11-05 Notes

From Moduli
Jump to: navigation, search

Uhlenbeck Gauge Fixing Lemma

Let \(A(t)\) be a solution to \(\frac{\partial A}{\partial t}=D^*_AF_A\) and thus \(\mathcal{E}(A(t_1))-\mathcal{E}(A(t_0))= \int_{t_1}^{t_0}\left\|{D^*_AF_A}\right\|^2_{L^2}dt\).

Theorem 1. (Uhlenbeck Foundamental Lemma) Let \(P=B^n\times G\longrightarrow B\) and \(n\leq 4\). There exist \(C,\ \epsilon>0\) so that: if \(A\in\mathcal{A}_{L^2_1}\) and \(\int_B|F_A|^2<\epsilon\), then there exists \(u\in\mathcal{G}_{L^2_1}\) such that if \(u\cdot A=\Gamma+a\), where \(\Gamma\) is the trivial connection, then (a) \(\int_B|\nabla_{\Gamma}a|^2+|a|^2\leq C\int_B|F_A|^2\); (b) \(d^*_{\Gamma}a=0 and *a|_{\partial B^n}=0\).

Remark 1. (b) is the "slice" condition. And (b) is equivalent to \(\int<d_{\Gamma}\xi,a>=0\), \(\forall \xi\in\Omega^0(B,adP)\).

Since \(d-tr(\xi*a)=-tr(d_A\xi\wedge *a)=tr(\xi\wedge d_A*a)\), thus \(\int_{\partial B}-tr(\xi*a)=-\int_Btr(d_A\xi\wedge *a)=\int_Btr(\xi\wedge *(*d_A*)a)\). If \(A\longrightarrow\Gamma+a\), \(d_A^*a=0\), \(*a|_{\partial B}=0\) and \(\left\|{a}\right\|_{L^2_1}\) is small, then the estimate (a) follows. Note that \(\left\|{F_A}\right\|_{L^2(B)}=\left\|{F_{\Gamma+a}}\right\|_{L^2(B)}\), \(F_{\Gamma+a}=da+a\wedge a\), thus \(\int_B|da+a\wedge a|^2\geq\int_B|da|^2-C\int_B|a|^4\). Since \(n\leq 4\), thus \(L^2_1\hookrightarrow L^4\) and \(\int_B|a|^4\leq C_1\left\|{a}\right\|_{L^2_1}^4\).

If \(\int_B|da|^2\geq C_2\left\|{a}\right\|_{L^2_1}\), denoted by (*), then \[\begin{aligned} \int_B|F_A|^2&\geq C_2\left\|{a}\right\|_{L^2_1}^2-CC_1\left\|{a}\right\|^4_{L^2_1}\\ &\geq (C_2-CC_1\left\|{a}\right\|^2_{L^2_1})\left\|{a}\right\|_{L^2_1}^2. \end{aligned}\] Now the question is that when does (*) hold.

Theorem 2. Suppose \(M\) is a manifold with boundary and \(H^1(M)\) is represented by harmonic form, i.e. \(\{\theta|(d+d^*)\theta=0 and *\theta|_{\partial M}=0\}\). Find \(\theta\) by minimizing \(\left\|{\theta+df}\right\|_{L^2}\) with \(d\theta=0\). If \(H^1(M)=0\), then the map \(d+d^*:\ \{\alpha|*\alpha|_{\partial M}=0\}\longrightarrow\Omega^0_{L^2}+\Omega^2_{L^2}\) is injective.

Since \((d+d^*)^2=\nabla^*\nabla+\)Riemannian curvature, which is zero in flat ball, thus \(\int|(d+d^*)a|^2=\int|\nabla a|^2\), which holds on the \(B^n\) with \(*a|_{\partial B}=0\).

There are two problems:(1)There might be a small energy connection without gauge representative near trivial connection; (2)Assume the connection was gauge-equivalent into the slice.

We showed that if we fix \(\Gamma\in\mathcal{A}_{L^2_k}\), then there exists an \(\epsilon\) such that \[\tilde{S}_{\Gamma,\epsilon}=\{\Gamma+a|d_A^*a=0, *a|_{\partial M}=0, \left\|{a}\right\|_{L^n}<\epsilon\}\] is injective into the quotient space \(\mathcal{A/G}\), that is \(\tilde{S}_{\Gamma,\epsilon}\times \mathcal{G}_{stab(\Gamma)}\longrightarrow\mathcal{A}\) is an injection.

Hope: if \(A\) is an \(L^n-\)connection, then there exists \(u\int\mathcal{G}_{L^n_1}\) so that \(u\cdot A=\Gamma+a\) and \(a\in\tilde{S}_{\Gamma,\epsilon}\).

Lemma 1. There exists a \(\mathcal{G}-\)invariant neighborhood \(U\) of \(\Gamma\) in the \(L^n\) topology so that every connection \(A\in U\) is gauge-equivalent into the slice \(\tilde{S}_{\Gamma,\epsilon}\).

Note that rescaling decreases the Yang-Mills energy. We want to prove for \(L^n\) ball of radius \(\delta\), but we cannot use implicit function theorem since we are on the border line. Instead, we use open/closed argument. We want to sharpen result to show that \[m: (\tilde{S}_{\Gamma,\epsilon}\times \mathcal{G}_{L^2_{k+1}})/stab(\Gamma)\longrightarrow \mathcal{A}_{L^2_k}\] is a diffeomorphism onto its image. We need to show that the differential is an isomorphism. This map is equivariant, so it suffices to check this at \((\Gamma+a,1)\). note that \(\mathcal{D}_{(\Gamma+a,1)}m(\alpha,\xi) =d_{\Gamma}\xi+[a,\xi]+\alpha\).

For simplicity, given that \(\ker(d_{\Gamma})=0\), prove the injectivity. Suppose \(\mathcal{D}_{(\Gamma+a,1)}m(\alpha,\xi)=0\). Take inner product with \(d_A\xi\),thus

\[\begin{aligned} \int|d_A\xi|^2+\int<[a,\xi],d_A\xi> +\int<\alpha,d_A\xi>&= 0.\\ \end{aligned}\] Since \(\alpha\) is in slice and \(\frac{1}{n}+\frac{n-2}{2n}+\frac{1}{2}=1\), thus \[\begin{aligned} \int|d_{\Gamma}\xi|^2&\leq \left\|{a}\right\|_{L^n} \left\|{\xi}\right\|_{L^{\frac{2n}{n-2}}} \left\|{d_{\Gamma}\xi}\right\|_{L^2}. \end{aligned}\] Since \(\ker(d_\Gamma)=0\) and \(\left\|{\xi}\right\|_{L^2_1}\leq C\left\|{d_\Gamma\xi}\right\|_{L^2}\), denoted by (**), thus \[\begin{aligned} \int|d_\Gamma\xi|^2&\leq C\left\|{a}\right\|_{L^n}\left\|{d_\Gamma\xi}\right\|^2_{L^2}. \end{aligned}\] Hence \(\left\|{d_\Gamma}\right\|_{L^2}=0\) if \(\left\|{a}\right\|_{L^n}\) is small in the case \(n\neq 0\).

If \(\ker(d_\Gamma)\neq 0\), then we can assume that \((\alpha,\xi)\) is a representative of the tangent space of the quotient \((\tilde{S}_{\Gamma,\epsilon}\times\mathcal{G}) /stab(\Gamma)\) at \((\Gamma+a,1)\), that is, \((\alpha,\xi)\) is orthogonal to \(([\alpha,\xi],\xi_0)\), \(\xi_0\in\ker(d_\Gamma)\), and thus \(\xi\perp\ker(d_\Gamma)\). For such \(\xi\), the estimate (**) holds and the argument above works. Therefore \(\xi=0\) implies that \(\alpha=0\) and \(\mathcal{D}_{(\Gamma+a,1)}m\) is injective. \(\mathcal{D}_{(\Gamma+a,1)}m\) is a continuously varying family of Fredholm opertors. Since \(\mathcal{D}_{(\Gamma,1)}m\) is invertible, thus \(index=0\), and the injectivity implies the isomorphism.

Retrieved from "/math/moduli/index.php?title=11-05_Notes&oldid=255"
Views
Navigation