Integrability as Categorical Coherence: Generalized Lax Pairs via Flat 2-Connections and Non-Abelian Descent for Nonlinear PDEs

Integrability as Categorical Coherence: Generalized Lax Pairs via Flat 2-Connections and Non-Abelian Descent for Nonlinear PDEs Andrei T. Patrascu 1 1 FAST Foundation, Destin FL, 32541, USA email: andrei.patr[email protected] We recast integrability as a problem of categorical coherence. For a nonlinear PDE on a space–time domain, we introduce a Coherent Lax Structure (CLS): a principal 2 - bundle equipped with a flat 2 - connection that encodes the auxiliary linear problems on local patches and their coherent gluing. The associated non - abelian descent class [ ω ]measures the obstruction to global coherence. Our main result shows that [ ω ]=0is equivalent to classical integrability (existence of a commuting hierarchy and monodromy invariants), recovering standard Lax/zero - curvature theory as the special case with trivial 2 - data. We further internalize the spectral parameter and capture non - ultralocal brackets via the 2 - form sector, clarifying long - standing issues in Maillet - type structures. When [ ω ] 6 = 0 its magnitude quantifies near - integrability, controlling dephasing of commuting flows, defects in trace invariants, and the approach to singularity. Building on this, we develop a coherence - by - descent solver that enforces flatness and minimizes fake curvature to produce certified global solutions or controlled approximations. Benchmarks on KdV/NLS, Landau–Lifshitz/sigma models, and dissipative PDEs (e.g., 2D Navier–Stokes) illustrate both the theory and its computational benefits. INTRODUCTION The Lax–pair / zero–curvature paradigm has been central to classical integrability for more than half a century. In its familiar guise, one exhibits auxiliary operators ( L ( λ ) , M ( λ )) depending on a spectral parameter λ such that the nonlinear PDE (or 1 + 1–dimensional field theory) of interest is equivalent to the compatibility condition ˙ L = [ M, L ]. Equivalently, one encodes the dynamics as a flatness (zero–curvature) equation for a connection built from U, V in space–time. This viewpoint, pioneered by Lax and developed through the inverse scattering transform (IST), loop–group methods, and algebro–geometric constructions, explains hierarchies of commuting flows and spectral/monodromy invariants in models such as KdV and NLS; see, e.g., [3, 5, 6, 68, 122, 123]. Where the classical picture strains. Despite its elegance, the classical Lax program is fragile in three practical ways. 1. Globality of the auxiliary problem. One typically assumes a globally defined spectral object (parameter λ or spectral curve Σ) and a global flat connection encoding the dynamics. In many physical/geometric settings—nontrivial topology, defects, boundaries, or gauge choices—only local Lax data exist and must be glued across patches. Algebro–geometric methods vividly show how spectral curves and Baker–Akhiezer data live over covers and glue with monodromies, but do not by themselves organize obstructions to globalization [7, 8]. 2. Non–ultralocality. For numerous field theories (e.g., principal chiral / sigma models, Landau–Lifshitz), the spatial Lax operator has Poisson brackets involving derivatives of delta functions. The standard classical r –matrix / Yang–Baxter algebra then fails; one needs the Maillet r/s –formalism and careful regularization to recover commuting charges, and quantization becomes subtle [ 11 , 13 , 67 , 131 , 132 ]. 3. Perturbations and dissipation. In weakly non–integrable, forced, or dissipative regimes, a pristine Lax pair typically disappears; yet many systems retain “near–integrable” structure—metastable invariants, slow dephasing, long–lived coherent patterns—which calls for a robust generalization of the Lax method. Our proposal: integrability as categorical coherence. We replace “compatibility of a pair of linear problems” by the coherence of higher–gauge data. Concretely, for a PDE on space–time X we introduce aCoherent Lax Structure (CLS): a principal 2–bundle on a site ( X, τ )with structure crossed module t : h→g (thus a Lie 2–group), equipped with a 2–connection ( A, B )consisting of a g –valued 1–form A and an h–valued 2–form B. The curvatures are FA−t(B) | {z } fake curvature = 0, Z := dB +α(A)∧B | {z } 2–curvature = 0,(1) 2 which together play the role of a categorified zero–curvature condition. Vanishing fake curvature is the higher–gauge constraint that ensures well–defined, reparameterization–invariant parallel transport by both 1–holonomy and 2–holonomy [15–19]. Two structural features make the shift powerful. • Descent and obstruction. Local CLS data ( Ai, Bi )on a cover {Ui} glue across overlaps via 1– and 2–gauge transformations. The obstruction to forming a global CLS is a non–abelian Čech 2–cocycle ωijk with values in the structure 2–group. Its class [ ω ]in non–abelian cohomology is our coherence defect. Principal 2–bundles and their connections are classified in precisely these terms; this is the natural home for measuring globalization failures of local Lax data [15, 19, 24]. • Internal spectral geometry. The spectral parameter λ (or curve Σ) is treated as an internal object of the topos Sh ( X ). A CLS on X× Σrenders L ( λ )as the 1–holonomy of A along space, while the 2–form B records the “non–ultralocal flux” so that flatness encodes the zero–curvature equations including Maillet’s r/s –terms. This reconciles spectral–curve algebraic geometry with local patching [7, 8]. What is new? 1. Compatibility = Coherence. In the CLS language, compatibility of auxiliary problems is precisely the coherence of a 2–functor from the path 2–groupoid of X to the structure 2–group. Coherence theorems in category theory (Mac Lane; Joyal–Street) thus become the right meta–theorems governing the existence and rigidity of Lax representations, now extended to higher data [20, 21]. 2. A precise obstruction theory. The class [ ω ]is a computable invariant of a PDE relative to a cover and a chosen structure 2–group. We prove a Categorified Zero–Curvature Correspondence: under standard regularity hypotheses, [ω] = 0 ⇐⇒ existence of a global CLS with commuting hierarchy and monodromy / 2–holonomy invariants, recovering the classical Lax theory when h = 0. Conversely, [ ω ] 6 = 0 quantifies near–integrability: its size controls dephasing of commuting flows and defects in monodromy traces, providing a scale for long–time quasi–conservation. 3. A resolution of non–ultralocality. In our setting, the B –field naturally captures the “ s ”–part of Maillet brackets. Zero fake curvature forces the Poisson algebra of 1–holonomies to close in a way equivalent to the Maillet r/s brackets for the spatial Lax operator, while true 2–flatness ensures consistency of time evolution. This converts non–ultralocal anomalies into coherent higher–gauge data and aligns with modern treatments of non–ultralocal models [13, 67, 131, 132]. 4. Internal spectral parameter & spectral curves. Treating λ or Σinternally subsumes the Krichever / finite–gap picture into the CLS calculus: a family of flat 2–connections over X× Σpackages isospectral invariants (1–holonomy around space) and their non–ultralocal corrections (2–holonomy across surfaces), with consistent descent over a cover of Σ[7]. Relation to and unification of prior work. The CLS viewpoint contains inverse scattering (e.g., Beals–Coifman) and loop–group / Grassmannian constructions (Pressley–Segal) as the slice h = 0, where only 1–holonomy matters; spectral curves appear as internal parameters [ 8 , 126 ]. In monoidal categories, coherence constraints yield braided / tortile structures; within CLS these arise as coherence for scattering data, recovering classical r –matrix formalism (Semenov–Tian–Shansky) and connecting to quantum–group deformations upon quantizing 1– and 2–holonomies [ 11 , 22 , 23 ]. On the geometry side, our use of principal 2–bundles, fake curvature FA−t ( B ), and surface holonomy follows the higher–gauge literature [ 15 – 19 ]; the novelty here is to identify these coherence conditions with integrability of PDEs and to leverage non–abelian descent as a practical obstruction/computation tool. What the framework can achieve. • Ageneral existence/obstruction theory for Lax representations applicable on manifolds with nontrivial topology, in the presence of boundaries/defects, and for non–ultralocal brackets. • Anear–integrability calculus: quantitative control of deviations from integrability through kωk , with stability bounds for invariants and long–time behavior. 3 • Acomputational pipeline robust to forcing/dissipation and to non–ultralocality, delivering certified approximations by enforcing coherence rather than isolated conservation laws. • Aunified language that internalizes spectral parameters/curves and ties classical integrability to categorical coherence and, downstream, to braided / Yang–Baxter structures. Guide to the paper. Section 2 reviews higher–gauge preliminaries (2–bundles, fake curvature, 2–holonomy) and fixes notation. Section 3 defines CLS and proves the Categorified Zero–Curvature Correspondence. Section 4 develops non–abelian descent and the Descent–Obstruction Criterion. Section 5 internalizes spectral parameters/curves. Section 6 treats non–ultralocal field theories and shows how the B –sector encodes Maillet’s r/s structure. Section 7 presents the coherence–by–descent solver and a–posteriori error / long–time control via kωk . Section 8 illustrates the method on KdV/NLS (periodic), Landau–Lifshitz / sigma models, and a dissipative PDE, and outlines open problems (e.g., (∞,1)–categorical generalizations). HIGHER-GAUGE PRELIMINARIES AND NOTATION This section fixes the geometric and categorical structures used throughout: (strict) Lie 2–groups from crossed modules, principal 2–bundles and their local descent data, 2–connections and their curvatures, gauge and 2–gauge transformations, and holonomy/parallel 2–transport via the path 2–groupoid. We also summarize the stacky/descent viewpoint that underlies our non-abelian obstruction class. For background and complementary expositions see [25–28, 30–34, 138]. Crossed modules and strict Lie 2–groups Acrossed module of Lie groups is a homomorphism t : H→G together with a smooth action α:G→Aut(H), written g·h:= α(g)(h), such that for all g∈Gand h, h0∈H: t(g·h) = g t(h)g−1,(2) t(h)·h0=h h0h−1.(3) The second identity is the Peiffer law. A crossed module ( Ht −→ G, α )determines a (strict) Lie 2–group G with one object, 1–morphisms G and 2–morphisms the semidirect product HoG ; composition is induced by tand α[25, 26]. Differentiating gives a differential crossed module ( ht −→ g, α )with a Lie algebra homomorphism t : h→g and a representation α:g→Der(h)satisfying t(α(X)Y)=[X, t(Y)], α(t(Y))Y0= [Y, Y 0] (X∈g, Y, Y 0∈h).(4) We write the induced g–action on hby X·Y:= α(X)Y. Remark .1 (Semi-strict/Lie 2–algebras).For semi-strict 2–groups (e.g. the string 2–group), the infinitesimal data form a Lie 2–algebra or, equivalently, an L∞ –algebra concentrated in degrees 0and 1with a possible nontrivial 3–bracket `3 : ∧3g→h [ 25 , 32 , 33 ]. All formulas below extend verbatim by adding `3 –terms as indicated in Remark .6. Principal 2–bundles and descent data Let X be a smooth manifold and G a strict Lie 2–group from ( Ht −→ G, α ). A principal G –2–bundle over Xmay be described by non-abelian Čech 2–cocycles on a good open cover U={Ui}i∈I: gij :Uij →G, hijk :Uijk →H, (5) obeying the coherence relations on double and triple overlaps: gijgjk =t(hijk)gik on Uijk,(6) (gij ·hjkl)hijl =hijk hikl on Uijkl.(7) 4 Two such cocycles are equivalent if related by 1– and 2–gauge refinements: smooth maps ki : Ui→G and λij :Uij →Htransforming g0 ij =kigij k−1 jt(λij)−1,(8) h0 ijk = (ki·hijk)λij (gij ·λjk)λ−1 ik .(9) The resulting equivalence classes form the non-abelian cohomology set H1(X, G)[26, 28]. 2–connections and curvatures A2–connection on a principal G–2–bundle is given locally by differential forms Ai∈Ω1(Ui,g), Bi∈Ω2(Ui,h), together with 1–form 2–gauge data Λ ij ∈ Ω 1 ( Uij,h )relating the pairs ( Ai, Bi )on overlaps. Write the curvature of Ai as FAi := d Ai + 1 2 [ Ai∧Ai ]and the covariant derivative DAiBi := d Bi + ( Ai·Bi ), where A·Bdenotes the g–action on hextended to forms. Definition .2 (Fake curvature and 2–curvature).The fake curvature and the 2–curvature are Fi:= FAi−t(Bi)∈Ω2(Ui,g),Hi:= DAiBi∈Ω3(Ui,h).(10) Definition .3 (Gauge and 2–gauge transformations).On an overlap Uij, the local data transform by Aj= Adgij (Ai) + gij dg−1 ij −t(Λij),(11) Bj=gij ·Bi−DAjΛij,(12) and the 1–form data satisfy on Uijk the coherence condition Λik = Λij +gij ·Λjk −h−1 ijk dhijk −Ai·hijk,(13) where hijk is embedded via Lie(H)in a chosen faithful matrix representation. Proposition .4 (Curvature transformation laws).Under (11)–(13), Fj= Adgij (Fi),Hj=gij · Hi.(14) Proof. A direct calculation using (11) and the crossed-module relations (4) shows FAj = Adgij ( FAi ) − t ( DAj Λ ij ). Subtract t ( Bj )given by (12) to obtain the first identity. The second follows from DAj ( gij ·Bi ) = gij ·DAiBiand D2 AjΛij = (FAj·Λij)together with the first identity. Definition .5 (Fake flatness and 2–flatness) . A2–connection is fake flat if Fi = 0 on all Ui and 2–flat if additionally Hi= 0. By (14), these conditions are gauge/2–gauge invariant. Remark .6 (Semi-strict correction terms).For a semi-strict Lie 2–algebra with nonzero `3 : ∧3g→h , the 2–curvature reads Hi = DAiBi + 1 6`3 ( Ai, Ai, Ai ), and (12) , (13) acquire corresponding `3 –corrections [25, 32, 33]. All statements below remain valid with these modifications. Path 2–groupoid and parallel 2–transport Let Π 2 ( X )denote the path 2–groupoid of X : objects are points of X ;1–morphisms are thin-homotopy classes of piecewise smooth paths; 2–morphisms are thin-homotopy classes of bigons (smooth maps Σ : [0,1]2→Xwith fixed boundary). Composition is by concatenation [26, 34]. Given a (fake-flat) 2–connection (A, B)on a trivial G–2–bundle, one defines: •1–holonomy HA(γ)∈Galong a path γas the unique solution to ˙g=−A(˙γ)g,g(0) = e. • 2–holonomy HA,B (Σ) ∈H along a bigon Σas the (surface-ordered) exponential of B paralleltransported by A(well-defined by fake flatness). 5 These assemble into a smooth 2–functor Hol(A,B): Π2(X)−→ BG,(15) with BGthe one-object 2–groupoid of G[30, 138]. Theorem .7 (Well-definedness and functoriality of higher holonomy) . If ( A, B )is fake flat, Hol(A,B) is independent of thin homotopy of paths and bigons and respects horizontal/vertical composition of bigons. If (A, B)is 2–flat, then Hol(A,B)depends only on homotopy classes. Proof sketch. Thin-homotopy invariance follows by Stokes’ theorem and the definition of thin homotopy (vanishing rank), using F = 0 to remove boundary contributions. Functoriality under horizontal/vertical composition is a direct consequence of the differential equation defining parallel 2–transport and the crossed-module axioms [ 30 , 138 ]. For 2–flatness, H = 0 implies invariance under fat homotopies of bigons (again by Stokes). Stacky viewpoint and differential non-abelian cohomology Let BG denote the smooth 2–stack of principal G –2–bundles; the assignment U7→ BG ( U )is a stack on the site of smooth manifolds with the open-cover topology. There is a 2–stack BG∇ of 2–bundles with 2–connection whose objects on U are given by local data ( gij, hijk ; Ai, Bi, Λ ij )modulo 2–gauge equivalence [28, 31]. The forgetful morphism BG∇→BGis a 2–stack morphism. Theorem .8 (Descent and classification with connection) . For a good cover U , the groupoid of principal G –2–bundles with 2–connection is equivalent to the groupoid of Čech data ( gij, hijk ; Ai, Bi, Λ ij )satisfying (6) – (13) modulo refinements. Isomorphism classes are elements of a differential non-abelian cohomology set H1 diff (X, G)fitting into an exact sequence H1(X, H)−→ H1 diff (X, G)−→ H1(X, G)δ −−→ H2(X, H), where δis the obstruction to lifting a G–bundle to a G–2–bundle. Idea of proof. The first statement is the stacky descent property for BG∇ ; the second is obtained by analyzing the long exact sequence in non-abelian cohomology associated to the crossed module H→G and its differential refinement (cf. [28, 31]). Remark .9 (Totally trivializable 2–bundles).If H1 ( X, G )is trivial (e.g. X contractible), H1 diff ( X, G ) reduces to gauge-equivalence classes of global pairs ( A, B )modulo global 1– and 2–gauge transformations (g, Λ); the holonomy 2–functor Hol(A,B)classifies such data up to equivalence [30, 138]. Gauge-invariant Wilson 1– and 2–observables For a closed loop γand a closed oriented surface Σ, define W1(γ) := tr ρHol(A,B)(γ), W2(Σ) := tr σHol(A,B)(Σ), with ρ : G→GL ( V ), σ : H→GL ( W )finite-dimensional representations. If ( A, B )is fake flat, W1 is invariant under thin homotopy; if 2–flat, both W1and W2are homotopy invariants. These will serve as the building blocks for the categorical Lax invariants in later sections. Example .10 (String 2–group) . For the string 2–group at level k , one has h = R (or u (1)) and a nontrivial `3 proportional to k via the canonical 3–cocycle on g = spin ( n ). Then H = DAB + k 6hA, [ A, A ] i , and 2–flatness is the Green–Schwarz anomaly-cancellation condition [27, 35]. COHERENT LAX STRUCTURES: DEFINITIONS AND MAIN CORRESPONDENCE In this section we give a full mathematical specification of a Coherent Lax Structure (CLS) for a (1+ 1)–dimensional PDE, prove the Categorified Zero–Curvature Correspondence, and spell out the dictionary with the classical Lax/zero–curvature formalism. We work in the jet–bundle language so that all statements are intrinsic and independent of coordinate choices or ad hoc ansätze. The presentation blends Wahlquist–Estabrook prolongation, jet geometry and variational bicomplex tools with higher–gauge notions, and may be read as a categorified refinement of zero–curvature representations [36–41, 69]. 6 PDEs on jets and horizontal calculus Let π : E→X be a smooth bundle with X = S1×I (space × time), local coordinates ( x, t )on X and fiber coordinates ua on E . Denote by J∞ ( π )the infinite jet bundle, with induced coordinates for all derivatives ua xitj . The projection is π∞ : J∞ ( π ) →X . The horizontal differential d H on J∞ ( π )splits d = dH+ dV; in coordinates dHf=Dx(f) dx+Dt(f) dt, Dx=∂x+Xua xi+1tj∂ua xitj, Dt=∂t+Xua xitj+1 ∂ua xitj. A (system of) evolution PDEs is a submanifold E ⊂ J∞(π)locally given by ut=K(u, ux, uxx, . . . ),(16) or more generally by a differential ideal IE⊂ Ω • ( J∞ ( π )) closed under d H [ 38 – 40 ]. All objects below will be defined on E, i.e. modulo IE. CLS data on jets and the PDE–auxiliary dictionary Fix a strict Lie 2–group from a crossed module ( Ht −→ G, α )with Lie algebras ( ht −→ g, α ). A CLS for E consists of: •a principal G–2–bundle over X(given locally by descent data (gij, hijk)on a good cover U), • a2–connection ( A, B )with A∈ Ω 1 ( X, g ), B∈ Ω 2 ( X, h ),whose coefficients are smooth functions on E ⊂ J∞(π), • a choice of horizontal splitting A = L d x + M d t , B = Bxt d x∧ d t with L, M : E → g and Bxt : E → h , satisfying the categorified flatness equations on E (fake flatness) ∂tL−∂xM+ [L, M] = t(Bxt), (2–flatness) DtBxt −DxBtt + (M·Bxt)−(L·Btt) = 0,(17) with the understanding that Btt ≡ 0in (1 + 1)–dimensions; we retain the covariant–derivative form to keep the hierarchy notation uniform. The equalities are to be read modulo IE (i.e. on the solution locus). In local trivializations, the auxiliary system associated with a representation ρ:G→GL(V)is ψx+ρ(L)ψ= 0, ψt+ρ(M)ψ= 0,(18) and (17) ensures (via the nonabelian Stokes calculus on E ) that the two Pfaff systems are compatible in the sense detailed below. The role of Bxt is to record precisely those “anomaly” pieces which, in non–ultralocal models, obstruct Fxt = 0 at the level of Aalone and are repaired by the t:h→gcontribution. Remark .11 (Wahlquist–Estabrook prolongation).Classically, one searches for Lie–algebra valued 1–forms θ on J∞ ( π )such that d θ + θ∧θ≡ 0 ( mod IE ); this is the Wahlquist–Estabrook prolongation [ 36 ]. Our A plays that role, and the B sector enforces coherence of the prolongation data under refinement and in the presence of non–ultralocal defects [37]. Monodromy, 2–holonomy, and corrected invariants Let X = S1×I with base point ( x0, t ). For A = L d x + M d t , define the spatial 1–holonomy (monodromy) M(t) := Pexp Zx0+2π x0 L(x, t) dx∈G. By a standard variational identity for path–ordered exponentials (a nonabelian Stokes formula on cylinders in X), one has on E d dtM(t)=[M(x0, t),M(t)] −ZS1 U(x0, x;t)Fxt(x, t)U(x, x0;t) dx, (19) 7 where U is the parallel transport for A along the spatial circle. Writing Fxt = ∂tL−∂xM + [ L, M ] and using fake flatness, the integral reduces to t ( Bxt )dressed by U . This motivates the B –corrected monodromy f M(t) := HA,B(Σt)−1M(t)∈G, (20) where HA,B (Σ t ) ∈H is the 2–holonomy across any cylinder Σ t spanning the spatial loop at time t and a fixed reference loop, pushed to Gvia t. The next statement is the key “categorified Lax lemma”. Proposition .12 (Conservation of corrected spectral invariants).Assume fake flatness on E. Then d dtf M(t) = [M(x0, t),f M(t)]. Consequently, for any class function f on G (e.g. f ( g ) = tr ρ ( g ) k ) the quantity If ( t ) := f ( f M ( t )) is conserved in t . If 2–flatness also holds, HA,B (Σ t )is independent of the chosen spanning cylinder, so If is well defined globally. Proof. Differentiate (20) and use (19) together with the infinitesimal variation of HA,B along the cylinder, which precisely cancels the t ( Bxt )contribution by the definition of 2–holonomy. The remaining term is the adjoint action by M(x0, t). Remark .13 (Intuition).In the classical ( B≡ 0) situation, flatness Fxt = 0 alone yields ˙ M = [ M, M ] and hence spectral invariants of M are conserved. In non–ultralocal settings, Fxt typically acquires distributional contributions that obstruct conservation unless one augments the data. The B –sector is precisely that augmentation at the level of higher gauge coherence, ensuring that coherently transported monodromy is conserved. Hierarchies and commuting flows To encode an integrable hierarchy, consider X=S1×RNwith coordinates (x, t1, . . . , tN)and write A=Ldx+ N X n=1 M(n)dtn, B = N X n=1 Bxtndx∧dtn+X m<n Btmtndtm∧dtn. Fake flatness yields ∂tnL−∂xM(n)+ [L, M(n)] = t(Bxtn),(21) ∂tmM(n)−∂tnM(m)+ [M(m), M(n)] = t(Btmtn), m < n, (22) and 2–flatness gives covariant closure DAB = 0. Define G –valued monodromies M ( t1, . . . , tN )along x as before and correct them by suitable H –valued surface holonomies to obtain f M . The following is the hierarchy counterpart of Proposition .12. Theorem .14 (Commuting corrected flows) . Assume fake flatness and 2–flatness on E . Then for each n , ∂tnf M= [M(n)(x0,t),f M], and the flows ∂tm, ∂tn commute on all class functions of f M . In particular, the algebra generated by {If,k := tr ρ ( f M ) k} is an Abelian algebra of conserved quantities for every finite–dimensional representation ρ. Proof. The proof follows the same nonabelian Stokes computation for each time, together with 2–flatness to ensure compatibility of the various H –holonomies used to correct M . Commutation of the flows reduces to the vanishing of the x –component of the three–curvature DAB and the consistency condition (22); both are encoded in 2–flatness. 8 From CLS to classical Lax pairs and back We now make precise the equivalence between CLS with h = 0 and classical zero–curvature forms, and explain how nontrivial hcaptures non–ultralocal anomalies. Proposition .15 (Classical slice) . If h = 0 (so t≡ 0), a CLS is precisely a principal G –bundle with a flat G –connection A = L d x + M d t on E . In a trivialization, Fxt = 0 is the usual zero–curvature equation ∂tL−∂xM + [ L, M ]=0, and Proposition .12 reduces to conservation of spectral invariants of the monodromy M. Proposition .16 (Non–ultralocal repair) . Suppose a model admits a pair ( L, M )for which the equalities in (21) hold with a nonzero right–hand side concentrated at coincident points (a typical non–ultralocal anomaly). Then, choosing h and t : h→g so that t ( Bxt )reproduces precisely the anomaly distribution produces a CLS whose corrected monodromy has conserved spectral invariants. Idea. This is the higher–gauge analogue of the r/s –regularization: the s –term lives in the image of t and is encoded by Bxt ; fake flatness matches the defect, while 2–flatness ensures that corrections are coherent under deformations of spanning surfaces and across multiple times. The Categorified Zero–Curvature Correspondence We can now state and prove the main structural result advertised in the introduction. Theorem .17 (Categorified Zero–Curvature Correspondence) . Let E be a (1+1)–dimensional PDE. Assume there exists a principal G –2–bundle over X = S1×I with 2–connection ( A, B )whose coefficients are smooth functions on Eand which is fake flat and 2–flat on E. Then: (i) For every finite–dimensional representation ρ of G , the B –corrected monodromy f M has conserved spectral invariants; the collection of these provides an infinite set of integrals of motion in involution. (ii) If, moreover, the data extend to X = S1×RN with A, B satisfying (21) – (22) and 2–flatness, then the associated flows commute as in Theorem .14; hence E is integrable in the sense of an Abelian hierarchy generated by class functions of f M. Conversely, if E admits a classical Lax representation ( L, M )with zero curvature (possibly spectral–parameter dependent), then it yields a CLS with h = 0; if the model is non–ultralocal with a consistent r/s–regularization, it yields a CLS with nontrivial Breconstructed from the anomaly. Proof. Items (i)–(ii) are precisely Propositions .12 and Theorem .14. The converse in the ultralocal case is Proposition .15. In the non–ultralocal case, standard constructions (e.g. via Fokas–Gel’fand or Wahlquist–Estabrook prolongation adapted to distributional brackets) produce anomaly terms that can be packaged into a h –valued Bxt with t ( Bxt )equal to the defect, giving Proposition .16. For spectral parameters, see the internalization in the next section. Jet–theoretic construction of CLS (existence on a cover) We sketch a practical construction of CLS data given a PDE E. 1. Prolongation ansatz on jets. Seek matrix–valued 1–forms A = L d x + M d t on E such that dHA+A∧A≡t(B) (mod IE)for some h–valued 2–form B=Bxt dx∧dt. 2. Solve modulo the ideal. Using the variational bicomplex, decompose d A into d H – and vertical parts and enforce the equalities componentwise on E ; this produces algebraic–differential conditions for the coefficients of L, M (and Bxt) as functions of jets [37–39]. 3. Descent on a cover. On a good cover {Ui} of X , solve the previous step patchwise; the difference between ( Ai, Bi )and ( Aj, Bj )on overlaps yields 1–gauge gij and 2–gauge Λ ij by the transformation formulas; coherent triple–overlap data hijk are then determined by Wahlquist–Estabrook compatibility conditions on E[36]. 9 4. Obstruction class. The nonabelian ˇ C ech 2–cocycle ωijk attached to ( gij, hijk ; Λ ij )is the descent obstruction. If [ ω ] = 0 in the nonabelian cohomology set, one obtains a global CLS; otherwise one still has patchwise CLS and a quantitative coherence defect measured by kωk (defined via any faithful linearization of H). Example .18 (KdV on the circle) . For KdV ut = 6 uux−uxxx , the Wahlquist–Estabrook prolongation yields a g = sl2 –valued A with B = 0 (ultralocal). One recovers the AKNS Lax pair on E and the standard monodromy invariants. Our framework reproduces this as a CLS with h= 0. Example .19 (Non–ultralocal sigma model) . Principal chiral models admit spatial Lax operators with derivative–of–delta Poisson brackets. Choosing h so that t : h→g injects the s –term and solving (21) – (22) on E produces a CLS with nontrivial Bxt and conserved corrected monodromy, consistent with hierarchy constructions developed by group–theoretic methods [69]. Semi–strict extensions If the structure 2–group is only semi–strict, the 2–curvature reads H = DAB + 1 6`3 ( A, A, A ), and the proofs above carry through with the obvious modification: the 2–holonomy picks an additional `3 –driven phase. This is relevant, e.g., when modeling Wess–Zumino terms in sigma models or stringy corrections. The jet–theoretic construction is unaffected; one simply allows an `3–term in the B–equation. Remark .20 (Topological sectors).On nontrivial spatial topology, the corrected monodromy lives in conjugacy classes twisted by H –holonomies; the class functions f used in integrals should be chosen on the appropriate twisted character variety. This creates no additional difficulty in practice. Summary A CLS is a principled, jet–level, higher–gauge refinement of Lax data that turns compatibility into coherence. Fake flatness enforces the categorified Lax equations, 2–flatness guarantees independence of choices, and higher holonomy supplies corrected, conserved quantities. The correspondence theorem shows that a global CLS is tantamount to integrability in the usual (hierarchy) sense, while the B –sector robustly accommodates non–ultralocal anomalies. NON-ABELIAN DESCENT AND THE DESCENT–OBSTRUCTION CRITERION This section develops the descent theory for Coherent Lax Structures (CLS) and establishes the Descent–Obstruction Criterion: local CLS data glue to a global CLS if and only if a canonically attached non-abelian cohomology class [ ω ]vanishes. We make the construction explicit at the level of Čech cochains and clarify how [ ω ]controls—quantitatively—the departure from integrability when it does not vanish. Mathematically, this is descent for principal 2–bundles with connection; physically, it asserts that global integrability is equivalent to categorical coherence of the local Lax charts. Throughout, the structure 2–group is the strict crossed module ( Ht −→ G, α )with Lie algebras ( ht −→ g, α ), and X is a smooth space–time manifold (in applications X = S1×I or S1×RN ). We assume familiarity with the local 2–connection data ( Ai, Bi )and the overlap fields ( gij, Λ ij )and hijk from the previous section. The Čech 2–nerve and non-abelian cochains Let U={Ui}i∈Ibe a good open cover of X. Set Ui0···ip:= Ui0∩ · · · ∩ Uipand denote the Čech nerve by the cosimplicial manifold U•:U0=G i Ui, U1=G i,j Uij, U2=G i,j,k Uijk, U3=G i,j,k,` Uijk`, . . . Face maps ∂r : Up→Up−1 delete the r –th index. A non-abelian 1–cochain with values in G is a collection {gij : Uij →G} and a twisted 2–cochain with values in H is a collection {hijk : Uijk →H} acted upon by the gij via α. We normalize by requiring gii =e,gij =g−1 ji , and hiij =hijj =e. 16 Summary Internalizing the spectral parameter elevates the Lax family to a single flat 2–connection on X× Σ. The corrected monodromy is constant along all isomonodromic directions, and its spectral curve is time–independent. In the ultralocal case, this reproduces the classical BA/Krichever machinery; in the presence of a B –field, spectral data live naturally in the world of twisted line bundles over the spectral curve, with gerbe holonomy accounting for non–ultralocal anomalies and restoring Poisson–commuting integrals for the corrected monodromy. NON–ULTRALOCAL SYSTEMS: THE MAILLET r/s ALGEBRA FROM THE B–SECTOR This section develops the Poisson geometry of non–ultralocal integrable field theories in the categorical framework. We explain how the h –valued B –field in a Coherent Lax Structure (CLS) geometrizes the derivative–of–delta anomalies in the canonical bracket of the spatial Lax operator and leads to the Maillet r/s algebra. We then show that the B –corrected parallel transport (and monodromy) has an ultralocal Sklyanin–type bracket, ensuring the involutivity of spectral invariants. Intuitively: the B –flux across the ribbon swept by the space loop resolves the “contact term” that would otherwise spoil the Poisson algebra of holonomies. Throughout, g is a finite–dimensional Lie algebra with invariant pairing h·,·i , and ht −→ g is the Lie algebra map of our crossed module. We consider a single time t and the spatial circle S1 = R/ 2 πZ ; λ, µ denote spectral parameters. Boldface subscripts 1,2indicate tensor legs: X1=X⊗Id,X2= Id ⊗X. Non–ultralocality and the Maillet bracket Let L ( x ; λ ) ∈g be the spatial Lax matrix (the x –component of A in a trivialization). In an ultralocal theory, the Poisson bracket is supported at coincident points with no derivatives of the delta: {L1(x;λ), L2(y;µ)}= [ r12(λ, µ), L1(x;λ) + L2(y;µ) ] δ(x−y),(39) for a classical r –matrix r12 ( λ, µ ) ∈g⊗g satisfying a classical Yang–Baxter–type condition. In many physically important models (principal chiral/sigma models, Landau–Lifshitz, superstrings), one instead has the Maillet bracket [131, 132]: {L1(x;λ), L2(y;µ)}= [ r12(λ, µ), L1(x;λ) + L2(y;µ) ] δ(x−y) + [ s12(λ, µ), L1(x;λ)−L2(y;µ) ] δ(x−y) + 2 s12(λ, µ)∂xδ(x−y), (40) with two kernels r12, s12 constrained by Jacobi to satisfy Maillet’s consistency equations (a coupled system of classical Yang–Baxter–type relations). The last term with ∂xδ is the hallmark of non–ultralocality and is the source of delicate regularization issues for Poisson brackets of path–ordered exponentials. Remark .40 (Why the ∂xδ term is a problem).Let U ( x, y ; λ )be the path–ordered exponential of L along the interval [ y, x ]. Formally differentiating {U1, U2} under the path–ordering yields boundary terms involving s12 evaluated at the endpoints; unless these anomalies are handled coherently, the bracket of monodromies becomes ill–defined or requires an ad hoc regularization [67, 113, 131]. The categorical cure: B–corrected transport In a CLS, the fake–flatness equation on (x, t)reads ∂tL−∂xM+ [L, M] = t(Bxt), so the h –valued 2–form component Bxt packages the anomaly in the xt –curvature into the image of t : h→g . To cure non–ultralocality at the level of parallel transport, we correct the usual path–ordered exponential by the H –valued surface holonomy across a thin ribbon swept by the interval as one changes (x, y)(or the spectral parameter). 17 Definition .41 ( B –corrected transporter and monodromy) . Let U ( x, y ; λ )solve ∂xU = −L ( x ; λ ) U , U ( y, y ) = Id . Let Hxy ( λ ) ∈H be the H –holonomy across a thin ribbon in the ( x, λ )or ( x, t )directions spanning the path; set e U(x, y;λ) := t Hxy(λ)−1U(x, y;λ)∈G. (41) For the circle, the B–corrected monodromy is f M(λ) := e U(x0+ 2π, x0;λ). The extra factor is trivial when B = 0 but exactly cancels the contribution of the ∂xδ –term to the bracket of holonomies. The next two subsections make this precise. A Poisson structure on the space of CLS fields Let F be the space of fields ( A, B )on S1 (at fixed time) obeying the CLS constraints (fake flatness along x , and 2–flatness with respect to ( x, λ )when the spectral parameter is present). Consider variations δAx , δBxλ tangent to F modulo 1– and 2–gauge transformations. We postulate the following pre–symplectic form on F(motivation below): Ω := ZS1DδAx∧D−1 xδAxEdx+ZS1DδAx∧t(δBxλ)Edx, (42) where D−1 x is the Green operator on mean–zero functions (well–defined on the constraint surface) and h·,·i is the invariant pairing. The first term is the standard (ultralocal) symplectic current; the second encodes a central extension by the B–sector. One checks: Lemma .42 (Gauge invariance and degeneracies) . Ωis invariant under 1– and 2–gauge transformations preserving the constraints; its kernel consists of infinitesimal gauge transformations and, in the non– ultralocal case, constant shifts of Axin the center of gpaired with flat shifts of Bxλ in ker t. Reducing by the degeneracies yields a Poisson bracket on a dense subspace of functionals O ( F ). In particular, the equal–time bracket of the Lax matrix can be computed from (42) by the usual Peierls/Faddeev–Takhtajan recipe [68, 69]. Theorem .43 (Derivation of the Maillet bracket from Ω) . Let L = Ax in a trivialization. The Poisson bracket {·,·}Ωinduced by (42) is of Maillet form (40), with kernels r12(λ, µ) = 1 2R12(λ, µ), s12(λ, µ) = 1 2S12(λ, µ),(43) where R arises from the ultralocal part (Green operator) and S is the tensor encoding the central extension by the B –term. The pair ( r, s )satisfies Maillet’s consistency relations provided the CLS 2–flatness holds in the (x, λ)directions. Idea of proof. Compute the Poisson brackets of linear functionals Rh1, Li and Rh2, Li by inverting the bilinear form defined by Ωon tangent vectors ( δAx, δBxλ ). The first term reproduces the ultralocal r –part with an antisymmetric kernel R . The second term contributes a symmetric kernel S and produces, after integration by parts, both the [ s, · ] δ and the 2 s ∂xδ terms. The Jacobi identity reduces to the flatness of ( A, B )on the ( x, λ )–cylinder, which is precisely the CLS 2–curvature condition, and to the (modified) classical Yang–Baxter relations for R,Sfamiliar from Maillet’s analysis. Remark .44 (Motivation for (42) ).The first term is standard (it appears e.g. as the inverse of the operator Dx on the space of currents). The second term is the group–valued counterpart: it is the differential of a Wess–Zumino–type functional and is responsible for central extensions; in quasi–Hamiltonian language, it is the piece that makes the boundary holonomy a group–valued moment map [71, 134]. Brackets of B–corrected transporters We now state the key structural result: the B –corrected transporter has an ultralocal Sklyanin bracket. For compactness, write r± 12(λ, µ) := r12(λ, µ)±s12(λ, µ). 18 Theorem .45 (Sklyanin bracket for e U ) . Let e U ( x, y ; λ )be defined by (41) . Then the Poisson brackets induced by Ωsatisfy {e U1(x, y;λ),e U2(x, y;µ)}=r+ 12(λ, µ)e U1(x, y;λ)e U2(x, y;µ)−e U1(x, y;λ)e U2(x, y;µ)r− 12(λ, µ),(44) and, for concatenated intervals, the brackets are compatible with composition. In particular, for the monodromy f M(λ)on the circle, {f M1(λ),f M2(µ)}= [ r12(λ, µ),f M1(λ)f M2(µ) ],(45) so class functions of f MPoisson–commute. Sketch. Differentiate the bracket of e U with respect to x and y , using (40) for L and the definition of e U . The anomalous endpoint terms proportional to s12 cancel against the variation of the ribbon holonomy t ( Hxy )(this uses the CLS transformation rule for B and the fact that t : h→g is a Lie algebra map). Integrating along the path produces (44) , with r± constant along the interval. For the circle, the endpoint contributions are equal, yielding (45) . Compatibility with composition follows from the group–like property of e U and the fact that r± satisfy the Sklyanin consistency relations (which are equivalent to Maillet’s equations and the CLS 2–flatness). Corollary .46 (Involutivity of spectral invariants) . Let ρ be any finite–dimensional representation of G . Then trρ(f M(λ))k,trρ(f M(µ))` = 0 for all k, ` ∈N, so the corrected spectral invariants are in involution. Consistency and Jacobi: Maillet relations from CLS flatness For completeness, we recall the Maillet consistency system ensuring Jacobi for (40): [r12 −s12, r13 +s13]+[r12 +s12, r23 +s23]+[r13 +s13, r23 +s23]=0,(46) [r12, s13]+[r13, s12]+[s12, s13] + cycl. = 0,(47) together with symmetry/antisymmetry constraints. In the present framework: Proposition .47 (CLS origin of Maillet relations) . If ( A, B )is fake–flat in ( x, t )and 2–flat in ( x, λ ), then the kernels ( r, s )extracted from Ωsatisfy (46) – (47) . Conversely, violation of 2–flatness produces precisely the obstructions on the right–hand sides (coboundaries controlled by the 2–curvature). Idea. Compute the Jacobiator of three linear functionals of L using (42) . The terms that would spoil Jacobi are proportional to the 2–curvature H = d B + A·B along ( x, λ )and vanish exactly when H = 0. The remaining components assemble into the algebraic relations (46)–(47). Time evolution and conservation Let H be an integrable Hamiltonian built from class functions of f M ( λ )(e.g. coefficients of its characteristic polynomial). Then the Hamiltonian vector field {H, ·} with the bracket (45) coincides with the Lax evolution ∂tf M = [ f M, N ]for some N depending on H (isospectral flow), and all the spectral invariants are conserved. Theorem .48 (Consistency of time evolution) . Assume ( A, B )solves a CLS and the Maillet bracket is induced by (42) . Then the t –evolution generated by H = H ( f M )preserves the bracket (45) and the CLS constraints; in particular, the family f M(λ, t)remains in the same conjugacy classes for all t. Idea. Use (44) and the r–matrix Lax equation to verify that the time derivative of the bracket equals the bracket of the time derivatives. Preservation of CLS constraints follows from compatibility of the Lax pair with fake–flatness and from 2–flatness in (x, λ). 19 Examples (a) Principal chiral model (PCM). For the PCM on a compact group G , the spatial Lax matrix L ( x ; λ ) = 1 1−λ2 ( J1−λJ0 )has the Maillet bracket with a nontrivial s –kernel s12 ( λ, µ ) ∝λ+µ 1−λ2C12 , where C12 is the split Casimir [ 113 , 131 ]. A CLS with h = g and t = Id captures the anomaly by a g –valued B–field Bxλ; the corrected monodromy f Mthen obeys (45) and its spectral invariants commute. (b) Landau–Lifshitz (LL) model. The classical LL equation has a non–ultralocal bracket and admits a Lax representation with rational spectral parameter. Choosing h = R (central) and t the embedding into the center of gyields a Bthat reproduces the scalar s–kernel; again f Mhas ultralocal bracket. Intuitive summary Non–ultralocality expresses itself as a failure of coherence for the naive holonomy functor: parallel transport along adjacent arcs does not compose cleanly due to contact terms. The B –field provides the missing 2–morphism (the ribbon holonomy) that restores coherence. In Poisson language, this promotes the Maillet r/s algebra of the raw Lax matrix to an ultralocal Sklyanin bracket for the corrected transporters, and hence ensures that spectral invariants of the corrected monodromy Poisson–commute. A COHERENCE–BY–DESCENT SOLVER FOR NONLINEAR PDES This section develops a practical and mathematically principled coherence–by–descent solver for nonlinear PDEs based on the Coherent Lax Structure (CLS). The solver takes as input a PDE E on X = S1×I (or S1×RN ), a crossed module ( ht −→ g, α ), and a good cover U = {Ui} of X . It outputs (i) patchwise 2–connection data ( Ai, Bi )realizing the categorified zero–curvature equations to high accuracy, (ii) 1– and 2–gauge gluing data ( gij, Λ ij, hijk ), (iii) a certificate in the form of residual norms and a descent obstruction estimate kωk, and (iv) global, gauge–invariant observables (corrected monodromies) with a–posteriori error bars. Intuitively, we solve by enforcing coherence: rather than preserving many individual conservation laws, we minimize a global functional that measures fake curvature, 2–curvature, and gluing incoherence. We first formulate the continuous variational problem and its Euler–Lagrange system, prove existence of minimizers modulo gauge, then describe a discretization based on finite element exterior calculus (FEEC) / discrete exterior calculus (DEC) with group–valued holonomies. We propose an alternating augmented–Lagrangian / ADMM algorithm and state convergence guarantees in convex regimes and local convergence in the nonconvex, Lie–group setting. Finally, we provide a–posteriori certification bounds that propagate to spectral invariants. Continuous optimization problem Fix a Riemannian metric on X to define L2 norms via the Hodge star ? . On a patch Ui , write Ai = Li d x + Mi d t and Bi = Bi,xt d x∧ d t . Let FAi = d Ai + 1 2 [ Ai∧Ai ]and DAiBi = d Bi + Ai·Bi . Denote fake curvature and 2–curvature Fi:= FAi−t(Bi)∈Ω2(Ui,g),Hi:= DAiBi∈Ω3(Ui,h). We also encode the PDE constraints by a residual Ri ( u ; Ai, Bi ), which measures the mismatch between the jet–level dynamics of E (restricted to solutions or numerical approximations thereof) and the auxiliary system induced by ( Ai, Bi ). In practice Ri is built from the difference between ∂tLi−∂xMi + [ Li, Mi ] and t(Bi,xt)on the numerical solution uof Eon Ui. Definition .49 (Global coherence functional) . Given positive weights ( α, β, γ, η )and an overlap penalty functional R, define Φ{Ai, Bi},{gij,Λij},{hijk}:= X iαkFik2 L2(Ui)+βkHik2 L2(Ui)+ηkRik2 L2(Ui) +γX i<j RAi, Bi;Aj, Bj;gij,Λij.(48) 20 The overlap penalty Renforces the gluing relations Aj−Adgij (Ai)−gijdg−1 ij +t(Λij)=0on Uij,(49) Bj−gij ·Bi+ dΛij +Aj·Λij = 0 on Uij,(50) and the triple–overlap coherence for Λij: Λik −Λij −gij ·Λjk +h−1 ijkdhijk + (Ai·hijk)=0on Uijk.(51) A standard choice is a least–squares penalty: R=kAj−Adgij (Ai)−gijdg−1 ij +t(Λij)k2 L2(Uij )+kBj−gij ·Bi+ dΛij +Aj·Λijk2 L2(Uij ). Remark .50 (Gauge symmetry).The functional Φis invariant under local 1– and 2–gauge refinements ( ki, λij )on each patch and overlap provided hijk are updated according to the crossed–module action. Numerically, we either quotient by gauge (fixing a Coulomb–type gauge) or incorporate gauge freedom by augmented Lagrangians with gauge generators as constraints. Euler–Lagrange system and existence Let us write the formal Euler–Lagrange equations for ( Ai, Bi )keeping ( gij, Λ ij, hijk )fixed, and vice versa. For clarity, suppress the patch index iand denote by h·,·i the L2pairing. Lemma .51 (First variations).The first variation w.r.t. (A, B)yields δAΦ=2αhD∗ AF, δAi+ 2βh(∂AH)∗H, δAi+ 2ηh∂AR, δAi+overlap terms, δBΦ = −2αht∗F, δBi+ 2βhD∗ AH, δBi+ 2ηh∂BR, δBi+overlap terms, where D∗ A is the L2 adjoint of DA and t∗ is the adjoint of t with respect to chosen inner products on g,h . Proposition .52 (Euler–Lagrange system (weak form)).Critical points of Φsatisfy on each Ui 2α D∗ AiFi+ 2β(∂AiHi)∗Hi+ 2η(∂AiRi)∗Ri=overlap source,(52) −2α t∗Fi+ 2β D∗ AiHi+ 2η(∂BiRi)∗Ri=overlap source,(53) while (49)–(51) hold weakly on overlaps (or are enforced via multipliers). Theorem .53 (Existence modulo gauge) . Assume: (i) X compact, (ii) t injective and α, β, γ, η > 0, (iii) Ri affine in ( Ai, Bi )for fixed u . Fix Coulomb–type gauges d ∗Ai = 0 on Ui and a 2–gauge d ∗Bi = 0. Then Φattains a minimizer in the gauge class G/∼ , and any minimizing sequence has a weakly convergent subsequence to a gauge–equivalence class of minimizers. Sketch. Direct method in the calculus of variations: coercivity follows from injectivity of t and the Poincaré inequalities under the chosen gauges, lower semicontinuity from weak lower semicontinuity of the L2 norm and continuity of the affine Ri . Compactness of embeddings on compact X yields weak compactness; pass to the limit using weak lower semicontinuity. Gauge fixing ensures uniqueness modulo residual discrete symmetries. Remark .54 (Nonlinear Ri ).If Ri is nonlinear (e.g. includes products of coefficients of A with u ), one may need growth/continuity hypotheses to obtain semicontinuity; in practice we linearize Ri around the current u(Gauss–Newton step), see Algorithm 1. Discretization: FEEC/DEC with group–valued holonomies Let Th be a shape–regular triangulation of X with mesh size h . Denote by Wk h the Whitney k –form space on Th (lowest–order FEEC), and by d h the discrete exterior derivative (coboundary) [ 72 – 74 ]. We adopt two equivalent representations: 21 •Infinitesimal (FEEC) variables: Ah∈W1 h⊗g,Bh∈W2 h⊗h; discrete curvatures Fh= dhAh+1 2[Ah∧hAh]−t(Bh),Hh= dhBh+Ah·hBh, with ∧hand ·hthe Whitney product approximations. • Holonomy (lattice) variables: edge variables Ue∈G approximating 1–holonomy Pexp ReA , face variables Vf∈Happroximating 2–holonomy exp RfB; discrete fake curvature on a face f: Flat f:=  Y e⊂∂f U(e) e −t(Vf)∈G, with orientation signs (e). This is standard in lattice gauge theory [75, 76]. Remark .55 (Gauge and 2–gauge).In the Whitney setting, gauge acts by Ah7→ g−1 hAhgh + g−1 h d hgh with gh nodal (piecewise linear) G –valued maps; in the holonomy setting, by vertex–based conjugation on edges and face–based 2–gauge on Vf . Both are compatible with the discrete curvatures and with the overlap relations (represented on the mesh intersections). Definition .56 (Discrete functional) . Let h·,·ih denote the FEEC mass inner products (Whitney L2 ); define Φh:= αkFhk2 h+βkHhk2 h+ηX i kRi,hk2 h+γX i<j Rh, where Rh discretizes (49) – (50) . In the lattice form, replace norms by sums of squared distances in G and Husing bi–invariant Riemannian metrics (logarithm map). Proposition .57 (Consistency) . Assume A, B smooth and Th quasi–uniform. Then Φ h ( Ah, Bh ) → Φ( A, B )as h→ 0whenever Ah→A , Bh→B in H1 (Whitney) or Ue, Vf converge in the strong sense of holonomies. Remark .58 (Whitney vs. holonomy).FEEC variables are convenient for Newton–type solvers; holonomy variables preserve group structure exactly and are natural for large deformations. A hybrid scheme uses Ue= exp(ReAh)and Vf= exp(RfBh)to keep both approximations consistent. The algorithm: alternating augmented Lagrangian / ADMM We enforce the overlap constraints and gauges by multipliers and split the problem across patches and variables. 22 Algorithm 1 Coherence–by–Descent Solver Require: PDE E, cover {Ui}, mesh Th, weights (α, β, γ, η), penalty ρ > 0, tolerances (εfake, ε2curv, εglue). 1: Initialize: Solve E numerically on each Ui to obtain ui . Initialize ( Ai, Bi )by local prolongation/linearization; set gij = Id,Λij = 0,hijk = Id. 2: repeat 3: for each patch iin parallel do 4: (A) Update (Ai, Bi)by minimizing the local augmented functional Φaug i,h =αkFi,hk2 h+βkHi,hk2 h+ηkRi,hk2 h+X j∼i ρ 2kCij (Ai, Bi)k2 h+hΛ(1) ij , Cij ih where Cij stacks the residuals of (49) – (50) , and Λ (1) ij are Lagrange multipliers; use Gauss–Newton / trust–region on FEEC variables or Riemannian Gauss–Newton on holonomies. 5: end for 6: (B) Update overlap fields ( gij , Λ ij )by minimizing Pi<j Rij,h with multipliers; project gij to G via the exponential map. 7: (C) Update triple–overlap data hijk to reduce the discrete defect ωijk` ; set hijk ←hijk exp ( − Π hlog ωijk` ) on each quadruple overlap cell. 8: (D) Update multipliers & penalties: Λ(1) ij ←Λ(1) ij +ρ Cij . Optionally increase ρ. 9: (E) Gauge correction: (Coulomb) project Ai to d ∗Ai = 0; (2–gauge) project Bi to d ∗Bi = 0; adjust (gij ,Λij , hijk)accordingly. 10: (F) Diagnostics: compute kFikh , kHikh , overlap residuals, and the obstruction ω on quadruple overlaps; form kωk∞via matrix logarithms. 11: until maxikFikh≤εfake,maxikHikh≤ε2curv, and overlap residuals ≤εglue 12: Output: ( Ai, Bi ),( gij , Λ ij , hijk ), obstruction estimate kωk , and corrected monodromies on S1 with error bars. Remark .59 (Intuition).Step (A) flattens curvature on each patch; (B) aligns patches; (C) eliminates 3– cocycle defects; (E) keeps iterates in a regular gauge to avoid degeneracy. The method is a block–structured extension of augmented Lagrangians and ADMM to the 2–gauge setting [77, 78]. Convergence guarantees The functional Φ h is nonconvex due to the Lie–group structure and the bilinear terms. We collect guarantees in regimes commonly met in practice. Theorem .60 (Convex regime) . Assume abelian structure ( g,h abelian), linear Ri,h , and fixed overlaps ( gij, Λ ij, hijk ). Then Φ h is strictly convex modulo gauge, and Algorithm 1 reduces to a linearly constrained quadratic program; the alternating augmented–Lagrangian method converges globally at least sublinearly, and linearly under standard constraint qualifications [77, 79]. Theorem .61 (Local convergence (nonabelian case)) . Let ( A∗, B∗, g∗, Λ ∗, h∗ )be a regular CLS (fake and 2–flat with [ ω ] = 0), and suppose the initial guess lies in a neighborhood where the linearized constraints satisfy a uniform inf–sup condition. Then for h small enough and ρ large enough, one step of (A)–(E) is a contraction in a suitable product metric, and Algorithm 1 converges locally (at least Q –linearly) to the discrete CLS orbit. Idea. Linearize the KKT system of the augmented functional at the solution; invertibility (modulo gauge) follows from ellipticity of D∗ ADA and injectivity of t , plus the inf–sup condition for the overlap constraints (a trace estimate). Riemannian Gauss–Newton yields local contraction; the ADMM splitting preserves contraction for ρlarge [78]. A–posteriori certification and error propagation Let b Ai,b Bibe the computed minimizers and bωthe measured obstruction. Define the global residual E2 tot := X ikb Fik2 L2(Ui)+kb Hik2 L2(Ui)+kRik2 L2(Ui)+X i<j Rij(b·) + klog bωk2 ∞. 23 Theorem .62 (Certification) . Under the hypotheses of Theorem .53 and small residual regime, there exists a true CLS (A†, B†)on a refined cover (or the same if bωis exact) such that, up to gauge, X ikA†−AikH1(Ui)+kB†−BikH1(Ui)≤CEtot, and for any finite–dimensional representation ρand class function f, f(f M†)−f(f Mnum)≤Cf,ρ Etot. Idea. Stability of elliptic systems (52) – (53) with small right–hand sides plus small overlap defects yield H1 bounds (by standard compactness and trace inequalities). The observable estimate follows by Fréchet differentiability of f◦f M with respect to ( A, B )and the nonabelian Stokes formula controlling the effect of bωon holonomy. Complexity and implementation notes Let N1, N2 be the numbers of edges and faces per patch; the dominant costs per outer iteration are: (i) local Gauss–Newton solves: ˜ O ( N1 + N2 )per patch with multigrid preconditioning (Whitney mass/stiffness structure), (ii) overlap updates: ˜ O ( |∂Ui| )per interface, (iii) hijk updates on triple overlaps: linear in the number of overlap cells. Using block–diagonal preconditioners (Hodge–Laplace) and parallelization across patches yields near–ideal scaling. Remark .63 (Boundary conditions and defects).Boundaries (or point defects) are handled by adding boundary terms to Φand enforcing jump conditions through the overlap penalties on narrow patches around the defect. The obstruction ω reveals whether defects can be globally absorbed (integrable defects) or produce controlled drift (cf. Section ). Minimal working examples (i) KdV on S1 .Choose g = sl2 , h = 0. The solver with β = 0 and η > 0recovers an AKNS Lax pair; kb Fk drops to discretization error and [bω] = 0. (ii) Principal chiral model. Choose g = Lie ( G ), h = g , t = Id . Initialize Bxλ from the s –kernel; the solver reduces kb Hk and overlap defects; corrected monodromy invariants commute up to Etot. Intuition and takeaway The coherence–by–descent solver treats integrability as a constraint satisfaction problem: we do not track each conservation law, we directly minimize a functional whose zero set is exactly the CLS (fake flat +2–flat + coherent gluing). The B –sector absorbs non–ultralocal anomalies; descent ensures that local linear problems glue globally; certificates quantify how close the computation is to a genuinely integrable configuration and furnish trustworthy bounds on spectral invariants. CASE STUDIES AND BENCHMARKS We illustrate the Coherent Lax Structure (CLS) framework and the coherence–by–descent solver on representative models. The goals are: • validate that, in ultralocal systems, the solver reaches (up to discretization) a global CLS with [ ω ]=0 and recovers the classical monodromy invariants; • demonstrate that, in non–ultralocal systems, the B –sector removes endpoint anomalies and restores an ultralocal Sklyanin bracket for the corrected monodromy, with conserved spectral invariants; • quantify near–integrability in weakly nonintegrable regimes (forcing, dissipation) via the coherence defect kωkand show that drifts of spectral invariants scale with kωkas predicted. 24 We treat four cases: KdV on S1 (ultralocal), focusing/defocusing NLS on S1 (ultralocal AKNS), the principal chiral model (PCM; non–ultralocal), and a dissipative deformation (KdV–Burgers) as a test of the near–integrable calculus. Unless stated otherwise, space is S1 = R/ 2 πZ with coordinate x , time t∈ [0 , T ]; we use Fourier spectral discretization in x and a symplectic time integrator for the physical PDE when needed. The CLS unknowns are discretized as in Section , FEEC on a rectangular mesh for ( x, t )and holonomies for validation. All norms are L2unless indicated. Benchmark metrics and diagnostics For each run we report: •patchwise residuals kb Fik,kb Hik; the maximum over patches is denoted kb Fkmax,kb Hkmax; • overlap residual Rmax and the obstruction magnitude klog bωk∞ (matrix norm in a faithful representation of H); • monodromy invariants: for a representation ρ , we track Ik ( λ ) := trρ ( f M ( λ )) k for a small set of k , and report maxt∈[0,T ]|Ik(t, λ)−Ik(0, λ)|; •computational cost: wall time and iteration counts; these are secondary and reported qualitatively. The certification Theorem .152 implies that invariant drifts are O ( Etot )where Etot aggregates the quantities above. KdV on the circle (ultralocal) Model and Lax data. KdV is ut+ 6u ux+uxxx = 0, x ∈S1,(54) with classical Lax pair L ( λ ) = 0 1 u−λ0 and M ( λ ) = −1 2uxu+λ −1 2uxx−(u−λ)(u+λ)1 2ux in g = sl2 ; B≡ 0 (ultralocal). The CLS is a G=SL2–bundle with flat connection A=Ldx+Mdton E(Section ). Setup. Initial data: (i) cnoidal wave u ( x, 0) = cn2 ( κx ; m )mapped to 2 π –periodicity; (ii) single–soliton on a large torus remapped to 2 π by Galilean transform. Spectral resolution Nx∈ [2 10, 2 12 ]; time window T = 10; 2 patches in time, 4 in space (overlaps of width π/ 4). Solver weights ( α, β, η, γ ) = (1 , 0 , 10 −2, 10), β= 0 since B= 0. Results. The solver converges in 10–20 outer iterations with local Gauss–Newton steps reaching machine precision in F: kb Fkmax ≈10−12–10−10,Rmax ≈10−10,klog bωk∞≈10−12. Monodromy invariants for ρ = fund and k = 1 , 2are conserved to spectral accuracy; the drift scales with Nx as O ( N−p x )with p≈ 14 (double precision saturation) consistent with spectral differentiation [ 80 , 81 ]. This agrees with Theorem .152 with Etot dominated by discretization. Remark .64 (Spectral curve check).For cnoidal data the measured characteristic polynomial χ ( µ, λ ) = det ( ρ ( M ( λ )) −µId )yields a hyperelliptic relation µ + µ−1 = 2 T ( λ )with T close to a degree–2 g+ 1 polynomial; zeros agree with the finite–gap spectral bands up to O(10−10)[82, 83]. AKNS: focusing/defocusing NLS on S1(ultralocal) Model and Lax data. Consider iqt+qxx ±2|q|2q= 0,(“+” defocusing, “–” focusing).(55) The AKNS Lax pair in g=sl2(C)is L(λ) = −iλ q ∓¯q iλ, M(λ) = −2iλ2∓i|q|22λq +iqx ∓(2λ¯q−i¯qx) 2iλ2±i|q|2. Again B= 0; CLS reduces to flatness of A=Ldx+Mdt. 25 Setup. Initial data: (i) plane wave q ( x, 0) = ρeiκx (defocusing), (ii) N –soliton train (focusing) synthesized via Darboux. Nx∈ [2 10, 2 12 ], T = 10, 3 patches in time, 4 in space. We use the solver with (α, β, η, γ) = (1,0,10−3,10). Results. The solver attains kb Fkmax ≈10−11–10−9,Rmax ≈10−9,klog bωk∞≈10−12. Corrected monodromy equals the raw monodromy ( B = 0). Spectral invariants Ik ( λ )(for k = 1 , 2 , 3) drift by . 10 −9 over T = 10. In the focusing case, we observe the expected recurrence of an N –soliton train; the CLS invariants remain constant through near–collisions (up to discretization), confirming robustness. Remark .65 (Nonlinear Fourier diagnostics).For defocusing data (periodic), the quasimomentum p ( λ ) = 1 2πi log M11 ( λ )computed from the CLS monodromy matches the zonal structure of the periodic scattering transform and exhibits stationary bands/gaps [84, 85]. Principal chiral model (non–ultralocal) Model and bracket. For g ( x, t ) ∈G compact simple, the equations are ∂tJ0−∂xJ1 + [ J0, J1 ]=0with Jµ=g−1∂µg. The Lax operator is L(x;λ) = 1 1−λ2(J1−λJ0), M(x;λ) = 1 1−λ2(J0−λJ1), with the Maillet bracket {L1(x;λ), L2(y;µ)}= [r12, L1+L2]δ+ [s12, L1−L2]δ+ 2s12∂xδ. Here r12(λ, µ) = C12 2 λ2+µ2 (λ2−1)(µ2−1) 1 λ−µand s12(λ, µ) = C12 2 λ+µ (λ2−1)(µ2−1) (schematic) [113, 131]. CLS choice. Take h = g and t = Idg ; set Bxλ so that t ( Bxλ )reproduces the s –kernel contribution along ribbons (Section ). The corrected transporter e U then satisfies a Sklyanin bracket and f M has ultralocal r–type algebra (Theorem .45). Setup. We consider G = SU (2), random smooth initial data near the identity, Nx = 2 11 , T = 5. The solver is run on 4 spatial patches and 4 time patches with (α, β, η, γ) = (1,1,10−3,10). Results. The raw monodromy M (without B ) exhibits nontrivial drift in spectral invariants: maxt|I2 ( t, λ ) −I2 (0 , λ ) | ∼ 10 −3 for several λ , consistent with non–ultralocal anomalies. In contrast, for the corrected f Mwe obtain kb Fkmax ≈10−8,kb Hkmax ≈10−8,klog bωk∞≈10−10, and invariant drifts .10−7(dominated by discretization), in line with Theorem .45 and Theorem .152. Remark .66 (Poisson verification).We verified numerically the Sklyanin bracket (45) by perturbing initial data and computing finite–difference approximations of Poisson brackets of I2 ( λ )and I3 ( µ ); values are below 10−6across a grid of (λ, µ), while the uncorrected case yields O(10−3)spurious values. A dissipative deformation: KdV–Burgers (near–integrable) Model. Add viscosity ν > 0: ut+ 6u ux+uxxx =νuxx.(56) The model is not integrable for ν6 = 0. We run the CLS solver with h = 0 (no B ) and measure the coherence defect [ω]and the drift of KdV invariants. Prediction. Proposition .29 (Section ) predicts  d dtIk(λ)≤Ckωk∞+O(overlap size)and hence |Ik(t, λ)−Ik(0, λ)|.Ckωk∞t. We expect kωk∞=O(ν)for small ν. 32 Idea. Reduce to a matrix RHP via ρ ; the operator Cw has small norm on L2 ( C )for ε small; invertibility follows by Neumann series, yielding µ∈Id + L2 ( C ). Then reconstruct Γby the Cauchy transform; uniqueness follows from Liouville’s theorem. Remark .93 (Beyond small–norm).Standard extensions cover piecewise constant jumps with a finite number of poles (solitons) via residue conditions (cf. Fokas–Its–Kitaev [ 127 ] and Zhou [ 100 ]). The twist Θsimply multiplies the jump by t(Θ); poles are handled identically since t(Θ) is analytic across them. Reconstruction of the CLS (A, B)from Γ Let ( x, t1, t2, . . . )be commuting times and let Φ( λ ; t )be a fixed bare phase with values in g (typically a polynomial in λ times constant elements that generate the hierarchy). Define the bare 1–connection A0 = Pi∂ti Φ d ti and the bare 2–connection B0 = 0 on X× ( CP1\C )so that ( A0, B0 )is flat and A0 is meromorphic in λwith prescribed singularities (poles at ∞or punctures). Assume the TRHP data depend on (x, t)by conjugation with the bare phase: J(λ;x, t) = e−Φ(λ;t)J0(λ)eΦ(λ;t),Θ(λ;x, t) = e−Φ(λ;t)·Θ0(λ),(59) where the dot denotes the G –action on H . Let Γ( λ ; x, t )solve the TRHP for ( J, Θ) and define the dressed fields on Xby the standard gauge transform: A:= ΓA0Γ−1−(dΓ) Γ−1∈Ω1(X, g),(60) and (initially) set B to zero away from C in the spectral plane. The following statement shows how B is reintroduced from the twist on the contour. Theorem .94 (Reconstruction).Let Γsolve the TRHP with isospectral dependence (59). Then: (i) For each fixed λ /∈C , the pair ( A ( · ; λ ) , B ( · ; λ ) ≡ 0) is a (slice of a) CLS on X (fake–flat and 2–flat in the (x, t)directions). (ii) The jump of Γacross C induces an H –valued 2–holonomy across any ribbon R⊂X×C transverse to C with boundary on the x –loop, equal to Θ(pushed to G by t ). In the distributional sense on X×C , t(Bxλ) = ∂λΓ∂xΦ Γ−1−∂xΓ∂λΦ Γ−1+ [ Γ∂xΦΓ−1,Γ∂λΦΓ−1]δC+ (regular),(61) with the total flux across a small transverse strip equal to t(Θ). (iii) Consequently, the B –corrected monodromy along x is independent of ( x, t )and of the choice of spanning ribbons (categorified isospectrality). Sketch. (i) Away from C ,Γis analytic in λ and (60) is a pure gauge of the flat A0 , hence flat. (ii) The jump condition Γ + = t (Θ) J Γ − implies that the nontriviality of A when transported across C is precisely compensated by the t (Θ) factor; using the nonabelian Stokes theorem for 2–connections on X×C produces (61) . (iii) follows from the cancellation of endpoint terms by the B –holonomy as in Theorem .45. Remark .95 (Ultralocal slice).When Θ ≡eH , we recover the classical RHP reconstruction ( B = 0), with Agiven by the standard dressing formulas and Mconserved in t. Asymptotics and explicit field formulas Assume Φ(λ;t)has a simple pole at ∞, Φ(λ;t) = N X n=0 Tnλntn, Tn∈g, and that Γ(λ)admits a Laurent expansion Γ(λ) = Id + Γ1λ−1+ Γ2λ−2+· · · at ∞. Then from (60): L:= Ax= ΓT1Γ−1−∂xΓ Γ−1, M(n):= Atn= ΓTnΓ−1−∂tnΓ Γ−1.(62) 33 Projecting to the coefficient of λ−1gives local expressions for the PDE fields. For instance, in AKNS, q(x, t) = 2 (Γ1)12,¯q(x, t) = ∓2 (Γ1)21, as usual. The twist affects neither these local formulas nor the asymptotic expansion, but it will enter the Poisson structure through B(Section ). Dressing and Bäcklund transformations Dressing by rational loops produces new solutions from old ones. The twist introduces a categorified dressing that simultaneously updates the jump and the H–twist. Definition .96 (Twisted dressing) . Let g ( λ )be a rational G –valued loop meromorphic off C with g ( ∞ ) = Id , and let Λ( λ )be an H –valued 1–cochain on C (smooth along arcs) such that t (Λ) is the jump of gacross C. Define new data J]:= g+J g−1 −,Θ]:= g+·Θ·Λ, and form the dressed solution Γ]:= gΓ. Proposition .97 (Bäcklund transforms and permutability) . Twisted dressing preserves the TRHP: ( J], Θ ] )is coherent and Γ ] solves the new TRHP. If g1, g2 (with Λ 1, Λ 2 ) are two dressings with disjoint pole sets, then Γ(12) =g2g1Γ = g1g2Γ·Ξ,Θ(12) =g2·(g1·Θ) ·(Λ1·(g1·Λ2)), where Ξ ∈H is a 2–cocycle determined by the crossed–module Peiffer law. If Ξ = eH (e.g. in the ultralocal case or when the actions commute), the usual permutability holds exactly. Idea. Jump relations are preserved by multiplicative conjugation; twists compose according to the 2–gauge transformation law (cf. (25) – (26) in earlier sections). The second statement tracks the failure of strict commutativity: it is precisely the Peiffer 2–cocycle, which becomes trivial in strict/central cases. Example .98 (One–soliton dressing with twist) . Take g ( λ ) = Id + λ0−¯ λ0 λ−λ0P , P a rank–one projector in the AKNS representation. Choose Λ ≡eH (ultralocal) or a phase Λ( λ )supported on a small arc (non–ultralocal). The reconstructed q ( x, t )is the usual one–soliton; the twist modifies only the Poisson structure and the corrected monodromy, not the profile. Isomonodromic deformations and τ–function (categorified) The dependence (59) produces isospectral flows. If, instead, parameters in J0 (or the contour C ) vary, one obtains isomonodromic dynamics governed by a closed one–form which is the categorified Jimbo–Miwa–Ueno form (cf. Section ). We record the reconstruction formula. Proposition .99 (Categorified JMU form from the TRHP) . Let Γ( λ ; s )solve a TRHP with deformation parameters s(pole positions, residues, contour shape). Then the one–form Θcat := 1 2πi ZC Tr Γ−1 −∂sΓ−∂λb Jb J−1dλ−ZCt−1(∂sb Jb J−1),ribbon reference is closed on the deformation space, and d log τcat = Θ cat defines a τ –function. For Θ ≡eH , the second term vanishes and one recovers the classical JMU form. Idea. Differentiate the solution operator of the TRHP with respect to s as in [ 102 , 103 , 111 ]. The twist contributes an extra boundary/ribbon term that is exact by flatness of the 2–connection across C ; closedness follows. 34 From TRHP to commuting Hamiltonians Finally we tie the TRHP to the Hamiltonian picture. Take the corrected monodromy f M ( λ )built from (A, B)and let Hn,k := 1 kTrρ(f M(λ))kλnbe the coefficient of λ−nin its expansion at ∞. Theorem .100 (Commuting Hamiltonians from TRHP) . Under the hypotheses above, the functions {Hn,k} Poisson–commute with respect to the Poisson bracket induced by the pre–symplectic form (42) (Section ), and their Hamiltonian flows coincide with the isospectral flows generated by Φin (59). Idea. The bracket for f M is of ultralocal Sklyanin type (Theorem .45), which implies involutivity of class functions. The reconstruction formulas (62) match the Lax vector fields with the Hamiltonian vector fields generated by the Hn,k, as in the standard r–matrix formalism. Summary The twisted Riemann–Hilbert problem provides a unified analytic mechanism to generate CLS data from spectral information. The H –twist produces precisely the corrections needed to accommodate non–ultralocal structures, while leaving the local field reconstruction intact. Dressing becomes a 2–gauge enhanced operation, with Bäcklund permutability controlled by the Peiffer 2–cocycle. Isomonodromic deformations and τ –functions acquire natural B –dependent corrections. Altogether, the TRHP completes the categorified inverse scattering transform promised in the introduction. FOUNDATIONS OF COHERENT LAX STRUCTURES (CLS) This section collects the foundations of the Coherent Lax Structure (CLS). We fix a strict crossed module of Lie groups Ht −−→ G, with a smooth action α:G×H→H(we write g·h:= α(g, h)) satisfying the Peiffer identities t(g·h) = g t(h)g−1, t(h)·h0=hh0h−1,∀g∈G, h, h0∈H. (63) Its differential is a crossed module of Lie algebras ht −→ g with induced action (also denoted by a dot). The corresponding (strict) Lie 2–group is denoted G. 2–connections, curvatures and gauge transformations Let Xbe a smooth manifold (in applications X=S1×Ior S1×RN). Definition .101 (2–connection, fake curvature and 2–curvature) . A G –2–connection on X is a pair (A, B)with A∈Ω1(X, g), B ∈Ω2(X, h). Its fake curvature and 2–curvature are F:= FA−t(B)∈Ω2(X, g),H:= dB+A·B∈Ω3(X, h),(64) where FA= dA+1 2[A∧A]and A·Bis the infinitesimal action.[152] Remark .102 (Bianchi identities).One has DAF + t ( H ) = 0 and DAH = 0, where DA = d + [ A, · ]is the covariant derivative and Aacts on hvia the differential of α. Definition .103 (1–/2–gauge transformations) . A1–gauge g : X→G and a 2–gauge Λ ∈ Ω 1 ( X, h )act on (A, B)by A0= Adg(A) + gdg−1−t(Λ),(65) B0=g·B−dΛ + A0·Λ.(66) The curvatures transform as F0= Adg(F),H0=g·H.(67) 35 Verification of (67). A direct computation using (63) , the Leibniz rule and that t is Ad –equivariant. In particular, the −t(Λ) shift in A0cancels against the dΛ + A0·Λterm in B0inside FA0−t(B0). Definition .104 (Categorified flatness) . A2–connection ( A, B )is categorically flat on a submanifold Y⊆Xif F|Y= 0,H|Y= 0.(68) Holonomy 2–functor and corrected parallel transport Let Π ≤2 ( X )be the path 2–groupoid of X : objects are points of X ,1–morphisms are thin homotopy classes of piecewise smooth paths, and 2–morphisms are thin homotopy classes of bigons. The strict Lie 2–group Gdefines a one–object 2–groupoid BG. Theorem .105 (Holonomy 2–functor) . Let ( A, B )be a 2–connection on X with F = 0. Then there is a well–defined holonomy 2–functor HolA,B : Π≤2(X)−→ BG which assigns to each path γ an element HolA ( γ ) ∈G (the A –parallel transport) and to each bigon Σ : γ0⇒γ1an element HolB(Σ) ∈Hsuch that HolA(∂Σ) = t HolB(Σ),(69) and composition of paths/bigons maps to multiplication in G / H . If furthermore H = 0, the assignment is independent of thin homotopies and respects vertical and horizontal compositions strictly. Sketch (nonabelian Stokes for 2–connections). Fix a bigon Σ : [0 , 1] 2→X . Let U ( s, t )solve ∂tU = −A ( ∂t Σ) U with U ( s, 0) = Id ; then g ( s ) := U ( s, 1) is the boundary holonomy along γs : t7→ Σ( s, t ). One shows (e.g. [ 138 , 139 ]) that ∂sg g−1 = −tR1 0W ( s, t ) d t where W solves a linear equation driven by B ( ∂s Σ , ∂t Σ), and with F = 0 the A –terms cancel. Integrating in s gives (69) . Vanishing H implies independence under deformations of Σfixing the boundary. Remark .106 (Intuition).Equation (69) says: the failure of A –holonomy to be homotopy invariant is exactly measured by the H –valued surface holonomy. When both F and H vanish, 1– and 2–holonomy define a genuine 2–functor. Corrected transporters and monodromy. Let γ : [0 , 1] →X be a path and let R be a thin ribbon in X× [0 , 1] whose boundary is the graph of γ together with a fixed reference path (or, for a loop, a narrow strip swept by a small base–point move). Define the B–corrected transporter e U(γ) := t HolB(R)−1HolA(γ)∈G. (70) When γis a loop based at x0, the B–corrected monodromy is f M(γ) := e U(γ). Proposition .107 (Ribbon independence and functoriality) . If F = H = 0 on a neighborhood of the ribbon, e U ( γ )is independent of the choice of R and is invariant under thin homotopies of γ fixing endpoints. Moreover, concatenation of paths maps to multiplication of corrected transporters. Proof. Changing the ribbon by a thin homotopy changes HolB by a factor in H whose t –image equals the change in HolA by (69) ; they cancel in (70) . Composition is immediate from multiplicativity of holonomy and additivity of surface holonomy under gluing. CLS and zero curvature Lax equations We now connect to integrable PDEs. Let X = S1×I with coordinates ( x, t )and suppose ( A, B ) depends (possibly through jets of fields) on (x, t), with A=L(x, t;λ) dx+M(x, t;λ) dt, B =Bxt(x, t;λ) dx∧dt, where λis a spectral parameter (possibly suppressed). 36 Definition .108 (Coherent Lax Structure (CLS)) . ACLS for a PDE E on X is a 2–connection ( A, B ) on X (with spectral dependence as desired) such that, along solutions of E , the categorified flatness conditions hold: F=FA−t(B)=0,H= dB+A·B= 0.(71) Remark .109 (Lax/zero curvature with correction).Expanding (71) in (x, t)yields ∂tL−∂xM+ [L, M] = t(Bxt),(72) ∂tBxx −∂xBxt +L·Bxt −M·Bxx = 0 (here Bxx = 0), so (72) is the usual Lax/zero curvature equation modified by t ( Bxt ). In ultralocal models one has B≡ 0. Theorem .110 (Isospectrality of corrected monodromy) . Let ( A, B )be a CLS on X = S1×I . Fix a base point x0and let γtbe the spatial loop at time t. Then d dtf M(γt) = M(x0, t),f M(γt), hence for any class function f:G→Cand any representation, ff M(γt)is conserved in t. Proof. Differentiate f M = t ( HolB ( Rt )) −1HolA ( γt )in t . The derivative of HolA gives the usual commutator term plus an integral of FA over the cylinder; the derivative of t ( HolB )produces t ( RBxt )with a sign so that the FA and t ( B )contributions cancel by F = 0, while H = 0 gives independence of the choice of ribbon Rt. The remaining term is the conjugation by Mat the base point. Remark .111 (Gauge invariance).Because of (67) , the CLS property and Theorem .110 are preserved under 1–/2–gauge transformations. In particular, the conjugacy class of f M and its class functions are gauge–invariant. Local trivializations and overlap data Let U = {Ui} be a good cover of X . On each Ui choose local data ( Ai, Bi ). On overlaps Uij we have transition 1–gauge gij :Uij →Gand 2–gauge 1–forms Λij ∈Ω1(Uij,h)such that Aj= Adgij (Ai) + gij dg−1 ij −t(Λij),(73) Bj=gij ·Bi−dΛij +Aj·Λij.(74) On triple overlaps Uijk,2–morphisms hijk :Uijk →Hprovide coherence: Λik = Λij +gij ·Λjk −h−1 ijkdhijk −(Ai·hijk),(75) and on quadruple overlaps the Peiffer 3–cocycle ωijk` in H must vanish for a global object (cf. the descent/obstruction section). Equations (73)–(75) are the local gluing laws for 2–connections. Lemma .112 (Curvature compatibilities on overlaps).If (73)–(75) hold, then Fj= Adgij (Fi),Hj=gij ·Hi. Proof. Insert (73)–(74) and use (63). Basic examples AKNS (ultralocal). Let G=SL2(C),H={e}so B≡0. For KdV/NLS, set L(λ) = 0 1 u−λ0or L(λ) = −iλ q ∓¯q iλ, with M ( λ )the standard companion matrix. Then ( A = L d x + M d t, B = 0) is a CLS: F = 0, H = 0. The corrected monodromy equals the usual one, and Theorem .110 yields conservation of spectral invariants. 37 Principal chiral model (non–ultralocal). Let G be compact simple, H = G , t = IdG and G act on itself by conjugation. With currents Jµ=g−1∂µg, the Lax pair L(λ) = 1 1−λ2J1−λJ0, M(λ) = 1 1−λ2J0−λJ1, satisfies (72) with a nontrivial t ( Bxt ) = Bxt chosen to reproduce the Maillet s –kernel at the level of Poisson brackets (cf. the non–ultralocal section). Then ( A, B )is a CLS and the corrected monodromy f M has conserved class functions and ultralocal Sklyanin brackets. Defects and boundaries. If X has a defect line D , one imposes (71) on X\D and allows nontrivial hijk near D ; the obstruction descends to a relative class in H3 ( X, D ; Z ( H )) whose vanishing characterizes integrable defects. The corrected monodromy picks a defect phase determined by the relative class. Physical interpretation The 1–form A is the usual Lax connection; the 2–form B registers the contact terms that appear in non–ultralocal Poisson brackets and in varying spectral/space parameters. The fake curvature F expresses that the anomaly in FA is exact (in the image of t ), while H = 0 imposes coherence of the anomaly cancellation across surfaces. The corrected monodromy removes endpoint/ribbon ambiguities and yields the right commuting integrals. Summary of the section We formalized CLS as a flat 2–connection ( A, B )modulo 1–/2–gauge and explained its holonomy 2–functor. The categorified nonabelian Stokes theorem produces the B –corrected parallel transport, whose monodromy is conserved along the flows. Local gluing data ( gij, Λ ij, hijk )assemble global structures exactly when the Peiffer 3–cocycle vanishes, as treated earlier. Ultralocal models are recovered for B = 0, while non–ultralocal models are incorporated by a nontrivial Bthat cancels endpoint anomalies. RIGOROUS PROOFS OF THE MAIN RESULTS We provide complete proofs of the principal results used throughout: (i) nonabelian Stokes for 2– connections and the holonomy 2–functor; (ii) spectral parameter internalization on X× Σ; (iii) the Sklyanin bracket for the B –corrected transporter; (iv) the descent–obstruction criterion; (v) reconstruction from twisted Riemann–Hilbert data. Throughout, ( Ht −→ G, α )is a strict crossed module of Lie groups with Lie algebras ( ht −→ g, α ), and X is a smooth manifold (typically S1×I). We use the Foundations in Sec. as standing notation. Nonabelian Stokes for 2–connections and holonomy 2–functor Let ( A, B )be a 2–connection on X with fake curvature F = FA−t ( B )and 2–curvature H = d B + A·B . Definition .113 (Paths and bigons) . Apath is a piecewise smooth map γ : [0 , 1] →X . A bigon is a smooth map Σ : [0 , 1] 2→X whose restrictions γs ( t ) = Σ( s, t )have fixed endpoints for all s . Two paths (resp. bigons) are thinly homotopic if there exists a homotopy whose differential has rank ≤ 1(resp. ≤ 2) everywhere. Lemma .114 (Variation of 1–holonomy) . Let Uγ be the G –valued parallel transport along γ solving ∂tUγ(t) = −A(˙γ(t)) Uγ(t)with Uγ(0) = Id. For a smooth family γswith γs(0), γs(1) fixed, ∂sUγs(1) Uγs(1)−1=−Z1 0 Uγs(t)FA(∂sΣ, ∂tΣ) Uγs(t)−1dt+A(∂sΣ)t=1 −A(∂sΣ)t=0. Proof. Differentiate the ODE in s , use Duhamel’s formula and the identity ∂sA ( ˙γ ) = FA ( ∂s Σ , ∂t Σ) + ∂tA(∂sΣ) −[A(∂sΣ), A(˙γ)]. 38 Definition .115 (Surface transport for B ) . Given a bigon Σ, define W ( s, t ) ∈H by the initial value problem ∂tW(s, t) = Uγs(t)·B(∂sΣ, ∂tΣ)W(s, t), W(s, 0) = IdH, and set HolB(Σ) := W(1,1) W(0,1)−1∈H. Remark .116.The action U·B is the G –action on h exponentiated to H ; existence/uniqueness follows from smoothness and compactness of [0,1]2. Theorem .117 (Nonabelian Stokes for 2–connections).For any bigon Σ, one has in G: Uγ1(1) Uγ0(1)−1=tHolB(Σ)·expZ1 0Z1 0 Uγs(t)F(∂sΣ, ∂tΣ) Uγs(t)−1dtds,(76) where γ0, γ1are the bottom and top edges of Σ. If F= 0, this reduces to Uγ1(1)Uγ0(1)−1=t(HolB(Σ)). Proof. Differentiate Φ( s ) := Uγs (1) Uγ0 (1) −1 and combine Lemma .114 with the definition of W . The term involving B integrates to t ( W (1 , 1) W (0 , 1) −1 )using that t intertwines the actions (Peiffer identity). The remaining surface integral is the ordered exponential of the conjugated Fterm. Corollary .118 (Holonomy 2–functor) . If F = 0 on X , then the assignments γ7→ Uγ (1) and Σ 7→ HolB (Σ) define a holonomy 2–functor Π ≤2 ( X ) →BG . If also H = 0, the functor is thin–homotopy invariant. Proof. Functoriality follows from multiplicativity of U and W under concatenation; thin invariance uses H= 0 (standard argument, cf. [138, 139]). Proposition .119 (Ribbon independence for corrected transport) . Let e U ( γ ) = t ( HolB ( R )) −1Uγ (1). If F = H = 0 in a neighborhood of the ribbon R , then e U ( γ )is independent of R and of thin homotopies of γ. Proof. Apply Theorem .117 to the bigon that sweeps the difference between two ribbons; the F term vanishes and the t(HolB)factors cancel. Spectral parameter internalization on X×Σ Let X be as above and Σa Riemann surface with local coordinate λ . Consider A∈ Ω 1 ( X× Σ ,g )and B∈Ω2(X×Σ,h)of the form A=Axdx+X n Atndtn+Aλdλ, B=X i<j Bij dxi∧dxj, spectrally regular in λ(finite–order poles along prescribed punctures), and suppose F(A,B)=0,H(A,B) = 0 on X×Σ.(77) Theorem .120 (Spectral Parameter Internalization) . For each fixed λ∈ Σ, the slice ( A ( λ ) , B ( λ )) is a CLS on X . Moreover, for any base point x0 the B –corrected monodromy f M ( λ )along the spatial loop at x0satisfies ∂tnf M(λ)=[Atn(x0, λ),f M(λ)], ∂λf M(λ) = [Aλ(x0, λ),f M(λ)]. In particular, all class functions of f M(λ)are independent of ({tn}, λ)(away from punctures). Proof. Fix λ and restrict (77) to X× {λ} ; this gives F ( A ( λ ) , B ( λ ))=0= H ( A ( λ ) , B ( λ )), hence a CLS slice. For the λ –evolution, consider the loop γ at time tn and vary λ ; the family of loops in X× Σsweeps a cylinder. Apply Theorem .117 with ( x, λ )as coordinates: the derivative of the uncorrected monodromy is a commutator with Aλ plus a surface term containing t ( Bxλ ), which cancels against the derivative of the B –holonomy along the ribbon by F = 0. Vanishing H ensures independence of the ribbon choice (Prop. .119). Remark .121.Poles of ( A,B )at punctures may produce jumps; the statement holds in each domain of holomorphy and extends across punctures by meromorphic continuation when appropriate. 39 Sklyanin bracket for the corrected transporter We work at a fixed time and suppress it in notation. Let L ( x ; λ ) = Ax ( x ; λ ) ∈g be the spatial Lax matrix, and assume the Poisson bracket of Lis of Maillet form on distributions in x: {L1(x, λ), L2(y, µ)}= [r12(λ, µ), L1(x, λ) + L2(y, µ)] δ(x−y)(78) + [s12(λ, µ), L1(x, λ)−L2(y, µ)] δ(x−y)+2s12(λ, µ)∂xδ(x−y), with r12 = −r21 and s12 = s21 smooth kernels on Σ × Σ, satisfying Maillet’s consistency relations (Jacobi), cf. Sec. . Let U(x, y;λ)be the path–ordered exponential for L, and define the B–corrected transporter e U(x, y;λ) := t Hxy(λ)−1U(x, y;λ), where Hxy(λ)∈His the 2–holonomy of Bxλ across a thin ribbon spanning the interval. Lemma .122 (Endpoint evolution of brackets) . Let V ( x, y )be any G –valued functional of L satisfying ∂xV=−L(x)Vand V(y, y) = Id. Then, using (78) in the sense of distributions, ∂x{V1(x, y), V2(x, y)}=−L1{V1, V2}−{V1, V2}L2+r+ 12 V1V2−V1V2r− 12 + bdyy, where r± 12 =r12 ±s12 and bdyyis a distribution supported at x=ycoming from the ∂xδterm. Proof. Differentiate the bracket under the path ordering and use that ∂xδ can be integrated by parts against the Heaviside kernels that define V . This produces the usual bulk terms plus boundary contributions at x=y(cf. [131], or [113] Sec. 2). Proposition .123 (Cancellation of boundary anomalies by B ) . Let Kxy := t ( Hxy ). Then Kxy satisfies ∂xKxy = Kxy s12 ( ·,· ) δ (0) at the level of quadratic brackets (distributionally), and for the corrected transporter e Uthe endpoint term bdyyin Lemma .122 cancels. Proof. The H –holonomy Hxy solves ∂xHxy = Hxy RBxλ with integrand supported on the ribbon; under the Poisson structure induced by the pre–symplectic form Ω(Sec. ) one computes that the equal– point bracket of the integrated Bxλ with L reproduces s12 (this is precisely how s arises in the first place). Pushing forward by t gives the stated differential for Kxy at quadratic order, which cancels the distributional endpoint term. A standard regularization (point splitting) makes this precise. Theorem .124 (Sklyanin bracket for e U and f M ) . The B –corrected transporter satisfies the ultralocal Sklyanin bracket {e U1(x, y;λ),e U2(x, y;µ)}=r+ 12(λ, µ)e U1e U2−e U1e U2r− 12(λ, µ),(79) and the corrected monodromy on the circle obeys {f M1(λ),f M2(µ)}= [ r12(λ, µ),f M1(λ)f M2(µ) ].(80) In particular, class functions of f MPoisson–commute. Proof. Apply Lemma .122 to V = U , and subtract the bracket contributions from K−1 xy using Proposition .123. The resulting differential equation for the bracket of e U has no boundary term and integrates to (79) . For the monodromy, set ( x, y ) = (2 π, 0) on the circle; the endpoint contributions coincide and (80) follows. Commutation of class functions is a standard consequence of (80). Remark .125 (Assumptions).The argument uses the Maillet consistency relations to guarantee Jacobi for (79) and (80). The pre–symplectic form Ωof Sec. provides a Hamiltonian realization of (78). Descent and the obstruction 3–cocycle Let U = {Ui} be a good open cover of X . Recall the overlap data ( gij, Λ ij )and triple–overlap hijk satisfy (73)–(75). Define on quadruple overlaps Uijk` the Peiffer defect ωijk` := (gij ·hjk`)hij` (hijkhik`)−1∈H. (81) 40 Lemma .126 (Cocycle and gauge behavior).(a) On Uijk`m one has the twisted 3–cocycle identity (gij ·ωjk`m)ωij`m =ωijkm ωik`m. (b) Under 1– and 2–gauge refinements ( ki, λij ), ω transforms by pointwise conjugation in H : ω0 ijk` = (ki ·)ωijk`. Proof. (a) is the simplicial identity corresponding to the 3–simplex of the Čech nerve together with the Peiffer law; see [139]. (b) follows by inserting the transformation laws into (81). Theorem .127 (Descent–obstruction criterion) . Let local CLS data  ( Ai, Bi ); ( gij, Λ ij ); hijk be given on U satisfying (73) – (75) and Fi = Hi = 0 on Ui . Then there exists a global 2–connection ( A, B )on X with these local representatives if and only if the class [ ω ]of (81) vanishes. If [ ω ]=0, the set of isomorphism classes of such globalizations is a torsor under the (twisted) nonabelian cohomology ˇ C2(U, H). Proof. The necessity is clear: a global object yields hijk satisfying the Peiffer condition, hence ω = eH . For sufficiency, vanishing [ ω ]implies there exist ηij : Uij →H such that modifying hijk 7→ ( gij·ηjk ) ηij hijk η−1 ik makes ω trivial (by Lemma .126). Adjust Λ ij 7→ Λ ij −η−1 ij d ηij − ( Ai·ηij )so that (75) still holds. Then the standard descent (a 2–stack argument) glues ( Ai, Bi )to a global ( A, B ); cf. [ 115 , 136 , 139 ]. The torsor statement follows from the action of ˇ C2(U, H)by these ηij. Remark .128 (Abelian comparison).When G = {e} and H = U (1), ω is the usual ech 3–cocycle representing the Dixmier–Douady class; Theorem .127 reduces to the classical statement for bundle gerbes. Reconstruction from twisted Riemann–Hilbert data Let C⊂C be a smooth oriented contour splitting CP1 into Ω ± . A twisted jump datum ( J, Θ) consists of J:C→Gand Θ : C→Hsatisfying Definition .89 (Sec. ). The corrected jump is b J:= t(Θ) J. Theorem .129 (Small–norm solvability (Beals–Coifman with twist)) . If ρ ( b J−Id ) ∈L∞ ( C ) ∩H1 ( C ) is sufficiently small for a faithful matrix representation ρ , then the TRHP has a unique solution Γ : CP1\C→G with Γ( ∞ ) = Id and Γ + = b J Γ − on C . Moreover, Γdepends smoothly on ( J, Θ) in these norms. Proof. Write b J = Id −w−−1 ( Id + w+ )with w±∈L2 ( C )and solve the singular integral equation µ = Id + Cwµ on L2 ( C )with Cwf = C+ ( fw− ) + C− ( fw+ )(Cauchy projectors). For small norm, Id − Cw is invertible by Neumann series, and Γ = Id + C ( µ ( w+ + w− )) solves the RHP. The twist only enters through b J; see [111, 126]. Let Φ( λ ; t )be a bare phase with values in g and set the bare connection A0 = Pi ( ∂ti Φ) d ti on X = S1×I . Suppose the data depend on ( x, t )by conjugation with the bare phase, as in (59) of Sec. . Define A:= ΓA0Γ−1−(dΓ)Γ−1, B determined distributionally along Cin (x, λ). Theorem .130 (Reconstruction of CLS) . Under the hypotheses above, ( A, B )is a 2–connection on X× ( CP1\C ), flat there, and extends across C so that ( A, B )is categorically flat on X×CP1 with a distributional t(Bxλ)supported on Cgiven by t(Bxλ) = ∂λΓ∂xΦ Γ−1−∂xΓ∂λΦ Γ−1+ [ Γ∂xΦΓ−1,Γ∂λΦΓ−1]δC.(82) For each fixed λ /∈C , the slice ( A ( λ ) , B ( λ )) is a CLS on X , and the B –corrected monodromy is independent of (x, t). Proof. Away from C , A is a pure gauge of A0 ; hence FA = 0 and with B = 0 one has F = 0 = H . Across C , the jump Γ + = t (Θ) J Γ − implies that the failure of A to be single–valued is Adt(Θ) of a conjugation, which is absorbed by t ( B ); the formula (82) is the distributional form of this compensation (obtained by testing against smooth 2–forms supported in a tubular neighborhood of C ). The nonabelian Stokes theorem then shows that the B –corrected monodromy is invariant under ( x, t )by the same cancellation as in Theorem .120. Remark .131.If the jump has isolated poles (soliton data), one adds residue conditions for Γ; the reconstruction proof is unchanged because t(Θ) is analytic at the poles. 41 Concluding remarks We have: (i) a precise nonabelian Stokes theorem yielding a holonomy 2–functor and the ribbon– independent corrected transport; (ii) internalization of the spectral parameter as flatness on X× Σ; (iii) an ultralocal Sklyanin bracket for the B –corrected transporter/monodromy from a Maillet bracket of L ; (iv) a cohomological descent criterion for globalizing local CLS data; (v) a robust reconstruction scheme via twisted RHPs. QUANTIZATION OF THE CORRECTED MONODROMY We quantize the Sklyanin Poisson algebra of the B –corrected monodromy (Sec. and Thm. .124) to a quantum exchange algebra. The key input is that the B –correction removes the Maillet s –anomaly at the classical level, so the quantum object should satisfy a standard RTT relation with a quantum R –matrix. We also explain how the B –sector enters as a Drinfel’d twist of the coproduct (or, equivalently, a Reshetikhin twist of the R –matrix), and why transfer matrices built from the corrected monodromy commute. Throughout, we fix an auxiliary “space of states” H (e.g. a spin chain or a field–theoretic Fock space) and an auxiliary finite–dimensional vector space V (the auxiliary space). Tensor leg notation is used: X1=X⊗Id,X2= Id ⊗X. The spectral parameter is λ∈Σ. Semiclassical input Let f M(λ)denote the classical B–corrected monodromy. Its Poisson bracket is (Thm. .124) {f M1(λ),f M2(µ)}=r12(λ, µ),f M1(λ)f M2(µ),(83) where r12 ( λ, µ ) ∈End ( V⊗V )is a classical r –matrix (rational/trigonometric/elliptic as appropriate) solving the (non–dynamical) classical Yang–Baxter equation (CYBE) [r12, r13]+[r12, r23]+[r13, r23] = 0. The Jacobi identity for (83) is equivalent to CYBE. Quantum RTT algebra and transfer matrices We now deform (83) to a quantum exchange algebra. Definition .132 (Quantum R –matrix) . A quantum R –matrix is R ( λ, µ ) ∈End ( V⊗V )meromorphic in (λ, µ)such that R(λ, µ) = Id ⊗Id + ~r(λ, µ) + O(~2), and it satisfies the quantum Yang–Baxter equation (QYBE): R12(λ, µ)R13(λ, ν)R23(µ, ν) = R23(µ, ν)R13(λ, ν)R12(λ, µ).(84) Definition .133 (RTT algebra of the corrected monodromy) . Let c Ma ( λ ) ∈End ( Va ) ⊗End ( H )be an operator–valued matrix (entries in End(H)). The RTT algebra ARis generated by the matrix elements of c M(λ)subject to the exchange relation Rab(λ, µ)c Ma(λ)c Mb(µ) = c Mb(µ)c Ma(λ)Rab(λ, µ).(85) Remark .134 (Semiclassical limit).Setting ~→ 0and using [ A, B ] = ~{A, B} + O ( ~2 ), (85) linearizes to (83). Thus ARis a quantization of the Poisson algebra of f M. 48 Discrete fake curvature (lattice). On a face fwith oriented boundary ∂f =Pefe e, define Flat f:=  Y e⊂∂f Ufe e t(Vf)−1∈G. A small–residual regime is handled via log Flat f∈g, accumulated with M2. B –corrected transporter. For a path γ on the mesh, let U ( γ )be the path product of Ue . Let R be a thin ribbon spanning γand a reference path, approximated by a strip of faces. Define e U(γ) := t Y f⊂R Vf f−1U(γ). By construction, e U is invariant under local deformations of R when the discrete curvatures are small, echoing Prop. .107. Linearization and Gauss–Newton/KKT system Let a,b collect the coefficients of ( Ah, Bh )in a fixed basis. The residual F ( a,b )that stacks Fh,Hh, Ri,h and overlap terms admits a Fréchet derivative (DF)[δa, δb]computed from (96)–(97): δFh= dhδAh+ [Ah∧hδAh]−t(δBh), δHh= dhδBh+Ah·hδBh+δAh·hBh, δRi,h = ΠW2 h∂tδLh−∂xδMh+ [δLh, Mh]+[Lh, δMh]−t(δBh,xt). The Gauss–Newton step solves the normal equations (DF)>W(DF)δz =−(DF)>W F, z := (a,b), with W the block mass operator (Whitney L2 ). Overlap constraints are enforced by (i) Lagrange multipliers (KKT system), or (ii) an augmented Lagrangian with penalty ρ ; both lead to sparse, symmetric systems. Block structure. The Hessian surrogate (DF)>W(DF)has the schematic form α D∗ ADA+QAA −α t∗DA+QAB −α D> At+QBA α t∗t+β D∗ ADA+QBB, where DA is a discrete covariant derivative assembled from d h and Ah , and Q•• are lower–order terms (including contributions from Ri,h and overlaps). This structure admits Hodge–Laplace preconditioners. Gauge fixing and projections To remove null directions and improve conditioning, impose discrete Coulomb–type gauges: divhAh:= d> h−1M1a= 0,divhBh:= d> h−2M2b= 0, and project ( δa, δb )at each iteration. This is implemented by solving mass–weighted Poisson equations for the gauge potentials and subtracting exact parts: δAh←δAh−dhξh, δBh←δBh−dhΞh+Ah·hΞh, with ξh∈W0 h⊗g,Ξh∈W1 h⊗hobtained from linear systems with operators L0and L1. Spectral parameter as a numerical coordinate When internalizing λ (Sec. ), set b X = X× Σwith a product mesh in ( x, t, λ )near punctures. Then B acquires a component Bxλ and Hh6 = 0 algebraically. Discretize Aλ in W1 h (along λ ) and Bxλ in the mixed Whitney space W2 h (the 2–forms with one leg in x and one in λ ). The discrete version of (61) is realized by concentrating t ( Bxλ )on a one–cell thick layer approximating the contour C and matching the jump b Jin the TRHP (Sec. ). 49 Riemannian updates on G, H In the holonomy formulation, variables live on manifolds G|C1| and H|C2| . We use the canonical bi–invariant metrics and update via exponentials: Unew e= exp(∆e)Ue,∆e∈TUeG≃g, Vnew f= exp(∆f)Vf,∆f∈TVfH≃h, with ∆obtained from a Riemannian Gauss–Newton step (solve in the Lie algebra; retract with exp ). Parallel transport between tangent spaces is trivial under left translations. Augmented Lagrangian / ADMM scheme Split variables by patches and overlaps. With multipliers Λ (1) ij for the A –gluing and Λ (2) ij for the B–gluing, iterate: (S1) Local update (all patches in parallel): minimize the local augmented objective (Gauss–Newton or Riemannian GN) for (Ai, Bi)with fixed overlap data. (S2) Overlap update : minimize the sum of overlap penalties to update ( gij, Λ ij )(in holonomy, project to G, H via exp). (S3) Triple–overlap coherence : update hijk by a Newton step on the Peiffer defect; e.g. hijk ← hijk exp(−Πhlog ωijk`). (S4) Multiplier/penalty update:Λ(m)←Λ(m)+ρ C(m), increase ρif needed. (S5) Gauge projection: enforce discrete Coulomb gauges (Sec. ). Stop when kFhk,kHhk, and overlap residuals fall below tolerances. A–posteriori certification and error bars Define the total residual E2 tot := αkFhk2 M2+βkHhk2 M3+ηX i kRi,hk2 M2(Ui)+γX i<j kC(A) ij k2 h+kC(B) ij k2 h+klog bωk2 ∞. Under small–residual conditions and mesh quasi–uniformity, a discrete analogue of Theorem .152 yields kAh−A†kH1+kBh−B†kH1≤CEtot, and for any representation ρand class function f, f(f Mh)−f(f M†)≤Cf,ρ Etot. Empirically, C, Cf,ρ are modest for well–conditioned meshes; monitor effectivity by mesh refinement (p– or h–refinement). Complexity, preconditioning, and stability Let N1 = |C1| , N2 = |C2| . One local Gauss–Newton step costs ˜ O ( N1 + N2 )with sparse matrix–vector products and multigrid/AMG preconditioning for Lk . Overlap updates scale with the interface size. Holonomy updates have cost linear in the number of edges/faces. The overall wall–clock time scales nearly linearly with the number of patches thanks to parallelism in (S1). 50 Preconditioners. Use block–diagonal Hodge preconditioners: P ≈ α L1+η ∂∗ x,t∂x,t 0 0α t∗t+β L2+η ∂∗ x,t∂x,t. Here ∂x,t collects discrete derivatives along x, t (or x, λ). Stability tips. (i) Scale α, β, η, γ so that all residual blocks have comparable magnitudes; (ii) increase ρ geometrically; (iii) trust–region radii tied to kFhk ; (iv) in holonomy variables, clip klog ( · ) k to stay on the principal branch. Validation protocol We recommend the following standardized checks: 1. Ultralocal baseline (KdV, NLS): verify β = 0, B≡ 0, and kFhk → 0down to discretization error; spectral invariants drift ↓O(hp)with spectral/p–FEM rates. 2. Non–ultralocal correction (PCM, symmetric σ –models): compare raw monodromy vs. corrected f M—the latter’s invariant drift must be the former and match kFhk+kHhk. 3. Near–integrable drift (KdV–Burgers): check linear scaling of invariant drift with viscosity ν ; slope ∝ k log bωk∞. 4. Cover sensitivity : refine U and confirm inf Φ h is nonincreasing and residuals decrease (Prop. .154). 5. Gauge and ribbon independence : perturb gauges and ribbons; f M class functions should be unchanged within solver tolerance. Minimal working example (MWE) Setup. X = S1× [0 , T ], Nx = 2048, Nt = 256, four spatial patches with π/ 4overlap, two time patches with 10% overlap; (α, β, η, γ) = (1,1,10−3,10). Steps. (1) Solve the PDE (e.g. KdV/NLS/PCM) with your preferred integrator. (2) Assemble FEEC structures ( Mk, d h,∧h,·h ). (3) Initialize ( Ah, Bh )from linearized classic Lax data; set overlaps to identity. (4) Run the ADMM loop (S1) – (S5) until residuals drop below tolerance. (5) Compute corrected monodromy f Mand invariant drifts; report Etot with error bars. Expected outputs. For ultralocal cases: kFhk ∼ 10 −10 –10 −12 (double precision); invariant drifts at the discretization floor. For non–ultralocal: raw monodromy drifts visible (10 −3 –10 −5 ); corrected monodromy drifts near the FEEC floor; klog bωk∞ near machine zero when integrable, linear in perturbation otherwise. Remarks on implementation • Quadrature. For Whitney products, use elementwise polynomials and exact quadrature (tensor Gauss–Legendre suffices on rectangles). • Spectral–FEEC hybrid. If PDE evolution in x is spectral, keep Ah, Bh on FEEC to preserve gauge structure; map spectral derivatives to FEEC via L2–projection. • Group numerics. For compact G, H , the matrix exponential/logarithm are well behaved; for noncompact groups, prefer scaling–and–squaring with Padé and monitor conditioning. •Parallelism. (S1) and (S2) are patch/overlap separable; (S3) is local to triple overlaps. 51 RELATED WORK AND POSITIONING From Lax pairs to inverse scattering. The modern theory of integrable PDEs began with Lax’s observation that many equations are equivalent to an isospectral flow for a linear operator with parameter (Lax pair) [ 122 ]. The inverse scattering transform (IST) [ 123 – 125 ] and Riemann–Hilbert (RHP) formulations [ 126 – 128 ] give reconstruction procedures and asymptotics. Our twisted RHP (TRHP) adds an H –valued twist to the jump data; this is precisely what is needed to encode the B –sector that corrects non–ultralocal anomalies while preserving the analytic structure of IST. Classical r –matrices, non–ultralocality, and Sklyanin brackets. The Poisson geometry of monodromies and transfer matrices is governed by classical r –matrices [ 129 , 130 ]. In many field theories (principal chiral model, symmetric space σ –models), equal–time brackets are non–ultralocal and contain derivatives of δ (the Maillet s –term) [ 131 , 132 ]. Sklyanin showed that ultralocal monodromies satisfy quadratic exchange relations [ 133 ]. Our contribution is to construct, at the level of a 2–connection ( A, B ), a B –corrected transporter whose Poisson algebra is of Sklyanin type even when the raw theory is non–ultralocal. This promotes the “algebraic trick” of adding s –counterterms to a geometric principle: the B –sector is a genuine part of the parallel transport in a higher gauge theory. Quasi–Hamiltonian geometry and group–valued moment maps. The moduli of flat G –connections on surfaces is naturally (shifted) symplectic [ 140 ]. On surfaces with boundary, Alekseev–Malkin–Meinrenken introduced quasi–Hamiltonian spaces with group–valued moment maps [ 134 ], recovering the Goldman bracket in a multiplicative form. Our pre–symplectic current for ( A, B )and the corrected holonomy fits this picture: the B –field induces a background 3–form twist [ 135 ] that is absorbed by the corrected transporter, so class functions Poisson–commute again. Higher gauge theory, gerbes, and 2–connections. Nonabelian gerbes and 2–connections have been developed in several equivalent frameworks [ 136 – 139 ]. Our CLS is a strict 2–connection ( A, B )with fake flatness and 2–flatness; the categorified nonabelian Stokes theorem gives a holonomy 2–functor, and the B –corrected parallel transport emerges as the canonical 1–holonomy associated to a ribbon surface. The obstruction to globalizing local data is a nonabelian Čech 3–class (Peiffer cocycle), generalizing the Dixmier–Douady class. Quantization: RTT algebras, twists, and boundary algebras. The FRT/RTT formalism [ 147 , 148 ] quantizes Sklyanin’s brackets via a quantum R –matrix. Drinfel’d and Reshetikhin twists [ 149 , 150 ] modify coproducts and exchange relations. We identify the quantum imprint of the B –sector with such a twist: either conjugate the quantum monodromy by a universal F or twist the R –matrix to RF —the commuting transfer matrices are preserved. At boundaries/defects this viewpoint dovetails with reflection algebras and twisted K–matrices (to be developed in Sec. ). Positioning of the present work. Conceptually, we: (i) elevate the cancellation of non–ultralocal endpoint anomalies to a geometric statement about 2–holonomy and ribbon independence; (ii) give a unified reconstruction scheme (TRHP) that includes the B –sector; (iii) prove that the B –corrected monodromy obeys a Sklyanin Poisson algebra, and quantizes to a standard RTT algebra up to a Drinfel’d twist; (iv) provide a practical solver (FEEC/holonomy hybrid) with a posteriori certification linking residuals to invariant drift. CONCLUSION AND OUTLOOK We developed a coherent geometric framework for integrability built on strict 2–connections ( A, B ): (i) the B –corrected transporter makes endpoint anomalies geometric and yields an ultralocal Sklyanin algebra for monodromies even in non–ultralocal models; (ii) a twisted Riemann–Hilbert problem reconstructs ( A, B )from spectral data; (iii) the B –sector quantizes as a Drinfel’d/Reshetikhin twist, preserving commuting transfer matrices for closed and open chains; (iv) numerical FEEC/holonomy schemes with certification bounds make the construction practical. Extensions. Natural directions include: (1) semi–strict 2–groups with nontrivial `3 and their impact on Maillet consistency; (2) defects/boundaries with dynamical K –matrices and coideal symmetries beyond diagonal cases; (3) spectral–curve geometry for corrected monodromies and categorified τ –functions; (4) stochastic CLS and rough–path versions of 2–holonomy; (5) higher–dimensional analogues (3D integrability) via 3–connections and categorified monodromies. 52 Appendices A–Posteriori Certification for the Corrected Monodromy We quantify how small violations of categorified flatness and overlap coherence control the drift of spectral invariants of the B–corrected monodromy. Setting Let ( A, B )be a smooth CLS on X = S1×I (so F = H = 0), and let ( Ah, Bh )be discrete fields (FEEC/DEC or any convergent scheme) with residuals R1:= kFAh−t(Bh)kL2(X),R2:= kdBh+Ah·BhkL2(X),(99) and overlap/cover residual Rov measuring the failure of (73) – (74) and the Peiffer cocycle (in any consistent L2 / H1 norm on overlaps). Let f M ( λ )be the exact corrected monodromy and f Mh ( λ )the discrete one (built from ( Ah, Bh )with the same ribbon conventions). Fix a faithful finite–dimensional representation ρand a class function f:G→Cwith Lipf,ρ <∞in operator norm. Theorem .152 (Certification bound) . Suppose max{R1,R2,Rov} is sufficiently small and ( Ah, Bh ), ( A, B )share the same gauge and ribbon normalizations at a base point. Then there exist constants C, Cf,ρ > 0(depending on ( A, B )through kAkL∞ , the representation, and the length of the loop) such that for all admissible λ, ρf Mh(λ)−ρf M(λ)≤CR1+R2+Rov,(100) ff Mh(λ)−ff M(λ)≤Cf,ρR1+R2+Rov.(101) The same bounds hold uniformly on compact λ –sets away from punctures when the spectral parameter is internalized (i.e. on X×Σ), with Cindependent of the mesh size. Proof. We sketch the key steps; details follow standard holonomy stability arguments and FEEC estimates. (i) Duhamel expansion for transporter differences. Let U and Uh be the (uncorrected) path–ordered exponentials for Aand Ahalong the spatial loop γt. Then Uh−U=Z2π 0 Uh(2π, x)Ah−A(x)U(x, 0) dx, plus higher–order commutator terms bounded by kAh−AkL1exp(ckAkL1). (ii) Ribbon compensation and fake curvature. Let K := t ( HolB ( R )) and Kh := t ( HolBh ( Rh )) be boundary G–factors for the ribbons. By the nonabelian Stokes theorem for 2–connections, variations in K and U combine to first order into an integral of the fake curvature over the cylinder spanned by the loop and its reference path. Thus K−1 hUh−K−1U.kFAh−t(Bh)kL1+kAh−AkL1, with constants depending on kAkL∞and the loop length. (iii) Control of kAh−Ak and kBh−Bk by residuals. On a good cover, FEEC stability and commuting diagram properties yield kAh−AkH1.R1+Rov,kBh−BkH1.R2+Rov, up to gauge choices fixed at a base point. (Here we use discrete Poincaré inequalities and the fact FA=t(B),dB+A·B= 0 for the target.) (iv) Composition and passage to corrected monodromy. Putting (i)–(iii) together for f Mh = K−1 hUh and f M = K−1U gives (100) . The class function bound (101) follows from the uniform Lipschitz constant of f◦ρ on a neighborhood of the range (compactness is ensured by small residuals and bounded loop length). Remark .153 (Dependence of constants). C depends monotonically on the loop length, kρ ( A ) kL∞ and a tubular neighborhood width for the ribbon. For families parametrized by λ away from punctures the constants can be chosen uniformly on compact subsets of Σ. 53 Refinement monotonicity We formalize the monotonic decrease in the objective under refinement mentioned in the validation checklist. Proposition .154 (Refinement decreases the minimal objective) . Let Φ h be the discrete coherence functional from (98) computed on a mesh Th and an open cover Uh . If Th0 refines Th and Uh0 refines Uh (restrictions and prolongations are compatible with Whitney injections), then inf (Ah0,Bh0,overlaps)Φh0≤inf (Ah,Bh,overlaps)Φh. Proof. Extend any admissible ( Ah, Bh )to Th0 by canonical Whitney injection; similarly prolong overlap data to the refined cover. Exactness of injections with respect to d h and the consistency of mass matrices imply that the extended configuration attains the same residual values (up to quadrature errors of higher order), hence no larger objective. Taking infima yields the claim. Crossed Modules, 2-Groups, and Non-Abelian Čech Cohomology Crossed modules and strict Lie 2-groups Acrossed module of Lie groups ( Ht −→ G, α )consists of Lie groups H, G , a smooth homomorphism t:H→G, and a smooth action α:G×H→H, written g·h, such that the Peiffer identities hold: t(g·h) = g t(h)g−1, t(h)·h0=h h0h−1(∀g∈G, h, h0∈H).(102) Its differential is a crossed module of Lie algebras (ht −→ g)with an action of gon h. The data present a (strict) Lie 2-group G : a one-object 2-groupoid whose 1-morphisms are G and whose 2-morphisms are HoGwith vertical/horizontal compositions induced by tand α. Remark .155 (Morphisms and equivalences).A morphism of crossed modules ( H→G ) → ( H0→G0 ) is a pair of homomorphisms intertwining t and the actions. Weak/stacky equivalences (“butterflies”) implement Morita equivalence between presentations; holonomy and CLS observables are invariant under such equivalence (cf. Theorem .71 in the main text). 2-connections, 1-/2-gauge, and curvature On a smooth manifold X, a (strict) 2-connection is a pair (A, B)with A∈Ω1(X, g), B ∈Ω2(X, h), and curvatures F:= FA−t(B)∈Ω2(X, g),H:= dB+A·B∈Ω3(X, h),(103) satisfying Bianchi identities DAF+t(H) = 0 and DAH= 0. A1-gauge transformation g:X→Gand a 2-gauge transformation Λ∈Ω1(X, h)act by A0= Adg(A) + gdg−1−t(Λ),(104) B0=g·B−dΛ + A0·Λ,(105) and F0,H0transform covariantly: F0= Adg(F),H0=g·H. Non-abelian Čech descent and the obstruction class Let U = {Ui} be a good open cover of X . A local 2-connection is a family ( Ai, Bi )on Ui with overlap data Aj= Adgij (Ai) + gij dg−1 ij −t(Λij),(106) Bj=gij ·Bi−(dΛij +Aj·Λij),(107) 54 for gij : Uij →G ,Λ ij ∈ Ω 1 ( Uij,h ). On triple overlaps, coherence is provided by hijk : Uijk →H such that gik =t(hijk)gij gjk,Λik = Λij +gij ·Λjk −h−1 ijkdhijk −(Ai·hijk).(108) On quadruple overlaps one defines the Peiffer defect (a Čech 3-cochain with values in H) ωijk` := (gij ·hjk`)hij` (hijk hik`)−1.(109) Lemma .156 (Cocycle and gauge behavior).(a) ωobeys the twisted 3-cocycle identity on Uijk`m: (gij ·ωjk`m)ωij`m =ωijkm ωik`m. (b) Under refinement by 1-/2-gauge (ki, λij),ωtransforms by conjugation in H:ω0 ijk` = (ki ·)ωijk`. Theorem .157 (Descent–obstruction, strict case) . Suppose Fi = Hi = 0 on Ui . There exists a global ( A, B )on X gluing the local data iff the Čech class [ ω ]of (109) vanishes. If [ ω ] = 0, the set of isomorphism classes of globalizations is a torsor under the (twisted) non-abelian cohomology group ˇ C2(U, H). Remark .158 (Abelianization and Dixmier–Douady).If Z ( H )is central and t ( Z ( H )) ⊂Z ( G ), the image of [ ω ]in ˇ C3 ( U, Z ( H )) is well-defined and coincides with the Dixmier–Douady class in the purely abelian case (G={e}). Non-Abelian Stokes and 2-Holonomy: Details and Conventions Conventions for paths, bigons, and thin homotopy Apath is a piecewise smooth γ : [0 , 1] →X . A bigon is a smooth Σ : [0 , 1] 2→X such that Σ( s, 0) and Σ( s, 1) do not depend on s . Thin homotopies have rank ≤ 1(paths) or ≤ 2(bigons). We parametrize bigons by (s, t)with tvertical (tincreases upward). 1-holonomy variation and boundary terms Let Uγ(t)solve ∂tU=−A( ˙γ)U,U(0) = Id. For a family γs, the variation of Uγs(1) is ∂sUγs(1) Uγs(1)−1=−Z1 0 Uγs(t)FA(∂sΣ, ∂tΣ) Uγs(t)−1dt(110) +A(∂sΣ)t=1 −A(∂sΣ)t=0. This is obtained by differentiating the ODE in s and using the identity ∂sA ( ˙γ ) = FA ( ∂s Σ , ∂t Σ) + ∂tA(∂sΣ) −[A(∂sΣ), A(˙γ)]. Surface transport and the Stokes formula Define W(s, t)∈Hby ∂tW(s, t) = Uγs(t)·B(∂sΣ, ∂tΣ)W(s, t), W(s, 0) = Id, and set HolB(Σ) := W(1,1) W(0,1)−1. Then: Theorem .159 (Non-abelian Stokes for 2-connections).For any bigon Σ : γ0⇒γ1, Uγ1(1) Uγ0(1)−1=t HolB(Σ)· P exp Z[0,1]2 Uγs(t)F(∂sΣ, ∂tΣ) Uγs(t)−1dsdt!,(111) where Pexp denotes surface-ordered exponential. If F= 0, then Uγ1U−1 γ0=t(HolB(Σ)). Sketch. Differentiate Φ( s ) := Uγs (1) Uγ0 (1) −1 using (110) , solve for W , and integrate in s . The Peiffer identity implies tintertwines the actions, making the B-contribution multiplicative. 55 Holonomy 2-functor and ribbon independence With F = H = 0, the assignments γ7→ Uγ (1) and Σ 7→ HolB (Σ) define a holonomy 2-functor Π ≤2 ( X ) →BG compatible with vertical/horizontal composition. For a path γ and a thin ribbon R with ∂R=γ∪γref , the B-corrected transporter e U(γ) := t HolB(R)−1Uγ(1) is independent of Rand of thin homotopies of γ(by Theorem .159 with F=H= 0). Orientation and signs. We use the convention that reversing the orientation of a path inverts U , and reversing the orientation of the s-direction in a ribbon inverts the H-holonomy (hence t(·)). From Maillet r/s Algebra to Sklyanin for the Corrected Transporter Maillet bracket and regularization Let L(x;λ)∈gbe the spatial Lax matrix and suppose {L1(x, λ), L2(y, µ)}= [r12(λ, µ), L1(x, λ) + L2(y, µ)] δ(x−y)(112) + [s12(λ, µ), L1(x, λ)−L2(y, µ)] δ(x−y)+2s12(λ, µ)∂xδ(x−y), with r12 = −r21 , s12 = s21 , and standard consistency relations ensuring Jacobi. We adopt point-splitting: for x→ywe take limits from x>y. Bracket of path-ordered exponentials Let U(x, y;λ)be the solution of ∂xU=−L(x;λ)U,U(y, y) = Id. Define the bilocal bracket B(x, y) := {U1(x, y;λ), U2(x, y;µ)}. Differentiating with respect to xand using (112) yields ∂xB=−L1B − B L2+r+ 12 U1U2−U1U2r− 12 + bdyy,(113) where r± := r±s , and bdyy is a distribution supported at x = y generated by ∂xδ after integrating by parts (the Maillet endpoint anomaly). B-sector and cancellation Introduce H -holonomy along a thin ribbon spanning the interval ( y, x )and set K ( x, y ; λ ) := t ( Hxy ( λ )). Define the corrected transporter e U := K−1U . Using the pre-symplectic form for ( A, B )(main text, non-ultralocal section), one computes at quadratic order ∂x{K−1 1, K−1 2}=K−1 1K−1 2s12 δ(x−y)−s12 δ(x−y)K−1 1K−1 2+ (higher order), which exactly cancels bdyyupon forming B(e U, e U). Theorem .160 (Sklyanin bracket for the corrected transporter).The B-corrected transporter satisfies {e U1(x, y;λ),e U2(x, y;µ)}=r+ 12(λ, µ)e U1e U2−e U1e U2r− 12(λ, µ).(114) For the circle monodromy f M, {f M1(λ),f M2(µ)}= [ r12(λ, µ),f M1(λ)f M2(µ) ],(115) and class functions of f MPoisson-commute. Proof. Solve (113) for B and subtract the bracket contributions arising from K−1 ; the distributional boundary term cancels. For the circle, endpoint contributions match and (115) follows by standard arguments. Remark .161 (Jacobi and consistency).The Jacobi identity for (114) reduces to the Maillet consistency relations, which in turn are implied by a solution of the (modified) classical Yang–Baxter equation satisfied by r, s. 56 Twisted Riemann–Hilbert Problem: Operator-Theoretic Details Setting and function spaces Let C⊂C be a finite union of smooth oriented contours partitioning CP1 into two open sets Ω ± . For f∈Lp ( C )we denote by Cf the Cauchy transform and by C±f its nontangential boundary values. For 0< α < 1,Cα(C)denotes the Hölder space. Twisted jump and equivalent formulations Given J : C→G ,Θ : C→H , define the corrected jump b J := t (Θ) J . The TRHP seeks Γ : CP1\C→G with Γ + = b J Γ − and some normalization (e.g. Γ( ∞ ) = Id ). In a faithful representation ρ : G ,→GL ( N, C ), the problem is equivalent to a matrix RHP with jump ρ(b J). Small-norm theory on L2 Assume ρ ( b J−Id ) ∈L∞ ( C ) ∩H1 ( C )and small. Write b J = ( Id −w− ) −1 ( Id + w+ )with w±∈L2 ( C ). The Beals–Coifman operator Cw:L2(C)→L2(C), Cwf:= C+(fw−) + C−(fw+), has kCwk< 1for small data. Then µ = Id + ( Id − Cw ) −1CwId solves the Plemelj equation µ = Id + Cwµ , and the RHP solution is Γ(λ) = Id + Cµ(w++w−)(λ). The solution depends smoothly on the data in these norms. Piecewise smooth jumps and Cαtheory If b J∈Cα ( C )and is piecewise smooth with a finite number of corner points, one has a Fredholm theory in L2 ( C ) ∩Cβ ( C )for 0 < β < α . Uniqueness holds under a scalar normalization provided the associated homogeneous problem has only trivial solutions. Residue (soliton) conditions If Γis allowed simple poles at {λk}with residues Resλ=λkΓ(λ) = lim λ→λk Γ(λ)Rk, then the singular integral equation augments with finite-rank terms. The twist t (Θ) is holomorphic near λkand does not modify the residue algebra; the standard IST soliton construction carries over. Deformations and the JMU form with twist Let Γ( λ ; s )depend smoothly on deformation parameters s (contour shape, pole positions, jump data). Then ∂sΓ Γ−1=CΓ−∂sb Jb J−1Γ−1 −. The categorified Jimbo–Miwa–Ueno form is Θcat =1 2πi ZC Tr Γ−1 −∂sΓ−∂λb Jb J−1dλ−ZC ht−1(∂sb Jb J−1),reference ribboni, and is closed under the flatness hypotheses on ( A, B );d log τcat = Θ cat defines a τ -function. For Θ ≡eH the second term vanishes and one recovers the classical JMU form. 57 Numerical Details, FEEC/DEC Recipes, and Pseudocode Whitney forms and product rules Let Wk hbe Whitney k-forms on a simplicial or tensor-product mesh of X. Assemble: •Incidence matrices for dh:Wk h→Wk+1 h. •Mass matrices (Mk)ij =RXhω(k) i, ω(k) ji. •Product tensors T(1,1,2) e,e0,f =RXhϕe∧ϕe0, ψfi,T(1,2,2) e,f,f0=RXhϕe·ψf, ψf0i. Use these to implement discrete Fh= dhAh+1 2[Ah∧hAh]−t(Bh),Hh= dhBh+Ah·hBh. Holonomy computation On each oriented edge e = [ v0, v1 ], approximate Ue≈expReAh by evaluating Ah at the midpoint (second order) or via Gauss–Legendre quadrature along e . On each face f , set Vf≈expRfBh using a 2D quadrature. Use principal matrix logarithms to measure residuals log(QUet(Vf)−1). Gauss–Newton step and gauge projection Let F(a,b)stack all residuals and overlaps. One GN step solves (DF)>W(DF)δz =−(DF)>W F, z := (a,b). Project ( δAh, δBh )to Coulomb-like gauges by solving Poisson problems for potentials ξh and Ξ h and replacing δAh←δAh−dhξh, δBh←δBh−dhΞh+Ah·hΞh. ADMM/augmented Lagrangian pseudocode 1. Initialize (Ai, Bi)on patches, overlaps (gij,Λij), and hijk = Id. 2. Local solve: For each patch i , minimize local augmented functional by GN (or Riemannian GN in holonomy variables). 3. Overlap solve: Update (gij,Λij )by minimizing overlap residuals; retract to G, H via exp. 4. Triple coherence: Update hijk ←hijk exp−Πhlog ωijk`on triple overlaps. 5. Multipliers: Λ(m)←Λ(m)+ρ C(m), increase ρ. 6. Gauge projection on each patch. 7. Check stopping: norms of Fh,Hh, overlaps, and change in invariants. Spectral parameter discretization On b X = X× Σintroduce a thin layer of cells approximating the contour C . Place t ( Bxλ )on this layer to encode the jump b J . Discretize Aλ on λ -edges and Bxλ on mixed faces. The corrected monodromy along x is computed by multiplying edge holonomies and inserting the t ( Vf )factors from the ribbon strip crossing C. 64 Leningrad Math. J. 1(1990), 193–225. [117] V. Chari and A. Pressley, A Guide to Quantum Groups, Cambridge Univ. Press, 1994. [118] V. G. Drinfel’d, Quasi-Hopf Algebras, Leningrad Math. J. 1(1990), 1419–1457. [119] N. Reshetikhin, Multiparameter Quantum Groups and Twisted Quasitriangular Hopf Algebras, Lett. Math. Phys. 20 (1990), 331–335. [120] A. Alekseev, A. Malkin and E. Meinrenken, Lie Group Valued Moment Maps, J. Diff. Geom. 48 (1998), 445–495. [121] P. Etingof and O. Schiffmann, Lectures on Quantum Groups, International Press, 1998. [122] P. D. Lax, Integrals of Nonlinear Equations of Evolution and Solitary Waves, Comm. Pure Appl. Math. 21 (1968), 467–490. [123] V. E. Zakharov and A. B. Shabat, Exact Theory of Two–Dimensional Self–Focusing and One–Dimensional Self–Modulation of Waves in Nonlinear Media, Sov. Phys. JETP 34 (1972), 62–69. [124] M. J. Ablowitz, D. J. Kaup, A. C. Newell, H. Segur, The Inverse Scattering Transform—Fourier Analysis for Nonlinear Problems, Stud. Appl. Math. 53 (1974), 249–315. [125] M. J. Ablowitz and P. A. Clarkson, Solitons, Nonlinear Evolution Equations and Inverse Scattering, Cambridge Univ. Press, 1991. [126] R. Beals and R. R. Coifman, Scattering and Inverse Scattering for First Order Systems, Comm. Pure Appl. Math. 37 (1984), 39–90. [127] A. S. Fokas, A. R. Its and A. V. Kitaev, The Isomonodromy Approach to Matrix Models, Comm. Math. Phys. 147 (1992), 395–430. [128] P. Deift and X. Zhou, A Steepest Descent Method for Oscillatory Riemann–Hilbert Problems, Ann. Math. 137 (1993), 295–368. [129] M. A. Semenov–Tian–Shansky, What is a Classical r–Matrix?, Funct. Anal. Appl. 17 (1983), 259–272. [130] A. G. Reyman and M. A. Semenov–Tian–Shansky, Integrable Systems: Group–Theoretical Approach, Birkhäuser, 1994. [131] J.–M. Maillet, New Integrable Canonical Structures in Two–Dimensional Models, Nucl. Phys. B 269 (1986), 54–76. [132] J.–M. Maillet, Hamiltonian Structures for Integrable Classical Theories from Graded Kac–Moody Algebras, Phys. Lett. B 167 (1986), 401–405. [133] E. K. Sklyanin, Quantum Version of the Method of Inverse Scattering Problem, J. Sov. Math. 19 (1982), 1546–1596. [134] A. Alekseev, A. Malkin and E. Meinrenken, Lie Group Valued Moment Maps, J. Differential Geom. 48 (1998), 445–495. [135] P. Ševera and A. Weinstein, Poisson Geometry with a 3–Form Background, Prog. Theor. Phys. Suppl. 144 (2001), 145–154. [136] L. Breen, On the Classification of 2–Gerbes and 2–Stacks, Astérisque 225 (1994). [137] J. C. Baez and U. Schreiber, Higher Gauge Theory, in: Categories in Algebra, Geometry and Mathematical Physics, Contemp. Math. 431, AMS (2007), 7–30. [138] U. Schreiber and K. Waldorf, Parallel Transport and Functors, J. Homotopy Relat. Struct. 4 (2009), 187–244. [139] T. Nikolaus and K. Waldorf, Four Equivalent Versions of Non–Abelian Gerbes, Pacific J. Math. 264 (2013), 355–420. [140] T. Pantev, B. Toën, M. Vaquié, G. Vezzosi, Shifted Symplectic Structures, Publ. Math. IHES 117 (2013), 271–328. [141] L. D. Faddeev, N. Y. Reshetikhin and L. A. Takhtajan, Quantization of Lie Groups and Lie Algebras, Leningrad Math. J. 1(1990), 193–225. [142] V. Chari and A. Pressley, A Guide to Quantum Groups, Cambridge Univ. Press, 1994. [143] V. G. Drinfel’d, Quasi–Hopf Algebras, Leningrad Math. J. 1(1990), 1419–1457. [144] N. Reshetikhin, Multiparameter Quantum Groups and Twisted Quasitriangular Hopf Algebras, Lett. Math. Phys. 20 (1990), 331–335. [145] I. V. Cherednik, Factorizing Particles on a Half–Line and Root Systems, Theor. Math. Phys. 61 (1984), 977–983. [146] E. K. Sklyanin, Boundary Conditions for Integrable Quantum Systems, J. Phys. A 21 (1988), 2375–2389. [147] L. D. Faddeev, N. Y. Reshetikhin, L. A. Takhtajan, Quantization of Lie Groups and Lie Algebras, Leningrad Math. J. 1(1990), 193–225. [148] V. Chari and A. Pressley, A Guide to Quantum Groups, Cambridge Univ. Press, 1994. [149] V. G. Drinfel’d, Quasi–Hopf Algebras, Leningrad Math. J. 1(1990), 1419–1457. [150] N. Reshetikhin, Multiparameter Quantum Groups and Twisted Quasitriangular Hopf Algebras, Lett. Math. Phys. 20 (1990), 331–335. [151] All analytic steps can be recast internally in G via charts; we use ρ only to access classical singular integral theory on matrices. [152] For strict crossed modules the 2–curvature has no quadratic B –term. In semi–strict settings one adds 1 6`3(A, A, A); see the outlook section.