Indexed text attachment
connections-curvature-higher-geometry.tex
Source for Connections, Curvature, and Higher Geometry: From covariant differentiation to differential cohomology
View paper · Download original file
Lines 1701–1800 of 4453. Source text is displayed literally.
1701 =-1 .1702\]1703The integral is computed in polar coordinates, since the omitted point has measure zero for a smooth top form. The bundle's topological first Chern class has the same sign: the dual line \(\mathcal O(1)\) has a section defined by a nonzero linear functional on \(\mathbb C^2\), with one positively oriented transverse zero. Its Euler number and hence its degree are \(1\). The duality of lines reverses the Chern class. One can also recover the sign by using the transition \(z^{-1}\) from the frame \((1,z)\) to the frame \((z^{-1},1)\).1704 1705The standard comparison with integral Chern classes reduces to this calculation~\cite[Chapter~IV]{bott-tu}. For a line bundle, naturality applied to the universal line over projective space reduces normalization to the projective-line inclusion, which detects its degree-two generator. For a bundle \(E\), pass to the full flag bundle \(\pi\colon\operatorname{Flag}(E)\to X\). The successive projectivizations have cohomology free over the cohomology of their bases by the projective bundle theorem. The unit is one of the module generators, so each pullback is injective. In particular \(\pi^*\) is injective in real cohomology. The pullback of \(E\) has a flag with line-bundle quotients; choosing a metric splits the filtration as a direct sum of these lines. On the flag bundle, both the real image of the topological total Chern class and the curvature-defined total class are the product of \(1+c_1\) of the line factors. Injectivity proves equality downstairs.1706 1707\section{Real bundles and Pontryagin classes}1708 1709For a real bundle \(V\), set1710\[1711 p_j(V)=(-1)^j c_{2j}(V\otimes_{\mathbb R}\mathbb C)1712 \in H^{4j}(X;\mathbb Z).1713\]1714A metric connection on \(V\) complexifies to a unitary connection, so the formula gives a Chern--Weil representative for the real image. In a real orthonormal frame \(F^{\mathsf T}=-F\), and transposing the matrix in \(\det(I+iF/(2\pi))\) shows that this polynomial is even in \(F\). Thus the odd-index Chern forms of the complexification vanish. The integral odd-index Chern classes may still be two-torsion. Indeed conjugation identifies the complexification with its conjugate, and \(c_k(\overline E)=(-1)^k c_k(E)\). It follows that the Pontryagin total class satisfies the product formula after inverting \(2\). Over the integers, products of two odd-index classes in the Whitney formula can contribute two-torsion~\cite[\S15]{milnor-stasheff}.1715 1716For an explicit torsion calculation, let \(\lambda\) be the tautological real line on \(\mathbb{RP}^4\) and let \(u=w_1(\lambda)\). The real bundle underlying \(\lambda_{\mathbb C}\) is \(\lambda\oplus\lambda\), whose second Stiefel--Whitney class is \(u^2\). Thus \(a=c_1(\lambda_{\mathbb C})\) reduces to \(u^2\), so \(a^2\) reduces to \(u^4\ne0\). For \(V=\lambda\oplus\lambda\) the integral \(p_1(V)=-a^2\) is nonzero, whereas \(p_1(\lambda)=0\) by rank. The transition functions can be chosen to take values in \(\{\pm1\}\), and the two line summands admit flat connections. Their real characteristic forms vanish. This example exhibits both the two-primary qualification on the Whitney identity and the failure of real forms to detect torsion.1717 1718\section{Orientation and the Pfaffian}1719 1720For an oriented real bundle of rank \(2m\), take an oriented orthonormal frame. Define the Euler form1721\[1722 e(\nabla)=\operatorname{Pf}\left(\frac{F}{2\pi}\right).1723\]1724We use the Pfaffian with \(\operatorname{Pf}\left(\begin{smallmatrix}0&a\\-a&0\end{smallmatrix}\right)=a\). One definition is the coefficient of the ordered top exterior basis vector in the \(m\)-th divided power of \(\sum_{i<j}F_{ij}e_i\wedge e_j/(2\pi)\). Conjugation by an orthogonal matrix multiplies this coefficient by its determinant. The polynomial is therefore invariant under \(SO(2m)\), and reversing orientation changes its sign. On a complex line with its complex orientation, the skew matrix for multiplication by \(i\) is \(\left(\begin{smallmatrix}0&-1\\1&0\end{smallmatrix}\right)\). For \(F=i\,d\alpha\) this gives \(e(\nabla)=-d\alpha/(2\pi)=c_1(\nabla)\). The Pfaffian product on oriented block sums then implies equality of the Euler form and the top Chern form for the realification of a unitary bundle. This can be checked on block diagonal matrices with real scalar entries, which determine the restricted invariant polynomial on \(\mathfrak u(m)\) by unitary diagonalization. Polarization extends the equality to matrices of two-forms.1725 1726The comparison of the Pfaffian class with the real image of the integral Euler class is the Euler Chern--Weil theorem. The latter class is obtained by pulling the Thom class back along the zero section~\cite[\S25]{tu}. In odd rank, negation of each fiber reverses orientation and fixes the zero section. It follows that the integral Euler class is killed by \(2\). Its real image is zero.1727 1728Take \(E=\mathcal O(1)\oplus\mathcal O(1)\) on \(\mathbb{CP}^2\) and let \(h=c_1(\mathcal O(1))\). The top Chern class is \(h^2\). Since \((\mathcal O(1))_{\mathbb R}\otimes\mathbb C\cong\mathcal O(1)\oplus\mathcal O(-1)\), the first Pontryagin class of \(E_{\mathbb R}\) is \(2h^2\). Two independent linear functionals give hyperplanes intersecting in one positively oriented point, so \(\int_{\mathbb{CP}^2}h^2=1\). The resulting Euler and Pontryagin numbers are \(1\) and \(2\), respectively, for the same oriented real bundle. They detect different invariant polynomials of its curvature.1729\chapter[Chern--Simons transgression]{Chern--Simons forms and coherent transgression}1730\label{ch:chern-simons}1731 1732A form that witnesses the equality of two de Rham classes contains additional information. It records how the two curvature representatives are related, and it changes when the comparison is changed. Chern and Simons used this principle to construct secondary invariants~\cite{chern-simons}. Here we make the comparison depend on a simplex of connections, so comparisons between comparisons have their own differential identities.1733 1734\section{An oriented path of connections}1735 1736Let \(P\) be an invariant symmetric \(m\)-linear form with \(m\geq1\), and let \(A_0,A_1\) be connections of a fixed principal bundle \(Q\to X\). Set \(B=A_1-A_0\) and \(A_t=A_0+tB\). Define the relative transgression by1737\[1738 T_1(A_0,A_1)=m\int_0^1P(B,F_t^{m-1})\,dt.1739\]1740Although written in a frame, the integrand is a global form because \(B\) and \(F_t\) transform by simultaneous conjugation. Its exterior derivative is \(P(F_1^m)-P(F_0^m)\) by \eqref{eq:hdg-cw-path}. A second proof differentiates the curvature formula:1741\[1742 \dot F_t= D_{A_t}B.1743\]1744The derivative of \(P(F_t^m)\) is \(mP(D_{A_t}B,F_t^{m-1})\). Invariance converts covariant differentiation to exterior differentiation inside \(P\). Terms differentiating \(F_t\) vanish by Bianchi, so this derivative equals \(m\,d_X P(B,F_t^{m-1})\).1745 1746For the quadratic polynomial \(p(X)=\operatorname{tr}(X^2)\) on a matrix Lie algebra, expand1747\[1748 F_t=F_0+tD_{A_0}B+t^2 B\wedge B.1749\]1750Integration gives the three-form1751\[1752 T_1(A_0,A_1)1753 =\operatorname{tr}\!\left(2B\wedge F_01754 +B\wedge D_{A_0}B+\tfrac23 B\wedge B\wedge B\right).1755\]1756On a trivial bundle with the reference connection \(A_0=0\), this reduces to the usual local expression \(\operatorname{tr}(A\wedge dA+\tfrac23 A^{\wedge 3})\). The definition of \(T_1\) also makes sense when neither connection has a global potential, since it uses a reference connection on the same bundle.1757 1758\section{The sign from integration over a simplex}1759 1760Orient the simplex with ordered vertices \(0,\ldots,q\) by coordinates \(t_1,\ldots,t_q\), with \(t_0=1-\sum_{i=1}^q t_i\). On \(\Delta^q\times X\) use the connection \(\widetilde A=\sum_{i=0}^q t_i A_i\) in the \(X\) directions. It has zero components in the simplex directions in local frames pulled back from \(X\). For a form \(\omega\) on the product, let \(I_q\omega\) mean the integral of the component with simplex degree \(q\), with all simplex differentials written first. Its degree on \(X\) is \(\deg(\omega)-q\). For a closed \(\omega\),1761\begin{equation}1762 d_X I_q\omega=(-1)^{q+1}I_{\partial\Delta^q}\omega.1763 \label{eq:hdg-fiber-stokes}1764\end{equation}1765Indeed, write \(\omega=\sum_p\omega_p\) by simplex degree. In degree \(q\), the closedness equation is the sum of the simplex derivative of \(\omega_{q-1}\) and \((-1)^q\) times the \(X\)-differential of the coefficients of \(\omega_q\). Stokes on the parameter simplex gives \eqref{eq:hdg-fiber-stokes}.1766 1767Apply this to the closed form \(P(\widetilde F^m)\) and define, for \(q\geq0\),1768\begin{equation}1769 T_q(A_0,\ldots,A_q)1770 =(-1)^{q(q-1)/2} I_q P(\widetilde F^m).1771 \label{eq:hdg-transgression-simplex}1772\end{equation}1773At \(q=0\) this is \(P(F_0^m)\). Each oriented face pulls the connection back to the affine family defined by the surviving vertices. Since \((-1)^{q(q-1)/2}(-1)^{q+1}=(-1)^{(q-1)(q-2)/2}\), the sign convention gives1774\begin{equation}1775 d_X T_q(A_0,\ldots,A_q)1776 =\sum_{i=0}^q(-1)^iT_{q-1}(A_0,\ldots,\widehat A_i,\ldots,A_q).1777 \label{eq:hdg-transgression-desc}1778\end{equation}1779The hat omits the indicated connection. The case \(q=2\) is the relation1780\[1781 T_1(A_0,A_1)+T_1(A_1,A_2)-T_1(A_0,A_2)1782 = d_XT_2(A_0,A_1,A_2).1783\]1784It says that making the comparison in two steps and taking the direct comparison differ by an exact form. For a polynomial of degree \(m\), the form \(T_q\) vanishes for \(q>m\), since each curvature factor carries at most one simplex differential. The relations still hold at those levels because the boundary sum in \eqref{eq:hdg-transgression-desc} then vanishes.1785 1786\section{A two-parameter calculation}1787 1788For \(q=2\), put \(B_i=A_i-A_0\) for \(i=1,2\). The curvature of the affine family consists of \(F_{\boldsymbol t}\) and the terms \(dt_i\wedge B_i\). Each product of the two mixed curvature terms contributes a minus sign when \(dt_2\) is moved past \(B_1\). The prefactor in \eqref{eq:hdg-transgression-simplex} is also negative, so for \(m\geq2\),1789\[1790 T_2(A_0,A_1,A_2)=m(m-1)\int_{\Delta^2}1791 P(B_1,B_2,F_{\boldsymbol t}^{m-2})\,dt_1\,dt_2 .1792\]1793If \(m=2\) the integral is over a coefficient independent of the parameter, and \(\operatorname{area}(\Delta^2)=1/2\) in these coordinates. Thus \(T_2=P(B_1,B_2)\). With \(p=c_1^2\) and \(A_i=-2\pi i\,a_i\) for a unitary line, write \(b_i=a_i-a_0\), which is a globally defined real one-form. The identity becomes \(T_2=b_1\wedge b_2\). Expanding the closed curvature expressions directly gives1794\[1795 T_1(A_0,A_1)+T_1(A_1,A_2)-T_1(A_0,A_2)1796 =db_1\wedge b_2-b_1\wedge db_2 .1797\]1798The two sides agree with \(d(b_1\wedge b_2)\), confirming the orientation choice in a case where the two-form can be nonzero.1799 1800\section{Gauge transformations and periods}