Patrick Lei

All notes

Singular spaces

Lectures by Paul Gunnells, UMass, Spring 2020.

Topology and Geometry of Singular Spaces
Math 797D
Notes by Patrick Lei, Spring 2020
Lectures by Paul Gunnells

University of Massachusetts Amherst

Disclaimer

These notes were taken during lecture using the vimtex package of the editor neovim. Any errors are mine and not the instructor’s. In addition, my notes are picture-free (but will include commutative diagrams) and are a mix of my mathematical style (omit lengthy computations, use category theory) and that of the instructor. If you find any errors, please contact me at plei@umass.edu.

1 January 21

1.1 Course Description

Singular spaces arise naturally in many contexts, including algebraic geometry and representation theory. Any singular space has a decomposition into manifolds (a "stratification"), and so the study of them is a mixture of topology, geometry, and combinatorics. The goal of this course is to present the basics of stratified spaces and to illustrate the general theory with some important examples. Potential topics: general material, including stratifications, Whitney conditions, local structures, intersection homology; isolated singularities of complex hypersurfaces (the Milnor fibration and related topics), including connections to knot theory; and compactifications of locally symmetric spaces.

1.2 Organization

This class has a webpage: www.math.umass.edu/~gunnells/singspc/singspc.html. Grading will be based on an expository paper of around 10 pages. More information is on the website. He says he does this because most grad students don’t do enough writing until the thesis.1 Occasionally Paul will suggest exercises that we can think about.

1.3 Overview

Obviously this courses is about singular spaces, in particular stratified spaces. Recall that a nice space is a smooth manifold, which is locally Euclidean. An example of this is a torus. However, not all useful spaces are manifolds.

Example 1.1.

Consider the nodal cubic given by collapsing a circle on the torus to a point. Algebraically, this is given by y2=x3+x2. This is not a manifold, but it has a decomposition (stratification) into manifolds (S1×S1∖S1,pt).

Example 1.2.

Affine algebraic varieties have a natural stratification (take the smooth locus, then repeat on the singular locus). For example, for a hypersurface (f=0), the singular locus is given by the vanishing of all partial derivatives of f.

For a simple example, consider f=x3−y2. This is singular at the origin. If we consider S3∩Vf, we obtain an S1. Viewing S3=ℝ3∪{∞}, our S1 is a trefoil knot.2 The stratification into manifolds is obvious.

Example 1.3.

Consider a manifold M and topological group G acting on M. Sometimes the space M/G is a manifold (for example when M=Sn,G=ℤ/2⁢ℤ, M/G=ℙn⁡(ℝ)). However, typically the quotient is not a manifold. Sometimes the space is not even Hausdorff.3 If we suppose that the action is proper (this is automatically satisfied if G is compact ((S1)k,S⁢O⁢(n))), then M/G will be a stratified space. If G is finite, then M/G is a global orbifold.

Definition 1.4.

Let G be a Lie group and K a maximal compact subgroup. For example, consider G=S⁢Ln⁢(ℝ) and K=S⁢O⁢(n). Alternatively, consider G=S⁢Ln⁢(ℂ) and K=S⁢U⁢(n). The space X=G/K is a contractible smooth manifold diffeomorphic to Euclidean space. This is a global symmetric space. Now let Γ⊂G be a discrete group (for example S⁢Ln⁢(ℤ)). Another example is S⁢Ln⁢(ℤ⁡[i])⊂S⁢Ln⁢(ℂ). Now Γ has a natural action on G/K. Thus a locally symmetric space is Γ\G/K.

Example 1.5.

Consider G=S⁢L2⁢(ℝ) and K=S⁢O⁢(2). Then X=G/K≃ℋ, the upper half plane. Now Γ=S⁢L2⁢(ℤ) acts by Möbius transformations. Then Γ\ℋ≃ℂ.

In general, locally symmetric spaces are orbifolds. However, they are not compact, so we want to build good compactifications for these spaces. For this, we have a well-developed theory that leads to very interesting singular spaces. In the previous example, we may consider principal congruence subgroups Γ⁡(N), and it turns out Γ⁡(N)\ℋ is a curve of some genus minus finitely many points.

It should be apparent that this subject is all over the place. For this reason, Paul finds this subject very interesting. Resources can be found on the course webpage.

Definition 1.6.

Let M be a smooth manifold and Z be a subset of M. Then a stratification of Z is a filtration by subsets Z0⊂Z1⊂⋯⊂Zn=Z such that:

  1. 1.

    Si=Zi∖Zi−1 is a locally closed regular submanifold of M of dimension i (possibly disconnected or even empty).4 The connected components of the Si are strata.

  2. 2.

    If S,T are strata and S¯∩T≠∅, then T⊂S¯. This gives a poset structure on the set of strata.

  3. 3.

    The stratification is locally finite. Equivalently, any point is contained in the closure of finitely many strata.

Next lecture we will see the Whitney conditions, which guarantee that the geometry is locally homogeneous.

2 January 23

2.1 Basic Examples

Last time we defined what a stratification is. Here is an example that is not very exciting:

Example 2.1.

A manifold with boundary has a very simple stratification. More generally, a manifold with corners has a stratification given by the codimension of the corner. Note that topologically, a manifold with corners is the same thing as a manifold with boundary. A tetrahedron is an example of a manifold with corners, but an octahedron is not.

Example 2.2.

Let V be a finite-dimensional vector space over ℝ or ℂ. Let Z be a finite collection of affine subspaces. This defines a stratification of V. Note this works for any locally finite connection of subspaces, e.g. a periodization of Z by some lattice.

Example 2.3.

Let X be a smooth manifold and consider Xn. Then for some I⊂[n] with |I|≥2, we can consider the set ΔI={xj=xk⁢ iff ⁢j,k∈I}. We will consider the space

This is related to configuration spaces. Locally this looks like a subspace arrangement.

Example 2.4.

Let G be a finite group acting on X. X has a stratification given by the various stabilizers. For example, we can obtain the previous example by considering the action of Sn on Xn.

Example 2.5.

Let M=G⁢r⁢(k,n). For example, G⁢r⁢(1,n)=ℙn−1. Now choose a fixed flag in ℝn. We will obtain a stratification by looking at intersections with the flag. Considering G⁢r⁢(2,4), we projectivize our setup and find six possible configurations. We can form a poset by taking limiting configurations, and the closures of strata are called Schubert varieties. The strata are called Schubert cells.

There is another construction (the matroid stratification) that we can perform on the Grassmannian. Fix a basis e1,…,en. Then consider the n! fixed flags that come from this basis. Now take all possible intersections of all Schubert cells. This is not a stratification because the axiom of the frontier fails.

Exercise 2.6.

Figure out what the Schubert varieties of G⁢r⁢(2,4) look like.

Example 2.7.

Let Z be a complex projective variety. Z has a natural stratification given by iteratively taking the singular locus. However, this is not always the best stratification. Consider the Whitney cusp given by (x3+z2⁢x2−y2=0). The singular locus is the z-axis. The generic cross-section looks like a nodal curve, but the the cross-section at z=0 is is cuspidal curve.

2.2 Some Geometry

Recall that two regular submanifolds M,N intersect transversely inside X if Tp⁢M+Tp⁢N⊂Tp⁢X has maximal possible dimension. We ran out of time, so next time we will define normal slices and links.

3 January 28

3.1 Normal Slices and Links

Let M⊃Z with 𝒮 the set of strata. Then let p∈Z be contained in stratum S. Choose a small disk N∋p and transverse to all strata containing p with dimN+dimS=dimM.

Definition 3.1.

THe normal slice to p is N∩Z and the link is defined as ∂N∩Z.

Example 3.2.

On the Whitney cusp, the normal slice of a generic singular point is an X-shape and the normal slice of the origin is a V-shape. The links are four points and two points, respectively.

Example 3.3.

Let p be contained in a codimension k corner. Then it is not difficult to see that L⁡(p)=Δk, the (k−1)-simplex.

Exercise 3.4.

Stratify ℂn by the coordinate subspaces. Compute the links. As a warm up, try this with ℝn, or if you feel ambitious, do this with Vn for any real vector space.

3.2 The Whitney Conditions

3.2.1 Condition A

Let Z=⋃Sα be a stratification. Then

  1. 1.

    Suppose Sα⊂Sβ¯.

  2. 2.

    In addition, suppose that we have a sequence {xi}⊂Sβ and y∈Sα.

Then if Txi⁢Sβ→P, we have Ty⁢Sα⊂P. Here the limiting process of spaces takes place in the Grassmannian.1

3.2.2 Condition B

Suppose that conditions 1 and 2 hold has above. Additionally suppose that for y∈Sα, {xi}⊂Sβ, {yi}⊂Sα with xi→y,yi→y, consider the secant lines ℓi through xi,yi. Then if ℓi→ℓ, we have ℓ⊂P.

Example 3.5.

On the Whitney cusp, this condition fails at the origin.

Exercise 3.6.

Prove that condition B implies condition A.

Definition 3.7.

Let M⊃Z with 𝔖 the set of strata. This is a Whitney stratification if it is a stratification and satisfies Whitney’s condition B.

Example 3.8.

If we make the origin a stratum on the Whitney cusp, then we obtain a Whitney stratification.

Definition 3.9.

Suppose Z is Whitney stratified. Then

  1. 1.

    The homeomorphism types of links and normal slices are uniquely determined as soon as the disk N is sufficiently small.

  2. 2.

    The normal slice is homeomorphic to the cone on the link.

  3. 3.

    Locally near p, Z looks like S×c⁢L⁢(p) (local triviality along the strata). More precisely, given S⊂Z, there exists a closed neighborhood TS⊂Z and a locally trivial fibration TS→S such that f−1⁢(P)≃c⁢Lp⁢(S).

Proof of this theorem can be found in Mather’s notes.

Definition 3.10.

A map f:X→Y of spaces is a fibration if it satisfies the path lifting property: If x∈f−1⁢(y) and we have a closed loop in Y based at y, it can be lifted to a path in X based at x.

Definition 3.11.

A fibration is locally trivial if there exists a covering {U} such that f−1⁢(U) is homeomorphic to a product U×F. Here F is called the fiber.

Example 3.12.

Any fiber bundle is an example of a locally trivial fibration. If F is equipped with the action of a topological group G, the fiber bundle is determined by a covering {U} and a G-valued cocycle. Let F=T2=ℝ2/ℤ2 and γ∈S⁢L2⁢(ℤ). Then let M=T2×I/(t,0)∼(γ⁢t,1). This gives a torus fibration M over the circle.

Exercise 3.13.

Compute the homology of M.

Theorem 3.14.

Let X be a real or complex semianalytic semianalytic variety. Then X admits a filtration by semianalytic varieties whose connected components form the strata of a Whitney stratification.2

A recent proof of this was given by Kaloshin.

4 February 4

4.1 Whitney, Continued

Last time we discussed the Whitney conditions and their consequences. Here is an example of a nontrivial link bundle.

Example 4.1.

Recall the construction in Example 3.12. Then take the matrix

This is a torus bundle T2→M→S1 over the circle, but is not trivial. Now we will construct three singular spaces:

  • •

    Set X¯=M×[0,1). This is a manifold with boundary.

  • •

    In ∂X¯=M, collapse the fibers of M over S1. This is a stratified space with a circle glued onto a four dimensional manifold. Call this X̂.

  • •

    Now collapse all of the boundary S1 to a single point. Call this X∗.

Now consider the boundary strata and the link of a point.

Boundary Stratum Link of point
X¯ M pt
X̂ S1 T2
X∗ pt M
Remark 4.2.

These singularities appear in the study of Hilbert modular surfaces.

4.2 Thom’s Presentation of Stratified Spaces

Our goal here is to explain how to assemble a stratified space from simpler ones (manifolds with corners) by gluing them together. The main reference is a paper in the Liverpool Singularities Symposium. This will allow us to compute links.

Consider a torus with an open slice taken out, so it has boundary two circles. Then if we glue the two boundary components together and collapse them, we obtain the nodal cubic. Here, we use

Theorem 4.3 (Collar Neighborhood Theorem).

Let ∂X=Y. Then there exists a neighborhood of Y in X that locally looks like Y⊃U×[0,ε).

Now we generalize this to manifolds with corners.1 Suppose Mn is a manifold with corners. Then let Mi,n be a codimension 1 corner where 0≤i≤n−1. Now let Ui,n⊂M be a neighborhood of Mi,n. Note that we have maps

Choosing small ε, we set Ui,nε≔ρi−1⁢([0,ε)). If ε is sufficiently small, we can ensure that:

  1. 1.

    Ui,nε→πi×ρiMi,n×[0,ε) is an isomorphism;

  2. 2.

    On Ui,nε∩Uj,nε, we have πi⁢πj=πj⁢πi;

  3. 3.

    We also have ρj⁢πi=ρj;

  4. 4.

    If S⊂{0,…,n−1}, set

    Then MS,n is a corner of codimension |S|, and similarly to the above, we have US,nε≃MS,n×[0,ε)|S|.

Given all of this data, we can recover the manifold with corners. We can see the links of corners. If p is contained in a corner, then we can recover the link by

4.2.1 Thom Presentation

Let Mi be a manifold with corners for i=0,…,n. For j<i let Mj⁢i be a codimension 1 boundary component of Mi. Define MS similarly for S⊂{0,…,n} with largest element i. Now suppose we have MS⁢S′ where every element of S′ is larger than every element of S. Then we have

  • •

    A proper fibration π:MS⁢S′→MS;

  • •

    An inclusion MS⁢S′↪MS′;

Both the inclusions and fibrations satisfy the expected compatibility relations.

5 February 6

5.1 Thom, Continued

Last time we bagan the Thom presentation of stratified spaces. Recall that we have manifolds with boundary Mi with MS boundary components for S with largest element i. Also recall we have proper fibrations deleting on the right and inclusions deleting on the left. Also assume that the fibrations and inclusions satisfy the expected compatibility relations. Now we will construct a singular space from this as

where ∼ is determined by the fibrations.

Theorem 5.1.

Any Whitney stratified space has a Thom presentation.

We discussed the Whitney cusp, a picture of which is included. Now the strata are images of open strata in the manifolds with corners. We now describe how to compute links. Suppose p∈Si. Consider all sets {i}≠S∋i such that i is the smallest element. Then for each S let MSp⊂MS be the inverse image of p in MS. Then on this collection, we obtain an induced Thom presentation. The link is the associated singular space.

Example 5.2.

For the Whitney cusp, the link at a generic singular point is four points. At the origin, the link is three circles.

Exercise 5.3.

Do the same for ℂ2 with the coordinate axes, do ℂn,ℝn, and then do the Whitney cusp over ℂ.

Example 5.4.

Recall the space X¯ in Example 4.1. This has the Thom presentation M0,M01,M1 where M0=M01=Y, the torus bundle over S1. Therefore the link at a boundary component is a point.

For X̂, the presentation is now M0=S1. Then the fibration is π:Y→S1, so the link at a point is the torus.

Finally, for X∗, the presentation now has M0=pt. Clearly here the link is now all of Y.

We will give a geometric explation of why the links work this way. The idea is that the link parameterizes the directions we can move from the boundary component into the space.

5.2 Thin Singularities

A modification of this scheme works for some singularities. The idea is that we know the links of the corners in a manifold with corners. A codimension k corner has link Δk−1. We will try to use this in a nontrivial way. The resulting links will look like Δ×{spaces}/∼ on proper faces of Δ.

Example 5.5.

For ℝ2 stratified by the coordinate axes, our thin scheme is given by taking each of the quadrants. We can recover the link from this presentation.

Next time we will define the thin presentation.

6 February 11

Last time we began an overview of thin singularities. Recall that this is a variation of the Thom presentation that uses knowledge of manifolds with corners.

6.1 Thin Singularities in Detail

We will consider manifolds with corners MS,T, where S,T are disjoint subsets of {0,…,n−1}. Here, S tracks the depth of the corner and T tracks the codimension of this piece in the whole picture.

  • •

    Clearly the top (N) dimensional manifold with corners is M∅,∅;

  • •

    Ma,∅ is a codimension 1 corner in M∅,∅;

  • •

    Similarly, MT,∅ is a codimension |T| corner in M∅,∅;

  • •

    M∅,a is a (N−1)-dimensional manifold with corners;

  • •

    MS,T is a codimension |S| corner in a manifold with corners of dimension N−|T|.

6.1.1 Fibrations

For a∉S∪T with S∩T=∅, then we have a fibration MS∪{a},T→MS,T∪{a}. In addition, we assume that all natural commutativity relations hold. In particular, if S,T,A are pairwise disjoint, we should obtain a unique fibration MS∪A,T→MS,T∪A.

6.1.2 Fullness

Note that this is not some sort of alternative healing thing.1. Recall that the fiber product of two maps is the limit X×ZY of

Here, consider the diagram

We will require that the map MS⁢a⁢b,T→MS⁢b,T⁢a×MS,T⁢a⁢bMS⁢a,T⁢b is surjective. In addition, we will require this for all such diagrams.

Now we can assemble our singular space.

Definition 6.1.

Set

where the equivalences come from the fibrations.

Example 6.2.

We can do this for the plane stratified by the coordinate axes, where the closure of each quadrant is an M∅,∅.

6.2 Constructing Links

Let p∈M∅,T. We will consider the image of p in X. Then we consider the full inverse image of p in MT,∅→M∅,T. Here, the link of p is the quotient

where on the face where the coordinates corresponding to A⊂T are nonzero, we collapse the fibers of MT,∅p→MA,Bp.

Remark 6.3.

Note that if the third coordinate is nonzero, that means you are moving towards the point (0,0,1) because you are moving away from the opposite edge where the third coordinate is zero.2

Example 6.4.

In the plane stratified by the coordinate axes, the link of a generic point on a coordinate axis is just a point. The link of the origin is a circle.

Example 6.5.

Consider the product of the first quadrant and the torus. Then we can collapse to obtain two cylinders, and then further collapse to a point. Then we have a four-dimensional stratum, two two-dimensional strata, and a point. We want to compute the link of S0. Note that the map from M01,∅→M1,0 is collapsing vertically, so we have a solid torus and two circles. Paul claims that the link is S3 and that this example is really ℂ2 stratified by the coordinate axes. In addition, S3 meets the two coordinate axes in circles, which form a Hopf link.

Remark 6.6.

S3 is the union of two solid tori. To see this, note that

Remark 6.7.

In ℝ2, the link of the point is S1 which contains two S0, which are linked in some sense.3

7 February 13

7.1 Toric Varieties

Recall Example 6.5 from last time. We can generalize this example in the following way: Let P⊂ℝn be a lattice polytope of dimension n. We will assume that P is simple, which means that each vertex meets n facets.

Example 7.1.

Both the cube and the tetrahedron are simple polytopes. However, each vertex of the octahedron meets 4 facets, so it is not simple.

Note that for a simple polytope P, each vertex looks like a corner in ℝn. In particular, any face of codimension k is determined by k facets containing it.

We can consider the space X=P×Tn/∼. We will have to collapse certain things on the boundary of P. For each facet F⊂P, choose a primitive normal vector vF∈ℤn that points inward. Then each vF determines a subgroup TF⊂𝕋n. Similarly, any face determines a subgroup, which is the direct sum of the subgroups determined by the facets containing it.

Now for each p∈P, let F⁡(p) be the face that contains it in its relative interior. Now we can define the equivalence relation on P×Tn by (p,g)∼(q,h) if p=q and g⁢h−1∈TF⁡(p). This gives us a presentation of a stratified space with thin singularities, which happen to be projective toric varieties attached to simple polytopes.

Example 7.2.

If P=I=[0,1], then X=S2. If P=In, then X=(S2)n. Also, if P is the simplex, then X=ℙℂ2.

Remark 7.3.

Note that toric varieties are compactification of the (algebraic) torus ℂ×.1 The reason that this is the same as our presentation is because ℂ×≃S1×ℝ.

Proposition 7.4.

X is a manifold if and only if for each vertex p∈P, the associated normal vectors v1,…,vn corresponding to facets F1,…,Fn form a ℤ-basis for ℤn.

Remark 7.5.

This condition guarantees that locally at each vertex we get the picture from Example 6.5. This also answers a question of Tetsuya about whether the space is uniquely determined by the homeomorphism type of the polytope.

Example 7.6.

Consider the triangle with vertices (0,0),(1,0),(1,3). Then at the origin, X the two normals do not generate ℤ2, so X is singular at the origin. Then the link at this point is a quotient of S3 called a lens space.2

7.2 Intersection Homology

Note that algebraic topology is not a prerequisite for this course, so students are not assumed to know what homology is.3. The idea is to attach a collection of vector spaces that are topological invariants. This type of construction is ubiquitous throughout mathematics.

7.2.1 Homology Crash Course

We will define groups Hi⁢(X,G), where G is typically ℤ,ℚ,ℝ and where 0≤i≤dimX. We expect this to be an invariant under homeomorphism.4

Example 7.7.

Hi⁢(S1,ℤ)=ℤ for i=0,1. Similarly, Hi⁢(Sn)=ℤ for i=0,n and is zero otherwise.

Example 7.8.

Hi⁢(Tn)=ℤ(ni) for all i.

Example 7.9.

Let L⁡(p,q) be the lens space corresponding to the cyclic quotient singularity determined by ℤ/p⁢ℤ acting by character (1,q). Then Hk⁢(L⁡(p,q),ℤ)=ℤ for k=0,3 and H1⁢(L⁡(p,q),ℤ)=ℤ/α⁢ℤ.

The general formula5 for homology is as follows:

  1. 1.

    From X, produce a chain complex of abelian groups with boundary map ∂.

  2. 2.

    Compute the kernel Zi⁢(X)=ker⁢∂i.

  3. 3.

    Compute the image Bi⁢(X)=Im⁢∂i+1.

  4. 4.

    Compute the homology Hi⁢(X,ℤ)=Zi/Bi.

8 February 20

Last time we discussed homology. There are many ways to define homology, but we will use a combinatorial approach in this course.

Pros Cons
Geometric meaning easy to understand Depends on extra data
Easy to compute Not clearly an invariant

The idea is to decompose our space X by simplices in a reasonable way. For example, the tetrahedron i s a model of the sphere. Unfortunately, not all spaces (or even topological manifolds) can be triangulated.1 Fortunately, smooth manifolds and Whitney stratified spaces can be triangulated. We can require that the stratification is compatible with the stratification.

Remark 8.1.

There is a class of triangularizable spaces between 𝐓𝐨𝐩 and 𝐃𝐢𝐟𝐟, namely the category 𝐏𝐋 of piecewise-linear manifolds.

8.1 Piecewise Linear Manifolds

Definition 8.2.

An abstract simplicial complex Δ is a family of nonempty finite subsets of a set S such that:

  1. 1.

    If A∈Δ and ∅≠B⊂A, then B∈Δ;

  2. 2.

    Any B∈Δ is only contained in finitely many A∈Δ.

The subsets are called faces and the dimension of face A is defined to be |A|−1.

Definition 8.3.

An abstract simplicial complex Δ is pure if all facets have the same dimension.

We now define the geometric realization of a simplicial complex. Suppose A∈Δ has dimension n. Then F⁡(A) is the standard n-simplex on ℝn. If we choose a total ordering on S, then B⊂A can be identified with a face F⁡(B)⊂F⁡(A). Then we can define the geometric realization to be the colimit

Definition 8.4.

A triangulation of a space X is an abstract simplicial complex Δ and a homeomorphism |Δ|→X.

Definition 8.5.

An abstract simplicial complex Δ′ is a subdivision of Δ if |Δ′|≃|Δ| and every simplex of Δ′ lines in some simplex of Δ.

Definition 8.6.

A morphism f:Δ→Δ′ of simplicial complexes is a map f:|Δ|→Δ′ such that f|A maps linearly onto a simplex of Δ′.

Definition 8.7.

A piecewise-linear map f:Δ1→Δ2 is a map f such that there exist subdivisions Δ1′→Δ2′ such that f:Δ1′→Δ2′ is a morphism.

Definition 8.8.

A combinatorial n-manifold is a triangulated manifold such that links of p-simplices are PL-isomorphic to the boundary of an (n−p)-simplex.

Definition 8.9.

A PL-manifold is a manifold where the transition functions are PL.

Remark 8.10.

Piecewise linear implies combinatorial, so PL-manifolds have triangulations.

8.2 Defining Homology

Given a simplicial complex Δ, we can make a chain complex C∗⁢(Δ). We consider a total ordering on the set S and we consider symbols [v0,…,vn], where permuting v0,…,vn changes the sign of the symbol. Now we define

and the boundary map ∂:Cn⁢(Δ)→Cn−1⁢(Δ) by

For example, we can see that ∂([v0,v1])=[v1]−[v0].

Exercise 8.11.

Check that ∂2=0 (just check what happens when you delete two vertices in the opposite order).

Remark 8.12.

The purpose of the sign in the definition of δ is to respect the orientation of the simplices.2

Example 8.13.

Consider the circle S1 represented as a triangle with vertices 0,1,2. Then we see that

Then the boundary ∂ is defined by the matrix

Then we see that H1⁢(S1)≃ℤ and H0⁢(S1)≃ℤ.

Remark 8.14.

In order to compute the homology, we need to use Smith normal form, which can be found in musty old British algebra textboks.

8.2.1 A Variation

There is a variation which goes by various names (local coefficients, twisted coefficients, local systems), where we have a nontrivial action on the coefficients G of the homology.

Definition 8.15.

A local system G~ is given by a representation π1⁢(X,x)→Aut⁡G.

There is a construction due to Steenrod of H∗⁢(X,G~) which is not seen in a first topology course but is very important. Combinatorially, for each face B of a simplex A, choose a point xB in the relative interior of the geometric realization. Then over each xB we place a copy GB of G. Then for each face, we choose isomorphisms GA→GB. Then we require all natural compatibility relations to hold.

Remark 8.16.

Local systems are intermediate between ordinary homology and sheaf cohomology.

9 February 25

9.1 Homology with Local Systems

Last time we began our discussion of homology with coefficient systems. Today we will define this using simplicial homology.

Let Δ be a triangulation of X. Over each simplex σ∈Δ, choose a point xσ in the relative interior of σ. Now attach a copy Gσ of G at each xσ. If τ<σ, choose an isomorphism hτ⁢σ:Gσ→Gτ. Then we require all compositions to commute.

For a reference, see Steenrod’s paper or Whitehead’s big book on homotopy theory.

We now need to create a homology theory out of this. Define the chains Cn⁢(X,G~) to be G-linear combinations of symbols [x0,…,xn] with boundary map

where hi is the isomorphism hσ′⁢σ obtained by deleting xi. Then we define H∗⁢(X,G~) as the homology of the chain complex.

Example 9.1.

Let X=S1 and G=ℚ2. Here the local system will be given by γ=(1101). Then we can represent S1 as a triangle with the xσ of the edges being their midpoints. We choose all maps to be the identity except for one, which is γ.

The 0-chain is C0⁢(X,G~)≃ℚ6 and C1⁢(X,G~)≃ℚ6. Now we need to understand the boundary map. Now the boundary is given by

We can check the rank of ∂, which is 5, so H0≃H1≃ℚ.

Remark 9.2.

If we replace γ is the identity matrix, then the rank of ∂ is now 4, so H0≃H1≃ℚ2. If we set γ=(2111), then H1≃H1≃0.

Remark 9.3.

Recall our manifold Mγ given by a T2-bundle over S1. Our question is: Given a fiber bundle F↪E→B, can we compute H∗⁢(E) in terms of H∗⁢(B),H∗⁢(F)? Recall that if E=B×F, then H∗⁢(E)=H∗⁢(B)⊗H∗⁢(F). In general, there is the Leray-Serre spectral sequence. Under favorable circumstances, the E2 page will determine H∗⁢(E). For Example 9.1, we computed Hp⁢(S1,H1⁢(T2)~). Then the homology of Mγ is determined by taking direct sums along the diagonals of the E2 page.1

9.2 Some Technical Remarks

Remark 9.4.

There is a way to eliminate dependence on the triangulation, namely PL geometric chains. Write C∗Δ⁢(X) for the chain complex attached to the triangulation. Suppose that Δ′ is another triangulation and that Δ,Δ′ have a common refinement Δ′′. Then we can take the direct limit of this system and call the elements PL geometric chains. This makes sense for any PL manifold.

Remark 9.5.

Simplicial homology is not the only way to work. It is the easiest way to work in the elementary treatment, but there are others:

Regular Cell Complexes

The closure of a cell must be homeomorphic to (B,∂B), and ∂B must be a union of cells. The boundary map is slightly more complicated.

CW Complexes

These are much more efficient and need very few cells, but writing down the boundary is much more difficult.

Singular Homology

Here, we work with maps Δn→X.

Remark 9.6.

Homology is a functor H∗:𝐇𝐨𝐓𝐨𝐩→𝐀𝐛. There is a similar construction for local systems as long as the local systems are related.

Theorem 9.7 (Poincare Duality).

Let X be a compact oriented manifold. Then there exists a perfect pairing

This implies that Hi⁢(X)≃Hn−i⁢(X)∗. The pairing is called the intersection form on X.2

10 February 27

Recall Poincare duality from last time. The classical picture is using dual cell decompositions. Let Δ be a triangulation of X and let Δ′ be the barycentric subdivision. Then we can obtain the dual cell decomposition by taking unions of simplices that are transverse to the original σ∈Δ. Note that this is a regular cell complex Z, but not a simplicial complex. If we compute the homology H∗⁢(Z∗), we obtain the same result as if we compute H∗⁢(Δ∗).

Remark 10.1.

This fails if we take integral coefficients. For example, if we take the lens space Λ⁡(a,b). This has homology groups ℤ,ℤ/a⁢ℤ,0,ℤ, so we see Poincare duality does not hold. This is an example of a rational homology sphere.

10.1 Cohomology

Cohomology is the dual theory to homology. We will consider cochain complexes with maps d:Ci→Ci+1. Here d is defined as the adjoint of ∂ with respect to the pairing Ci⊗Ci→ℤ given by1

Then recall that cohomology has a cup product structure, which is dual to the intersection product and turns cohomology into a ring.

If X is a compact oriented manifold, then we can phrase everything in terms of the cup product. The reason cohomology is easier to define a product for is because the construction is local.

10.2 Intersection Homology

We want a homology theory for singular spaces that has as many nice properties of usual homology as possible. The first problem is that Poincare duality fails for ordinary homology.

Example 10.2.

Consider X={x⁢y=0}⊂ℙ2. Then X is a union of two ℙ1. Then we can see that H0⁢(X)=ℚ, but H2⁢(X)=ℚ2.

Example 10.3.

Now consider the suspension Σ⁢T2. Recall that H∗⁢(T2)=ℚ,ℚ2,ℚ. It is a good exercise to determine H∗⁢(Σ⁢T2). However, note that the suspensions of two independent cycles do not intersect in a cycle, so we cannot define an intersection product.

Now let α be a cycle on the torus and W∈H2 be the suspension of the other cycle β on T2. Then α,W meet transversely and intersect in one point. Then we see that α is a boundary, so it has no dual.

Remark 10.4.

Returning to Example 10.2, if we do not allow chains to pass through the intersection point, then we recover Poincare duality. Thus the main idea of intersection homology is to introduce data to control how chains meet the singular locus. More generally, we will impose conditions on how chains meet all strata.

Doing this will give the opportunity to both eliminate and create homology. This new data is called perversity.2 This will allow us to pair cycles created with different perversities and recover Poincare duality.

11 March 3

Note: I was away on this day, notes were provided by Arthur Wang.

Definition 11.1.

A pseudomanifold is a stratified space with a singular locus of codimension at least 2. Alternatively, a pseudomanifold has a triangulation where every (n−1)-simplex is a face of exactly two n-simplices.

Recall that a compact oridnted manifolds have a fundamental class, which generates the top homology. This comes from the sum of all the n-simplices.

Example 11.2.

Every complex projective variety is a pseudomanifold.

Note we want the stratification to be compatible with the PL structure. Recall that in a Whitney stratification, we have a neighborhood of p that looks like Bi×co⁢(L⁡(p)). We need this to be compatible with the PL structure. In particular, we need a filtered space V=Vn⊃⋯⊃Vi=pt and a PL i-ball Bi such that we have a map V×Bi→X that is a PL homeomorphism onto a neighborhood of p∈X. Fortunately, this can always be arranged for a Whitney stratified space.

To define intersection homology, we will use PL geometric chains. We work with chains ξ∈CΔ⁢(X) for some triangulation in the direct system.

Definition 11.3.

The support |ξ| of a geometric chain ξ js the union of the closures of the simplices where the coefficient of ξ is nonzero.

Definition 11.4.

A perversity p¯ is a sequence p¯=(o2,p3,…) such that

  1. 1.

    p2=0;

  2. 2.

    pk≤pk+1≤pk+1.

Example 11.5.

If X=Σ⁢T2, then p¯=(p2,p3), so there are two perversities: (0,0) and (0,1).

Remark 11.6.

We always have two distinguished perversities:

These are called the top and zero perversities.

There are two other interesting perversities:

Note that m¯+n¯=t¯.

Definition 11.7.

If p¯+q¯=t¯, then they are complementary.

For a complex projective variety, all strata have even dimension, so we can define p¯⁢(2⁢n)=n−1. This is called middle perversity and is most common in the literature.

The condition of slow growth is necessary to prove the topological invariance of intersection homology. In particular, the result is independent of the stratification.

Example 11.8.

Consider a homogeneous variety X (like a Grassmannian or flag variety). This has a stratification into Schubert varieties. We get many interesting constructions from this stratification going beyond the topology of X, which leads to perverse sheaves.

Definition 11.9.

A subspace Y⊂X is called (p¯,i)-allowable if

  1. 1.

    dimY≤i;

  2. 2.

    dim(Y∩Xn−k)≤i−k+pk for all k≥2.

Definition 11.10.

The intersection chain complex consists of chains ξ such that both ξ,∂ξ are allowable.

The intersection homology is defined exactly as expected.

Remark 11.11.

The second condition in allowability controls how much a chain can meet the singular locus and corresponds to a deviation from transversality.

Example 11.12.

Consider the two lines x⁢y=0 in ℙ2. This has intersection homology ℚ2,0,ℚ2, which satisfies Poincare duality. On the other hand, the ordinary homology is ℚ,0,ℚ2.

Exercise 11.13.

Do the same for the nodal cubic.

12 March 5

Today we compute some examples of perversity. Last time, we considered the example x⁢y=0 in ℙ2, and we saw that the perversity is identically 0.

Example 12.1.

Consider the nodal cubic. Then we see that the perversity is the zero vector, so there are no allowable 0-chains and 1-chains that meet the singular point, but there is an allowable 2-chain. Note that each ordinary homology group is ℚ. Then we can see that the intersection homology changes in dimension 1, where it is now zero.

12.1 Normal Spaces and IH

Let X be a stratified pseudomanifold with singular set Σ. Then we say X is normal if each point has a fundamental system of neighborhoods U such that U∖Σ is connected. In algebraic geometry, a space is normal if the local ring at each point is integrally closed.

If a space is normal, then IH with the top perversity recovers the usual homology. In general, IH of a space with top perversity will recover the homology of its normalization.In addition, if X is normal, IH with the zero perversity recovers the cohomology of X. In general, IH with zero perversity is the cohomology of the normalization.

Example 12.2.

Consider X=Σ⁢T2. Then the possible perversities are the zero perversity and the top perversity. First we note that the ordinary homology groups are H0=ℚ,H1=0,H2=ℚ2,H3=ℚ. This matches the intersection homology with the top perversity.

i i−k top zero
0 0−3 −2 −3
1 1−3 −1 −2
0−3 −2 −3
2 2−3 0 −1
1−3 −1 −2
3 3−3 1 0
2−3 0 −1
Table 12.1: Consitions for Allowable Chains.

Now we can compute the intersection homology with the zero perversity, which is ℚ,ℚ2,0,ℚ. Then Poincare Duality1 says there is a pairing

Our goal is to explain local properties of intersection homology groups. This is a key point in the passage from IH1 (geometric) to IH2 (sheafy).2

Example 12.3.

Consider an isolated point singularity. Then locally we obtain the open cone on the link L of the singularity, so we obtain co⁢L=X. When can a class survive? Only somewhere in between pn∈{0,…,n−2}.

Exercise: Find answer.

13 March 10

Note: I was not here on this day; notes were provided by Arthur Wang.

Consider the open cone on a link of an isolated singularity. Recall that this is the same as a neighborhood of the singularity. Note: There was a probably incorrect calculation done, that is not reproduced here.

We will consider two generalizations:

  1. 1.

    Suppose X has only isolated singularities. Then some discussion shows that

  2. 2.

    Suppose L is an (n−1)-dimensional stratified space. Then clearly c⁢L=X is a stratified n dimensional space with strata the cone point and the cones on the strata of L. Then we can compute

13.1 Local Coeffieicents for IH

Let X=Xn⊃Xn−2⊃⋯. To specify a local system on X for IH, we only need to do it on X∖Σ. For example, if X∖Σ is connected, then we only need a representation of π1⁢(X∖Σ). Note this is not true for regular homology (recall how regular homology with local systems worked).

Note that I⁢Cip¯⁢(X)⊂Ci⁢(X). Then (p¯,i) allowability means that we only need the data for a local system in X∖Σ. Suppose Δ is an i-simplex in |ξ| and ∂Δ is an (i−1)-simplex in |∂ξ|. Then we require

How large can these dimensions be? With the top perversity, then we have

Our goal is now to present Deligne’s definition, but we must beware of cheating. The key to the definition are local intersection homology groups. These have very spevific structure involving the links of strata.

13.2 Local Homology

We present some more facts about regular homology. Let X⊃Y and suppose X is triangulated so that Y is compatibly triangulated. Then Ci⁢(X)⊃Ci⁢(Y). We can make a quotient Ci⁢(X,Y)≔Ci⁢(X)/Ci⁢(Y). Then we obtain a boundary map making C∗⁢(X,Y) into a complex. Then H∗⁢(X,Y) is defined to be the homology of that complex.

Note that cycles in Ci⁢(X,Y) are chains in X with boundaries in Ci−1⁢(Y). Also if Y=∅, then H∗⁢(X,∅)=H∗⁢(X).

Example 13.1.

Let X=S2 and Y be the two poles. Note that we have a new class joining the two poles. Thus Hi⁢(X,Y)≠0. We can compute things using the long exact sequence of a pair:

Definition 13.2.

The local homology group at x∈X is H∗⁢(X,X∖{x}).

14 March 24

Because of the coronavirus, we moved lectures online. Paul says he practiced, so hopefully things go well. All lectures will go on YouTube and a link will be sent later today. In an email, Paul asked that we do not share the link, so the link will not be in the notes.

14.1 Cleaning up Loose Ends

Last time we computed the intersection homology on the link of an isolated singularities. Paul took information from different sources and there was conflicting information. After discussing this, we will go back to local systems in IH.

Recall that for compact manifolds there is only one homology theory that is really useful to think about. For noncompact manifolds, there are two different useful theories: usual homology, which is compactly supported, and H∗B⁢M⁢(M), which is called Borel-Moore homology or homology with closed support. Here we allow infinitely supported chains. For noncompact manifolds, these usually give different answers.

Example 14.1.

Recall that the usual homology of ℝn is ℚ is degree 0 and 0 everywhere else. On the other hand, the Borel-Moore homology of ℝn is ℚ is degree n and 0 everywhere else.

Remark 14.2.

Recall that Hi⁢(X)≃Hi⁢(X×ℝ). On the other hand, HiB⁢M⁢(X)≃Hi+1B⁢M⁢(X×ℝ). Thus Borel-Moore homology does not introduce new information, but is not truly homotopy invariant.

Let L be a compact (n−1)-dimensional compact manifold and X=c⁢L, the open cone on L. Then we have X=Xn and X0 is the cone point. Then, the usual intersection homology of X is

Here, pn=n−ℓ.

In the Borel-Moore situation, we can compute

Remark 14.3.

Nobody else uses this tilde notation. Please do not use it ever when you talk to other people.

Remark 14.4.

What happens when L is itself a stratified space and not a manifold? Then we should have the same story but replacing normal homology with intersection homology. We have

There is a similar statement for the usual intersection homology, which is left as an exercise.

14.2 Local Homology

Recall the definition of relative homology H∗⁢(X,Y). If Y=X∖{x}, this only makes sense in dimX=n, and we have Hn⁢(X,X∖{x})=Hn⁢(Bn,Bn∖{0}) by the excision theorem. By the long exact sequence of a pair, this becomes Hn−1⁢(Bn∖{0})≃ℤ.

We can now consider relative intersection homology I⁢H~∗p¯⁢(X,X∖{x}). Suppose x∈Sk. We should still have some form of excision and we expect some information about the link of x. Recall that a neighborhood of x∈Sk looks like ℝk×c⁢L. We have

A remark of this is in IH2, Borel seminar volume. Our goal is to motivate Deligne’s construction of IH via sheaf theory. Note this will involve a lot of lying as Paul will not tell us what a sheaf is, what a theory is, or motivate anything.

Remark 14.5.

Sheaves are nice because there are many canonical operations we can do with them without really thinking about what is going on geometrically.

Our goal is to construct a complex computing IH. We begin with the nonsingular part X∖Σ. Then set Si=Xi∖Xi−1 and Ui=X−Xn−i.

We have a complex

attached to the open set U2. We denote this by P⁢(ℚ)2∙. Then we grow this to U3 by

15 March 26

15.1 Sheafy Stuff (with caveats)

Let X=Xn⊇Xn−2⊇⋯⊆X0 be a stratified space and Si=Xi∖Xi−1 be the strata. Consider the open sets Ui=X∖Xn−i. Recall that we defined P⁢(ℚ)2∙ to be the complex with ℚ at degree 0 and is 0 everywhere. Then we define P⁢(ℚ)3∙ to be

Here

  • •

    (i2)∗ is the pushforward from U2 to U3 along the inclusion.1

  • •

    R means to take the derived version of the pushforward.2 What this means is to replace P⁢(ℚ)2∙ with an injective resolution and then apply (i2)∗.

  • •

    The τ≤p⁡(2) truncates the complex, killing anything in degree above p⁡(2) and replacing the object in degree p⁡(2) with ker⁡dp⁡(2).

Then we turn this into an inductive process:

Once we finish, we have a complex that computes intersection homology after taking the derived functor of global sections of the complex.3 In fact, the two complexes are quasi-isomorphic. This is different from the complex we obtained from the other definition, but it computes the same homology.

Remarks 15.1.
  1. 1.

    We could start with a local system instead of ℚ.

  2. 2.

    This definition makes sense in a broader setting because it uses definitions that make sense whenever we have the six functors formalism.

  3. 3.

    A good reference for this is the IH2 paper, where they explain what they did and how it relates to what came before.

15.2 Isolated Singularities of Complex Hypersurfaces

Now we make a shift in the course to discussing examples and applications of this theory. The first example, isolated singularities, does not rely on intersection homology. The standard reference for this is the book by Milnor.

Let f∈ℂ⁡[x1,…,xn] and suppose that ∇f=0 only at the origin. Assume f⁡(0)=0. Then V⁡(f) is a hypersurface in ℂn which has an isolated singularity at the origin.

Example 15.2.

Take n=2 and f∈ℂ⁡[x,y]. The link of 0 in ℂ2 is S3. The link of the singular point is simply V⁡(f)∩S3. This has real dimension 1, so L is a link in S3. If the hypersurface is smooth, then the link is the usual circle in S3.

If V⁡(f) is singular, then the link can have arbitrarily many connected components and can be arbitrarily complicated. For example, if f=x⁢y, then the link is the Hopf link. If f is a product of independent linear forms, then L is a collection of circles. Each circle is standardly embedded, but the components are typically linked in a complicated way. This depends on the actual linear forms.

Example 15.3.

Let f=x3−y2. Let (a,b) be a fixed real solution. Then for all angles θ, (a⁢e2⁢i⁢θ,b⁢e3⁢i⁢θ) is also a solution. Every solution of this form, so we can see what L looks like. First note that all points sit on a torus |x|=a,|y|=b. We can see that we obtain a line of slope 3/2 in the torus. Thus L is a trefoil knot embedded in the torus.

More generally, if we have f=xp−yq, then we get the (q,p) torus knot if (p,q)=1.

Question 15.4.

What kinds of knots and links can we obtain can arise as links of isolated singularities of hypersurfaces in ℂ2? Do those that arise have special properties? Classify the ones that arise.

Question 15.5.

How do we compute the link given f?

We will discuss these question in more depth later.

16 March 31

16.1 More Examples of Isolated Singularities

We will continue our discussion of isolated singularities and links of complex hypersurfaces. Recall that we suppose ∇f=0 only at the origin and that L=S2⁢n−1∩V⁡(f).

Example 16.1.

Consider f=x2+y3+z5∈ℂ⁡[x,y,z]. The link is a 3-manifold inside S5, which is known as the Poincare dodecahedral space. This is an interesting object in geometric topology which has the same integral homology as S3.1 The fundamental group of L turns out to be a perfect group, which means that it has trivial abelianization.

L can be realized as the configuration space of the regular dodecahedron in ℝ3. Thus L can be realized as L=S⁢O⁢(3)/A5, which is the same as S⁢U⁢(2)/A~5, where A~5 is the preimage of A5 under the covering map S⁢U⁢(2)→S⁢O⁢(3).2 The fundamental domain for the action of the binary icosahedral group on S3 is actually one of the facets of the 120-cell.

Example 16.2.

Consider f=x12+x22+x32+x42+x56⁢k−1. The link is a 7-manifold in S4. In fact, we actually get exotic spheres.

Definition 16.3.

An exotic sphere is a manifold that is homeomorphic but not diffeomorphic to a sphere.

Remark 16.4.

The first 14 examples were due to Milnor. By a result of Kervaire-Milnor, there are 28 different smooth structures on S7. The examples are all obtained by setting k=1,…,28.

16.2 Links of Plane Curve Singularities

We will study the links that appear in the two-dimensional case. First, L is the trivial knot (isotopic to S1⊂S3) if and only if V⁡(f) is smooth at the origin. For a proof of this, see Milnor’s book.

For example, if f=x3−y2, we can rewrite this as y=x3/2. Then we write y=a⁢ei⁢θ1,x=b⁢ei⁢θ2, and we can obtain θ1=32⁢θ2. We want to generalize this to see similar information with more complicated polynomials. The technique we will use is Puiseaux series.

Consider the field ℂ⁡((t)) of Laurent series over ℂ. Then Puiseaux series are elements of the algebraic closure of ℂ⁡((t)). It turns out adding all roots of t is enough, so all Puiseaux series have the form

for some fixed N.

Given f∈ℂ⁡[x,y], we can find solutions of the form y=∑aμ⁢xμ in Puiseaux series. In fact, there is an algorithm to compute them due to Newton. We will use this expression to compute the type of the link.

Example 16.5.

Consider f=y4−2⁢x3⁢y2+x6−4⁢x5⁢y−x7. We want to solve for y. This is an iterative algorithm, but where do we begin? We use the Newton polygon, which is simply everything above the segment between (0,4) and (6,0). Then if we consider the equation α+32⁢β=6, this gives us a notion of the weighted degree of xα⁢yβ. Thus all monomials along the segment have weighted degree 6, and others have higher weighted degree.

Assume that y has a solution that to first approximation is y=t⁢x3/2. Then we obtain

If t is a root of t4−2⁢t2+1, then this gives a good approximation. If we take t=1, then we see that y=x3/2 is a good approximation to the solution.

We continue, setting x1=x1/2, so y=x13⁢(1+y1). Then we obtain

Our new line segment is α+12⁢β=1. We set y1=t⁢x11/2 and compute t=1. Then we can see that f⁡(x1,y1)=0. Therefore we have

Next time we will discuss the general process and how to interpret the result.

17 April 2

17.1 Puiseaux Series

We will discuss the algorithm due to Newton to compute Puiseaux series. This is an iterative algorithm that starts with an initial approximation and then refines it. All of these intermediate steps are computed using Newton polygons.

The steepest nonvertical edge of the Newton polygon will be α+μ0⁢β=ν. Then we can rewrite

If we set y=t⁢xμ0, then the first terms become xν⁢g⁢(t). Choose a root t0 of g. This kills the first terms, so write μ0=p0q0. Then we write x=x1q0 and y=x1p0⁢(t0+y1). Then we substitute this into f and obtain something of the form

Apply the method again to f1 and continue. There are a few issues with this:

  1. 1.

    The procedure may not terminate. Fortunately, this is why we have power series.

  2. 2.

    The denominators may not be bounded. This will be explained later, but apparently denominator growth is controlled by the geometry of the Newton polygon.

To find the answer, we have something like

We claim that the denominators are bounded, so we really get a Puiseaux series. If ui is the height of the lowest point of the Newton polygon on the y-axis, then ui≥ui+1 and ui=ui+1 if and only if μi+1∈ℤ. By the well-ordering principle, we cannot have indefinite denominator growth. Suppose the solution looks like

and assume that not all λ are integers.

Let λ1 be the smallest nonintegral exponent and write it as λ1=n1m1. Then if λ2 is the next exponent not of the form α/m1, write it as λ2=n2m1⁢m2. We keep goiny and get a sequence of pairs (m1,n1),(m1,n2),…,(mg,ng). These are known as Puiseaux pairs and are uniquely determined by the series. A set of pairs gives rise to a standard Puiseaux series

17.2 Computing the Knots

Consider the map (x,y)↦x from ℂ2→ℂ. Then let x⁡(θ) be the small circle δ⁢ei⁢θ about the origin and consider its inverse image.

Example 17.1.

Consider y=x3/2. The inverse image has two branches:

We can graph this on a cylinder as θ goes from 0 to 2⁢π. The ends will join together to give us a knot.

Example 17.2.

Consider y=x3/2+x7/4. We have four branches in our map ℂ2→ℂ. If δ is very small, then the δ3/2 term will dominate.

The four branches will correspond to the two branches of the previous example (two new branches for each old one), but we will have a small perturbation and there is nontrivial motion as we move along the cylinder. Recall that the trefoil knot is the torus knot for x3/2. Then if we thicken the trefoil to a torus, the new knot is simply a perturbation.

Example 17.3.

Consider y=x3/2+x37/2. There is a single Puiseaux pair, so we get a perturbation but no new branches.

In general, we get an iterated torus knot. The depth of the iteration is the number of Puiseaux pairs. Unfortunately, this description does not make clear what knot we actually obtain. We know that the torus knot (p,q) is simply the the image of the line of slope qp. We will understand the iterated torus know by an interative procedure:

Order 0:

Standard S1⊂S3.

Order 1:

T2⊂S3 the boundary of a tubular neighborhood of S1.

Order k:

Body of closure of tubular neighborhood of order k−1 torus.

Definition 17.4.

A K⊂S3 is admissable embedded if

  1. 1.

    K⊂S1×D2.

  2. 2.

    π:K→S1 is a covering map.

Let T be a tubular neighborhood of K such that for each t∈S1 the intersection T∩{t}×D2 is a union of disjoint discs around the points K∩({t}×D2). Let the first S1 inside T be the boundary of any one of these distinct discs. To get the other circle, choose a curve on ∂T such that the direction from the curve to the knot is constant when measured in each disc.

18 April 7

18.1 Plane Curve Singularities

Last time we discussed a way to generate Puiseaux pairs from a function f∈ℂ⁡[x,y].

Theorem 18.1.

Suppose V⁡(f) is irreducible at 0. Then let (m1,n1),…,(mg,ng) be the sequence of Puiseaux pairs. Then L=L⁡(0) is an iterated torus know of type given by the Puiseaux pairs.

Recall from last time how we trivialized the tubular neighborhood of a torus knot. Now, it makes sense to discuss the (m2,n2) cable on the (m1,n2) torus knot. We can continually iterate this construction until we have exhausted all Puiseaux pairs.

Remark 18.2.

Now a natural question is to ask whether all sequences of relatively prime pairs arise from this process. Unfortunately, this is not true, and there are some conditions:

  • •

    m1<n1;

  • •

    nj−1⁢mj<nj.

For example, (2,3),(2,5) cannot arise from this process.

When the singularity is not irreducible, we get a link, not just a knot. We can ask what kinds of links can appear. The invariant we consider is the linking number l⁢k⁢(J,K) of two components J,K of a link.

  1. 1.

    If the two knots are unlinked, then lk(J,,K)=0;

  2. 2.

    Otherwise, we consider crossings where J goes under K, signed by orientation, and then take the sum.

Remark 18.3.

The definition of linking number is invariant under Reidemeister moves.1

Theorem 18.4.

The topology of a curve singularity is determined by the Puiseaux pairs for each branch and the pairwise linking numbers.

Remark 18.5.

Suppose we have two curves C,C′ in ℙ2 of degrees m,n. Then the linking number of C,C′ at an intersection point is exactly their intersection multiplicity. In particular, there cannot be unlinked components or negative linking numbers.

Remark 18.6.

Linking numbers do not determine a link. For example, consider the Borromean rings compared to the trivial link. In fact, a similar configuration exists for any number of components.

Remark 18.7.

Some references are Plane Algebraic Curves by Brieskorn and Knörrer and Walter Neumann’s survey paper.

18.2 Higher-Dimensional Hypersurface Singularities

The main reference for this is Milnor’s book. While it is somewhat old-fashioned, it does a lot with very little technology and accomplishes a lot.

Let f∈ℂ⁡[z0,…,zn] and that V⁡(f) is singular at the origin. Then recall L=V⁡(f)∩S2⁢n+1.

Theorem 18.8 (Milnor Fibration).

For ε sufficiently small, the map p:S2⁢n+1∖L→S1 given by

is a locally trivial fibration.

Now we want to study the fiber of this map. If the singularity is isolated, then each fiber is a manifold whose closure is a manifold with boundary L. If we go around the circle, we obtain a monodromy map F0→F0, and we have an induced automorphism of H∗⁢(F0).

Example 18.9.

Consider the case of a plane curve singularity. Then L is a knot in S3, and p gives us a collection of surfaces (over ℝ) in S3. These are called Seifert surfaces.

Definition 18.10.

A Seifert surface for a link L is a bicollared surface with boundary L. Here, bicollared means the boundary has two collar neighborhoods.

To construct a Seifert surface from a link, we can trivialize crossings, fill the circles into discs, and then untrivialize the crossings again.

Refer to caption
Figure 18.1: Seifert surface for the trefoil knot

Here are some more properties of Milnor fibers:

  1. 1.

    Each fiber is parallelizable;

  2. 2.

    Each fiber has the homotopy type of a cell complex of dimension n.

19 April 9

Paul used GP/Pari to force Zoom to record at a higher resolution.

19.1 More Milnor Fibration

Recall the definition of the Milnor fibration from last time. Then recall that F is parallelizable and has homotopy type of an n-dimensional cell complex. We give some more properties of the fibration:

  1. 1.

    One can show that π1⁢(L),…,πn−2⁢(L) are all trivial.

  2. 2.

    Each fiber is diffeomorphic to an open (in the analytic topology) subset of a nonsingular complex hypersurface of the form f=c, where c is a nonzero constant.

  3. 3.

    Suppose that 0 is an isolated singularity. Then F has the homotopy type of a bouquet of n-spheres.

Proof of these claims can be found in Milnor’s book, but they are too long for Paul to give them in class.

The key invariant of an isolated singularity is the Milnor number.

Definition 19.1 (Milnor number).

Let Jf be the ideal generated by the partial derivatives of f. Then the Milnor number is simply

Definition 19.2 (Milnor number, second definition).

Let J:ℂn+1→ℂn+1 be given by the ∂f∂zi. This induces a map on a smapp sphere Sε2⁢n+1→S2⁢n+1. The degree of the map is the induced map on H2⁢n+1⁢(S2⁢n+1).

Theorem 19.3.

The Milnor number μ is the number of spheres in the bouquet attached to the isolated singularity. In particular, Hn⁢(F)=ℤμ.

Example 19.4.

The Milnor number of z02+x13 is 2. From the explicit drawing of the surface, we can see a homotopy down to a graph with the same homotopy type as S1∧S1.

This is an example of something called a Brieskorn variety, which is a variety of the form x0a0+⋯+xnan. In this case, we can simply compute

Recall that we can consider the monodromy of the fibration. Then this gives an automorphism h of F, and we can regard h∗:Hn⁢(F)→Hn⁢(F) as an element of G⁢Lμ⁢(ℤ).

Theorem 19.5 (Deligne-Griffiths).

h∗ is a unipotent matrix. This means here that h∗ is conjugate to an upper triangular matrix with eigenvalues roots of unity. Moreover, all Jordan blocks of h∗ have dimension at most n+1.

Example 19.6.

Let f=∑i=0nziai. Then we can explicitly compute the eigenvalues of the monodromy map. In fact, they are all products of the form

where ωi is a nontrivial ai-th root of unity. For example, in the case f=z02+z13, the two eigenvalues are the primitive sixth roots of unity.

We will now sketch a proof of the formula. First, we can extend the fibration to all of ℂn+1∖V⁡(f). We can show that the fiber above y∈S1 is F×ℝ. We can now compute the monodromy of this fibration, which is simply

Then the monodromy is simply h2⁢π. By a result of Pham, we see that the fiber above 1 deformation retracts onto the iterated join of the Ωai. For example, the join of Ωp∗Ωq is simply the complete bipartite graph Kp,q.

Example 19.7.

If we label some vertices on the Seifert surface we drew last time, then we can see the graph.

We can now compute the reduced homology of F, and we see that the top homology is simply

Remark 19.8.

Results extend to weighted homogeneous polynomials f. Here, f⁡(x1,…,xm) is weighted homogeneous of tope a1,…,am if it is a linear combination of monomials z1i1⁢⋯⁢zmim where

20 April 14

Paul calculated e and π in GP/Pari.

20.1 Globally Symmetric Spaces

Example 20.1.

Let ℋ be the upper half plane and Γ=S⁢L2⁢(ℤ) (or some finite-index subgroup), and then we can consider the space Γ\ℋ. This is a locally symmetric space, which is simply a punctured Riemann surface. There are two obvious compactifications:

  1. 1.

    Fill in the missing points;

  2. 2.

    Fill in the missing points with a bounding circle.

Why is this a locally symmetric space? It is a quotient of a global symmetric space. Here, if G=S⁢L2⁢(ℝ) and K=S⁢O⁢(2), then ℋ=G/K.

Definition 20.2.

Let M be a Riemannian manifold with an involution sp for each p∈M that fixes p but reverses every geodesic through p. Then we call M a symmetric space. Note that any such M is complete because we can extend geodesics by flipping. Also note that M has a transitive isometry group.

Example 20.3 (Basic Examples).
  1. 1.

    ℝn is always a symmetric space. Here the geodesics are straight lines and the involutions are reflections.

  2. 2.

    Any sphere is a symmetric space.

  3. 3.

    Any hyperbolic space Hn is a symmetric space.

20.1.1 Classification

These spaces were classified by Elie Cartan, which was one of the first major results in the classification of Lie groups and Lie algebras. Here, M is a product of three types of symmetric spaces:

  • •

    Euclidean type (ℝn);

  • •

    Compact type;

  • •

    Noncompact type;

These are characterized by conditions on their sectional curvature (identically zero, nonnegative, nonpositive).

Next, any symmetric space can be factored into irreducible symmetric spaces. Therefore, we simply need to classify the irreducible symmetric spaces. These were classified using algebra, and are all G/K, where G is a connected Lie group and K is a maximal compact subgroup.

First, we classify all possible Lie algebras, then see which Lie groups have a given Lie algebra. Finally, using the theory of root systems and weights, we obtain 8 classical families (A,B,C,D) of symmetric spaces and 12 exceptional examples (E,F,G). Each one comes in a compact version and a noncompact version. In fact, there is a duality between the compact spaces and noncompact spaces.

20.1.2 Basic Examples

  1. 1.

    Let G=S⁢Ln⁢(ℝ) and K=S⁢O⁢(n). Then M=G/K and has dimension (n+2)⁢(n−1)2. The symmetry comes from the involution X↦(X−1)T.

  2. 2.

    Let G=S⁢p2⁢n⁢(ℝ). Then K=U⁡(n) and M=G/K has dimension n⁢(n+1) and is called the Siegel upper half-space. This space arises in the study of moduli spaces of abelian varieties. In fact, this is a Hermitian symmetric space, which means M is a complex manifold.

  3. 3.

    Let G=S⁢L2⁢(ℂ) and K=S⁢U⁢(2). Then G/K=H3, which is hyperbolic 3-space.

  4. 4.

    Now consider G=S⁢L2⁢(ℝ)×S⁢L2⁢(ℝ). Then K=S⁢O⁢(2)×S⁢O⁢(2) and H/K=ℋ×ℋ. This example and the previous one arise when studying algebraic groups over number fields. The upper half-plane corresponds to ℚ, H3 corresponds to imaginary quadratic fields, and ℋ×ℋ corresponds to the real quadratic fields. Quotients of H3 give Bianchi manifolds, quotients of ℋ×ℋ give Hilbert modular surfaces, and quotients of ℋ give modular curves.

20.2 Locally Symmetric Space

Here, we only require involutions that reflect geodesics in neighborhoods of points. However, every locally symmetric space arises as the quotient of a global symmetric space by a discrete subgroup of G.

Example 20.4.

We can consider G=S⁢L2⁢(ℝ) and Γ=S⁢L2⁢(ℤ). Alternatively, we can consider G=S⁢L2⁢(ℂ) and Γ=S⁢L2⁢(K) where K is an imaginary quadratic field. If G=S⁢L2⁢(ℝ)×S⁢L2⁢(ℝ), then we can take Γ=S⁢L2⁢(K), where K is a real quadratic field. Here, Γ acts by γ⋅g=γ⁢g on the first factor and by γ⋅g=γ¯⁢g on the second factor.

We can understand the geometry of these space by interpreting them as configuration spaces. We will focus on the case where G=S⁢Ln⁢(ℝ). Our goal is to understand compactifications of the locally symmetric spaces geometrically. Here, points in Γ\X correspond to lattices in ℝn modulo equivalence relations and with some decorations.

21 April 16

Paul computed some values of the Riemann Zeta function.

21.1 Compactifications of the Space of Lattices

Let G=S⁢Ln⁢(ℝ),K=S⁢O⁢(n),Γ=S⁢Ln⁢(ℤ). Then consider the global symmetric space X=G/K. note that any coset g⁢K is a basis of g obtained by rotation. Therefore G/K corresponds to the set of volume 1 ordered bases of ℝn up to rotation.

Next, consider the locally symmetric space Γ\X. Here, Γ⁢g gives the lattice generated by the rows of g, and thus Γ\X can be identified by the space of lattices in ℝn up to rotation and scaling.

Remark 21.1.

Double coset spaces like Γ\G/K are ubiquitous in geometry. Any double coset space really corresponds to the set of G-orbits on pairs.

Proof.

Let X1=G/H1 and X2=G/H2 for two subgroups H1,H2 of G. Then consider the diagonal action of G by G:X1×X2 diagonally. Then we can show that G-orbits are in bijection with H1\G/H2. □

Example 21.2.

Let G=S⁢Ln⁢(ℂ) and B be the maximal Borel subgroup. Then G/B is the flag variety, and B\G/B encodes the relative positions of two flags in ℂn. The Bruhat decomposition shows that this is actually a finite set, indexed by the Weyl group W.

Example 21.3.

Consider G=S⁢Ln⁢(ℝ) and K=S⁢O⁢(n). Then we can encode X=G/K as the set of volume 1 ellipsoids in ℝn and the points give metrics on ℝn. Then the space K\G/K, which parameterizes ellipsoids of volume 1 up to rotation. By a result of Euler, this is paramaterized by positive real numbers whose product is 1.

Therefore, we can consider Γ\G/K as the set of G orbits in G/Γ×G/K={lattices}×{metrics}. We can consider the space of lattices in the presence of a fixed metric or the space of metrics in the presence of a fixed lattice.

Remarks 21.4.

The space Γ\X is very close to being a manifold. It has singularities, but they are very mild (in fact just quotient singularities), so the space is an orbifold. The singularities come from lattices with nongeneric symmetry (with symmetry that is more than just x↦−x).

  1. 1.

    We can give up and call this space an orbifold, stack, or some other new term, or we can incorporate decorations with the lattices. We will take the second approach later. This corresponds to replacing S⁢Ln⁢(ℤ) with a finite index subgroup.

  2. 2.

    This space is not compact. If we scale the x-axis up and the y-axis down, this sequence of lattices clearly does not converge. In fact, a theorem of Mahler says that a sequence of lattices diverges to infinity if and only if the length of the minimal nonzero vector converges to 0.

21.1.1 Topology and Geometry for Small n

Let n=2. Then this gives lattices in ℝ2 modulo rotation and scaling. We are looking for a fundamental domain for Γ acting on X together with an identification on its boundary. This is known as explicit reduction theory.1

First, we will find a shortest nonzero vector v∈L. If we rotate and scale v onto e1, we can choose the minimal length vector w from the row immediately above the x-axis. Then |w|≥1 and lies in the strip between −12 and 12. This gives the famous picture of the fundamental domain in the upper half plane. The two vertical lines are clearly identified, and then the arc the arc is folded in half. Therefore, the space Γ\X is topologically equivalently to a disc.

22 April 21

Paul computed some values of L-functions of number fields in GP/Pari. He then mentioned the class number formula.

Last time, we discussed the space of lattices modulo rotation and scaling. In addition, we computed the fundamental domain of the action of S⁢L2⁢(ℤ) on the upper half-plane.

22.1 Geometry for n=3

We now consider the space of lattices in ℝ3. We can rescale to make the shortest nonzero vector (1,0,0). Then we can rotate the next shortest vector to lie in the x⁢y-plane, and we can assume that v2 is in the right half of the fundamental domain of S⁢L2. Finally, we choose a v3 lying above the plane. To do this, consider the quotient of ℝ3 by the plane, and we can choose v3 in the inverse image of the shortest nonzero vector in the quotient lattice.

Remark 22.1.

The image of a lattice in a quotient space is not necessarily a lattice.

Next, we consider where we can arrange v3 to be. We can move along the lattice and move v3 to lie over a polygon determined by v1,v2. In addition, v3 cannot lie inside the unit hemisphere. Therefore, our configuration has dimension 5 and should have a deformation retract onto a compact space of dimensnio 3.

First, let’s ignore the noncompact dimensions and focus on the compact part. To see this object, we move v2 to the top of the semicircle and look at the possible polygons. Here, the polygon is the set of points whose distance to the origin is smaller than the distance to any other lattice points. This is called the Dirichlet-Voronoi domain.1 Then this three dimensional space is very complicated, and it is easier to describe a finite version. This is simply a cube with four of the vertices truncated. Note the center of the triangular face is the A1×A1×A1 lattice while the center of a hexagonal face is the A2×A1 lattice.

Figure 22.1: Soulé’s cube. This was used to compute the cohomology of S⁢L3⁢(ℤ).
Exercise 22.2.

Find the three remaining special lattices (names can be found in a crystollagraphy book).

We will not consider the higher-dimensional analogues. The quotients are well-understood up to n=7, but the case n=8 is unknown. In n=8, we have the E8 lattice, which makes the geometry of the polytope extremely complicated.2 We have no idea how many faces there are, but we do have a few crude estimates. In addition, our algorithm for finding a nice basis of our lattices fails in dimension greater than 3. It barely works for n=4, but fails for n=5.

Example 22.3.

Consider the lattice L⊂ℝ5 with basis e1,e2,e3,e4,12⁢(e1+⋯+e5). Note that ℤ5 is a strict subset of L. However, if we run our algorithm, we only obtain the ℤ5 lattice, which is not L.

Remark 22.4.

Even worse, in high dimensions, there exist lattices such that no basis contains a minimal vector (Conway-Sloane).

22.2 Singularities of the space of lattices

We can eliminate these singularities by introducing extra data. We will call these “marked lattices” and our extra data will kill these automorphisms.

Definition 22.5.

A marking on a lattice is a surjective map m:L→S such that there exists a finite index sublattice L′ such that m is fixed by translation by L′.

Example 22.6.

We can mark L=ℤn using S=(ℤ/p⁢ℤ)n by reduction mod p. Here, L′=p⁢ℤn.

Remark 22.7.

We can think of markings as coming from coloring data on our lattices.

We can consider “Γ⁡(p) markings.” Then the space of marked lattices is

where Γi is a finite-index subgroup.

Theorem 22.8.

For p≥3, all Γ⁡(p)-marked lattices have the same automorphisms. Therefore, the moduli space of marked lattices is smooth. The analogous procedure for abelian varieties is taking a principal polarization.

23 April 23

23.1 Compactifications of Locally Symmetric Spaces

Example 23.1.

Consider ℝn. How do we compactify it? How do we add limit points for rays that go off to infinity? If the lines are parallel, do we want them to have the same limit point? This is one of the reasonable ideas, and we obtain a Sn−1 at infinity.

If we forget about rays and just consider lines, we obtain ℙn⁡(ℝ), with a ℙn−1⁡(ℝ) at infinity. In both these examples, we see that points at infinity can be recorded via equivalence relations on geodesics.

Example 23.2.

Let Γ=ℤ/6⁢ℤ act on ℝ2 by rotation. Then note that any cone with angle π/3 is a fundamental domain. How do we compactify this?

Method 1:

We use what we did for ℝ2. We can extend the action to the compactification of ℝ2 (say the S1) at infinity, and we obtain an honest cone.

Method 2:

We now have 6 distinguished rays. We can attempt to use the 1-parameter family of parallel rays for each of these. Now we glue these 6 lines onto ℝ2. We will dictate that the two rays bounding one fundamental domain give lines that intersect at a single point. All rays between the two distinguished rays will converge to that single point. In some sense, this is dual to what we did before and we have a hexagon at infinity.

Why would we consider the second method?

  1. 1.

    We gain information that we lost in the first method.

  2. 2.

    This is what happens when we consider toric varieties.

  3. 3.

    When we consider Γ\X, we need something more like this method.

Now we will consider the space of lattices up to rotation and scaling. Here, our points have a geometric interpretation, so we can try to leverage this. We have two tools to work with:

  1. 1.

    Geodesics in Γ\X. We can think of adding limit points for families of geodesics.

  2. 2.

    In configuration spaces and moduli problems, we want the points at infinity to have a geometric meaning.1 For us, this means that lattices should degenerate to something more general. There are other contexts:

    1. (a)

      Consider the moduli space of genus g curves Mg. When we go to infinity, we pass to stable nodal curves with several components.

    2. (b)

      Consider the moduli space of abelian varieties. These degenerate to various geometric objects.

Now if we consider lattices in ℝ2, consider the family of lattices L⁡(t)=⟨2−t⁢v1,2t⁢v2⟩, this should converge to the x-axis at infinity and the y-axis at −∞. The moral of the story is that lattices degenerate to subspaces. This may not seem too complicated for n=2, but it is very complex when n=3.

Remark 23.3.

This approach does use geodesics! We are moving around in the upper half-plane.

Now we consider the space of lattices in ℝ3. Recall this is isomorphic to a quotient of the Soulé cube times (ℝ≥0)2. If we start with the standard basis vectors, we can consider the family

As t→∞, we simply obtain the x⁢y-plane.

For another example, consider

In the limit at infinity, we simply obtain the x-axis. We can also see “in between” phenomena. Consider the family

We can see that the x-direction collapses faster than the y-direction, so if we rescale by 22⁢t, we see that the limit is a flag 0⊂(y=z=0)⊂(z=0).

Remark 23.4.

We have seen all possible degeneration types. Now we can attempt to look for more refined degeneration data. We observe that the image of the diagonal matrices in S⁢L3⁢(ℝ)/S⁢O⁢(3) is a maximal flat geodesic submanifolds. Therefore, the points at infinity should be parameterized by flags with some extra data.

If we return to the first example L⁡(t)=⟨2−t⁢e1,2−t⁢e2,22⁢t⁢e3⟩. Recall that we obtain the plane as the limit. However, for t≫1, the images of e1,e2 give a lattice in the plane. This gives a point in our space for n=2! If we change e1,e2 to different v1,v2 (say the basis for the A2 lattice), then we obtain a different lattice in our intermediate stage.

24 April 28

Last time we discussed limits of lattices in ℝn. We were able to obtain flags by degenerating lattices. Unfortunately, not every flag can arise as the degeneration of a lattice family.

Example 24.1.

Consider n=2 with X=ℋ. Then suppose Γ=S⁢L2⁢(ℤ). Suppose we head down to α∈ℝ along a geodesic. If α is rational, then we go off to infinity. However, if α is irrational, the picture is much more subtle.

  • •

    If α is contained in a real quadratic field, then the path approaches a periodic geodesic.

  • •

    Otherwise, the path wanders around the quotient and does not escape to infinity. The behavior here is completely unknown.

In terms of flags, only certain flags will arise as limit points, and in fact, we want rational flags. These are flags whose subspaces are generated by lattice points. What about limiting shapes of lattices?

When n=3, we have three types of limiting flags: a plane, a line, and a line inside a plane. When we have a plane, we have a shape inside the limiting plane. In the case of a line, we expect a shape inside the quotient of ℝ3 by the line. Finally, in the case of a full flag, we expect no shape data. More generally, for a flag

we can look at pairwise successive quotients and find lattice shapes there. Putting it all together, we obtain the reductive Borel-Serre compactification. Strata are induced by rational flags in ℝn including the empty flag.

For example, we can add three different types of strata when n=3. We can index them by partitions of 3 where the order of the parts matters.

Table 24.1: Partitions and Strata
Flag Partition Strata
∅ 3 5-dimensional strata
ℝ2 2+1 2-dimensional strata at infinity
ℝ1 1+2 2-dimensional strata at infinity
ℝ1⊂ℝ2 1+1+1 0-dimensional strata at infinity

In fact, the strata at infinity are lower-rank locally symmetric spaces that are being compactified themselves in this process. The space for n=3 looks something like this:

22032112111
Figure 24.1: Simplified picture of strata for n=3

This space is singular. We have attached two-dimensional strata to a five-dimensional space. We can compte links by realizing the space as a quotient of a manifold with corners. The bigger space is called the Borel-Serre compactification.1 There will be a five-dimensional stratum and two four-dimensional strata that meet in a three-dimensional corner. In fact, the four-dimensional strata are fiber bundles over the locally symmetric spaces for n=2.

Consider the case when n=2. Suppose γ1,γ2 are parallel geodesics that go off to infinity. Should they go to different limiting points or the same point? Recall our discussion about compactifying ℝn. There are two answers here:

  1. 1.

    We can send them to the same point because of the metric on the upper half plane. This is because the two geodesics get arbitrarily close together.

  2. 2.

    The geodesics hit different points on various parallel horocycles.2 Thus sending them to different points also makes sense.

To add this data in our lattice/flag description, we add a real transverse flag to the data. The transverse flag corresponds to adding an S1 at infinity. This S1 is really the matrices of the form

In the quotient, this circle is collapsed down to a point.

When n=3, we add flags to the strata corresponding to the compositions 21,12,111. In the Lie groups picture, the three strata correspond to parabolic subgroups. Then the fiber above each point in the two dimensional strata is a torus. Then the three-dimensional stratum is the image of the matrices of the form

and is called a nilmanifold of the Heisenberg manifold.

Exercise 24.2.

Compute the links of points in the strata using the presentation as a space with thin singularities.

Remark 24.3.

It is an open question to generalize these constructions to the nonarchimedean setting. Here, we want to replace real Lie groups with reductive groups over ℚp and K with the same group over ℤp. For example, we replace S⁢Ln⁢(ℝ) with S⁢Ln⁢(ℚp) and K=S⁢O⁢(n) with K=S⁢Ln⁢(ℤp). The analogue of X in this setting is called a building. Compactifications of X in this setting have been studied, but locally summetric spaces have not been studied.