Integrable probability and Macdonald processescorwin/HIM_Lecture1.pdf · Here is an example of a...
Transcript of Integrable probability and Macdonald processescorwin/HIM_Lecture1.pdf · Here is an example of a...
![Page 1: Integrable probability and Macdonald processescorwin/HIM_Lecture1.pdf · Here is an example of a Markov process preserving the class of the q-Whittaker processes (Macdonald processes](https://reader033.fdocuments.us/reader033/viewer/2022050313/5f75a3e1435ddb6ba828ee31/html5/thumbnails/1.jpg)
Integrable probability and
Macdonald processes
Ivan Corwin(Clay Mathematics Institute, Massachusetts Institute of Technology and Microsoft Research)
HIM_Lecture1 Page 1
![Page 2: Integrable probability and Macdonald processescorwin/HIM_Lecture1.pdf · Here is an example of a Markov process preserving the class of the q-Whittaker processes (Macdonald processes](https://reader033.fdocuments.us/reader033/viewer/2022050313/5f75a3e1435ddb6ba828ee31/html5/thumbnails/2.jpg)
Concise and exact formulas for expectations of rich
class of interesting observables.
1.
Scaling limits of systems and formulas provide access
to exact descriptions of large universality classes of
physical and mathematical systems.
2.
Integrable probabilistic systems have two characteristics:
Focus on the Kardar-Parisi-Zhang universality class
where representation theory (Macdonald symmetric
functions) serves as a significant source of integrable
probabilistic systems
HIM_Lecture1 Page 2
![Page 3: Integrable probability and Macdonald processescorwin/HIM_Lecture1.pdf · Here is an example of a Markov process preserving the class of the q-Whittaker processes (Macdonald processes](https://reader033.fdocuments.us/reader033/viewer/2022050313/5f75a3e1435ddb6ba828ee31/html5/thumbnails/3.jpg)
Totally asymmetric simple exclusion process (TASEP)
Two (1-parameter) deformations:
Rate 1
ASEP [Spitzer '70] q-TASEP [Borodin-C '11]
RateRate q Rate p
Step initial data
Object of study:
Height above site x or equivalently,
current of particles to pass site x.
p+q=1
p-q=
q/p= 0<q<1
(simulation courtesy of Patirk Ferrari)
HIM_Lecture1 Page 3
![Page 4: Integrable probability and Macdonald processescorwin/HIM_Lecture1.pdf · Here is an example of a Markov process preserving the class of the q-Whittaker processes (Macdonald processes](https://reader033.fdocuments.us/reader033/viewer/2022050313/5f75a3e1435ddb6ba828ee31/html5/thumbnails/4.jpg)
Local growth1.
Smoothing mechanism2.
Slope dependent growth rate3.
Independent space-time noise4.
Key properties of models in the KPZ universality class
Many probabilistic/physical systems share these features
HIM_Lecture1 Page 4
![Page 5: Integrable probability and Macdonald processescorwin/HIM_Lecture1.pdf · Here is an example of a Markov process preserving the class of the q-Whittaker processes (Macdonald processes](https://reader033.fdocuments.us/reader033/viewer/2022050313/5f75a3e1435ddb6ba828ee31/html5/thumbnails/5.jpg)
Kardar-Parisi-Zhang equation '86 in 1+1 dimension:
Hopf-Cole solution [Bertini-Cancrini '95, Bertini-Giacomin '97]
Define: where solves the (well-posed)
multiplicative stochastic heat equation (SHE)
(standard scalings: )
HIM_Lecture1 Page 5
![Page 6: Integrable probability and Macdonald processescorwin/HIM_Lecture1.pdf · Here is an example of a Markov process preserving the class of the q-Whittaker processes (Macdonald processes](https://reader033.fdocuments.us/reader033/viewer/2022050313/5f75a3e1435ddb6ba828ee31/html5/thumbnails/6.jpg)
Two (weak) scalings preserve the KPZ equation
Rescaling the KPZ equation:
KPZ scaling: [Forster-Nelson-Stephen '77]
"KPZ fixed point" is universal limit process
Weak nonlinearity scaling: , scale nonlinearity by
Weak noise scaling: , scale noise by
Useful proxies for finding approximation schemes HIM_Lecture1 Page 6
![Page 7: Integrable probability and Macdonald processescorwin/HIM_Lecture1.pdf · Here is an example of a Markov process preserving the class of the q-Whittaker processes (Macdonald processes](https://reader033.fdocuments.us/reader033/viewer/2022050313/5f75a3e1435ddb6ba828ee31/html5/thumbnails/7.jpg)
1+1 dimensional
semi-discrete
and discrete SHERate q Rate p
ASEP
q-TASEP
Rate
TASEP
Directed polymers
KPZ equation
KPZ fixed point
HIM_Lecture1 Page 7
![Page 8: Integrable probability and Macdonald processescorwin/HIM_Lecture1.pdf · Here is an example of a Markov process preserving the class of the q-Whittaker processes (Macdonald processes](https://reader033.fdocuments.us/reader033/viewer/2022050313/5f75a3e1435ddb6ba828ee31/html5/thumbnails/8.jpg)
Consider TASEP with step initial data
Theorem (Johansson '99): For TASEP with step initial data,
See also [Baik-Deift-Johansson '99, Prahofer-Spohn '02]
Source of integrability: determinantal structure of Schur
measure and process [Okounkov-Reshetikhin '03, Borodin-Ferrari '08]
tight [Cator-C-Pimentel-Quastel '12]->KPZ fixed point exists HIM_Lecture1 Page 8
![Page 9: Integrable probability and Macdonald processescorwin/HIM_Lecture1.pdf · Here is an example of a Markov process preserving the class of the q-Whittaker processes (Macdonald processes](https://reader033.fdocuments.us/reader033/viewer/2022050313/5f75a3e1435ddb6ba828ee31/html5/thumbnails/9.jpg)
TASEP is one of a few growth models in the KPZ class that can
be analyzed via the techniques of determinantal point processes
(or free fermions, nonintersecting paths, Schur processes).
Other examples include
Discrete time TASEPs with sequential/parallel update
PushASEP or long range TASEP
Directed last passage percolation in 2d with
geometric/Bernoulli/exponential weights
Polynuclear growth processes
HIM_Lecture1 Page 9
![Page 10: Integrable probability and Macdonald processescorwin/HIM_Lecture1.pdf · Here is an example of a Markov process preserving the class of the q-Whittaker processes (Macdonald processes](https://reader033.fdocuments.us/reader033/viewer/2022050313/5f75a3e1435ddb6ba828ee31/html5/thumbnails/10.jpg)
Strong experimental evidence that real life systems follow1.
the KPZ class universal laws
Direct well-posedness of the KPZ equation and some weak
universality of the equation
2.
Non-determinantal models whose large time behaviour has
been analyzed
3.
Recent advances on the KPZ front:
HIM_Lecture1 Page 10
![Page 11: Integrable probability and Macdonald processescorwin/HIM_Lecture1.pdf · Here is an example of a Markov process preserving the class of the q-Whittaker processes (Macdonald processes](https://reader033.fdocuments.us/reader033/viewer/2022050313/5f75a3e1435ddb6ba828ee31/html5/thumbnails/11.jpg)
ASEP [Tracy-Widom, 2009], [Borodin-C-Sasamoto, 2012]
KPZ equation / stochastic heat equation (SHE)
[Amir-C-Quastel, 2010], [Sasamoto-Spohn, 2010], [Dotsenko, 2010+],
[Calabrese-Le Doussal-Rosso, 2010+], [Borodin-C-Ferrari, 2012]
q-TASEP [Borodin-C, 2011+]
Semi-discrete stochastic heat equation
[O'Connell, 2010], [Borodin-C, 2011, Borodin-C-Ferrari, 2012]
Fully discrete log-Gamma polymer (stochastic heat equation)
[C-O'Connell-Seppalainen-Zygouras, 2011] [Borodin-C-Remenik, 2012]
Non-determinantal models whose large time behaviour has
been analyzed:
HIM_Lecture1 Page 11
![Page 12: Integrable probability and Macdonald processescorwin/HIM_Lecture1.pdf · Here is an example of a Markov process preserving the class of the q-Whittaker processes (Macdonald processes](https://reader033.fdocuments.us/reader033/viewer/2022050313/5f75a3e1435ddb6ba828ee31/html5/thumbnails/12.jpg)
Theorem (Borodin-C '11): q-TASEP step initial data
q-Laplace transform [Hahn '49] identifies distribution
Good for asymptotics [Borodin-C-Ferrari '12]
where
q<1
HIM_Lecture1 Page 12
![Page 13: Integrable probability and Macdonald processescorwin/HIM_Lecture1.pdf · Here is an example of a Markov process preserving the class of the q-Whittaker processes (Macdonald processes](https://reader033.fdocuments.us/reader033/viewer/2022050313/5f75a3e1435ddb6ba828ee31/html5/thumbnails/13.jpg)
Theorem (Borodin-C-Sasamoto '11): ASEP step initial data
Both q-TASEP and ASEP have second type of Fredholm
determinant formula (harder for asymptotics).
For ASEP, second formula matches [Tracy-Widom '09].
where same as upto
p-q=
q/p=
HIM_Lecture1 Page 13
![Page 14: Integrable probability and Macdonald processescorwin/HIM_Lecture1.pdf · Here is an example of a Markov process preserving the class of the q-Whittaker processes (Macdonald processes](https://reader033.fdocuments.us/reader033/viewer/2022050313/5f75a3e1435ddb6ba828ee31/html5/thumbnails/14.jpg)
Discrete time q-TASEPs
q-TASEP log-Gamma discrete
polymer
semi-discrete stochastic
heat eqn.
KPZ equation / stochastic heat equation
universal limits (Tracy-Widom distributions, Airy processes)
ASEP
HIM_Lecture1 Page 14
![Page 15: Integrable probability and Macdonald processescorwin/HIM_Lecture1.pdf · Here is an example of a Markov process preserving the class of the q-Whittaker processes (Macdonald processes](https://reader033.fdocuments.us/reader033/viewer/2022050313/5f75a3e1435ddb6ba828ee31/html5/thumbnails/15.jpg)
q-TASEP:
Semi-discrete SHE:
[Alberts-Khanin-Quastel '12],
[Moreno-Remenik-Quastel '12]
[Borodin-C '11,
Borodin-C-Sasamoto '12]
Continuum SHE:
ASEP:
[Bertini-Giacomin '97],
[Gartner '88]
HIM_Lecture1 Page 15
![Page 16: Integrable probability and Macdonald processescorwin/HIM_Lecture1.pdf · Here is an example of a Markov process preserving the class of the q-Whittaker processes (Macdonald processes](https://reader033.fdocuments.us/reader033/viewer/2022050313/5f75a3e1435ddb6ba828ee31/html5/thumbnails/16.jpg)
Theorem (Amir-C-Quastel '10): Let then
Formula discovered independently and in parallel in
non-rigorous work of [Sasamoto-Spohn '10].
Corollary: KPZ equation is in KPZ universality class.
Let with
Formally h solves the KPZ equation
HIM_Lecture1 Page 16
![Page 17: Integrable probability and Macdonald processescorwin/HIM_Lecture1.pdf · Here is an example of a Markov process preserving the class of the q-Whittaker processes (Macdonald processes](https://reader033.fdocuments.us/reader033/viewer/2022050313/5f75a3e1435ddb6ba828ee31/html5/thumbnails/17.jpg)
Scaled one point marginal distribution for KPZ equation
HIM_Lecture1 Page 17
![Page 18: Integrable probability and Macdonald processescorwin/HIM_Lecture1.pdf · Here is an example of a Markov process preserving the class of the q-Whittaker processes (Macdonald processes](https://reader033.fdocuments.us/reader033/viewer/2022050313/5f75a3e1435ddb6ba828ee31/html5/thumbnails/18.jpg)
Macdonald processesRuijsenaars-Macdonald system
Representations of Double Affine Hecke Algebras
Hall-Littlewood processesRandom matrices over finite fields
Spherical functions for p-adic groups
GeneralRandom matrices over
Calogero-Sutherland, Jack polynomials
Spherical functions for Riem. Symm. Sp.
RMT
q-Whittaker processesq-TASEP, 2d dynamics
q-deformed quantum Toda lattice
Representations of
Whittaker processesDirected polymers and their hierarchies
Quantum Toda lattice, repr. of
Schur processesPlane partitions, tilings/shuffling, TASEP, PNG, last passage percolation, GUE
Characters of symmetric, unitary groups
Kingman partition structuresCycles of random permutations
Poisson-Dirichlet distributions
HIM_Lecture1 Page 18
![Page 19: Integrable probability and Macdonald processescorwin/HIM_Lecture1.pdf · Here is an example of a Markov process preserving the class of the q-Whittaker processes (Macdonald processes](https://reader033.fdocuments.us/reader033/viewer/2022050313/5f75a3e1435ddb6ba828ee31/html5/thumbnails/19.jpg)
Macdonald processes: a source of (many parameter) integrable
probabilistic systems. Specializations and degenerations include
q-TASEP, continuum/semi-discrete/discrete SHE, KPZ equation
ASEP does not fit. But it does share certain parallel formulas
Structure
leads to
Formulas
suggests new
HIM_Lecture1 Page 19
![Page 20: Integrable probability and Macdonald processescorwin/HIM_Lecture1.pdf · Here is an example of a Markov process preserving the class of the q-Whittaker processes (Macdonald processes](https://reader033.fdocuments.us/reader033/viewer/2022050313/5f75a3e1435ddb6ba828ee31/html5/thumbnails/20.jpg)
(Ascending) Macdonald processes are probability measures on
interlacing triangular arrays (Gelfand-Tsetlin patterns)
HIM_Lecture1 Page 20
![Page 21: Integrable probability and Macdonald processescorwin/HIM_Lecture1.pdf · Here is an example of a Markov process preserving the class of the q-Whittaker processes (Macdonald processes](https://reader033.fdocuments.us/reader033/viewer/2022050313/5f75a3e1435ddb6ba828ee31/html5/thumbnails/21.jpg)
Macdonald polynomials
with partitions form a basis in symmetric
polynomials in N variables over They diagonalize
with (generically) pairwise different eigenvalues
They have many remarkable properties that include orthogonality
(dual basis ), simple reproducing kernel (Cauchy type identity),
Pieri and branching rules, index/variable duality, explicit generators
of the algebra of (Macdonald) operators commuting with , etc. HIM_Lecture1 Page 21
![Page 22: Integrable probability and Macdonald processescorwin/HIM_Lecture1.pdf · Here is an example of a Markov process preserving the class of the q-Whittaker processes (Macdonald processes](https://reader033.fdocuments.us/reader033/viewer/2022050313/5f75a3e1435ddb6ba828ee31/html5/thumbnails/22.jpg)
We are able to do two basic things:
•Construct relatively explicit Markov operators that map Macdonald
processes to Macdonald processes;
Evaluate averages of a broad class of observables.•
The construction is based on commutativity of Markov operators
an idea from [Diaconis-Fill '90], and Schur process dynamics
from [Borodin-Ferrari '08].
HIM_Lecture1 Page 22
![Page 23: Integrable probability and Macdonald processescorwin/HIM_Lecture1.pdf · Here is an example of a Markov process preserving the class of the q-Whittaker processes (Macdonald processes](https://reader033.fdocuments.us/reader033/viewer/2022050313/5f75a3e1435ddb6ba828ee31/html5/thumbnails/23.jpg)
Evaluation of averages is based on the following observation.
Let be an operator that is diagonalized by the Macdonald
polynomials (for example, a product of Macdonald operators),
Applying it to the Cauchy type identitywe obtain
If all the ingredients are explicit (as for products of Macdonald
operators), we obtain meaningful probabilistic information. Contrast
with the lack of explicit formulas for the Macdonald polynomials.
HIM_Lecture1 Page 23
![Page 24: Integrable probability and Macdonald processescorwin/HIM_Lecture1.pdf · Here is an example of a Markov process preserving the class of the q-Whittaker processes (Macdonald processes](https://reader033.fdocuments.us/reader033/viewer/2022050313/5f75a3e1435ddb6ba828ee31/html5/thumbnails/24.jpg)
Here is an example of a Markov process preserving the class of the
q-Whittaker processes (Macdonald processes with t=0).
Each coordinate of the triangular
array jumps by 1 to the right
independently of the others with
The set of coordinates forms q-TASEP
Rate
HIM_Lecture1 Page 24
![Page 25: Integrable probability and Macdonald processescorwin/HIM_Lecture1.pdf · Here is an example of a Markov process preserving the class of the q-Whittaker processes (Macdonald processes](https://reader033.fdocuments.us/reader033/viewer/2022050313/5f75a3e1435ddb6ba828ee31/html5/thumbnails/25.jpg)
Taking the observables corresponding to powers of the first
Macdonald operator yields
with
A rigorous version of the physics "replica trick" HIM_Lecture1 Page 25
![Page 26: Integrable probability and Macdonald processescorwin/HIM_Lecture1.pdf · Here is an example of a Markov process preserving the class of the q-Whittaker processes (Macdonald processes](https://reader033.fdocuments.us/reader033/viewer/2022050313/5f75a3e1435ddb6ba828ee31/html5/thumbnails/26.jpg)
q-TASEP Semi-discrete SHE Continuum SHE
[Molchanov '86] [Kardar '87] observe
satisfies
Bethe ansatz [c<0 Lieb-Liniger '63, c>0 McGuire '64] gives eigenbasis
"Replica trick" [Dotsenko '10, Calabrese-Le Doussal-Rosso '10]
Divergent series! Risky to draw conclusions (originally obtained incorrect answer)
HIM_Lecture1 Page 26
![Page 27: Integrable probability and Macdonald processescorwin/HIM_Lecture1.pdf · Here is an example of a Markov process preserving the class of the q-Whittaker processes (Macdonald processes](https://reader033.fdocuments.us/reader033/viewer/2022050313/5f75a3e1435ddb6ba828ee31/html5/thumbnails/27.jpg)
Theorem (Borodin-Corwin, '11): For the integral
(Here the -integration is over with )
solves the delta Bose gas for all and for
Clear symmetry between attractive (c>0) and repulsive (c<0) cases•
Bethe eigenstates are very different in attractive/repulsive cases•
Formula can be found in [Heckman-Opdam '97] Plancherel theorem
for delta Bose gas; ideas trace back to [Harish Chandra, Helgason]
•
HIM_Lecture1 Page 27
![Page 28: Integrable probability and Macdonald processescorwin/HIM_Lecture1.pdf · Here is an example of a Markov process preserving the class of the q-Whittaker processes (Macdonald processes](https://reader033.fdocuments.us/reader033/viewer/2022050313/5f75a3e1435ddb6ba828ee31/html5/thumbnails/28.jpg)
For semi-discrete SHE, satisfies [Borodin-C '11]
In all cases, the "nested contour integral ansatz" solves Bose gas
For q-TASEP, satisfies [Borodin-C-Sasamoto '12]
HIM_Lecture1 Page 28
![Page 29: Integrable probability and Macdonald processescorwin/HIM_Lecture1.pdf · Here is an example of a Markov process preserving the class of the q-Whittaker processes (Macdonald processes](https://reader033.fdocuments.us/reader033/viewer/2022050313/5f75a3e1435ddb6ba828ee31/html5/thumbnails/29.jpg)
Self-duality of ASEP [Schutz '97] -> moments satisfy Bose gas•
Nested contour integral ansatz applies [Borodin-C-Sasamoto '12]•
Leads to two Fredholm determinants (one new and one TWs)•
TW compute ASEP k-particle Green's function via Bethe ansatz•
ASEP is not solved by Macdonald process. However,
For q-TASEP: Nested contour integral formulas and Bose gas are
consequences of structural properties of the Macdonald polynomials
•
For ASEP: No structure to predict existence of nested contour
integral formulas (duality is from symmetry)
•
Formulas suggest search for new structure:
HIM_Lecture1 Page 29
![Page 30: Integrable probability and Macdonald processescorwin/HIM_Lecture1.pdf · Here is an example of a Markov process preserving the class of the q-Whittaker processes (Macdonald processes](https://reader033.fdocuments.us/reader033/viewer/2022050313/5f75a3e1435ddb6ba828ee31/html5/thumbnails/30.jpg)
ASEP and q-TASEP are important systems in the KPZ
universality class, which can be scaled to the KPZ equation
•
Macdonald processes are a source of integrable probabilistic models•
Generalize Schur processes but are not determinantal•
Integrability from structural properties of Macdonald polynomials
(lead to nice Markov dynamics and concise formulas for averages)
•
Turning averages into asymptotics remains challenging•
Rigorous replica trick developed for q-TASEP and ASEP •
Nested contour integral ansatz formulas for ASEP moments
suggest search for new structure parallel to Macdonald processes
•
To summarize:
HIM_Lecture1 Page 30
![Page 31: Integrable probability and Macdonald processescorwin/HIM_Lecture1.pdf · Here is an example of a Markov process preserving the class of the q-Whittaker processes (Macdonald processes](https://reader033.fdocuments.us/reader033/viewer/2022050313/5f75a3e1435ddb6ba828ee31/html5/thumbnails/31.jpg)
Lecture 1: Overview and intro to symmetric functions.
Lecture 2: Schur processes
Lecture 3: Macdonald processes I
Lecture 4: Macdonald processes II
Lecture 5: Duality and Bose gas methods
Lecture 6: Analysis of ASEP, conjectures and open problems
Exercise handout and office hours (Wed., Thur. 3-5pm 3.040)
Lectures times Wed. 10-12 and Thur. 9-11.
Website: http://math.mit.edu/~icorwin/Lipschitz.html
HIM_Lecture1 Page 31