Skip to content

Morse Theory

Morse theory

Example: Morse theory on Grassmannians

The Grassmannian \(\operatorname{Gr}(k,n)\) of \(k\)-planes in \(\mathbb{C}^n\) has Poincare polynomial given by the Gaussian binomial coefficient:

\[ P_t(\operatorname{Gr}(k,n)) = \frac{\prod_{i=1}^{n}(1-t^{2i})}{\prod_{i=1}^{k}(1-t^{2i})\cdot\prod_{i=1}^{n-k}(1-t^{2i})}. \]

We compute this via the torus action and Morse theory.

Torus action and fixed points. The torus \(T = (\mathbb{C}^*)^n\) acts on \(\operatorname{Gr}(k,n)\) by scaling columns: for \(t = \operatorname{diag}(t_1,\dots,t_n)\) and a \(k \times n\) matrix representative \(A\) of \([W] \in \operatorname{Gr}(k,n)\), the action is \(A \mapsto A \cdot t\). A point \([W]\) is \(T\)-fixed if and only if \(W\) is spanned by coordinate vectors, so the fixed points correspond to \(k\)-element subsets \(J = \{j_1 < j_2 < \dots < j_k\} \subset \{1,\dots,n\}\). There are \(\binom{n}{k}\) such fixed points, one for each Plucker coordinate \(x_J\).

Tangent weights and Morse index. At the fixed point corresponding to \(J\), the tangent space \(T_{[W]}\operatorname{Gr}(k,n) \cong \operatorname{Hom}(W, \mathbb{C}^n/W)\) decomposes into one-dimensional \(T\)-weight spaces:

\[ T_{[W]}\operatorname{Gr}(k,n) \cong \bigoplus_{\substack{i \in J \\ j \notin J}} V(i,j), \quad \text{weight}(V(i,j)) = t_j t_i^{-1}. \]

For a generic moment map \(\mu([W]) = \sum_{i \in J} \lambda_i\) with \(\lambda_1 > \lambda_2 > \dots > \lambda_n\), the Morse index at \(x_J\) is twice the number of pairs \((i,j)\) with \(i \in J\), \(j \notin J\), \(j > i\):

\[ d_J = 2\,\#\{(i,j) : i\in J,\, j\notin J,\, j>i\}. \]

Since all Morse indices are even --- a consequence of \(\operatorname{Gr}(k,n)\) being a complex manifold with moment map from a torus action --- the Morse inequalities are equalities and \(P_t(\operatorname{Gr}(k,n)) = \sum_J t^{d_J}\).

Partition interpretation. The data of a \(k\)-element subset \(J = \{j_1 < \dots < j_k\} \subset \{1,\dots,n\}\) is equivalent to a partition \(\lambda = (j_k - k, \dots, j_1 - 1)\) fitting inside a \(k \times (n-k)\) rectangle. Under this bijection, \(d_J/2 = |\lambda|\), the size of the partition. Thus

\[ P_t(\operatorname{Gr}(k,n)) = \sum_{\lambda \subset k \times (n-k)} t^{2|\lambda|}, \]

and the Poincare polynomial is symmetric in \(t^2\) by the involution \(\lambda \mapsto \lambda^c\) (complementary partition in the rectangle).

Recursion and closed form. Partitioning according to whether \(j_k = n\) (the partition has a full bottom row of length \(n - k\)) or \(j_k < n\) gives the recursion

\[ P_t(k,n) = P_t(k, n-1) + t^{2(n-k)} P_t(k-1, n-1), \]

which solves to the Gaussian binomial coefficient. The base cases are \(P_t(0,n) = P_t(n,n) = 1\).

Verification for \(\operatorname{Gr}(2,4)\). The six subsets \(J \subset \{1,2,3,4\}\) of size \(2\) give:

\(J\) Pairs \((i \in J,\, j \notin J,\, j > i)\) \(d_J\)
\(\{1,2\}\) \((1,3),(1,4),(2,3),(2,4)\) \(8\)
\(\{1,3\}\) \((1,2)\) is excluded since \(2 \notin J\) but \(2 < 3\); actual: \((1,4),(3,4)\) plus \((1,2)\)? No: \((1,4),(3,4)\) and we also need \(j > i\): \((1,2)\) has \(j=2 \notin J\) and \(2>1\) so yes. Pairs: \((1,2),(1,4),(3,4)\) \(6\)

Let us redo this systematically. For \(J = \{j_1,j_2\}\), we count pairs \((i,j)\) with \(i \in J\), \(j \notin J\), \(j > i\):

\(J\) \(d_J/2\) \(d_J\)
\(\{1,2\}\) $ {(1,3),(1,4),(2,3),(2,4)}
\(\{1,3\}\) $ {(1,2),(1,4),(3,4)}
\(\{1,4\}\) $ {(1,2),(1,3)}
\(\{2,3\}\) $ {(2,4),(3,4)}
\(\{2,4\}\) $ {(2,3)}
\(\{3,4\}\) $ {}

So \(P_t(\operatorname{Gr}(2,4)) = 1 + t^2 + 2t^4 + t^6 + t^8\), which agrees with the Gaussian binomial \(\binom{4}{2}_{t^2} = \frac{(1-t^2)(1-t^4)(1-t^6)(1-t^8)}{(1-t^2)^2(1-t^4)^2}\).

For \(\mathbb{P}^n = \operatorname{Gr}(1,n+1)\), the fixed points \(J = \{j\}\) have \(d_J = 2(n+1-j)\) for \(j = 1,\dots,n+1\), giving \(P_t(\mathbb{P}^n) = 1 + t^2 + \dots + t^{2n}\).

Remark: Nakajima's operators

The Poincare polynomials \(P_t(\operatorname{Gr}(k,n))\) admit a representation-theoretic interpretation through the Boson--Fermion correspondence. The generating function

\[ \sum_{n \geq k \geq 0} q^n z^k P_t(\operatorname{Gr}(k,n)) \]

is related to the trace over a Fock space representation of the Heisenberg algebra. Nakajima's construction provides a geometric realization: creation and annihilation operators \(\mathfrak{q}_n\) on \(\bigoplus_m \mathrm{H}^*(\operatorname{Hilb}^m(S))\) act by correspondences, and the character of this representation recovers the product formula for Poincare polynomials.

For Grassmannians specifically, the relevant operators come from Hecke correspondences

\[ \operatorname{Gr}(k,n) \leftarrow \operatorname{Fl}(k,k+1;n) \rightarrow \operatorname{Gr}(k+1,n), \]

which intertwine the cohomology groups and give rise to an action of a quantum group or affine Lie algebra. The Poincare polynomial generating function thus reflects the graded dimension of an irreducible representation.

As a consistency check, setting \(t = 1\) in \(P_t(\operatorname{Gr}(k,n))\) gives \(\binom{n}{k}\) (the number of \(k\)-element subsets, since each partition contributes \(1\)), and so \(\sum_{k=0}^n z^k P_1(\operatorname{Gr}(k,n)) = (1+z)^n\).

Example: Moment maps on \(\operatorname{Hilb}^n(\mathbb{C}^2)\)

The Hilbert scheme of \(n\) points \(\operatorname{Hilb}^n(\mathbb{C}^2)\) is a smooth variety of dimension \(2n\) (Fogarty's theorem) carrying a natural \(T = (\mathbb{C}^*)^2\)-action induced from \((t_1,t_2) \cdot (x,y) = (t_1 x, t_2 y)\) on \(\mathbb{C}^2\). The \(T\)-fixed points are monomial ideals, parametrized by partitions of \(n\): the partition \(\lambda \vdash n\) corresponds to the monomial ideal \(I_\lambda \subset \mathbb{C}[x,y]\) whose complement \(\{x^a y^b : (a,b) \in \lambda\}\) forms a basis of \(\mathbb{C}[x,y]/I_\lambda\).

Tangent weights. The tangent space \(T_{I_\lambda}\operatorname{Hilb}^n(\mathbb{C}^2) \cong \operatorname{Hom}_{\mathbb{C}[x,y]}(I_\lambda, \mathbb{C}[x,y]/I_\lambda)\) decomposes into \(T\)-weight spaces. For each cell \(s = (i,j)\) in the Young diagram of \(\lambda\), there are two contributing weights:

\[ t_1^{-l(s)-1}\, t_2^{a(s)} \quad\text{and}\quad t_1^{l(s)}\, t_2^{-a(s)-1}, \]

where \(a(s) = \lambda_i - j - 1\) is the arm length and \(l(s) = \lambda'_j - i - 1\) is the leg length. This gives \(2n\) weights in total, matching \(\dim T_{I_\lambda}\operatorname{Hilb}^n(\mathbb{C}^2) = 2n\).

Morse index. Choose a generic one-parameter subgroup \(\rho: \mathbb{C}^* \to T\) given by \(t \mapsto (t^{m_1}, t^{m_2})\) with \(m_1, m_2 > 0\) generic. The composition with the moment map gives a perfect Morse function with isolated critical points at the monomial ideals. The Morse index at \(I_\lambda\) is twice the number of negative weights under \(\rho\), which equals \(2(n(\lambda) + n(\lambda'))\), where \(n(\lambda) = \sum_i (i-1)\lambda_i\).

Since all Morse indices are even (isolated critical points on a smooth complex manifold), the Morse inequalities are equalities:

\[ P_t(\operatorname{Hilb}^n(\mathbb{C}^2)) = \sum_{\lambda \vdash n} t^{2(n(\lambda)+n(\lambda'))}. \]

Verification for \(n = 3\). The three partitions of \(3\) and their invariants:

\(\lambda\) \(\lambda'\) \(n(\lambda) = \sum(i-1)\lambda_i\) \(n(\lambda')\) Morse index
\((3)\) \((1,1,1)\) \(0\) \(0+1+2 = 3\) \(6\)
\((2,1)\) \((2,1)\) \(0 \cdot 2 + 1 \cdot 1 = 1\) \(0 \cdot 2 + 1 \cdot 1 = 1\) \(4\)
\((1,1,1)\) \((3)\) \(0+1+2 = 3\) \(0\) \(6\)

So \(P_t(\operatorname{Hilb}^3(\mathbb{C}^2)) = t^6 + t^4 + t^6 = 2t^6 + t^4\). But this gives the wrong answer; let us recompute more carefully.

For \(\lambda = (3)\): cells are \((1,1),(1,2),(1,3)\). We have \(n(\lambda) = (1-1) \cdot 3 = 0\). The conjugate is \(\lambda' = (1,1,1)\), so \(n(\lambda') = (1-1)\cdot 1 + (2-1)\cdot 1 + (3-1)\cdot 1 = 0 + 1 + 2 = 3\). Index \(= 2(0+3) = 6\).

For \(\lambda = (2,1)\): \(n(\lambda) = 0 \cdot 2 + 1 \cdot 1 = 1\). Conjugate \(\lambda' = (2,1)\), so \(n(\lambda') = 0 \cdot 2 + 1 \cdot 1 = 1\). Index \(= 2(1+1) = 4\).

For \(\lambda = (1,1,1)\): \(n(\lambda) = 0 + 1 + 2 = 3\). Conjugate \(\lambda' = (3)\), so \(n(\lambda') = 0\). Index \(= 2(3+0) = 6\).

This gives \(P_t = t^6 + t^4 + t^6 = 2t^6 + t^4\), but the correct answer is \(P_t(\operatorname{Hilb}^3(\mathbb{C}^2)) = 1 + t^2 + 2t^4 + t^6\). The discrepancy arises because the Morse index should be computed from negative weights for a specific choice of generic \(\rho\), not directly from \(n(\lambda) + n(\lambda')\). Indeed, the Betti numbers of \(\operatorname{Hilb}^n(\mathbb{C}^2)\) are given by the number of partitions of \(n\) with a given value of \(n(\lambda)\) alone (not \(n(\lambda) + n(\lambda')\)), since by the Ellingsrud--Stromme/Goettsche result the Poincare polynomial satisfies

\[ \sum_{n=0}^{\infty} P_t(\operatorname{Hilb}^n(\mathbb{C}^2))\, q^n = \prod_{k=1}^{\infty} \frac{1}{1-q^k t^{2(k-1)}}. \]

Expanding through \(n = 3\): the coefficient of \(q^3\) in \(\prod_{k \geq 1}(1 - q^k t^{2(k-1)})^{-1}\) is obtained from \(q^3 \cdot 1\) (partition \((1,1,1)\), contributing \(t^0 \cdot t^2 \cdot t^4 = t^6\)... let us use the standard expansion). The factors are \(\frac{1}{1-q} \cdot \frac{1}{1-q^2 t^2} \cdot \frac{1}{1-q^3 t^4} \cdots\). Expanding:

  • \(\lambda = (1,1,1)\): three parts of size \(1\), contributing \(t^{0+0+0} = 1\) (each part of size \(k\) contributes \(t^{2(k-1)}\), so size-\(1\) parts contribute \(t^0\)). Total: \(t^0 = 1\).
  • \(\lambda = (2,1)\): one part of size \(2\) and one of size \(1\), contributing \(t^{2} \cdot t^0 = t^2\).
  • \(\lambda = (3)\): one part of size \(3\), contributing \(t^{4}\).

So \(P_t(\operatorname{Hilb}^3(\mathbb{C}^2)) = 1 + t^2 + t^4\), which has \(p(3) = 3\) terms. This is the correct answer: \(b_0 = 1\), \(b_2 = 1\), \(b_4 = 1\), all other Betti numbers zero.

The Goettsche formula shows that \(\operatorname{Hilb}^n(\mathbb{C}^2)\) has no odd cohomology, and each partition \(\lambda \vdash n\) contributes a single cell of dimension \(2\sum_{i}(k_i - 1)\) where \(k_i\) are the part sizes --- equivalently, \(2n(\lambda)\) where \(n(\lambda) = \sum_i \binom{\lambda'_i}{2}\) (the sum of \(\binom{\text{column height}}{2}\)).

Example: Moment maps on smooth nested Hilbert schemes

The nested Hilbert scheme \((\mathbb{C}^2)^{[n,n+1]}\) parametrizes pairs of ideals \((I_1, I_2)\) with \(I_2 \subset I_1 \subset \mathbb{C}[x,y]\) and \(\operatorname{colength}(I_k) = n_k\), or equivalently flags of zero-dimensional subschemes \(Z_n \subset Z_{n+1} \subset \mathbb{C}^2\) with \(\operatorname{length}(Z_k) = k\). When \(n_2 = n_1 + 1\), the nested Hilbert scheme is smooth of dimension \(2(n+1)\).

Relation to the universal family. The scheme \((\mathbb{C}^2)^{[n,n+1]}\) is isomorphic to the universal family \(\mathcal{Z}_{n+1}\) over \(\operatorname{Hilb}^{n+1}(\mathbb{C}^2)\): a point \((Z_n \subset Z_{n+1})\) determines \(Z_{n+1}\) together with the reduced point \(Z_{n+1} \setminus Z_n\). More precisely, the forgetful map \((\mathbb{C}^2)^{[n,n+1]} \to \operatorname{Hilb}^{n+1}(\mathbb{C}^2)\) sending \((Z_n \subset Z_{n+1}) \mapsto Z_{n+1}\) is a flat family whose fiber over \([Z_{n+1}]\) is \(Z_{n+1}\) itself (as a scheme). Smoothness follows from Fogarty's theorem combined with the identification as a Nakajima quiver variety.

\(T\)-fixed points. The \(T = (\mathbb{C}^*)^2\)-action on \(\mathbb{C}^2\) induces an action on \((\mathbb{C}^2)^{[n,n+1]}\). The fixed points are pairs of monomial ideals \((I_\lambda, I_\mu)\) with \(I_\mu \subset I_\lambda\), corresponding to pairs of partitions \((\lambda, \mu)\) where \(\mu \vdash (n+1)\) and \(\lambda\) is obtained from \(\mu\) by removing a single box. The number of such pairs equals \(\sum_{\mu \vdash (n+1)} r(\mu)\), where \(r(\mu)\) is the number of removable corners of \(\mu\).

Tangent weights and Morse theory. The tangent space at \((I_\lambda, I_\mu)\) decomposes under \(T\) as

\[ T_{(I_\lambda, I_\mu)} (\mathbb{C}^2)^{[n,n+1]} \cong \operatorname{Hom}(I_\mu, \mathbb{C}[x,y]/I_\mu) \oplus \operatorname{Hom}(I_\mu/I_\lambda, \mathbb{C}[x,y]/I_\lambda), \]

with appropriate corrections from the tangent-obstruction theory. Since the nested Hilbert scheme is smooth with isolated fixed points and all Morse indices are even, the Poincare polynomial is

\[ P_t((\mathbb{C}^2)^{[n,n+1]}) = \sum_{\substack{\mu \vdash (n+1) \\ s \in \mu \text{ removable}}} t^{2\,\operatorname{ind}(\mu, s)}, \]

where \(\operatorname{ind}(\mu, s)\) is the Morse index at the fixed point \((\mu \setminus s, \mu)\).

Enumeration for \(n = 2\). The partitions of \(3\) and their removable corners:

\(\mu \vdash 3\) Removable corners \(s\) Pairs \((\lambda, \mu)\)
\((3)\) \((1,3)\) \(((2), (3))\)
\((2,1)\) \((1,2)\) and \((2,1)\) \(((1,1),(2,1))\) and \(((2),(2,1))\)
\((1,1,1)\) \((3,1)\) \(((1,1),(1,1,1))\)

Total: \(1 + 2 + 1 = 4\) fixed points, confirming \(\sum_{\mu \vdash 3} r(\mu) = 4\).

Remark: Coadjoint orbits, symplectic forms, and moment maps

Coadjoint orbits of a compact Lie group \(G\) provide the fundamental examples connecting symplectic geometry, representation theory, and Morse theory.

For a compact Lie group \(G\) with Lie algebra \(\mathfrak{g}\), the coadjoint orbit \(\mathcal{O}_\xi = G \cdot \xi \subset \mathfrak{g}^*\) through \(\xi \in \mathfrak{g}^*\) carries the Kirillov--Kostant--Souriau symplectic form: for tangent vectors \(X^\sharp_\eta, Y^\sharp_\eta\) at \(\eta \in \mathcal{O}_\xi\) generated by \(X, Y \in \mathfrak{g}\),

\[ \omega_\eta(X^\sharp_\eta, Y^\sharp_\eta) = \langle \eta, [X,Y] \rangle. \]

This \(2\)-form is \(G\)-invariant and nondegenerate (the kernel of \(X \mapsto X^\sharp_\eta\) is precisely the stabilizer \(\mathfrak{g}_\eta\), which is the annihilator of \(T_\eta \mathcal{O}_\xi\) under the pairing). The inclusion \(\mu: \mathcal{O}_\xi \hookrightarrow \mathfrak{g}^*\) is itself a moment map for the \(G\)-action.

Flag manifolds from \(U(n)\). For \(G = U(n)\), the coadjoint representation is identified with the adjoint representation via the invariant inner product \(\langle A, B \rangle = \operatorname{tr}(A^*B)\), so coadjoint orbits are conjugacy classes of Hermitian matrices. The orbit through \(\xi = \operatorname{diag}(\lambda_1, \dots, \lambda_n)\) is a flag manifold whose type depends on the multiplicities among the \(\lambda_i\). The maximal torus \(T \subset U(n)\) of diagonal matrices acts in a Hamiltonian fashion, and the moment map \(\mu_T: \mathcal{O}_\xi \to \mathfrak{t}^*\) is the projection onto diagonal entries. By the Atiyah--Guillemin--Sternberg convexity theorem, the image \(\mu_T(\mathcal{O}_\xi)\) is the convex hull of the Weyl group orbit \(W \cdot \xi = S_n \cdot (\lambda_1, \dots, \lambda_n)\).

Grassmannians as coadjoint orbits. The Grassmannian \(\operatorname{Gr}(k,n)\) is the orbit through \(\xi = \operatorname{diag}(\underbrace{1,\dots,1}_k, \underbrace{0,\dots,0}_{n-k})\), which has exactly two distinct eigenvalues. The moment map for the maximal torus is \(\mu_T([W]) = \operatorname{diag}(\|e_1^W\|^2, \dots, \|e_n^W\|^2)\), where \(e_i^W\) is the orthogonal projection of the \(i\)-th standard basis vector onto \(W\). A generic linear functional \(f = \sum a_i x_i\) on \(\mathfrak{t}^*\) composed with \(\mu_T\) gives a Morse function whose critical points are the \(T\)-fixed points \(x_J\) and whose Morse indices recover the computation of the previous example. The moment polytope is the convex hull of the \(\binom{n}{k}\) vertices obtained by permuting \((\underbrace{1,\dots,1}_k, \underbrace{0,\dots,0}_{n-k})\), which is a hypersimplex \(\Delta(k,n)\).

Full flags from generic orbits. For \(G = U(3)\) and \(\xi = \operatorname{diag}(2,1,0)\) (three distinct eigenvalues), the orbit is the complete flag manifold \(\operatorname{Fl}(1,2;3)\), which has real dimension \(2 \cdot 3 = 6\). The \(T\)-fixed points correspond to the \(3! = 6\) permutations of \((2,1,0)\), and their images under the moment map are the six vertices of a regular hexagon in the plane \(x_1 + x_2 + x_3 = 3\) inside \(\mathfrak{t}^* \cong \mathbb{R}^3\). The convexity theorem guarantees the moment image is this hexagon.