def _CRsweep(A, Findex, Cindex, nu, thetacr, method):
n = A.shape[0] # problem size
numax = nu
z = np.zeros((n,))
e = np.ones((n,))
e[Cindex] = 0.0
enorm = norm(e)
rhok = 1
for it in range(1, numax+1):
if method == 'habituated':
gauss_seidel(A, e, z, iterations=1)
e[Cindex] = 0.0
elif method == 'concurrent':
gauss_seidel_indexed(A, e, z, indices=Findex, iterations=1)
else:
raise NotImplementedError('method not recognized: need \
habituated or concurrent')
enorm_old = enorm
enorm = norm(e)
rhok_old = rhok
rhok = enorm / enorm_old
# criteria 1
if (abs(rhok - rhok_old) / rhok < 0.1) and (it >= nu):
return rhok, e
# criteria 2
if rhok < 0.1 * thetacr:
return rhok, e
def _CRsweep(A, Findex, Cindex, nu, thetacr, method):
n = A.shape[0] # problem size
numax = nu
z = np.zeros((n, ))
e = np.ones((n, ))
e[Cindex] = 0.0
enorm = norm(e)
rhok = 1
for it in range(1, numax + 1):
if method == 'habituated':
gauss_seidel(A, e, z, iterations=1)
e[Cindex] = 0.0
elif method == 'concurrent':
gauss_seidel_indexed(A, e, z, indices=Findex, iterations=1)
else:
raise NotImplementedError, 'method not recognized: need habituated or concurrent'
enorm_old = enorm
enorm = norm(e)
rhok_old = rhok
rhok = enorm / enorm_old
# criteria 1
if (abs(rhok - rhok_old) / rhok < 0.1) and (it >= nu):
return rhok, e
# criteria 2
if rhok < 0.1 * thetacr:
return rhok, e
def smoother(self, x, b, nu, tol=1.e-10):
if self.withscipy:
gauss_seidel(self.A, x, b, iterations=nu)
return x
else:
# if self.withPETSc and importPETSc:
if False:
_b = PETSc.Vec().createWithArray(b, comm=PETSc.COMM_SELF)
_x = PETSc.Vec().createWithArray(np.zeros_like(b), comm=PETSc.COMM_SELF)
self.ksp_smt.rtol = tol
self.ksp_smt.max_it = nu
# self.ksp_smt.setConvergenceHistory()
self.ksp_smt.solve(_b, _x)
return _x.getArray()
else:
gauss_seidel(self._A, x, b, iterations=nu) # TODO: to remove
return x
def relax(A, x):
fn, kwargs = unpack_arg(prepostsmoother)
if fn == 'gauss_seidel':
gauss_seidel(A, x, numpy.zeros_like(x),
iterations=candidate_iters, sweep='symmetric')
elif fn == 'gauss_seidel_nr':
gauss_seidel_nr(A, x, numpy.zeros_like(x),
iterations=candidate_iters, sweep='symmetric')
elif fn == 'gauss_seidel_ne':
gauss_seidel_ne(A, x, numpy.zeros_like(x),
iterations=candidate_iters, sweep='symmetric')
elif fn == 'jacobi':
jacobi(A, x, numpy.zeros_like(x), iterations=1,
omega=1.0 / rho_D_inv_A(A))
elif fn == 'richardson':
polynomial(A, x, numpy.zeros_like(x), iterations=1,
coeffients=[1.0/approximate_spectral_radius(A)])
elif fn == 'gmres':
x[:] = (gmres(A, numpy.zeros_like(x), x0=x,
maxiter=candidate_iters)[0]).reshape(x.shape)
else:
raise TypeError('Unrecognized smoother')
def relax(A, x):
fn, kwargs = unpack_arg(prepostsmoother)
if fn == 'gauss_seidel':
gauss_seidel(A, x, numpy.zeros_like(x),
iterations=candidate_iters, sweep='symmetric')
elif fn == 'gauss_seidel_nr':
gauss_seidel_nr(A, x, numpy.zeros_like(x),
iterations=candidate_iters, sweep='symmetric')
elif fn == 'gauss_seidel_ne':
gauss_seidel_ne(A, x, numpy.zeros_like(x),
iterations=candidate_iters, sweep='symmetric')
elif fn == 'jacobi':
jacobi(A, x, numpy.zeros_like(x), iterations=1,
omega=1.0 / rho_D_inv_A(A))
elif fn == 'richardson':
polynomial(A, x, numpy.zeros_like(x), iterations=1,
coeffients=[1.0/approximate_spectral_radius(A)])
elif fn == 'gmres':
x[:] = (gmres(A, numpy.zeros_like(x), x0=x,
maxiter=candidate_iters)[0]).reshape(x.shape)
else:
raise TypeError('Unrecognized smoother')
def CR(S, method='habituated', maxiter=20):
"""Use Compatible Relaxation to compute a C/F splitting
Parameters
----------
S : csr_matrix
sparse matrix (n x n) usually matrix A of Ax=b
method : {'habituated','concurrent'}
Method used during relaxation:
- concurrent: GS relaxation on F-points, leaving e_c = 0
- habituated: full relaxation, setting e_c = 0
maxiter : int
maximum number of outer iterations (lambda)
Returns
-------
splitting : array
C/F list of 1's (coarse pt) and 0's (fine pt) (n x 1)
References
----------
.. [1] Livne, O.E., "Coarsening by compatible relaxation."
Numer. Linear Algebra Appl. 11, No. 2-3, 205-227 (2004).
Examples
--------
>>> from pyamg.gallery import poisson
>>> from pyamg.classical.cr import CR
>>> A = poisson((20,20),format='csr')
>>> splitting = CR(A)
"""
# parameters (paper notation)
ntests = 3 # (nu) number of random tests to do per iteration
nrelax = 4 # (eta) number of relaxation sweeps per test
smagic = 1.0 # (s) parameter in [1,5] to account for fill-in
gamma = 1.5 # (gamma) cycle index. use 1.5 for 2d
G = 30 # (G) number of equivalence classes (# of bins)
tdepth = 1 # (t) drop depth on parse of L bins
delta = 0 # (delta) drop threshold on parse of L bins
alphai = 0.25 # (alpha_inc) quota increase
# initializations
alpha = 0.0 # coarsening ratio, quota
beta = np.inf # quality criterion
beta1 = np.inf # quality criterion, older
beta2 = np.inf # quality criterion, oldest
n = S.shape[0] # problem size
nC = 0 # number of current Coarse points
rhs = np.zeros((n, 1)) # rhs for Ae=0
if not isspmatrix(S):
raise TypeError('expecting sparse matrix')
S = binormalize(S)
splitting = np.zeros((S.shape[0], 1), dtype='intc')
# out iterations ---------------
for m in range(0, maxiter):
mu = 0.0 # convergence rate
E = np.zeros((n, 1)) # slowness measure
# random iterations ---------------
for k in range(0, ntests):
e = 0.5*(1 + scipy.rand(n, 1))
e[splitting > 0] = 0
enorm = norm(e)
# relaxation iterations ---------------
for l in range(0, nrelax):
if method == 'habituated':
gauss_seidel(S, e, np.zeros((n, 1)), iterations=1)
e[splitting > 0] = 0
elif method == 'concurrent':
raise NotImplementedError('not implemented: need an \
F-smoother')
else:
raise NotImplementedError('method not recognized: need \
habituated or concurrent')
enorm_old = enorm
enorm = norm(e)
if enorm <= 1e-14:
# break out of loops
ntests = k
nrelax = l
maxiter = m
# end relax
# check slowness
E = np.where(np.abs(e) > E, np.abs(e), E)
# update convergence rate
mu = mu + enorm/enorm_old
# end random tests
mu = mu/ntests
# work
alpha = float(nC)/n
W = (1 + (smagic-1)*gamma*alpha)/(1-gamma*alpha)
# quality criterion
beta2 = beta1
beta1 = beta
beta = np.power(max([mu, 0.1]), 1.0 / W)
# check if we're doing well
if (beta > beta1 and beta1 > beta2) or \
m == (maxiter-1) or max(E) < 1e-13:
return splitting.ravel()
# now add points
#
# update limit on additions to splitting (C)
if alpha < 1e-13:
alpha = 0.25
else:
alpha = (1-alphai) * alpha + alphai * (1/gamma)
nCmax = np.ceil(alpha * n)
L = np.ceil(G * E / E.max()).ravel()
binid = G
# add whole bins (and t-depth nodes) at a time
u = np.zeros((n, 1))
# TODO This loop may never halt...
# Perhaps loop over nC < nCmax and binid > 0 ?
while nC < nCmax:
if delta > 0:
raise NotImplementedError
if tdepth != 1:
raise NotImplementedError
(roots,) = np.where(L == binid)
for root in roots:
if L[root] >= 0:
cols = S[root, :].indices
splitting[root] = 1 # add roots
nC += 1
L[cols] = -1
binid -= 1
#L[troots] = -1 # mark t-rings visited
#u[:]=0.0
#u[roots] = 1.0
#for depth in range(0,tdepth):
# u = np.abs(S) * u
#(troots,tmp) = np.where(u>0)
return splitting.ravel()
def CR(S, method='habituated', maxiter=20):
"""Use Compatible Relaxation to compute a C/F splitting
Parameters
----------
S : csr_matrix
sparse matrix (n x n) usually matrix A of Ax=b
method : {'habituated','concurrent'}
Method used during relaxation:
- concurrent: GS relaxation on F-points, leaving e_c = 0
- habituated: full relaxation, setting e_c = 0
maxiter : int
maximum number of outer iterations (lambda)
Returns
-------
splitting : array
C/F list of 1's (coarse pt) and 0's (fine pt) (n x 1)
References
----------
.. [1] Livne, O.E., "Coarsening by compatible relaxation."
Numer. Linear Algebra Appl. 11, No. 2-3, 205-227 (2004).
Examples
--------
>>> from pyamg.gallery import poisson
>>> from pyamg.classical.cr import CR
>>> A = poisson((20,20),format='csr')
>>> splitting = CR(A)
"""
# parameters (paper notation)
ntests = 3 # (nu) number of random tests to do per iteration
nrelax = 4 # (eta) number of relaxation sweeps per test
smagic = 1.0 # (s) parameter in [1,5] to account for fill-in
gamma = 1.5 # (gamma) cycle index. use 1.5 for 2d
G = 30 # (G) number of equivalence classes (# of bins)
tdepth = 1 # (t) drop depth on parse of L bins
delta = 0 # (delta) drop threshold on parse of L bins
alphai = 0.25 # (alpha_inc) quota increase
# initializations
alpha = 0.0 # coarsening ratio, quota
beta = np.inf # quality criterion
beta1 = np.inf # quality criterion, older
beta2 = np.inf # quality criterion, oldest
n = S.shape[0] # problem size
nC = 0 # number of current Coarse points
rhs = np.zeros((n, 1))
# rhs for Ae=0
if not isspmatrix(S):
raise TypeError('expecting sparse matrix')
S = binormalize(S)
splitting = np.zeros((S.shape[0], 1), dtype='intc')
# out iterations ---------------
for m in range(0, maxiter):
mu = 0.0 # convergence rate
E = np.zeros((n, 1)) # slowness measure
# random iterations ---------------
for k in range(0, ntests):
e = 0.5 * (1 + scipy.rand(n, 1))
e[splitting > 0] = 0
enorm = norm(e)
# relaxation iterations ---------------
for l in range(0, nrelax):
if method == 'habituated':
gauss_seidel(S, e, np.zeros((n, 1)), iterations=1)
e[splitting > 0] = 0
elif method == 'concurrent':
raise NotImplementedError, 'not implemented: need an F-smoother'
else:
raise NotImplementedError, 'method not recognized: need habituated or concurrent'
enorm_old = enorm
enorm = norm(e)
if enorm <= 1e-14:
# break out of loops
ntests = k
nrelax = l
maxiter = m
# end relax
# check slowness
E = np.where(np.abs(e) > E, np.abs(e), E)
# update convergence rate
mu = mu + enorm / enorm_old
# end random tests
mu = mu / ntests
# work
alpha = float(nC) / n
W = (1 + (smagic - 1) * gamma * alpha) / (1 - gamma * alpha)
# quality criterion
beta2 = beta1
beta1 = beta
beta = np.power(max([mu, 0.1]), 1.0 / W)
# check if we're doing well
if (beta > beta1
and beta1 > beta2) or m == (maxiter - 1) or max(E) < 1e-13:
return splitting.ravel()
# now add points
#
# update limit on additions to splitting (C)
if alpha < 1e-13:
alpha = 0.25
else:
alpha = (1 - alphai) * alpha + alphai * (1 / gamma)
nCmax = np.ceil(alpha * n)
L = np.ceil(G * E / E.max()).ravel()
binid = G
# add whole bins (and t-depth nodes) at a time
u = np.zeros((n, 1))
# TODO This loop may never halt...
# Perhaps loop over nC < nCmax and binid > 0 ?
while nC < nCmax:
if delta > 0:
raise NotImplementedError
if tdepth != 1:
raise NotImplementedError
(roots, ) = np.where(L == binid)
for root in roots:
if L[root] >= 0:
cols = S[root, :].indices
splitting[root] = 1 # add roots
nC += 1
L[cols] = -1
binid -= 1
#L[troots] = -1 # mark t-rings visited
#u[:]=0.0
#u[roots] = 1.0
#for depth in range(0,tdepth):
# u = np.abs(S) * u
#(troots,tmp) = np.where(u>0)
return splitting.ravel()
def gauss_seidel_iter(A, x, b, iterations=1):
gauss_seidel(A, x, b, iterations=iterations)
