Skip to content

matthieugomez/InfinitesimalGenerators.jl

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

296 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Build status

This package provides a set of tools to work with Markov Processes defined on a 1-dimensional grid.

Markov Processes

The package allows you to compute expectations involving Markov processes.

using InfinitesimalGenerators

# Create a diffusion process (here, the Ornstein-Uhlenbeck dx = -0.03 * x * dt + 0.01 * dZ_t)
# Note that the package assumes reflecting boundaries at the limits
x = range(-1, 1, length = 100)
μx = .- 0.03 .* x
σx = 0.01 .* ones(length(x))
X = DiffusionProcess(x, μx, σx)

# Convenience constructors are also available:
# X = OrnsteinUhlenbeck(; xbar = 0.0, κ = 0.03, σ = 0.01)
# X = CoxIngersollRoss(; xbar = 0.1, κ = 0.1, σ = 1.0)

# Return its stationary distribution
g = stationary_distribution(X)

# Return the associated generator as a matrix (i.e. the operator `f -> ∂_tE[f(x_t)|x_0=x]`)
MX = generator(X)

# Use the generator to compute E[∫_0^T e^{-∫_0^t v(x_s)ds}f(x_t)dt + e^{-∫_0^T v(x_s)ds}ψ(x_T) | x_0 = x]
feynman_kac(MX, range(0, 100, step = 1/12); f = zeros(length(x)), ψ = ones(length(x)), v = zeros(length(x)))

Additive Functionals

Given a Markov process X, an additive functional m is defined by dm = μm(x) dt + σm(x) dZm with corr(dZm, dZ) = ρ.

# Create an additive functional with drift μm and volatility σm
m = AdditiveFunctionalDiffusion(X, μm, σm; ρ = 0.0)

# Return its associated generator (i.e. the operator `f -> ∂_tE[e^{m}f(x_t)|x_0=x]`)
generator(m)

# Return the long run scaled CGF of m, i.e. ξ -> lim_{t→∞} log(E[e^{ξ m_t}])/t
cgf(m)(1.0)

# Return the tail index of the stationary distribution of e^m
tail_index(m)

Derivatives

The package also allows you to compute (lazy) first and second derivatives of a function on a grid using finite difference schemes.

using InfinitesimalGenerators
x = range(-1, 1, length = 100)
f = sin.(x)
FirstDerivative(x, f; direction = :forward, bc = (0, 0))
FirstDerivative(x, f; direction = :backward, bc = (0, 0))
SecondDerivative(x, f; bc = (0, 0))

The argument bc refers to the value of the first derivative at each limit of the grid. This argument defaults to zero, which is the right condition when solving problems with reflecting boundaries.

Joint Operator

For coupled Markov processes switching between N regimes, jointoperator combines the individual generators with a transition matrix.

Q = [-0.1 0.1; 0.2 -0.2]  # regime transition matrix
J = jointoperator([generator(X1), generator(X2)], Q)

Related Packages

  • SimpleDifferentialOperators contains more general tools to define operators with different boundary conditions. In contrast, InfinitesimalGenerators always assumes reflecting boundaries.

About

A set of tools to work with Markov Processes

Resources

License

Stars

Watchers

Forks

Packages

 
 
 

Contributors

Languages