from matplotlib import pyplot as plt
# shape = (3,3,3)
shape = (100,100,100)
# b_rand = [np.random.randn(*shape).astype(np.complex128) for k in range(3)]
omega = 0.4
b = [np.zeros(shape).astype(np.complex128) for k in range(3)]
b[1][50,50,50] = 1.
# b[0][30,30,30] = 1.
ops, b = maxwell.get_ops(omega, b)
# a0 = b.dup()
# ops['multA'](b, a0)
# a1 = b.dup()
# ops['multAT'](b, a1)
# for k in range(3):
# diff = a0.f[k].g.get() - a1.f[k].g.get()
# print np.linalg.norm(diff.flatten(), ord=2)
# a = a1.f[0].g.get().flatten()
# for t in a:
# print t
x, err = bicg.solve_asymm(b, max_iters=1000, **ops)
# for k in range(err.size):
# print k, err[k]
# print err.size, 'iterations ending with', err[-3:]
v = 0.01
plt.imshow(np.real(np.squeeze(x.f[1].g.get()[50,:,:])), \
cmap=plt.cm.jet, vmin=-v, vmax=v, interpolation='nearest')
plt.show()
import numpy as np from my_math import bicg from my_phys import simple_wave import pycuda.autoinit shape = (1, 1, 400) omega = 0.3; b = np.zeros(shape).astype(np.complex128) b[0,0,shape[2]/2] = 1 + 0j; # b = ga.to_gpu(b) # b.set(np.arange(shape[2]).astype(np.complex128)) # y = ga.zeros_like(b) b, ops = simple_wave.get_ops(shape, omega, b) # multAT(b, y) # print y.get() # multA(b, y) # print y.get() x, err = bicg.solve_asymm(b, **ops) # x, err = bicg.solve_asymm(multA, multAT, b, \ # dot=my_dot, axby=my_axby, copy=my_copy) print err.size, 'iterations ending with', err[-3:]
A = np.dot(A2, A1) - omega**2 * np.eye(N)
def mA(x, y):
y[:] = np.dot(A, x)
return
def mAT(x, y):
y[:] = np.dot(A.T, x)
return
b = np.zeros(N).astype(np.complex128)
b[N/2] = 1 + 0j
# print mAT(b)
# Find "exact" solution.
x = np.linalg.solve(A, b)
print 'error from "exact" solution', np.linalg.norm(np.dot(A, x) - b)
# Solve using bi-cg.
x = np.zeros_like(b)
x, err = bicg.solve_asymm(mA, mAT, b)
print len(err), 'iterations'
print 'ending in:'
for err_val in err[-5:]:
print err_val
print 'error is', np.linalg.norm(np.dot(A, x) - b)
# plt.plot(np.abs(x), 'b.-')
# plt.show()
def mA(x, y):
y[:] = np.dot(A, x)
return
def mAT(x, y):
y[:] = np.dot(A.T, x)
return
b = np.zeros(N).astype(np.complex128)
b[N / 2] = 1 + 0j
# print mAT(b)
# Find "exact" solution.
x = np.linalg.solve(A, b)
print 'error from "exact" solution', np.linalg.norm(np.dot(A, x) - b)
# Solve using bi-cg.
x = np.zeros_like(b)
x, err = bicg.solve_asymm(mA, mAT, b)
print len(err), 'iterations'
print 'ending in:'
for err_val in err[-5:]:
print err_val
print 'error is', np.linalg.norm(np.dot(A, x) - b)
# plt.plot(np.abs(x), 'b.-')
# plt.show()
from matplotlib import pyplot as plt
# shape = (3,3,3)
shape = (100, 100, 100)
# b_rand = [np.random.randn(*shape).astype(np.complex128) for k in range(3)]
omega = 0.4
b = [np.zeros(shape).astype(np.complex128) for k in range(3)]
b[1][50, 50, 50] = 1.
# b[0][30,30,30] = 1.
ops, b = maxwell.get_ops(omega, b)
# a0 = b.dup()
# ops['multA'](b, a0)
# a1 = b.dup()
# ops['multAT'](b, a1)
# for k in range(3):
# diff = a0.f[k].g.get() - a1.f[k].g.get()
# print np.linalg.norm(diff.flatten(), ord=2)
# a = a1.f[0].g.get().flatten()
# for t in a:
# print t
x, err = bicg.solve_asymm(b, max_iters=1000, **ops)
# for k in range(err.size):
# print k, err[k]
# print err.size, 'iterations ending with', err[-3:]
v = 0.01
plt.imshow(np.real(np.squeeze(x.f[1].g.get()[50,:,:])), \
cmap=plt.cm.jet, vmin=-v, vmax=v, interpolation='nearest')
plt.show()