On the Harder-Narasimhan filtration for finite dimensional representations of quivers (Q2452178)

A quiver is a directed graph. Let \(Q\) be a finite quiver, with \(Q_0\) the vertices and \(Q_1\) the arrows. Denote an arrow between vertices \(v,w\in Q_0\) by \(v\to w\in Q_1\). Let \(M=((M_v)_{v}, (M_{v\to w})_{v\to w})\), where \(M_v\) are finite dimensional vector spaces (over an algebraically closed field \(\Bbbk\)) indexed by \(Q_0\), and \(M_{v\to w}:M_{v}\to M_{w}\) are linear maps indexed by \(Q_1\). \(M\) is called a representation of \(Q\). Let \(\mathcal{R}=\bigoplus_{v\to w\in Q_1} \mathrm{Hom}_{\Bbbk}(M_v,M_w)\) be the space of all representations for fixed spaces \(\{M_v\;|\;v\in Q_0\}\). The reductive group \(\mathcal{G}=\prod_{v\in Q_0}\mathrm{GL}(M_v)\) acts on \(\mathcal{R}\) by: \[ (g_v)_v\cdot (M_{v\to w})_{v\to w}=(g_wM_{v\to w}g^{-1}_v)_{v\to w}. \] Fix integers \(\{\Theta_v\;|\;v\in Q_0\}\) and define \(\Theta=\sum_{v\in Q_0}\Theta_v \mathrm{dim}_{\Bbbk}M_v.\) Likewise, fix positive integers \(\{\sigma_v\;|\;v\in Q_0\}\) and define \(\sigma=\sum_{v\in Q_0}\sigma_v \mathrm{dim}_{\Bbbk}M_v\). Now define a character of \(\mathcal{G}\) by \(\chi((g_v)_v)=\prod_{v\in Q_0}\mathrm{det}(g_v)^{\Theta\sigma_v-\sigma\Theta_v}\). Using this character, \(\chi\)-semistability is defined. Let \(\mu=\Theta/\sigma\). A representation \(M\) is \((\Theta,\sigma)\)-semistable if for all subrepresentations \(M'\subset M\) we have \(\mu (M')\leq \mu(M)\). It is shown in this paper that \(\chi\)-semistability and \((\Theta, \sigma)\)-semistability coincide. As discussed in the paper, every representation \(M\) has a unique filtration \(0\subset M_1\subset M_2\subset\cdots\subset M\) such that \(\mu(M_i/M_{i-1})>\mu(M_{i+1}/M_i)\) for all \(i\) and each \(M_i/M_{i-1}\) is \((\Theta,\sigma)\)-semistable. This filtration is called the \textit{Harder-Narasimhan} filtration of \(M\). Additionally, if \(M\) corresponds to a \(\chi\)-unstable representation in \(\mathcal{R}\), then there exists a unique weighted filtration of \(M\) called the \textit{Kempf filtration} of \(M\) The main theorem of this paper states that the Kempf filtration of \(M\) is the Harder-Narasimhan filtration of \(M\).