Cohomology of noncommutative Hilbert schemes (Q2644169)

Let \(A=k\langle x_1,\dots,x_m\rangle\) be the free associative \(k\)-algebra, \(k\) algebraically closed. A representation of \(A\) on a finite dimensional vector space \(W\) of dimension \(d\) consists of a tuple \(\varphi_\ast\in\text{End}_k(W)^m.\) The orbits of \(\text{GL}(W)\) in \(\text{End}(W)^m\) correspond bijectively to the isomorphism classes of \(d\)-dimensional representations of \(A\). Fix another \(k\)-vector space \(V\) of dimension \(n\), together with a basis \(v_1,\dots,v_n.\) Then \(X=\text{Hom}(V,W)\oplus\text{End}(W)^m\) parametrizes \(d\)-dimensional representations of \(A\), together with a fixed linear map from \(V\) to \(W\), and again the group \(\text{GL}(W)\) acts on \(X\) via base change in \(W\). Define a tuple \((f,\varphi_\ast)\) to be stable if the image of \(f\) generates \(W\) as a representation of \(A\), and let \(X^s\) denote the subset of \(X\) consisting of stable tuples. Then \(H_{d,n}^{(m)}=X^s/\text{GL}(W)\) is the quotient variety of \(X^s\) by \(\text{GL}(W).\) It is then known that \(H_{d,n}^{(m)}\) is smooth and irreducible, of dimension \(N=nd+(m-1)d^2,\) and that the set \(X^s\) of stable tuples is a principal \(\text{GL}(W)\) bundle over \(H_{d,n}^{(m)}.\) The variety \(H_{d,n}^{(m)}\) has several interpretations, that is, the \(k\)-points parameterize each of the following sets: (1) Equivalence classes of \(d\)-dimensional representations \(W\) of \(A\) together with an \(n\)-tuple of vectors generating \(W\) as a representation of \(A.\) (2) Equivalence classes of \(d\)-dimensional representations \(W\) of \(A,\) together with a surjective \(A\)-homomorphism from the free representation \(A^n\) to \(W\). (3) \(A\)-subrepresentations of codimension \(d\) of the free representation \(A^n.\) (4) Isomorphism classes of stable representations of the quiver \(Q_n^{(m)}.\) In particular, the variety \(H_{d,1}^{m}\) parametrizes left ideals of codimension \(d\) in \(A\). Thus it can be viewed as a noncommutative Hilbert scheme for the free algebra in \(m\) generators, in the same way as the Hilbert scheme \(\text{Hilb}^d(\mathbb{A}^m)\) parametrizes ideals of codimension \(d\) in the polynomial ring \(k[x_1,\dots,x_m].\) Denote the quotient variety \(\text{End}(W)^m//\text{GL}(W)\) by \(V_d^{(m)}.\) Then \(V_d^{(m)}=\text{Spec}(k[\text{End}(W)^m]^{\text{GL}(W)}),\) such that its \(k\)-points are in bijection with the semisimple representations of \(A\). By a result of C. Procesi, the ring \(R=k[\text{End}(W)^m]^{\text{GL}(W)}\) is generated by the functions \((\varphi_1,\dots,\varphi_m)\mapsto\text{tr}(\varphi_{i_1},\dots,\varphi_{i_s})\) for sequences \((i_1,\dots,i_s)\in \{1,\dots,m\}.\) In fact, \(R\) is already generated by such functions for \(s\leq d^2+1.\) One of the problems in dealing with the varieties \(V_d^{(m)}\) is that they are highly singular, except in the cases \(m=1,\) and \(d=2=m\). Although no explicit desingularizations of the varieties \(V_d^{(m)}\) are known (except in case \(d=2\)), the noncommutative Hilbert schemes \(H_{d,n}^{(m)}\) form a class of closely related smooth varieties: The obvious map \(X^s\rightarrow\text{End}(W)^m,\) is \(\text{GL}(W)\)-equivariant. Thus it descends to a projective morphism \(p:H_{d,n}^{(m)}\rightarrow V_d^{(m)}\) on the level of quotients by \(\text{GL}(W).\) The fibres of \(p\) are difficult to determine in general, but they are at least tractable using the Luna stratification of \(V_d^{(m)}\) and the theory of nullcones of quiver representations. The morphism \(p\) extends the canonical map from the Hilbert scheme \(\text{Hilb}^d(\mathbb{A}^m)\) to the \(d\)th symmetric power \((\mathbb{A}^m)^d/S_d.\) The main aim of this paper is to prove the following results on the geometry of the varieties \(H_{d,n}^{(m)}\) The variety \(H_{d,n}^{(m)}\) has a cell decomposition, whose cells are parametrized by \(m\)-ary forests with \(n\) roots and \(d\) nodes. As a consequence, the Betti numbers in the cohomology of \(H_{d,n}^{(m)}(\mathbb{C})\) can be described in a compact form by assembling the Poincaré polynomials into a generating function. This leads to the result that the cohomological Euler characteristic of \(H_{d,n}^{(m)}\) is given by \[ \chi(H_{d,n}^{(m)})=\frac{n}{(m-1)d+n}\left(\begin{matrix} md+n-1\\d\end{matrix}\right). \] The last main result of the article considers the asymptotic behavior of both the Euler characteristic and the Poincaré polynomials: After a suitable normalization, the distribution of the Betti numbers of \(H_{d,1}^{(m)}\) for large \(d\) has the Airy distribution as a limit law. The author thoroughly defines words and forests and uses this to construct a cell decomposition. This is very well written, and the applications that follows are then easy to understand. That is, the normal forms for representations and submodules, and in some sense the intersection theory of \(H_{d,n}^{(m)}\). Also, the generating functions are well treated. The article ends with the final result about the asymptotics of the Euler characteristic of \(H_{d,n}^{(m)}\).