Step 0 (symmetry of mixed partials). Let WβRm be open, g:WβR a smooth map, and i,j indices. For aβW and small h,k>0 consider
Ξ(h,k)=g(a+heiβ+kejβ)βg(a+heiβ)βg(a+kejβ)+g(a),
where eiβ,ejβ are standard basis vectors. Applying the mean value theorem first in the eiβ-variable and then in the ejβ-variable gives Ξ(h,k)=hkβjββiβg(ΞΎ) for some ΞΎ in the rectangle spanned by the four points; applying it in the other order gives Ξ(h,k)=hkβiββjβg(Ξ·) similarly. Letting (h,k)β(0,0) and using continuity of the second partial derivatives yields βjββiβg=βiββjβg on W.
Step 1 (pullback of a decomposable form). A straightforward induction from Wedge Product of Differential Forms on Euclidean Space and Associativity of the Wedge Product of Differential Forms on Euclidean Space, using Sign of a Permutation, shows that for 1-forms Ξ±1β,β¦,Ξ±kβ and vectors v1β,β¦,vkβ,
(Ξ±1ββ§β―β§Ξ±kβ)xβ(v1β,β¦,vkβ)=det[(Ξ±rβ)xβ(vsβ)]r,sβ,
with the determinant of the kΓk matrix indicated. Write F=(F1β,β¦,Fmβ) in components; each Fiβ is a smooth 0-form on U and its exterior derivative is dFiβ=βl=1nββxlββFiββdxlβ. For a C1 function a:VβR and increasing indices i1β<β―<ikβ, I claim
Fβ(adyi1βββ§β―β§dyikββ)=(aβF)dFi1βββ§β―β§dFikββ.
Indeed, by Pullback of a Differential Form by a C^1 Map the left side at x on (v1β,β¦,vkβ) equals a(F(x)) times the value of dyi1βββ§β―β§dyikββ on the matrix-vector products JFβ(x)v1β,β¦,JFβ(x)vkβ; by the determinant identity this is a(F(x))det[(JFβ(x)vsβ)irββ], and (JFβ(x)vsβ)irββ=βlββxlββFirβββ(x)(vsβ)lβ=(dFirββ)xβ(vsβ), which is the right side by the same determinant identity.
By Coordinate Expansion of Differential Forms on Euclidean Open Sets, Ο=βIβaIβdyi1βββ§β―β§dyikββ over increasing multi-indices I=(i1β,β¦,ikβ), and the pullback is additive in Ο directly from Pullback of a Differential Form by a C^1 Map; hence
FβΟ=Iββ(aIββF)dFi1βββ§β―β§dFikββ.
Evaluating on basis vectors ej1ββ,β¦,ejkββ (increasing J) and using the uniqueness in Coordinate Expansion of Differential Forms on Euclidean Open Sets, the coefficient of FβΟ on dxj1βββ§β―β§dxjkββ is
cJβ=Iββ(aIββF)MI,Jβ,MI,Jβ=det[βxjsβββFirβββ]r,sβ.
Each aIββF is C1 by the chain rule (using C^1 Maps on Euclidean Open Sets are Differentiable), each MI,Jβ is obtained from the smooth partial derivatives of F by finitely many sums and products via Determinant of a Real Square Matrix, and products and sums of C1 functions are C1 by Products and Quotients of C^k Real-Valued Maps on Euclidean Open Sets Are C^k; hence FβΟ is a C1 differential k-form.
Step 2 (the case k=0). For a C1 function b:VβR, Fβ(db)=d(bβF): at x on a vector v, the left side is βjββyjββbβ(F(x))(JFβ(x)v)jβ=βj,lββyjββbβ(F(x))βxlββFjββ(x)vlβ, which equals d(bβF)xβ(v) by the chain rule.
Step 3 (conclusion). Compute d(FβΟ) from Exterior Derivative of a C^1 Differential Form on a Euclidean Open Set:
d(FβΟ)=Jββl=1βnββxlββcJββdxlββ§dxj1βββ§β―β§dxjkββ.
Expand βxlββcJββ by the product rule for partial derivatives (the one-dimensional product rule applied to each coordinate slice): the terms in which the derivative falls on aIββF combine, by Step 2 and the two determinant identities of Step 1, into exactly the coordinate expansion of
Fβ(dΟ)=Iββj=1βmβ(βyjββaIβββF)dFjββ§dFi1βββ§β―β§dFikββ,
which is the pullback of dΟ by Step 1 applied to the expansion of dΟ. The remaining terms carry a derivative of an entry of MI,Jβ, i.e. a second partial βxlββxjsβββ2Firβββ; summed against dxlββ§dxj1βββ§β―β§dxjkββ these assemble into expressions of the form d(dFiβ)β§(remainingΒ factors), whose coefficients are antisymmetric in the pair of differentiation indices by Permutation Rule for Wedge Products of Coordinate 1-Forms while the second partials are symmetric in that pair by Step 0; hence each such sum vanishes. Therefore d(FβΟ)=Fβ(dΟ). β