def run_example(with_plots=True):
r"""
An example for IDA with scaled preconditioned GMRES method
as a special linear solver.
Note, how the operation Jacobian times vector is provided.
ODE:
.. math::
\dot y_1 - y_2 &= 0\\
\dot y_2 -9.82 &= 0
on return:
- :dfn:`imp_mod` problem instance
- :dfn:`imp_sim` solver instance
"""
#Defines the residual
def res(t, y, yd):
res_0 = yd[0] - y[1]
res_1 = yd[1] + 9.82
return N.array([res_0, res_1])
#Defines the Jacobian*vector product
def jacv(t, y, yd, res, v, c):
jy = N.array([[0, -1.], [0, 0]])
jyd = N.array([[1, 0.], [0, 1]])
j = jy + c * jyd
return N.dot(j, v)
#Initial conditions
y0 = [1.0, 0.0]
yd0 = [0.0, -9.82]
#Defines an Assimulo implicit problem
imp_mod = Implicit_Problem(
res, y0, yd0, name='Example using the Jacobian Vector product')
imp_mod.jacv = jacv #Sets the jacobian
imp_sim = IDA(imp_mod) #Create an IDA solver instance
#Set the parameters
imp_sim.atol = 1e-5 #Default 1e-6
imp_sim.rtol = 1e-5 #Default 1e-6
imp_sim.linear_solver = 'SPGMR' #Change linear solver
#imp_sim.options["usejac"] = False
#Simulate
t, y, yd = imp_sim.simulate(
5, 1000) #Simulate 5 seconds with 1000 communication points
#Basic tests
nose.tools.assert_almost_equal(y[-1][0], -121.75000000, 4)
nose.tools.assert_almost_equal(y[-1][1], -49.100000000)
#Plot
if with_plots:
P.plot(t, y)
P.xlabel('Time')
P.ylabel('State')
P.title(imp_mod.name)
P.show()
return imp_mod, imp_sim
def DAE_integration_assimulo(self, **kwargs):
"""
Perform time integration for DAEs with the assimulo package
"""
assert self.set_time_setting == 1, 'Time discretization must be specified first'
if self.tclose > 0:
close = True
else:
close = False
# Control vector
self.U = interpolate(self.boundary_cntrl_space, self.Vb).vector()[self.bndr_i_b]
if self.discontinous_boundary_values == 1:
self.U[self.Corner_indices] = self.U[self.Corner_indices]/2
# Definition of the sparse solver for the DAE res function to
# be defined next M should be invertible !!
my_solver = factorized(csc_matrix(self.M))
rhs = self.my_mult(self.J, self.my_mult(self.Q,self.A0)) + self.my_mult(self.Bext,self.U* self.boundary_cntrl_time(0.,self.tclose))
self.AD0 = my_solver(rhs)
# Definition of the rhs function required in assimulo
def res(t,y,yd):
"""
Definition of the residual function required in the DAE part of assimulo
"""
if close:
if t < self.tclose:
z = self.my_mult(self.M,yd) - self.my_mult(self.J, self.my_mult(self.Q,y)) - self.my_mult(self.Bext,self.U* self.boundary_cntrl_time(t,self.tclose))
else:
z = self.my_mult(self.M,yd) - self.my_mult((self.J - self.R), self.my_mult(self.Q,y))
else:
z = self.my_mult(self.M,yd) - self.my_mult(self.J, self.my_mult(self.Q,y)) - self.my_mult(self.Bext,self.U* self.boundary_cntrl_time(t,self.tclose))
return z
# Definition of the jacobian function required in assimulo
def jac(c,t,y,yd):
"""
Definition of the Jacobian matrix required in the DAE part of assimulo
"""
Matrix = csr_matrix(self.my_mult(self.J,self.Q))
if close and t > self.tclose:
Matrix = csr_matrix(self.my_mult(self.J - self.R, self.Q))
return c*csr_matrix(self.M) - Matrix
# Definition of the jacobian matrix vector function required in assimulo
def jacv(t,y,yd,res,v,c):
"""
Jacobian matrix-vector product required in the DAE part of assimulo
"""
w = self.my_mult(self.Q, v)
z = self.my_mult(self.J, w)
if close and t > self.tclose:
z -= self.my_mult(self.R, w)
return c*self.my_mult(self.M,v) - z
print('DAE Integration using assimulo built-in functions:')
model = Implicit_Problem(res,self.A0,self.AD0,self.tinit)
model.jacv = jacv
#sim = Radau5DAE(model,**kwargs)
#
# IDA method from Assimulo
#
sim = IDA(model,**kwargs)
sim.algvar = [1 for i in range(self.M.shape[0])]
sim.atol = 1.e-6
sim.rtol = 1.e-6
sim.report_continuously = True
ncp = self.Nt
sim.usejac = True
sim.suppress_alg = True
sim.inith = self.dt
sim.maxord = 5
#sim.linear_solver = 'SPGMR'
time_span, DAE_y, DAE_yd = sim.simulate(self.tfinal,ncp)
#print(sim.get_options())
print(sim.print_statistics())
A_dae = DAE_y.transpose()
# Hamiltonian
self.Nt = A_dae.shape[1]
self.tspan = np.array(time_span)
Ham_dae = np.zeros(self.Nt)
for k in range(self.Nt):
#Ham_dae[k] = 1/2 * A_dae[:,k] @ self.M @ self.Q @ A_dae[:,k]
Ham_dae[k] = 1/2 * self.my_mult(A_dae[:,k].T, \
self.my_mult(self.M, self.my_mult(self.Q, A_dae[:,k])))
# Get q variables
Aq_dae = A_dae[:self.Nq,:]
# Get p variables
Ap_dae = A_dae[self.Nq:,:]
# Get Deformation
Rho = np.zeros(self.Np)
for i in range(self.Np):
Rho[i] = self.rho(self.coord_p[i])
W_dae = np.zeros((self.Np,self.Nt))
theta = .5
for k in range(self.Nt-1):
W_dae[:,k+1] = W_dae[:,k] + self.dt * 1/Rho[:] * ( theta * Ap_dae[:,k+1] + (1-theta) * Ap_dae[:,k] )
self.Ham_dae = Ham_dae
return Aq_dae, Ap_dae, Ham_dae, W_dae, np.array(time_span)
