458 lines
15 KiB
Python
458 lines
15 KiB
Python
|
from sympy.concrete.products import Product
|
||
|
from sympy.concrete.summations import Sum
|
||
|
from sympy.core.basic import Basic
|
||
|
from sympy.core.function import Lambda
|
||
|
from sympy.core.numbers import (I, pi)
|
||
|
from sympy.core.singleton import S
|
||
|
from sympy.core.symbol import Dummy
|
||
|
from sympy.functions.elementary.complexes import Abs
|
||
|
from sympy.functions.elementary.exponential import exp
|
||
|
from sympy.functions.special.gamma_functions import gamma
|
||
|
from sympy.integrals.integrals import Integral
|
||
|
from sympy.matrices.expressions.matexpr import MatrixSymbol
|
||
|
from sympy.matrices.expressions.trace import Trace
|
||
|
from sympy.tensor.indexed import IndexedBase
|
||
|
from sympy.core.sympify import _sympify
|
||
|
from sympy.stats.rv import _symbol_converter, Density, RandomMatrixSymbol, is_random
|
||
|
from sympy.stats.joint_rv_types import JointDistributionHandmade
|
||
|
from sympy.stats.random_matrix import RandomMatrixPSpace
|
||
|
from sympy.tensor.array import ArrayComprehension
|
||
|
|
||
|
__all__ = [
|
||
|
'CircularEnsemble',
|
||
|
'CircularUnitaryEnsemble',
|
||
|
'CircularOrthogonalEnsemble',
|
||
|
'CircularSymplecticEnsemble',
|
||
|
'GaussianEnsemble',
|
||
|
'GaussianUnitaryEnsemble',
|
||
|
'GaussianOrthogonalEnsemble',
|
||
|
'GaussianSymplecticEnsemble',
|
||
|
'joint_eigen_distribution',
|
||
|
'JointEigenDistribution',
|
||
|
'level_spacing_distribution'
|
||
|
]
|
||
|
|
||
|
@is_random.register(RandomMatrixSymbol)
|
||
|
def _(x):
|
||
|
return True
|
||
|
|
||
|
|
||
|
class RandomMatrixEnsembleModel(Basic):
|
||
|
"""
|
||
|
Base class for random matrix ensembles.
|
||
|
It acts as an umbrella and contains
|
||
|
the methods common to all the ensembles
|
||
|
defined in sympy.stats.random_matrix_models.
|
||
|
"""
|
||
|
def __new__(cls, sym, dim=None):
|
||
|
sym, dim = _symbol_converter(sym), _sympify(dim)
|
||
|
if dim.is_integer == False:
|
||
|
raise ValueError("Dimension of the random matrices must be "
|
||
|
"integers, received %s instead."%(dim))
|
||
|
return Basic.__new__(cls, sym, dim)
|
||
|
|
||
|
symbol = property(lambda self: self.args[0])
|
||
|
dimension = property(lambda self: self.args[1])
|
||
|
|
||
|
def density(self, expr):
|
||
|
return Density(expr)
|
||
|
|
||
|
def __call__(self, expr):
|
||
|
return self.density(expr)
|
||
|
|
||
|
class GaussianEnsembleModel(RandomMatrixEnsembleModel):
|
||
|
"""
|
||
|
Abstract class for Gaussian ensembles.
|
||
|
Contains the properties common to all the
|
||
|
gaussian ensembles.
|
||
|
|
||
|
References
|
||
|
==========
|
||
|
|
||
|
.. [1] https://en.wikipedia.org/wiki/Random_matrix#Gaussian_ensembles
|
||
|
.. [2] https://arxiv.org/pdf/1712.07903.pdf
|
||
|
"""
|
||
|
def _compute_normalization_constant(self, beta, n):
|
||
|
"""
|
||
|
Helper function for computing normalization
|
||
|
constant for joint probability density of eigen
|
||
|
values of Gaussian ensembles.
|
||
|
|
||
|
References
|
||
|
==========
|
||
|
|
||
|
.. [1] https://en.wikipedia.org/wiki/Selberg_integral#Mehta's_integral
|
||
|
"""
|
||
|
n = S(n)
|
||
|
prod_term = lambda j: gamma(1 + beta*S(j)/2)/gamma(S.One + beta/S(2))
|
||
|
j = Dummy('j', integer=True, positive=True)
|
||
|
term1 = Product(prod_term(j), (j, 1, n)).doit()
|
||
|
term2 = (2/(beta*n))**(beta*n*(n - 1)/4 + n/2)
|
||
|
term3 = (2*pi)**(n/2)
|
||
|
return term1 * term2 * term3
|
||
|
|
||
|
def _compute_joint_eigen_distribution(self, beta):
|
||
|
"""
|
||
|
Helper function for computing the joint
|
||
|
probability distribution of eigen values
|
||
|
of the random matrix.
|
||
|
"""
|
||
|
n = self.dimension
|
||
|
Zbn = self._compute_normalization_constant(beta, n)
|
||
|
l = IndexedBase('l')
|
||
|
i = Dummy('i', integer=True, positive=True)
|
||
|
j = Dummy('j', integer=True, positive=True)
|
||
|
k = Dummy('k', integer=True, positive=True)
|
||
|
term1 = exp((-S(n)/2) * Sum(l[k]**2, (k, 1, n)).doit())
|
||
|
sub_term = Lambda(i, Product(Abs(l[j] - l[i])**beta, (j, i + 1, n)))
|
||
|
term2 = Product(sub_term(i).doit(), (i, 1, n - 1)).doit()
|
||
|
syms = ArrayComprehension(l[k], (k, 1, n)).doit()
|
||
|
return Lambda(tuple(syms), (term1 * term2)/Zbn)
|
||
|
|
||
|
class GaussianUnitaryEnsembleModel(GaussianEnsembleModel):
|
||
|
@property
|
||
|
def normalization_constant(self):
|
||
|
n = self.dimension
|
||
|
return 2**(S(n)/2) * pi**(S(n**2)/2)
|
||
|
|
||
|
def density(self, expr):
|
||
|
n, ZGUE = self.dimension, self.normalization_constant
|
||
|
h_pspace = RandomMatrixPSpace('P', model=self)
|
||
|
H = RandomMatrixSymbol('H', n, n, pspace=h_pspace)
|
||
|
return Lambda(H, exp(-S(n)/2 * Trace(H**2))/ZGUE)(expr)
|
||
|
|
||
|
def joint_eigen_distribution(self):
|
||
|
return self._compute_joint_eigen_distribution(S(2))
|
||
|
|
||
|
def level_spacing_distribution(self):
|
||
|
s = Dummy('s')
|
||
|
f = (32/pi**2)*(s**2)*exp((-4/pi)*s**2)
|
||
|
return Lambda(s, f)
|
||
|
|
||
|
class GaussianOrthogonalEnsembleModel(GaussianEnsembleModel):
|
||
|
@property
|
||
|
def normalization_constant(self):
|
||
|
n = self.dimension
|
||
|
_H = MatrixSymbol('_H', n, n)
|
||
|
return Integral(exp(-S(n)/4 * Trace(_H**2)))
|
||
|
|
||
|
def density(self, expr):
|
||
|
n, ZGOE = self.dimension, self.normalization_constant
|
||
|
h_pspace = RandomMatrixPSpace('P', model=self)
|
||
|
H = RandomMatrixSymbol('H', n, n, pspace=h_pspace)
|
||
|
return Lambda(H, exp(-S(n)/4 * Trace(H**2))/ZGOE)(expr)
|
||
|
|
||
|
def joint_eigen_distribution(self):
|
||
|
return self._compute_joint_eigen_distribution(S.One)
|
||
|
|
||
|
def level_spacing_distribution(self):
|
||
|
s = Dummy('s')
|
||
|
f = (pi/2)*s*exp((-pi/4)*s**2)
|
||
|
return Lambda(s, f)
|
||
|
|
||
|
class GaussianSymplecticEnsembleModel(GaussianEnsembleModel):
|
||
|
@property
|
||
|
def normalization_constant(self):
|
||
|
n = self.dimension
|
||
|
_H = MatrixSymbol('_H', n, n)
|
||
|
return Integral(exp(-S(n) * Trace(_H**2)))
|
||
|
|
||
|
def density(self, expr):
|
||
|
n, ZGSE = self.dimension, self.normalization_constant
|
||
|
h_pspace = RandomMatrixPSpace('P', model=self)
|
||
|
H = RandomMatrixSymbol('H', n, n, pspace=h_pspace)
|
||
|
return Lambda(H, exp(-S(n) * Trace(H**2))/ZGSE)(expr)
|
||
|
|
||
|
def joint_eigen_distribution(self):
|
||
|
return self._compute_joint_eigen_distribution(S(4))
|
||
|
|
||
|
def level_spacing_distribution(self):
|
||
|
s = Dummy('s')
|
||
|
f = ((S(2)**18)/((S(3)**6)*(pi**3)))*(s**4)*exp((-64/(9*pi))*s**2)
|
||
|
return Lambda(s, f)
|
||
|
|
||
|
def GaussianEnsemble(sym, dim):
|
||
|
sym, dim = _symbol_converter(sym), _sympify(dim)
|
||
|
model = GaussianEnsembleModel(sym, dim)
|
||
|
rmp = RandomMatrixPSpace(sym, model=model)
|
||
|
return RandomMatrixSymbol(sym, dim, dim, pspace=rmp)
|
||
|
|
||
|
def GaussianUnitaryEnsemble(sym, dim):
|
||
|
"""
|
||
|
Represents Gaussian Unitary Ensembles.
|
||
|
|
||
|
Examples
|
||
|
========
|
||
|
|
||
|
>>> from sympy.stats import GaussianUnitaryEnsemble as GUE, density
|
||
|
>>> from sympy import MatrixSymbol
|
||
|
>>> G = GUE('U', 2)
|
||
|
>>> X = MatrixSymbol('X', 2, 2)
|
||
|
>>> density(G)(X)
|
||
|
exp(-Trace(X**2))/(2*pi**2)
|
||
|
"""
|
||
|
sym, dim = _symbol_converter(sym), _sympify(dim)
|
||
|
model = GaussianUnitaryEnsembleModel(sym, dim)
|
||
|
rmp = RandomMatrixPSpace(sym, model=model)
|
||
|
return RandomMatrixSymbol(sym, dim, dim, pspace=rmp)
|
||
|
|
||
|
def GaussianOrthogonalEnsemble(sym, dim):
|
||
|
"""
|
||
|
Represents Gaussian Orthogonal Ensembles.
|
||
|
|
||
|
Examples
|
||
|
========
|
||
|
|
||
|
>>> from sympy.stats import GaussianOrthogonalEnsemble as GOE, density
|
||
|
>>> from sympy import MatrixSymbol
|
||
|
>>> G = GOE('U', 2)
|
||
|
>>> X = MatrixSymbol('X', 2, 2)
|
||
|
>>> density(G)(X)
|
||
|
exp(-Trace(X**2)/2)/Integral(exp(-Trace(_H**2)/2), _H)
|
||
|
"""
|
||
|
sym, dim = _symbol_converter(sym), _sympify(dim)
|
||
|
model = GaussianOrthogonalEnsembleModel(sym, dim)
|
||
|
rmp = RandomMatrixPSpace(sym, model=model)
|
||
|
return RandomMatrixSymbol(sym, dim, dim, pspace=rmp)
|
||
|
|
||
|
def GaussianSymplecticEnsemble(sym, dim):
|
||
|
"""
|
||
|
Represents Gaussian Symplectic Ensembles.
|
||
|
|
||
|
Examples
|
||
|
========
|
||
|
|
||
|
>>> from sympy.stats import GaussianSymplecticEnsemble as GSE, density
|
||
|
>>> from sympy import MatrixSymbol
|
||
|
>>> G = GSE('U', 2)
|
||
|
>>> X = MatrixSymbol('X', 2, 2)
|
||
|
>>> density(G)(X)
|
||
|
exp(-2*Trace(X**2))/Integral(exp(-2*Trace(_H**2)), _H)
|
||
|
"""
|
||
|
sym, dim = _symbol_converter(sym), _sympify(dim)
|
||
|
model = GaussianSymplecticEnsembleModel(sym, dim)
|
||
|
rmp = RandomMatrixPSpace(sym, model=model)
|
||
|
return RandomMatrixSymbol(sym, dim, dim, pspace=rmp)
|
||
|
|
||
|
class CircularEnsembleModel(RandomMatrixEnsembleModel):
|
||
|
"""
|
||
|
Abstract class for Circular ensembles.
|
||
|
Contains the properties and methods
|
||
|
common to all the circular ensembles.
|
||
|
|
||
|
References
|
||
|
==========
|
||
|
|
||
|
.. [1] https://en.wikipedia.org/wiki/Circular_ensemble
|
||
|
"""
|
||
|
def density(self, expr):
|
||
|
# TODO : Add support for Lie groups(as extensions of sympy.diffgeom)
|
||
|
# and define measures on them
|
||
|
raise NotImplementedError("Support for Haar measure hasn't been "
|
||
|
"implemented yet, therefore the density of "
|
||
|
"%s cannot be computed."%(self))
|
||
|
|
||
|
def _compute_joint_eigen_distribution(self, beta):
|
||
|
"""
|
||
|
Helper function to compute the joint distribution of phases
|
||
|
of the complex eigen values of matrices belonging to any
|
||
|
circular ensembles.
|
||
|
"""
|
||
|
n = self.dimension
|
||
|
Zbn = ((2*pi)**n)*(gamma(beta*n/2 + 1)/S(gamma(beta/2 + 1))**n)
|
||
|
t = IndexedBase('t')
|
||
|
i, j, k = (Dummy('i', integer=True), Dummy('j', integer=True),
|
||
|
Dummy('k', integer=True))
|
||
|
syms = ArrayComprehension(t[i], (i, 1, n)).doit()
|
||
|
f = Product(Product(Abs(exp(I*t[k]) - exp(I*t[j]))**beta, (j, k + 1, n)).doit(),
|
||
|
(k, 1, n - 1)).doit()
|
||
|
return Lambda(tuple(syms), f/Zbn)
|
||
|
|
||
|
class CircularUnitaryEnsembleModel(CircularEnsembleModel):
|
||
|
def joint_eigen_distribution(self):
|
||
|
return self._compute_joint_eigen_distribution(S(2))
|
||
|
|
||
|
class CircularOrthogonalEnsembleModel(CircularEnsembleModel):
|
||
|
def joint_eigen_distribution(self):
|
||
|
return self._compute_joint_eigen_distribution(S.One)
|
||
|
|
||
|
class CircularSymplecticEnsembleModel(CircularEnsembleModel):
|
||
|
def joint_eigen_distribution(self):
|
||
|
return self._compute_joint_eigen_distribution(S(4))
|
||
|
|
||
|
def CircularEnsemble(sym, dim):
|
||
|
sym, dim = _symbol_converter(sym), _sympify(dim)
|
||
|
model = CircularEnsembleModel(sym, dim)
|
||
|
rmp = RandomMatrixPSpace(sym, model=model)
|
||
|
return RandomMatrixSymbol(sym, dim, dim, pspace=rmp)
|
||
|
|
||
|
def CircularUnitaryEnsemble(sym, dim):
|
||
|
"""
|
||
|
Represents Circular Unitary Ensembles.
|
||
|
|
||
|
Examples
|
||
|
========
|
||
|
|
||
|
>>> from sympy.stats import CircularUnitaryEnsemble as CUE
|
||
|
>>> from sympy.stats import joint_eigen_distribution
|
||
|
>>> C = CUE('U', 1)
|
||
|
>>> joint_eigen_distribution(C)
|
||
|
Lambda(t[1], Product(Abs(exp(I*t[_j]) - exp(I*t[_k]))**2, (_j, _k + 1, 1), (_k, 1, 0))/(2*pi))
|
||
|
|
||
|
Note
|
||
|
====
|
||
|
|
||
|
As can be seen above in the example, density of CiruclarUnitaryEnsemble
|
||
|
is not evaluated because the exact definition is based on haar measure of
|
||
|
unitary group which is not unique.
|
||
|
"""
|
||
|
sym, dim = _symbol_converter(sym), _sympify(dim)
|
||
|
model = CircularUnitaryEnsembleModel(sym, dim)
|
||
|
rmp = RandomMatrixPSpace(sym, model=model)
|
||
|
return RandomMatrixSymbol(sym, dim, dim, pspace=rmp)
|
||
|
|
||
|
def CircularOrthogonalEnsemble(sym, dim):
|
||
|
"""
|
||
|
Represents Circular Orthogonal Ensembles.
|
||
|
|
||
|
Examples
|
||
|
========
|
||
|
|
||
|
>>> from sympy.stats import CircularOrthogonalEnsemble as COE
|
||
|
>>> from sympy.stats import joint_eigen_distribution
|
||
|
>>> C = COE('O', 1)
|
||
|
>>> joint_eigen_distribution(C)
|
||
|
Lambda(t[1], Product(Abs(exp(I*t[_j]) - exp(I*t[_k])), (_j, _k + 1, 1), (_k, 1, 0))/(2*pi))
|
||
|
|
||
|
Note
|
||
|
====
|
||
|
|
||
|
As can be seen above in the example, density of CiruclarOrthogonalEnsemble
|
||
|
is not evaluated because the exact definition is based on haar measure of
|
||
|
unitary group which is not unique.
|
||
|
"""
|
||
|
sym, dim = _symbol_converter(sym), _sympify(dim)
|
||
|
model = CircularOrthogonalEnsembleModel(sym, dim)
|
||
|
rmp = RandomMatrixPSpace(sym, model=model)
|
||
|
return RandomMatrixSymbol(sym, dim, dim, pspace=rmp)
|
||
|
|
||
|
def CircularSymplecticEnsemble(sym, dim):
|
||
|
"""
|
||
|
Represents Circular Symplectic Ensembles.
|
||
|
|
||
|
Examples
|
||
|
========
|
||
|
|
||
|
>>> from sympy.stats import CircularSymplecticEnsemble as CSE
|
||
|
>>> from sympy.stats import joint_eigen_distribution
|
||
|
>>> C = CSE('S', 1)
|
||
|
>>> joint_eigen_distribution(C)
|
||
|
Lambda(t[1], Product(Abs(exp(I*t[_j]) - exp(I*t[_k]))**4, (_j, _k + 1, 1), (_k, 1, 0))/(2*pi))
|
||
|
|
||
|
Note
|
||
|
====
|
||
|
|
||
|
As can be seen above in the example, density of CiruclarSymplecticEnsemble
|
||
|
is not evaluated because the exact definition is based on haar measure of
|
||
|
unitary group which is not unique.
|
||
|
"""
|
||
|
sym, dim = _symbol_converter(sym), _sympify(dim)
|
||
|
model = CircularSymplecticEnsembleModel(sym, dim)
|
||
|
rmp = RandomMatrixPSpace(sym, model=model)
|
||
|
return RandomMatrixSymbol(sym, dim, dim, pspace=rmp)
|
||
|
|
||
|
def joint_eigen_distribution(mat):
|
||
|
"""
|
||
|
For obtaining joint probability distribution
|
||
|
of eigen values of random matrix.
|
||
|
|
||
|
Parameters
|
||
|
==========
|
||
|
|
||
|
mat: RandomMatrixSymbol
|
||
|
The matrix symbol whose eigen values are to be considered.
|
||
|
|
||
|
Returns
|
||
|
=======
|
||
|
|
||
|
Lambda
|
||
|
|
||
|
Examples
|
||
|
========
|
||
|
|
||
|
>>> from sympy.stats import GaussianUnitaryEnsemble as GUE
|
||
|
>>> from sympy.stats import joint_eigen_distribution
|
||
|
>>> U = GUE('U', 2)
|
||
|
>>> joint_eigen_distribution(U)
|
||
|
Lambda((l[1], l[2]), exp(-l[1]**2 - l[2]**2)*Product(Abs(l[_i] - l[_j])**2, (_j, _i + 1, 2), (_i, 1, 1))/pi)
|
||
|
"""
|
||
|
if not isinstance(mat, RandomMatrixSymbol):
|
||
|
raise ValueError("%s is not of type, RandomMatrixSymbol."%(mat))
|
||
|
return mat.pspace.model.joint_eigen_distribution()
|
||
|
|
||
|
def JointEigenDistribution(mat):
|
||
|
"""
|
||
|
Creates joint distribution of eigen values of matrices with random
|
||
|
expressions.
|
||
|
|
||
|
Parameters
|
||
|
==========
|
||
|
|
||
|
mat: Matrix
|
||
|
The matrix under consideration.
|
||
|
|
||
|
Returns
|
||
|
=======
|
||
|
|
||
|
JointDistributionHandmade
|
||
|
|
||
|
Examples
|
||
|
========
|
||
|
|
||
|
>>> from sympy.stats import Normal, JointEigenDistribution
|
||
|
>>> from sympy import Matrix
|
||
|
>>> A = [[Normal('A00', 0, 1), Normal('A01', 0, 1)],
|
||
|
... [Normal('A10', 0, 1), Normal('A11', 0, 1)]]
|
||
|
>>> JointEigenDistribution(Matrix(A))
|
||
|
JointDistributionHandmade(-sqrt(A00**2 - 2*A00*A11 + 4*A01*A10 + A11**2)/2
|
||
|
+ A00/2 + A11/2, sqrt(A00**2 - 2*A00*A11 + 4*A01*A10 + A11**2)/2 + A00/2 + A11/2)
|
||
|
|
||
|
"""
|
||
|
eigenvals = mat.eigenvals(multiple=True)
|
||
|
if not all(is_random(eigenval) for eigenval in set(eigenvals)):
|
||
|
raise ValueError("Eigen values do not have any random expression, "
|
||
|
"joint distribution cannot be generated.")
|
||
|
return JointDistributionHandmade(*eigenvals)
|
||
|
|
||
|
def level_spacing_distribution(mat):
|
||
|
"""
|
||
|
For obtaining distribution of level spacings.
|
||
|
|
||
|
Parameters
|
||
|
==========
|
||
|
|
||
|
mat: RandomMatrixSymbol
|
||
|
The random matrix symbol whose eigen values are
|
||
|
to be considered for finding the level spacings.
|
||
|
|
||
|
Returns
|
||
|
=======
|
||
|
|
||
|
Lambda
|
||
|
|
||
|
Examples
|
||
|
========
|
||
|
|
||
|
>>> from sympy.stats import GaussianUnitaryEnsemble as GUE
|
||
|
>>> from sympy.stats import level_spacing_distribution
|
||
|
>>> U = GUE('U', 2)
|
||
|
>>> level_spacing_distribution(U)
|
||
|
Lambda(_s, 32*_s**2*exp(-4*_s**2/pi)/pi**2)
|
||
|
|
||
|
References
|
||
|
==========
|
||
|
|
||
|
.. [1] https://en.wikipedia.org/wiki/Random_matrix#Distribution_of_level_spacings
|
||
|
"""
|
||
|
return mat.pspace.model.level_spacing_distribution()
|