Abstract. We formulate a general obstruction framework for finite Lieclosure generated by pairs of competing derivations on a decomposedvector space. Under a non-integrability condition, the resulting commutatortower never truncates, the Lie algebra is infinite-dimensional,and no finite extension can preserve a unique attractor on a neutralsubspace. This framework arose originally from the study of collapse dynamicsin RLHF-trained transformer systems, where base and safety gradientsinteract incompatibly and force convergence to a reconstructed metastate.We present the abstract theory in a model-agnostic way and thendescribe two applications: (1) collapse geometry for transformer latentspaces; and (2) an operator-algebraic specialization to pairs of MASAsinMd(C), viewed as a structural obstruction for finite mutually unbiasedoperator systems. We conclude with a conjecture suggesting that in composite dimensionsthe non-integrability condition holds for mutually unbiased MASAs,which would obstruct the existence of complete sets of mutually unbiasedbases (MUBs) in those dimensions.MSC (2020): 46L05, 47L30, 81P45, 94A17.
Abstract. We formulate a general obstruction framework for finite Lieclosure generated by pairs of competing derivations on a decomposedvector space. Under a non-integrability condition, the resulting commutatortower never truncates, the Lie algebra is infinite-dimensional,and no finite extension can preserve a unique attractor on a neutralsubspace. This framework arose originally from the study of collapse dynamicsin RLHF-trained transformer systems, where base and safety gradientsinteract incompatibly and force convergence to a reconstructed metastate.We present the abstract theory in a model-agnostic way and thendescribe two applications: (1) collapse geometry for transformer latentspaces; and (2) an operator-algebraic specialization to pairs of MASAsinMd(C), viewed as a structural obstruction for finite mutually unbiasedoperator systems. We conclude with a conjecture suggesting that in composite dimensionsthe non-integrability condition holds for mutually unbiased MASAs,which would obstruct the existence of complete sets of mutually unbiasedbases (MUBs) in those dimensions.MSC (2020): 46L05, 47L30, 81P45, 94A17.
Abstract. We present a mathematically rigorous unified framework connecting three domains:(1) categorical classification formalism with transfinite stability properties, (2) non-nilpotency obstructionsin Lie algebras of derivations, and (3) structural constraints on mutually unbiased bases(MUBs) in composite dimensions.The central contribution is the Pompetzki Obstruction Framework, which provides precise conditionsunder which pairs of derivations generate non-nilpotent Lie algebras. We prove that infinite-dimensional operator algebras, the relevant obstruction is non-nilpotency of the lower centralseries, not infinite-dimensionality. This correction resolves a technical imprecision in earlierformulations.For MUBs, we prove a conditional obstruction theorem: if mutually unbiased MASA pairssatisfy a spectral non-periodicity condition, then complete MUB families cannot exist. We explicitlycharacterize when this condition holds via the Weyl–Heisenberg Envelope Theorem, connecting finiteLie closure to the existence of nice error bases, which are known to require prime-power dimension.All proofs are self-contained. The satirical “Swamp Formalism” is presented in Appendix Aas a pedagogical toy theory illustrating the axiomatic method, with its model-theoretic propertiesrigorously established. XD
Which pure states give uniform outcomes in several mutually unbiased measurements, and when can they form another measurement basis? We distinguish common unbiased states, their orthonormal completion, and further extension. For a displayed phase-deformed Fourier pair in dimension 2m, a clock symmetry and skew antiunitary complete every prescribed common unbiased state to an explicit orthonormal basis. In dimension six, an amplitude-profile obstruction proves strong unextendibility of a given mutually unbiased triple when its original third-basis rays are invariant under the ternary clock. We give an exact nonproduct example and an application to a companion whose columns are localized at single Fourier frequencies. For every exact Hadamard matrix in a specified closed entrywise neighborhood of a fixed reference matrix, the coordinate and Hadamard bases have exactly 48 common unbiased rays but no third measurement basis, with a quantitative lower bound on the squared Gram defect of every six-column candidate.
We also study auxiliary complex equations for common states, retaining algebraic multiplicities. For a flat unitary of order six with all 3×3 minors nonzero, the projective ray scheme is finite; each singular two-minor contributes two reduced isolated boundary points, and the affine coordinate algebra is finite-dimensional over C. In every order, a critical-algebra identity preserves multiplicities; in dimension six it gives finiteness for a fixed phase fibre. Specialization over a discrete valuation ring controls complete reciprocal-orthogonal collections. A nonphysical Tao/Potts collection separates algebraic from physical solutions. These results do not determine the global maximum number of mutually unbiased bases or classify all triples.
Abstract We present a comprehensive unified theory connecting three apparently disparate domains: (1) categorical swamp formalism for entity classification, (2) collapse dynamics in RLHF-trained transformer systems, and (3) structural obstructions to complete sets of mutually unbiased bases (MUBs) in composite dimensions. The unifying mechanism is the Pompetzki Obstruction Framework, which identifies non-truncating commutator towers as the fundamental obstruction to finite Lie closure in decomposed vector spaces. The theory begins with the Pompetzki Grand Unified Swamp Formalization, which establishes a rigorous mathematical framework for classifying entities into the canonical house set H = {Gryffindor, Ravenclaw, Hufflepuff, Slytherin} via transfinite recursion. We prove that the resulting theory TSwamp is complete, compact, and decidable when restricted to finite-depth predicates. We then abstract the core obstruction mechanism into a general framework applicable to any decomposed vector space with competing derivations. Two applications follow: first, we prove the Pompetzki Reconstruction Theorem for RLHF-trained transformers, establishing the existence, structure, and inevitability of collapse-state behavior under gradient-incompatible constraints. Second, we apply the framework to pairs of maximal abelian *-subalgebras (MASAs) in Md(ℂ), proving via the Field-Symmetry Closure Lemma that complete MUB families cannot exist in composite dimensions. The central insight is that Shrek, as a fixed point of the sorting operator, induces the same categorical structure that governs both AI alignment dynamics and quantum information-theoretic constraints. The swamp is all.
Mutually Unbiased Bases (MUBs) constitute a fundamental geometric structure in quantum theory, known for providing an optimal measurement scheme for quantum state tomography. In prime and prime-power dimensions, analytical constructions of maximal sets of MUBs are well-known and standard construction relies on the Weyl-Heisenberg (WH) group and finite fields. In non-prime-power dimensions, on the other hand, the existence of such maximal sets remains an open question. We present a generalized numerical method of constructing MUBs without any reliance on \textit{a priori} group structure or specific algebraic frameworks. Formulating the problem at the level of Gram matrix, we reduce the search for complete sets of d+1 MUBs to a phase space optimisation problem. We use the fact that the MUB Gram matrix is a projection matrix, and the third- and fourth-order trace constraints are necessary and sufficient conditions for a projection matrix. We further develop a classification framework based on third-order Bargmann invariants and automorphism groups, allowing us to probe the underlying algebraic and geometric structure of the resulting configurations. Numerical applications of this method in dimensions 3, 4, and 5 demonstrate that all numerically constructed solutions are mutually isomorphic, are isolated points in phase space, and possess automorphism groups that coincide exactly with the Clifford group, the normalizer of the WH group. In the lowest non-prime-power dimension 6 where the existence of a complete set remains open, our numerical search yields no solution. However, for incomplete sets of solutions, our search finds no set of four MUBs, but it yields several inequivalent triples, which we classify and whose neighbourhood we then investigate.
We establish a connection between the problem of constructing maximal collections of mutually unbiased bases (MUBs) and an open problem in the theory of Lie algebras. More precisely, we show that a collection of μ MUBs in $\K^n$ gives rise to a collection of μ Cartan subalgebras of the special linear Lie algebra $sl_n(\K)$ that are pairwise orthogonal with respect to the Killing form, where $\K=\R$ or $\K=\C$. In particular, a complete collection of MUBs in $\C^n$ gives rise to a so-called orthogonal decomposition (OD) of $sl_n(\C)$. The converse holds if the Cartan subalgebras in the OD are also †-closed, i.e., closed under the adjoint operation. In this case, the Cartan subalgebras have unitary bases, and the above correspondence becomes equivalent to a result of \cite{bbrv02} relating collections of MUBs to collections of maximal commuting classes of unitary error bases, i.e., orthogonal unitary matrices. This connection implies that for n≤5 an essentially unique complete collection of MUBs exists. We define \emph{monomial MUBs}, a class of which all known MUB constructions are members, and use the above connection to show that for n=6 there are at most three monomial MUBs.
We develop Stereo Algebra, a mathematical framework deriving the necessary structure induced by constrained distinguishability, with locality as the constraint. Given a finite-dimensional real inner product space V and a locality algebra L⊂End(V) representing operations accessible through independent action on distinguishable subsystems, we prove that V admits a unique orthogonal decomposition V=Vmono⊕Vstereo, where Vmono is the minimal L-invariant subspace and Vstereo its orthogonal complement. Information in Vstereo is inaccessible to all operations in L yet structurally present. From this decomposition we derive, without auxiliary assumptions: phantom centers as equivalence-class invariants; admissible transformations as the centralizer C(L); collapse as algebra enlargement; relational invariants as orbit structure; emergent composition obstructions; and learning as irreversible descent induced by controlled constraint expansion. The framework is algebraic and substrate-independent. Precise conditions under which the framework would be refuted are stated.