Independent scholarship · Open accessThe monthly digest

Indexed text attachment

representation-theory-in-higher-categories.tex

Source for Representation Theory in Higher Categories

View paper · Download original file

Lines 2661–2760 of 3021. Source text is displayed literally.

2661  \sum_{\pi\in\widehat G}(\dim M_\pi)(\dim N_\pi).2662\]2663For scalar \(L^2(S^2)\), \(M_\ell\) is one-dimensional for every integer \(\ell\geq0\), because the \(H=\mathrm{SO}(2)\)-fixed subspace of each \(V_\ell\) has dimension one. With \(c\) input channels and \(d\) output channels each spectral block is an arbitrary \(d\)-by-\(c\) matrix, applied in the same way to all \(2\ell+1\) coordinates of \(V_\ell\).2664 2665\section{Regular convolution and its scope}2666On the left regular representation \(L^2(G)\), a right convolution operator \(T_k f=f*k\) for \(k\in L^1(G)\) commutes with left translation by change of variables. With Haar probability, use2667\[2668 (f*k)(x)=\int_Gf(y)k(y^{-1}x)\,dy,\qquad2669 \widehat f(\pi)=\int_G f(g)\pi(g)^*\,dg.2670\]2671The Fourier multiplier is \(\widehat{f*k}(\pi)=\widehat k(\pi)\widehat f(\pi)\): substitute \(x=yz\) in the double integral and multiply \(\pi(z)^*\pi(y)^*\) in that order. Left translation by \(a\) acts on \(\widehat f(\pi)\) by right multiplication by \(\pi(a)^*\), so the position of \(\widehat k(\pi)\) agrees with \eqref{eq:learning-blocks}. Peter--Weyl orthogonality gives an actual continuous kernel for every finite-support family of matrix multipliers.2672 2673For \(k\in L^1(G)\) the convolution norm is bounded by \(\|k\|_1\). Indeed, rewriting the convolution as \(\int_G k(z)f(xz^{-1})\,dz\) expresses it as an integral of right translations, all \(L^2\)-isometries. Approximation of \(k\) in \(L^1\) by finite Peter--Weyl sums shows that every such \(T_k\) is compact on \(L^2(G)\). For an infinite compact group the identity is bounded and equivariant and has infinite-dimensional range, so it cannot have an \(L^1\) convolution kernel. It is convolution by the Dirac measure at the identity. Formula \eqref{eq:learning-blocks} applies to all bounded intertwiners independently of such kernel realizations.2674 2675\section{Pointwise layers and feature types}2676Let \(X=G/H\), and let \(V,W\) be finite-dimensional \(H\)-representations with actions \(\rho_V,\rho_W\). A pointwise activation sends the fiber coordinate \(v\) of \([g,v]\) to \([g,\sigma(v)]\). The relation \([gh,v]=[g,\rho_V(h)v]\) makes this rule independent of the representative exactly when2677\[2678   \sigma(\rho_V(h)v)=\rho_W(h)\sigma(v)\qquad(h\in H).2679\]2680The same formula then implies \(G\)-equivariance under left translation of \(g\). An arbitrary scalar activation works for scalar fields, since the stabilizer acts trivially on the fiber. A coordinatewise activation on a vector fiber need not satisfy the relation. For a unitary fiber a map \(v\mapsto a(\|v\|)v\) satisfies it. If the activation is Lipschitz, it induces a Lipschitz map on \(L^2\)-sections over the finite invariant base measure: the pointwise estimate integrates, and \(\|\sigma(v)\|\leq\|\sigma(0)\|+L\|v\|\) ensures square integrability. A nonlinear layer need not preserve the spectral truncation of its input.2681 2682\section{A rotation-invariant quadratic feature}2683Give \(S^2\) its rotation-invariant probability measure. For real scalar inputs define2684\[2685  (Tf)(x)=5\int_{S^2}P_2(x\cdot y)f(y)\,d\mu(y),2686  \qquad P_2(t)=\tfrac12(3t^2-1).2687\]2688The kernel uses only the dot product, so it is unchanged by simultaneous rotation of \(x\) and \(y\). For fixed \(y\), it is a traceless quadratic in \(x\) on the sphere. To determine its action, take \(f_A(y)=y^{\mathsf T}Ay\) for a real symmetric traceless matrix \(A\). The moments are2689\[2690 \int y_i y_j\,d\mu=\tfrac13\delta_{ij},\qquad2691 \int y_i y_j y_k y_l\,d\mu2692  =\tfrac1{15}(\delta_{ij}\delta_{kl}2693                 +\delta_{ik}\delta_{jl}+\delta_{il}\delta_{jk}).2694\]2695For the fourth moment, the polynomial in \(v\) obtained by integrating \((v\cdot y)^4\) is rotation invariant and homogeneous. Its value at a unit vector is \(1/5\), from the sphere's area measure in the height coordinate \(t\in[-1,1]\). Comparing coefficients, or polarizing \(\|v\|^4/5\), gives the tensor written above. The second moment is obtained from \(\|y\|^2=1\) and rotation symmetry. Consequently2696\[2697  \int f_A\,d\mu=0,\qquad2698  \int (x\cdot y)^2 f_A(y)\,d\mu(y)=\tfrac{2}{15}x^{\mathsf T}Ax.2699\]2700Thus \(Tf_A=f_A\). The kernel is symmetric and the range lies in the traceless quadratics, so \(T\) is their orthogonal projection.2701 2702Apply a scalar square after the projection and then pool by integration. The projection is finite rank with continuous outputs, so the square is defined and integrable for every \(L^2\) input. It is an equivariant pointwise layer followed by an invariant pooling layer.2703\[2704\begin{tikzcd}[column sep=large]2705 L^2(S^2;\mathbf R) \arrow[r,"T"]2706 & V_2^{\mathbf R}\arrow[r,"u\mapsto u^2"]2707 & C(S^2;\mathbf R)\arrow[r,"\int_{S^2}(-)\,d\mu"]2708 & \mathbf R2709\end{tikzcd}2710\]2711Its value on \(f_A\) is \(2\operatorname{tr}(A^2)/15\), from the fourth moment. For example, \(A=\operatorname{diag}(1,-1,0)\) gives \(4/15\), and a constant input gives zero. A channel version first projects each channel and then applies any matrix of linear coefficients to those projected channels, as allowed by \eqref{eq:learning-blocks}; each squared pooled output is invariant for every choice of those coefficients.2712 2713\section{Kernels for associated bundles}2714Take the spaces of sections of \(E=G\times_H V\) and \(F=G\times_H W\) with invariant measures and Hermitian metrics. Suppose an intertwiner \(T\) has finite spectral support. For each retained \(G\)-type there are only finitely many copies by compact Frobenius reciprocity, since \(V\) and \(W\) are finite dimensional. The retained subspaces consist of continuous sections, given by matrix coefficients as in Chapter~\ref{ch:synthesis}. For an orthonormal basis \(s_j\) of the retained input, define \(K(x,y):E_y\to F_x\) by2715\[2716   K(x,y)v=\sum_j(Ts_j)(x)\langle s_j(y),v\rangle.2717\]2718Use the Hermitian product linear in the second argument. The orthonormal expansion gives \((Ts)(x)=\int K(x,y)s(y)\,d\mu(y)\) and makes the formula independent of the chosen basis. Replacing \(s_j\) by the translated basis and using \(T U_a=U_a T\) shows \(K(ax,ay)=a K(x,y)a^{-1}\), with the two actions interpreted on the appropriate fibers. Conversely, a continuous kernel satisfying this covariance defines a bounded equivariant operator by change of variables and the uniform bound on the kernel over the compact base. For each output point one can transport the kernel at \(o\) to that point, and the result is independent of the transporting element precisely because the kernel at \(o\) intertwines the \(H\)-actions on the fibers over \(y\) and \(hy\). This describes how global rotation equivariance and local fiber coordinates constrain the same layer.2719\par2720\chapter{An orbit, its stabilizer, and its invariants}2721\label{ch:synthesis}2722The sphere carries an ordinary topology, an action of \(\mathrm{SO}(3)\), and several complexes of sections. Each operation applied to this data has a different output. Taking the quotient as a set gives one point. Taking the quotient as a groupoid retains the rotations about an axis. Passing to a Hilbert space of sections takes a representation of those rotations and makes a representation of the whole group. The computation below compares these operations for the same line bundles.2723 2724\section{A transitive action as a groupoid}2725Let \(G\) be a compact Lie group, \(H\) a closed subgroup, and \(X=G/H\). The action groupoid \(G\ltimes X\) has objects \(x\in X\) and arrows \((g,x):x\to gx\). The map from the one-object groupoid of \(H\) sends its object to \(o=1H\) and its arrows to the arrows \(o\to o\). This map is fully faithful because \(\{g:go=o\}=H\). It is smoothly essentially surjective because \(g\mapsto go\) is a surjective submersion with local sections. At the level of equivariant vector bundles, the resulting comparison is the fiber construction of Chapter~\ref{ch:equivariant-k}. The bundle action records how to transport a fiber along every arrow of the action groupoid, including the automorphisms of \(o\).2726 2727The quotient stack has a related description that also keeps track of families. Over a smooth test manifold \(S\), an object of \([X/G]\) is a principal right \(G\)-bundle \(P\to S\) with a map \(u:P\to X\) such that \(u(pg)=g^{-1}u(p)\). The subset \(Q=u^{-1}(o)\) is a principal right \(H\)-bundle. To see its local triviality, choose a local section \(p:S_0\to P\), lift \(u(p(s))\) to \(a(s)\in G\) on smaller neighborhoods using sections of \(G\to G/H\), and use \(p(s)a(s)\) as the local section of \(Q\). Right multiplication on this section by \(h\in H\) gives all of \(Q\) over the neighborhood.2728 2729Conversely, an \(H\)-bundle \(Q\to S\) gives \(P=Q\times_H G\) with \(u([q,g])=g^{-1}H\). This is well defined because replacing \((qh,g)\) by \((q,hg)\) leaves \(g^{-1}H\) unchanged. Its fiber over \(o\) recovers \(Q\) by \(q\mapsto[q,1]\). In the other direction \([q,g]\mapsto qg\) is a \(G\)-bundle isomorphism giving the starting \(u\). Both constructions are functorial in bundle isomorphisms and commute with pullback to another test manifold. This proves the equivalence of quotient stacks \([G/H\,/\,G]\simeq BH\), where \(BH\) in the stack description assigns principal \(H\)-bundles to tests.2730 2731The Borel construction \(EG\times_G X\) has a corresponding statement for classifying spaces. The map \([e,gH]\mapsto egH\) is a homeomorphism onto \(EG/H\) for a contractible free universal \(G\)-space \(EG\), with inverse \(eH\mapsto[e,H]\). The latter quotient is a model of the classifying space of \(H\). For \(H=\mathrm{SO}(2)\simeq S^1\), one can use \(\mathbf{CP}^{\infty}\) as this classifying space. Its second integral cohomology is infinite cyclic, and the universal line associated to the weight-one character has first Chern class a generator \cite[Ch.~3]{hatcher-vbkt}. These are cohomological invariants of the homotopy quotient. The ordinary set orbit captures only transitivity.2732 2733\section{Weights, bundles, and section representations}2734Return to \(G=\mathrm{SO}(3)\), \(H=\mathrm{SO}(2)\), with \(o\) the north pole of \(S^2\). Write \(\chi_m(e^{it})=e^{imt}\) for the weight \(m\in\mathbf Z\) character and \(L_m=G\times_H\mathbf C_m\). With \([gh,v]=[g,\chi_m(h)v]\), a section is represented by a function \(f:G\to\mathbf C\) satisfying \(f(gh)=\chi_m(h)^{-1}f(g)\). Completing under the invariant probability measure gives the compact induced representation2735\[2736  L^2(S^2,L_m)=\operatorname{Ind}_H^G\chi_m.2737\]2738Compact Frobenius reciprocity calculates the multiplicity of any continuous finite-dimensional irreducible \((V,\pi)\) as the dimension of \(\operatorname{Hom}_H(V|_H,\mathbf C_m)\). Concretely, an \(H\)-map \(b\) gives the intertwiner with \(v\)-section \(f_v(g)=b(\pi(g^{-1})v)\). The covariance equation follows by moving \(\pi(h^{-1})\) across \(b\). On the continuous matrix coefficients, evaluation at \(1\) gives the inverse formula. Haar projection onto an irreducible isotypic type gives the same comparison in the Hilbert completion, as in compact Frobenius reciprocity \cite[Sec.~8]{robert}.2739 2740The weights of \(V_\ell\) on \(H\) are the integers from \(-\ell\) through \(\ell\), each with multiplicity one. Hence2741\begin{equation}2742   L^2(S^2,L_m)\simeq2743      \widehat{\bigoplus}_{\ell\geq |m|}V_\ell .2744   \label{eq:synthesis-spherical-sections}2745\end{equation}2746For \(m=0\), the bundle is the scalar line and this yields the multiplicities of the rotation layers in Chapter~\ref{ch:learning}. For any two weights \(m,n\), an equivariant bounded linear operator between their section spaces consists of one scalar for every \(\ell\geq\max(|m|,|n|)\), with uniformly bounded moduli, by \eqref{eq:learning-blocks}. The scalars depend on choices of unitary identifications of the common irreducibles. All absent types give zero blocks. This describes an operator between different field types even when they live over the same base.2747 2748\section{Opposite weights with the same spectral types}2749Orient \(S^2\) so that the action on its tangent plane at \(o\) is the standard complex rotation. The equivariant real tangent bundle, viewed as a complex line by the oriented metric, is \(L_1\), since its fiber representation at \(o\) is \(\chi_1\). Associated bundles commute with tensor product and duals, so \(L_m=L_1^{\otimes m}\) with dual powers for negative \(m\). The first Chern class of the tangent line is the Euler class of the tangent plane \cite[Ch.~3]{hatcher-vbkt}. Evaluation on the fundamental class gives \(\chi(S^2)=2\), hence2750\[2751   c_1(L_m)=2m\,u,\qquad u\in H^2(S^2;\mathbf Z),\quad2752       \langle u,[S^2]\rangle=1.2753\]2754The equivariant first Chern class is \(m c\), where \(c\) is the weight-one generator on \(EG/H\). The map from a fiber \(S^2\) of \(EG\times_G S^2\to BG\) pulls back the associated equivariant line to the ordinary line \(L_m\), so it sends \(c\) to \(2u\).2755\[2756 \begin{tikzcd}[column sep=large, row sep=large]2757 S^2 \arrow[r,"\mathrm{fiber\ inclusion}"]2758 & EG\times_G S^2 \arrow[r,"\simeq"] & BH\\2759 H^2(S^2;\mathbf Z)2760 & H^2_G(S^2;\mathbf Z) \arrow[l,swap,"\mathrm{restriction}"]