def set_TR(self, radius, B_op):
assert self.parameters["zero_initial_guess"]
self.TR_radius_2 = radius * radius
self.update_x = self.update_x_with_TR
self.B_op = B_op
self.Bx = Vector()
self.B_op.init_vector(self.Bx, 0)
def trace(self, W=None):
"""
Compute the trace of A.
If the weight W is given compute the trace of W^1/2AW^1/2.
This is equivalent to
tr_W(A) = \sum_i lambda_i,
where lambda_i are the generalized eigenvalues of
A x = lambda W^-1 x.
Note if U is a W-orthogonal matrix then
tr_W(A) = \sum_i D(i,i).
"""
if W is None:
diagUtU = np.sum(self.U * self.U, 0)
tr = np.sum(self.d * diagUtU)
else:
WU = np.zeros(self.U.shape, dtype=self.U.dtype)
u, wu = Vector(), Vector()
W.init_vector(u, 1)
W.init_vector(wu, 0)
for i in range(self.U.shape[1]):
u.set_local(self.U[:, i])
W.mult(u, wu)
WU[:, i] = wu.array()
diagWUtU = np.sum(WU * self.U, 0)
tr = np.sum(self.d * diagWUtU)
return tr
def nonzero_values(function):
serialized_vector = Vector(MPI.COMM_SELF)
function.vector().gather(
serialized_vector, array(range(function.function_space().dim()),
"intc"))
indices = nonzero(serialized_vector.get_local())
return sort(serialized_vector.get_local()[indices])
def __init__(self,
form,
Space,
bcs=[],
name="x",
matvec=[None, None],
method="default",
solver_type="cg",
preconditioner_type="default"):
Function.__init__(self, Space, name=name)
self.form = form
self.method = method
self.bcs = bcs
self.matvec = matvec
self.trial = trial = TrialFunction(Space)
self.test = test = TestFunction(Space)
Mass = inner(trial, test) * dx()
self.bf = inner(form, test) * dx()
self.rhs = Vector(self.vector())
if method.lower() == "default":
self.A = A_cache[(Mass, tuple(bcs))]
self.sol = Solver_cache[(Mass, tuple(bcs), solver_type,
preconditioner_type)]
elif method.lower() == "lumping":
assert Space.ufl_element().degree() < 2
self.A = A_cache[(Mass, tuple(bcs))]
ones = Function(Space)
ones.vector()[:] = 1.
self.ML = self.A * ones.vector()
self.ML.set_local(1. / self.ML.array())
def get_diagonal(A, d, solve_mode=True):
"""
Compute the diagonal of the square operator A
or its inverse A^{-1} (if solve_mode=True).
"""
ej, xj = Vector(), Vector()
if hasattr(A, "init_vector"):
A.init_vector(ej, 1)
A.init_vector(xj, 0)
else:
A.get_operator().init_vector(ej, 1)
A.get_operator().init_vector(xj, 0)
ncol = ej.size()
da = np.zeros(ncol, dtype=ej.array().dtype)
for j in range(ncol):
ej[j] = 1.
if solve_mode:
A.solve(xj, ej)
else:
A.mult(ej, xj)
da[j] = xj[j]
ej[j] = 0.
d.set_local(da)
def estimate_diagonal_inv2(Asolver, k, d):
"""
An unbiased stochastic estimator for the diagonal of A^-1.
d = [ \sum_{j=1}^k vj .* A^{-1} vj ] ./ [ \sum_{j=1}^k vj .* vj ]
where
- vj are i.i.d. ~ N(0, I)
- .* and ./ represent the element-wise multiplication and division
of vectors, respectively.
REFERENCE:
Costas Bekas, Effrosyni Kokiopoulou, and Yousef Saad,
An estimator for the diagonal of a matrix,
Applied Numerical Mathematics, 57 (2007), pp. 1214-1229.
"""
x, b = Vector(), Vector()
if hasattr(Asolver, "init_vector"):
Asolver.init_vector(x, 1)
Asolver.init_vector(b, 0)
else:
Asolver.get_operator().init_vector(x, 1)
Asolver.get_operator().init_vector(b, 0)
d.zero()
for i in range(k):
x.zero()
Random.normal(b, 1., True)
Asolver.solve(x, b)
x *= b
d.axpy(1. / float(k), x)
def estimate_diagonal_inv2(Asolver, k, d):
"""
An unbiased stochastic estimator for the diagonal of :math:`A^{-1}`.
:math:`d = [ \sum_{j=1}^k v_j .* A^{-1} v_j ] ./ [ \sum_{j=1}^k v_j .* v_j ]`
where
- :math:`v_j` are i.i.d. :math:`\mathcal{N}(0, I)`
- :math:`.*` and :math:`./` represent the element-wise multiplication and division
of vectors, respectively.
Reference:
`Costas Bekas, Effrosyni Kokiopoulou, and Yousef Saad,
An estimator for the diagonal of a matrix,
Applied Numerical Mathematics, 57 (2007), pp. 1214-1229.`
"""
x, b = Vector(d.mpi_comm()), Vector(d.mpi_comm())
if hasattr(Asolver, "init_vector"):
Asolver.init_vector(x,1)
Asolver.init_vector(b,0)
else:
Asolver.get_operator().init_vector(x,1)
Asolver.get_operator().init_vector(b,0)
d.zero()
for i in range(k):
x.zero()
parRandom.normal(1., b)
Asolver.solve(x,b)
x *= b
d.axpy(1./float(k), x)
def __init__(self, A, solve_mode=False, accurancy = 1e-1, init_vector=None, random_engine=rademacher_engine, mpi_comm=mpi_comm_world()):
"""
Constructor:
- :code:`A`: an operator
- :code:`solve_mode`: if :code:`True` we estimate the trace of :code:`A`:math:`^{-1}`, otherwise of :code:`A`.
- code:`accurancy`: we stop when the standard deviation of the estimator is less then
:code:`accurancy`*tr(:code:`A`).
- :code:`init_vector`: use a custom function to initialize a vector compatible with the
range/domain of :code:`A`.
- :code:`random_engine`: which type of i.i.d. random variables to use (Rademacher or Gaussian).
"""
if solve_mode:
self.A = Solver2Operator(A)
else:
self.A = A
self.accurancy = accurancy
self.random_engine = random_engine
self.iter = 0
self.z = Vector(mpi_comm)
self.Az = Vector(mpi_comm)
if init_vector is None:
A.init_vector(self.z, 0)
A.init_vector(self.Az, 0)
else:
init_vector(self.z, 0)
init_vector(self.Az, 0)
def __init__(self,
A,
solve_mode=False,
accurancy=1e-1,
init_vector=None,
random_engine=rademacher_engine):
"""
Constructor:
- A: an operator
- solve_mode: if True we estimate the trace of A^{-1}, otherwise of A.
- accurancy: we stop when the standard deviation of the estimator is less then
accurancy*tr(A).
- init_vector: use a custom function to initialize a vector compatible with the
range/domain of A
- random_engine: which type of i.i.d. random variables to use (Rademacher or Gaussian)
"""
self.A = A
self.accurancy = accurancy
self.random_engine = random_engine
self.iter = 0
self.z = Vector()
self.Az = Vector()
if solve_mode:
self._apply = self._apply_solve
else:
self._apply = self._apply_mult
if init_vector is None:
A.init_vector(self.z, 0)
A.init_vector(self.Az, 0)
else:
init_vector(self.z, 0)
init_vector(self.Az, 0)
def estimate_diagonal_inv2(Asolver, k, d):
"""
An unbiased stochastic estimator for the diagonal of :math:`A^{-1}`.
:math:`d = [ \sum_{j=1}^k v_j .* A^{-1} v_j ] ./ [ \sum_{j=1}^k v_j .* v_j ]`
where
- :math:`v_j` are i.i.d. :math:`\mathcal{N}(0, I)`
- :math:`.*` and :math:`./` represent the element-wise multiplication and division
of vectors, respectively.
Reference:
`Costas Bekas, Effrosyni Kokiopoulou, and Yousef Saad,
An estimator for the diagonal of a matrix,
Applied Numerical Mathematics, 57 (2007), pp. 1214-1229.`
"""
x, b = Vector(d.mpi_comm()), Vector(d.mpi_comm())
if hasattr(Asolver, "init_vector"):
Asolver.init_vector(x, 1)
Asolver.init_vector(b, 0)
else:
Asolver.get_operator().init_vector(x, 1)
Asolver.get_operator().init_vector(b, 0)
d.zero()
for i in range(k):
x.zero()
parRandom.normal(1., b)
Asolver.solve(x, b)
x *= b
d.axpy(1. / float(k), x)
def trace2(self, W=None):
"""
Compute the trace of A*A (Note this is the square of Frob norm, since A is symmetic).
If the weight W is provided, it will compute the trace of (AW)^2.
This is equivalent to
tr_W(A) = \sum_i lambda_i^2,
where lambda_i are the generalized eigenvalues of
A x = lambda W^-1 x.
Note if U is a W-orthogonal matrix then
tr_W(A) = \sum_i D(i,i)^2.
"""
if W is None:
UtU = np.dot(self.U.T, self.U)
dUtU = self.d[:, None] * UtU #diag(d)*UtU.
tr2 = np.sum(dUtU * dUtU)
else:
WU = np.zeros(self.U.shape, dtype=self.U.dtype)
u, wu = Vector(), Vector()
W.init_vector(u, 1)
W.init_vector(wu, 0)
for i in range(self.U.shape[1]):
u.set_local(self.U[:, i])
W.mult(u, wu)
WU[:, i] = wu.array()
UtWU = np.dot(self.U.T, WU)
dUtWU = self.d[:, None] * UtWU #diag(d)*UtU.
tr2 = np.power(np.linalg.norm(dUtWU), 2)
return tr2
def __init__(self, prior, d, U):
self.prior = prior
self.LowRankH = LowRankOperator(d, U)
dsolve = d / (np.ones(d.shape, dtype=d.dtype) + d)
self.LowRankHinv = LowRankOperator(dsolve, U)
self.help = Vector()
self.init_vector(self.help, 0)
self.help1 = Vector()
self.init_vector(self.help1, 0)
def __init__(self, S, mpi_comm=mpi_comm_world(), init_vector=None):
self.S = S
self.tmp = Vector(mpi_comm)
self.my_init_vector = init_vector
if self.my_init_vector is None:
if hasattr(self.S, "init_vector"):
self.my_init_vector = self.S.init_vector
elif hasattr(self.S, "operator"):
self.my_init_vector = self.S.operator().init_vector
elif hasattr(self.S, "get_operator"):
self.my_init_vector = self.S.get_operator().init_vector
def build_pressure_null_space(self):
# Prepare function used to correct pressure values
null_fcn = Function(self.subspace("p"))
null_fcn.vector()[:] = 1.0
# Create vector that spans the null space and normalize
null_vec = Vector(null_fcn.vector())
null_vec *= 1.0 / null_vec.norm("l2")
# FIXME: Check what are relevant norms for different Krylov methods
# Create null space basis object
null_space = VectorSpaceBasis([null_vec])
return (null_space, null_fcn)
def AorthogonalityCheck(A, U, d):
"""
Test the frobenious norm of D^{-1}(U^TAU) - I_k
"""
V = np.zeros(U.shape)
AV = np.zeros(U.shape)
Av = Vector()
v = Vector()
A.init_vector(Av,0)
A.init_vector(v,1)
nvec = U.shape[1]
for i in range(0,nvec):
v.set_local(U[:,i])
v *= 1./math.sqrt(d[i])
A.mult(v,Av)
AV[:,i] = Av.array()
V[:,i] = v.array()
VtAV = np.dot(V.T, AV)
err = VtAV - np.eye(nvec, dtype=VtAV.dtype)
# plt.imshow(np.abs(err))
# plt.colorbar()
# plt.show()
print "i, ||Vt(i,:)AV(:,i) - I_i||_F, V[:,i] = 1/sqrt(lambda_i) U[:,i]"
for i in range(1,nvec+1):
print i, np.linalg.norm(err[0:i,0:i], 'fro')
def doublePass(A,Omega,k):
"""
The double pass algorithm for the HEP as presented in [1].
Inputs:
- A: the operator for which we need to estimate the dominant eigenpairs.
- Omega: a random gassian matrix with m >= k columns.
- k: the number of eigenpairs to extract.
Outputs:
- d: the estimate of the k dominant eigenvalues of A
- U: the estimate of the k dominant eigenvectors of A. U^T U = I_k
"""
w = Vector()
y = Vector()
A.init_vector(w,1)
A.init_vector(y,0)
nvec = Omega.shape[1]
assert(nvec >= k )
Y = np.zeros(Omega.shape)
for ivect in range(0,nvec):
w.set_local(Omega[:,ivect])
A.mult(w,y)
Y[:,ivect] = y.array()
Q,_ = np.linalg.qr(Y)
AQ = np.zeros(Omega.shape)
for ivect in range(0,nvec):
w.set_local(Q[:,ivect])
A.mult(w,y)
AQ[:,ivect] = y.array()
T = np.dot(Q.T, AQ)
d, V = np.linalg.eigh(T)
sort_perm = d.argsort()
sort_perm = sort_perm[::-1]
d = d[sort_perm[0:k]]
V = V[:, sort_perm[0:k]]
U = np.dot(Q,V)
return d, U
def __init__(self, prior, d, U):
self.prior = prior
ones = np.ones(d.shape, dtype=d.dtype)
self.d = ones - np.power(ones + d, -.5)
self.lrsqrt = LowRankOperator(self.d, U)
self.help = Vector()
self.init_vector(self.help, 0)
class Solver2Operator:
def __init__(self, S, mpi_comm=mpi_comm_world(), init_vector=None):
self.S = S
self.tmp = Vector(mpi_comm)
self.my_init_vector = init_vector
if self.my_init_vector is None:
if hasattr(self.S, "init_vector"):
self.my_init_vector = self.S.init_vector
elif hasattr(self.S, "operator"):
self.my_init_vector = self.S.operator().init_vector
elif hasattr(self.S, "get_operator"):
self.my_init_vector = self.S.get_operator().init_vector
def init_vector(self, x, dim):
if self.my_init_vector:
self.my_init_vector(x, dim)
else:
raise NotImplementedError("Solver2Operator.init_vector")
def mult(self, x, y):
self.S.solve(y, x)
def inner(self, x, y):
self.S.solve(self.tmp, y)
return self.tmp.inner(x)
def __init__(self, form, Space, bcs=[],
name="x",
matvec=[None, None],
method="default",
solver_type="cg",
preconditioner_type="default"):
Function.__init__(self, Space, name=name)
self.form = form
self.method = method
self.bcs = bcs
self.matvec = matvec
self.trial = trial = TrialFunction(Space)
self.test = test = TestFunction(Space)
Mass = inner(trial, test) * dx()
self.bf = inner(form, test) * dx()
self.rhs = Vector(self.vector())
if method.lower() == "default":
self.A = A_cache[(Mass, tuple(bcs))]
self.sol = Solver_cache[(Mass, tuple(
bcs), solver_type, preconditioner_type)]
elif method.lower() == "lumping":
assert Space.ufl_element().degree() < 2
self.A = A_cache[(Mass, tuple(bcs))]
ones = Function(Space)
ones.vector()[:] = 1.
self.ML = self.A * ones.vector()
self.ML.set_local(1. / self.ML.array())
class LowRankHessian:
"""
Operator that represents the action of the low rank approximation
of the Hessian and of its inverse.
"""
def __init__(self, prior, d, U):
self.prior = prior
self.LowRankH = LowRankOperator(d, U)
dsolve = d / (np.ones(d.shape, dtype=d.dtype) + d)
self.LowRankHinv = LowRankOperator(dsolve, U)
self.help = Vector(U[0].mpi_comm())
self.init_vector(self.help, 0)
self.help1 = Vector(U[0].mpi_comm())
self.init_vector(self.help1, 0)
def init_vector(self,x, dim):
self.prior.init_vector(x,dim)
def inner(self,x,y):
Hx = Vector(self.help.mpi_comm())
self.init_vector(Hx, 0)
self.mult(x, Hx)
return Hx.inner(y)
def mult(self, x, y):
self.prior.R.mult(x,y)
self.LowRankH.mult(y, self.help)
self.prior.R.mult(self.help,self.help1)
y.axpy(1, self.help1)
def solve(self, sol, rhs):
self.prior.Rsolver.solve(sol, rhs)
self.LowRankHinv.mult(rhs, self.help)
sol.axpy(-1, self.help)
def __init__(self, form, Space, bcs=[],
name="x",
matvec=[None, None],
method="default",
solver_type="cg",
preconditioner_type="default"):
Function.__init__(self, Space, name=name)
self.form = form
self.method = method
self.bcs = bcs
self.matvec = matvec
self.trial = trial = TrialFunction(Space)
self.test = test = TestFunction(Space)
Mass = inner(trial, test)*dx()
self.bf = inner(form, test)*dx()
self.rhs = Vector(self.vector())
if method.lower() == "default":
self.A = A_cache[(Mass, tuple(bcs))]
self.sol = KrylovSolver(solver_type, preconditioner_type)
self.sol.parameters["preconditioner"]["structure"] = "same"
self.sol.parameters["error_on_nonconvergence"] = False
self.sol.parameters["monitor_convergence"] = False
self.sol.parameters["report"] = False
elif method.lower() == "lumping":
assert Space.ufl_element().degree() < 2
self.A = A_cache[(Mass, tuple(bcs))]
ones = Function(Space)
ones.vector()[:] = 1.
self.ML = self.A * ones.vector()
self.ML.set_local(1. / self.ML.array())
class LowRankHessian:
"""
Operator that represents the action of the low rank approximation
of the Hessian and of its inverse.
"""
def __init__(self, prior, d, U):
self.prior = prior
self.LowRankH = LowRankOperator(d, U)
dsolve = d / (np.ones(d.shape, dtype=d.dtype) + d)
self.LowRankHinv = LowRankOperator(dsolve, U)
self.help = Vector(U[0].mpi_comm())
self.init_vector(self.help, 0)
self.help1 = Vector(U[0].mpi_comm())
self.init_vector(self.help1, 0)
def init_vector(self, x, dim):
self.prior.init_vector(x, dim)
def inner(self, x, y):
Hx = Vector(self.help.mpi_comm())
self.init_vector(Hx, 0)
self.mult(x, Hx)
return Hx.inner(y)
def mult(self, x, y):
self.prior.R.mult(x, y)
self.LowRankH.mult(y, self.help)
self.prior.R.mult(self.help, self.help1)
y.axpy(1, self.help1)
def solve(self, sol, rhs):
self.prior.Rsolver.solve(sol, rhs)
self.LowRankHinv.mult(rhs, self.help)
sol.axpy(-1, self.help)
def les_setup(u_, mesh, KineticEnergySGS, assemble_matrix, CG1Function,
nut_krylov_solver, bcs, **NS_namespace):
"""
Set up for solving the Kinetic Energy SGS-model.
"""
DG = FunctionSpace(mesh, "DG", 0)
CG1 = FunctionSpace(mesh, "CG", 1)
dim = mesh.geometry().dim()
delta = Function(DG)
delta.vector().zero()
delta.vector().axpy(1.0, assemble(TestFunction(DG) * dx))
delta.vector().set_local(delta.vector().array()**(1. / dim))
delta.vector().apply('insert')
Ck = KineticEnergySGS["Ck"]
ksgs = interpolate(Constant(1E-7), CG1)
bc_ksgs = DirichletBC(CG1, 0, "on_boundary")
A_mass = assemble_matrix(TrialFunction(CG1) * TestFunction(CG1) * dx)
nut_form = Ck * delta * sqrt(ksgs)
bcs_nut = derived_bcs(CG1, bcs['u0'], u_)
nut_ = CG1Function(nut_form,
mesh,
method=nut_krylov_solver,
bcs=bcs_nut,
bounded=True,
name="nut")
At = Matrix()
bt = Vector(nut_.vector())
ksgs_sol = KrylovSolver("bicgstab", "additive_schwarz")
#ksgs_sol.parameters["preconditioner"]["structure"] = "same_nonzero_pattern"
ksgs_sol.parameters["error_on_nonconvergence"] = False
ksgs_sol.parameters["monitor_convergence"] = False
ksgs_sol.parameters["report"] = False
del NS_namespace
return locals()
class Solver2Operator:
def __init__(self,S,mpi_comm=mpi_comm_world(), init_vector = None):
self.S = S
self.tmp = Vector(mpi_comm)
self.my_init_vector = init_vector
if self.my_init_vector is None:
if hasattr(self.S, "init_vector"):
self.my_init_vector = self.S.init_vector
elif hasattr(self.S, "operator"):
self.my_init_vector = self.S.operator().init_vector
elif hasattr(self.S, "get_operator"):
self.my_init_vector = self.S.get_operator().init_vector
def init_vector(self, x, dim):
if self.my_init_vector:
self.my_init_vector(x,dim)
else:
raise NotImplementedError("Solver2Operator.init_vector")
def mult(self,x,y):
self.S.solve(y,x)
def inner(self, x, y):
self.S.solve(self.tmp,y)
return self.tmp.inner(x)
def build_pressure_null_space(self):
# Prepare function used to correct pressure values
W_ns = self.function_spaces()[1]
# NOTE: The following works only for nodal elements
# null_fcn = Function(W_ns)
# W_ns.sub(1).dofmap().set(null_fcn.vector(), 1.0)
null_fcn = self._get_constant_pressure(W_ns)
# Create vector that spans the null space and normalize
null_vec = Vector(null_fcn.vector())
null_vec *= 1.0 / null_vec.norm("l2")
# FIXME: Check what are relevant norms for different Krylov methods
# Create null space basis object
null_space = VectorSpaceBasis([null_vec])
return (null_space, null_fcn)
def __init__(self, prior, d, U):
self.prior = prior
self.LowRankH = LowRankOperator(d, U)
dsolve = d / (np.ones(d.shape, dtype=d.dtype) + d)
self.LowRankHinv = LowRankOperator(dsolve, U)
self.help = Vector(U[0].mpi_comm())
self.init_vector(self.help, 0)
self.help1 = Vector(U[0].mpi_comm())
self.init_vector(self.help1, 0)
def apply_sqrtM(self, vect):
"""Compute M^{1/2}.vect from Vector() vect"""
sqrtMv = Function(self.Vm)
for ii in range(self.Vm.dim()):
r, c, rx, cx = self.eigsolM.get_eigenpair(ii)
RX = Vector(rx)
sqrtMv.vector().axpy(
np.sqrt(r) * np.dot(rx.array(), vect.array()), RX)
return sqrtMv.vector()
def trace(A):
"""
Compute the trace of a sparse matrix A.
"""
v = Vector()
A.init_vector(v)
mpi_comm = v.mpi_comm()
nprocs = MPI.size(mpi_comm)
if nprocs > 1:
raise Exception("trace is only serial")
n = A.size(0)
tr = 0.
for i in range(0, n):
[j, val] = A.getrow(i)
tr += val[j == i]
return tr
def __init__(self):
self.parameters = {}
self.parameters["rel_tolerance"] = 1e-9
self.parameters["abs_tolerance"] = 1e-12
self.parameters["max_iter"] = 1000
self.parameters["zero_initial_guess"] = True
self.parameters["print_level"] = 0
self.A = None
self.B = None
self.converged = False
self.iter = 0
self.reasonid = 0
self.final_norm = 0
self.r = Vector()
self.z = Vector()
self.d = Vector()
if dolfin.__version__[2] == '5': self.vrs150 = True
else: self.vrs150 = False
def __init__(self, parameters=CGSolverSteihaug_ParameterList(),comm = mpi_comm_world()):
self.parameters = parameters
self.A = None
self.B_solver = None
self.B_op = None
self.converged = False
self.iter = 0
self.reasonid = 0
self.final_norm = 0
self.TR_radius_2 = None
self.update_x = self.update_x_without_TR
self.r = Vector(comm)
self.z = Vector(comm)
self.Ad = Vector(comm)
self.d = Vector(comm)
self.Bx = Vector(comm)
def __init__(self,S,mpi_comm=mpi_comm_world(), init_vector = None):
self.S = S
self.tmp = Vector(mpi_comm)
self.my_init_vector = init_vector
if self.my_init_vector is None:
if hasattr(self.S, "init_vector"):
self.my_init_vector = self.S.init_vector
elif hasattr(self.S, "operator"):
self.my_init_vector = self.S.operator().init_vector
elif hasattr(self.S, "get_operator"):
self.my_init_vector = self.S.get_operator().init_vector
def estimate_diagonal_inv2(Asolver, k, d):
"""
An unbiased stochastic estimator for the diagonal of A^-1.
d = [ \sum_{j=1}^k vj .* A^{-1} vj ] ./ [ \sum_{j=1}^k vj .* vj ]
where
- vj are i.i.d. ~ N(0, I)
- .* and ./ represent the element-wise multiplication and division
of vectors, respectively.
REFERENCE:
Costas Bekas, Effrosyni Kokiopoulou, and Yousef Saad,
An estimator for the diagonal of a matrix,
Applied Numerical Mathematics, 57 (2007), pp. 1214-1229.
"""
x, b = Vector(), Vector()
if hasattr(Asolver, "init_vector"):
Asolver.init_vector(x, 1)
Asolver.init_vector(b, 0)
else:
Asolver.get_operator().init_vector(x, 1)
Asolver.get_operator().init_vector(b, 0)
num = np.zeros(b.array().shape, dtype=b.array().dtype)
den = np.zeros(num.shape, dtype=num.dtype)
for i in range(k):
x.zero()
b.set_local(np.random.randn(num.shape[0]))
Asolver.solve(x, b)
num = num + (x.array() * b.array())
den = den + (b.array() * b.array())
d.set_local(num / den)
def set_operator(self, A):
"""
Set the operator :math:`A`.
"""
self.r = Vector()
self.z = Vector()
self.Ad = Vector()
self.d = Vector()
self.A = A
self.A.init_vector(self.r, 0)
self.A.init_vector(self.z, 0)
self.A.init_vector(self.d, 0)
self.A.init_vector(self.Ad, 0)
def update_x_with_TR(self,x,alpha,d):
x_bk = x.copy()
x.axpy(alpha,d)
self.Bx.zero()
self.B_op.mult(x, self.Bx)
x_Bnorm2 = self.Bx.inner(x)
if x_Bnorm2 < self.TR_radius_2:
return False
else:
# Move point to boundary of trust region
self.Bx.zero()
self.B_op.mult(x_bk, self.Bx)
x_Bnorm2 = self.Bx.inner(x_bk)
Bd = Vector()
self.B_op.init_vector(Bd,0)
Bd.zero()
self.B_op.mult(self.d,Bd)
d_Bnorm2 = Bd.inner(d)
d_Bx = Bd.inner(x_bk)
a_tau = alpha*alpha*d_Bnorm2
b_tau_half = alpha* d_Bx
c_tau = x_Bnorm2- self.TR_radius_2
# Solve quadratic for tau
tau = (-b_tau_half + math.sqrt(b_tau_half*b_tau_half - a_tau*c_tau))/a_tau
x.zero()
x.axpy(1,x_bk)
x.axpy(tau*alpha, d)
return True
def __init__(self):
"""
Construct the solver with default parameters
:code:`tolerance = 1e-4`
:code:`print_level = 0`
:code:`verbose = 0`
"""
self.parameters = {}
self.parameters["tolerance"] = 1e-4
self.parameters["print_level"] = 0
self.parameters["verbose"] = 0
self.A = None
self.converged = False
self.iter = 0
self.b = Vector()
self.r = Vector()
self.p = Vector()
self.Ap = Vector()
def pointwise_variance(self, method="Exact", path_len=8):
"""
Compute/Estimate the pointwise variance of the posterior, prior distribution
and the pointwise variance reduction informed by the data.
See _Prior.pointwise_variance for more details.
"""
pr_pointwise_variance = self.prior.pointwise_variance(method, path_len)
correction_pointwise_variance = Vector()
self.init_vector(correction_pointwise_variance, 0)
self.Hlr.LowRankHinv.get_diagonal(correction_pointwise_variance)
post_pointwise_variance = pr_pointwise_variance - correction_pointwise_variance
return post_pointwise_variance, pr_pointwise_variance, correction_pointwise_variance
def pointwise_variance(self, **kwargs):
"""
Compute/estimate the pointwise variance of the posterior, prior distribution
and the pointwise variance reduction informed by the data.
See :code:`_Prior.pointwise_variance` for more details.
"""
pr_pointwise_variance = self.prior.pointwise_variance(**kwargs)
correction_pointwise_variance = Vector(self.prior.R.mpi_comm())
self.init_vector(correction_pointwise_variance, 0)
self.Hlr.LowRankHinv.get_diagonal(correction_pointwise_variance)
post_pointwise_variance = pr_pointwise_variance - correction_pointwise_variance
return post_pointwise_variance, pr_pointwise_variance, correction_pointwise_variance
def __init__(self, times, tol=1e-10):
"""
Constructor:
- times: time frame at which snapshots are stored
- tol : tolerance to identify the frame of the
snapshot.
"""
self.nsteps = len(times)
self.data = []
for i in range(self.nsteps):
self.data.append(Vector())
self.times = times
self.tol = tol
def __init__(self, times, tol=1e-10, mpi_comm = mpi_comm_world()):
"""
Constructor:
- :code:`times`: time frame at which snapshots are stored.
- :code:`tol` : tolerance to identify the frame of the snapshot.
"""
self.nsteps = len(times)
self.data = [];
for i in range(self.nsteps):
self.data.append( Vector(mpi_comm) )
self.times = times
self.tol = tol
class Operator2Solver:
def __init__(self,op, mpi_comm=mpi_comm_world()):
self.op = op
self.tmp = Vector(mpi_comm)
def init_vector(self, x, dim):
if hasattr(self.op, "init_vector"):
self.op.init_vector(x,dim)
else:
raise
def solve(self,y,x):
self.op.mult(x,y)
def inner(self, x, y):
self.op.mult(y,self.tmp)
return self.tmp.inner(x)
def get_diagonal(A, d):
"""
Compute the diagonal of the square operator :math:`A`.
Use :code:`Solver2Operator` if :math:`A^{-1}` is needed.
"""
ej, xj = Vector(d.mpi_comm()), Vector(d.mpi_comm())
A.init_vector(ej,1)
A.init_vector(xj,0)
g_size = ej.size()
d.zero()
for gid in range(g_size):
owns_gid = ej.owns_index(gid)
if owns_gid:
SetToOwnedGid(ej, gid, 1.)
ej.apply("insert")
A.mult(ej,xj)
if owns_gid:
val = GetFromOwnedGid(xj, gid)
SetToOwnedGid(d, gid, val)
SetToOwnedGid(ej, gid, 0.)
ej.apply("insert")
d.apply("insert")
def solve(self, F, u, grad = None, H = None):
if grad is None:
print "Using Automatic Differentiation to compute the gradient"
grad = derivative(F,u)
if H is None:
print "Using Automatic Differentiation to compute the Hessian"
H = derivative(grad, u)
rtol = self.parameters["rel_tolerance"]
atol = self.parameters["abs_tolerance"]
gdu_tol = self.parameters["gdu_tolerance"]
max_iter = self.parameters["max_iter"]
c_armijo = self.parameters["c_armijo"]
max_backtrack = self.parameters["max_backtracking_iter"]
prt_level = self.parameters["print_level"]
cg_coarsest_tol = self.parameters["cg_coarse_tolerance"]
Fn = assemble(F)
gn = assemble(grad)
g0_norm = gn.norm("l2")
gn_norm = g0_norm
tol = max(g0_norm*rtol, atol)
du = Vector()
self.converged = False
self.reason = 0
if prt_level > 0:
print "{0:3} {1:15} {2:15} {3:15} {4:15} {5:15} {6:5}".format(
"It", "Energy", "||g||", "(g,du)", "alpha", "tol_cg", "cg_it")
for self.it in range(max_iter):
Hn = assemble(H)
Hn.init_vector(du,1)
solver = PETScKrylovSolver("cg", "petsc_amg")
solver.set_operator(Hn)
solver.parameters["nonzero_initial_guess"] = False
cg_tol = min(cg_coarsest_tol, math.sqrt( gn_norm/g0_norm) )
solver.parameters["relative_tolerance"] = cg_tol
lin_it = solver.solve(du,gn)
self.total_cg_iter += lin_it
du_gn = -du.inner(gn)
if(-du_gn < gdu_tol):
self.converged=True
self.reason = 3
break
u_backtrack = u.copy(deepcopy=True)
alpha = 1.
bk_converged = False
#Backtrack
for j in range(max_backtrack):
u.assign(u_backtrack)
u.vector().axpy(-alpha, du)
Fnext = assemble(F)
if Fnext < Fn + alpha*c_armijo*du_gn:
Fn = Fnext
bk_converged = True
break
alpha = alpha/2.
if not bk_converged:
self.reason = 2
break
gn = assemble(grad)
gn_norm = gn.norm("l2")
if prt_level > 0:
print "{0:3d} {1:15f} {2:15f} {3:15f} {4:15f} {5:15f} {6:5d}".format(
self.it, Fn, gn_norm, du_gn, alpha, cg_tol, lin_it)
if gn_norm < tol:
self.converged = True
self.reason = 1
break
self.final_grad_norm = gn_norm
if prt_level > -1:
print self.termination_reasons[self.reason]
if self.converged:
print "Inexact Newton CG converged in ", self.it, \
"nonlinear iterations and ", self.total_cg_iter, "linear iterations."
else:
print "Inexact Newton CG did NOT converge after ", self.it, \
"nonlinear iterations and ", self.total_cg_iter, "linear iterations."
print "Final norm of the gradient", self.final_grad_norm
print "Value of the cost functional", Fn
def __init__(self,op, mpi_comm=mpi_comm_world()):
self.op = op
self.tmp = Vector(mpi_comm)
def to_dense(A, mpi_comm = mpi_comm_world() ):
"""
Convert a sparse matrix A to dense.
For debugging only.
"""
v = Vector(mpi_comm)
A.init_vector(v)
nprocs = MPI.size(mpi_comm)
if nprocs > 1:
raise Exception("to_dense is only serial")
if hasattr(A, "getrow"):
n = A.size(0)
m = A.size(1)
B = np.zeros( (n,m), dtype=np.float64)
for i in range(0,n):
[j, val] = A.getrow(i)
B[i,j] = val
return B
else:
x = Vector(mpi_comm)
Ax = Vector(mpi_comm)
A.init_vector(x,1)
A.init_vector(Ax,0)
n = Ax.get_local().shape[0]
m = x.get_local().shape[0]
B = np.zeros( (n,m), dtype=np.float64)
for i in range(0,m):
i_ind = np.array([i], dtype=np.intc)
x.set_local(np.ones(i_ind.shape), i_ind)
x.apply("sum_values")
A.mult(x,Ax)
B[:,i] = Ax.get_local()
x.set_local(np.zeros(i_ind.shape), i_ind)
x.apply("sum_values")
return B
class OasisFunction(Function):
"""Function with more or less efficient projection methods
of associated linear form.
The matvec option is provided for letting the right hand side
be computed through a fast matrix vector product. Both the matrix
and the Coefficient of the required vector must be provided.
method = "default"
Solve projection with regular linear algebra using solver_type
and preconditioner_type
method = "lumping"
Solve through lumping of mass matrix
"""
def __init__(self, form, Space, bcs=[],
name="x",
matvec=[None, None],
method="default",
solver_type="cg",
preconditioner_type="default"):
Function.__init__(self, Space, name=name)
self.form = form
self.method = method
self.bcs = bcs
self.matvec = matvec
self.trial = trial = TrialFunction(Space)
self.test = test = TestFunction(Space)
Mass = inner(trial, test) * dx()
self.bf = inner(form, test) * dx()
self.rhs = Vector(self.vector())
if method.lower() == "default":
self.A = A_cache[(Mass, tuple(bcs))]
self.sol = Solver_cache[(Mass, tuple(
bcs), solver_type, preconditioner_type)]
elif method.lower() == "lumping":
assert Space.ufl_element().degree() < 2
self.A = A_cache[(Mass, tuple(bcs))]
ones = Function(Space)
ones.vector()[:] = 1.
self.ML = self.A * ones.vector()
self.ML.set_local(1. / self.ML.array())
def assemble_rhs(self):
"""
Assemble right hand side (form*test*dx) in projection
"""
if not self.matvec[0] is None:
mat, func = self.matvec
self.rhs.zero()
self.rhs.axpy(1.0, mat * func.vector())
else:
assemble(self.bf, tensor=self.rhs)
def __call__(self, assemb_rhs=True):
"""
Compute the projection
"""
timer = Timer("Projecting {}".format(self.name()))
if assemb_rhs:
self.assemble_rhs()
for bc in self.bcs:
bc.apply(self.rhs)
if self.method.lower() == "default":
self.sol.solve(self.A, self.vector(), self.rhs)
else:
self.vector().zero()
self.vector().axpy(1.0, self.rhs * self.ML)
def inner(self,x,y):
Hx = Vector(self.help.mpi_comm())
self.init_vector(Hx, 0)
self.mult(x, Hx)
return Hx.inner(y)
class CGSampler:
"""
This class implements the CG sampler algorithm to generate samples from :math:`\mathcal{N}(0, A^{-1})`.
Reference:
`Albert Parker and Colin Fox Sampling Gaussian Distributions in Krylov Spaces with Conjugate Gradient
SIAM J SCI COMPUT, Vol 34, No. 3 pp. B312-B334`
"""
def __init__(self):
"""
Construct the solver with default parameters
:code:`tolerance = 1e-4`
:code:`print_level = 0`
:code:`verbose = 0`
"""
self.parameters = {}
self.parameters["tolerance"] = 1e-4
self.parameters["print_level"] = 0
self.parameters["verbose"] = 0
self.A = None
self.converged = False
self.iter = 0
self.b = Vector()
self.r = Vector()
self.p = Vector()
self.Ap = Vector()
def set_operator(self, A):
"""
Set the operator :code:`A`, such that :math:`x \sim \mathcal{N}(0, A^{-1})`.
.. note:: :code:`A` is any object that provides the methods :code:`init_vector()` and :code:`mult()`
"""
self.A = A
self.A.init_vector(self.r,0)
self.A.init_vector(self.p,0)
self.A.init_vector(self.Ap,0)
self.A.init_vector(self.b,0)
parRandom.normal(1., self.b)
def sample(self, noise, s):
"""
Generate a sample :math:`s ~ N(0, A^{-1})`.
:code:`noise` is a :code:`numpy.array` of i.i.d. normal variables used as input.
For a fixed realization of noise the algorithm is fully deterministic.
The size of noise determine the maximum number of CG iterations.
"""
s.zero()
self.iter = 0
self.converged = False
# r0 = b
self.r.zero()
self.r.axpy(1., self.b)
#p0 = r0
self.p.zero()
self.p.axpy(1., self.r)
self.A.mult(self.p, self.Ap)
d = self.p.inner(self.Ap)
tol2 = self.parameters["tolerance"]*self.parameters["tolerance"]
rnorm2_old = self.r.inner(self.r)
if self.parameters["verbose"] > 0:
print("initial residual = {0:g}".format( math.sqrt(rnorm2_old) ))
while (not self.converged) and (self.iter < noise.shape[0]):
gamma = rnorm2_old/d
s.axpy(noise[self.iter]/math.sqrt(d), self.p)
self.r.axpy(-gamma, self.Ap)
rnorm2 = self.r.inner(self.r)
beta = rnorm2/rnorm2_old
# p_new = r + beta p
self.p *= beta
self.p.axpy(1., self.r)
self.A.mult(self.p, self.Ap)
d = self.p.inner(self.Ap)
rnorm2_old = rnorm2
if rnorm2 < tol2:
self.converged = True
else:
rnorm2_old = rnorm2
self.iter = self.iter+1
if self.parameters["verbose"] > 0:
print("Final residual {0} after {1} iterations".format( math.sqrt(rnorm2_old), self.iter))
class CGSolverSteihaug:
"""
Solve the linear system A x = b using preconditioned conjugate gradient ( B preconditioner)
and the Steihaug stopping criterion:
- reason of termination 0: we reached the maximum number of iterations (no convergence)
- reason of termination 1: we reduced the residual up to the given tolerance (convergence)
- reason of termination 2: we reached a negative direction (premature termination due to not spd matrix)
The operator A is set using the method set_operator(A).
A must provide the following two methods:
- A.mult(x,y): y = A*x
- A.init_vector(x, dim): initialize the vector x so that it is compatible with the range (dim = 0) or
the domain (dim = 1) of A.
The preconditioner B is set using the method set_preconditioner(B).
B must provide the following method:
- B.solve(z,r): z is the action of the preconditioner B on the vector r
To solve the linear system A*x = b call solve(x,b). Here x and b are assumed
to be FEniCS::Vector objects
Maximum number of iterations, tolerances, verbosity level etc can be
set using the parameters attributes.
"""
reason = ["Maximum Number of Iterations Reached",
"Relative/Absolute residual less than tol",
"Reached a negative direction"
]
def __init__(self):
self.parameters = {}
self.parameters["rel_tolerance"] = 1e-9
self.parameters["abs_tolerance"] = 1e-12
self.parameters["max_iter"] = 1000
self.parameters["zero_initial_guess"] = True
self.parameters["print_level"] = 0
self.A = None
self.B = None
self.converged = False
self.iter = 0
self.reasonid = 0
self.final_norm = 0
self.r = Vector()
self.z = Vector()
self.d = Vector()
if dolfin.__version__[2] == '5': self.vrs150 = True
else: self.vrs150 = False
def set_operator(self, A):
self.A = A
if self.vrs150:
self.A.init_vector(self.r,0)
self.A.init_vector(self.z,0)
self.A.init_vector(self.d,0)
else:
self.r = self.A.init_vector130()
self.z = self.A.init_vector130()
self.d = self.A.init_vector130()
def set_preconditioner(self, B):
self.B = B
def solve(self,x,b):
self.iter = 0
self.converged = False
self.reasonid = 0
betanom = 0.0
alpha = 0.0
beta = 0.0
if self.parameters["zero_initial_guess"]:
self.r.zero()
self.r.axpy(1.0, b)
x.zero()
else:
self.A.mult(x,self.r)
self.r *= -1.0
self.r.axpy(1.0, b)
self.z.zero()
self.B.solve(self.z,self.r) #z = B^-1 r
self.d.zero()
self.d.axpy(1.,self.z); #d = z
nom0 = self.d.inner(self.r)
nom = nom0
if self.parameters["print_level"] == 1:
print " Iterartion : ", 0, " (B r, r) = ", nom
rtol2 = nom * self.parameters["rel_tolerance"] * self.parameters["rel_tolerance"]
atol2 = self.parameters["abs_tolerance"] * self.parameters["abs_tolerance"]
r0 = max(rtol2, atol2)
if nom <= r0:
self.converged = True
self.reasonid = 1
self.final_norm = math.sqrt(nom)
if(self.parameters["print_level"] >= 0):
print self.reason[self.reasonid]
print "Converged in ", self.iter, " iterations with final norm ", self.final_norm
return
self.A.mult(self.d, self.z) #z = A d
den = self.z.inner(self.d)
if den <= 0.0:
self.converged = True
self.reasonid = 2
self.final_norm = math.sqrt(nom)
if(self.parameters["print_level"] >= 0):
print self.reason[self.reasonid]
print "Converged in ", self.iter, " iterations with final norm ", self.final_norm
return
# start iteration
self.iter = 1
while True:
alpha = nom/den
x.axpy(alpha,self.d) # x = x + alpha d
self.r.axpy(-alpha, self.z) # r = r - alpha A d
self.B.solve(self.z, self.r) # z = B^-1 r
betanom = self.r.inner(self.z)
if self.parameters["print_level"] == 1:
print " Iteration : ", self.iter, " (B r, r) = ", betanom
if betanom < r0:
self.converged = True
self.reasonid = 1
self.final_norm = math.sqrt(betanom)
if(self.parameters["print_level"] >= 0):
print self.reason[self.reasonid]
print "Converged in ", self.iter, " iterations with final norm ", self.final_norm
break
self.iter += 1
if self.iter > self.parameters["max_iter"]:
self.converged = False
self.reasonid = 0
self.final_norm = math.sqrt(betanom)
if(self.parameters["print_level"] >= 0):
print self.reason[self.reasonid]
print "Not Converged. Final residual norm ", self.final_norm
break
beta = betanom/nom
self.d *= beta
self.d.axpy(1., self.z) #d = z + beta d
self.A.mult(self.d,self.z) # z = A d
den = self.d.inner(self.z)
if den <= 0.0:
self.converged = True
self.reasonid = 2
self.final_norm = math.sqrt(nom)
if(self.parameters["print_level"] >= 0):
print self.reason[self.reasonid]
print "Converged in ", self.iter, " iterations with final norm ", self.final_norm
break
nom = betanom
