes the torsion term with a negative sign, $$ Q_T=-\frac{\hbar c\hbar/(mc)}{8}\,T_{\mu\nu\rho}T_{\mu\nu\rho} , $$ and adds it to the standard Bohm term. The paper therefore carries both signs within one section, with only the three-line "ultra-compact derivation" between them, and with no statement of which is the correction: Eq. (23) is positive, the box is negative, and the two later displays that repeat the term — the Pauli limit of Section 13.2 and the torsion potential of Eq. (26) — both carry the plus sign, so the box is the outlier by three displays to one. Nothing in the source fixes the sign, and a torsion-squared scalar whose coefficient is $+\frac{\hbar^{2}}{8\,m}$ in one display and $-\frac{\hbar^{2}}{8\,m}$ in the next is not a derived result.

The box carries a third line as well, $Q_H=Q_{\text{Bohm}}+Q_{\text{Torsion}}+\tfrac{c\hbar/(mc)}{2}\epsilon^{ij\eta}\frac{\partial T_{j\eta0}}{\partial x_i}$, labelled "quaternionic correction". That term is built on exactly the components $T_{j\eta0}$ that Eq. (5) makes zero for every field, so on the source's own torsion it vanishes identically and the box collapses to its first two lines. It is the same disjoint-support failure that separates Eq. (5) from the axial pseudo-vector $\mathcal{T}=\epsilon_{ij\eta}T_{j\eta0}$ used in Eq. (26).

Two of the three terms are not energies. Eq. (26), repeated in the comparison box with the same coefficients, reads $$ V_T(\mathbf r)=\frac{\hbar c\hbar/(mc)}{8}\sum_{\mu,\nu,\rho}T_{\mu\nu\rho}T_{\mu\nu\rho}+\frac{c\hbar/(mc)}{4}\,\sigma\cdot(\vec\nabla_\Phi\times T)+\frac{c\hbar/(mc)}{2}\,\vec\nabla_\Phi\cdot T , \qquad T=\epsilon_{ij\eta}T_{j\eta0} , $$ in which $\sigma$ is the source's Pauli vector and $\vec\nabla_\Phi$ its spatial gradient, neither of them the bivector $C_{\mu\nu}$ of the frame actions. The three terms do not share a dimension. With $[\frac{\hbar^{2}}{m}]=\mathrm{J\,m^2}$ and torsion in inverse length, the first term is $\mathrm{J\,m^2\cdot m^{-2}}=\mathrm{J}$; the second is $\mathrm{m^2\,s^{-1}\cdot m^{-2}}=\mathrm{s^{-1}}$, and the third is $\mathrm{s^{-1}}$ as well. The last two are frequencies where the first is an energy, and the whole display is short by one factor of $\hbar$ in those two terms. The defect is the sort that a units check catches and a symbol-by-symbol reading does not, and it is visible in the source's own numbers: at its stated $T\simeq3\times10^{-32}\,\mathrm{m^{-1}}$ the first term is $6\times10^{-108}\,\mathrm{J}$ while the other two are $\sim10^{-41}\,\mathrm{s^{-1}}$, and the ratio is a time.

Two further objections are independent of the sign and of the dimension. The equation $\tfrac12[q_\mu,q_\nu]=2T_{\mu\nu}{}_\rho q_\rho$ is the same conflation as elsewhere in the paper, and it is not even the paper's own normalisation of it: Appendix A.4 writes the same identity as $[q_\mu,q_\nu]=2T_{\mu\nu}{}_\rho q_\rho$, without the $\tfrac12$, so the two displays differ by a factor two. Both conflate a constant of the quaternion algebra, six fixed entries and no others, with a field, and neither introduces any field content. And the claim that $Q_T$ is "the unique scalar–torsion correction at $O(\hbar^{2})$" is asserted and not shown; uniqueness over scalar invariants built from torsion is a claim about a classification that the source does not perform. The scalar it squares is in any case built on the three surviving components of Eq. (5), so $Q_T$ as written is a function of one bilinear of $\tilde{Ξ}$ and not of a torsion field.

Torsion Waves and the Entropy Current

Section 11 derives a wave equation for the torsion, $\Box_\Phi T_{\lambda\mu\nu}=0$, by acting twice with the covariant derivative on the Dirac equation and imposing the teleparallel condition $\mathrm{Riem}_{\mu\nu ab}=0$. It then introduces the pseudo-vector $$ J_{S\mu}=\alpha\,\epsilon_{\mu\nu\rho\sigma}T_{\nu\rho\sigma}, \qquad \alpha=\frac{1}{3\cdot8\pi G} , $$ written here with every index low; raising the free index costs one sign, $\epsilon^{0123}=-\epsilon_{0123}$. The source states that "the wave equation for $T$ together with antisymmetry of $\epsilon_{\mu\nu\rho\sigma}$ implies $\partial/\partial x_\mu J_{S\mu}=0$", reading the conservation as "the field-theoretic avatar of the second law of thermodynamics".

The wave equation itself is not derived. The chain that produces $\Box_\Phi T=0$ runs: apply $\sum_\nu\varepsilon_\nu D_\nu D_\nu$ to the Dirac equation, obtain $$ (\Box_\Phi-(k_m)^{2})\tilde{Ξ}=-\tfrac14 \mathrm{Riem}_{\mu\nu\alpha\beta}\,C_{\mu\nu}\sigma_{\alpha\beta}\tilde{Ξ}-\tfrac12\sum_\rho\varepsilon_\rho T_{\rho\mu\nu}\,C_\mu D_\rho\tilde{Ξ} , $$ pair the result with $C_\lambda\tilde{Ξ}$, and invoke the Fierz relation $\langle\tilde{Ξ},C_\lambda C_{\mu\nu}\tilde{Ξ}\rangle=2\kappa T_{\lambda\mu\nu}$. Two things go wrong at once. The curvature term is set to zero by the teleparallel condition, which leaves the single term $-\tfrac12\sum_\rho\varepsilon_\rho T_{\rho\mu\nu}\langle\tilde{Ξ},C_\lambda C_\mu D_\rho\tilde{Ξ}\rangle$ — one power of $T$ multiplying a derivative bilinear, not a wave operator acting on $T$. And on the left there is $\langle\tilde{Ξ},C_\lambda(\Box_\Phi-(k_m)^{2})\tilde{Ξ}\rangle$, which is not $\Box_\Phi$ of any bilinear either; $\Box_\Phi T$ would be $2\sum_\alpha\varepsilon_\alpha\partial_\alpha\langle\tilde{Ξ},\partial_\alpha\tilde{Ξ}\rangle$ at leading order, the outer derivative acting on both slots, and what Eq. (24) supplies for $\sum_\mu\varepsilon_\mu\partial_\mu\partial_\mu\tilde{Ξ}$ is a term proportional to $T\,\partial_\mu\tilde{Ξ}$. Substituting gives $\Box_\Phi T$ proportional to $T\,\partial_\mu\tilde{Ξ}\,\tilde{Ξ}$ and to $(\partial_\mu\tilde{Ξ})^2$, neither of which vanishes. The source's footnote, "up to total derivatives", does not bridge the gap: a total derivative changes the action, not an order in $T$. So $\Box_\Phi T=0$ is asserted at the point where the derivation should be, and the three results built on it — light-speed torsion waves, the conserved entropy current, and the pulse-propagation estimate — inherit the assertion.

The implication does not hold, and the failure is an index count. The divergence is an epsilon contraction of the gradient, $$ \sum_\mu\varepsilon_\mu\partial_\mu J_{S\mu}=\alpha\sum_\mu\varepsilon_\mu\,\epsilon_{\mu\nu\rho\sigma}\,\partial_\mu T_{\nu\rho\sigma} , $$ and the epsilon needs four distinct indices, so the contraction keeps only the part of the gradient that is antisymmetric in all four — contracting the antisymmetrised gradient, which in four dimensions has the single independent component $\frac{\partial T_{123]}}{\partial x_{[0}}$. Hence the divergence is $4!\,\alpha\,\frac{\partial T_{123]}}{\partial x_{[0}}$, verified numerically on random fields, and it vanishes only when that component does — that is, only when the three-form part of $T$ is closed — while $\Box_\Phi T=0$ neither requires nor supplies closedness: the wave equation constrains the four components one at a time and says nothing about the antisymmetrised gradient. An explicit counterexample settles it: the field $T_{123}(x)=x_0$, all other components zero, satisfies $\Box_\Phi T_{\lambda\mu\nu}=0$ identically, while — taking $\epsilon_{0123}=+1$, so that $\partial^0=\partial/\partial h_t$ — $$ J_{S0}=\alpha\,\epsilon_{0123}\,T_{123}=\alpha\,x_0, \qquad \frac{\partial J_{S\mu}}{\partial x_\mu}=\alpha\,\epsilon_{0123}\,\frac{\partial T_{123}}{\partial x^0}=\alpha\ne0 , $$ verified directly. Conservation would hold if $T_{\nu\rho\sigma}$ were the dual of the gradient of a harmonic potential, $T_{\nu\rho\sigma}=\epsilon_{\nu\rho\sigma\lambda}\partial/\partial x_\lambda\vartheta_{\mathrm{fld}}$, where $\vartheta_{\mathrm{fld}}$ is a real scalar potential with $\Box_\Phi\vartheta_{\mathrm{fld}}=0$; the source neither assumes that form nor derives it. The claimed second-law avatar therefore rests on an identity that fails.

A separate objection, which would stand even if the wave equation had been derived: since the torsion is defined from the field by Eq. (5) and carries no independent dynamics, $\Box_\Phi T=0$ is not an independent wave equation but a constraint on $\tilde{Ξ}$; and Eq. (5)'s $T$ has three nonzero components, all of them the same bilinear $-\kappa\mathrm{ReTr}\langle\tilde{Ξ},\,\tilde{Ξ}\rangle$. The source's step 4 reads this as a virtue, calling the torsion waves "genuine emergent, massless degrees of freedom" that propagate "without enlarging the field content". The second half of that is right and is exactly the difficulty: nothing is added because nothing independent is there. A field whose every component is one and the same bilinear of $\tilde{Ξ}$ has the dynamics of $\tilde{Ξ}$ and none of its own, so "degrees of freedom" in the counting sense is the wrong description of it.

The Fierz Chain of Appendix B

The Appendix B chain is the source of the factor-2 relation used in Section 11, and it is where an exact error lives. The appendix needs the trace of the chirality action with two frame actions and a bivector, and prints it as $$ \mathrm{Tr}\bigl[\mathsf{V}_{\mathrm{vol}}\,C_\rho\,C_\lambda\,C_{\mu\nu}\bigr]=4\,\epsilon_{\rho\lambda\mu\nu} . $$ This $\mathrm{Tr}$ is the plain trace of the product of four actions and not its real part. The distinction is the whole content of the error. The trace is pure imaginary on every index quadruple, so $\mathrm{ReTr}$ of the same argument vanishes; the printed right-hand side is therefore a statement about the plain trace, and against the plain trace it is off by the unit $i$. The exact value is $$ \boxed{\;\mathrm{Tr}\bigl[\mathsf{V}_{\mathrm{vol}}\,C_\rho\,C_\lambda\,C_{\mu\nu}\bigr]=-\,4i\,\epsilon_{\rho\lambda\mu\nu}\;} $$ with the four indices all low and $\epsilon_{0123}=+1$, verified at residual exactly zero over all index quadruples: the trace takes the two values $\pm4i$ on the twenty-four quadruples with four distinct indices and vanishes on all others. Its $\epsilon_{\rho\lambda\mu\nu}$ also carries two raised indices, each of which costs a sign on the standing conventions, and the chain built on the display inherits both defects.

The second error is structural. The appendix evaluates $B_{\lambda\mu\nu}=\langle\tilde{Ξ},C_\lambda C_{\mu\nu}\tilde{Ξ}\rangle$, finds that "the only nonzero contribution comes from the axial-vector channel", and concludes that $B_{\lambda\mu\nu}$ is $\epsilon_\rho{}_{\lambda\mu\nu}$ times the axial bilinear $A_\rho=\langle\tilde{Ξ},\mathsf{V}_{\mathrm{vol}} C_\rho\tilde{Ξ}\rangle$. On the twenty-four triples with three distinct indices that conclusion is exact, $B_{\lambda\mu\nu}=-i\epsilon_{\lambda\mu\nu\rho}\varepsilon_\rho A_\rho$. For the six of them whose first index is the time index the arithmetic is even transparent, because $C_tC_\lambda=C_0^2=+1$ collapses the product to the bare bivector and a bivector bilinear is by construction the epsilon-dual of the axial one, $$ \bigl\langle\tilde{Ξ},C_{\mu\nu}\tilde{Ξ}\bigr\rangle=B_{0\mu\nu}=-i\,\epsilon_{0\mu\nu\rho}\,\varepsilon_\rho\,A_\rho , $$ so there the channel count and the conclusion agree.

The disagreement begins where an index repeats. On the twenty-four triples in which the first index coincides with exactly one of the other two, the epsilon vanishes and the bilinear does not: $B_{001}$ is nonzero and real for a generic field. The reason is that $C_0C_{01}=C_0^2C_1=C_1$: on a repeated first index the product collapses to a single frame action and not to the bivector, and $\langle\tilde{Ξ},C_\eta\tilde{Ξ}\rangle$ is real on every field tested, to machine precision. The bare bivector would not do the job: $\langle\tilde{Ξ},C_{\mu\nu}\tilde{Ξ}\rangle$ is pure imaginary on all six pairs, the pairing contributing one unit $i$ and the bivector two. Hermiticity is a statement about the source's own adjoint form and not about this pairing: $\tilde{Ξ}^{\dagger}C_{0\eta}\tilde{Ξ}$ is real because $C_{0\eta}$ is Hermitian, and $\tilde{Ξ}^{\dagger}C_{ij}\tilde{Ξ}$ is pure imaginary because $C_{ij}$ is anti-Hermitian, on all six pairs. So $B$ is not the epsilon channel, and the printed display cannot be an identity for the bilinear altogether: it holds on forty of the sixty-four triples, and the twenty-four that fail are exactly those where the first index coincides with one — not both — of the other two. What does hold, exactly and at every index triple, is the statement for the antisymmetrised projection, $$ \operatorname{Alt}\bigl\langle\tilde{Ξ},\,C_\lambda C_\mu C_\nu\,\tilde{Ξ}\bigr\rangle =-\,i\,\epsilon_{\lambda\mu\nu\rho}\,\varepsilon_\rho\,A_\rho , \qquad A_\rho=\bigl\langle\tilde{Ξ},\mathsf{V}_{\mathrm{vol}}\,C_\rho\,\tilde{Ξ}\bigr\rangle , $$ with $\varepsilon_t=+1$ and $\varepsilon_\eta=-1$ the sign vector, all indices low; the operator form of the same identity, $\operatorname{Alt}[C_\lambda C_\mu C_\nu]=-i\epsilon_{\lambda\mu\nu\rho}\varepsilon_\rho\mathsf{V}_{\mathrm{vol}} C_\rho$, was verified at all sixty-four index triples. The weight is what carries the sign: without $\varepsilon_\rho$ the same display with a bare $+i$ holds on the eighteen triples for which the remaining index $\rho$ is spacelike and fails on the six for which it is the time index, which is the form a reader is most likely to copy wrongly.

The source's factor-2 relation is therefore obtained only after antisymmetrising, and the antisymmetrisation is what the derivation omits. For three distinct indices, where $C_{\mu\nu}=C_\mu C_\nu$, the left side is the corpus's spin density $\Sigma_{\lambda\mu\nu}$, and the corpus's torsion–spin relation is $T_{\lambda\mu\nu}=\kappa\,\Sigma_{\lambda\mu\nu}$ — the Einstein–Cartan equation, verified in Torsion, Contorsion and the Spin Connection in Biquaternionic Form — so on its solutions $$ \bigl\langle\tilde{Ξ},C_\lambda\,C_{\mu\nu}\,\tilde{Ξ}\bigr\rangle=\Sigma_{\lambda\mu\nu}=\kappa^{-1}T_{\lambda\mu\nu} , \qquad \lambda,\mu,\nu\ \text{distinct}. $$ The source's display has $2\kappa$ multiplying the torsion instead of $\kappa^{-1}$, so the printed constant is the reciprocal of the right one up to the factor $2$: the ratio of the two is $2\kappa^{2}$, which no choice of sign of $\kappa$ repairs. On repeated indices the left side is nonzero where the epsilon channel vanishes, so the display is not an identity either way.

The identity proper needs no torsion at all, and it is the one recorded for the corpus, $$ \Sigma_{\lambda\mu\nu}=-\,i\,\epsilon_{\lambda\mu\nu\rho}\,\varepsilon_\rho\,A_\rho , \qquad A_\rho=\bigl\langle\tilde{Ξ},\mathsf{V}_{\mathrm{vol}}\,C_\rho\,\tilde{Ξ}\bigr\rangle , $$ with $\epsilon_{0123}=+1$ and every index low. The source's reading of its factor-2 display as a definition of the torsion, $$ T_{\lambda\mu\nu}=\frac{1}{8\pi G}\,\epsilon_\rho{}_{\lambda\mu\nu}\,A_\rho , $$ carries $1/\kappa$ where the matching $T=\kappa\Sigma$ requires $\kappa$, and it drops both the unit $i$ and the weight $\varepsilon_\rho$ that the antisymmetrised display carries. The two displays agree with each other because each is the inverse of the other, and both are the inverse of the Einstein–Cartan relation: $\Sigma=2\kappa T$ inverted gives $T=\Sigma/2\kappa$, and the second display's $1/\kappa$ is the reciprocal of the first's $\kappa$, so the pair looks self-consistent while sitting on the wrong side of the coupling throughout. The factor $2$ in that display is the one the appendix attributes to the normalisation of $C_{\mu\nu}=\tfrac12[C_\mu,C_\nu]$, and Section 11 carries the note "verifying the factor 2 matches your normalization in (App. B)", which addresses a reader and survived proof-reading.

The Empirical Claims, and the Limits of the Paper

Section 13.3 and its appendices reach furthest. The source lists magnetar X-ray polarimetry, entangled atom interferometry and a cryogenic high-frequency torsion-wave search, with required torsion amplitudes around $7\times10^{-28}\,\mathrm{m}^{-1}$ and current $3\sigma$ limits within a factor of a few, and adds a Stern–Gerlach phase bifurcation and an electro-torsional Aharonov–Bohm phase in its appendices. These are quantitative and, in form, falsifiable, which is a virtue the corpus's gravity articles do not exhibit.

Their evidentiary problem is upstream of them. Each number is computed from a coupling that failed the checks above. The torsion amplitude comes from Eq. (5), which vanishes on exactly the components $T_{j\eta0}$ that the axial pseudo-vector $\mathcal{T}=\epsilon_{ij\eta}T_{j\eta0}$ of Eq. (26) is built from, and is nonzero only on the three spatial triples that no observable quoted in the paper uses. Both the Stern–Gerlach phase and the atom-interferometric path shift trace back to $\mathcal{T}$ and to $\sigma\cdot(\vec\nabla_\Phi\times\mathcal{T})$, so they inherit that. The Aharonov–Bohm phase comes from a quaternionic connection whose torsion part is the same object. A sensitivity calculation built on a wave equation such as $\Box_\Phi T=0$, or on a current that is not conserved, cannot constrain the theory it is meant to test.

Appendix D carries its own dimensional gap, in an equation of two terms. Its phase is $$ \delta\theta=\frac{qΞ_{B}}{\hbar}+\frac{\beta\,Ξ_{T}}{\hbar}\bigl(\delta s_z\bigr), \qquad \beta=\tfrac12 , $$ with $\delta s_z$ the difference of the two spin projections, B minus A, and $Ξ_T$ called "the torsion flux" and never defined. The magnetic term is dimensionally complete: a flux times a charge is an action, and an action over $\hbar$ is a phase. The torsion term cannot be, whatever $Ξ_T$ is taken to be. If it is the flux of the torsion, $\int T\cdot dA$, its dimension is a length; the ratio to $\hbar$ is then length over action, which is not a phase, and the dimensionless $\beta=\tfrac12$ supplies nothing. If instead $Ξ_T$ is given units of action, the appendix never says so, and no display connects it to the torsion that the rest of the paper computes. The coefficient that the term needs is a coupling with the dimension of $\hbar$ per unit torsion flux, and that is exactly the slot where the missing power of $\hbar$ of Eq. (26) sits. The tables are recorded here as the source's claims, with the observation that none of them survives the recomputations above, and therefore that their numerical agreement with current sensitivities should not be read as evidence for the framework.

Their arithmetic is also worth one line, because it can be checked from the printed page alone. The worked example after Eq. (27) evaluates the phase as $\ell/(\hbar\,v)$ times a potential of order $10^{-30}$, quotes $\ell=1\,\mathrm{m}$, $v=10^3\,\mathrm{m\,s^{-1}}$ and $\hbar\simeq10^{-34}\,\mathrm{J\,s}$, and prints the answer as $10^{-3}\,\mathrm{rad}$; the prefactor written in its own denominator is $10^{-34}\cdot10^{3}=10^{-31}$, so $10^{-30}/10^{-31}$ is of order ten, not $10^{-3}$. The printed line is off by four orders of magnitude from the product printed beside it. The potential inserted in it, $10^{-30}\,\mathrm{J}$, is quoted rather than computed: with $\beta=\frac{\hbar^{2}}{8\,m}$ from Eq. (26) and the caesium mass that section's clock reference implies, $\beta\,\lvert T\rvert^2$ for the paper's own $\lvert T\rvert\simeq3\times10^{-32}\,\mathrm{m^{-1}}$ is of order $10^{-107}\,\mathrm{J}$, so the estimate's input is as unsourced as its arithmetic.

Two smaller items belong in the record. Section 13.2's torsion-free limit displays $\sqrt{-g}\,\tfrac{1}{2\cdot8\pi G}R$, which is $\tfrac{1}{16\pi G}R$ and agrees with Eq. (15) — the one place the paper's coupling arithmetic closes, and it closes only when the $2$ is read as multiplying $8\pi G$; and Section 13.2's claim that "setting $T_{\mu\nu}{}_\rho=0$ forces $[q_\mu,q_\nu]=0$" is false, since the commutator of distinct quaternion units never vanishes — only a bilinear of it can. Sections 5.6 and 11 contain editorial remnants ("your normalization", "Palatini variation sign convention consistency verification"), which together indicate a manuscript that did not pass a proof-reading stage.

Finally, Section 13.1's Table 1 asserts $w_2(TM)=0$ for the spin structure and $w_4(TM)=0$ for "quaternionic triviality" and global $\mathbb{H}$-sections. The first is the standard spin condition; the second is presented in the table as a check although the paper calls it an assumption elsewhere, and either way it is a global topological statement that no pointwise algebra can supply. The corpus's position on exactly this point is stated in Curved Spacetime and the Biquaternion Framework and is not softened by the source: whether the fields exist on the manifold at all is a topological question, and a bilinear construction at each point cannot answer it.

What Would Make It a Framework Claim

It is worth stating what the source would have to show for the corpus to record its programme as a candidate rather than as a claim with errors. The corpus's standard for a constructive claim is that the object be built, the identities checked, and the boundary stated.

  • A nonzero composite tetrad on the free solutions. The predicate fails at the first step: Eq. (4) is zero for every plane wave. A construction whose frame vanishes on the field it is built from needs a different bilinear, and the correction is not cosmetic, since the index structure and the derivative count both change.
  • A torsion of the size it claims. Eq. (5) must agree with the totally antisymmetric spin density $\langle\tilde{Ξ},\operatorname{Alt}[C_\rho C_\mu C_\nu]\tilde{Ξ}\rangle$, and it cannot while it is built from the four-element commutator: three components against four, with disjoint support.
  • An identity chain that carries information. The trace identity must carry its factor $i$, the bilinear must be antisymmetrised where the axial channel is claimed, the coupling must sit on the side Einstein–Cartan puts it, and the Fierz rearrangement must be one in which more than the scalar channel survives.
  • A constraint that constrains, or an action that is what it says. If the Lagrange multipliers only restate definitions and their own equation of motion is not solved, the action does not rise to a theory of the frame.
  • A conservation law that follows. The entropy current requires the torsion to be the dual of a harmonic potential; that assumption must be made or the conservation dropped.
  • One dimension per object. Eq. (4), Eq. (6), the teleparallel constraint and the torsion potential each mix powers of mass inside a single display. A construction that is to be read as geometry has to survive a units check before it is read at all.

None of these is a philosophical objection, and none requires the corpus's conventions. They are the arithmetic and index checks that any construction of this kind has to pass, and on this pass it does not.

Summary

The paper records a bi-quaternionic field as the primary object, builds a tetrad, a torsion and a contorsion as bilinear composites of it, and claims to derive the Dirac, Einstein–Cartan and spin–torsion equations from a single constrained action, together with an intrinsic arrow of time, a torsion quantum potential and a torsion entropy current.

Recomputation shows that the algebraic displays do not reproduce. The printed block map is off by the factor $i$, so it is not a map of the required square for the quaternion units the same section defines: the declared units give $G_\eta^2=+1$ where the signs need $-1$. The trace normalisation $\mathrm{Tr}_{\mathbb H}(q_\mu q_\nu)=2\delta_{\mu\nu}$ contradicts that embedding, for which the computed value is $2(-\varepsilon_\mu\delta_{\mu\nu})$; one substitution, $q_\eta\mapsto iq_\eta$, repairs both displays, and it is the one the paper never makes. The completeness relation's listed elements span four independent quaternion directions rather than the sixteen graded products it later uses. And the Fierz identity of Eq. (3) has a coefficient that depends on a notation it does not fix: $1$ in the reading of $\mathrm{Re}$ the display prints, $\tfrac14$ in the two-dimensional trace its own derivation cites, $\tfrac1{16}$ in the four-dimensional $\mathrm{ReTr}$ of Eqs. (4)–(6), which is the convention the composites use; the printed $\tfrac14$ is right for one of the three and for one only. Either way three of its four terms vanish identically, so it asserts only that the quaternion norm is multiplicative.

The composite geometry fails at its first display. The composite tetrad vanishes identically on every plane wave, verified to $4\times10^{-16}$, so the construction assigns no metric to the free Dirac field. The torsion of Eq. (5) is nonzero on three of its twenty-four independent components — the three cyclic spatial triples, all equal to $-\kappa\mathrm{ReTr}\langle\tilde{Ξ},\,\tilde{Ξ}\rangle$ — while the totally antisymmetric spin density it is matched to carries four and is nonzero where Eq. (5) vanishes; and the two displays that define that torsion put the output index at opposite ends, which for a tensor antisymmetric in one pair only is a difference of substance, with the paper's own worked example mis-signing the component it computes. The "vanishing symmetric part" check fails on the ten index triples where the product collapses to a scalar, and the vector current the display would need is zero in its three spatial components, $\mathrm{ReTr}\langle\tilde{Ξ},\,q_\eta\tilde{Ξ}\rangle=0$ identically. The contorsion–torsion relation carries its last index in the wrong order, which on totally antisymmetric torsion reverses the sign of the contorsion — $-\tfrac12$ where the standard relation gives $+\tfrac12$ — and the paper never states the sign convention that would make it right.

The action. The couplings in Eqs. (4) and (6) do not close dimensionally, $M^6$ and $M^2$ against a dimensionless tetrad and a contorsion of dimension $M$; that one false power propagates into the teleparallel constraint, whose two right-hand sides become $M^3$ and $M^4$ where the curvature they reconstruct has $M^2$. The action's multipliers restate definitions, leaving a coefficient $2\kappa$ in the Einstein equation that becomes $\kappa$ only if an unexplained multiplier equals $\tfrac12T$. The step offered for the Dirac equation declares a sum of graded basis elements zero "by orthogonality", which kills such a sum only when every coefficient already vanishes.

The sections beyond the geometry. The teleparallel section is standard geometry carrying a composite that does not exist, and its two consecutive displays of the Weitzenböck contorsion disagree in the sign of one term. The Kozyrev time gradient is the constant covector $(-1,0,0,0)$ in every field when computed with the paper's own $q_5=-1$, so the intrinsic time is not recovered. The torsion Bohm potential appears with both signs within one section, its "quaternionic correction" is built on components that Eq. (5) sets to zero, and the conflation of a constant commutator with a torsion field is the same one as in Eq. (5); and in the three-term potential the last two terms carry $\hbar$ where their own dimension requires $\hbar^{2}$, so two of the three are frequencies beside an energy. The entropy current is not conserved: $\Box_\Phi T=0$ does not imply $\partial J_{S\mu}/\partial x_\mu=0$, with the counterexample $T_{123}=x_0$. The Appendix B trace identity is missing a factor $i$ — the exact value is $-4i\epsilon_{\rho\lambda\mu\nu}$ — the bilinear is not the epsilon channel, so the axial channel is reached only after antisymmetrisation, and the factor-2 relation puts $2\kappa$ where Einstein–Cartan puts $\kappa^{-1}$, a discrepancy of $2\kappa^2$ that the paper's own second display then inverts back. The empirical tables inherit all of this, the electro-torsional Aharonov–Bohm phase puts a dimensionless coefficient in front of a torsion flux that no display defines, and the topological table asserts a global condition that no pointwise algebra can establish.

Two equations carry the word Dirac in this article, and they are kept apart. The source's is two-sided, with the time derivative acting from the right and the mass from the left, and its kinetic term exchanges $L$ and $R$. The corpus's is the one-sided pair, in which the gradient keeps the letter and the mass is the flip.

What the source does supply is a reminder, in a form the corpus's own articles state more carefully: the algebra can hold the frame and the connection, and the step from holding to generating is where the programme is decided. This article records the step taken, the displays that survive it, and the displays that do not.

Summary of Notation

symbol meaning
$\tilde{Ξ}$ the source's bi-quaternionic field
$\bigl\langle\tilde{Ξ},Ξ\bigr\rangle$ the pairing $\tilde{Ξ}^{\dagger}i\,Ξ\,\vec e_x$ that the source's Dirac adjoint denotes: it pairs two states and acts on neither
$q_\mu$ the quaternion units, $q_0=1$ and $q_\eta$ the three imaginary ones, acting by left multiplication $\tilde{Ξ}\mapsto e_\mu\tilde{Ξ}$
$C_\mu$ the frame action, $C_t:\tilde{Ξ}\mapsto i\tilde{Ξ}\vec e_x$ and $C_\eta:\tilde{Ξ}\mapsto i\vec e_\eta\tilde{Ξ}\vec e_y$
$L$, $R$ the two values the chirality subscript takes: $L$ is the minus value, the Dirac convention, and $R$ the plus value. A product takes the letter of its right factor and left multiplication keeps it
$\mathsf{V}_{\mathrm{vol}}$ the chirality action $\tilde{Ξ}\mapsto-i\tilde{Ξ}\vec e_z$
$\Gamma_A$ the sixteen graded products of the $G_\mu$, read as arrays: the graded basis of Appendix B.1, and the basis in which the source declares the constraint term zero
$G_\mu$ the source's four generators of Eq. (1), read as the actions displayed in the text: $G_0:\tilde{Ξ}\mapsto i\vec e_x\tilde{Ξ}$, $G_x:\tilde{Ξ}\mapsto-\vec e_y\tilde{Ξ}\vec e_x$, $G_y:\tilde{Ξ}\mapsto\vec e_y\tilde{Ξ}\vec e_y$, $G_z:\tilde{Ξ}\mapsto-\vec e_y\tilde{Ξ}\vec e_z$
$C_{\mu\nu}$ the commutator $\tfrac12[C_\mu,C_\nu]$ of two frame actions, the slot of the source's $\tfrac12[q_\mu,q_\nu]$: the product $C_\mu C_\nu$ for distinct indices, left multiplication by the bivector for two spatial indices, two-sided for one time and one spatial index
$\tilde\nabla_\Phi$, $(\tilde\nabla_\Phi)^{\natural}$, $\Box_\Phi$ the corpus's gradient, conjugate gradient and d'Alembertian, defined in the introduction: the only derivative operators this article uses of its own, all acting from the left
$D_\mu$ the source's covariant derivative, a derivative and not an action: it carries the spin connection and is never a basis operator
$\operatorname{Alt}$ the normalised antisymmetrisation, the average over the six permutations with signs
$\varepsilon_\mu$ the sign vector $\varepsilon_t=+1$, $\varepsilon_\eta=-1$, which writes a raised index low
$\kappa=8\pi G$ the source's single coupling, used both as $\kappa$ and as $\kappa^{-1}$
$\mathrm{ReTr}(X)$ the real part of the trace of $X$, which is $n$ times its central coefficient: $n=2$ for $\mathrm{Tr}_{\mathbb H}$ and $n=4$ for the four-dimensional trace of Eqs. (4)–(6)
$\exp(a){}_\mu$ the source's tetrad $e^a{}_\mu$, Eq. (4)
$T_\rho{}_{\mu\nu}$ the source's composite torsion, Eq. (5)
$N_{\mathrm{cont},\mu ab}$ the source's composite contorsion, Eq. (6)
$\Sigma_{\lambda\mu\nu}$ the corpus’s spin density, $\langle\tilde{Ξ},\operatorname{Alt}[C_\lambda C_\mu C_\nu]\tilde{Ξ}\rangle$, equal to $\kappa^{-1}T_{\lambda\mu\nu}$ on the source’s own solutions; $\Sigma_{\rho\mu\nu}$ is the same object carrying the index of the torsion it is matched to
$A_\rho$ the axial bilinear $\langle\tilde{Ξ},\mathsf{V}_{\mathrm{vol}} C_\rho\tilde{Ξ}\rangle$
$\mathrm{Sd}_{\mu\nu ab}$ the source's teleparallel Lagrange multiplier
$\lambda_{\mu a}$, $\Lambda_{\mu ab}$ the source's Lagrange multipliers
$q_5=q_1q_2q_3$ in the source, "the imaginary unit"; in fact the central scalar $-1$
$\tau=\mathrm{Arg}_{\mathbb{H}}\tilde{Ξ}$ the source's Kozyrev time field
$J_{S\mu}$ the source's torsion entropy pseudo-vector
$\mathcal{T}$ the axial pseudo-vector $\epsilon_{ij\eta}T_{j\eta0}$ on which the source's Pauli limit and torsional phases are built

Further Reading

  • J. G. B. Northey, "An External Construction of Space: Unifying Dirac and Einstein–Cartan through Quaternionic Spinors", International Journal of Quantum Foundations 11 (2025) 444–469 — the paper this article records and reviews; all quotations are from it.
  • F. W. Hehl, P. von der Heyde, G. D. Kerlick and J. M. Nester, "General relativity with spin and torsion: foundations and prospects", Reviews of Modern Physics 48 (1976) 393–416, for the standard Einstein–Cartan–Sciama–Kibble objects the source claims to reproduce.
  • R. Aldrovandi and J. G. Pereira, Teleparallel Gravity: An Introduction (Springer, 2013), for the Weitzenböck connection and the teleparallel gauge used in the source's Section 9.1.
  • Torsion, Contorsion and the Spin Connection in Biquaternionic Form, for the independently verified torsion, contorsion, spin-density and Weitzenböck identities against which the source's Eqs. (4)–(6) and Appendix B were checked.
  • The Dirac Equation in Biquaternionic Form, for the one-sided pair set against the source's two-sided equation, and for the boundary between the gradient and the contraction a frame imposes on it.
  • Curved Spacetime and the Biquaternion Framework, for the corpus's boundary statement that the algebra carries the frame and the connection without generating either.
  • The Schrödinger Equation in the Non-Quantum Limit, for the corpus's quantum potential, against which the source's torsion Bohm term was compared.
  • The Biquaternion Basis of a Unified-Field Fibre Bundle, for the corpus's earlier record of an external construction of the same algebra, and for the editorial treatment this article follows of an external source that is recorded and not endorsed.