4.4.3 Differential forms taking values in a fiber bundle. List of formulas
The next section reproduces the presentation of [BOU 82a], sections 7.3 and 7.8, with slight changes to the order and a few simplifications.
(I) Let π : N → B be a vector bundle and consider the vector bundle Altp (M; N) (M = T (B)) whose fiber over an arbitrary point b ∈ B is Altp (M; N)b := Altp (Mb; Nb) (section 4.2.6, Example 4.14(3)). If N is the trivial bundle B × F, where F is a Banach space, write Altp (M; F) for Altp (M; N).
Definition 4.29
A differential p-form on an open subset U of B taking values in N is a morphic section of Altp (M; N), i.e. an element of Ωp (U; N):=Γ (U, Altp (M; N)).
Clearly, for any open subset U of B, Ωp (U ; N) is a CrU-module and the mapping U → Ωp (U ; N) is a sheaf of Cr-Modules.
Remark 4.30
i)In the above, we can replace M by an arbitrary vector bundle of class Crand base B. We can alternatively write Ωp (U ; N):=Γ (U, Altp (M; N)), where the fiber bundle M is implicit on the left-hand side.
ii)With p = 0, Ω0 (U; N) is the set of sections of class Crof the form G : U → N : b ↦ (b, nb) that are liftings of 1U(thus, nb ∈ Nband π ° G = 1U).
iii)If N is the trivial bundle B × F, whereFis a Banach space, then Ωp (U ; N) is written as Ωp (U ; F).
(II) Exterior product Let πi : Ni → B (i = 1,2) and π : N → B be three vector bundles of class Cr and Φ a mapping from N1 ×ΒN2 into N. Suppose that, for every b0 ∈ B, there exist an open neighborhood U of b0 in B and vector charts ti = (Ui, φi, Fi) of πi (i = 1, 2) and t = (U, φ, F) of π, together with a mapping λ of class Cr in ℒF1×F2F, satisfying the following condition: for every b ∈ U and all (x1, x2) ∈ F1 × F2 (with the notation of Definition 3.22(i), Condition (V)),
tb∘λbx1x2=Φtb1x1tb2x2.
Definition 4.31
The mapping Φ defined above is said to be a coupling of the fiber product N1 ×ΒN2(section 2.3.9) in N.
Suppose that one such coupling is given and let M be a fiber bundle with base B and of class Cr. For every b ∈ B, there is a continuous bilinear mapping
Φb:AltpMN1b×AltqMN2b→Altp+qMNb.
The collection of these continuous bilinear mappings determines a coupling
u:AltpMN1×AltqMN2→Altp+qMN.
Given sections ω1, ω2 of Altp(M; N1) and Altq(M; N2) on U, the section u (ω1, ω2) of Altp + q(M; N) on U is written as ω1 ∧Φω2 (see Remark 4.18).
Definition 4.32
We say that ω1 ∧Φω2is the exterior product of ω1and ω2. This exterior product is written as ω1 ∧ ω2whenever the coupling Φ is implicitly clear.
If the si (i = 1, …, p + q) are morphic sections of M on U, then:
ω1∧Φω2s1…sp+q=∑σεσΦω1sσ1…sσpω2(sσp+1…sσp+q),
where the sum ranges over permutations of {1, …, p + q} such that σ (1) < … < σ (p) and σ (p + 1) < … < σ (p + q) (see Lemma-Definition 4.10(i)).
(III) De Rham algebra: generalization An algebra bundle of base B is a vector bundle A of base B equipped with a coupling from A ×ΒA into A. In the following, the fibers Ab (b ∈ B) are assumed to be associative and commutative algebras with neutral element eb ([P1], section 2.3.10(I)). Let π : M → B be a vector bundle with base B (e.g. T (B)). Write Ωp (U; A) (where U is an open subset of B) for the CrU-module formed by the morphic sections of Altp (M; A) on U.
The exterior product allows the direct sum Ω(U; A) = ⊕p ≥ 0Ωp(U; A) to be equipped with the structure of an associative, anticommutative and graduated algebra, again called the de Rham algebra (section 4.4.1(III)).
If ω = ω1 ∧ … ∧ ωp, where ωi ∈ Ω1 (U; A), and si is a morphic section of M on U (i = 1,…, p), then (see [4.21]):
ωs1…sp=detωisj.
(IV) Interior product Let p ≥ 1. There exists a coupling i from M ×Β Altp (M; A) into Altp–1 (M; A) whose restriction to each fiber is the interior product (Remark 4.12). If s is a section of M on the open subset U of B and ω ∈ Ωp (U; A), write i (s) ω for the section i (s, ω) of Altp–1 (M; A) on U. Then, for every b ∈ U,
isωb=sb┘ωb.
We say that i (s) ω is the interior product of s and ω. From this definition and the formulas listed in section 4.2.5, we can deduce the following relations (where the sj are sections of M):
isω⏟degreeps1…sp−1=ωss1…sp−1,is∘is=0,isω⏟degree1=ωs,is.ω⏟degreep∧ω′=isω∧ω′+−1pω∧isω′,isω1∧…∧ωp⏟forms of degree1=∑i=1p−1i+1ωisω1∧…∧ωi^∧…∧ωp.
The interior product is an antiderivation of degree – 1 of the de Rham algebra ([P1], section 2.3.12).
(V) Preimage Let f : B′ → B be a morphism of manifolds and ω ∈ Ωp (B; N). Let f⁎ (N) be the preimage of N under f (section 3.4.6). There exists a uniquely determined differential p-form f⁎ (ω) such that [4.22] holds (mutatis mutandis). For every family of vector fields X′1, …, X′p of class Cr on B′, the mapping
In particular, a differential 0-form G on B′ taking values in N is in fact just a lifting of class Cr of f into N, i.e. a mapping (of class Cr) G : B′ → N such that G (b′) ∈ Nf (b′) for all b′ ∈ B′.
(VI) Vector-valued differential forms The fiber bundle Ωp (B; N) is trivial if and only if N is trivial (section 3.4.1, Example 3.26(a)), or in other words N = B × F, where F is a Banach space (exercise). If so, the fibers Nb can be identified with F. We write Altp (M; F) for Altp (M; N), Ωp (B; F) for Ωp (B; N), and we say that ω is a vector-valued differential p-form taking values in F. In particular, if K=ℝ, N = T (B), and F=ℂ, then ω is a complex differential p-form on the real manifold B.
The usual fiber bundle AltpTBK can also be written as AltpTBKB=AltpTBB×K, and
ΩpB=ΓAltpTBB×K.
[4.25]
Remark 4.33
The space of differential p-forms on anℱNorSNmanifold can be defined using[4.25]([KRI 97],Chapter VII, section 33).
4.4.4 Orientation
For the rest of this chapter, excluding section 4.5.1, every manifold is a locally finite-dimensional differential manifold and F always denotes a Banach space.
(I) Orientation of a vector space Let E be a real m-dimensional vector space. We know that dim∧mE=1 (section 4.2.3(I)), so detE≔∧mE is the union of two closed half-lines on the real axis with opposite directions and origin 0. These half-lines are written as O and –O. The set {O, –O} formed by these half-lines is written as Or (E).
(II) Orientation of a real manifold Let Β be a manifold and M = T (B). Write
OrM=∪•b∈BOrMb.
Lemma 4.34
Letπ˜:OrM→Bbe the mapping defined byπ˜bO=bfor every b ∈ Β. There exists a unique topological space structure on OrMfor which the following two conditions are satisfied:
i)π˜is continuous.
ii)Ifsis a continuous, everywhere non-zero section of the vector bundle det (M) (whose fibers are the det (Mb), b ∈ Β, with the notation of (I), implying thats(b) ∈ det (Mb) = O(b) ∪ (− O((b))) on an open subset U of Β ands (b) ∈ O (s (b)) for every b ∈ U, then the mapping B → OrM : b ↦ O (s (b)) is continuous (exercise).
Suppose that the topological space OrM is equipped with the manifold structure determined by taking the preimage under π˜ of the manifold structure of Β (Remark 2.44), and consider the fibration π˜:OrM→B. The multiplicative group {± 1} acts simply transitively (and hence freely) ([P1], section 2.2.8(II)) on OrM by O ↦ –O. The manifold of orbits of this action is OrM \{± 1} ≅ B.
Corollary-Definition 4.35
1)The fibration π : OrM → Β is a principal bundle with structural group {± 1}. This principal bundleB˜is a covering of Β with fiber type {± 1} (section 3.3.3), namely a covering of two leaves.
2)The principal bundleB˜is said to be the orientation covering. An orientation of Β is a continuous section O : b ↦ (b, Ob) ofB˜Ob∈±1. If any such section exists, the pair (B, O) is said to be an oriented manifold. The orientations O and –O are said to be opposite.
3)A manifold B is said to be orientable if there exists an orientation on B.
4)A pure m-dimensional manifold B is orientable if and only if one of the following equivalent conditions is satisfied:
i)The orientation coveringB˜is the trivial bundle B × {± 1}.
ii)There exists a continuous differential m-form v0such that v0 (b) ≠ 0 for every b ∈ B; if so, v0is of class C∞, i.e. v0 ∈ Ωm (B). Since dim (Ωm (B))b = 1, the sign of v0 (b) must be constant on B, and v0determines an orientation O : b ↦ (b, sgn (v0 (b))) of B (where sgn denotes the sign).
iii)There exists an atlas of B whose charts (Ui, φi, ni) satisfy the property that, if Ui ∩ Uj ≠ ∅ (which implies ni = nj), then
∂φi1…φini∂φj1…φjnj>0
on Ui ∩ Uj.
5)Hence, if B is a pure orientable m-dimensional manifold, then the relation ∼ on Ωm (B) defined by v ∼ v′ if sgn (v (b)) = sgn (v′ (b)) (where b is an arbitrary point of B) is an equivalence relation. The orientation O : b ↦ (b, sgn (v (b))) is the equivalence class of v, written asv¯.
Proof
(4): See [DIE 93], Volume 3, (16.21.1), (16.21.16). (5): exercise.
Corollary 4.36
1)Let B be a manifold and b some point of B. ByCorollary-Definition 4.35(4), there exists an open neighborhood U of b in B that is orientable.
2)A manifold B is orientable if and only if each of its connected components is orientable.
3)Let (B, O) be a pure m-dimensional oriented manifold and v0 ∈ O. Every differential m-form ω ∈ Ωm (B) can be uniquely written in the form ω = f.v0, wheref:B→ℝis of class C∞. Given b ∈ B, writeωb><¯0iffb><¯0.
4)Let (B, O) be a pure m-dimensional oriented manifold. If ω ∈ Ωm (B; F), there exists a unique mapping f : B → Fof class C∞satisfying ω = f.v0(exercise).
Definition 4.37
Let (B, O) be an oriented manifold and v0 ∈ O. A sequence (Z1,…, Zm) of vector fields is said to be positive or direct (respectively negative or retrograde) if, for every b ∈ B,
v0bZ1b∧…∧Zmb>0resp.<0.
Example 4.38
i)The spaceℝmis orientable and the canonical m-form dx1 ∧ … ∧ dxm(where xiis the i-th coordinate function in the canonical basis) defines its canonical orientation.
ii)More generally, the underlying manifold of a finite-dimensional real Lie groupGis always orientable. Indeed, suppose thatGis m-dimensional, and let z∨be an m-covector such that ze∨ ≠ 0 (where e is the neutral element); then the differential m-form g ↦ γ (g) z∨(of class Cω), where γ is left translation (section 2.4.1(I)), is non-zero at every point.
iii)Every simply connected manifold and every parallelizable manifold (Definition 3.28) is orientable ([NAR 73], Corollary 2.7.6;[LEE 02], Proposition 10.5).
iv)In particular, the sphereSn(see footnote 2, p. 98) is orientable.
v)Every finite product of orientable manifolds is orientable. If B1, B2are two manifolds with orientations O1, O2respectively, the mapping (b1, b2) ↦ Ob1Ob2is an orientation of B1 × B2written as O1 ⊗ O2.
vi)Let B bean oriented manifold with orientation O. Let U be a submanifold of B. The mapping O |Uis an orientation of U. Hence, every submanifold of an orientable manifold is orientable.
vii)It can be shown that any finite-dimensional pure differential manifold B0underlying a holomorphic manifold B is orientable ([DIE 93], Volume 3, (16.21.13)).
viii)The Möbius strip (seeFigure 3.2insection 3.3.1and footnote 3, p. 98) and the Klein bottle (see footnote 6, p. 70) are not orientable. The Möbius strip is a striking example of a non-orientable manifold: any reader who wishes to experiment with the concept of orientation can make a Möbius strip by gluing together the two ends of a strip of paper with a half-twist. Now, draw a pencil line along the middle of the strip – the line will almost magically reach “the other side” of the strip from the starting point.
Let (B, O) be an oriented manifold. If we write this oriented manifold as B↷, we can write B↶ or −B↷ for the manifold equipped with the opposite orientation.
4.4.5 Integral of a differential form of maximal degree
(I) Volume integrals inℝm
Lemma 4.39
Let U, U′ be open subsets ofℝmand u : U → U′ a diffeomorphism. For every x ∈ U, letJ=det∂u∂xbe the Jacobian of u (section 1.2.2(IV)). Let λU⊗ m ≔ λ⊗ m |Uand λU′⊗ mbe the Radon measures induced on U and U′, respectively, by the Lebesgue measure onℝm([P2], section 4.1.5(I)). Then, the image of | J |. λU⊗ munder u is λU′⊗ m([P2], section 4.1.5(II)); in other words, for every functionf∈KU′, whereKU′denotes the space (of typeℒsℱ) of compactly supported continuous functions from U′ intoℝ([P2], section 4.1.4(IV)), we have the following change of variable formula, by[4.24]:
∫U′fx′.dλ⊗mx′=∫Ufux.Jx.dλ⊗mx.
This relation still holds if we replace f∈KU′ by a λ⊗ m-integrable mapping f : U′ ↦ F ([P2], section 4.1.2).
Example 4.40
Let U and U′ be two open intervals on the real line and u a diffeomorphism from U onto U′. Let f be a continuous function on U taking values in the Banach spaceF. The function t ↦ u (t) is monotone, sou.thas constant sign ε on U, and:
∫Uft.dt=∫U′fut.ε.u.t.dt=∫U′fut.u.t.dt.
Example 4.41
Let us calculate the volume V of a sphere of radius R.
1)Cartesian coordinates. Pick the center of the sphere as the origin, and begin by calculating the volume of the hemisphere z ≥ 0. We can use the Fubini-Tonelli theorem ([P2], section 4.1.3(III)) to do this by cutting the hemisphere into “slices” of infinitely small thickness dz. Each slice is a cylinder with a circular cross-section of radiusR2−z2and thickness dz. The volume of the hemisphere is therefore given by
∫0RπR2−z2dz=23πR3,
which gives us the classical formulaV=43πR3.
2)Spherical coordinates. In mathematics, the radial, azimuthal, and zenithal coordinates {r, ϕ, θ} are defined as shown inFigure 4.1, with r ≥ 0, 0 ≤ θ ≤ 2π, and 0 ≤ ϕ ≤ π (in physics, the symbols θ and ϕ are often swapped). These coordinates satisfy the relations:
Let C be the point with Cartesian coordinates (x, y, z). Then, dC→=e1→dx+e2→dy+e3→dz, wheree1→, e2→, e3→ are the vectors of the canonical basis. In spherical coordinates, writeur→, uϕ→, uθ→ for the unit vector along the direction ofOC→, the unit vector tangent to the meridian in the direction of increasing ϕ, and the unit vector parallel to C in the direction of increasing θ, respectively. Then
, OC→=rur→, OC→=ur→.dr+r∂ur→∂ϕ.dϕ+r∂ur→∂θ.dθ=∂∂rdr+∂∂φdφ+∂∂θdθ with ur→=cosθsinϕ.e1→+sinθsinϕ.e2→+cosϕ.e3→, so
dOC→=dr.ur→+1r.uϕ→+1rsinϕ.uθ→,
[4.28]
∂∂r=ur→,∂∂ϕ=r∂ur→∂ϕ=ruϕ→,∂∂θ=r∂ur→∂θ=rsinϕ.uθ→.
[4.29]
The orthonormal frameur→uϕ→uθ→ is positively oriented, since the determinant[4.27]is positive.
3)Calculation based on the radial coordinate, the longitude, and the latitude. The coordinates are now (r, φ, λ) (see Example 2.12,section 2.2.1(II)). In radians, the latitude φ and the longitude λ may now be expressed as a function of the zenith φ and the azimuth θ byφ=π2−ϕand λ = θ – π. Starting from the expressions[2.1], an analogous calculation gives dxdydz = r2cos φdrdφdλ, and
Let B be a pure m-dimensional manifold. There exists a positive Radon measure μ on B ([P2], section 4.1.5(V)) with the following properties:
i)For every chart c = (U, φ, m) of Β, the image under φ ([P2], section 4.1.5(II)) of the induced measure μUis of the form fc. λφ(U)⊗ m, where fcis a function of class C∞that does not vanish on φ (U).
ii)In other words, for every functiong∈KU,
∫Ugbdμb=∫φUgςfcςdλ⊗mς.
iii)Any Radon measure μ with the property (i) is said to be Lebesgue on B. Any two Lebesgue measures μ, μ′ on B are equivalent in the sense that each is absolutely continuous with respect to the other ([P2], section 4.1.6(III)) and each has a density function ofclass C∞with respect to the other.
(III) Integral of a form of maximal degree over anm-dimensional oriented manifold Let B↷ be a pure oriented manifold of dimension m ≥ 0 and let ω be a differential m-form taking values in a Banach space F. our next task is to give meaning to the quantity
∫B↷ω.
By Corollary-Definition 4.35(5), there exists a differential m-form v0 ∈ Ωm (B) that belongs to the orientation of B↷. Let c = (U, φ, m) be a chart of B such that the open set U is connected. For every point ζ = (ζ1,…, ζm) ∈ φ (U), we can write (with the same notation as above):
v0φ−1ς=fcς.dς1∧…∧dςm.
Since U is connected, fc is of class C∞ and has constant sign in φ (U). Let μv0, U, U be the positive Radon measure on U defined by
μv0,U=φ−1fc.λφU⊗m.
If we proceed in the same way for another arbitrary chart c′ = (U′, φ′, m) of B, where U′ is also connected, we obtain a measure μv0, U′ that is positive on U′. It is easy to show ([DIE 93], Volume 3, (16.24.1)) that, if U ∩ U′ ≠ ∅, the restrictions of μv0, U and μv0, U′′ to U ∩ U′ are equal. The positive measures μv0, U (as U ranges over the set of connected open subsets of B, which form a covering of B) are, therefore, the restrictions to U of a unique positive Radon measure μv0 (which is Lebesgue) defined on B. With the same hypotheses and notation as Corollary 4.36(4):
Corollary-Definition 4.43
The differential m-form ω is said to be integrable (overB↷)if f is μv0-integrable. If so, define:
∫B↷ω≔∫B↷f.dμv0.
This quantity only depends on the orientation ofB↷and not on the particular choice of v0made to specify this orientation.
Let ω ∈ Ωm (B; F) be an integrable differential m-form on B↷ taking values in F. With the notation introduced at the end of section 4.4.4(II):
∫−B↷ω=∫B↷ω.
If g∈KB and ω ∈ Ωm (B; F), then g.ω is a continuous m-form and g↦∫B↷g.ω is a Radon measure [ω] taking values in F ([P2], section 4.1.5(VIII)).7
Definition 4.44
The Radon measure [ω] defined on B is called the volume form onB↷determined by the differential m-form ω.
The above leads to the following result ([DIE 93], Volume 3, (16.24.2)):
Corollary 4.45
If ω ∈ Ωm (B) belongs to the orientation of B↷, then the volume form
ω:KB→ℝ:g↦∫B↷g.ω
is a positive Lebesgue measure onB↷. Conversely, every positive Lebesgue measure onB↷is of the form [ω], ω ∈ Ωm (B).
(IV) Orientation of a morphism
Definition 4.46
Let B, B′ be locally finite-dimensional manifolds and f : B → B′ a morphism. An orientation of f is a morphismf˜:B˜→B˜′:bOb→b′Ob′that makes the diagram
B˜→f˜B′˜↓π↓π′B→fB′
commute and which is compatible with the group action {± 1}. In other words,f˜bOb±1=fb±Ofb′. We write thatf˜=fε, ε ∈ {–1, 1}.
Corollary-Definition 4.47
1)Let B, B′ be orientable manifolds, f : B′ → B a diffeomorphism, andf˜=fεan orientation of f. Let O be an orientation of B and writeB↷=BO. There exists a unique orientation O′ on B′ such that, settingB↷=B′O′, every differential form ω ∈ Ωm (B; F) satisfies
∫B↷ω=ε∫B′↷f∗ω
[4.30]
(exercise). The orientation O′ is said to be associated with O byf˜.
2)Conversely, letB↷, B′↷be two oriented manifolds and f : B′ → B a diffeomorphism. If therelation [4.30]is satisfied for every differential form ω ∈ Ωm (B; F) and ε = + 1, we say that f preserves orientations, or that f reverses orientations if ε = –1. The pairf˜=fεis an orientation of f, where f is viewed as a diffeomorphism from B′ onto B (and B and B′ are the orientable manifolds underlyingB↷andB′↷, respectively).
(V) Canonical orientation of the orientation covering Let B be a manifold and .π˜:B˜→B the orientation covering (Corollary-Definition 4.35(1)). Let b˜=bOb∈B˜. The projection .π˜:B˜↠B is the linear mapping b˜↦b. Its tangent linear mapping Tb˜π˜:Tb˜B˜→TbB is an isomorphism, so π˜ is a local diffeomorphism (Theorem 2.61(2)), i.e. a diffeomorphism from an open subset U of b˜ onto an open subset V of b. We can choose U and V to be orientable (Corollary 4.36(1)); thus, there exists an orientation O:b˜↦Ob′ of V taking the value Ob at the point b. We have π˜−1b=bObb−Ob. Let ω ∈ O, ω˜=π˜∗ω, and O˜=ω˜¯ (see Corollary-Definition 4.35(5)); then b˜↦O˜b˜ is an orientation of B ([DIE 93], Volume 3, (16.21.6)), which gives us the following result:
Theorem 4.48
The manifoldB˜is orientable.
Remark 4.49
(1)The space.π˜:B˜↠Bis a covering of two leaves. It therefore has a canonical involution8ι:B˜→B˜that permutes these leaves over each point b ∈ B. IfO˜is an orientation ofB˜, then the orientation ofB˜associated with ι (Corollary-Definition 4.47) is−O˜. Therefore, bothO˜and−O˜are orientations ofB˜; nevertheless, we say thatO˜is the canonical orientation. This ambiguity is irrelevant in practice becauseO˜is unique up to permutation of the two leaves ofB˜.
2)Conversely, ifπ¯:M→Bis an oriented covering of class C∞with two leaves and its canonical involution associates a given orientation of M with the opposite orientation, thenπ¯:M→Bis isomorphic toπ˜:B˜→B([LEB 82],Chapter I, section 5.C, Theorem 2).
4.4.6 Differential forms of odd type
(I) Definition The fiber bundle ℝ˜≔B˜×±1 associated with B˜ of fiber type ℝ (section 3.5.5) is called the bundle of scalars of odd type. Let N be a vector bundle of base B. A differential p-form ω¯∈ΩpBℝ˜⊗N is said to be a differential p-form of odd type on B taking values in N. Let .π˜:B˜↠B be the projection. There exists a bijection
•˜:ω¯↦∼ω˜
[4.31]
between the differential p-forms of odd type on B taking values in N and the differential p-forms on B˜ taking values in π˜∗N such that ω˜−O=−ω˜O for every orientation O∈B˜.
Remark 4.50
a)When B is equipped with an orientation O, the bijection
ΩpBN→ΩpBℝ˜⊗N:ω↦ω¯≔O⊗ω
[4.32]
allows us to identify the usual differential p-forms ω onB↷with the differential p-forms of odd typeω¯onB↷. Thus, in most cases, we do not need to talk about differential forms of odd type on an oriented manifold. However, the concept of differential p-form of odd type becomes crucial on manifolds without an orientation (especially when these manifolds are non-orientable). Whenever we talk about p-forms on these manifolds, we will always be referring to differential p-forms of odd type.
b)We also say that an ordinary differential form (belonging to Ωp (B; N))is even and that a differential form of odd type (belonging toΩpBℝ˜⊗N)is odd ([DER 84],Chapter II). This more simple terminology is adopted below.
If N is the trivial bundle B × F, we write ΩpBℝ˜⊗F for the space ΩpBℝ˜⊗N, and we say that the odd p-form ω¯∈ΩpBℝ˜⊗F takes values in F.
(II) Preimage Let f : B → B′ be a morphism, f˜:B˜→B˜′ an orientation of f (Definition 4.46), and ω¯∈ΩpBℝ˜⊗N.
Definition 4.51
The preimagef∗ω¯is the odd differential form inΩpBℝ˜⊗f∗Nuniquely determined by the following property ([BOU 82a], 10.4.2): letω˜(respectivelyf∗ω¯˜) be the differential p-form onB˜′(respectivelyB˜) associated withω¯(respectivelyf∗ω¯) by the bijection•˜from[4.31]; then the following relation holds:
f˜∗ω˜=f∗ω¯˜.
(III) Exterior and interior products Given three fiber bundles N1, N2, N with the same base B, a coupling Φ from N1 ×BN2 into N (Definition 4.31), and two odd differential forms ω¯1∈ΩpBℝ˜⊗N1, ω¯2∈ΩqBℝ˜⊗N2, we can define the exterior product ω¯1∧Φω¯2∈Ωp+qBℝ˜⊗N in the same way as in Definition 4.32. The coupling Φ uniquely determines a coupling Φ˜ from N˜1×BN˜2 into N˜, and (exercise)
ω¯1∧Φω¯2˜=ω˜1∧Φ˜ω˜2.
If ω1 and ω2 are differential forms of same parity (respectively of opposite parity) (Remark 4.50(b)), then ω1 ∧Φω2 is an even (respectively odd) differential form.
The formulas satisfied by the exterior product and the interior product (section 4.4.3(II),(IV)) also hold for odd differential forms.
(IV) Change of variable In practical settings, an odd differential p-form α¯ can be expressed locally in the form [4.20]. Even differential p-forms satisfy the change-of-variable formula [4.23], whereas the corresponding formula satisfied by odd differential p-forms is as follows:
(V) Measure defined by an oddm-form Let B be an m-dimensional Hausdorff pure manifold, ω¯∈ΩmBℝ˜⊗F, and Om the canonical orientation of ℝm (Example 4.38(i)). Let c =(U, φ, m) be a chart of B.
Lemma 4.52
There exists a unique mappingfc : φ (U) → Fso that
ω¯U=φ∗Om⊗fc.dx1∧…∧dxm.
We say that ω¯ is locally integrable if fc is locally λU⊗ m-integrable. If so, there exists a unique Radon measure αω¯ taking values in F that satisfies the following property: for every chart c = (U, φ, m) of B, φαω¯=fc.λU⊗m.
Definition 4.53
The Radon measureαω¯is said to be defined by the odd differential m-formω¯and is also written asω¯.
Remark 4.54
Suppose thatF=ℝ; letB↷be an oriented m-dimensional manifold with orientation O. Let ω ∈ Ωm (B), ω¯=O⊗ω(see[4.32]). Unlike the Radon measure [ω] fromDefinition 4.44, the Radon measureω¯fromDefinition 4.53is not necessarily positive. We have the relationω=ω¯([P2], section 4.1.5(VI)). In a certain sense, the notion of an odd differential form transfers the orientation of the manifold over to the form (seeExample 4.55).
Example 4.55
Let Β be the cube 0 < ξi < 1 inℝ3; this is an open subset ofℝ3and hence a submanifold (section 2.3.3). WriteB↷for the manifold B equipped with the orientation induced by the canonical orientation ofℝ3(Example 4.38(vi)).
Since the differential forms ω andω¯fromExample 4.55have the same expression (dξ1 ∧ dξ2 ∧ dξ3), they are often written in the same way (even though the first belongs to Ω3 (B) and the second belongs toΩ3Bℝ˜; strictly speaking, they are distinct objects).
4.4.7 Integration of a differential form over a chain
(I) Integration over an odd simplex Recall that the standard m-simplex Δm in ℝm is defined by
Δm=∑i=0mtivi:ti≥0∑0≤i≤mti=1,
where v0 = 0and {v1, …, vm} is the canonical basis of ℝm ([P1], section 3.3.8(V)). Let U be an open neighborhood of Δ in ℝm, F a Banach space, and ω ∈ Ωm (U; F) an even m-form (Remark 4.32(b)). Since the space ℝm is equipped with an orientation O, we can calculate ∫U↷χΔm.ω, where χΔm is the characteristic function of Δm and U↷ is equipped with the orientation induced by O.
Definition 4.57
Let B be a pure, metrizable, m-dimensional manifold. An odd m-simplex in B is a tripleτ¯=ΔmσO, where Δmis the standard m-simplex inℝm, U is an open neighborhood of Δm, σ is a mapping of class C∞from U into B and O is an orientation ofℝm.
In algebraic topology, B is a topological space, σ is only assumed to be continuous, O is the canonical orientation of ℝm and τ¯ can be identified with σ. In differential geometry, where σ is of class C∞, we sometimes specify that τ¯ is a smooth simplex.
Let α ∈ Ωm (B; F) be an even m-form. The preimage σ⁎ (α) is an even m-form in Ωm (U; F). Set:
∫τ¯α≔∫U↷χΔm.σ∗α.
(II) Integration over an even simplex Let Δm, U, F, σ, B be as defined above.
Definition 4.58
An even m-simplex in B is a pairτ=Δmσ˜, whereσ˜≔σεis an orientation of a mapping σ of class C∞from U into B (Definition 4.46).
Consider an odd differential m-form α¯∈ΩmBℝ˜⊗F. Its preimage σ∗α¯ is another odd differential m-form (Definition 4.51). Set:
∫τα¯=∫UχΔm.σ∗α¯.
[4.36]
Remark 4.59
If α∈ΩpBF(respectivelyα¯∈ΩpBℝ˜⊗F), where p < m, then ∫τ↷α=0(respectively∫τα¯=0), since we are integrating over a set of measure zero. Ifα∈ΩBFis an even non-homogeneous differential form Σ0 ≤ p ≤ mαp, then ∫τα = ∫ταm, and an analogous result holds when integrating an odd non-homogeneous differential form over an even simplex.
(III) Integration over a chain There are two cases to consider:
i)integration of an odd form over an even chain;
ii)integration of an even form over an odd chain.
The remarks up to and including part (IV) discuss the first case (i). The second case (ii) is similar and is left to the reader.
Let {τi : i ∈ I} be a set of even m-simplexes in the metrizable, pure, m-dimensional manifold B. The free group with this set as a basis is the set Sm (B) of linear combinations with integer coefficients
τ=∑i∈Iki.τi,
[4.37]
where all but finitely many of the ki are zero ([P1], section 3.3.8(VI)). It is useful to consider the real vector space SmB⊗ℤℝ formed by allowing these sums to have real coefficients; any such sum is called a chain of simplexes (ibid.).
Example 4.60
A closed, convex polyhedron10inℝmis a finite intersection of closed half-spaces. Any closed polyhedron is a finite union of convex closed polyhedra ([BOU 82a], 11.3.1). A closed polyhedron is a chain of simplexes (see [P1], section 3.3.8(VI) for a demonstration of how to express a square as the sum of two simplexes), so a chain of closed polyhedra is a chain of simplexes.
In the following, every chain is a chain of simplexes. If α¯∈ΩmBℝ˜⊗F, set
∫τα¯≔∑i∈Iki.∫τiα¯.
[4.38]
This definition still makes sense in the obvious way if the chain τ = ∑i ∈ Ikiτi. is allowed to be infinite (i.e. the sum is allowed to include infinitely many non-zero terms) and the set supp α¯∩τ is compact.
(IV) Change of variable Let B′ be another metrizable pure m-dimensional manifold, suppose that f : B → B′ is a morphism, and let f˜=fε′ be an orientation of f .If τ=Δmπ˜ is an even simplex in B, then fτ≔Δmf˜∘π˜ is an even simplex in B′. From [4.38], we can deduce the definition of f (τ) when τ is an even chain in B. By [4.36],
∫τα¯=∫fτf∗α¯.
(V) Boundary of a chain The i-th face of the standard m-simplex Δm in ℝm is ϵim : Δm − 1 ↦ Δm, where ([P1], section 3.3.8(VI))
ϵim:t0…tm−1↦0t0…tm−1ifi=0t0…ti^…tm−1if0≤i≤m
Given an m-simplex τ=Δmσ˜, its boundary ∂ τ is defined as follows for m ≥ 1 :
∂τ≔∑i=0m−1iσ∘ϵim.
If τ∈SmB⊗ℤℝ is the chain defined by [4.37], where the τi are m-simplexes, write:
∂τ=∑i∈Iki∂τi.
[4.39]
The boundary ∂(τ) is a chain with the same parity as the chain τ. If these chains are odd (which will be assumed henceforth), the orientation of ∂ (τ) above is said to be induced by the orientation of τ. Consider, for example, the triangle Δ2 in the plane, equipped with its canonical orientation, and write v0, v1, v2 for its vertices; then ([P1], section 3.3.8(V)) ∂Δ2=v0v1v2^+v0^v1v2+−v0v1^v2 : see Figure 4.2
.
Remark 4.61
Consider one of the segments [vi, vj]. The interior ]vi, vj[ of this segment is a submanifold of the plane (Oxy). Suppose that this submanifold is oriented as shown inFigure 4.2, writing ξ for this orientation and η for some point of ]vi, vj[. The canonical orientation O of the plane (Oxy) is positive in the direction (Oz) and contains each of the elements v ∧ u, where v is a vector pointing strictly outward at η for Δ2(with the obvious meaning in this context) and u belongs to ξ. The orientation ξ is induced by O.
(VI) Orientation of a boundary Our next task is to specify the notion of the boundary of an odd chain, as well as the orientation of this boundary. This is straightforward if we proceed as described in Remark 4.61. Let τ be an odd chain in an m-dimensional manifold B (m ≥ 2) and write Fr (τ) for its frontier. The latter is the union of a submanifold ∂ τ of dimension m – 1, called the regular boundary (or simply the boundary) of τ (in the example from Remark 4.61, this is the union of the open segments ]vi, vj[), and a set of points contained in a submanifold of dimension m – 2 (the endpoints of the segments, in the example).
Let b ∈ ∂ τ and suppose that ξ is an orientation of Tb (∂ τ). Write ι˜bξ for the orientation of Tb (B) containing each of the vectors v ∧ u, where v is a vector pointing strictly outward from τ at b and u is a non-zero element of ∧m−1Tb∂τ that belongs to the orientation ξ. The mapping ι˜b is a bijection from Or (Tx (∂ τ) onto Or (B) and the ι˜bb∈∂τ determine a morphism ι˜:∂τ˜→B˜ that is an orientation of the canonical injection i : ∂ τ ↪ B (section 4.4.5(IV)).
Definition 4.63
If O is an orientation of B, the orientation of ∂ τ associated with O byι˜(Corollary-Definition 4.47) is said to be induced by O.
4.5 Pseudo-Riemannian manifolds
4.5.1 Metric
Let B be a Banach manifold of class Cr (r ≥ ∞) and g :(X, Y) ↦ g (X, Y) = 〈X | Y〉 a twice covariant Hermitian tensor field of class Cr. For every b ∈ B,
gb∈T20TbB=ℒ2(TbB,TbBK.
Consider the condition (M) and the weaker condition (WM) stated below:
is non-degenerate (i.e. if g (Yb, Xb)b = 0 for all Xb ∈ Tb (B), then Yb = 0).
If B is locally finite-dimensional (which is typically the case in practice), the conditions (M) and (WM) are equivalent.
Definition 4.64
Let B be a manifold equipped with a twice covariant Hermitian tensor fieldgof class Cr. If g satisfies (M) (respectively (WM)), the manifold B is said to be strongly (respectively weakly) pseudo-Riemannian.11The fieldgis called the metric or the fundamental tensor field of the manifold. This manifold is said to be strongly (respectively weakly) Riemannian if the Hermitian formg (b) is also positive definite for every b ∈ B.
Suppose now that K=ℝ and let B be a pure m-dimensional manifold.
Definition 4.65
A frame (hα)1 ≤ α ≤ mof T (B) (Definition 3.23) is said to be orthonormal for the metricgif
hαbhβbb=ηbαβ.δαβ
[4.40]
for every b ∈ B, where ηb, (α, β) ∈ {1, –1}. We say thatg (b) has signature (pb, m – pb) if, for every β ∈ {1,…, m}, Card {α ∈ {1, …, m} : sgn (ηb(α, β)) = 1} = pb.
If B is connected, the signature of g is constant. Let (U, ξ, m) be a chart; we can express g locally (over U) by
g=∑i,j=1mgij.dξi⊗dξj,
[4.41]
where the square matrix G (b) = (gij (b)) is real, symmetric, and nowhere singular. Any such chart is said to be a system of normal pseudo-Riemannian coordinates at the point b ∈ B if the system of tangent vectors ∂∂ξ1b…∂∂ξmb is orthonormal for the real symmetric form 〈.|.〉b.
The real manifold B is Riemannian if the symmetric form g (b) has signature (m, 0) at every point; if m ≥ 2 and the signature is (1, m – 1) or (m – 1, 1) at every point, then the manifold B is said to be Lorentzian (the Lorentzian manifold of dimension 4 is encountered in general relativity; we will call it the Einstein manifold).
4.5.2 Pseudo-Riemannian volume element
Let {σα :1 ≤ α ≤ m} be the dual coframe of the orthonormal frame (hα)1 ≤ α ≤ m (Lemma-Definition 3.38). The pseudo-Riemannian volume element for which the frame (hα)1 ≤ α ≤ m is positively oriented is (Definition 4.37 and section 4.4.5(III)):
ω=σ1∧…∧σm.
Let (hβ′) be another, arbitrary frame. Let (σβ′) be its dual, and suppose that A = (Aα′α) is the inverse change-of-basis matrix satisfying hα′ = ∑αhα′Aα′α (see [1.2], section 1.2.1(II)). Then, by [4.40], using the fact that g is bilinear,
gα′β′=∑α,βAα′αAβ′βghαhβ′=∑αAα′αAβ′αη,
where gα′β′ = g(hα, hβ′). Writing (gα′β′) for the matrix with entries gα′β′,we therefore have det(gα′β′) = det (A)2η, which implies that
detA2=detgα′β′⇒detA=εdetgα′β′,
where ε =+1if the frame (hβ′) is positively oriented and ε = – 1 otherwise. Moreover, σα = ∑α′Aα′ασα′, so