Skip to content
Shadow Theory

Geometric Realization of Completed Source Relations Descent, Orbit Spaces, Invariant Relations, and Variational Response in Shadow Theory

Local data and global descent

Jeremy RodgersSection 4 of 12Original PDF ↗Publication DOI ↗

Local data and global descent

We now establish the first leg of the progression: compatible local data glue to global fields. This is the theorem that fixes the domain of geometric realization. The material is classical [6, 11, 3]; what is new is only the use to which it is put and the care with which the invariance and admissibility hypotheses are tracked.

Local data and compatibility equations

Let {Ui}i∈I\{U_i\}_{i\in I} be an open cover of MM. Suppose we are given, for each i∈Ii\in I, gi∈Γ(S2T∗Ui),Ai∈Ω1(Ui,g),Φi∈C∞(Ui,W),(10) g_i\in\Gamma\bigl(S^2T^*U_i\bigr), \qquad A_i\in\Omega^1(U_i,\mathfrak{g}), \qquad \Phi_i\in C^\infty(U_i,W),\tag{10} where each gig_i is an intrinsic smooth symmetric covariant 22-tensor field on the open set Ui⊆MU_i\subseteq M, of Lorentzian signature at every point; together with smooth transition functions hij ⁣:Ui∩Uj⟶G.(11) h_{ij}\colon U_i\cap U_j\longrightarrow G .\tag{11} If in addition each UiU_i is the domain of a chart ψi ⁣:Ui→ψi(Ui)⊆R4\psi_i\colon U_i\to\psi_i(U_i)\subseteq\mathbb{R}^4, we may pass to coordinate components. These live on the coordinate images, not on MM: setting (gˇi)ab:=((ψi−1)∗gi)abonψi(Ui)⊆R4,(12) \bigl(\check g_i\bigr)_{ab} :=\bigl((\psi_i^{-1})^*g_i\bigr)_{ab} \quad\text{on}\quad \psi_i(U_i)\subseteq\mathbb{R}^4 ,\tag{12} and writing ψij:=ψi∘ψj−1\psi_{ij}:=\psi_i\circ\psi_j^{-1} for the coordinate transition map on ψj(Ui∩Uj)\psi_j(U_i\cap U_j), the components of a single intrinsic tensor in two charts are related by the usual pullback law gˇj  =  ψij ∗ gˇionψj(Ui∩Uj),i.e.(gˇj)cd=∂ψij a∂xc ∂ψij b∂xd (gˇi)ab∘ψij.(13) \check g_j \;=\; \psi_{ij}^{\,*}\,\check g_i \qquad\text{on}\qquad \psi_j(U_i\cap U_j), \qquad\text{i.e.}\qquad \bigl(\check g_j\bigr)_{cd} =\frac{\partial \psi_{ij}^{\,a}}{\partial x^{c}}\, \frac{\partial \psi_{ij}^{\,b}}{\partial x^{d}}\, \bigl(\check g_i\bigr)_{ab}\circ\psi_{ij}.\tag{13} Equation (13) is a statement about components on coordinate images. It is distinct from the gluing condition (15) below, which is a relation among the intrinsic tensors gig_i on their overlaps.

Definition 3.1 (Compatible local geometric data).

The collection {gi,hij,Ai,Φi}\{g_i,h_{ij},A_i,\Phi_i\} is compatible if, on every nonempty overlap (respectively triple overlap): (14)hii=1,hji=hij−1,hijhjk=hik,(15)gi∣Ui∩Uj=gj∣Ui∩Uj,(16)Aj=Ad⁡hij−1Ai+hij∗θG,(17)Φj=ρ(hij−1)Φi.\begin{aligned} \qquad\text{(14)} h_{ii}&=1, & h_{ji}&=h_{ij}^{-1}, & h_{ij}h_{jk}&=h_{ik} , \\[0.3em] \qquad\text{(15)} g_i\big|_{U_i\cap U_j} &= g_j\big|_{U_i\cap U_j} , \\[0.3em] \qquad\text{(16)} A_j &= \operatorname{Ad}_{h_{ij}^{-1}}A_i + h_{ij}^{*}\theta_G , \\[0.3em] \qquad\text{(17)} \Phi_j &= \rho\bigl(h_{ij}^{-1}\bigr)\Phi_i . \end{aligned} Here (14) is the Čech cocycle condition; (15) is the gluing condition (agreement of restrictions) for a section of the sheaf S2T∗MS^2T^*M; and (16), (17) are the compatibility conditions for local connection forms and local matter representatives, θG\theta_G denoting the left Maurer–Cartan form on GG. For a matrix group G⊆GL(n)G\subseteq\mathrm{GL}(n), equation (16) takes the familiar form Aj=hij−1Ai hij+hij−1 dhij.A_j = h_{ij}^{-1}A_i\,h_{ij} + h_{ij}^{-1}\,\mathrm{d}h_{ij} .

Remark 13 (Čech H1H^1 for nonabelian GG).

A cocycle {hij}\{h_{ij}\} satisfying (14) represents a class in Hˇ1({Ui},G‾)\check H^1(\{U_i\},\underline{G}), where G‾\underline{G} is the sheaf of smooth GG-valued functions. For nonabelian GG this object is a pointed set of cocycle classes, the distinguished point being the class of the trivial cocycle, and not a group; two cocycles are equivalent when they are cohomologous in the sense hij′=λi−1hijλjh'_{ij}=\lambda_i^{-1}h_{ij}\lambda_j for smooth λi ⁣:Ui→G\lambda_i\colon U_i\to G. Passing to the direct limit over refinements gives the pointed set Hˇ1(M,G‾)\check H^1(M,\underline{G}), which classifies principal GG-bundles over MM up to isomorphism [11]. Only for abelian GG does this pointed set carry a natural group structure; we use it below only as a pointed set.

Assumption 3.3 (Regularity of the descent data).

The cover {Ui}\{U_i\} is locally finite; the gig_i, AiA_i, Φi\Phi_i, hijh_{ij} are smooth on their domains; and each gig_i has the signature fixed in Convention 2.1. If a weaker regularity class is used in Cgeo\mathcal{C}_{\mathrm{geo}}, the same statements hold with “smooth” replaced uniformly by that class, provided the class is closed under restriction and under the gauge transformations appearing in (16)–(17).

The descent theorem

Theorem 3.4 (Geometric descent).

Let {gi,hij,Ai,Φi}\{g_i,h_{ij},A_i,\Phi_i\} be compatible local geometric data in the sense of Definition 3.1, satisfying Assumption 3.3. Then these data determine:

  1. a global Lorentzian metric gg on MM with g∣Ui=gig|_{U_i}=g_i;

  2. a principal GG-bundle π ⁣:P→M\pi\colon P\to M with local trivializations over {Ui}\{U_i\} whose transition functions are {hij}\{h_{ij}\};

  3. a global principal connection AA on PP whose local representatives (pullbacks along the canonical local sections sis_i) are the AiA_i;

  4. a global section Φ\Phi of P×ρWP\times_\rho W whose local representatives are the Φi\Phi_i.

The resulting global tuple (g,P,A,Φ)(g,P,A,\Phi) is unique up to the bundle isomorphisms induced by change of local trivialization. Conversely, every global tuple (g,P,A,Φ)(g,P,A,\Phi) over MM arises in this way from compatible local data, for any cover trivializing PP.

The theorem reconstructs the metric gg; it does not by itself produce a time orientation. Membership of (M,g)(M,g) in the time-oriented admissibility class Cgeo\mathcal{C}_{\mathrm{geo}} of Definition 2.4 is an additional global condition on the glued metric, discussed in Remark 3.6.

Proof. (1) The metric. Condition (15) says exactly that the intrinsic tensors gig_i and gjg_j restrict to the same tensor field on Ui∩UjU_i\cap U_j. Since Γ(S2T∗−)\Gamma(S^2T^*-) is a sheaf on MM, the gluing axiom yields a unique global section g∈Γ(S2T∗M)g\in\Gamma(S^2T^*M) with g∣Ui=gig|_{U_i}=g_i for all ii. Nondegeneracy and signature are pointwise properties of the symmetric bilinear form gxg_x on TxMT_xM; every x∈Mx\in M lies in some UiU_i, where gx=(gi)xg_x=(g_i)_x is Lorentzian by hypothesis. Hence gg is Lorentzian. (If one prefers to present the data in coordinate components, the equivalent hypothesis is (13), which expresses that the gˇi\check g_i are the component arrays of a single intrinsic tensor; the invariance of signature under the linear isomorphism dψij(x)\mathrm{d}\psi_{ij}(x) is then Sylvester’s law of inertia.)

(2) The bundle. Form the disjoint union ∐i∈IUi×G×{i}\coprod_{i\in I}U_i\times G\times\{i\} and impose (x,a,i)  ∼  (x,  hji(x) a,  j),x∈Ui∩Uj,  a∈G.(18) (x,a,i)\;\sim\;\bigl(x,\;h_{ji}(x)\,a,\;j\bigr), \qquad x\in U_i\cap U_j,\; a\in G .\tag{18} Reflexivity follows from hii=1h_{ii}=1, symmetry from hij=hji−1h_{ij}=h_{ji}^{-1}, and transitivity from hkjhji=hkih_{kj}h_{ji}=h_{ki}; hence (18) is an equivalence relation. Let PP be the quotient with the quotient topology, π[(x,a,i)]:=x\pi[(x,a,i)]:=x.

We record the trivializations rigorously. Fix ii and x∈Uix\in U_i. Every equivalence class lying over xx has exactly one representative whose third entry is ii: existence, because if a class contains (x,b,j)(x,b,j) then it contains (x,hij(x)b,i)(x,h_{ij}(x)b,i); uniqueness, because (x,a,i)∼(x,a′,i)(x,a,i)\sim(x,a',i) forces a′=hii(x)a=aa'=h_{ii}(x)a=a. Hence the map φi ⁣:π−1(Ui)⟶Ui×G,φi([(x,a,i)])=(x,a)(19) \varphi_i\colon \pi^{-1}(U_i)\longrightarrow U_i\times G, \qquad \varphi_i\bigl([(x,a,i)]\bigr)=(x,a)\tag{19} is a well-defined bijection, with inverse φi−1(x,a)=[(x,a,i)]\varphi_i^{-1}(x,a)=[(x,a,i)]; it is a homeomorphism by definition of the quotient topology, and we give PP the smooth structure for which the φi\varphi_i are diffeomorphisms (consistent, since the change of trivialization φi∘φj−1(x,a)=(x,hij(x)a)\varphi_i\circ\varphi_j^{-1}(x,a)=(x,h_{ij}(x)a) is smooth). The right GG-action [(x,a,i)]⋅b:=[(x,ab,i)][(x,a,i)]\cdot b:=[(x,ab,i)] is well defined, because hjih_{ji} acts by left multiplication and hence commutes with right multiplication; it is free and transitive on fibres, and each φi\varphi_i is GG-equivariant. Thus π ⁣:P→M\pi\colon P\to M is a principal GG-bundle with transition functions hijh_{ij}.

(3) The connection. Let si ⁣:Ui→Ps_i\colon U_i\to P be the canonical local sections si(x):=φi−1(x,1)s_i(x):=\varphi_i^{-1}(x,1), so that sj=si⋅hijs_j=s_i\cdot h_{ij} on Ui∩UjU_i\cap U_j. A principal connection A∈Ω1(P,g)A\in\Omega^1(P,\mathfrak{g}) is determined by its local representatives Ai:=si∗AA_i:=s_i^*A, and a family {Ai}\{A_i\} arises from a (necessarily unique) global AA if and only if it obeys the law induced by sj=si⋅hijs_j=s_i\cdot h_{ij}, namely sj∗A  =  Ad⁡hij−1(si∗A)+hij∗θG,s_j^*A \;=\; \operatorname{Ad}_{h_{ij}^{-1}}\bigl(s_i^*A\bigr)+h_{ij}^*\theta_G, which is precisely (16). Given (16), define AA over UiU_i by the pullback identity (φi−1)∗A  =  Ad⁡a−1(πU∗Ai)  +  θGonUi×G,(20) \bigl(\varphi_i^{-1}\bigr)^{*}A \;=\; \operatorname{Ad}_{a^{-1}}\bigl(\pi_U^{*}A_i\bigr)\;+\;\theta_G \qquad\text{on}\qquad U_i\times G,\tag{20} where πU ⁣:Ui×G→Ui\pi_U\colon U_i\times G\to U_i and the projection Ui×G→GU_i\times G\to G are understood, aa denotes the GG-coordinate, and θG\theta_G is pulled back along that projection. Formula (20) is an equality of g\mathfrak{g}-valued 11-forms on the manifold Ui×GU_i\times G; it is not an evaluation at a point. Two such local definitions agree on π−1(Ui∩Uj)\pi^{-1}(U_i\cap U_j): transporting (20) for jj through φi∘φj−1(x,a)=(x,hij(x)a)\varphi_i\circ\varphi_j^{-1}(x,a)=(x,h_{ij}(x)a) and using the standard identity (Lh∘)∗θG=θG(L_{h}\circ)^*\theta_G=\theta_G together with the Maurer–Cartan expansion of d(hija)\mathrm{d}(h_{ij}a) reproduces (20) for ii exactly when (16) holds. The resulting 11-form is GG-equivariant, Rb∗A=Ad⁡b−1AR_b^*A=\operatorname{Ad}_{b^{-1}}A, and reproduces the fundamental vector fields, A(ζX)=XA(\zeta_X)=X; hence it is a principal connection [6].

(4) The matter section. Sections of P×ρWP\times_\rho W correspond bijectively to ρ\rho-equivariant maps Φ~ ⁣:P→W\tilde\Phi\colon P\to W with Φ~(p⋅a)=ρ(a−1)Φ~(p)\tilde\Phi(p\cdot a)=\rho(a^{-1})\tilde\Phi(p), and such maps correspond to families Φi:=Φ~∘si\Phi_i:=\tilde\Phi\circ s_i obeying Φj=Φ~∘(si⋅hij)=ρ(hij−1)Φi\Phi_j=\tilde\Phi\circ(s_i\cdot h_{ij})=\rho(h_{ij}^{-1})\Phi_i, which is (17). Hence the Φi\Phi_i glue to a unique global section Φ\Phi.

Uniqueness up to isomorphism. Suppose two systems of local trivializations over a common cover (or over covers admitting a common refinement, after passing to it) produce compatible data {gi,hij,Ai,Φi}\{g_i,h_{ij},A_i,\Phi_i\} and {gi,hij′,Ai′,Φi′}\{g_i,h'_{ij},A'_i,\Phi'_i\} with cohomologous cocycles, hij′=λi−1hijλjh'_{ij}=\lambda_i^{-1}h_{ij}\lambda_j for smooth λi ⁣:Ui→G\lambda_i\colon U_i\to G, together with Ai′=Ad⁡λi−1Ai+λi∗θGA'_i=\operatorname{Ad}_{\lambda_i^{-1}}A_i+\lambda_i^*\theta_G and Φi′=ρ(λi−1)Φi\Phi'_i=\rho(\lambda_i^{-1})\Phi_i. Then the maps φi′−1(x,a)↦φi−1(x,λi(x)a)\varphi_i'^{-1}(x,a)\mapsto\varphi_i^{-1}(x,\lambda_i(x)a) agree on overlaps, exactly because the cocycles are cohomologous, and assemble into a GG-equivariant bundle isomorphism P′→PP'\to P covering idM\mathrm{id}_M, carrying A′A' to AA and Φ′\Phi' to Φ\Phi. That is, it is a vertical arrow of Ggeo\mathfrak{G}_{\mathrm{geo}}. Conversely, an arrow of Ggeo\mathfrak{G}_{\mathrm{geo}} between global tuples, read in local trivializations, exhibits the two systems of local data as related in exactly this way. Hence the global tuple is well defined precisely as an object of Ggeo\mathfrak{G}_{\mathrm{geo}} up to isomorphism.

Converse. Given a global tuple, choose any cover {Ui}\{U_i\} trivializing PP (possible since MM is paracompact and PP is locally trivial); the induced gi:=g∣Uig_i:=g|_{U_i}, hijh_{ij}, Ai=si∗AA_i=s_i^*A, Φi\Phi_i satisfy (14)–(17) by the computations above. ◻

Remark 16 (Changing the coordinate atlas is not a base diffeomorphism).

The uniqueness clause above is stated for changes of local trivialization, which induce vertical bundle isomorphisms covering idM\mathrm{id}_M. A change of coordinate atlas is a a different operation: re-charting MM changes the component arrays gˇi\check g_i of (12) according to (13) while leaving the intrinsic tensor gg, and hence the object Ψ\Psi and its class [Ψ]∈Sgeo[\Psi]\in S_{\mathrm{geo}}, entirely unchanged. It does not automatically produce a nontrivial base diffeomorphism, and it does not move a point of SgeoS_{\mathrm{geo}}. Nontrivial base diffeomorphisms enter only through the arrows of Ggeo\mathfrak{G}_{\mathrm{geo}} with f≠idMf\neq\mathrm{id}_M, which act on the geometry itself and not merely on its coordinate presentation.

Remark 17 (Descent and time orientation).

Theorem 3.4 glues the local metrics to a global Lorentzian gg, but it does not produce the time orientation t\mathfrak{t} that Definition 2.4 demands. Whether gg admits any time orientation at all is a global topological condition on (M,g)(M,g) (it holds if and only if the timelike cone bundle of gg has a global continuous section of one of its components), and no purely local compatibility condition can force it. Two consistent options are available, and a model must declare one.

Option (a): impose it after descent. Glue gg as above, then require, as an additional global admissibility condition, that gg be time-orientable and that a time orientation t\mathfrak{t} be chosen and recorded in B\mathcal{B}. Under this option Theorem 3.4 reconstructs the metric, and membership of the reconstructed configuration in Cgeo\mathcal{C}_{\mathrm{geo}} is a further global condition, verified separately.

Option (b): descend it. Adjoin to the local data a choice, for each ii, of a gig_i-timelike vector field τi∈Γ(TUi)\tau_i\in\Gamma(TU_i), and impose the additional compatibility condition that on every nonempty overlap τi\tau_i and τj\tau_j lie in the same component of the timelike cone, gi(τi,τj)<0on Ui∩Uj(21) g_i(\tau_i,\tau_j)<0 \qquad\text{on } U_i\cap U_j\tag{21} (with signature (−,+,+,+)(-,+,+,+), so that two timelike vectors have negative inner product exactly when they are causally co-oriented). Condition (21) is a genuine gluing condition: the relation “lies in the same cone component as” is an equivalence relation on timelike vectors at each point, and the local choices τi\tau_i define, by (21), a coherent global selection of one component of the cone bundle at every x∈Mx\in M. That selection is by definition a time orientation t\mathfrak{t} of the glued gg, and it is unique given the τi\tau_i. Descent then reconstructs the pair (g,t)(g,\mathfrak{t}), and no further global condition on time orientation is needed. (A partition of unity subordinate to {Ui}\{U_i\} shows that t\mathfrak{t} is realized by an actual global timelike vector field, since the cone components are convex.)

Corollary 3.7 (Realizability boundary).

A collection of proposed local metric, connection, and matter expressions represents a global geometric source configuration only if the cocycle and overlap conditions (14)–(17) hold on all overlaps, and only if the resulting global tuple lies in the declared admissibility class Cgeo\mathcal{C}_{\mathrm{geo}}, which, by Remark 3.6, includes the global time-orientation condition. Consequently, an abstract tuple of relation values that admits no compatible descent data lies outside the geometric realization domain: it is a point of the abstract product T×∏αVαT\times\prod_\alpha V_\alpha that is not in the image of any geometric evaluation map.

Proof. Immediate from Theorem 3.4: any global configuration restricts to compatible local data, so failure of compatibility for every presentation of the proposed values is an obstruction to the existence of a configuration presenting them. Membership in Cgeo\mathcal{C}_{\mathrm{geo}} is a further restriction, since Theorem 3.4 produces a global tuple but does not certify that it satisfies the declared causal, boundary, or topological conditions. ◻

Corollary 3.7 is the first and most basic restriction on the abstract completion of Paper 3: its coordinates must be jointly realizable by global geometric objects. Two further restrictions, invariance (Section 4) and separation (Section 5), follow.

Remark 19 (Obstructions are genuine).

That the constraint is not vacuous is visible already in cohomological terms: the isomorphism class of PP is the class of {hij}\{h_{ij}\} in the pointed set Hˇ1(M,G‾)\check H^1(M,\underline{G}) (Remark 3.2), and characteristic numbers computed from AA are constrained by that class [4, 8]. A demand such as “a U(1)U(1) bundle over S2S^2 with first Chern number 12\tfrac12” fails at the cocycle stage; there is no compatible descent datum, hence no configuration, hence no point of SgeoS_{\mathrm{geo}}, hence no point of the realized completion. The Cartesian product of independently nominated relation values is, in general, strictly larger than what geometry attains.

Back to top ↑