Initial Conditions - hyschive/gamer-fork GitHub Wiki

This page covers the following topics about constructing a simulation initial condition (IC) and GAMER initialization:

Setting IC from Analytical Functions

Grids

Set OPT__INIT=1 and edit the following grid IC function:

  • TESTPROB_ID=0: edit the function Init_Function_User() in src/Model_Hydro/Hydro_Init_ByFunction_AssignData.cpp.

  • TESTPROB_ID≠0: edit a problem-specific initialization function (usually named SetGridIC()). See also Adding New Simulations.

The grid IC function (either Init_Function_User() or SetGridIC()) has the following prototype:

void SetGridIC( real fluid[], const double x, const double y, const double z, const double Time,
                const int lv, double AuxArray[] )

It should set the variable fluid[] at a given location x/y/z and time Time, where fluid[] is a 1D array to store different fluid fields accessible by the keys DENS, MOMX, MOMY, MOMZ, ENGY corresponding to gas mass density, momentum density along x/y/z, and gas energy density (excluding potential energy), respectively. The following example shows SetGridIC() of the blast wave test (src/TestProblem/Hydro/BlastWave/Init_TestProb_Hydro_BlastWave.cpp):

//-------------------------------------------------------------------------------------------------------
// Function    :  SetGridIC
// Description :  Set the problem-specific initial condition on grids
//
// Note        :  1. This function may also be used to estimate the numerical errors when OPT__OUTPUT_USER is enabled
//                   --> In this case, it should provide the analytical solution at the given "Time"
//                2. This function will be invoked by multiple OpenMP threads when OPENMP is enabled
//                   --> Please ensure that everything here is thread-safe
//                3. Even when DUAL_ENERGY is adopted for HYDRO, one does NOT need to set the dual-energy variable here
//                   --> It will be calculated automatically
//
// Parameter   :  fluid    : Fluid fields to be initialized
//                x/y/z    : Physical coordinates
//                Time     : Physical time
//                lv       : Target refinement level
//                AuxArray : Auxiliary array
//
// Return      :  fluid
//-------------------------------------------------------------------------------------------------------
void SetGridIC( real fluid[], const double x, const double y, const double z, const double Time,
                const int lv, double AuxArray[] )
{

   const double Blast_Engy_Exp_Density = Blast_Engy_Exp/(4.0*M_PI/3.0*Blast_Radius*Blast_Radius*Blast_Radius);
   const double r = SQRT( SQR(x-Blast_Center[0]) + SQR(y-Blast_Center[1]) + SQR(z-Blast_Center[2]) );

   fluid[DENS] = Blast_Dens_Bg;
   fluid[MOMX] = 0.0;
   fluid[MOMY] = 0.0;
   fluid[MOMZ] = 0.0;

   if ( r <= Blast_Radius )   fluid[ENGY] = Blast_Engy_Exp_Density;
   else                       fluid[ENGY] = Blast_Engy_Bg;

} // FUNCTION : SetGridIC

If the IC does not have an analytical form and relies on the table interpolation, one can use the helper functions Aux_LoadTable() to load text tables from the disk and Mis_InterpolateFromTable() to conduct linear interpolation. See src/TestProblem/Hydro/ClusterMerger_vs_Flash/Init_TestProb_ClusterMerger_vs_Flash.cpp for example.

See also Add Problem-specific Grid Fields and Particle Attributes for adding user-defined fluid fields.

Caution

When enabling --openmp, the grid IC function must be thread-safe since it will be invoked by multiple threads in parallel.

One can disable OpenMP parallelization for the grid IC function by adopting OPT__INIT_GRID_WITH_OMP=0.

Magnetic Field

One can either set the magnetic field directly or set the vector potential; for the latter, the built-in routines will construct a corresponding divergence-free magnetic field automatically. We describe the two approaches separately in the following.

To set the magnetic field directly, set OPT__INIT=1 and OPT__INIT_BFIELD_BYVECPOT=0. Edit the following magnetic field IC function:

  • TESTPROB_ID=0: edit the function Init_Function_BField_User() in src/Model_Hydro/Hydro_Init_ByFunction_AssignData.cpp.

  • TESTPROB_ID≠0: edit a problem-specific initialization function (usually named SetBFieldIC()). See also Adding New Simulations.

The magnetic field IC function (either Init_Function_BField_User() or SetBFieldIC()) has the following prototype:

void SetBFieldIC( real magnetic[], const double x, const double y, const double z, const double Time,
                  const int lv, double AuxArray[] )

It is similar to the grid IC function SetGridIC(). One should set the variable magnetic[] at a given location x/y/z and time Time, where magnetic[] is a 1D array to store different magnetic field components accessible by the keys MAGX, MAGY, MAGZ.

To set the vector potential, set OPT__INIT=1 and OPT__INIT_BFIELD_BYVECPOT=2. Then edit a problem-specific initialization function linked to the function pointer Init_BField_ByVecPot_User_Ptr, which has the following prototype:

double Init_BField_ByVecPot_User_Template( const double x, const double y, const double z, const double Time,
                                           const int lv, const char Component, double AuxArray[] )

It should return the vector potential at a given location x/y/z and time Time, where Component specifies the vector potential component ('x', 'y', or 'z') to be returned.

Example: Init_BField_ByVecPot_User_Template() in src/Model_Hydro/MHD_Init_BField_ByVecPot_Function.cpp.

Caution

When enabling --openmp, the magnetic field or the vector potential IC function must be thread-safe since it will be invoked by multiple threads in parallel.

One can disable OpenMP parallelization for the magnetic field IC function by adopting OPT__INIT_GRID_WITH_OMP=0.

Particles

Set PAR_INIT=1 and edit the following particle IC function:

  • TESTPROB_ID=0: edit the function Par_Init_ByFunction() in src/Particle/Par_Init_ByFunction.cpp.

  • TESTPROB_ID≠0: edit a problem-specific initialization function (e.g., Par_Init_ByFunction_Merger() in src/TestProblem/Hydro/ClusterMerger_vs_Flash/Par_Init_ByFunction_Merger.cpp). See also Adding New Simulations.

The particle IC function has the following prototype:

void Par_Init_ByFunction( const long NPar_ThisRank, const long NPar_AllRank,
                          real *ParMass, real *ParPosX, real *ParPosY, real *ParPosZ,
                          real *ParVelX, real *ParVelY, real *ParVelZ, real *ParTime,
                          real *ParType, real *AllAttribute[PAR_NATT_TOTAL] )

It should set the particle IC in the arrays ParMass, ParPosX/Y/Z, ParVelX/Y/Z, ParTime, ParType, and, optionally, the pointer array *AllAttribute[PAR_NATT_TOTAL], all of which have the size of NPar_ThisRank — the number of particles to be set by this MPI rank. Note that particles set by this function are only temporarily stored in this MPI rank and will later be redistributed automatically to their host leaf patches. Therefore, there is no constraint on which particles should be set by this function as long as each MPI rank initializes exactly NPar_ThisRank particles.

The built-in particle types (defined in include/Macro.h) include PTYPE_TRACER, PTYPE_GENERIC_MASSIVE, PTYPE_DARK_MATTER, and PTYPE_STAR. For PTYPE_TRACER, one must also enable the compilation option --tracer.

The following example shows Par_Init_ByFunction() in src/Particle/Par_Init_ByFunction.cpp:

//-------------------------------------------------------------------------------------------------------
// Function    :  Par_Init_ByFunction
// Description :  User-specified function to initialize particle attributes
//
// Note        :  1. Invoked by Init_GAMER() using the function pointer "Par_Init_ByFunction_Ptr"
//                   --> This function pointer may be reset by various test problem initializers, in which case
//                       this funtion will become useless
//                2. Periodicity should be taken care of in this function
//                   --> No particles should lie outside the simulation box when the periodic BC is adopted
//                   --> However, if the non-periodic BC is adopted, particles are allowed to lie outside the box
//                       (more specifically, outside the "active" region defined by amr->Par->RemoveCell)
//                       in this function. They will later be removed automatically when calling Par_Aux_InitCheck()
//                       in Init_GAMER().
//                3. Particles set by this function are only temporarily stored in this MPI rank
//                   --> They will later be redistributed when calling Par_LB_Init_RedistributeByRectangular()
//                       and LB_Init_LoadBalance()
//                   --> Therefore, there is no constraint on which particles should be set by this function
//
// Parameter   :  NPar_ThisRank : Number of particles to be set by this MPI rank
//                NPar_AllRank  : Total Number of particles in all MPI ranks
//                ParMass       : Particle mass     array with the size of NPar_ThisRank
//                ParPosX/Y/Z   : Particle position array with the size of NPar_ThisRank
//                ParVelX/Y/Z   : Particle velocity array with the size of NPar_ThisRank
//                ParTime       : Particle time     array with the size of NPar_ThisRank
//                ParType       : Particle type     array with the size of NPar_ThisRank
//                AllAttribute  : Pointer array for all particle attributes
//                                --> Dimension = [PAR_NATT_TOTAL][NPar_ThisRank]
//                                --> Use the attribute indices defined in Field.h (e.g., Idx_ParCreTime)
//                                    to access the data

//
// Return      :  ParMass, ParPosX/Y/Z, ParVelX/Y/Z, ParTime, ParType, AllAttribute
//-------------------------------------------------------------------------------------------------------
void Par_Init_ByFunction( const long NPar_ThisRank, const long NPar_AllRank,
                          real *ParMass, real *ParPosX, real *ParPosY, real *ParPosZ,
                          real *ParVelX, real *ParVelY, real *ParVelZ, real *ParTime,
                          real *ParType, real *AllAttribute[PAR_NATT_TOTAL] )
{

// synchronize all particles to the physical time on the base level
// and assign particle type
   for (long p=0; p<NPar_ThisRank; p++)
   {
      ParTime[p] = Time[0];
      ParType[p] = PTYPE_GENERIC_MASSIVE;
   }

// set other particle attributes randomly
   real *ParPos[3] = { ParPosX, ParPosY, ParPosZ };
   real *ParVel[3] = { ParVelX, ParVelY, ParVelZ };

   const uint RSeed     = 2;                                         // random seed
   const real MassMin   = 1.0e-2;                                    // minimum value of particle mass
   const real MassMax   = 1.0;                                       // maximum value of particle mass
   const real PosMin[3] = { 0.0, 0.0, 0.0 };                         // minimum value of particle position
   const real PosMax[3] = { real( amr->BoxSize[0]*(1.0-1.0e-5) ),    // maximum value of particle position
                            real( amr->BoxSize[1]*(1.0-1.0e-5) ),
                            real( amr->BoxSize[2]*(1.0-1.0e-5) ) };
   const real VelMin[3] = { -1.0, -1.0, -1.0 };                      // minimum value of particle velocity
   const real VelMax[3] = { +1.0, +1.0, +1.0 };                      // maximum value of particle velocity

   srand( RSeed );

   for (long p=0; p<NPar_ThisRank; p++)
   {
      ParMass[p] = ( (real)rand()/RAND_MAX )*( MassMax - MassMin ) + MassMin;

      for (int d=0; d<3; d++)
      {
         ParPos[d][p] = ( (real)rand()/RAND_MAX )*( PosMax[d] - PosMin[d] ) + PosMin[d];
         ParVel[d][p] = ( (real)rand()/RAND_MAX )*( VelMax[d] - VelMin[d] ) + VelMin[d];
      }
   }

} // FUNCTION : Par_Init_ByFunction

See also Add Problem-specific Grid Fields and Particle Attributes for adding user-defined particle attributes.

GAMER has a built-in routine for constructing a particle initial condition in equilibrium (e.g., Plummer, NFW, tabular). Please refer to the test problem Hydro/ParticleEquilibriumIC for its usage.

Setting IC from Files

Grids

Set OPT__INIT=3 to load the grid initial condition from a uniform-mesh binary file named UM_IC. This file will be used to provide the initial grid data of the entire computational domain fully refined to the AMR level OPT__UM_IC_LEVEL (but see also OPT__UM_IC_DOWNGRADE and OPT__UM_IC_REFINE described below). The dimension of this uniform-mesh file (assuming a row-major array) can be either [NFIELD][NZ][NY][NX] (for OPT__UM_IC_FORMAT=1) or [NZ][NY][NX][NFIELD] (for OPT__UM_IC_FORMAT=2), where NFIELD is the number of fluid fields set by OPT__UM_IC_NVAR and NX/Y/Z are the grid dimensions (i.e., number of cells) along the x/y/z directions, respectively. Since UM_IC should store the initial condition of a uniform mesh corresponding to level OPT__UM_IC_LEVEL, one must ensure that

For example, for NX0_TOT_X=16, NX0_TOT_Y=32, NX0_TOT_Z=48, OPT__UM_IC_LEVEL=1, and NCOMP_PASSIVE_USER=0, UM_IC should have the dimension [5+0][48*2^1][32*2^1][16*2^1]=[5][96][64][32] for OPT__UM_IC_FORMAT=1 or [96][64][32][5] for OPT__UM_IC_FORMAT=2, assuming the array indices are row-major. The following C++ example sets up a static and uniform gas with mass density of 1 and total energy density of 2 (assuming OPT__UM_IC_FORMAT=1).

#include <cstdio>

int main()
{
   const int NX     = 32;
   const int NY     = 64;
   const int NZ     = 96;
   const int NFIELD = 5;

   float (*IC)[NZ][NY][NX] = new float [NFIELD][NZ][NY][NX];

   for (int k=0; k<NZ; k++)
   for (int j=0; j<NY; j++)
   for (int i=0; i<NX; i++)
   {
      IC[0][k][j][i] = 1.0;   // mass density
      IC[1][k][j][i] = 0.0;   // momentum density x
      IC[2][k][j][i] = 0.0;   // momentum density y
      IC[3][k][j][i] = 0.0;   // momentum density z
      IC[4][k][j][i] = 2.0;   // total energy density
   }

   FILE *File = fopen( "UM_IC", "wb" );
   fwrite( IC, sizeof(float), NFIELD*NZ*NY*NX, File );
   fclose( File );

   delete [] IC;
}

The entire computational domain will always be first fully refined to the AMR level OPT__UM_IC_LEVEL. After that, if OPT__UM_IC_DOWNGRADE is enabled, the initialization routine will remove grids on levels 1 — OPT__UM_IC_LEVEL that do not satisfy any refinement criterion. Also, if OPT__UM_IC_REFINE is enabled, the initialization routine will add grids to levels OPT__UM_IC_LEVEL+1 — MAX_LEVEL in the regions satisfying any refinement criterion.

The initial condition file UM_IC can be loaded concurrently by multiple MPI processes using OPT__UM_IC_LOAD_NRANK.

To support AMR data in UM_IC, set OPT__UM_IC_NLEVEL>1 and edit the input table Input__UM_IC_RefineRegion. See example/input/Input__UM_IC_RefineRegion and the example code tool/inits/create_UM_IC.cpp for details.

Caution

OPT__INIT=3 does not fully support user-defined passively advected scalars (i.e., --passive>0) yet. Ask developers for help if needed.

Magnetic Field

Set OPT__INIT=1 and OPT__INIT_BFIELD_BYVECPOT=1 to load the vector potential IC from a uniform-mesh HDF5 file named B_IC. The built-in routines will construct a corresponding divergence-free magnetic field automatically.

Example: tool/inits/gen_vec_pot.py.

Particles

Set PAR_INIT=3 to load the particle initial condition from a binary file named PAR_IC. The dimension of this file (assuming a row-major array) can be either [NUM_ATTRIBUTE][NUM_PARTICLE] (for PAR_IC_FORMAT=1) or [NUM_PARTICLE][NUM_ATTRIBUTE] (for PAR_IC_FORMAT=2), where NUM_ATTRIBUTE is the number of particle attributes to be loaded and NUM_PARTICLE is the total number of particles (i.e., PAR_NPAR). By default, NUM_ATTRIBUTE is equal to 8 + --par_attribute, corresponding to particle mass, position x/y/z, velocity x/y/z, type, and user-specified attributes (and in exactly this order). One can also use PAR_IC_MASS / PAR_IC_TYPE to assign the same particle mass / type to all particles, in which case the file PAR_IC should not store particle mass / type.

The following C++ example constructs a particle initial condition file with 1000 particles assuming PAR_IC_MASS<0, PAR_IC_TYPE<0, and PAR_IC_FORMAT=1.

#include <cstdio>

int main()
{
   const int NUM_PARTICLE  = 1000;
   const int NUM_ATTRIBUTE = 8;

   float (*ParIC)[NUM_PARTICLE] = new float [NUM_ATTRIBUTE][NUM_PARTICLE];

   for (int p=0; p<NUM_PARTICLE; p++)
   {
//    replace the following lines by your particle initial condition
      ParIC[0][p] = 1.1;   // mass
      ParIC[1][p] = 2.2;   // position x
      ParIC[2][p] = 3.3;   // position y
      ParIC[3][p] = 4.4;   // position z
      ParIC[4][p] = 5.5;   // velocity x
      ParIC[5][p] = 6.6;   // velocity y
      ParIC[6][p] = 7.7;   // velocity z
      ParIC[7][p] = 1.0;   // type (generic massive)
   }

   FILE *File = fopen( "PAR_IC", "wb" );
   fwrite( ParIC, sizeof(float), NUM_PARTICLE*NUM_ATTRIBUTE, File );
   fclose( File );

   delete [] ParIC;
}

The built-in particle types (defined in include/Macro.h) include PTYPE_TRACER=0.0, PTYPE_GENERIC_MASSIVE=1.0, PTYPE_DARK_MATTER=2.0, and PTYPE_STAR=3.0. They have the floating-point types for now but will be changed to integers in the future. For PTYPE_TRACER, one must also enable the compilation option --tracer.

Note that it is not required to adopt OPT__INIT=3 and PAR_INIT=3 at the same time. In other words, it is perfectly fine to set the grid initial condition from an analytical function and load the particle initial condition from a file (and vice versa).

Compilation Options

Related options: --passive,   --par_attribute  

Runtime Parameters

Parameters described on this page: OPT__INIT,   OPT__INIT_BFIELD_BYVECPOT,   RESTART_LOAD_NRANK,   OPT__RESTART_RESET,   OPT__UM_IC_LEVEL,   OPT__UM_IC_NLEVEL,   OPT__UM_IC_NVAR,   OPT__UM_IC_FORMAT,   OPT__UM_IC_FLOAT8,   OPT__UM_IC_DOWNGRADE,   OPT__UM_IC_REFINE,   OPT__UM_IC_LOAD_NRANK,   OPT__INIT_RESTRICT,   INIT_SUBSAMPLING_NCELL,   OPT__FFTW_STARTUP  

Other related parameters: PAR_INIT,   PAR_IC_FORMAT,   PAR_IC_MASS,   OPT__INIT_GRID_WITH_OMP  

Parameters below are shown in the format:   Name   (Valid Values)   [Default Value]

  • RESTART_LOAD_NRANK   (>0)   [1]

    • Description: Number of parallel I/O for restart. In other words, RESTART_LOAD_NRANK MPI processes will load the restart file in parallel.
    • Restriction:

  • OPT__RESTART_RESET   (0=off, 1=on)   [0]

    • Description: During restart, reset some of the simulation status parameters (e.g., step, time, snapshot ID) to their initial values as if the simulation starts over again.
    • Restriction:

  • OPT__UM_IC_NVAR   (0 ≤ input < total number of gas fields)   [total number of gas fields]

  • OPT__UM_IC_FORMAT   (1=[v][z][y][x], 2=[z][y][x][v]; row-major and v=field)   [1]

  • OPT__UM_IC_FLOAT8   (<0: same as --double, 0=single precision, 1=double precision)   [-1]

    • Description: Floating-point precision of the uniform-mesh initial condition file.
    • Restriction:

  • OPT__UM_IC_DOWNGRADE   (0=off, 1=on)   [1]

    • Description: Downgrade (i.e. derefine) the uniform-mesh initial condition data for cells not satisfying any refinement criteria. See Setting IC from Files — Grids for details.
    • Restriction:

  • OPT__UM_IC_REFINE   (0=off, 1=on)   [1]

    • Description: Refine the uniform-mesh initial condition data from level OPT__UM_IC_LEVEL to MAX_LEVEL for cells satisfying the adopted refinement criteria. See Setting IC from Files — Grids for details.
    • Restriction:

  • OPT__UM_IC_LOAD_NRANK   (>0)   [1]

    • Description: Number of parallel I/O for loading the uniform-mesh initial condition file. Specifically, it allows OPT__UM_IC_LOAD_NRANK MPI processes to load the initial condition file concurrently. But the actually achieved parallel I/O depends on the system specifications. See also Setting IC from Files — Grids.
    • Restriction:

  • OPT__INIT_RESTRICT   (0=off, 1=on)   [1]

    • Description: For non-leaf patches, replace fluid data by the volume-weighted average of their child patch data. It is similar to the option OPT__FIXUP_RESTRICT except that it only applies to the initial condition.
    • Restriction:

  • INIT_SUBSAMPLING_NCELL   (0=off, >0 → number of sub-cells along each direction)   [0]

    • Description: Perform sub-sampling when constructing the grid IC to make it smoother. Specifically, each cell will be divided into Nsub3 sub-cells when calling the grid IC function, where Nsub = INIT_SUBSAMPLING_NCELL, and then take the volume-weighted average of these sub-cells.
    • Restriction: Only applicable when adopting OPT__INIT=1.

  • OPT__FFTW_STARTUP   (-1 → set to default, 0=ESTIMATE, 1=MEASURE, 2=PATIENT)   [-1]

    • Description: Initialize FFTW plans. MEASURE is recommended for the balance between FFTW plan initialization time and FFT performance. Note that simulation results can vary in each run on the level of machine precision for OPT__FFTW_STARTUP != ESTIMATE.
    • Restriction: PATIENT is not supported by FFTW2. Must use ESTIMATE when enabling --bitwise_reproducibility.

Remarks


Links

Add a custom footer
⚠️ **GitHub.com Fallback** ⚠️