# This file is part of the pyMOR project (http://www.pymor.org).
# Copyright 2013-2018 pyMOR developers and contributors. All rights reserved.
# License: BSD 2-Clause License (http://opensource.org/licenses/BSD-2-Clause)
import numpy as np
from pymor.core.interfaces import ImmutableInterface
from pymor.reductors.basic import GenericRBReductor
from pymor.reductors.residual import ResidualReductor, ImplicitEulerResidualReductor
from pymor.operators.constructions import IdentityOperator
from pymor.algorithms.timestepping import ImplicitEulerTimeStepper
[docs]class ParabolicRBReductor(GenericRBReductor):
r"""Reduced Basis Reductor for parabolic equations.
This reductor uses :class:`~pymor.reductors.basic.GenericRBReductor` for the actual
RB-projection. The only addition is the assembly of an error estimator which
bounds the discrete l2-in time / energy-in space error similar to [GP05]_, [HO08]_
as follows:
.. math::
\left[ C_a^{-1}(\mu)\|e_N(\mu)\|^2 + \sum_{n=1}^{N} \Delta t\|e_n(\mu)\|^2_e \right]^{1/2}
\leq \left[ C_a^{-1}(\mu)\Delta t \sum_{n=1}^{N}\|\mathcal{R}^n(u_n(\mu), \mu)\|^2_{e,-1}
+ C_a^{-1}(\mu)\|e_0\|^2 \right]^{1/2}
Here, :math:`\|\cdot\|` denotes the norm induced by the problem's mass matrix
(e.g. the L^2-norm) and :math:`\|\cdot\|_e` is an arbitrary energy norm w.r.t.
which the space operator :math:`A(\mu)` is coercive, and :math:`C_a(\mu)` is a
lower bound for its coercivity constant. Finally, :math:`\mathcal{R}^n` denotes
the implicit Euler timestepping residual for the (fixed) time step size :math:`\Delta t`,
.. math::
\mathcal{R}^n(u_n(\mu), \mu) :=
f - M \frac{u_{n}(\mu) - u_{n-1}(\mu)}{\Delta t} - A(u_n(\mu), \mu),
where :math:`M` denotes the mass operator and :math:`f` the source term.
The dual norm of the residual is computed using the numerically stable projection
from [BEOR14]_.
Parameters
----------
d
The |InstationaryDiscretization| which is to be reduced.
RB
|VectorArray| containing the reduced basis on which to project.
basis_is_orthonormal
Indicate whether or not the basis is orthonormal w.r.t. `product`.
product
The energy inner product |Operator| w.r.t. which the reduction error is
estimated and `RB` is orthonormalized.
coercivity_estimator
`None` or a |Parameterfunctional| returning a lower bound :math:`C_a(\mu)`
for the coercivity constant of `d.operator` w.r.t. `product`.
"""
def __init__(self, d, RB=None, basis_is_orthonormal=None,
product=None, coercivity_estimator=None):
assert isinstance(d.time_stepper, ImplicitEulerTimeStepper)
super().__init__(d, RB, basis_is_orthonormal=basis_is_orthonormal, product=product)
self.coercivity_estimator = coercivity_estimator
self.residual_reductor = ImplicitEulerResidualReductor(
self.RB,
d.operator,
d.mass,
d.T / d.time_stepper.nt,
rhs=d.rhs,
product=product
)
self.initial_residual_reductor = ResidualReductor(
self.RB,
IdentityOperator(d.solution_space),
d.initial_data,
product=d.l2_product,
riesz_representatives=False
)
def _reduce(self):
with self.logger.block('RB projection ...'):
rd = super()._reduce()
with self.logger.block('Assembling error estimator ...'):
residual = self.residual_reductor.reduce()
initial_residual = self.initial_residual_reductor.reduce()
estimator = ParabolicRBEstimator(residual, self.residual_reductor.residual_range_dims,
initial_residual, self.initial_residual_reductor.residual_range_dims,
self.coercivity_estimator)
rd = rd.with_(estimator=estimator)
return rd
[docs]class ParabolicRBEstimator(ImmutableInterface):
"""Instantiated by :class:`ParabolicRBReductor`.
Not to be used directly.
"""
def __init__(self, residual, residual_range_dims, initial_residual, initial_residual_range_dims,
coercivity_estimator):
self.residual = residual
self.residual_range_dims = residual_range_dims
self.initial_residual = initial_residual
self.initial_residual_range_dims = initial_residual_range_dims
self.coercivity_estimator = coercivity_estimator
def estimate(self, U, mu, d, return_error_sequence=False):
dt = d.T / d.time_stepper.nt
C = self.coercivity_estimator(mu) if self.coercivity_estimator else 1.
est = np.empty(len(U))
est[0] = (1./C) * self.initial_residual.apply(U[0], mu=mu).l2_norm2()[0]
est[1:] = self.residual.apply(U[1:len(U)], U[0:len(U)-1],
mu=mu).l2_norm2()
est[1:] *= (dt/C**2)
est = np.sqrt(np.cumsum(est))
return est if return_error_sequence else est[-1]
def restricted_to_subbasis(self, dim, d):
if self.residual_range_dims and self.initial_residual_range_dims:
residual_range_dims = self.residual_range_dims[:dim + 1]
residual = self.residual.projected_to_subbasis(residual_range_dims[-1], dim)
initial_residual_range_dims = self.initial_residual_range_dims[:dim + 1]
initial_residual = self.initial_residual.projected_to_subbasis(initial_residual_range_dims[-1], dim)
return ParabolicRBEstimator(residual, residual_range_dims,
initial_residual, initial_residual_range_dims,
self.coercivity_estimator)
else:
self.logger.warning('Cannot efficiently reduce to subbasis')
return ParabolicRBEstimator(self.residual.projected_to_subbasis(None, dim), None,
self.initial_residual.projected_to_subbasis(None, dim), None,
self.coercivity_estimator)