Table of Contents
Course plan
Lecture 1. Introduction, motivation and course outline
Historical motivation (non-perturbative string theory description) along the lines of [FGZ95, pp.4–12]. Modern development according to [Mor25].
Lecture 2. Feynman graphs in QFT
Feynman graphs, Wick theorem, asymptotic series, connected and disconnected diagrams [Ski18, pp.13–26].
Lecture 3. Ribbon Feynman graphs. QCD. Large limit
Ribbon Feynman graphs — [FGZ95, sec.1.2]. To understand the connection with fancy, though accurate eq. (2.39) and (2.40) from [Ski18] (Lecture 2) see also [EKR15, sec.2.1], especially eqns. (2.1.20) and (2.1.21). QCD, large limit and planar diagrams — [Ton18, pp.305–313].
Lecture 4. Eigenvalue matrix models. Leading behaviour
You'll learn quick and dirty hack of obtaining leading behaviour of one-point correlators (as well as directly related one-point eigenvalue distributions). Main reference — [Zee10, pp.396–402]. For more convincing Vandermonde determinant derivations see footnote 8 in [FGZ95], as well as Appendix C.
Just in case of questions about the so-called Haar measure, see Example 5 here.
For more convincing Coloumb gas method explanations see also [FGZ95, sec.2.1].
Lecture 5. Loop equations. Subleading behaviour
Subleading behaviour can be obtained by straightforward expansion around saddle point from previous lecture. The recipe is as follows: expand the exponent around equillibrium . You'll have something like
so will be the propagator and all the rest — vertices. Next, you can calculate all the correlators using this Feyman diagrams technique. This procedure is quite tedious and even if you are patient enough to calculate first few orders, you may notice the beautiful pattern behind, which can be discovered independently and which is the subject of the current lecture.
Main goal is to understand [Mor25, sec.2.2]. It can be done with the help of [Mor94, sec.2.2], elementary matrix calculus [PP12] could also be of use. As a glimpse into the realm of AMM-CEO topological recursion describe the recursive procedure for obtaining all the correlators of the matrix model with the generating function given as sum over Young diagrams
by substituting it in the Virasoro constraints. For a result obtained from somewhat different persperctive see eq. (4.1.10) in [EKR15].
Lecture 6. Origin of loop equations. QCD
Lecture 7. -representation
The subject of the lecture is [Mor25, sec.2.3]. For the origin of this approach, as well as the notion of and matrix calculus derivation, see [MS09]. Grading and dilatation operator are also explained in my notes.
Lecture 8. Integrability
We need to understand [Mor25, sec.2.4]. For details consult [Mir94, secs.4.2, 4.5].
Lectures 9–11. Superintegrability
[MO22, sec.2] is the way to go. Also consult my note.
Lecture 12. Multi-matrix models. Ising model on a random 2 surface
See [Kaz86] and [EKR15, sec.2.2.1].
Lecture 13. Matrix model in external field. Kontsevich model
[Kon92, sec.4]: Feynman rules for cubic model in external field, its determinant formula and proof that it's a -function. [Mor94, sec.5]: Double scaling limit of one matrix model equals to Kontsevich one. [Wit90]: motivation.
Lecture 14. Unitary models. Itzykson-Zuber integral. Large 2 lattice QCD
See [IZ80]. Character expansion of the Itzykson-Zuber integral is also explained in my note. For determinant formula consult either paper above, or Terry Tao's blog post. Uniform model is described in works [BG80; GW80] and used for description of large 2 lattice QCD. Itzykson-Zuber integral is the "building block" of Kazakov-Migdal lattice -dimensional QCD [KM93].
Problem set
Problems 1-7. (Matrix) models
For the Gaussian Hermitian matrix model calculate the generating function up to the grading 4 via
- Ward identities
- -representation
- Wick's theorem
- Graphs/triangulations counting
- Determinantal representation
- Superintegrability:
- as a sum over fixed-point-free involutions of symmetric group
- as a sum of Schur functions.
Problem 8. Connected vs. disconnected
Find the expression for the GHMM generating function of cumulants up to 8-th order (the least order, where disconnected diagrams appear in ). Check that correlators obtained from this generating function actually count connected only ribbon graphs.
Problem 9. Two cuts
For the quartic model
find the large eigenvalue distribution .
Grading
Both talks and HW are graded on 5-point scale at first. Talks grade is just an average of individual talks grades. Homework grade is calculated as follows
Next, raw grade is calculated
and converted in final grade on 10-point scale
This function almost perfectly maps unsatisfactory, satisfactory, good, and excellent grades from one scale to another.
| Аня | Радомир | Саша | Юля | Вася Пупкин | |
| Talk 1 | 5 | 5 | 5 | 5 | 5 |
| Talk 2 | 5 | 5 | 5 | 5 | 5 |
| Talk 3 | 5 | 5 | 5 | 5 | |
| Problem 1 | 1 | 1 | |||
| Problem 2 | 1 | 1 | |||
| Problem 3 | 1 | 1 | |||
| Problem 4 | 1 | ||||
| Problem 5 | 1 | ||||
| Problem 6 | 1 | 1 | |||
| Problem 7 | 1 | ||||
| Problem 8 | 1 | ||||
| Problem 9 | 1 | ||||
| Grade | 5 | 2 | 3 | 3 | 10 |