Dynamical Higher-Gauge Theory of Categorical (Co)limit Failure and Adaptive Symmetry in Condensed Matter Systems
Abstract
We develop a dynamical higher–gauge framework in which the failure of categorical limits andcolimits to remain within the same structural class—for instance, a colimit of local symmetry groupsthat ceases to be a group—is described geometrically by curvature fields. In this construction,the standard gauge connection (A, B) of a 2–group or crossed module encodes the obstruction tostrict composition, while an additional “structure field” Φ promotes the underlying categorical data(tΦ, .Φ, ω2,3(Φ)) to dynamical variables. The resulting action describes systems whose symmetry type can evolve self–consistently with their dynamics. Applied tocondensed–matter settings, this formalism provides an analytic description of transitions betweendistinct emergent gauge structures—for example the adaptive U(1) ↔ Z2 spin–liquid transition andthe melting of fracton order—and yields curvature–based predictions for topological responses, criticalexponents, and defect–bound excitations. The theory generalises fixed–group gauge and topologicalfield theories by allowing both the curvature measuring the failure of categorical coherence and thecategorical structure itself to become dynamical degrees of freedom, offering a unified geometriclanguage for adaptive symmetry and self–organising order in correlated matter.
Full text
Dynamical Higher-Gauge Theory of Categorical (Co)limit Failure and Adaptive Symmetry in Condensed Matter Systems Andrei T. Patrascu FAST Foundation, Destin FL, 32541, USA email: andrei.patr[email protected] We develop a dynamical higher–gauge framework in which the failure of categorical limits and colimits to remain within the same structural class—for instance, a colimit of local symmetry groups that ceases to be a group—is described geometrically by curvature fields. In this construction, the standard gauge connection ( A, B )of a 2–group or crossed module encodes the obstruction to strict composition, while an additional “structure field” Φpromotes the underlying categorical data (tΦ, .Φ, ω2,3(Φ)) to dynamical variables. The resulting action S[A, B, Φ] =Zh1 2g2kFΦk2+1 2h2kHΦk2+κ 2kDΦk2+V(Φ) + Ltop(Φ, A, B) + Lmatti describes systems whose symmetry type can evolve self–consistently with their dynamics. Applied to condensed–matter settings, this formalism provides an analytic description of transitions between distinct emergent gauge structures—for example the adaptive U(1) ↔Z2 spin–liquid transition and the melting of fracton order—and yields curvature–based predictions for topological responses, critical exponents, and defect–bound excitations. The theory generalises fixed–group gauge and topological field theories by allowing both the curvature measuring the failure of categorical coherence and the categorical structure itself to become dynamical degrees of freedom, offering a unified geometric language for adaptive symmetry and self–organising order in correlated matter. I. INTRODUCTION Motivation: persistent puzzles beyond fixed symmetry A central theme of modern condensed matter physics is that long–wavelength behavior is governed not only by which degrees of freedom are present, but by how they can be consistently transformed and glued together. Quantum spin liquids (QSLs), topologically ordered phases, symmetry–enriched topological (SET) phases, and fracton phases all exemplify situations in which the familiar paradigm of a fixed, global or local symmetry group acting on microscopic variables is insufficient to capture emergent constraints and responses [ 1 , 3 – 7 , 85 , 89 , 111 ]. In QSLs, projective implementations of lattice and spin–rotation symmetries lead to emergent gauge structures and anyonic excitations; in fracton phases, constrained mobility reflects generalized Gauss–law–type constraints and submanifold symmetries. At the same time, generalized global symmetries and higher–form symmetries have reframed aspects of long–distance order and response [10, 11]. Two empirical challenges stand out: 1. Adaptive/emergent symmetry. As interactions, density, or external fields are tuned, candidate materials can reorganize not only their order parameters but the effective symmetry structure governing low–energy degrees of freedom (e.g., U(1) spin liquids that “Higgs” to Z2 spin liquids, or vice versa; SET phases in which symmetry fractionalization patterns change under doping/pressure). Standard effective field theories must be rewritten by hand across such changes. 2. Coherence beyond groups. In many systems, local symmetry data glue only up to phases or higher equivalences. Microscopically, one can patch local symmetry groups {Gi} , but their (co)limit fails to be a genuine group at low energies—instead producing higher categorical structures (2–groups, groupoids, or fusion categories) that govern anyon fusion/associativity, projective actions, and subsystem constraints [5, 11–13]. These features point to a gap between standard gauge/topological field theories (which presuppose a fixed structure group) and the realities of strongly correlated matter. A framework is needed that (i) measures the failure of local symmetry data to glue strictly, and (ii) allows the very object that encodes symmetry to evolve dynamically with the state of the system.
2 Statement of the gap: fixed–group EFTs versus adaptive symmetry Conventional lattice and continuum gauge theories are built on a fixed compact Lie group G with a connection A and curvature F = dA + A∧A [ 15 , 16 ]. Topological theories like BF and Chern–Simons similarly assume a fixed underlying algebraic input. These frameworks succeed spectacularly when the emergent symmetry structure is stable (e.g., deconfined Z2 gauge theory for gapped spin liquids, U(1) gauge fields for gapless QSLs [17, 111]). However: • They do not intrinsically encode how local symmetry charts {Gi} fail to glue globally into a single group when composition is only defined up to phases/2–morphisms (projective and higher actions). • They cannot describe within one theory transitions in which the effective symmetry type changes (e.g., U(1) ↔Z2; onset/melting of fracton mobility constraints). • In tensor–category and anyon theories, associativity up to F –symbols (a 3–cocycle) is primary data, but a continuum geometric language that dynamically tracks such coherence (rather than fixing it) is missing. Put differently: even when higher symmetries and 2–group global symmetries are recognized [ 10 , 11 , 18 ], the structure of symmetry is treated as background data. There is no generally adopted condensed–matter framework in which both (i) the obstruction to strict gluing and (ii) the symmetry structure itself are dynamical degrees of freedom. Key idea: geometrizing (co)limit failure as curvature, and making it dynamical Our starting point is categorical: consider a diagram of local symmetry groups and homomorphisms D:··· → Gi→Gj→ ··· obtained from patching microscopic symmetries on overlapping regions. The categorical limit/colimit of {Gi} generally exists, but need not lie in the category Grp : composition may close only up to phases, inverses may exist only up to higher equivalence, and associativity may hold only coherently. The appropriate recipient is a higher symmetry object—most minimally, a 2–group (crossed module) ( G1 t −→ G0, . )with Peiffer identities [ 12 – 14 ]. On a spacetime M , a principal 2–bundle carries a 2–connection (A, B)with Lie–algebra–valued forms A∈Ω1(M, g0), B ∈Ω2(M, g1), and curvatures F=dA +A∧A−t(B), H =dB +A . B. (1) When the microscopic symmetry data glue to a true group, the obstruction vanishes and (in suitable gauge) F = H = 0. Nonzero F and H are the geometric avatars of coherence failure: the categorical difference between the (co)limit and a strict group shows up as curvature. The second step is dynamical. We promote these obstruction forms to fields with an action (schematically) S[A, B] = ZM 1 2g2kFk2+1 2h2kHk2+Ltop(A, B) + Lmatter(A, B;Ψ),(2) whose Euler–Lagrange equations couple ( F, H )to currents and implement generalized Bianchi identities (e.g., DF + t ( H )=0, DH = 0). This already yields a dynamical gauge theory of (co)limit failure: the obstruction can propagate, condense, and mediate responses, unifying deconfinement and topological responses across settings. The third step is genuinely new: we allow the symmetry structure itself to be dynamical. Introduce a structure field Φ(x)∈ M2-group,
3 valued in the moduli of crossed modules (or an effective submanifold thereof). The maps t and . and cohomology data (ω2, ω3)that specify the 2–group now depend on Φ: t7→ tΦ, . 7→ .Φ, ω2,37→ ω2,3(Φ). The curvatures (1) become FΦ=dA +A∧A−tΦ(B), HΦ=dB +A .ΦB, (3) and the full action reads S[A, B, Φ] = ZM"1 2g2kFΦk2+1 2h2kHΦk2+κ 2kDΦk2+V(Φ) + Ltop(Φ;A, B) + Lmatter(Φ;A, B;Ψ)#. (4) Equation (4) is a dynamical higher–gauge theory of adaptive symmetry: the same fields that measure failure of strict gluing also drive the local symmetry type (encoded by Φ) to evolve so as to reduce incoherence or follow material constraints. In condensed matter language, ( A, B )carry the emergent gauge fluxes (line and surface holonomies), while Φselects among U(1)–like, Z2 –like, or more exotic 2–group structures; V (Φ) and anomaly/topological couplings Ltop encode energetic and quantized preferences (e.g., projective symmetry classes and associator data) [1, 5, 11]. Anticipated results and roadmap We show that Eqs. (3)–(4) produce concrete, testable consequences that are not readily accessible in fixed–group frameworks: 1. Analytic classification of emergent gauge types. Minimizing V (Φ) plus topological couplings yields phase selection between U(1) and Z2 spin liquids and their symmetry–enriched variants. The cohomology classes ω2,3 (Φ) computed from the underlying lattice symmetries classify allowed phases without enumerating mean–field ansätze [1, 111]. 2. New critical exponents for type–changing transitions. Linearizing the coupled ( A, B, Φ) equations gives dispersions and scaling near transitions where the symmetry structure changes (“structural Higgsing”): e.g., the emergent photon gaps as Φselects Z2 ; a rank–2 tensor gauge mode softens as fracton constraints melt [7, 85]. 3. Quantized responses from curvature integrals. Integrals of HΦ over 3–cycles, and mixed anomaly couplings Ltop (Φ; A, B ), produce quantized topological responses (ground–state degeneracy on manifolds, magnetoelectric and thermal Hall coefficients) that jump when Φcrosses domain walls. These are computed directly from curvature classes, providing a geometric route to SET response catalogues [5, 6, 11]. 4. Defect–bound excitations at symmetry–type interfaces. Spatial textures of Φ( x )create domain walls along which the 2–group structure changes; we show they bind protected modes (anyon–condensation interfaces) with calculable spectra, offering STM/neutron probes. 5. Lattice formulation for simulation. We discretize ( A, B, Φ) on triangulations: link variables U`∈G0 , face variables Wf∈G1 , and site/cell variables Φ p ; discrete curvatures FΦ ( c ) , HΦ (4 –cell ) yield Monte–Carlo–measurable loop and surface operators that diagnose deconfinement and structure selection in a single simulation. Physical case studies. We will work out two detailed applications. (i) An adaptive U(1) ↔Z2 spin–liquid transition on kagome/triangular lattices, with predictions for the photon gap, vison mass, specific heat crossover ( C ( T ): T3→e−m/T ), and quantized changes in thermal Hall response when ω3 (Φ) flips. (ii) A fracton “melting” transition described by a dynamical rank–2 (higher–form) sector in which Φcontrols which multipole moments are conserved; we derive mobility constraints from DHΦ = 0 and identify the conditions under which single–fracton motion becomes allowed.
4 Contributions. Conceptually, we provide the first condensed–matter field theory that (a) geometrizes the failure of categorical (co)limits of symmetry groups as curvature and (b) promotes the symmetry structure (crossed–module data) to a dynamical field. Technically, we give explicit continuum and lattice actions, derive coupled equations and Bianchi identities with variable structure maps, and compute spectra and responses. Phenomenologically, we propose measurable signatures (loop/surface operator laws, thermal transport, defect spectroscopy) that distinguish type–changing transitions from conventional Higgs/deconfinement transitions. Organization. Section 2 reviews categorical preliminaries and derives the higher–gauge curvatures as obstructions. Section 3 introduces the dynamical obstruction theory (2). Section 4 promotes the symmetry structure to a field and analyzes the coupled dynamics (4). Sections 5–6 develop condensed–matter applications and predictions; Section 7 presents a lattice discretization and simulation protocol; Section 8 discusses broader implications and experimental outlook. II. FROM CATEGORICAL (CO)LIMITS TO HIGHER–GAUGE CURVATURE A. Categorical background This subsection fixes the categorical language and proves the basic structural facts we shall use repeatedly. We begin with diagrams and their (co)limits, specialize to the category of groups Grp, and then explain what happens when gluing succeeds only up to higher equivalence: the natural recipient is a 2–group (equivalently, a crossed module). Throughout, we use standard references for category theory and higher groups [21–23]. Diagrams and universal properties. Let C be a category and J a small indexing category. A diagram of shape Jin Cis a functor D:J−→ C. Acone to D with apex X∈Ob ( C )is a family of morphisms {φj:X→D ( j ) }j∈Ob(J) such that for every u:j→j0 in J one has D ( u ) ◦φj = φj0 . A limit of D is a terminal cone ( L, {πj} ): for any other cone (X, {φj})there exists a unique u:X→Lwith φj=πj◦ufor all j. Dually, a cocone from D with apex X is a family {ψj:D ( j ) →X} compatible with D ( u ), and a colimit (C, {ij})is an initial cocone: for any cocone (X, {ψj})there is a unique v:C→Xwith ψj=v◦ij. Intuition. Limits encode compatible choices (pulling data back to a common core); colimits encode compatible amalgamations (pushing data forward to a universal glue). The power of universal properties is that they specify objects up to unique isomorphism and behave functorially under change of diagrams [21]. Special cases. Products, equalizers, coproducts, and coequalizers are the building blocks: •The product A×Bis the limit of the diagram A←·→B. •The equalizer Eq(f, g)⊆Xof f, g :X⇒Yis the limit of that parallel pair. •The coproduct AqBis the colimit of · → A, · → B. •The coequalizer Coeq(f, g)is the colimit of X⇒Y. Limits and colimits in Grp .We recall (and rederive constructively) that the category of groups is complete and cocomplete. Proposition 2.1. The category Grp of (discrete) groups and homomorphisms has all small limits and colimits. Proof. Limits: Given a diagram D:J→Grp , the limit can be realized as a subgroup of the product QjD ( j )consisting of all tuples ( xj )compatible with the arrows of J . This is standard and uses that products and equalizers exist in Grp: YjD(j)with EqYuD(u),id.
5 Colimits: For a diagram D , form the free product ∗jD ( j ), then quotient by the normal closure of the relations that encode the arrows of J. Explicitly, for each u:j→j0and x∈D(j)impose ιj0D(u)(x)=ιj(x) where ιj:D(j)→ ∗kD(k)is the canonical injection. The resulting quotient colim D∼ =∗jD(j).DDιj0(D(u)(x))ιj(x)−1:u, x EE satisfies the universal property of a colimit. Thus products/equalizers and free products/normal quotients yield all small (co)limits [21, Ch. V]. Remark. The proof is constructive and will be mirrored later in our lattice discretizations: products/equalizers produce constraint subspaces; coproducts/coequalizers produce glued spaces modulo relations. Subcategories with additional structure. While Grp is complete and cocomplete, structured subcategories need not be. For instance: • The category TopGrp of (Hausdorff) topological groups is complete and cocomplete (free topological groups exist), but subtle issues of final topologies can arise for large colimits. • The category LieGrpfd of finite–dimensional Lie groups is complete but not cocomplete in general: coproducts (free products) typically leave the finite–dimensional setting; coequalizers may force identifications that destroy a smooth structure. One must pass to larger categories (pro–Lie groups or convenient locally convex Lie groups) to recover cocompleteness [25]. This is the first signal that a diagram of structured groups can have a (co)limit in Grp that exits the structured subcategory. Our condensed–matter applications live exactly in this regime: microscopic symmetry data glue in a way that cannot be captured by a single object of the intended subcategory (e.g., a finite–dimensional Lie group), but is naturally expressed by a higher symmetry object. Gluing that succeeds only up to equivalence: 2–groups and crossed modules From strict to weak composition. Suppose we try to glue local symmetry charts Gi on overlaps Ui∩Uj with transition homomorphisms φij :Gi→Gj. Strict cocycle conditions would demand φjk ◦φij =φik on Ui∩Uj∩Uk.(5) In many physical situations (projective symmetries, anyon fusion, subsystem symmetries), (5) fails but holds up to a controlled correction: φjk ◦φij = Adαijk ◦φik, αijk ∈G1,(6) where G1 is another group acting by automorphisms on the Gi (physically: a “phase” or defect transport). Consistency on quadruple overlaps imposes a higher relation among the αijk—a 3–cocycle condition. Equation (6) says that composition works up to a specified isomorphism. Categorically, this means we have left the world of strict groups and entered the world of group–like monoidal groupoids: 2–groups. Definition (2–group). A2–group is a (small) monoidal groupoid ( G,⊗, 1) in which every object is invertible up to isomorphism and every morphism is invertible. Equivalently, a 2–group is a one– object strict 2–category in which all 1– and 2–morphisms are invertible. The monoidal structure ⊗ is associative/unital up to coherent isomorphisms (the associator and unitors) satisfying Mac Lane’s pentagon and triangle identities [21]. There is a powerful algebraic model for 2–groups: crossed modules. Definition (crossed module). Acrossed module of groups consists of a pair of groups ( G1, G0 ), a homomorphism t:G1→G0 (“target”), and an action .:G0→Aut ( G1 )such that the Peiffer identities hold for all g∈G0,h, h0∈G1: t(g . h) = g t(h)g−1,(7) t(h). h0=h h0h−1.(8) Intuitively, t records which 1–morphisms are “boundaries” of 2–morphisms, and . tells how 1–morphisms act on 2–morphisms; the Peiffer identities assert compatibility between them.
6 From crossed modules to strict 2–groups. Given a crossed module ( G1 t −→ G0, . ), one builds a strict 2–group Gas follows [23]: •Objects are elements of G0. •A morphism gh =⇒g0is specified by h∈G1with t(h) = g0g−1. • Vertical composition of morphisms corresponds to multiplication in G1 ; horizontal composition (monoidal product) corresponds to multiplication in G0together with the action .. The identities (7) – (8) guarantee that source/target are respected and that horizontal/vertical compositions satisfy the interchange law. This construction yields a strict 2–group (associator/units are identities). Proposition 2.2. The assignment above defines a strict 2–group. Proof (sketch). Given gh =⇒g0 and g0k =⇒g00 with t ( h ) = g0g−1 and t ( k ) = g00g0−1 , vertical composition is gkh =⇒g00 since t ( kh ) = t ( k ) t ( h ) = g00g0−1g0g−1 = g00g−1 . For horizontal composition, on objects (g1, g2)7→ g1g2; on morphisms (g1 h1 =⇒g0 1)⊗(g2 h2 =⇒g0 2)is g1g2 h1(g1.h2) =======⇒g0 1g0 2 and the Peiffer identities ensure that sources/targets and associativity constraints hold strictly. All morphisms are invertible (use inverses in G1and G0). From 2–groups to crossed modules. Conversely, given a strict 2–group G , taking G0 = Ob ( G )with the monoidal product and G1 = AutG (1) (the automorphism group of the unit object) with the conjugation action recovers a crossed module. This sets up an equivalence of categories between strict 2–groups and crossed modules [23]. Physical intuition. A crossed module packages two layers of symmetry: G0 (ordinary transformations) and G1 (“gauge–of–gauge” transformations). Whenever microscopic gluing succeeds only up to controlled corrections, these corrections live as elements of G1 , and their interaction with G0 is encoded by t and . . The Peiffer identities express the compatibility constraints that appear (for example) in projective symmetry actions or anyon associativity. How (co)limit failure produces a 2–group We now make precise the statement used heuristically in the Introduction: a diagram of local groups whose gluing fails strictly but succeeds up to specified corrections admits a canonical enhancement to a diagram valued in 2–groups, whose (bi)categorical colimit is a 2–group capturing the failure. Let U = {Ui} be an open cover of a space and consider the Čech groupoid ˇ C ( U ): objects i , morphisms i→jfor each nonempty Ui∩Uj, and composable triples for Ui∩Uj∩Uk, etc. Suppose we are given: •Groups Gion each Ui. •Homomorphisms φij :Gi→Gjon double overlaps. •Elements αijk ∈G1,ijk (for some system G1,ijk) on triple overlaps, witnessing (6). These data amount to a pseudofunctor from ˇ C ( U )to the 2–category of groups, homomorphisms, and conjugacies. The obstruction to strict functoriality is the 2–cell αijk . Composition of 2–cells on quadruple overlaps yields a 3–cocycle constraint (the pentagon), i.e. a class in H3(ˇ C(U),center). Proposition 2.3. Under the hypotheses above, the bicategorical colimit (a homotopy colimit) of the pseudofunctor exists and is canonically a 2–group. Its associated crossed module (G1 t −→ G0, .)has: 1. G0the ordinary colimit of the Gimodulo the relations induced by φij; 2. G1 generated by the symbols αijk (and relations from quadruple overlaps), with t ( αijk ) = φjkφijφ−1 ik and the natural conjugation action.
7 The Peiffer identities follow from the coherence of the pseudofunctor (pentagon and triangle). Proof (sketch). Model the bicategorical colimit as a strict 2–group by transporting the weak gluing data into a crossed module: G0 is the 1–truncation (objects modulo 1–cell identifications), while G1 arises from the 2–cell corrections; source/target of 2–cells are precisely the defects of the 1–cell equalities, and the pentagon coherence yields (7) – (8) . A detailed construction can be found in [ 23 ]; the present setting is a Čech version of that construction. Takeaway. The difference between a strict colimit in Grp and the actual gluing pattern is captured by the crossed–module data ( G1 t −→ G0, . ). This difference is the higher symmetry we shall gauge in the next subsection: its curvature will measure the obstruction to strict gluing. Examples and intuitions relevant to condensed matter Projective symmetry groups. If φjkφij = Adαijk φik with αijk ∈ U(1), then t is trivial and (8) is abelian; G1 = U(1) centrally extends G0 . The class of α in group cohomology H2 ( G0, U(1)) is the obstruction to lifting a projective representation; its differential in H3 is the associator (F–symbol) that appears in string–net models and symmetry fractionalization [24]. Subsystem/fracton symmetries. When composition is restricted (e.g., only along submanifolds), the diagrams of groups exist but their colimits in Grp carry many idempotents/partial operations. Passing to 2–groups (or more generally to higher groupoids) packages these partial coherences into G1 and its action, preparing the ground for tensor–gauge theories. From categorical background to geometry. In Section II B we will equip a principal 2–bundle with a 2–connection (A, B)valued in (g1 t −→ g0, .)and show that its curvatures F=dA +A∧A−t(B), H =dB +A . B, are precisely the geometric incarnations of the obstructions just described: F measures the residual failure of 1–cell equalities (strict group composition) and H measures the residual failure of coherence among the corrections (associativity up to 2–cells). This is the bridge from categorical (co)limits to higher–gauge curvature that underlies our dynamical constructions. Notation. We write 2Grp for the 2–category of strict 2–groups (equivalently, crossed modules), g0 and g1 for their Lie algebras when smooth structure is present, and use t and . both at the group and Lie–algebra level when no confusion arises. B. Curvature as obstruction In this subsection we pass from the purely categorical discussion of §II A to geometric data on spacetime M: principal 2–bundles and their connections. We show that the curvatures of a 2–connection, F=dA +A∧A−t(B), H =dB +A . B, (9) are precisely the geometric incarnations of the obstructions identified categorically: •F measures the residual failure of 1–cell compatibility (strict group composition) when gluing local symmetry charts; •Hmeasures the residual failure of coherence among the corrections (associativity up to 2–cells). We make this precise by deriving the transformation laws, Bianchi identities, and a descent statement that identifies vanishing curvature with strict gluing (true colimits in Grp).
8 Set–up and notation. Let ( G1 t −→ G0, . )be a (Lie) crossed module with Lie algebras ( g1 t −→ g0, . ). The action .differentiates to an action of g0on g1, and tis g0–equivariant: t(X . Y ) = [X, t(Y)] for X∈g0, Y ∈g1.(10) We use the following conventions for differential forms with values in these Lie algebras: A∈Ω1(M, g0), B∈Ω2(M, g1), A∧A:= 1 2[A , A]∈Ω2(M, g0), A.B:= A∧.B∈Ω3(M, g1), where ∧. denotes the wedge product followed by the g0 –action on g1 . The g0 –covariant derivative on g0–valued forms is D:= d+ [A, ·], and on g1–valued forms is D:= d+A . (·). 2–connections and their curvatures Aprincipal 2–bundle with structure 2–group ( G1 t −→ G0, . )can be defined in several equivalent ways (e.g. via bibundles, anafunctors, or descent data); see [ 26 , 27 ]. For our purposes it suffices to fix a good cover {Ui} of M and describe a 2–connection by local differential forms ( Ai, Bi )on each patch and higher transition data on overlaps (we return to these explicit overlap formulas below). Globally, a 2–connection is specified by (A, B)as above, with curvatures (9). Definition II.1 (Curvatures of a 2–connection) . Given A∈ Ω 1 ( M, g0 )and B∈ Ω 2 ( M, g1 ), the 2– curvature (F, H)is F:= dA +A∧A−t(B)∈Ω2(M, g0), H := dB +A.B∈Ω3(M, g1). We call F the fake curvature (it would vanish for a strictly horizontal 2–connection) and H the 3–curvature. Remark II.2 (Physics intuition).The 1–form A gauges the ordinary (1–level) symmetry G0 ; the 2–form B gauges the “gauge–of–gauge” layer G1 . The combination t ( B )subtracts the part of dA + A∧A that arises from 2–morphisms, so that F measures the residual failure of strict 1–composition. The 3–curvature H measures how the 2–morphism corrections fail to be coherent under parallel transport by A. Gauge transformations and covariance of curvature There are two levels of gauge transformations [26, 28]: •a1–gauge transformation by a smooth map g:M→G0; • a2–gauge transformation by a g1 –valued 1–form a∈ Ω 1 ( M, g1 )(infinitesimally, or by a G1 –valued function at the finite level). Under a 1–gauge transformation g, the fields transform as A7→ Ag:= g−1Ag +g−1dg, B 7→ Bg:= g−1. B, (11) where g−1. B denotes the induced action. Under a 2–gauge transformation a, A7→ A0:= A−t(a), B 7→ B0:= B−Da −1 2[a∧a]g1,(12) where [·,·]g1is the Lie bracket on g1and Da := da +A.a. Lemma 2.4 (Curvature covariance). Under (11)–(12), the curvatures transform covariantly: F7→ F0=g−1Fg, H 7→ H0=g−1. H. (13) Proof. For a 1–gauge transformation, Ftransforms as in ordinary gauge theory: Fg=dAg+Ag∧Ag−t(Bg) = g−1(dA +A∧A−t(B))g=g−1Fg,
9 using the equivariance t(g . B) = g t(B)g−1. Likewise, Hg=d(g−1. B)+(g−1Ag +g−1dg).(g−1. B) = g−1.(dB +A.B) = g−1. H. For a 2–gauge transformation (12), F0=d(A−t(a)) + (A−t(a)) ∧(A−t(a)) −tB−Da −1 2[a∧a]. Expanding and using dt ( a ) = t ( da )and t ( A . a ) = [ A, t ( a )] (equivariance (10) ), and that t ([ a∧a ]) = [ t ( a ) ∧t ( a )], the a –dependent terms cancel pairwise, leaving F0 = F . A similar calculation (or functoriality of the action) shows H0=H. Consequence. The vanishing or non–vanishing of F and H is a gauge–invariant statement: if either curvature is nonzero in one gauge, it is nonzero in all. Bianchi identities The curvatures satisfy higher Bianchi identities that tie fake curvature and 3–curvature together. Proposition 2.5 (Higher Bianchi identities). With Dthe g0–covariant derivative, DF +t(H) = 0, DH =F . B. (14) Proof. Compute DF =dF + [A, F] = d(dA +A∧A−t(B)) + [A, dA +A∧A−t(B)]. The ordinary Bianchi identity for A (on g0 –valued forms) gives d ( dA + A∧A )+[ A, dA + A∧A ]=0. Thus DF =−d t(B)+[A, t(B)]=−tdB +A.B=−t(H), using equivariance (10). For H, DH =dH +A.H=d(dB +A.B) + A . (dB +A.B). Expanding and using d2 = 0, bilinearity of . , and the identity d ( A . B ) + A . ( dB ) = ( dA ) . B −A∧A . B+A . (dB),we obtain DH = (dA +A∧A). B =F+t(B). B =F . B, since t ( B ) . B = [ t ( B ) ,· ]acts as zero on B∈g1 by the Peiffer identity t ( h ) . h0 = hh0h−1 differentiated at the identity (this makes im ( t )act by inner automorphisms on g1 , annihilating the wedge–self–action in degree three). Interpretation. The first identity says that fake curvature cannot vary covariantly without being sourced by the 3–curvature; the second says that the 3–curvature fails to be covariantly closed exactly to the extent that fake curvature is present. In particular, in the fake–flat regime F = 0 we have DH = 0, so H defines a covariantly closed 3–form—the natural receptacle for surface holonomy and quantized responses. Descent data and curvature as obstruction We now relate ( F, H )to the categorical gluing data of §II A. Fix a good cover {Ui} of M . A 2–connection can be presented by local 1– and 2–forms ( Ai, Bi )on Ui , together with transition data on overlaps Uij , Uijk [26]: on Uij :gij :Uij →G0,Λij ∈Ω1(Uij,g1),(15) on Uijk :hijk :Uijk →G1,(16)
16 Topological couplings. Topological terms encode global data (extensions/associators) and produce quantized responses. We catalog the most common ones and their gauge invariance/quantization in spacetime dimension d. 1. BF–type coupling in d = 4.Assume a cross-pairing β:g1×g0→R compatible with t and . , e.g. β(U, X) := κ0t(U), X=κ1U, t†(X). Let F0:= dA +A∧A(without the t(B)term). Then SBF =k 2πZM βB∧F0(d= 4) (31) is gauge invariant modulo 2 πZ if k∈Z and the characteristic classes are integral.[ ? ] Its variation contributes k 2πDB to the A –equations and k 2πF0 to the B –equations, providing a topological mass mixing between the two sectors. 2. Chern–Simons–like coupling in d = 3.With the usual Chern–Simons 3–form CS ( A ) = κ0A∧dA + 2 3A∧A∧A, SCS =k0 4πZM CS(A) + k1 2πZM βB∧A(d= 3) (32) is topological; gauge invariance requires k0∈Z (the usual level quantization) and appropriate integrality for the mixed term.[?] 3. Cohomological couplings ω2, ω3 .Let ω2 and ω3 denote, respectively, a class that measures central extension (projective phases) and the associator (3–cocycle). In a continuum description these appear as background differential cocycles, Sω= 2πiZMω2,1 2πF0+ 2πiZVω3,1 2πdCS(A),(33) with ∂V = M (Wess–Zumino extension). The second term is well defined modulo 2 π i Z if ω3 is integral and the CS extension ambiguity is integral.[?] We group these contributions as Stop[A, B;ω2, ω3] = SBF +SCS +Sω,(34) and write the full gauge-field action as Sgauge[A, B] = Skin[A, B] + Stop[A, B;ω2, ω3].(35) Coupling to matter. Let Ψdenote matter fields. We distinguish: • Point excitations (spinons, chargons) transforming in a representation R of G0 . Minimal coupling uses the covariant derivative Dµ Ψ = ∂µ Ψ + Aµ·R Ψ. The associated conserved 1–form current is JA∈Ωd−1(M, g0). • String/membrane excitations (visons, vortex sheets) charged under the higher form; their worldvolumes couple through a conserved ( d− 2)–form current JB∈ Ω d−2 ( M, g1 )by Rκ1 ( B∧?JB ). Microscopically JBis sourced by the endpoints/edges of defect worldsheets. We write Smatt[Ψ; A, B] = ZMLmatt(Ψ, DΨ) + ZM κ0 A∧?JA+ZM κ1 B∧?JB,(36) where Lmatt is gauge invariant under both 1– and 2–gauge transformations. Field equations with sources. Varying Sgauge +Smatt with respect to Aand Byields 1 g2DF +1 h2ρ† B(H) + δStop δA =−? JA,(37) −1 g2t†(F) + 1 h2DH +δStop δB =−? JB,(38) with explicit topological contributions read off from (31) – (33) . For instance, in d = 4 the BF term contributes k 2πDB to (37) and k 2πF0to (38).
17 Ward identities (generalized continuity). Gauge invariance of Sgauge + Smatt under infinitesimal 1– and 2–gauge transformations with parameters ∈Ω0(M, g0)and η∈Ω0(M, g1), δA=D, δB= . B, δηA=−t(η), δηB=Dη, implies two Noether identities. Writing only the matter variations (the gauge-field part cancels by Bianchi identities (14) and topological quantization), we obtain hδA, ?JAi0+hδB, ?JBi1= 0 ⇒D†JA+ρ† B(JB)=0,(39) hδηA, ?JAi0+hδηB, ?JBi1= 0 ⇒ −t†(JA) + D†JB= 0,(40) where D† denotes the distributional adjoint acting on currents. Equations (39) – (40) generalize current conservation: the 1–form charge is not conserved in isolation if the 2–form sector carries charge, and vice versa; their exchange is mediated by tand .. Scaling and engineering dimensions. In d spacetime dimensions, [ κ0 ]=[ κ1 ]=1(dimensionless), [ A ]=1,[ B ]=2. Then [ F ]=2,[ H ]=3, and the kinetic couplings scale as [ g−2 ] = d− 4,[ h−2 ] = d− 6. In d = 4 the A sector is marginal while the B sector is irrelevant; topological BF couplings are marginal. In d = 3 both kinetic terms are relevant; Chern–Simons couplings are marginal and dominate the infrared. Boundary terms and anomaly inflow. On manifolds with boundary ∂M , the topological terms (34) reduce to boundary actions (e.g. Wess–Zumino and Chern–Simons functionals on ∂M ). Gauge variation produces boundary terms that must be cancelled by corresponding anomalies of edge degrees of freedom (anomaly inflow). For the Wess–Zumino term in (33) , this is the standard mechanism that relates 3–cocycles ω3to boundary associator phases. Summary of Sect. III A. • The kinetic action (27) together with the variation rules (28) yields coupled equations (29) – (30) that make the obstruction forms F, H propagating degrees of freedom. • Topological pieces (34) encode extension/associator data ( ω2, ω3 )and generate quantized responses (BF mixing, Chern–Simons levels, Wess–Zumino terms). • Matter couplings (36) give generalized continuity equations (39) – (40) , tying 1–form and 2–form charges through the crossed-module maps tand .. B. Equations of motion and conservation laws We now extract the dynamical content of the action constructed in §IIIA. We first reorganize the Euler–Lagrange equations into a compact “Maxwell form” with effective sources. Then we derive the corresponding Ward identities (generalized continuity equations) and the stress–energy tensor, showing explicitly how a nonvanishing t ( B )(fake curvature) and the higher sector H deform both Noether and energy–momentum conservation by mediating exchange between the 1– and 2–form sectors. When topological terms are present, we keep track of their (metric–independent) contributions separately. Preliminaries and notation. We retain the pairings κ0,1 , covariant derivatives D on g0 – and g1 –valued forms, and the Hodge operator ?from §IIIA. Curvatures are F=dA +A∧A−t(B)∈Ω2(M, g0), H =dB +A.B∈Ω3(M, g1), and satisfy the higher Bianchi identities (Proposition II.IIB): DF +t(H) = 0, DH =F . B. (41) We denote by JA∈ Ω d−1 ( M, g0 )and JB∈ Ω d−2 ( M, g1 )the matter currents defined in (36) and by δStop/δA and δStop/δB the (distributional) variational derivatives of the topological sector (34). Field equations in Maxwell form From (37)–(38) we define the effective source currents JA:= ?JA+δStop δA +1 h2ρ† B(H)∈Ω1(M, g0),(42) JB:= ?JB+δStop δB −1 g2t†(F)∈Ω2(M, g1).(43)
18 Then the equations of motion take the compact form 1 g2DF =−JA,1 h2DH =−JB.(44) Thus the 1–form sector is driven not only by the matter current JA and topological terms, but also by the higher sector through ρ† B ( H ); conversely, the 2–form sector is driven by JB , topological terms, and by the fake curvature via t†(F). Decoupled limits. If t = 0 and . = 0, then F = dA + A∧A and H = dB , and (44) reduces to two independent Yang–Mills/Maxwell systems with sources ?JA and ?JB . If B is constrained so that ρ† B = 0 (e.g. in the abelian case with infinitesimal B ), the first equation reduces further to the standard Yang–Mills form DF =−? JA. Generalized continuity (Ward) identities Gauge invariance of the full action under infinitesimal 1– and 2–gauge transformations implies the Ward identities (39) – (40) . We now incorporate the contributions of the dynamical gauge fields themselves by replacing the matter currents with the total currents JA,JB. Proposition 3.2 (Noether identities with mixing). Let JA,JBbe defined by (42)–(43). Then D†JA+ρ† BJB= 0,(45) −t†JA+D†JB= 0.(46) Proof. Apply the infinitesimal 1–gauge transformation δA = D , δB = .B to the action. Gauge invariance, together with the equations of motion in the form (44), give (after integrating by parts) 0 = δS=ZM κ0 , D†JA+κ0 , ρ† B(JB). Since is arbitrary, (45) follows. For 2–gauge transformations δηA = −t ( η ), δηB = Dη we similarly obtain 0 = δηS=ZM κ1 η, −t†(JA) + D†JB, which yields (46). Interpretation. Equations (45) – (46) state that 1–form and 2–form charges are not separately conserved when t and . are nontrivial; charge can flow between the two sectors through the crossed–module maps. In the fake–flat limit F = 0 (so t† ( F ) = 0) and with H = 0, the mixing terms vanish and the usual covariant conservations D†? JA= 0,D†? JB= 0 are recovered. Stress–energy tensor and its divergence We compute the Hilbert stress–energy tensor Tµν by varying the metric gµν in the kinetic part of the action; the topological sector (34) is metric–independent and contributes no bulk stress–energy.[ ? ] Using index notation with κ0,1to contract Lie algebra indices, we obtain Tµν A=1 g2κ0 Fµλ, Fνλ−1 4gµν κ0 Fαβ, Fαβ,(47) Tµν B=1 2h2κ1 Hµλσ, Hνλσ−1 6gµν κ1 Hαβγ, Hαβγ ,(48) Tµν =Tµν A+Tµν B+Tµν matt.(49) Note that F includes the −t ( B )piece; thus Tµν A already contains cross–terms in B , reflecting the fake–curvature mixing in the energy density.
19 The divergence of Tµν A obeys the standard Yang–Mills identity modified by the covariant derivative and the presence of F: ∇µTµν A=κ0 Fνλ,1 g2DµFµλ.(50) Similarly, for the 3–form kinetic term (see, e.g., [46–48]) ∇µTµν B=κ1 Hνλσ,1 h2DµHµλσ.(51) Substituting the field equations (44) into (50)–(51) yields ∇µTµν A=−κ0 Fνλ,JA,λ,(52) ∇µTµν B=−κ1 Hνλσ,JB,λσ.(53) Expanding JA,JB using (42) – (43) , we see explicitly the exchange of energy–momentum between the sectors mediated by the crossed–module maps: ∇µTµν A=−κ0 Fνλ, ?JA,λ−κ0 Fνλ,δStop δAλ−1 h2κ0 Fνλ, ρ† B(H)λ, ∇µTµν B=−κ1 Hνλσ, ?JB,λσ−κ1 Hνλσ,δStop δBλσ +1 g2κ1 Hνλσ, t†(F)λσ. The last terms in each line are equal in magnitude and opposite in sign once we use the adjoint relations (24) – (25) , so that ∇µ ( Tµν A + Tµν B )contains only matter and topological work terms. Including the matter stress–energy and using the matter equations of motion gives the standard diffeomorphism Ward identity ∇µTµν = 0,(54) i.e. total energy–momentum is covariantly conserved; the mixing induced by t ( B )and H merely redistributes it between the 1– and 2–form sectors. Physical content. Even in the absence of explicit matter currents, a nonzero fake curvature F (hence nontrivial t ( B )) and 3–curvature H cause the A and B sectors to exchange energy and momentum. This is the dynamical imprint of categorical (co)limit failure: the higher sector does physical work on the ordinary sector and vice versa. Component form in d= 3,4and decoupling checks For later use we record component forms in common dimensions (with κ0,1 taken as Tr for compact groups and 012··· = +1). In d= 4.Let Fµν ∈g0,Hµνρ ∈g1. Then 1 g2DµFµν =−Jν A,1 h2DµHµνρ =−Jνρ B,(55) Tµν A=1 g2Tr(FµλFνλ)−1 4gµνTr(FαβFαβ),(56) Tµν B=1 2h2Tr(HµλσHνλσ)−1 6gµνTr(HαβγHαβγ ).(57) The BF coupling (31) modifies (55) by the topological sources ∝k ( DB ) ν and ∝k Fνρ 0 and does not alter (56) in the bulk. In d = 3.Chern–Simons terms (32) shift the equations by ∝k0F0 and ∝k1A and, again, are metric–independent. The stress–energy is given by (47)–(48) with the obvious index ranges. Summary • The equations of motion of the higher–gauge theory can be written in a Maxwell form (44) with effective sources JA,JB that include matter, topological, and higher–sector mixing terms (42) – (43) .
20 • Gauge invariance enforces generalized continuity equations (45) – (46) , expressing the fact that 1– and 2–form charges exchange through the crossed–module structure maps. • The stress–energy tensors (47) – (48) are deformed by the fake curvature F = dA + A∧A−t ( B ); their divergences (52) – (53) show that energy–momentum is exchanged between the sectors in exactly the amounts dictated by the mixing terms, while the total Tµν is conserved. C. Comparison to existing theories This subsection situates the dynamical higher–gauge theory developed in §III A–§IIIB relative to familiar effective field theories. We give precise reductions to (i) Yang–Mills theory (fixed symmetry group, no higher sector), (ii) higher–form electrodynamics (Kalb–Ramond field), and (iii) topological theories of BF/Chern–Simons type; then we explain in what mathematical and physical sense our framework strictly generalizes all of them. A key distinction is the presence of the crossed–module maps t and . : when nontrivial, they encode the non–closure of symmetry composition (the obstruction measured by the curvatures Fand H) and couple the 1– and 2–form sectors dynamically. Reduction to Yang–Mills (fixed group, no higher sector) Set t = 0 and . = 0 in the crossed module ( g1 t −→ g0, . ), and decouple B (or equivalently set h−2 = 0 so the B–sector has no dynamics). Then F=dA +A∧A, H =dB, and the gauge transformations reduce to δA = D with δB = 0. The action (35) becomes ordinary Yang–Mills with matter: SYM[A] = 1 2g2ZM κ0(F∧?F) + ZM κ0(A∧?JA) + Stop[A]B=0,(58) and the equations of motion (44) reduce to 1 g2DF =−? JA−δStop δA .(59) The higher Bianchi identity (41) becomes the usual Yang–Mills Bianchi identity DF = 0. Thus: Proposition 3.3 (Yang–Mills limit). For t = . = 0 and B non–dynamical, the theory (35) – (44) reduces to Yang–Mills with the same compact structure group G0 and standard Ward identity D†? JA = 0. Physical reading. In this limit, symmetry composition closes strictly on G0 ; there is no obstruction to glue, and no higher sector to track it. Reduction to higher–form electrodynamics (Kalb–Ramond) Set A= 0 (or restrict to a topologically trivial 1–sector) and t= 0. Then F= 0, H =dB, and the action reduces to the free 2–form theory (Kalb–Ramond electrodynamics) SKR[B] = 1 2h2ZM κ1(H∧?H) + ZM κ1(B∧?JB),(60) with gauge symmetry B7→ B + d Λand equation of motion dH = −h2? JB . The conservation law is d†JB = 0 [ 49 ]. Hence the higher theory reproduces standard 2–form electromagnetism when the obstruction map tis trivial and the 1–sector is frozen.
21 Topological truncations: BF and Chern–Simons Topological theories are recovered by dropping the kinetic terms and retaining only metric–independent couplings. BF in d = 4.Take the abelian or nonabelian case with a cross–pairing β as in (31) and set g−2 = h−2= 0. Then SBF =k 2πZM β(B∧F0), F0=dA +A∧A, (61) is independent of the metric and yields the constraints F0 = 0 and DB = 0 (flatness), typical of a d = 4 TQFT [ 50 ]. In lattice formulations, SBF computes intersection numbers; continuum observables are Wilson loops/surfaces with area/perimeter law controlled by topology. Chern–Simons in d= 3.Set h−2= 0 and retain (32). The field equations enforce k0 4πF0+k1 2πB= 0,k1 2πA= 0, so that the gauge field is flat (modulo sources), and dynamics is purely topological [ 51 ]. Again, our theory reduces to a TQFT with quantized link invariants. Where the generalization is strict There are two intertwined directions in which (35) extends these theories: (A) Non–closure of symmetry composition and fake curvature. In Yang–Mills, group composition closes on G0and the curvature is F=dA +A∧A. In our theory, F=dA +A∧A−t(B), so any nontrivial t (originating from the 2–group/crossed–module structure) shifts the field strength by the 2–form gauge field. This fake curvature captures the categorical obstruction of §IIB: the difference between the strict colimit in Grp and the actual gluing. It is this term that allows energy, momentum, and charge to flow between the 1– and 2–form sectors, cf. (44) – (46) , a phenomenon absent in fixed–group theories. (B) Dynamical higher sector and topological mass without Higgs. Topological couplings such as BF in d = 4 and Chern–Simons in d = 3 are usually treated as pure TQFTs or as background interactions. In (35) they coexist with kinetic terms and with t –coupling, producing a genuine exchange between sectors and, in particular, topological mass generation for the propagating modes. In the abelian d = 4 case with t= 0, the quadratic action S=Z1 2g2F∧?F +1 2h2H∧?H +k 2πB∧F leads, after gauge fixing, to a mixed kinetic operator whose determinant has a massive pole at p2+m2= 0, m =k g 2π h,(62) with the remaining poles pure gauge [ 52 , 53 ]. Thus the 1– and 2–form excitations combine into a topologically massive multiplet without a Higgs scalar. Our framework generalizes this mechanism to nonabelian crossed modules with nonzero t and . , where the mixing is governed by the obstruction maps. Sketch of the derivation of (62) .In momentum space on R1,3 , work in Lorenz gauges d†A=d†B= 0. The quadratic form is 1 2g2A·(p2P(1) T)A+1 2h2B·(p2P(2) T)B+k 2πA·MB, where P(p) T projects onto transverse p –forms and M is the momentum–space mixing map implementing B∧F∼A· ( ?p ) ·B . The block determinant is det ( p2P(1) T ) det ( p2P(2) T−k2g2 4π2h2P(2) T ), giving the massive pole (62) and pure–gauge zeros; see [52] for a detailed computation.
22 Relation to gerbes and “fake–flatness” constraints In nonabelian gerbe theory, consistency of surface holonomy often imposes the fake–flatness condition F = 0 as a constraint, so that only H remains as a nontrivial curvature [ 54 ]. Our construction relaxes this: F is allowed and dynamical, measuring exactly the failure of strictness of the underlying categorical glue. The Bianchi identities (41) ensure that when F is nonzero, H is not covariantly closed (unless compensated), encoding the higher obstruction. Hence our theory contains nonabelian gerbes as a special case (impose F= 0 by a Lagrange multiplier), but is strictly more general and physically richer. Background vs. dynamical 2–group global symmetries Recent work uses background 2–group gauge fields to describe mixed anomalies and generalized global symmetries. In those settings the 2–group data ( t, . )are fixed and non–dynamical fields couple only as probes. In contrast, our fields ( A, B )are dynamical, and the crossed–module maps actively mediate charge/energy exchange via (44) – (46) . Moreover, in Section IV we will promote the structure ( t, . )itself to a field Φ, allowing the type of symmetry to evolve—a regime not accessible to fixed–background formulations. Comparison table Theory Curvatures Symmetry closure Dynamics / mass Yang–Mills F=dA +A∧Astrict on G0kinetic only (no BF/CS) Kalb–Ramond H=dB strict abelian kinetic for Bonly BF/CS (topological) constraints F0= 0, DB = 0 strict metric–independent, quantized This work F=dA +A∧A−t(B), H =dB +A.B non–strict via t, . kinetic + topological; BF mass, sector exchange Conclusion of §III D. Our dynamical higher–gauge theory contains Yang–Mills, Kalb–Ramond, and BF/CS as controlled corners but adds essential new structure precisely where condensed–matter systems require it: when local symmetries glue only up to higher equivalence. The obstruction to strictness becomes a field with its own curvature and dynamics, leading to novel conservation–law mixings and topological mass generation (§III B), and setting the stage for the symmetry–type dynamics of Section IV. D. Comparison to existing theories This subsection clarifies how the dynamical higher–gauge framework of Eqs. (35) – (44) relates to familiar effective field theories used in condensed matter and high–energy theory. We provide precise reductions to: (i) Yang–Mills theory (fixed symmetry group, no higher sector), (ii) higher–form electrodynamics (Kalb–Ramond two–form), and (iii) topological theories of BF/Chern–Simons type. We then show in what sense our construction strictly generalises these corners by accommodating the non–closure of symmetry composition through the crossed–module maps t and . , and by allowing kinetic and topological mixing between the 1– and 2–form sectors. Finally we present two complementary derivations — a diagonalisation of the quadratic action (topological mass generation) and a BRST analysis — that make the generalisation sharp. Yang–Mills as a strict–closure limit Set t = 0 and . = 0 in the crossed module ( g1 t −→ g0, . )and freeze the 2–form sector (or send h−2→ 0). Then F=dA +A∧A, H =dB,
23 and the action (35) reduces to SYM[A] = 1 2g2ZM κ0(F∧?F) + ZM κ0(A∧?JA) + Stop[A]B=0.(63) The equations of motion (44) become the standard Yang–Mills equations 1 g2DF =−? JA−δStop δA ,(64) and the Bianchi identity (41) reduces to DF = 0. The Ward identity (45) becomes D†? JA = 0. In this limit symmetry composition closes strictly on G0 ; no higher obstruction is present and the “fake” shift −t(B)in Fvanishes. Higher–form electrodynamics (Kalb–Ramond) as a decoupled limit If we instead set A = 0 (or restrict the 1–sector to be topologically trivial) and t = 0, then F = 0 and H=dB, and the action reduces to the free two–form theory SKR[B] = 1 2h2ZM κ1(H∧?H) + ZM κ1(B∧?JB).(65) Gauge invariance B7→ B + d Λand the equation dH = −h2? JB follow in the usual way, together with the conservation law d†JB = 0 (see, e.g., the canonical analysis in [ ? ]). The crucial difference to §III A is that no crossed–module mixing is present, so 1– and 2–form charges cannot exchange. Topological truncations: BF and Chern–Simons sectors Setting g−2 = h−2 = 0 in (35) eliminates metric dependence and leaves only the topological couplings (examples are given in (34)). Two standard cases are recovered. BF theory in d= 4.With a compatible cross–pairing βand F0=dA +A∧A, SBF =k 2πZM β(B∧F0) enforces the flatness constraints F0 = 0, DB = 0 and is a TQFT. This sector governs quantised linking of Wilson loops and Wilson surfaces; in lattice realisations it measures intersection numbers. Chern–Simons–like theory in d= 3.With CS(A) = κ0A∧dA +2 3A∧A∧A, SCS =k0 4πZCS(A) + k1 2πZβ(B∧A) is again metric–independent. Level quantisation of k0, k1 is required for invariance under large gauge transformations. Eliminating B constrains A to be flat (modulo sources), as in standard Chern–Simons theory. Why the present theory is strictly more general There are two structural extensions beyond the three limits above. (A) Non–closure of symmetry composition. In Yang–Mills one always has F = dA + A∧A . Here, the 2–group structure induces F=dA +A∧A−t(B), H =dB +A . B, so that −t ( B )measures the obstruction to strict 1–composition, and H measures the obstruction to higher coherence. These obstructions are not auxiliary: they are dynamical fields obeying the coupled equations (44) . Energy–momentum and charge exchange between sectors is controlled by t and . (cf. (45)–(46) and (52)–(53)).
24 (B) Topological mass generation with crossed–module mixing. The coexistence of kinetic terms and BF–type couplings yields topological masses for the propagating modes in a way that continuously deforms the BF mechanism to nonabelian crossed modules. We give two derivations (one by diagonalisation, one by dualisation), and we emphasise the role of tin the nonabelian case. Diagonalising the quadratic action (abelian d= 4) Consider the abelian sector with t = id , trivial . , on Minkowski M = R1,3 . The quadratic action (with Lorenz gauges ∂µAµ= 0,∂µBµν = 0) is S(2)[A, B] = 1 4g2ZFµνFµν +1 12h2ZHµνρHµνρ +k 4πZµνρσBµν ∂ρAσ,(66) with Fµν = ∂µAν−∂νAµ and Hµνρ = ∂[µBνρ] . Fourier transforming and writing the quadratic form in the transverse projectors P(1) T and P(2) T on 1– and 2–forms, the kinetic operator is block–off–diagonal due to the mixing. Integrating out Bgives the effective action Seff[A] = 1 4g2ZFµν ηµαηνβ +m2 Πµν,αβFαβ, where Πprojects onto the transverse two–form subspace and m=k g 2π h.(67) Thus the vector acquires a topological mass m while preserving gauge invariance; the physical spectrum is that of a massive spin–1 particle with three polarisations. Equivalently, dualising the 2–form B to a pseudoscalar ϕby Hµνρ =h2µνρσ∂σϕreduces (66) to a Stueckelberg–like action 1 4g2ZF2+h2 2Z(∂ϕ)2+k 2πZϕ µνρσ∂µFνρdxσ, demonstrating again that the U(1) gauge boson “eats” the shift symmetry of ϕ and becomes massive without a Higgs potential. This matches the well–known four–dimensional BF–mass mechanism, now embedded as a limit of our crossed–module dynamics. Nonabelian generalisation and the role of t In a nonabelian crossed module, the quadratic expansion of (35) around a background with small fields yields (schematically) L(2) ∼1 2g2A·P(1) TA+1 2h2B·P(2) TB+k 2πA·MB−1 2g2A·t(B), where the last term is the linearisation of −κ0 ( F, t ( B )) and is controlled by the Lie algebra map t:g1→g0 . The mass eigenvalues now depend on the representation of g1 under g0 and on the operator t†t . In particular, components in ker t do not contribute to fake curvature and remain massless in the absence of BF mixing, whereas components in im t mix strongly with A . This makes explicit how the categorical obstruction map t controls the low–energy spectrum — a phenomenon with no counterpart in fixed–group Yang–Mills. BRST structure and cohomological comparison Let c be the ghost 0–form for 1–gauge transformations and ξ the ghost 1–form for 2–gauge transformations (ghost–for–ghost). The BRST differential sacts on the fields by sA =Dc −t(ξ), sc =−1 2[c, c], sB =Dξ +c . B, sξ =−c . ξ,
25 and is nilpotent ( s2 = 0) if and only if the Peiffer identities of the crossed module hold (equivalently, the 2–group axioms). The curvatures transform as sF = [F, c], sH =c . H, so the gauge–invariant polynomials κ0 ( F∧?F )and κ1 ( H∧?H )are BRST–closed. In the strict Yang–Mills limit ( t = . = 0) the ξ –sector decouples and one recovers the standard BRST complex. In the pure BF/CS limits the cohomology collapses to the topological sector. This cohomological picture makes rigorous in which sense the present theory contains the known models as subcomplexes while adding a genuinely new ghost sector controlling 2–gauge symmetry and the obstruction maps; see [ ? ] for general BRST cohomology methods in reducible gauge systems. Conceptual summary • Yang–Mills corner: t = . = 0, B frozen. Strict symmetry closure, ordinary curvature F , standard Ward identities. No categorical obstruction. • Kalb–Ramond corner: A frozen, t = 0. Pure 2–form dynamics with conserved higher–form charge, no exchange with the 1–form sector. • BF/CS corner: kinetic terms off. Topological constraints F0 = 0, DB = 0, metric independence and quantised responses. • Present framework: t, . possibly nonzero, kinetic and topological terms on. Fake curvature F = dA + A∧A−t ( B )and H = dB + A . B are the dynamical measures of categorical (co)limit failure; they mediate exchange between sectors, produce topological masses, and reduce to all corners above in controlled limits. IV. MAKING THE SYMMETRY TYPE DYNAMICAL A. Field of structures In Sections II–III we fixed a crossed module ( g1 t −→ g0, . )(equivalently a strict Lie 2–group) and built dynamics for a 2–connection ( A, B )with curvatures F = dA + A∧A−t ( B ) , H = dB + A . B . We now promote the symmetry type itself to a dynamical field. Concretely, we introduce a field of structures Φ: M−→ M2-grp,(68) where M2-grp denotes the (stacky) moduli space of crossed modules on ( g1,g0 ). At each spacetime point x∈M, the value Φ(x)selects structure maps and higher cohomology data tΦ(x):g1→g0, .Φ(x):g0yg1, ω2,3 Φ(x),(69) subject to the Peiffer identities and coherence conditions at that point. This section develops the differential geometry of Φ, derives the induced variations of the curvatures FΦ, HΦ, formulates a sigma– model kinetic term on M2-grp , and obtains the Φ–equations of motion coupled to ( A, B ), including the generalised Bianchi identities in the presence of a space–time dependent structure. Model for the moduli and constraints. For definiteness we keep the underlying vector spaces g0,g1 and their Lie brackets [ ·,· ] 0 ,[ ·,· ] 1 fixed (the most common situation in applications). The structure space at a point is then Str := (t, ρ)t:g1→g0, ρ:g0→Der(g1)satisfy Peiffer identities×Coh,(70) where ρ ( X ) ·Y≡X . Y and Coh encodes admissible 2– and 3–cocycles ω2, ω3 (when these are present as background/topological data). The Peiffer identities read t ρ(X)Y= [X, t(Y)]0, X ∈g0, Y ∈g1,(71) ρ t(Y)Y0= [Y, Y 0]1, Y, Y 0∈g1,(72)
32 Coupled Euler–Lagrange equations in “wave–map” form Equations (91) – (93) of §IVB already provide a convenient first–order (in variations) form. Here we recast the Φ–equation into a wave–map form, making explicit the covariant d’Alembertian on the moduli M2-grp and the curvature sources. Definition 4.5 (Tension field). The tension field of Φ: (M, g)→(M2-grp, G)is τα(Φ) := ∇µ∂µΦα+ Γαβγ(Φ) ∂µΦβ∂µΦγ, where Γis the Levi–Civita connection of G and ∇ is the Levi–Civita connection of g acting on target indices through Γ. Proposition 4.6 (Wave–map EOM with curvature sources). The Φ–equation (93) can be written as κ τα(Φ) = −∂αV(Φ) + Sα[FΦ, HΦ;Φ] + ∂αLtop(Φ; A, B) + ∂αLmatt(Φ),(101) with Sα[FΦ, HΦ;Φ] = 1 g2DFΦ,(∂αtΦ)(B)E0−1 h2DHΦ,(∂αρΦ)(A, B)E1.(102) Proof. This is the coordinate–free rewriting of (93) ; the right–hand side collects the Φ–derivatives of the structure maps appearing in the kinetic sector plus any explicit Φ–dependence of the topological and matter couplings. Interpretation. Equation (101) exhibits the promised feedback loop: curvatures (FΦ, HΦ)⇒force on Φ⇒change of (tΦ, ρΦ)⇒change of curvatures When FΦ and HΦ are large in norm, the source Sα pushes Φin the direction that reduces the fake curvature and the 3–curvature by changing tΦand ρΦ. Gauge–field equations with Φ–dependent sources. The coupled gauge equations (restating (91) – (92) ) are 1 g2DFΦ+1 h2ρ† Φ,B(HΦ) + δLtop δA =−? JA(Φ; Ψ),(103) −1 g2t† Φ(FΦ) + 1 h2DHΦ+δLtop δB =−? JB(Φ; Ψ).(104) The mixing maps t† Φ and ρ† Φ,B depend on Φand mediate bidirectional exchange between the 1– and 2–form sectors. Hyperbolicity and gauges On a globally hyperbolic (M, g), impose Lorenz–type gauges DµAµ= 0, DµBµν = 0, and choose local coordinates on M2-grp for which the Christoffel symbols Γ αβγ are bounded on the region explored by Φ. Then the principal parts of (103)–(104)–(101) are 1 g2gAν+··· =sources,1 h2gBµν +··· =sources, κ gΦα+··· =sources, where g := ∇µ∇µ is the geometric d’Alembertian and the ellipses denote lower–order terms (at most first order in derivatives of the fields). Thus, modulo the usual Yang–Mills constraints, the coupled system is a quasilinear hyperbolic system; standard results for wave–maps and Yang–Mills–type systems give local well–posedness for smooth initial data [68, 69].
33 Linearisation about a homogeneous background Let ( A0, B0, Φ 0 )be a stationary background solving the static versions of (103) – (101) . Write fluctuations A=A0+a, B =B0+b, Φ=Φ0+ϕ, and denote by t0 := tΦ0 , ρ0 := ρΦ0 and by t,α , ρ,α the derivatives of the structure maps at Φ 0 . To linear order the curvature variations are δFΦ=D0a−t0(b)−t,α(B0)ϕα,(105) δHΦ=D0b+a .0B0+ρ,α(A0, B0)ϕα,(106) with D0 and .0 the covariant derivative and action at Φ 0 . In a background Lorenz gauge Dµ 0aµ = 0, Dµ 0bµν = 0, the linearised equations become 1 g2D0D0a+1 h2ρ† 0,B0(D0b) + M(A)a+K(A↔B)b=J(A) αϕα,(107) −1 g2t† 0(D0a) + 1 h2D0D0b+M(B)b+K(B↔A)a=eJ(B) αϕα,(108) κgϕα+∇2Vα βϕβ=Sα (1)[a, b] + Sα (0) ,(109) where M (·) and K (·) are background–dependent (Proca–type and mixing) operators coming from expanding the kinetic and topological sectors, while J(A) αϕα=1 g2D0 t,α(B0)ϕα−1 h2ρ† 0,B0ρ,α(A0, B0)ϕα, and similarly for eJ(B) α. The Φ–source splits into a part linear in (a, b), Sα (1)[a, b] = 1 g2DD0a−t0(b), t,α(B0)E0−1 h2DD0b+ (a .0B0), ρ,α(A0, B0)E1, and a background piece Sα (0) =−1 g2DF0, t,αβ(B0)ϕβE0+1 h2DH0, ρ,αβ(A0, B0)ϕβE1, where F0, H0 are background curvatures and commas on t,αβ denote second derivatives on the moduli. Equations (107) – (109) exhibit the back–reaction: fluctuations of Φsource the gauge sector and vice versa. In a homogeneous background with F0 = H0 = 0 (fake–flat and 3–flat), the leading Φ–mass matrix is ∇2Vand the mixing occurs only through Sα (1). Energy identity and feedback Define the (positive) energy densities EA=1 2g2κ0(FΦ, FΦ)(t),EB=1 2h2κ1(HΦ, HΦ)(t),EΦ=κ 2Gαβ ∂0Φα∂0Φβ+V(Φ), where the subscript ( t )indicates contraction with the spatial metric on a t = const slice (we suppress the matter and topological contributions for clarity). Using (103) – (104) – (101) and the Bianchi identities, a standard computation (contract, integrate by parts, and use D = ±? D? ) yields on a compact Cauchy slice Σt d dtZΣt (EA+EB+EΦ) = −ZΣt κ0 Ei,JA,i−ZΣt κ1 Hi,JB,i+ZΣt∂0ΦαSα,(110) with Ei and Hi the electric components of FΦ and HΦ , and JA,B the total currents of §IV B. The last term is the feedback: if curvature sources point along −∂0 Φ, the energy in the structure sector decreases, transferring to the gauge sector or to damping (if present). Physical upshot. Equation (110) shows explicitly that curvature acts as a generalised conjugate force to the structure variable: ( FΦ, HΦ )do work on Φ, pushing it toward symmetry types that reduce the obstruction energy.
34 Boundary conditions On a manifold with boundary ∂M , variation of (88) yields surface terms which vanish under natural boundary conditions, e.g. nµFΦ,µν∂M = 0,nµHΦ,µνρ∂M = 0, nµ∂µΦα ∂M = 0, (Neumann–type), or their Dirichlet counterparts A|∂M , B|∂M ,Φ |∂M fixed. Mixed conditions are possible and, in condensed–matter settings, encode physical interfaces (e.g. a domain wall where Φjumps), cf. the discussion after (86)–(87). Example: scalar structure parameter and explicit wave equation For the toy model of §IV A with a single scalar parameter α (Φ) controlling a central extension (cf. (82)), the structure source is S[α] = α0(Φ) g2DFΦ, ι(B)E0−∂Ltop ∂α (α;A, B), and the wave equation (101) reduces to κgα=−dV dα +α0(Φ) g2DFΦ, ι(B)E0−∂Ltop ∂α .(111) In homogeneous situations the curvature overlap drives α to the minimum of an effective potential Veff ( α ) = V ( α ) −1 g2αhF, ι ( B ) i0−Ltop ( α ), realising the intuitive picture that the system “learns” its symmetry type. Summary of §IV C • The structure field Φobeys a wave–map equation (101) with a source Sα [ FΦ, HΦ ;Φ] given by the overlap of the obstruction curvatures with the Φ–gradients of the structure maps. • The gauge fields ( A, B )satisfy Φ–dependent Maxwell–like equations (103) – (104) ; their mixing operators are controlled by tΦand ρΦ. • Linearising about a background yields a block–coupled hyperbolic system (107) – (109) that makes back–reaction explicit and provides the starting point for spectral and stability analyses in Sections 5– 6. • An energy identity (110) quantifies the feedback loop: curvature sources do work on Φ, driving symmetry–type transitions when Veff has multiple minima. V. CONDENSED–MATTER APPLICATIONS A. Adaptive U(1)↔Z2spin–liquid transition We now develop a concrete application of the dynamical symmetry–type framework to two–dimensional quantum magnets where an emergent compact U(1) gauge structure adapts into a Z2 spin liquid. We work with half–filled Mott insulators on frustrated lattices (kagome and triangular), derive a continuum effective action for spinons coupled to a 2–connection ( A, B )and a structure field Φ, and compute the phase diagram and observables. Our analysis interpolates between a gapless U(1) state (Dirac or Fermi surface spinons coupled to a compact gauge field) and a gapped Z2 state obtained by charge–2 Higgsing of the emergent U(1), in which the vison is the π –flux of the residual Z2 gauge sector. The adaptive element is that the strength of the extension map tΦ is itself a dynamical field controlled by Φ(Section IV), so that gauge structure selection is an output of the theory.
35 Microscopic starting point. We consider the Hubbard model at half filling H=−tX hiji,σ c† iσcjσ +UX i ni↑ni↓+··· on the kagome or triangular lattice. In the Mott regime Ut , a strong–coupling expansion produces a spin model with Heisenberg exchange J∼ 4 t2/U , ring exchanges, and further–neighbor couplings [ 70 , 71 ]. Numerics and experiments indicate proximate spin–liquid behavior both on the kagome lattice [ 72 , 73 ] and on organic triangular salts [ 74 , 75 ]. In parton mean–field language one writes Si = 1 2f† iασαβfiβ with single–occupancy and an emergent compact U(1) gauge redundancy; depending on the ansatz, low–energy spinons are Dirac fermions (kagome) [ 76 ] or possess a Fermi surface (triangular) [ 70 ]. Pairing of charge–2 matter (spinon pairs or Schwinger bosons) Higgses U(1) →Z2, producing a gapped Z2spin liquid [77]. Identification of (A, B, Φ) and the double–well V(Φ) We take the continuum limit of Section IV. The 1–form A gauges the emergent U(1); the 2–form B captures the higher (vison) sector; and the structure field Φselects the instantaneous crossed–module data. For concreteness we adopt the toy family (central extension controlled by one scalar) from §IV A: tΦ=α(Φ)ι, ι:u(1) ,→u(1), ρΦ≡ρ0fixed,(112) so that α(Φ) controls the charge–2 Higgsing strength. The curvatures are FΦ=dA −α(Φ)B, HΦ=dB, and the minimal gauge Lagrangian (in 2+1 dimensions) is Lgauge =1 2g2|FΦ|2+1 2h2|HΦ|2+k 2πB∧dA, (113) where the B∧dA mixing is the 2+1D BF term (topological mass mechanism). Time–reversal allows k = 0, in which case the mass arises entirely from the tΦ mixing (Higgs). The structure sector is a sigma model κ 2(∂Φ)2+V(Φ) with a double well V(Φ) = r 2Φ2+u 4Φ4, u > 0, r =r(λ)(114) that encodes material control (pressure p , strain , stoichiometry x collectively into λ ). The minima Φ = ± Φ 0 represent distinct symmetry types: one with α (Φ 0 ) 6 =0 ( Z2 state), one with α ( − Φ 0 ) ≈ 0(U(1) state).[?] Spinon sector and coupling to (A, B) For a Dirac U(1) state on kagome, the spinon Lagrangian is Lsp =¯ ψγµ(∂µ−iAµ)ψ+ ∆(Φ)ψTiσyψ+ h.c.+··· , with ∆(Φ) the pairing amplitude that tracks α (Φ) (∆ ∝α at mean field). On the triangular lattice the spinon Fermi surface case is similar, but the low– T thermodynamics differs. In both cases integrating out gapped pairs in the Z2 phase generates the BF term in (113) and, on breaking T , a Chern–Simons term for A[78] (we set k0=0 for T–symmetric cases). Phase diagram from minimisation of the action At Gaussian level, integrating out ( A, B )in (113) around a spatially uniform Φgives an effective potential for Φ, Veff(Φ) = V(Φ) + 1 2Zd3p (2π)3hlog p2+m2 γ(Φ)+ log p2+m2 B(Φ)−2log(p2)i,(115)
36 where the last term comes from gauge fixing/ghosts and m2 γ(Φ) = m2 t(Φ) + m2 BF, mt(Φ) := g h|α(Φ)|, mBF := k g 2π h.(116) In 2+1D, Rd3plog(p2+m2) = m3/(6π)(up to a constant), so Veff(Φ) = r 2Φ2+u 4Φ4+1 12πm3 γ(Φ) + m3 B(Φ)+··· ,(117) with mB (Φ) the mass of the dual 2–form mode (degenerate with mγ in the BF limit). The gauge fluctuations shift the quadratic coefficient, reff =r+c1α0(0)2g3 πh3m?+··· ,(118) where m? is the mass scale set by mBF or by the infrared regulator, and c1> 0. Thus, even when the bare potential V (Φ) has r > 0(favoring U(1)), gauge fluctuations can drive reff negative, selecting the Z2 minimum — a fluctuation–induced first– or second–order transition depending on u and higher terms (cf. the Coleman–Weinberg mechanism). The phase boundary is given implicitly by reff ( λ )=0with λ the control parameter; its slope with respect to kor g/h is determined by (118). Emergent “photon” gap and vison mass In the Z2phase, the gauge boson acquires a mass mγ(Φ0) = rg h|α(Φ0)|2+k g 2πh2,(119) where the second piece is present only if the BF mixing is nonzero. The vison is the π –flux excitation of the Z2 gauge sector; in the continuum it is a point excitation whose worldline couples minimally to B . A crude estimate of its gap follows from the energy cost of a localized HΦ flux tube of core size ξB∼ 1 /mB : mv∼1 2h2Zcore |HΦ|2∼µB h2mB∝g h2|α(Φ0)|,(120) where µB is a numerical constant fixed by the core profile. On approaching the transition, mγ→ 0and mv→0continuously if it is second–order; in a weakly first–order case both show a small jump. Thermodynamics: specific heat and crossovers Thermal signatures depend on dimensionality and on the spinon spectrum. Gauge contribution. A linearly dispersing boson in d spatial dimensions gives Cbos ( T ) ∝Td at Tmγ replaced by an activated form ∝e−mγ/T when gapped. Thus: • In quasi–2D magnets (kagome/triangular), a gapless U(1) mode contributes CA ( T ) ∝T2 ; in the Z2 phase it crosses over to CA(T)∝e−mγ/T . •In genuine 3D U(1) spin liquids, CA(T)∝T3[79]. Because our microscopic examples are 2D lattices, the predicted quasi–2D crossover is T2→e−mγ/T . If interlayer coupling makes the system effectively 3D at low T, the crossover becomes T3→e−mγ/T . Spinon contribution. For Dirac spinons (kagome), Cψ ( T ) ∝T2 ; pairing in the Z2 phase opens a gap ∆ so that Cψ ( T ) ∝e−∆/T . For a spinon Fermi surface (triangular), gauge fluctuations yield C ( T ) ∼γT with non–Fermi–liquid corrections; pairing again leads to activated behavior [82]. Total C(T)prediction. In the U(1) regime: C(T)≃a1T2+a2Tη+Cph(T), η ∈(1,2) from gauge–spinon corrections, while in the Z2regime C(T)≃b1e−mγ/T +b2e−∆/T +Cph(T), where Cph is the phonon background (dimension–dependent Debye law). Extracting ( mγ, ∆) from Arrhenius plots across the phase boundary gives a direct test of (119) and the Φ–driven selection.
37 Neutron scattering continuum: reweighting across the transition The dynamical spin structure factor S(q, ω)at low energies is dominated by two channels: 1. Spinon pair creation ψ¯ ψ : a broad continuum with threshold set by the spinon gap. In a Dirac U(1) state this extends down to ω→ 0at special momenta; pairing in the Z2 state opens a threshold 2∆, pushing spectral weight to higher ω. 2. Gauge fluctuations through current operators: in the U(1) regime these produce a low– ω tail; in the Z2regime they are suppressed by mγ. A simple RPA evaluation with current vertex J=¯ ψγψgives, in the Dirac case, S(q, ω)∼ = Πψψ(q, ω) 1−gRPA Πψψ(q, ω) with Π ψψ the Dirac bubble in 2+1D and gRPA an effective coupling that decreases as mγ grows. The prediction is a clear reweighting of the continuum: suppression of subgap weight and an onset near 2∆ upon entering the Z2 phase, consistent with DMRG/DMFT trends seen near kagome spin liquids [ 72 , 73 ]. Thermal Hall step from a mixed ω3(Φ)–CS term When time–reversal symmetry is (weakly) broken, integrating out gapped spinons induces a Chern– Simons term for A with level k0 and a Wess–Zumino coupling of the associator class ω3 (Φ) to dCS ( A ) (Section II): Ltop ⊃k0 4πCS(A)+2πiω3(Φ),1 2πdCS(A).(121) If Φtunnels between two symmetry types whose associator classes differ by an integer ∆ hω3i = n∈Z , the effective CS level jumps by ∆ k0 = n . The chiral central charge c− on the edge then jumps by n , producing a quantised change of thermal Hall conductance ∆κxy T=π2k2 B 3h∆c−=π2k2 B 3hn, (122) observable as a step when tuning λ across the Φdomain wall [ 80 , 81 ]. This effect is robust to microscopic details because it is topological (anomaly inflow). Phase diagram and experimental protocol Phase diagram. Combine (117) with (114) . For fixed ( g, h, k ), the phase boundary reff ( λ )=0gives λc . The slope ∂λc/∂k > 0(BF mixing favours Z2 ), while ∂λc/∂ ( g/h ) < 0(strong tΦ mixing favours Z2 ). Close to λc, mγ∼ |λ−λc|ν,∆∼ |λ−λc|ν∆, with mean–field exponents ν = ν∆ = 1 2 corrected by fluctuations. In a first–order scenario both masses jump weakly. Measurements. • Specific heat: Track C ( T )across λc ; fit low– T data to T2 or activated forms to extract mγ and ∆. • Neutron scattering: Measure S ( q, ω )at the Dirac nodes / M–points; look for suppression of subgap weight and an onset at 2∆. • Thermal Hall: In weak T –breaking (field tilt), search for a step of magnitude (122) when crossing λc.
38 Consistency checks and caveats Compact U(1) in 2+1D is confining in the absence of dynamical matter (Polyakov). Here, gapless spinons in the U(1) regime suppress monopoles at intermediate scales; the transition to Z2 corresponds to condensation of charge–2 matter rather than monopole proliferation. In strictly T –symmetric cases k = 0 so masses arise via tΦ ; in weak T –breaking, k6 = 0 adds a topological mass. Our continuum estimates (119–120) capture universal scaling but not the microscopic prefactors, which are material dependent. Summary of predictions 1. Adaptive selection of symmetry type: gauge fluctuations shift r→reff , moving the phase boundary and allowing U(1)→Z2under tuning. 2. Gaps: mγgiven by (119), vison mass scales as in (120). 3. Thermodynamics: C(T)crosses from T2(or T3in 3D) to activated behavior. 4. Spectroscopy: reweighting of the neutron continuum with an onset at 2∆. 5. Transport: quantised step ∆(κxy/T )=(π2k2 B/3h)nwhen ω3(Φ) changes by n. B. Fracton order as a non–associative limit Fracton phases exhibit subdimensional excitations and mobility constraints that do not fit within ordinary gauge theory. In this subsection we show how they arise naturally as a non–associative limit of the categorical framework developed earlier: local symmetry data glue only along submanifolds, and their compositions fail to associate. The resulting obstruction is encoded by the 3–curvature H , and the kinematics reduces to a higher–rank tensor gauge theory with tensor “photons” of quadratic dispersion ω∝k2 . We derive the mobility constraints from a covariant Gauss law that follows from the higher Bianchi identity DH = 0, and we indicate how a change of symmetry type Φcan restore associativity, driving a transition to a mobile phase (fracton condensation). From subsystem symmetries to non–associative glue. Consider a three–dimensional system endowed with submanifold (planar) symmetries: on each family of parallel planes (a foliation) we have a local group Gi that acts only on degrees of freedom supported on that plane. When composing symmetry operations supported on intersecting planes, the result depends on the order of composition; the obstruction to associativity localizes near the line of intersection and is recorded by a 3–cocycle (the “associator”). In the continuum limit, this is precisely the 3–curvature H of our 2–connection ( A, B )(cf. (9) ). We shall work in the abelian regime, set A≡ 0(no ordinary 1–form symmetry in the low–energy sector) and retain the 2–form gauge field Bwith curvature H=dB ∈Ω3(M),(123) and restricted (subsystem) gauge transformations B7→ B+dΛ,Λ∈Ω1(M)supported along foliations. (124) The restriction of Λto foliation one–forms implements the fact that only surface–attached deformations are allowed; this is a geometric way to encode subsystem symmetry. Dictionary to a symmetric rank–2 tensor gauge field To make contact with fracton tensor gauge theories [ 83 , 84 ], we now repackage the restricted 2–form B into a symmetric rank–2 tensor potential Aij . Choose three independent (local) foliation one–forms e(a) = e(a) idxi , a = 1 , 2 , 3, whose kernels integrate to planar layers.[ ? ] Restrict B to the foliation sector by writing B= 3 X a=1 e(a)∧C(a), C(a)=C(a) idxi.(125)
39 Define a symmetric tensor potential by Aij := 1 2 3 X a=1 e(a) iC(a) j+e(a) jC(a) i(so Aij =Aji).(126) The subsystem gauge transformation Λ = Pae(a)α(a)induces δC(a)=dα(a)and therefore δAij =∂i∂jα, α := 1 2X a e(a) ke(a) kα(a).(127) Thus the emergent low–energy gauge invariance is precisely the scalar–charge symmetry of higher–rank electrodynamics: Aij 7→ Aij +∂i∂jα[83]. Introduce a temporal scalar potential φthat transforms as φ7→ φ+∂tα, (128) so the electric tensor Eij is the gauge–invariant combination Eij := ∂tAij −∂i∂jφ. (129) Gauge invariance requires the magnetic tensor to be invariant under ∂i∂jα ; a minimal choice is the double curl Bij := iab jcd ∂a∂cAbd,(130) which is symmetric and annihilates pure gauge deformations as iab∂a∂bα = 0. In this dictionary, the components of H = dB (with B the 2–form) are linearly related to the “electric” components Eij and spatial double curls Bij. Fracton tensor electrodynamics from the 2–form sector The most general quadratic, rotation–invariant, α –gauge–invariant Lagrangian density built from Eij and Bij is Lfrac =1 2g2EijEij −c2 2g2BijBij +AijJij −φ ρ, (131) with implicit spatial sums, a stiffness g > 0, a characteristic velocity c , and couplings to a symmetric tensor current Jij and scalar charge density ρ. Varying with respect to φyields the Gauss law ∂i∂jEij =ρ, (132) while variation with respect to Aij gives the dynamical equation 1 g2∂tEij +c2 g2iab jcd ∂a∂cBbd =−Jij.(133) Using (129) – (130) and working in the (scalar–charge) Coulomb gauge ∂i∂jAij = 0, φ = 0, plane waves Aij ∝ei(k·x−ωt)obey ω2=c2k4for transverse, trace–free modes. (134) This quadratic (dynamical exponent z = 2) dispersion is the hallmark of fracton tensor photons [ 83 ]. The number of propagating polarizations equals the number of symmetric, trace–free components of Aij orthogonal (in the double–curl sense) to pure gauges; for our purposes the important point is that they are tensorial and subluminal.
40 Mobility constraints from DH = 0 In the abelian sector with A≡0, the higher Bianchi identity reduces to DH =ddB = 0 ⇐⇒ ∂µHµνρ = 0,(135) in the absence of 2–form sources (cf. (44) with JB = 0). Identify the electric tensor with Eij := H0ij and the spatial components with the double curls of Aij (this identification follows from the dictionary (125)–(130)). Then the time–component of (135) reads ∂t(∂i∂jEij) + ∂``ab icd ∂a∂c∂jAbd= 0.(136) By an identity of derivatives and epsilons (symmetry under exchange of the two curls) the second term is a total derivative that vanishes under suitable boundary conditions; hence, with ρ = ∂i∂jEij from (132) , we obtain the continuity equation ∂tρ+∂i∂jJij = 0,(137) with Jij defined by (133). Equation (137) implies two conservation laws: d dt ZR3 ρ d3x= 0,(138) d dt ZR3 xkρ d3x=−ZR3 xk∂i∂jJij d3x=−I∞ (xk∂jJij −δkiJij nj)dS = 0,(139) i.e. conservation of total charge and of the total dipole moment. As is standard [ 84 , 85 ], dipole conservation implies that an isolated charge (fracton) cannot move without changing the dipole moment; thus it is immobile. Bound dipoles can move only transversely to their dipole moment, as follows from (137) with ρ= 0 and Jij =∂(ivj)a pure gradient current. Summary so far. Restricting the 2–form gauge symmetry to foliation–supported transformations produces, via (125) – (126) , the scalar–charge tensor gauge theory (131) . The higher Bianchi identity DH = 0 yields the continuity law (137) responsible for fracton immobility and dipole conservation. The tensor “photon” has quadratic dispersion (134). Associativity restoration by Φand the mobility transition Within our dynamical framework (Section IV), the type of symmetry is controlled by the structure field Φ. In the fracton regime, the associator class encoded by H is nontrivial and the gauge transformations are of scalar–charge type Aij 7→Aij +∂i∂jα. There are two ways for mobility to be restored: (i) Dipole condensation. Introduce a dipole field Ψ k minimally coupled to Aij via Lmatt ⊃ Ψ † k (i ∂t− Akk )Ψ k−1 2mΨ| ( ∇i−Aik )Ψ k|2 + ··· . If Ψ k condenses, the effective gauge invariance enlarges to δAij = ∂iλj + ∂jλi (vector–charge theory) and the Gauss law becomes ∂iEij = ρj , allowing lineon mobility along λ . In our language, this corresponds to a Φ–flow that changes the action map .Φ so that higher corrections glue along lines instead of points (partial restoration of associativity). (ii) Full associativity restoration. If Φflows to a regime where the associator vanishes (i.e. ω3 (Φ) = 0 in the sense of Section IIC), the gauge structure reduces to an ordinary U(1) 1–form theory; charges become mobile and the gauge photon dispersion crosses over to linear. An effective Lagrangian that interpolates between the two regimes is Lint =1 2g2E2 ij −1 2g2c2 2BijBij +ε(Φ) c2 1(∂kAij)(∂kAij)+··· ,(140) with ε (Φ) a nonnegative function that vanishes in the fracton limit and becomes positive as associativity is restored.[?] Plane waves then satisfy ω2(k) = c2 2k4+ε(Φ)c2 1k2,(141) exhibiting a crossover scale k×∼ε (Φ) c2 1/c2 2 : for kk× the dispersion is fractonic ( ω∼c2k2 ), while for kk×it is linear (ω∼√ε c1k).
41 Relation back to DH = 0 and the categorical picture In the non–associative regime the obstruction 3–form H is the relevant curvature; mobility constraints follow from DH = 0. As Φrestores associativity, the obstruction class vanishes ( H→ 0in cohomology), and the gauge structure reverts to ordinary 1–form dynamics in which the relevant Gauss law is ∂iEi = ρ and the continuity equation is ∂tρ + ∂iJi = 0, allowing mobility. Thus the Φ–driven transition is precisely the passage from a non–associative (2–group–valued) glue to an associative (group–valued) one. Remarks on elasticity duality and foliations The mapping above dovetails with the duality between fracton tensor gauge theory and crystal elasticity [ 86 , 87 ]: the rank–2 gauge potential Aij maps to the linearized strain tensor, Eij to stress, and the Gauss law to Burgers/Frank constraints; immobile disclinations are the fractons. In our language, the foliation one–forms e(a) play the role of the crystalline axes, and the associator curvature H measures the failure of glide/tilt compositions to associate. Summary of §V B • Subsystem (planar) symmetries force a non–associative gluing of local symmetry data; the obstruction is the 3–curvature H=dB. • Restricting the 2–form sector to foliation–supported gauge transformations produces a symmetric rank–2 tensor gauge theory with gauge invariance Aij 7→ Aij + ∂i∂jα , electric tensor Eij and magnetic tensor Bij given by a double curl. • The higher Bianchi identity DH = 0 implies the continuity equation ∂tρ + ∂i∂jJij = 0, which in turn enforces dipole conservation and fracton immobility. Tensor photons have ω∼k2. • A change of symmetry type Φcan restore associativity, allowing a linear term ∝ ( ∂A ) 2 and driving a crossover ω2=c2 2k4+ε(Φ)c2 1k2, with concomitant melting of fracton constraints. C. Defects and domain walls in Φ Spatial textures of the structure field Φ :M→ M2-grp are boundaries between symmetry types: they separate regions where the crossed–module data ( tΦ, .Φ, ω2,3 (Φ)) take different values. In this subsection we develop a systematic theory of such defects (domain walls, vortices), compute the induced sources for the gauge sector using the Φ–modified Bianchi identities (86) – (87) , derive bound–state spectra localised on Φdomain walls, and obtain experimental signatures in tunnelling and local density of states. Static Φdomain walls: profile and tension For clarity we consider a single real modulus (the discussion extends componentwise). Let LΦ=κ 2(∂µΦ)2+V(Φ), V (Φ) = u 4Φ2−v22, so that the minima are Φ = ±v . A static wall located at x = 0 (normal ˆx ) obeys the Euler–Lagrange equation κ Φ 00 ( x ) = ∂ΦV (Φ) with boundary conditions Φ( −∞ ) = −v ,Φ(+ ∞ ) = + v . Completing the square (Bogomolny trick) yields the first–order system dΦ dx =±ru 2κv2−Φ2,(142) whose solution is the familiar kink/antikink Φwall(x) = vtanhx−x0 ξ, ξ =√2κ v√u,(143)
48 Discrete Bianchi identity. Let τ be a tetrahedron with oriented boundary faces f1, . . . , f4 , and denote by FΦ(fi)the fake curvature on fitransported to v(τ)by conjugation with Πv(τ)→v(fi). Then: Lemma VII.1 (Bianchi identity on a tetrahedron).For each tetrahedron τ, Y f⊂∂τ FΦ(f)σ(f,τ)=tΦv(τ)HΦ(τ)∈G0.(173) Proof. Expand each FΦ ( f )as U∂f t ( Wf ) −1 and transport to v ( τ ). The product of the transported U∂f around ∂τ cancels to the identity by oriented edge pairing. The remaining product is the image under t of the ordered product of the transported Wσ(f,τ) f , i.e. t ( HΦ ( τ )), using t -equivariance under conjugation and the Peiffer identity. Remark VII.2.On a 4–simplex σ4 the second Bianchi identity reads Qτ⊂∂σ4HΦ ( τ ) σ(τ,σ4) = 1 after parallel transport to a common basepoint, the discrete analogue of DHΦ−FΦ. B = 0. C. Lattice action and continuum limit We define a gauge–invariant lattice action that reduces to (88) as the lattice spacing a→ 0. Let Tr0 (resp. Tr1 ) be invariant, positive–definite class functions on G0 (resp. G1 ), normalised so that Tr0(1) = Tr1(1) = dim of the defining representation. Set SL[U, W, Φ] = βFX f∈K2 s0 FΦ(f)+βHX τ∈K3 s1 HΦ(τ) +κ 2X `∈K1 GαβΦm(`)Φα t(`)−Φα s(`)Φβ t(`)−Φβ s(`)+X p∈K0 V(Φp) +Stop L[U, W;Φ] + Smatt L[U, W;Φ],(174) where m(`)is the midpoint of `, and s0(g) := 1 −1 Tr0(1) <Tr0(g), s1(h) := 1 −1 Tr1(1) <Tr1 h,(175) are standard Wilson/Villain–type penalties. The Φ–gradient uses a nearest–neighbour discretisation with metric Gαβ evaluated at the midpoint, ensuring positivity. Topological terms. In d= 4, a BF coupling discretises to SBF L=k 2πX c∈K4DB(c),F0(c)E,(176) where B ( c )is the cochain obtained by averaging Wf over the six faces of the boundary of c and parallel transporting to a vertex, F0 ( c )is the sum of the plaquette holonomies U∂f on the same boundary (Whitney cochains/cup products give a systematic discretisation [ 99 , 100 ]), and h·,·i pairs G1 and G0 via t . In d = 3, Chern–Simons terms can be discretised by compact lattice CS actions [ 101 , 102 ], or via doubled BF (for abelian sectors). Continuum limit. Let the lattice spacing be a . Parametrise U` = exp{a Aµ ( x` ) } , Wf = exp{a2Bµν ( xf ) } with x`(resp. xf) a point on `(resp. f). A standard Baker–Campbell–Hausdorff expansion yields FΦ(f) = expna2Fµν −tΦBµν+O(a3)o, HΦ(τ) = expna3Hµνρ+O(a4)o,(177) so that SL→S in (88) provided βF = Tr0(1) 2g2ad−4 , βH = Tr1(1) 2h2ad−6 and κ rescales with ad−2 , as expected. D. Monte Carlo updates and ergodicity We sample the Boltzmann weight exp{−SL [ U, W, Φ] } with a sequence of local and gauge–covariant updates that alternate over (U, W, Φ). The basic ingredients are:
49 (A) Link updates (Metropolis/heat–bath). For each `∈K1 , propose U0 ` = R`U` with R` = exp{Paσaξa} drawn from an isotropic distribution near the identity (for G0 = SU (2) use a Kennedy– Pendleton heat–bath [ 103 ]; for U (1) use a uniform offset). Accept with probability min{ 1 ,exp [ − ∆ S(`) L ] } , where ∆ S(`) L includes only the local terms touching ` : the four (in d = 4) plaquettes f with `⊂∂f , the three tetrahedra τwith those f⊂∂τ, and the BF/CS contributions touching `. (B) Face updates (surface Metropolis / “surface worm”). For each f∈K2 , propose W0 f = SfWf with Sf = exp{PaTaζa} ∈ G1 (small steps). Accept with probability based on local change in s0 ( FΦ ( f )), the three tetrahedra τ sharing f (via HΦ ), and topological couplings. To reduce critical slowing down, augment with a surface worm update for abelian G1 : pick a random face and perform a random walk along adjacent faces, updating W by a fixed increment so as to build a closed surface before acceptance (this preserves δW = 0 and changes large–scale flux sectors). (C) Φ–updates (local Metropolis / HMC). For each p∈K0 , propose Φ 0 p = Φ p + δ Φin a coordinate chart on M2-grp (or move along a geodesic with respect to Gαβ ). Compute ∆ S from the Φkinetic term on the star of p , the local potential V (Φ p ), and the dependence of FΦ ( f ) , HΦ ( τ )on tΦ, ρΦ at the neighbouring cells. Accept with Metropolis. For continuous targets M one may accelerate with Hybrid Monte Carlo (HMC) [104] using the pullback metric Gαβ for the kinetic energy. (D) Gauge moves. Interleave gauge–orbit moves to decorrelate: vertex updates εp rotate all incident links and faces via (164) – (165) ; edge updates η` shift the local 2–gauge slice. Since the action is gauge–invariant, these are always accepted and help reduce autocorrelations. (E) Sector–tunnelling updates. To sample different topological sectors (integer H –fluxes on noncontractible 3–cycles), perform global face updates on a representative 2–cycle that winds around the torus, shifting Wf by a fixed group element on that cycle. Detailed balance is preserved by symmetric proposals. E. Observables We list gauge–invariant observables used to diagnose confinement/deconfinement, symmetry–type transitions, and topological responses. Wilson loops and surfaces. For a closed loop C⊂K1, W(C) := 1 Tr0(1) <Tr0 Y `∈C U`.(178) An area law, hW ( C ) i ∼ exp{−σArea ( C ) } , signals confinement; a perimeter law signals deconfinement. For a closed, oriented surface Σ⊂K2, W(Σ) := 1 Tr1(1) <Tr1 Y f∈Σf Wf,(179) with f Wf transported to a baseface via (170) . Area/perimeter laws of W (Σ) diagnose higher–form confinement. Creutz ratios and string tensions. On hypercubic sublattices define W ( R, T )for rectangular loops and the Creutz ratio χ ( R, T ) = −log W(R,T )W(R−1,T −1) W(R−1,T )W(R,T −1) [ 105 ], which approaches the string tension for large R, T. Curvature densities. The local fake curvature density and 3–curvature density EF:= 1 |K2|X f s0(FΦ(f)),EH:= 1 |K3|X τ s1(HΦ(τ)),(180) and their susceptibilities χF,H := |K|(hE2i−hEi2)locate crossovers and transitions. Structure factor of Φ.Define the site magnetisation m = 1 |K0|Pp Φ p (in a chosen component) and its Binder cumulant U4= 1 −hm4i/(3hm2i2)[106]. The Fourier–space structure factor SΦ(q) := 1 |K0|DX p Φpeiq·xp 2E,(181) gives a second–moment correlation length ξΦ.
50 Topological linkings. For abelian sectors with BF coupling, the mixed correlator hW ( C ) W (Σ) i measures the linking number L(C, Σ) via hW(C)W(Σ)i ∝ cos(2πk L)[101]. F. Finite–size scaling (FSS) Let L be the linear size (number of units along each periodic direction), V = Ld the volume, and λ a control parameter (e.g. r in (114) or the bare couplings). Close to a continuous transition at λ = λc , standard FSS ansatz gives χΦ(λ, L)∼Lγ/ν fχ (λ−λc)L1/ν,(182) U4(λ, L)∼fU(λ−λc)L1/ν,(183) mγ(λ, L)∼L−zfm (λ−λc)L1/ν,(184) with dynamical exponent z ( z = 1 for relativistic photons, z = 2 for fracton tensors). Crossings of U4 for different L locate λc ; data collapse extracts ν , γ , and z . Deconfinement is detected by perimeter–law scaling of W(C)and vanishing string tension in the Creutz ratio. G. Algorithmic details and error control We use the product of Haar measures Q`dU`QfdWfQpdµ (Φ p )for integration; acceptance targets 60% ± 10% for local Metropolis moves by adapting step sizes. Autocorrelation times τO are estimated via the integrated autocorrelation function; errors are obtained with binning/jackknife or bootstrap. Multi–level variance reduction [ 107 ] is effective for extended Wilson surfaces. For G0 = SU ( N )reunitarise after floating–point drift; for general compact groups sample proposals in Lie algebra coordinates and exponentiate. H. Extensions: foliated updates for fracton limits In the scalar–charge fracton regime (§V B), restrict 2–gauge proposals η to cochains supported on chosen foliations (planes) and limit face updates to those that preserve ∂i∂j Gauss constraints. Efficient planar worms update stacks of faces in a given foliation, enabling mobility of dipolar excitations while keeping isolated charges immobile, in line with (137). I. Benchmarks and consistency checks • Discrete Bianchi identities: verify (173) holds to machine precision at each sweep (a valuable code check). • Continuum limit: measure hs0 ( FΦ ) i and hs1 ( HΦ ) i versus a ; confirm the expected a4 and a6 scaling. • Topological sectors: on T3 with abelian G1 = Zk , histogram the triplet of H –fluxes ( nx, ny, nz )and check uniform weights (degeneracy k3, Proposition VI.2). • Deconfinement: reproduce the perimeter/area crossover of W ( C )across the U(1) ↔Z2 transition of §VA; extract mγand compare to (119). J. Worked abelian example: U(1) ×U(1) with t(θ) = q θ For G0 = G1 = U (1) and t (e iθ )=e iqθ with integer q (controlled by Φ), write U` = eiθ` , Wf = eibf . Then FΦ(f) = expni∆θf−iq(Φv(f))bfo, HΦ(τ) = expni∆bτo,(185)
51 where (∆ θ ) f is the lattice curl (sum of θ around ∂f ) and (∆ b ) τ the lattice curl of b on the faces of τ . The lattice action is SL=βFX f1−cos(∆θ)f−q bf+βHX τ1−cos(∆b)τ+κ 2X ` (Φt(`)−Φs(`))2+X p V(Φp),(186) with simple Metropolis updates: propose θ0 ` = θ` + δ and b0 f = bf + δ0 ,Φ 0 p = Φ p + δΦ . The deconfinement line is signalled by the vanishing of the string tension and the onset of perimeter law for W ( C ); the symmetry–type transition is tracked by U4and SΦ(q), with scaling as in (182). K. Summary of Section VII • We constructed a lattice 2–gauge theory with link variables U`∈G0 , face variables Wf∈G1 , and site structure variables Φp∈ M2-grp. • Discrete curvatures FΦ ( f )and HΦ ( τ )reproduce the continuum obstructions, satisfy an exact Bianchi identity (173), and lead to a Wilson action (174) approaching (88). • An alternating Metropolis/heat–bath scheme with gauge moves and surface worms samples the path integral efficiently; finite–size scaling of Wilson loops/surfaces and Φ–correlators diagnoses deconfinement and symmetry–type transitions. VIII. DISCUSSION We conclude by clarifying the conceptual departures of the present framework, by outlining concrete computational and experimental routes to falsifiable predictions, and by pointing to domains beyond condensed matter where the same mathematics is poised to organize phenomena with adaptive symmetry. Throughout, we keep the notation of Sections II–VII: the higher connection ( A, B ), the Φ-dependent curvatures FΦ, HΦin (85), and the full action (88). A. Conceptual novelty: symmetry as a field From fixed to adaptive symmetry. Conventional effective theories take the symmetry group (and its extension/associator classes) as fixed data, while order parameters are dynamical fields that break that data. Here, the type of symmetry itself is dynamical: the crossed–module maps ( tΦ, .Φ )and cohomological data ω2,3 (Φ) are selected by a structure field Φ. This elevates the categorical glue from background scaffolding to a participant in the dynamics, subject to equations of motion (93)–(94) and to Φ-sourced Bianchi identities (86)–(87). In practical terms: • the obstructions to strict composition (fake and 3–curvatures) are promoted to fields whose energy can be traded against the cost of changing symmetry type; •gauge, topological, and Landau sectors are unified under a single variational principle; • renormalization flows can now move within the moduli M2-grp of symmetry types, not merely within the coupling space of a fixed type. A rigidity statement. The following observation shows that promoting the symmetry type is not optional if one wants to allow space–time dependent obstruction terms while keeping locality and gauge covariance. Lemma VIII.1 (Necessity of a structure field) . Suppose one postulates local curvatures of the form F = dA + A∧A−bt ( x )( B )and H = dB + Ab. ( x ) B with bt, b. arbitrary x -dependent maps, while insisting on covariant Bianchi identities of the schematic form DF + bt ( x )( H ) = (tensorial) , DH −Fb. ( x ) B = (tensorial) . Then bt, b. must arise from a field Φ :M→ M2-grp with bt = tΦ , b. = .Φ , and the additional ∂bt, ∂b.terms assemble precisely into the Φ-gradient sources in (86)–(87).
52 Proof (sketch). Locality and covariance force ∂µbt , ∂µb. to enter the Bianchi identities in the bilinear forms ( ∂bt ) ∧B and ( ∂b. )( A, B ). Consistency with 1– and 2–gauge transformations fixes their transformation to be that of tangent vectors to a homogeneous space of crossed modules. This identifies bt, b. as the pullbacks of the universal maps t, . along a field Φ. The precise Φ-gradient corrections then follow from differentiating the Peiffer identities; Emergent laws as geometry. Because S [ A, B, Φ] contains a sigma model on M2-grp and a Φ-source Sα [ FΦ, HΦ ], the vectors of the structure flow are supplied by the physical state of the system itself. This provides a controlled sense in which “the rules change with the state”: the rules are the point Φin the moduli of rules, and its motion follows a gradient–flow with sources determined by observed curvatures. The language is geometric; the content is predictive. B. Computational and experimental outlook We summarize quantitative pipelines to extract the obstruction curvatures and structure parameters from data, anchored to the field equations. Neutron scattering: photon gaps and curvature densities The double differential cross section d2σ dΩdω ∝ |F ( q ) |2S ( q, ω )probes the dynamical spin structure factor S , which, in spinon parton descriptions, decomposes into matter and gauge contributions. The emergent gauge sector enters through current correlators that couple linearly to A ; to leading order one may write S(q, ω)≈Smatt(q, ω) + G =ΠT AA(q, ω; Φ),ΠT AA =Z(Φ) ω2−c2(Φ)q2−m2 γ(Φ) + i0+,(187) with G a (weakly q -dependent) form factor. Measuring the low– ω edge of the gauge continuum yields mγ (Φ) and c (Φ); combining with the specific heat crossover (Section V) fixes g/h and, via mγ = g h|α (Φ) | (when k= 0), the slope α0(0) appearing in the universal ratio (163). In 3D quantum spin ice, the static structure factor also contains pinch–point features proportional to the divergence–free projector of the emergent magnetic field b=∇×a; their suppression by mγquantifies h|FΦ|2idirectly. Muon spin rotation (µSR): internal field fluctuations Polarized muons precess in the local field B drawn from the distribution generated by emergent fluxes; the longitudinal relaxation rate is λ(T) = γ2 µZ∞ 0 dt cos(ωµt)hB⊥(t)B⊥(0)iT.(188) In a U(1) spin liquid, B⊥ couples to b , whose correlator is obtained from the photon propagator Π AA . For a linearly dispersing photon in 3D with gap mγ,hB⊥B⊥i(ω)∼ω2=ΠAA(ω), giving, at low T, λ(T)∝(T3, mγ= 0, e−mγ/T , mγ>0,(189) modulo form–factor prefactors. Equation (189) mirrors the C ( T )crossover and provides an independent route to mγ(Φ) and thus to α(Φ). Tomography of HΦby Wilson surfaces On toroidal samples or programmable simulators, one can access Wilson surfaces directly: hW (Σ) i with Σspanning a nontrivial 2–cycle. In abelian phases at level k , hW (Σ) i = exp{−σΣArea (Σ) }exp{ 2 π i nΣ/k} , where nΣ = 1 2πRCHΦ and C is any 3–cycle dual to Σ. Thus the phase of hWi measures the quantised flux sector nCof Proposition VI.1.
53 Candidate materials and protocols • Kagome (Herbertsmithite). A Dirac U(1) regime with a possible Z2 sink is consistent with §V A. Protocol: combine inelastic neutron scattering at the M points (extract mγ ), low– T specific heat (activated tail), and µ SR (Eq. (188) ) across a tuning parameter (pressure/strain) to map mγ ( λ ) and test the scaling mγ∼ |λ−λc|1/2. •Rare–earth pyrochlores. 3D U(1) spin liquids support emergent photons with T3specific heat and pinch points in neutron data; a Φ-driven flow to Z2would quench both, providing a clean test. •α -RuCl 3 .Although proximate to Kitaev physics (nonabelian in field), the mixed WZ–CS coupling (156) implies that if two domains differ by ∆ ω3 = n , sweeping across a boundary produces a step ∆(κxy/T)of magnitude (π2k2 B/3h)n(Section VI). Programmable quantum simulators Cold atoms, Rydberg arrays, trapped ions, and superconducting circuits now engineer local constraints and gauge couplings. To realize a 2–form sector, encode Wf on plaquette qudits and enforce the discrete Bianchi identity (173) energetically; couple to U` via a three–body term that implements the face penalty s0 ( FΦ ( f )). The structure field Φbecomes a classical or slow quantum register controlling the strength q (Φ) in the abelian prototype (186) . Wilson surfaces are measured by Ramsey interferometry on plaquette qubits; domain walls in Φare drawn by spatial light modulators. The ingredients match existing proposals for lattice gauge simulation once extended to plaquette–dressed constraints. C. Computation: algorithms, sign, and dualities The lattice scheme in Section VII is practical in abelian and compact nonabelian regimes. Two technical points deserve emphasis. Worldsheet dualities. For abelian G1 , Poisson resummation maps (186) to a worldsheet gas: integer– valued 2–cochains represent surfaces whose boundaries are electric lines. This eliminates the compact angles bf and turns the face update into a worm on surfaces, typically eliminating critical slowing down in the B sector. The Φfield enters as a local weight changing the effective tension of sheets; near transitions, multicanonical sampling of Φhelps flatten the free–energy barriers between symmetry types. Sign problems and remedies. Chern–Simons terms (in d = 3) and some nonabelian extensions induce complex weights. Complex Langevin with gauge cooling and Lefschetz thimble deformations alleviate the sign problem in moderate volumes; for abelian theories, doubled topological sectors (time–reversal pairs) cancel phases exactly. Where possible, one can simulate at ∆ k0 = 0 and add the quantised step by analytic continuation (protected by integrality). D. Beyond condensed matter The same geometry of adaptive symmetry applies more broadly. Neural networks (adaptive invariances). In overparameterized models, families of equivalent parameterizations form a groupoid; training can be viewed as motion on M2-grp of local reparametrization rules, with Φplaying the role of a hypernetwork selecting invariances. The obstruction curvature is the misfit between successive local symmetries (e.g. layerwise equivariances) and the backpropagated update; gradient noise then plays the role of FΦ, HΦ sources. The formal identities (86) – (87) translate into commutation constraints on update rules in multi–task settings. Cosmology (adaptive gauge sectors). Across cosmological phase transitions, gauge sectors can change effective gauge groups and matter representations. While standard Higgsing keeps the parent gauge group fixed, the present formalism captures more radical scenarios where the low–energy symmetry type itself evolves, for instance through changing extension classes (axionic strings attached to walls modelled by ∆ ω36 = 0) or through metastable phases with nontrivial HΦ trapped in compact three–cycles of spatial topology.
54 Cognitive models (higher–gauge theory of mind). Treating Φas a field of interpretative structures, the wave–map equation (101) driven by sensory curvature becomes a principled dynamics of adaptation: the system reduces the obstruction between current symmetry (internal model) and incoming data by moving on M2-grp . While speculative, the mathematics enforces consistency constraints akin to those discussed around Lemma VIII.1. E. Limitations, open problems, and opportunities 1. Unitarity/positivity in nonabelian 2–groups. Constructing reflection–positive lattice discretizations for general crossed modules remains nontrivial; a full Osterwalder–Schrader construction is open. 2. Classification of flows on M2-grp .The topology and singular strata (rank–jump loci of tΦ ) control defect types and selection rules; a Stratified Morse theory for S [ A, B, Φ] would systematize wall spectra. 3. Interplay with anomalies. When global symmetries mix with higher–form symmetries, the background 2–group anomaly inflow must match the dynamical one; a classification of admissible Φ paths under anomaly constraints is needed. 4. Nonabelian fractonization. Extending §V B to nonabelian 2–groups with nontrivial associators should produce nonabelian subdimensional excitations; lattice prototypes are lacking. 5. Algorithms. Sign–problem–free reformulations for nonabelian 2–form sectors (e.g. character expansions on 2–cells) and HMC on M2-grp with curved metrics are ripe for development. F. Closing perspective The central message is economical: many long–standing puzzles in quantum matter arise precisely where local symmetry data do not glue strictly. The appropriate response is not to bolt on more fields, but to let the gluing rules flow. Promoting the categorical structure to a field Φturns obstructions into curvatures, curvatures into forces on Φ, and symmetry selection into a dynamical, measurable process. The resulting theory is predictive: it yields quantised flux sectors, mixed anomaly steps, critical exponents, and universal ratios, together with concrete simulation algorithms and experimental protocols. Whether in frustrated magnets, quantum simulators, or adaptive systems more broadly, we expect this geometry of symmetry to provide a common language for emergent laws. IX. CONCLUSION We close by collecting the logical and mathematical thread of the work into a single chain, and by highlighting the concrete outputs that follow from allowing both the obstruction curvatures and the type of symmetry to evolve. The chain is: Failure of (co)limit ⇒higher–gauge curvature ⇒dynamical curvature ⇒ dynamical structure field. Failure of (co)limit ⇒higher–gauge curvature ⇒dynamical curvature ⇒ dynamical structure field. From categorical (co)limit failure to curvature Starting with local symmetry data {Gi}and homomorphisms along overlaps, the condition that they glue strictly into a group (co)limit fails precisely when associativity/inversion/closure hold only up to coherent morphisms. In Section II we encoded this by a Lie 2–group (crossed module) ( G1 t −→ G0, . )and its Lie algebra curvatures F=dA +A∧A−t(B), H =dB +A . B,
55 together with the Bianchi identities DF + t ( H ) = 0 and DH −F . B = 0. The core structural statement proved earlier may be rephrased as: Proposition (Restatement of the obstruction criterion).For a fixed crossed module, the following are equivalent on a contractible patch: 1. the diagram of local groups is a strict group (co)limit; 2. there exists a 2–connection (A, B)such that F= 0 and H= 0; 3. holonomies of (A, B)define a strict 2–functor whose 1–composition closes in G0. Thus the failure of (co)limit is measured by ( F, H ): the higher–gauge curvature is the geometric avatar of categorical non–closure. Making curvature dynamical Promoting (A, B)to dynamical fields (Section III) leads to the action S[A, B] = 1 2g2ZkFk2+1 2h2ZkHk2+ZLtop[A, B] + Smatt[A, B;Ψ], whose Euler–Lagrange equations are the higher–form Maxwell equations 1 g2DF +1 h2ρ† B(H) = −? JA,−1 g2t†(F) + 1 h2DH =−? JB. Already at this stage, curvature ( F, H )is not a passive witness of (co)limit failure; it becomes a source of forces and quantised responses via the topological sector (BF/CS), giving e.g. linking phases, topological masses, and—in condensed matter—proximity to deconfined spin liquids and fracton order. Letting the symmetry type evolve: the Φfield The decisive step (Section IV) was to promote the type of symmetry to a field: Φ :M→ M2-grp selects, pointwise, (tΦ, .Φ, ω2,3(Φ)) and thereby the Φ–dependent curvatures (recall (85)) FΦ=dA +A∧A−tΦ(B), HΦ=dB +A .ΦB. The full action (Eq. (88)) S[A, B, Φ] = 1 2g2ZkFΦk2+1 2h2ZkHΦk2+κ 2ZkDΦk2+ZV(Φ) + Ltop(Φ;A, B) + Lmatt(Φ) yields the wave–map equation for Φ(Eq. (101)) κ τ(Φ) = −∇V(Φ) + 1 g2FΦ, ∂tΦ(B)−1 h2HΦ, ∂ρΦ(A, B)+∂Ltop +∂Lmatt, together with Φ–sourced Bianchi identities DFΦ+tΦ(HΦ)+(∂tΦ)(DΦ) ∧B= 0, DHΦ−FΦ.ΦB−(∂ρΦ)(A, B)∧DΦ=0. In words: curvature pushes on Φ, and the motion of Φin turn deforms the curvatures and their constraints. This closes the promised chain: (co)limit failure ⇒(F, H)⇒forces on Φ⇒adapted (tΦ, .Φ)⇒ ···
56 Predictive control unavailable to fixed–group models Allowing Φto evolve unlocks quantitative predictions that are not accessible when the symmetry group is fixed: 1. Quantised responses from mixed anomalies. The coupling 2 π i Rhω3 (Φ) ,1 2πdCS ( A ) i yields steps ∆ σxy = e2 h ∆ hω3i and ∆( κxy/T ) = π2k2 B 3h ∆ hω3i , Eq. (159) , when Φcrosses a domain wall with ∆ω3∈Z. 2. Flux sectors & degeneracy. Integrality of 1 2πRHΦ on 3–cycles (Prop. VI.1) organises ground– state sectors, recovering k3 degeneracy on T3 (Prop. VI.2) and predicting its Φ–driven changes across interfaces. 3. Critical exponents and universal ratios. Near continuous symmetry–type transitions (at k= 0), mγ∼ |r−rc|1/2and ξΦ∼ |r−rc|−1/2, and the universal amplitude ratio mΦ mγ−−−−→ r→r− cRΦ/γ =√2u |α0(0)| h g (Eq. (163)) gives a parameter–free target for neutron/Raman data. 4. Interface spectra & imaging. Φdomain walls bind Pöschl–Teller gauge modes, carry anomaly– protected edge content when ∆ k0 or ∆ ω36 = 0, and imprint distinct LDOS/tunnelling signatures, providing experimental tomography of the Φtexture. A unifying geometric language At a conceptual level, the framework synthesizes three pillars under one categorical umbrella: •Gauge theory: the dynamical 1– and 2–form connections and their Maxwell–like equations; •Topological field theory: BF/CS/WZ sectors controlling quantised linking and anomalies; • Order–parameter dynamics: the sigma model on M2-grp with potential V (Φ) selecting symmetry types. The novelty is not the presence of these ingredients, but their coupling: the same curvatures that measure failure of strict gluing are the forces that move the system on the moduli of possible gluings. This provides a geometric language for emergent laws and self–organization of symmetry: not merely “symmetry breaking” within a fixed group, but symmetry selection among distinct group–like types. Limitations and prospects We emphasised open problems (Section VIII E) ranging from reflection positivity of general 2–group discretisations to anomaly–constrained flows and nonabelian fractonization. On the computational side, character/worldsheet dualities and HMC on curved M2-grp metrics promise scalable simulations; experimentally, combined neutron/ µ SR/thermal Hall protocols can overconstrain mγ , ξΦ , and anomaly steps. The same mathematics is poised to structure adaptive symmetry in programmable quantum platforms and, beyond condensed matter, in settings where the “model of the model” must evolve (learning systems, cosmological history, and proposed higher–gauge theories of cognition). Final remark The minimal slogan of this work is precise: let the obstruction flow. Once one recognises that categorical (co)limits need not hold strictly in complex materials, the rest follows inevitably: encode the failure as higher–gauge curvature; make curvature dynamical; then let the structure that curvature obstructs become dynamical as well. The outcome is a theory with genuine predictive power over phases and responses that fixed–group models cannot access, and a set of concrete numerical and experimental tools to test it.
57 Appendix A: Derivation of continuum EOM and Bianchi identities with variable structure maps This appendix provides a complete derivation of the field equations and the modified Bianchi identities when the crossed–module structure maps are promoted to space–time dependent objects selected by a structure field Φ :M→ M2-grp . We keep the notation of Sections II–IV and work on an oriented pseudo-Riemannian manifold (M, g)with Hodge operator ?. 1. Set-up and conventions A (differential) crossed module consists of Lie algebras ( g1,g0 ), a Lie algebra homomorphism t:g1→g0 and a left action ρ:g0→Der(g1),X7→ ρ(X)(we write X . Y =ρ(X)Y), obeying the Peiffer identities t(X . Y )=[X, t(Y)]0,(A1) t(Y). Y 0= [Y, Y 0]1.(A2) Promoting the structure to Φ-dependence, we write tΦ and ρΦ ( · ) = .Φ , with ∂αtΦ and ∂αρΦ their derivatives along local coordinates Φ α on the moduli M2-grp . Let κ0 and κ1 be ad-invariant, nondegenerate bilinear forms on g0and g1, and h·,·i0,h·,·i1the induced L2pairings of forms. A2–connection is a pair A∈Ω1(M, g0), B ∈Ω2(M, g1)with Φ–dependent curvatures FΦ=dA +A∧A−tΦ(B)∈Ω2(M, g0),(A3) HΦ=dB +A .ΦB∈Ω3(M, g1).(A4) We denote by D the g0 covariant derivative Dω := dω + [ A, ω ]on g0 –valued forms and by D := ±? D? its formal adjoint. 2. Modified Bianchi identities Claim. The Φ–dependent curvatures satisfy DFΦ+tΦ(HΦ)+(∂αtΦ)(DΦα)∧B= 0,(A5) DHΦ−FΦ.ΦB−(∂αρΦ)(A, B)∧DΦα= 0.(A6) Derivation. Start with FΦ = dA + A∧A−tΦ ( B ). Using the standard Bianchi identity D ( dA + A∧A ) = 0 and the Leibniz rule for the x–dependence of tΦ, DtΦ(B)=tΦ(DB)+(∂αtΦ)(DΦα)∧B, together with the equivariance [A, tΦ(B)] = tΦ(A .ΦB), yields DFΦ=−tΦ(DB +A .ΦB)−(∂αtΦ)(DΦα)∧B=−tΦ(HΦ)−(∂αtΦ)(DΦα)∧B, which is (A5). For HΦ, DHΦ=dHΦ+A .ΦHΦ=ddB +ρΦ(A)B+A .ΦHΦ. Expanding, d ( ρΦ ( A ) B ) = ρΦ ( dA ) B−ρΦ ( A ) dB + ( ∂αρΦ )( d Φ α )( A, B ). Rewriting dB = HΦ−ρΦ ( A ) B and using bilinearity, DHΦ=ρΦ(dA +A∧A)B+ (∂αρΦ)(dΦα)(A, B)=(FΦ+tΦ(B)) .ΦB+ (∂αρΦ)(DΦα)(A, B), and the differential Peiffer identity tΦ(B).ΦB= 0 gives (A6). 3. Variation of the action and Euler–Lagrange equations Consider the full action (suppressing matter for clarity) S[A, B, Φ] = 1 2g2Zκ0(FΦ∧?FΦ)+ 1 2h2Zκ1(HΦ∧?HΦ)+κ 2ZGαβ(Φ)DΦα·DΦβ+ZV(Φ)+Stop.(A7)
64 [81] T. Qin, Q. Niu, and J. Shi, “Energy Magnetization and the Thermal Hall Effect,” Phys. Rev. Lett. 107 , 236601 (2011). [82] P. A. Lee, N. Nagaosa, and X.-G. Wen, “Doping a Mott insulator: Physics of high-temperature superconductivity,” Rev. Mod. Phys. 78, 17 (2006). [83] M. Pretko, “Generalized electromagnetism of subdimensional particles: A spin liquid story,” Phys. Rev. B 96, 035119 (2017). [84] M. Pretko, “Subdimensional particle structure of higher rank U(1) spin liquids,” Phys. Rev. B 95 , 115139 (2017). [85] R. M. Nandkishore and M. Hermele, “Fractons,” Annu. Rev. Condens. Matter Phys. 10, 295–313 (2019). [86] A. Gromov, “Towards classification of fracton phases: the multipole algebra,” Phys. Rev. X 9 , 031035 (2019). [87] M. Pretko and L. Radzihovsky, “Fracton-elasticity duality,” Phys. Rev. Lett. 120, 195301 (2018). [88] K. Slagle and M. Pretko, “Foliated field theory and string-membrane-net condensation picture of fracton order,” Phys. Rev. B 99, 205106 (2019). [89] S. Vijay, J. Haah, and L. Fu, “Fracton topological order, generalized lattice gauge theory and duality,” Phys. Rev. B 94, 235157 (2016). [90] N. Seiberg and S.-H. Shao, “Exotic symmetries, duality, and fractons in 2+1-dimensional quantum field theory,” SciPost Phys. 10, 003 (2021). [91] R. Jackiw and C. Rebbi, “Solitons with fermion number 1/2,” Phys. Rev. D 13, 3398–3409 (1976). [92] M. A. Shifman, “Domain walls and decay rate of the excited vacua in the large N Yang–Mills theory,” Phys. Rev. D 59, 021501 (1998). [93] F. A. Bais and J. K. Slingerland, “Condensate-induced transitions between topologically ordered phases,” Phys. Rev. B 79, 045316 (2009). [94] A. Kitaev and L. Kong, “Models for gapped boundaries and domain walls,” Commun. Math. Phys. 313 , 351–373 (2012). [95] J.-L. Brylinski, Loop Spaces, Characteristic Classes and Geometric Quantization (Birkhäuser, 1993). [96] M. K. Murray and D. Stevenson, “Bundle gerbes: stable isomorphism and local theory,” J. London Math. Soc. 62, 925–937 (2000). [97] X.-L. Qi, T. L. Hughes, and S.-C. Zhang, “Topological field theory of time-reversal invariant insulators,” Phys. Rev. B 78, 195424 (2008). [98] A. M. Essin, J. E. Moore, and D. Vanderbilt, “Magnetoelectric polarizability and axion electrodynamics in crystalline insulators,” Phys. Rev. Lett. 102, 146805 (2009). [99] A. N. Hirani, Discrete Exterior Calculus, PhD Thesis, Caltech (2003). [100] M. Desbrun, A. N. Hirani, M. Leok, and J. E. Marsden, “Discrete Exterior Calculus,” arXiv:math/0508341. [101] R. Dijkgraaf and E. Witten, “Topological Gauge Theories and Group Cohomology,” Commun. Math. Phys. 129, 393–429 (1990). [102] M. Panero, “Numerical lattice gauge theory: Chern–Simons term and the sign problem,” JHEP 0505 , 066 (2005). [103] A. D. Kennedy and B. J. Pendleton, “Improved heatbath method for Monte Carlo calculations in lattice gauge theories,” Phys. Lett. B 156, 393–399 (1985). [104] S. Duane, A. D. Kennedy, B. J. Pendleton, and D. Roweth, “Hybrid Monte Carlo,” Phys. Lett. B 195 , 216–222 (1987). [105] M. Creutz, “Monte Carlo Study of Quantized SU(2) Gauge Theory,” Phys. Rev. D 21, 2308–2315 (1980). [106] K. Binder, “Finite size scaling analysis of Ising model block distribution functions,” Z. Phys. B 43 , 119–140 (1981). [107] M. Lüscher and P. Weisz, “Locality and exponential error reduction in numerical lattice gauge theory,” JHEP 0109, 010 (2001). [108] A. Bullivant, J. Faria Martins, and R. Picken, “Lattice 2-gauge theory and state-sum models,” J. Math. Phys. 58, 023501 (2017). [109] S. W. Lovesey, Theory of Neutron Scattering from Condensed Matter, Vols. 1–2 (Oxford University Press, 1984). [110] A. Yaouanc and P. Dalmas de Réotier, µ SR Spectroscopy: An Introduction (Oxford University Press, 2011). [111] L. Savary and L. Balents, “Quantum spin liquids: a review,” Rep. Prog. Phys. 80, 016502 (2017). [112] E. Zohar, J. I. Cirac, and B. Reznik, “Quantum simulations of lattice gauge theories using ultracold atoms in optical lattices,” Rep. Prog. Phys. 79, 014401 (2016). [113] D. Banerjee et al., “Atomic Quantum Simulation of Dynamical Gauge Fields coupled to Fermionic Matter: From String Breaking to Evolution after a Quench,” Phys. Rev. Lett. 109, 175302 (2012). [114] O. Benton, O. Sikora, and N. Shannon, “Seeing the light: Experimental signatures of emergent electromagnetism in a quantum spin ice,” Phys. Rev. B 86, 075154 (2012). [115] T. Fennell et al., “Magnetic Coulomb phase in the spin ice Ho2Ti2O7,” Science 326, 415–417 (2009). [116] S. M. Winter et al., “Models and materials for generalized Kitaev magnetism,” J. Phys.: Condens. Matter 29, 493002 (2017). [117] G. Aarts, “Can complex Langevin dynamics evade the sign problem?” Phys. Rev. Lett. 102 , 131601 (2009). [118] M. Troyer and U.-J. Wiese, “Computational Complexity and Fundamental Limitations to Fermionic Quantum Monte Carlo Simulations,” Phys. Rev. Lett. 94, 170201 (2005).