(co)homology

References

Summary

ker dn / im dn+1

Differential Operators

If we view d as a kind of linear differential operator, or a system of equations, then

The homology then tells you about the non-integrable solutions.

Example: Electromagnetism

In EM, Maxwell's equations tell us ∇ċ B = 0 So the magnetic field is a cycle. But when doing calculations, we often want to integrate the magnetic field, in order to compute things about it; so we need B to be integrable

This means that we need B = ∇ × A This says that there is a magnetic potential A, that generates the actual magnetic field.

In fact, looking at the de Rham complex, Helmholtz's theorem says that for RN, the (co)homology at the level of the magnetic field is zero (every magnetic field satisfying Maxwell's equations is integrable).

  1. Aharonov-Bohm Effect

    When you puncture R, you get non-trivial cohomology

Cohomology

Coboundaries are flat or path-independent or conservative

Using Poincaré duality, we have that <d f , γ> = <f, ∂ γ> where d is the codifferential, and ∂ is the differential, and the pairing is integrating the function (cochain) over the space (chain).

A coboundary is g = df. So for coboundaries <g, γ> = <f, ∂γ> So g is determined by just the boundary (paths through the bulk are irrelevant).

A cocycle will satisfy <f, ∂ γ> = <0, γ> = 0 for all γ. So it has no information on the boundary

But think about what this means: when we take two half-coverings, we're just taking two pieces of the boundary, and saying they stitch together to form the whole boundary (they don't literally need to have the same size, any 2-block partition will work). So there's a glue-line that you're stitching the coverings together on.

But no matter how you move, stretch, contract, etc. the covering pieces, the fact that they come together at the glue-line means they enclose a bulk — they're the boundary of something. So the integral of each piece will always be the same! So if you take just one of the two coverings, no matter how you move it around, its integral will never change, as long as you keep the glue-line fixed in place.

By the way, you need to keep the glue-line the same, because if you don't one will be bigger than the other, and so you can't stitch the two versions together to enclose a bulk.

But note that here, this only works when we're integrating over ∂ γ. So by assumption there's a bulk, γ. In other words, a cocycle is something that, if you have a bulk, becomes path independent on the boundary.

Cocycles can suck the information from the boundary into the bulk. The boundary data is trivial; all that matters is what's in the bulk

Mapping Cones

A mapping cone of a function f: X → Y glues a cone-version of X onto Y, along f. You first form the mapping cylinder X × I, and then glue an end on Y, and then collapse the other end to a point.

Our perspective on cocycles then tells us that a cocycle's integral won't change, as you deform X, as long as you keep f fixed.

Cohomology

Cohomology is then very interesting: think of the mapping cone, or half a sphere. Being a cocycle for the hemisphere means you can deform the hemisphere and nothing changes.

But that hemisphere is actually the bulk for the gluing line! Which means that deforming the hemisphere also won't change coboundaries on the gluing line.

Consider the pair of pants cobordism M. We break it into 3 parts:

  1. the 1d rims Γ = ∂ Σ
  2. the 2d surface Σ = ∂ M
  3. the 3d bulk M

What we found above is that the cocycles and coboundaries don't notice the shape of the cobordism at all, other than the fact that it has a certain number of rims.

Chern Classes

The first Chern class is a 2-cocycle So it's something that has no boundary data

Topological String Theory

In the A-model you basically work with a pair of pants where you collapse all three of the rims to points

The theory is topological in the sense that d ω = 0 (ω is a cycle), for an appropriate d

The idea of quantum cohomology is that you study maps f: S2 → X that hit three different cycles in the manifold X (the collapsed rims of the pants)

You think of the map f as a kind of non-local communication between the cycles (strings)

Definitions

Homology

Let C• be a chain complex $$ \ldots \rightarrow C_{n}_{+}_{1} \xrightarrow{d_{n+1}} C_{n} \rightarrow \ldots $$

The cycles are defined to be Zn = ker dn

The boundaries are Bn = im dn+1

The homology is Hn = Zn/Bn = ker dn/im dn+1.

Change of coefficients

Theorems

Universal coefficient theorem

de Rham's Theorem

Hde Rhamk(M) ≅ Hsingulark(M;ℝ)

Poincaré duality

Let M be a smooth, compact, n-dimensional manifold. Then

Hk(M) ≅ Hn − k(M)

Singularities: In the case of singularities, PD breaks, and we need to replace it with intersection homology

Eilenberg-MacLane spaces

Hgroupk(G) ≅ Hsingulark(K(G,1))

Examples

Simplicial

Probes: Simplices. We perform a triangulation of our space

Loop quantum gravity: In spin foam models, we discretize spacetime into a simplicial complex.

Ponzano-Regge models:

Turaev-Viro TQFT: 3d quantum gravity has no local degrees of freedom; so you get a tqft.

Definition

Let X be a simplicial complex and Ck(X) be the free module generated by the k-simplices in X.

Singular

Probes: Continuous maps of simplices

de Rham

Probes: Smooth differential forms

Sheaf

Probes: Sections of a sheaf

Our coefficients are a sheaf called the coefficient system. The constant sheaf ℝ recovers singular cohomology.

Intersection

Probes: Chain complexes

The goal is to restore Poincaré duality to spaces with singularities

String theory: This is relevant in string theory compactifications

Group

Probes:

Measures group actions

Lie algebra

Probes:

Morse

Elliptic

Quantum Elliptic

Big quantum

Frobenius manifold: The cohomology becomes a Frobenius manifold (a manifold where the tangent spaces are all Frobenius algebras)

Quantum

Correlation functions: This is related to the correlation functions of the A-model

Closed strings:

Floer

Open strings:

Equivariant