def implicit_restart(self, H, evals, p):
n = len(self.H)
k = n - p
Q = np.identity(n, np.complex128)
eye = np.identity(n, np.complex128)
t0 = g.time()
for i in range(p):
Qi, Ri = np.linalg.qr(H - evals[i] * eye)
H = Ri @ Qi + evals[i] * eye
Q = Q @ Qi
t1 = g.time()
if self.verbose:
g.message(f"Arnoldi: QR in {t1-t0} s")
r = g.eval(self.basis[k] * H[k, k - 1] +
self.basis[-1] * self.H[-1][-1] * Q[n - 1, k - 1])
rn = g.norm2(r)**0.5
t0 = g.time()
g.rotate(self.basis, np.ascontiguousarray(Q.T), 0, k, 0, n)
t1 = g.time()
if self.verbose:
g.message(f"Arnoldi: rotate in {t1-t0} s")
self.basis = self.basis[0:k]
self.basis.append(g.eval(r / rn))
self.H = [[H[j, i] for j in range(i + 2)] for i in range(k)]
self.H[-1][-1] = rn
def rotate_basis_to_evec(self, little_evec):
n = len(self.H)
t0 = g.time()
g.rotate(self.basis[0:n], np.ascontiguousarray(little_evec.T), 0, n, 0,
n)
t1 = g.time()
if self.verbose:
g.message(f"Arnoldi: rotate in {t1-t0} s")
return self.basis[0:n]
assert eps == 0.0
eps = g.norm2((b < a) - (a > b)) ** 0.5
g.message(f"Test a < b compatible with b > a: {eps}")
assert eps == 0.0
################################################################################
# Test basis rotate against linear combination
################################################################################
a = [g.complex(grid) for i in range(3)]
b = [g.complex(grid) for i in range(3)]
rng.cnormal(a)
c = [g.copy(x) for x in a]
Qt = np.array([[1, 2, 3], [9, 7, 13], [15, 17, 19]], dtype=np.complex128)
for i in range(3):
g.linear_combination(b[i], a, Qt[i])
g.rotate(a, Qt, 0, 3, 0, 3, True)
g.rotate(c, Qt, 0, 3, 0, 3, False)
for i in range(3):
eps = g.norm2(a[i] - b[i]) / g.norm2(a[i])
g.message(f"Test basis rotate {i} on accelerator: {eps}")
assert eps < 1e-13
eps = g.norm2(c[i] - b[i]) / g.norm2(a[i])
g.message(f"Test basis rotate {i} on host: {eps}")
assert eps < 1e-13
################################################################################
# Test mem_report
################################################################################
g.mem_report()
def __call__(self, mat, src, ckpt=None):
# verbosity
verbose = g.default.is_verbose("irl")
# checkpointer
if ckpt is None:
ckpt = g.checkpointer_none()
ckpt.grid = src.grid
self.ckpt = ckpt
# first approximate largest eigenvalue
pit = g.algorithms.eigen.power_iteration(eps=0.05,
maxiter=10,
real=True)
lambda_max = pit(mat, src)[0]
# parameters
Nm = self.params["Nm"]
Nk = self.params["Nk"]
Nstop = self.params["Nstop"]
assert Nm >= Nk and Nstop <= Nk
# tensors
dtype = np.float64
lme = np.empty((Nm, ), dtype)
lme2 = np.empty((Nm, ), dtype)
ev = np.empty((Nm, ), dtype)
ev2 = np.empty((Nm, ), dtype)
ev2_copy = np.empty((Nm, ), dtype)
# fields
f = g.lattice(src)
v = g.lattice(src)
evec = [g.lattice(src) for i in range(Nm)]
# advice memory storage
if not self.params["advise"] is None:
g.advise(evec, self.params["advise"])
# scalars
k1 = 1
k2 = Nk
beta_k = 0.0
# set initial vector
evec[0] @= src / g.norm2(src)**0.5
# initial Nk steps
for k in range(Nk):
self.step(mat, ev, lme, evec, f, Nm, k)
# restarting loop
for it in range(self.params["maxiter"]):
if verbose:
g.message("Restart iteration %d" % it)
for k in range(Nk, Nm):
self.step(mat, ev, lme, evec, f, Nm, k)
f *= lme[Nm - 1]
# eigenvalues
for k in range(Nm):
ev2[k] = ev[k + k1 - 1]
lme2[k] = lme[k + k1 - 1]
# diagonalize
t0 = g.time()
Qt = np.identity(Nm, dtype)
self.diagonalize(ev2, lme2, Nm, Qt)
t1 = g.time()
if verbose:
g.message("Diagonalization took %g s" % (t1 - t0))
# sort
ev2_copy = ev2.copy()
ev2 = list(reversed(sorted(ev2)))
# implicitly shifted QR transformations
Qt = np.identity(Nm, dtype)
t0 = g.time()
for ip in range(k2, Nm):
g.qr_decomposition(ev, lme, Nm, Nm, Qt, ev2[ip], k1, Nm)
t1 = g.time()
if verbose:
g.message("QR took %g s" % (t1 - t0))
# rotate
t0 = g.time()
g.rotate(evec, Qt, k1 - 1, k2 + 1, 0, Nm)
t1 = g.time()
if verbose:
g.message("Basis rotation took %g s" % (t1 - t0))
# compression
f *= Qt[k2 - 1, Nm - 1]
f += lme[k2 - 1] * evec[k2]
beta_k = g.norm2(f)**0.5
betar = 1.0 / beta_k
evec[k2] @= betar * f
lme[k2 - 1] = beta_k
if verbose:
g.message("beta_k = ", beta_k)
# convergence test
if it >= self.params["Nminres"]:
if verbose:
g.message("Rotation to test convergence")
# diagonalize
for k in range(Nm):
ev2[k] = ev[k]
lme2[k] = lme[k]
Qt = np.identity(Nm, dtype)
t0 = g.time()
self.diagonalize(ev2, lme2, Nk, Qt)
t1 = g.time()
if verbose:
g.message("Diagonalization took %g s" % (t1 - t0))
B = g.copy(evec[0])
allconv = True
if beta_k >= self.params["betastp"]:
jj = 1
while jj <= Nstop:
j = Nstop - jj
g.linear_combination(B, evec[0:Nk], Qt[j, 0:Nk])
B *= 1.0 / g.norm2(B)**0.5
if not ckpt.load(v):
mat(v, B)
ckpt.save(v)
ev_test = g.inner_product(B, v).real
eps2 = g.norm2(v - ev_test * B) / lambda_max**2.0
if verbose:
g.message("%-65s %-45s %-50s" % (
"ev[ %d ] = %s" % (j, ev2_copy[j]),
"<B|M|B> = %s" % (ev_test),
"|M B - ev B|^2 / ev_max^2 = %s" % (eps2),
))
if eps2 > self.params["resid"]:
allconv = False
if jj == Nstop:
break
jj = min([Nstop, 2 * jj])
if allconv:
if verbose:
g.message("Converged in %d iterations" % it)
break
t0 = g.time()
g.rotate(evec, Qt, 0, Nstop, 0, Nk)
t1 = g.time()
if verbose:
g.message("Final basis rotation took %g s" % (t1 - t0))
return (evec[0:Nstop], ev2_copy[0:Nstop])
def __call__(self, mat, src, ckpt=None):
# verbosity
verbose = g.default.is_verbose("irl")
# checkpointer
if ckpt is None:
ckpt = g.checkpointer_none()
ckpt.grid = src.grid
self.ckpt = ckpt
# first approximate largest eigenvalue
pit = g.algorithms.eigen.power_iteration(eps=0.05, maxiter=10, real=True)
lambda_max = pit(mat, src)[0]
# parameters
Nm = self.params["Nm"]
Nu = self.params["Nu"]
Nk = self.params["Nk"]
Nstop = self.params["Nstop"]
Np = Nm-Nk
MaxIter=self.params["maxiter"]
Np /= MaxIter
assert Nm >= Nk and Nstop <= Nk
print ( 'Nm=',Nm,'Nu=',Nu,'Nk=',Nk )
# tensors
dtype = np.float64
ctype = np.complex128
lme = np.zeros((Nu,Nm), ctype)
lmd = np.zeros((Nu,Nm), ctype)
lme2 = np.zeros((Nu,Nm), ctype)
lmd2 = np.empty((Nu,Nm), ctype)
Qt = np.zeros((Nm,Nm),ctype)
Q = np.zeros((Nm,Nm),ctype)
ev = np.empty((Nm,), dtype)
ev2_copy = np.empty((Nm,), dtype)
# fields
f = g.lattice(src)
v = g.lattice(src)
evec = [g.lattice(src) for i in range(Nm)]
w = [g.lattice(src) for i in range(Nu)]
w_copy = [g.lattice(src) for i in range(Nu)]
# advice memory storage
if not self.params["advise"] is None:
g.advise(evec, self.params["advise"])
# scalars
k1 = 1
k2 = Nk
beta_k = 0.0
rng=g.random("test")
# set initial vector
# rng.zn(w)
for i in range(Nu):
rng.zn(w[i])
if i > 0:
g.orthogonalize(w[i],evec[0:i])
evec[i]=g.copy(w[i])
evec[i] *= 1.0/ g.norm2(evec[i]) ** 0.5
g.message("norm(evec[%d]=%e "%(i,g.norm2(evec[i])))
if i > 0:
for j in range(i):
ip=g.innerProduct(evec[j],w[i])
if np.abs(ip) >1e-6:
g.message("inner(evec[%d],w[%d])=%e %e"% (j,i,ip.real,ip.imag))
# evec[i] @= src[i] / g.norm2(src[i]) ** 0.5
# initial Nk steps
Nblock_k = int(Nk/Nu)
for b in range(Nblock_k):
self.blockStep(mat, lmd, lme, evec, w, w_copy, Nm, b,Nu)
Nblock_p = int(Np/Nu)
# restarting loop
# for it in range(self.params["maxiter"]):
for it in range(MaxIter):
if verbose:
g.message("Restart iteration %d" % it)
Nblock_l = Nblock_k + it*Nblock_p;
Nblock_r = Nblock_l + Nblock_p;
Nl = Nblock_l*Nu
Nr = Nblock_r*Nu
# ev2.resize(Nr)
ev2 = np.empty((Nr,), dtype)
for b in range(Nblock_l, Nblock_r):
self.blockStep(mat, lmd, lme, evec, w, w_copy, Nm, b,Nu)
for u in range(Nu):
for k in range(Nr):
lmd2[u,k]=lmd[u,k]
lme2[u,k]=lme[u,k]
Qt = np.identity(Nr, ctype)
# diagonalize
t0 = g.time()
# self.diagonalize(ev2, lme2, Nm, Qt)
self.diagonalize(ev2,lmd2,lme2,Nu,Nr,Qt)
# def diagonalize(self, eval, lmd, lme, Nu, Nk, Nm, Qt):
t1 = g.time()
if verbose:
g.message("Diagonalization took %g s" % (t1 - t0))
# sort
ev2_copy = ev2.copy()
ev2 = list(reversed(sorted(ev2)))
for i in range(Nr):
g.message("Rval[%d]= %e"%(i,ev2[i]))
# rotate
# t0 = g.time()
# g.rotate(evec, Qt, k1 - 1, k2 + 1, 0, Nm)
# t1 = g.time()
# if verbose:
# g.message("Basis rotation took %g s" % (t1 - t0))
# convergence test
if it >= self.params["Nminres"]:
if verbose:
g.message("Rotation to test convergence")
# diagonalize
for k in range(Nr):
ev2[k] = ev[k]
# lme2[k] = lme[k]
for u in range(Nu):
for k in range(Nr):
lmd2[u,k]=lmd[u,k]
lme2[u,k]=lme[u,k]
Qt = np.identity(Nm, ctype)
t0 = g.time()
# self.diagonalize(ev2, lme2, Nk, Qt)
self.diagonalize(ev2,lmd2,lme2,Nu,Nr,Qt)
t1 = g.time()
if verbose:
g.message("Diagonalization took %g s" % (t1 - t0))
B = g.copy(evec[0])
allconv = True
if beta_k >= self.params["betastp"]:
jj = 1
while jj <= Nstop:
j = Nstop - jj
g.linear_combination(B, evec[0:Nr], Qt[j, 0:Nr])
g.message("norm=%e"%(g.norm2(B)))
B *= 1.0 / g.norm2(B) ** 0.5
if not ckpt.load(v):
mat(v, B)
ckpt.save(v)
ev_test = g.innerProduct(B, v).real
eps2 = g.norm2(v - ev_test * B) / lambda_max ** 2.0
if verbose:
g.message(
"%-65s %-45s %-50s"
% (
"ev[ %d ] = %s" % (j, ev2_copy[j]),
"<B|M|B> = %s" % (ev_test),
"|M B - ev B|^2 / ev_max^2 = %s" % (eps2),
)
)
if eps2 > self.params["resid"]:
allconv = False
if jj == Nstop:
break
jj = min([Nstop, 2 * jj])
if allconv:
if verbose:
g.message("Converged in %d iterations" % it)
break
t0 = g.time()
g.rotate(evec, Qt, 0, Nstop, 0, Nk)
t1 = g.time()
if verbose:
g.message("Final basis rotation took %g s" % (t1 - t0))
return (evec[0:Nstop], ev2_copy[0:Nstop])