Merge pull request #214 from andersonw1/navier_for_PyMFEM

Navier miniapp. Merged. Thank you for update.

README.md

@@ -140,4 +140,4 @@ Please refer the developers' web sites for the external libraries
 * Hypre: https://computing.llnl.gov/projects/hypre-scalable-linear-solvers-multigrid-methods
 * METIS: http://glaros.dtc.umn.edu/gkhome/metis/metis/overview
 * libceed: https://github.com/CEED/libCEED
-* gslib: https://github.com/Nek5000/gslib
+* gslib: https://github.com/Nek5000/gslib

mfem/_par/navier_solver.i

@@ -0,0 +1,53 @@
+%module(package="mfem._par") navier_solver
+%{
+#include "mfem.hpp"
+#include "numpy/arrayobject.h"
+#include "miniapps/navier/navier_solver.hpp"
+#include "../common/pyoperator.hpp"
+#include "../common/pysolvers.hpp"
+#include "../common/pycoefficient.hpp"
+#include "../common/pyintrules.hpp"
+#include "../common/pybilininteg.hpp"
+#include "../common/pynonlininteg.hpp"
+#include "../common/io_stream.hpp"
+%}
+
+%init %{
+import_array();
+%}
+
+%inline %{
+#include "miniapps/navier/navier_solver.cpp"
+%}
+
+
+%include "exception.i"
+%import "element.i"
+%import "../common/exception.i"
+
+%import "coefficient.i"
+%import "pgridfunc.i"
+%import "pmesh.i"
+%import "solvers.i"
+%import "operators.i"
+%import "../common/exception.i"
+%import "../common/io_stream_typemap.i"
+
+OSTREAM_TYPEMAP(std::ostream&)
+
+
+%typemap(in) double &time (double temp){
+  temp = PyFloat_AsDouble($input);
+  $1 = &temp;
+ }
+
+%typemap(argout) double &time {
+  %append_output(PyFloat_FromDouble(*$1));
+}
+
+%ignore *::AddVelDirichletBC(VecFuncT *f, Array &);
+%ignore *::AddPresDirichletBC(ScalarFuncT *f, Array &);
+%ignore *::AddAccelTerm(VecFuncT *f, Array &);
+
+%include "miniapps/navier/navier_solver.hpp"
+

mfem/_par/setup.py

@@ -121,7 +121,7 @@ def get_extensions():
                "tmop", "tmop_amr", "tmop_tools", "qspace", "qfunction",
                "quadinterpolator", "quadinterpolator_face",
                "submesh", "transfermap", "staticcond","sidredatacollection",
-               "psubmesh", "ptransfermap"]
+               "psubmesh", "ptransfermap","navier_solver"]
 
     if add_pumi == '1':
         from setup_local import puminc, pumilib

mfem/par.py

@@ -125,6 +125,11 @@
 except:
     pass
 
+try:
+    import mfem._par.navier_solver as navier_solver
+except:
+    pass
+
 try:   
     from mfem._par.schwarz import (SchwarzSmoother,
                                    ComplexSchwarzSmoother)

miniapps/navier/navier_mms.py

@@ -0,0 +1,225 @@
+'''
+   navier_mms.py
+
+   See c++ version in the MFEM library for more detail
+'''
+
+from mfem.par import intArray, doubleArray
+import mfem.par as mfem
+import os
+import sys
+from os.path import expanduser, join
+from numpy import sin, cos, exp, sqrt, zeros, abs, pi
+import numpy as np
+from mpi4py import MPI
+
+num_procs = 1
+myid = 0
+
+def run(ser_ref_levels=1,
+        order=5,
+        kinvis=1,
+        t_final = 10 * 0.25e-4,
+        dt = 0.25e-4,
+        pa = True,
+        ni = False,
+        visualization = True,
+        checkres = False,
+        numba = True):
+
+    mesh = mfem.Mesh("../../data/inline-quad.mesh",1,1)
+    mesh.EnsureNodes()
+    nodes = mesh.GetNodes()
+    nodes *= 2.0
+    nodes -= 1.0
+
+    for i in range(ser_ref_levels):
+        mesh.UniformRefinement()
+
+    if MPI.ROOT:
+        print("Number of elements: " + str(mesh.GetNE()))
+
+    pmesh = mfem.ParMesh(MPI.COMM_WORLD, mesh)
+
+    navsolv = mfem.navier_solver.NavierSolver(pmesh,order,kinvis)
+    navsolv.EnablePA(pa)
+    navsolv.EnableNI(ni)
+
+    u_ic = navsolv.GetCurrentVelocity()  
+    if numba:
+        @mfem.jit.vector(vdim=pmesh.Dimension(),td = True, interface = 'c++')
+        def u_excoeff(x,t,u):
+            xi = x[0]
+            yi = x[1]
+
+            u[0] = pi * sin(t) * pow(sin(pi * xi), 2.0) * sin(2.0 * pi * yi)
+            u[1] = -(pi * sin(t) * sin(2.0 * pi * xi) * pow(sin(pi * yi), 2.0))
+
+    else:
+            assert False, "numba required"
+    u_ic.ProjectCoefficient(u_excoeff)
+
+    if numba:
+        @mfem.jit.scalar(td = True)
+        def p_excoeff(x,t):
+            xi = x[0]
+            yi = x[1]
+
+            return cos(pi * xi) * sin(t) * sin(pi * yi)          
+    else:
+            assert False, "numba required"
+
+
+    attr = intArray([1]*pmesh.bdr_attributes.Max())
+    navsolv.AddVelDirichletBC(u_excoeff, attr)
+
+    if numba:
+        @mfem.jit.vector(vdim=pmesh.Dimension(),td = True, interface="c++",params={"kinvis":kinvis})
+        def accel(x,t,u):
+            xi = x[0]
+            yi = x[1]
+
+            u[0] = pi * sin(t) * sin(pi * xi) * sin(pi * yi) \
+             * (-1.0
+             + 2.0 * pow(pi, 2.0) * sin(t) * sin(pi * xi)
+             * sin(2.0 * pi * xi) * sin(pi * yi)) \
+          + pi \
+          * (2.0 * kinvis * pow(pi, 2.0)
+             * (1.0 - 2.0 * cos(2.0 * pi * xi)) * sin(t)
+             + cos(t) * pow(sin(pi * xi), 2.0)) \
+          * sin(2.0 * pi * yi)
+
+            u[1] = pi * cos(pi * yi) * sin(t) \
+          * (cos(pi * xi)
+             + 2.0 * kinvis * pow(pi, 2.0) * cos(pi * yi)
+             * sin(2.0 * pi * xi)) \
+          - pi * (cos(t) + 6.0 * kinvis * pow(pi, 2.0) * sin(t)) \
+          * sin(2.0 * pi * xi) * pow(sin(pi * yi), 2.0) \
+          + 4.0 * pow(pi, 3.0) * cos(pi * yi) * pow(sin(t), 2.0) \
+          * pow(sin(pi * xi), 2.0) * pow(sin(pi * yi), 3.0)           
+    else:
+            assert False, "numba required"
+    domain_attr = intArray([1]*pmesh.attributes.Max())
+    navsolv.AddAccelTerm(accel, domain_attr)
+
+    last_step = False
+
+    navsolv.Setup(dt)
+
+    err_u = 0
+    err_p = 0
+    u_gf = navsolv.GetCurrentVelocity()
+    p_gf = navsolv.GetCurrentPressure()
+    step = 0
+
+    time = 0.0
+
+    # if visualization:
+    #     visit_dc = mfem.VisItDataCollection("navier_mms", pmesh)
+    #     #visit_dc.SetLevelsOfDetail(4)
+    #     visit_dc.RegisterField("u", u_gf)
+    #     visit_dc.SetCycle(0)
+    #     visit_dc.SetTime(0.0)
+    #     visit_dc.Save()
+
+    while last_step == False:
+        if time + dt >= t_final - dt/2:
+            last_step = True
+
+        time = navsolv.Step(time, dt, step) #t should update in here
+
+        u_excoeff.SetTime(time)
+        p_excoeff.SetTime(time)
+        err_u = u_gf.ComputeL2Error(u_excoeff)
+        err_p = p_gf.ComputeL2Error(p_excoeff)
+
+        # if visualization and step % 10000 == 0:
+        #     visit_dc.SetCycle(step)
+        #     visit_dc.SetTime(time)
+        #     visit_dc.Save()
+
+        if MPI.ROOT:
+             print(" "*7 + "Time" + " "*10 + "dt" + " "*7 + "err_u" + " "*7 + "err_p")
+             print(f'{time:.5e} {dt:.5e} {err_u:.5e} {err_p:.5e}\n')
+
+
+
+
+        step = step + 1
+
+    navsolv.PrintTimingData()
+
+    if checkres:
+        tol = 1E-3
+        if err_u > tol or err_p > tol:
+            if MPI.ROOT:
+                print("Result has a larger error than expected.")
+
+    if visualization:
+        sock = mfem.socketstream("localhost", 19916)
+        sock.precision(8)
+        sock << "solution\n" << pmesh << u_ic
+        sock << "window_title 'Navier_MMS'\n"
+        sock << "keys\n maac\n" << "axis_labels 'x' 'y' 't'\n"
+        sock.flush()
+
+        visit_dc = mfem.VisItDataCollection("navier_mms", pmesh)
+        #visit_dc.SetLevelsOfDetail(4)
+        visit_dc.RegisterField("u_ic", u_ic)
+        visit_dc.SetCycle(0)
+        visit_dc.SetTime(0.0)
+        visit_dc.Save()
+
+if __name__ == "__main__":
+    from mfem.common.arg_parser import ArgParser
+
+    parser = ArgParser(description='navier_mms (translated from miniapps/navier/navier_mms.cpp)')
+
+    parser.add_argument('-rs', '--refine-serial',
+                        action='store', default=1, type=int,
+                        help="Number of times to refine the mesh uniformly in serial.")
+    parser.add_argument('-o', '--order',
+                        action='store', default=5, type=int,
+                        help="Order (degree) of the finite elements.")
+    parser.add_argument('-dt', '--time-step',
+                        action='store', default=0.25e-4, type=float,
+                        help="Time step.")
+    parser.add_argument('-tf', '--final-time',
+                        action='store', default=10 * 0.25e-4, type=float,
+                        help="Final time.")
+    parser.add_argument('-no-pa', '--disable-pa',
+                        action='store_false',
+                        help="Disable partial assembly.")
+    parser.add_argument('-ni', '--enable-pa',
+                        action='store_true',
+                        help="Enable numerical integration rules.")
+    parser.add_argument('-vis', '--visualization',
+                        action='store_true',
+                        help='Enable GLVis visualization')
+    parser.add_argument('-cr', '--checkresult',
+                        action='store_true',
+                        help="Enable or disable checking of the result. Returns -1 on failure.")
+    parser.add_argument("-n", "--numba",
+                        default=1, action='store', type=int,
+                        help="Use Number compiled coefficient")
+
+   #NOTE: There is no argument to change kinvis, but this is also true in c++ version
+
+    args = parser.parse_args()
+    parser.print_options(args)
+
+    # meshfile = expanduser(
+    #     join(os.path.dirname(__file__), '..', '..', 'data', args.mesh))
+    numba = (args.numba == 1)
+
+    run(ser_ref_levels = args.refine_serial,
+        order=args.order,
+        kinvis=1,
+        t_final=args.final_time,
+        dt=args.time_step,
+        pa=True,
+        ni=False,
+        # visualization=args.visualization,
+        checkres=False,
+        numba=numba)
+