def J(self, x, grad):
# x is ln(nu)
# write out nut_no.dat so that it can be read in
nu = exp(x)
temp = self.path; temp += "nut_no.dat"
file_io.write_field(temp, self.x_no, self.y_no, nu)
# solve flow at this nu
f = open('runopt.out','a')
f.write(''.join(['Running primal from ',self.restart,'...\n']))
f.close()
run_ifdual.runpri(self.path, self.restart, self.niter, self.np, self.ctolpri)
f = open('runopt.out','a')
f.write('Done running primal\n')
f.close()
src = ''.join([self.path,'ifDual.out'])
dst = ''.join([self.path,'ifDual_Pri.step',str(self.count),'.out'])
shutil.copy(src,dst)
# read in objective function value
J = file_io.read_J(self.path)
# calculate gradient - call adjoint
if grad.size > 0:
f = open('runopt.out','a')
f.write(''.join(['Running adjoint from ',self.restart,'...\n']))
f.close()
run_ifdual.runadj(self.path, self.restart, self.niter, self.np, self.ctoladj)
f = open('runopt.out','a')
f.write('Done running adjoint\n')
f.close()
# read in the gradient
temp = self.path; temp += "grad.dat"
nno, self.x_no, self.y_no, grad[:] = file_io.read_field(temp)
# move files around
f = open('runopt.out','a')
f.write('Writing files for step %d\n' % self.count )
f.close()
src = ''.join([self.path,'J.dat'])
dst = ''.join([self.path,'J.step',str(self.count),'.dat'])
shutil.copy(src,dst)
src = ''.join([self.path,'nut_no.dat'])
dst = ''.join([self.path,'nut_no.step',str(self.count),'.dat'])
shutil.copy(src,dst)
src = ''.join([self.path,'grad.dat'])
dst = ''.join([self.path,'grad.step',str(self.count),'.dat'])
shutil.copy(src,dst)
src = ''.join([self.path,'restart.out'])
dst = ''.join([self.path,'restart.step',str(self.count),'.out'])
shutil.copy(src,dst)
src = ''.join([self.path,'ifDual.out'])
dst = ''.join([self.path,'ifDual_Adj.step',str(self.count),'.out'])
shutil.copy(src,dst)
f = open('runopt.out','a')
f.write('Iter: %d\t' % self.count)
f.write('J = %.10e\n' % J)
f.close()
self.count+=1
return J
def lbfgs(self, maxeval, nu0, tol):
n = nu0.size
o = nlopt.opt(nlopt.LD_LBFGS, n)
# tolerance on objective function value
o.set_ftol_rel(tol)
o.set_maxeval(maxeval)
o.set_min_objective(self.J)
# read in the baseline RANS viscosity for bounds
temp = self.path; temp += "nut_noRANS.dat"
j1, j2, j3, nu_base = file_io.read_field(temp)
# constrain mu to be greater than 0.75 * baseline
# o.set_lower_bounds(log(0.10*nu_base))
# constrain mu to be less than 1.5 * baseline
# o.set_upper_bounds(log(10.0*nu_base))
# call the optimizer - pass in initial viscosity
f = o.optimize(log(nu0))
minf = o.last_optimum_value()
res = o.last_optimize_result()
return f, minf, res
def lbfgs(self, maxeval, nu0, tol):
n = nu0.size
o = nlopt.opt(nlopt.LD_LBFGS, n)
# tolerance on objective function value
o.set_ftol_rel(tol)
o.set_maxeval(maxeval)
o.set_min_objective(self.J)
# read in the baseline RANS viscosity for bounds
temp = self.path
temp += "nut_noRANS.dat"
j1, j2, j3, nu_base = file_io.read_field(temp)
# constrain mu to be greater than 0.75 * baseline
# o.set_lower_bounds(log(0.10*nu_base))
# constrain mu to be less than 1.5 * baseline
# o.set_upper_bounds(log(10.0*nu_base))
# call the optimizer - pass in initial viscosity
f = o.optimize(log(nu0))
minf = o.last_optimum_value()
res = o.last_optimize_result()
return f, minf, res
import pylab from numpy import * from numpy.linalg import * from opt import opt import file_io path = "reg_test/" restart = "restart.out" niter = 1 # max ifDual iterations # read in initial turbulent viscosity temp = path temp += "nut_no.dat" nno, x_no, y_no, nu0 = file_io.read_field(temp) # create an instance of opt o = opt(x_no, y_no, niter, restart, path) tol = 1e-9 nopt = 200 # number of optimization iterations x_opt, J_opt, res = o.lbfgs(nopt, nu0, tol) nu_opt = exp(x_opt) print J_opt # write out mu_opt temp = path temp += "nu_opt" file_io.write_field(temp, x_no, y_no, nu_opt)
def J(self, x, grad):
# x is ln(nu)
# write out nut_no.dat so that it can be read in
nu = exp(x)
temp = self.path
temp += "nut_no.dat"
file_io.write_field(temp, self.x_no, self.y_no, nu)
# solve flow at this nu
f = open('runopt.out', 'a')
f.write(''.join(['Running primal from ', self.restart, '...\n']))
f.close()
run_ifdual.runpri(self.path, self.restart, self.niter, self.np,
self.ctolpri)
f = open('runopt.out', 'a')
f.write('Done running primal\n')
f.close()
src = ''.join([self.path, 'ifDual.out'])
dst = ''.join([self.path, 'ifDual_Pri.step', str(self.count), '.out'])
shutil.copy(src, dst)
# read in objective function value
J = file_io.read_J(self.path)
# calculate gradient - call adjoint
if grad.size > 0:
f = open('runopt.out', 'a')
f.write(''.join(['Running adjoint from ', self.restart, '...\n']))
f.close()
run_ifdual.runadj(self.path, self.restart, self.niter, self.np,
self.ctoladj)
f = open('runopt.out', 'a')
f.write('Done running adjoint\n')
f.close()
# read in the gradient
temp = self.path
temp += "grad.dat"
nno, self.x_no, self.y_no, grad[:] = file_io.read_field(temp)
# move files around
f = open('runopt.out', 'a')
f.write('Writing files for step %d\n' % self.count)
f.close()
src = ''.join([self.path, 'J.dat'])
dst = ''.join([self.path, 'J.step', str(self.count), '.dat'])
shutil.copy(src, dst)
src = ''.join([self.path, 'nut_no.dat'])
dst = ''.join([self.path, 'nut_no.step', str(self.count), '.dat'])
shutil.copy(src, dst)
src = ''.join([self.path, 'grad.dat'])
dst = ''.join([self.path, 'grad.step', str(self.count), '.dat'])
shutil.copy(src, dst)
src = ''.join([self.path, 'restart.out'])
dst = ''.join([self.path, 'restart.step', str(self.count), '.out'])
shutil.copy(src, dst)
src = ''.join([self.path, 'ifDual.out'])
dst = ''.join([self.path, 'ifDual_Adj.step', str(self.count), '.out'])
shutil.copy(src, dst)
f = open('runopt.out', 'a')
f.write('Iter: %d\t' % self.count)
f.write('J = %.10e\n' % J)
f.close()
self.count += 1
return J