|
XGC1
|
|
XGC1
|
Data Types | |
| type | resamp_bin_type |
Public Member Functions | |
| subroutine | set_cce_buffer_id (id, inode1, inode2) |
| subroutine | set_cce_buffer_and_overlap_id (id1, id2, inode1, inode2) |
| subroutine | resample_init (grid, inode1, inode2, dsmu, dvp, nmu, nvp, vp_max, mu_max, tile_size, bins, cce_sync_sendrcv) |
| This routine initializes the resampling module and sets some global parameters such as v-grid size, etc. More... | |
| subroutine | resample_finalize (bins) |
| This routine cleans up the resampling module. More... | |
| subroutine | resample (grid, spall, psn, inode1, inode2, dsmu, dvp, nmu, nvp, vp_max, mu_max, cce_sync_sendrcv, tile_size) |
| Resample the marker particles in XGC to improve phase-space coverage and reduce marker noise. More... | |
| subroutine | build_bins (sp, bins, grid, cce_sync) |
| Loops over particles, puts them in the appropriate bin, computes partial mean This routine sorts particles in the appropriate phase-space bin and computes the total charge in the bin. More... | |
| subroutine | add_to_bin (bin, ptli, ind) |
| Adds the data of one particle to a phase-space bin. More... | |
| subroutine | destroy_bins (bins) |
| Cleans up bin memory without deallocating the array of bins itself (can be kept for next species) More... | |
| subroutine | update_cmat (grid, psn, nconstraint, nptl, vert, patch, bary, tri, psi, B, bvec, spec, pindex, part, Cmat, op_mode) |
| This subroutine sets up the matrix C (constraint matrix) with the coefficients needed to compute the conserved quantities charge, parallel momentum, magnetic moment, parallel and perp. kinetic energy and the grid charge on the vertices connected to the bin. Those quantities are then obtained by multiplying the matrix C with the delta-f (–> w1) weight vector from the right. This matrix forms the basis for the QP optimization program that finds the delta-f weights for the resampled particle set. More... | |
| subroutine | check_bin (bin, sp, cce_sync_sendrcv, resamp_auto_upsample_loc, resamp_auto_downsample_loc, resamp_var_resample_loc) |
| This routine checks whether a bin needs to be resampled and determines the type of resampling (up-/down-/re-sampling). The chosen resampling option is stored in the bin data structure: 0) No resampling needed 1) Resampling due to high weight variance 2) Upsampling due to low particle count 3) Downsampling due to high particle count. More... | |
| subroutine | resample_bin (bin, grid, sp, psn, cce_sync_sendrcv, start_wr_pos, resamp_failed_bins_loc, resamp_upsample_failed_loc, resamp_downsample_failed_loc, resamp_resample_failed_loc, resamp_retried_bins_loc, resamp_retried_failed_loc, resamp_fullf_failed_loc, resamp_phvol_added_loc) |
| The actual process of resampling the particle set in a bin happens in this routine. A bin is passed as input, the pre-determined type of resampling is executed and the weights of the new particle set are computed with a QP optimization. If the optimization is successful, the old particle set in the input variable "sp" is replaced with the new particle set. More... | |
| subroutine | load_new_ptl (grid, new_ptl, bin_id, itr, p, psi) |
| This routine loads new particles into a bin with uniform marker density in configuration space. Particles are drawn from a rectacngle enclosing the bin volume and are accepted if inside the Voronoi volume of the bin. A more efficient loading can be achieved by pre-selecting the triangle in which to load. This functionality will be provided in a separate function. More... | |
| subroutine | setup_fullf (ptli, spec_type, den, flow, temp_ev, bvec, B, psi) |
| Evaluate the background density, temperature, flow, magnetic field and psi at the input particle's position. More... | |
| subroutine | load_new_ptl2 (grid, new_ptl, bin_id, itr, p, psi) |
| This routine loads new particles into a bin with uniform marker density in configuration space. Particles are drawn from the bin volume. First, a triangle is chosen using importance sampling with the triangle area as criterion. Then a random position in the triangle is drawn. The drawn position is accepted if inside the Voronoi volume of the bin. More... | |
| subroutine | get_new_fullf_weight (grid, psn, part, psi, B, flow, den, temp_ev, spec, tri, bary, mark_den, err) |
| Compute the new full-f weight of the input particle \(w_0 = f/g\). Note that either \(w_0\) or \(f\) can be chosen freely, while the particle's phasespace volume \(\delta V = 1/g\) is determined by Liousville's theorem of phase space volume conservation. The distribution function \(f\) can in principle be anything here as long as the \(w_2\) weight is set accordingly. The goal is to choose \(w_0\) and \(w_2\) such that the weight evolution equation does not have to be changed. This is from Seung-Hoe Ku's notes: \(f_p = w_1 w_0 g\) \(\frac{\mathrm{D}f_p}{\mathrm{D}t} = w_0 g \frac{\mathrm{d}w_1}{\mathrm{d}t}\) \(=-\frac{\mathrm{D}f_0}{\mathrm{D}t} + S\) \(w_1(t_2)-w_1(t_1) = -\frac{f_0(t_2)-f_0(t_1)}{w_0 g}\) \(w_2(t_2)-w_2(t=0) = -\frac{f_0(t_2)-f_0(t=0)}{w_0 g}\) Easiest choice: \(w_0 g = f_0(t=0),\quad w_2(t=0) = 0\) Then: \(1-w_2 = \frac{f_0(t)}{w_0 g} = \frac{f_0(t)}{f_0(t=0)}\) and \(w_1(t_2)-w_1(t_1) = -\frac{f_0(t_2)}{f_0(t=0)} + (1-w_2(t_1))\). More... | |
| subroutine | setup_opt (nconstraint, part_num, Cmat, constraint, weights, n_eq_bin, n_ineq_bin, eq_bin, ineq_bin, grid_ineq_on, ineq_tol, positivity, ierr) |
| This routine sets up the input for the QP optimization that computes the delta-f weights of the new particles. There are two types of constraints, bin and grid constraints. Bin constraints correspond to moments of the distribution function evaluated with nearest neighbor interpolation on the configuration space grid. Grid constraints correspond to moments evaluated with linear (barycentric) interpolation on the conf. space grid. Bin constraints can be enforced as equality and inequality constraints: use idx_constraint and idx_constraint2 to choose which constraint is an equality and which is an inequality constraint. Grid constraints are enforced as inequality constraints. Since grid constraints make the optimization problem much harder due to their large number, using them is optional (grid_ineq_on) A positivity constraint for the weights can be activated with the variable "positivity". This is used for full-f optimization. More... | |
| subroutine | restart_resample () |
| Move phase space density from the 5D grid over to the particles for restarting simulatin with a different grid. More... | |
| subroutine | get_marker_velo (grid, psn, ptli, isp, velo, vrad) |
| Evaluates the marker particle equations of motion. The results are the time derivatives of the marker position and parallel velocity (or rho, to be precise), as well as the radial velocity v.grad(psi). The results can be used in the moment preserving QP weight optimization. More... | |
| subroutine | get_sampling_ratio (grid) |
| This routine upsamples the marker population in spall for a new configuration space grid. The new grid should be denser than the current grid, i.e. require upsampling, not downsampling. This is due to particle data locality. Upsampling to a denser mesh is done on the coarse mesh, while downsampling from a fine mesh is better done with particle data arranged for the coarse mesh. The routine reads in the vertex positions of the new mesh and runs a search for each of them to associate them with the vertices of the current mesh. The ratio of new mesh vertices per current mesh vertex is the upsampling ratio. If the upsampling ratio is smaller than unity, that vertex is skipped. More... | |
| subroutine | resamp_scale_phvol (grid, sp) |
| Loops over particles and scales their phase space volume to account for particles added to empty bins. More... | |
| subroutine | f0_upsampler (grid, psn, spall) |
| Tries to upsample so that velocity space is fully populated. Called when f0_upsample = .true., typically to fill up velocity space before grid operations when using the pseudo-inverse interpolation with f0_velocity_interp_use_pseudo_inv = .true. to ensure that the pseudo-inverse calculation will converge. More... | |
| subroutine | diag_velocity_space_filled (grid, sp, tile_size_in) |
| This is a diagnostics routine. Loops over particles to check how many 3D real-space mesh nodes are fully populated by particles in velocity space and prints to the log. More... | |
| subroutine | cce_fcoupling_resample (grid, psn, spall, istep) |
| subroutine | cce_reset_particle_outside (grid, sp) |
| subroutine | cce_buffer_snd_updated_bins (isp) |
| subroutine | cce_buffer_rcv_updated_bins (isp) |
| logical function | is_first_call () |
Public Attributes | |
| integer | resamp_rate = 2 |
| timesteps between subsequent resamples (multiplied by sml_f_source_period More... | |
| real(8) | resamp_var = 0.02D0 |
| threshold for relative standard deviation in bin for auto-resample More... | |
| real(8) | resamp_min_ratio = 0.2D0 |
| min ratio of (# of ptl)/(target # of ptl) in bin for auto-upsample More... | |
| real(8) | resamp_max_ratio = 2.0D0 |
| max ratio of (# of ptl)/(target # of ptl) in bin for auto-downsample More... | |
| integer | resamp_max_target = 4 |
| Overrides the number of constraints in determining the target # of ptl of a bin. More... | |
| integer | resamp_tile_size = 2 |
| Bin size on the velocity space grid in cells (not vertices) (input parameter) More... | |
| integer | resamp_tile_size_now = 2 |
| Used to override resamp_tile_size. More... | |
| real(8) | resamp_ineq_tol = 1D-7 |
| Threshold for relative error in the inequality constraints in the QP optimization. More... | |
| logical | resamp_restart = .false. |
| Perform resampling before dumping the final restart file. More... | |
| logical | resamp_retry = .false. |
| Retry QP optimization for failed bins with relaxed inequality constraints. More... | |
| real(8) | resamp_ineq_tol_max = 1D-3 |
| Maximal threshold for relative error in inequality constraints for retried bins. More... | |
| real(8) | resamp_highv_max = 10D0 |
| energy cutoff of the high velocity bins v_para>f0_vp_max and v_perp>f0_smu_max More... | |
| real(8) | resamp_highv_max_ratio = 4D0 |
| Downsampling threshold for high-velocity bins. More... | |
| character(len=200) | resamp_node_file |
| File containing the vertex positions of the mesh for which to resample. More... | |
| integer | resamp_nphi_new =1 |
| Number of poloidal planes in simulation with new mesh. More... | |
| logical | resamp_restart_read =.false. |
| Whether to read a restart file written from a simulation with different grid. More... | |
| logical | resamp_fill_empty =.false. |
| Whether to fill empty bins. More... | |
| logical | resamp_fullf_on =.false. |
| Whether to resample the full-f weights in addition to delta-f weights. More... | |
| logical | resamp_grid_ineq_on =.false. |
| Switch for using inequality constraints for the grid charge for resampling. More... | |
| logical | resamp_distribute_evenly_subbins =.false. |
| Whether to fill/remove evenly in 1x1 velocity cells in the bin when resamp_tile_size > 1 and upsampling/downsampling. More... | |
| logical | resamp_fill_empty_subbins =.false. |
| Whether to fill all empty 1x1 velocity cells in the bin when resamp_tile_size > 1 and there are already particles in the bin. More... | |
| integer | resamp_fill_empty_subbin_target =1 |
| If resamp_fill_empty_subbins=.true., target of sub bins. More... | |
| integer, dimension(:,:), allocatable | resamp_patches |
| !< node numbers of vertices contained in triangle patch of voronoi cell More... | |
| integer, dimension(:), allocatable | resamp_patch_size |
| total number of patch vertices in Voronoi cell More... | |
| real(kind=8), dimension(:,:), allocatable | resamp_patch_rzdims |
| R-Z boundaries of the Voronoi-triangle patch. More... | |
| integer | cce_buffer_id |
| Index of currently resampled buffer. More... | |
| integer | cce_buffer_first_node |
| integer | cce_buffer_last_node |
| node interval of current buffer More... | |
| integer | cce_grid_nnode |
| Number of node on the grid. More... | |
| integer, parameter | cce_nbconstraint =5 |
| real(kind=8), dimension(:,:,:,:,:,:), allocatable | fbins |
| integer | cce_inode1 |
| integer | cce_inode2 |
| logical | cce_skip_nodes |
Private Attributes | |
| real(kind=8), private | resamp_bin_max_fac |
| threshold for auto downsample More... | |
| real(kind=8), private | resamp_inv_bin_num |
| Inverse of the number of macro bins in v-space. More... | |
| integer, private | resamp_nmu_in |
| Number of cells on original vperp/mu grid (=f0_nmu) More... | |
| integer, private | resamp_nvp_in |
| Number of cells on original v_para grid (=f0_nvp) More... | |
| real(kind=8), private | resamp_dsmu |
| v_perp/mu spacing of original v-grid (=f0_dsmu) More... | |
| real(kind=8), private | resamp_dvp |
| v_para spacing of original v-grid (=f0_dvp) More... | |
| real(kind=8), private | resamp_mu_max |
| v_perp/mu cutoff of v-grid (=f0_mu_max) More... | |
| real(kind=8), private | resamp_vp_max |
| v_para cutoff of v-grid (=f0_vp_max) More... | |
| integer, private | resamp_musize |
| Number of sampling bins in v_perp/mu direction. More... | |
| integer, private | resamp_vpsize |
| Number of sampling bins in v_para direction. More... | |
| integer, private | resamp_inode1 |
| Index of the first node of local patch of the configuration space mesh (=f0_inode1) More... | |
| integer, private | resamp_inode2 |
| Index of the last node of local patch of the configuration space mesh (=f0_inode2) More... | |
| real(kind=8), dimension(:), allocatable, private | resamp_den |
| Reference density for normalization. More... | |
| real(kind=8), dimension(:,:), allocatable, private | resamp_t_ev |
| Reference temperature for normalization. More... | |
| real(kind=8), dimension(:,:), allocatable, private | resamp_inv_ph_vol |
| Inverse phase-space volume element. More... | |
| real(kind=8), dimension(:), allocatable, private | resamp_samp_ratio |
| Upsampling ratio for restarts with finer grid. More... | |
| logical, private | resamp_restart_mode =.false. |
| real(kind=8), dimension(:), allocatable, private | resamp_phvol_sum |
| Sum of phase space volume of all particles on a node. More... | |
| real(kind=8), dimension(:), allocatable, private | resamp_phvol_added |
| Sum of phase space volume of particles added in empty bins. More... | |
| integer, parameter, private | resamp_nongridmom = 8 |
| Number of bin moments to be conserved. More... | |
| integer, dimension(:), allocatable, private | resamp_off_vgrid |
| Counter for number of particles that are not on the v-grid. More... | |
| integer, dimension(:), allocatable, private | resamp_on_vgrid |
| Counter for number of particles that are on the v-grid. More... | |
| integer, private | resamp_empty_bins |
| Number of empty sampling bins. More... | |
| integer, private | resamp_resampled_bins |
| Number of bins that were picked for re-/up-/down-sampling. More... | |
| integer, private | resamp_failed_bins |
| Number of bins for which re-/up-/down-sampling failed. More... | |
| integer, private | resamp_var_resample |
| Number of variance resampled bins. More... | |
| integer, private | resamp_resample_failed |
| Number of bins for which variance resampling failed. More... | |
| integer, private | resamp_auto_upsample |
| Number of upsampled bins. More... | |
| integer, private | resamp_upsample_failed |
| Number of bins for which upsampling failed. More... | |
| integer, private | resamp_auto_downsample |
| Number of downsampled bins. More... | |
| integer, private | resamp_downsample_failed |
| Number of bins for which downsampling failed. More... | |
| integer, private | resamp_retried_bins |
| Number of bins retried after failure. More... | |
| integer, private | resamp_retried_failed |
| Number of bins failed again after retry. More... | |
| integer, private | resamp_fullf_failed |
| Number of full-f failed bins. More... | |
| integer, private | resamp_wrong_particles |
| Number of particles in wwrong bins. More... | |
| real(kind=8), private | resamp_smu_fac |
| Factor used in the context of tapering off the marker density at high velocity. More... | |
| real(kind=8), private | resamp_vp_fac |
| Factor used in the context of tapering off the marker density at high velocity. More... | |
| subroutine resamp_module::add_to_bin | ( | type(resamp_bin_type), intent(inout) | bin, |
| type(ptl_type), intent(in) | ptli, | ||
| integer, intent(in) | ind | ||
| ) |
Adds the data of one particle to a phase-space bin.
| [in,out] | bin | Phase-space bin data structure, type(resamp_bin_type) |
| [in] | ptli | Data of a single particle, type(ptl_type) |
| [in] | ind | Index of that particle in the corresponding species data structure, integer |
| subroutine resamp_module::build_bins | ( | type(species_type), intent(inout) | sp, |
| type(resamp_bin_type), dimension(resamp_inode1:resamp_inode2,resamp_musize+1,0:resamp_vpsize+1), intent(inout) | bins, | ||
| type(grid_type), intent(in) | grid, | ||
| logical, intent(in) | cce_sync | ||
| ) |
Loops over particles, puts them in the appropriate bin, computes partial mean This routine sorts particles in the appropriate phase-space bin and computes the total charge in the bin.
| [in,out] | sp | Particle data structure for one species, type(species_type) |
| [in,out] | bins | Phase-space bin data structure, type(resamp_bin_type) |
| [in] | grid | XGC grid data structure, type(grid_type) |
| subroutine resamp_module::cce_buffer_rcv_updated_bins | ( | integer, intent(in) | isp | ) |
| subroutine resamp_module::cce_buffer_snd_updated_bins | ( | integer, intent(in) | isp | ) |
| subroutine resamp_module::cce_fcoupling_resample | ( | type(grid_type), pointer | grid, |
| type(psn_type), pointer | psn, | ||
| type(species_type), dimension(:), pointer | spall, | ||
| integer, intent(in) | istep | ||
| ) |
| subroutine resamp_module::cce_reset_particle_outside | ( | type(grid_type), intent(in) | grid, |
| type(species_type), intent(out) | sp | ||
| ) |
| subroutine resamp_module::check_bin | ( | type(resamp_bin_type), intent(inout) | bin, |
| type(species_type), intent(in) | sp, | ||
| integer, intent(in) | cce_sync_sendrcv, | ||
| integer, intent(inout) | resamp_auto_upsample_loc, | ||
| integer, intent(inout) | resamp_auto_downsample_loc, | ||
| integer, intent(inout) | resamp_var_resample_loc | ||
| ) |
This routine checks whether a bin needs to be resampled and determines the type of resampling (up-/down-/re-sampling). The chosen resampling option is stored in the bin data structure: 0) No resampling needed 1) Resampling due to high weight variance 2) Upsampling due to low particle count 3) Downsampling due to high particle count.
| [in,out] | bin | Phase-space bin data structure, type(resamp_bin_type) |
| [in] | sp | Particle data, type(species_type) |
| [in] | cce_sync_sendrcv | for the coupling |
| [in,out] | resamp_auto_upsample_loc | Counter for upsampled bins |
| [in,out] | resamp_auto_downsample_loc | Counter for downsampled bins |
| [in,out] | resamp_var_resample_loc | Counter for variance-resampled bins |
We must pass all resamp counters to be updated correctly with OMP.
| subroutine resamp_module::destroy_bins | ( | type(resamp_bin_type), dimension(resamp_inode1:resamp_inode2,1:resamp_musize+1,0:resamp_vpsize+1), intent(inout) | bins | ) |
Cleans up bin memory without deallocating the array of bins itself (can be kept for next species)
| [in,out] | bins | Bin data structure, type(resamp_bin_type) |
| subroutine resamp_module::diag_velocity_space_filled | ( | type(grid_type), intent(in) | grid, |
| type(species_type), intent(in) | sp, | ||
| integer, intent(in) | tile_size_in | ||
| ) |
This is a diagnostics routine. Loops over particles to check how many 3D real-space mesh nodes are fully populated by particles in velocity space and prints to the log.
| [in] | grid | XGC grid data structure, type(grid_type) |
| [in] | sp | Particle data structure for one species, type(species_type) |
| [in] | tile_size_in | Bin size on the velocity space grid in cells |
| subroutine resamp_module::f0_upsampler | ( | type(grid_type), intent(in) | grid, |
| type(psn_type), intent(in) | psn, | ||
| type(species_type), dimension(0:ptl_nsp_max), intent(inout) | spall | ||
| ) |
Tries to upsample so that velocity space is fully populated. Called when f0_upsample = .true., typically to fill up velocity space before grid operations when using the pseudo-inverse interpolation with f0_velocity_interp_use_pseudo_inv = .true. to ensure that the pseudo-inverse calculation will converge.
| [in] | grid | XGC grid data structure, type(grid_type) |
| [in] | psn | XGC E-field/potential data, type(psn_type) |
| [in,out] | spall | Particle data, type(species_type) |
| subroutine resamp_module::get_marker_velo | ( | type(grid_type), intent(in) | grid, |
| type(psn_type), intent(in) | psn, | ||
| type(ptl_type), intent(in) | ptli, | ||
| integer, intent(in) | isp, | ||
| real (kind=8), dimension(4), intent(out) | velo, | ||
| real (kind=8), intent(out) | vrad | ||
| ) |
Evaluates the marker particle equations of motion. The results are the time derivatives of the marker position and parallel velocity (or rho, to be precise), as well as the radial velocity v.grad(psi). The results can be used in the moment preserving QP weight optimization.
| [in] | grid | XGC configuration space grid, type(grid_type) |
| [in] | psn | E-field/potential data structure, type(psn_type) |
| [in] | ptli | Particle data (phase+constants), type(ptl_type) |
| [in] | isp | Particle species index, integer |
| [out] | velo | dX/dt and drho/dt, real(8) |
| [out] | vrad | Radial particle velocity (dX/dt).grad(psi) |
| subroutine resamp_module::get_new_fullf_weight | ( | type(grid_type), intent(in) | grid, |
| type(psn_type), intent(in) | psn, | ||
| type(ptl_type), intent(inout) | part, | ||
| real (8), intent(in) | psi, | ||
| real (8), intent(in) | B, | ||
| real (8), intent(in) | flow, | ||
| real (8), intent(in) | den, | ||
| real (8), intent(in) | temp_ev, | ||
| integer, intent(in) | spec, | ||
| integer, intent(in) | tri, | ||
| real (8), dimension(3), intent(in) | bary, | ||
| real (8), intent(in) | mark_den, | ||
| logical, intent(out) | err | ||
| ) |
Compute the new full-f weight of the input particle \(w_0 = f/g\). Note that either \(w_0\) or \(f\) can be chosen freely, while the particle's phasespace volume \(\delta V = 1/g\) is determined by Liousville's theorem of phase space volume conservation. The distribution function \(f\) can in principle be anything here as long as the \(w_2\) weight is set accordingly. The goal is to choose \(w_0\) and \(w_2\) such that the weight evolution equation does not have to be changed. This is from Seung-Hoe Ku's notes: \(f_p = w_1 w_0 g\) \(\frac{\mathrm{D}f_p}{\mathrm{D}t} = w_0 g \frac{\mathrm{d}w_1}{\mathrm{d}t}\) \(=-\frac{\mathrm{D}f_0}{\mathrm{D}t} + S\) \(w_1(t_2)-w_1(t_1) = -\frac{f_0(t_2)-f_0(t_1)}{w_0 g}\) \(w_2(t_2)-w_2(t=0) = -\frac{f_0(t_2)-f_0(t=0)}{w_0 g}\) Easiest choice: \(w_0 g = f_0(t=0),\quad w_2(t=0) = 0\) Then: \(1-w_2 = \frac{f_0(t)}{w_0 g} = \frac{f_0(t)}{f_0(t=0)}\) and \(w_1(t_2)-w_1(t_1) = -\frac{f_0(t_2)}{f_0(t=0)} + (1-w_2(t_1))\).
Over time \(f_0(t) = f_M + f_g(t)\) can become negative, so \(w_0=f_0(t)/g\) with \(w_2(t)=0\) would be negative. But we want positive full-f weights. Therefore, we choose \(w_0=\max(f_0(t_{resample}),\alpha f_M)/g > 0\). This yields \(1-w_2(t) = 1-w_2(t_{resample})-\frac{f_0(t_{resample})}{w_0 g} + \frac{f_0(t)}{w_0 g}\) In order to get \(1-w_2 = \frac{f_0(t)}{w_0 g}\), we need to set \(w_2(t_{resample})\) such that \(1- w_2(t_{resample})-\frac{f_0(t_{resample})}{w_0 g}=0\).
| [in] | grid | XGC grid data structure, type(grid_type) |
| [in] | psn | Field data, type(psn_type) |
| [in,out] | part | Input particle data, type(ptl_type) |
| [in] | psi | Poloidal flux at particle position, real(8) |
| [in] | B | Magnetic field |B_0|, real(8) |
| [in] | flow | Background mean parallel flow, real(8) |
| [in] | den | Background density, real(8) |
| [in] | temp_ev | Background temperature |
| [in] | spec | Species type of input particle, integer |
| [in] | tri | Triangle in which the particle is, integer |
| [in] | bary | Corresponding barycentric coordinates, real(8) |
| [in] | mark_den | Inverse of particle's phase space volume g=1/dV, real(8) |
| [out] | err | Error flag, logical |
| subroutine resamp_module::get_sampling_ratio | ( | type(grid_type), intent(in) | grid | ) |
This routine upsamples the marker population in spall for a new configuration space grid. The new grid should be denser than the current grid, i.e. require upsampling, not downsampling. This is due to particle data locality. Upsampling to a denser mesh is done on the coarse mesh, while downsampling from a fine mesh is better done with particle data arranged for the coarse mesh. The routine reads in the vertex positions of the new mesh and runs a search for each of them to associate them with the vertices of the current mesh. The ratio of new mesh vertices per current mesh vertex is the upsampling ratio. If the upsampling ratio is smaller than unity, that vertex is skipped.
| [in] | grid | XGC configuration space grid, type(grid_type) |
| logical function resamp_module::is_first_call | ( | ) |
| subroutine resamp_module::load_new_ptl | ( | type(grid_type), intent(in) | grid, |
| type(ptl_type), intent(inout) | new_ptl, | ||
| integer, dimension(3), intent(in) | bin_id, | ||
| integer, intent(out) | itr, | ||
| real (kind=8), dimension(3), intent(out) | p, | ||
| real (kind=8), intent(out) | psi | ||
| ) |
This routine loads new particles into a bin with uniform marker density in configuration space. Particles are drawn from a rectacngle enclosing the bin volume and are accepted if inside the Voronoi volume of the bin. A more efficient loading can be achieved by pre-selecting the triangle in which to load. This functionality will be provided in a separate function.
| [in] | grid | XGC configuration space grid, type(grid_type) |
| [in,out] | new_ptl | Particle to store data in |
| [in] | bin_id | ID of the bin into which particle is loaded |
| [out] | itr | Triangle into which the particle was loaded, integer |
| [out] | p | Barycentric coordinates of the loaded triangle, real(8) |
| [out] | psi | Pol. flux at new particle's location, real(8) |
| subroutine resamp_module::load_new_ptl2 | ( | type(grid_type), intent(in) | grid, |
| type(ptl_type), intent(inout) | new_ptl, | ||
| integer, dimension(3), intent(in) | bin_id, | ||
| integer, intent(out) | itr, | ||
| real (kind=8), dimension(3), intent(out) | p, | ||
| real (kind=8), intent(out) | psi | ||
| ) |
This routine loads new particles into a bin with uniform marker density in configuration space. Particles are drawn from the bin volume. First, a triangle is chosen using importance sampling with the triangle area as criterion. Then a random position in the triangle is drawn. The drawn position is accepted if inside the Voronoi volume of the bin.
| [in] | grid | XGC configuration space grid, type(grid_type) |
| [in,out] | new_ptl | Particle to store data in |
| [in] | bin_id | ID of the bin into which particle is loaded |
| [out] | itr | Triangle into which the particle was loaded, integer |
| [out] | p | Barycentric coordinates of the loaded triangle, real(8) |
| [out] | psi | Pol. flux at new particle's location, real(8) |
| subroutine resamp_module::resamp_scale_phvol | ( | type(grid_type), intent(in) | grid, |
| type(species_type), intent(inout) | sp | ||
| ) |
Loops over particles and scales their phase space volume to account for particles added to empty bins.
| [in] | grid | XGC grid data structure, type(grid_type) |
| [in,out] | sp | Particle data structure for one species, type(species_type) |
| subroutine resamp_module::resample | ( | type (grid_type), intent(in) | grid, |
| type (species_type), dimension(0:ptl_nsp_max), intent(inout) | spall, | ||
| type (psn_type), intent(in) | psn, | ||
| integer, intent(in) | inode1, | ||
| integer, intent(in) | inode2, | ||
| real (kind=8), intent(in) | dsmu, | ||
| real (kind=8), intent(in) | dvp, | ||
| integer, intent(in) | nmu, | ||
| integer, intent(in) | nvp, | ||
| real (kind=8), intent(in) | vp_max, | ||
| real (kind=8), intent(in) | mu_max, | ||
| integer, intent(in) | cce_sync_sendrcv, | ||
| integer, optional | tile_size | ||
| ) |
Resample the marker particles in XGC to improve phase-space coverage and reduce marker noise.
| [in] | grid | XGC grid data structure, type(grid_type) |
| [in,out] | spall | Particle data, type(species_type) |
| [in] | psn | Field data, type(psn_type) |
| [in] | inode1 | Index of first mesh vertex in domain, integer |
| [in] | inode2 | Index of last mesh vertex in domain, integer |
| [in] | dsmu | v_perp/mu resolution of XGC's total-f velocity grid, integer |
| [in] | dvp | v_para resolution of XGC's total-f v-grid, integer |
| [in] | nmu | Number of v_perp/mu bins in total-f v-grid, integer |
| [in] | nvp | Half the number of v_para bins in total-f v-grid, integer |
| [in] | vp_max | v_para cutoff of total-f v-grid, real(8) |
| [in] | mu_max | v_perp/mu cutoff of total-f v-grid, real(8) |
| [in] | tile_size | Resampling combines tile_size x tile_size bins of the total-f v-grid to one macro bin, integer, optional |
| [in] | cce_sendrcv | Takes values 0(nothing), 1(sender) , 2(receiver) |
| subroutine resamp_module::resample_bin | ( | type (resamp_bin_type), intent(inout) | bin, |
| type (grid_type), intent(in) | grid, | ||
| type (species_type), intent(inout) | sp, | ||
| type (psn_type), intent(in) | psn, | ||
| integer, intent(in) | cce_sync_sendrcv, | ||
| integer, intent(in) | start_wr_pos, | ||
| integer, intent(inout) | resamp_failed_bins_loc, | ||
| integer, intent(inout) | resamp_upsample_failed_loc, | ||
| integer, intent(inout) | resamp_downsample_failed_loc, | ||
| integer, intent(inout) | resamp_resample_failed_loc, | ||
| integer, intent(inout) | resamp_retried_bins_loc, | ||
| integer, intent(inout) | resamp_retried_failed_loc, | ||
| integer, intent(inout) | resamp_fullf_failed_loc, | ||
| real (8), intent(inout) | resamp_phvol_added_loc | ||
| ) |
The actual process of resampling the particle set in a bin happens in this routine. A bin is passed as input, the pre-determined type of resampling is executed and the weights of the new particle set are computed with a QP optimization. If the optimization is successful, the old particle set in the input variable "sp" is replaced with the new particle set.
| [in,out] | bin | Phase-space bin data structure, type(resamp_bin_type) |
| [in] | grid | XGC grid data structure, type(grid_type) |
| [in,out] | sp | Particle data structure, type(species_type) |
| [in] | psn | Field data, type(psn_type) |
| [in] | cce_sync_sendrcv | 0:no f-coupling, 1 sender, 2 receiver |
| [in] | start_wr_pos | Write position in sp, integer |
| [in,out] | resamp_failed_bins_loc | Counter for failed bins |
| [in,out] | resamp_upsample_failed_loc | Counter for failed upsampled bins |
| [in,out] | resamp_downsample_failed_loc | Counter for failed downsampled bins |
| [in,out] | resamp_resample_failed_loc | Counter for failed var-resampled bins |
| [in,out] | resamp_retried_bins_loc | Counter for retried bins |
| [in,out] | resamp_retried_failed_loc | Counter for bins that failed on retry |
| [in,out] | resamp_fullf_failed_loc | Counter for bins with failed full-f resampling |
| [in,out] | resamp_phvol_added_loc | Phase space volume of particles added to empty bins |
We must pass all resamp counters to be updated correctly with OMP.
| subroutine resamp_module::resample_finalize | ( | type(resamp_bin_type), dimension(:,:,:), allocatable | bins | ) |
This routine cleans up the resampling module.
| [in,out] | bins | Resampling bin data structure, type(resamp_bin_type) |
| subroutine resamp_module::resample_init | ( | type(grid_type), intent(in) | grid, |
| integer, intent(in) | inode1, | ||
| integer, intent(in) | inode2, | ||
| real (kind=8), intent(in) | dsmu, | ||
| real (kind=8), intent(in) | dvp, | ||
| integer, intent(in) | nmu, | ||
| integer, intent(in) | nvp, | ||
| real (kind=8), intent(in) | vp_max, | ||
| real (kind=8), intent(in) | mu_max, | ||
| integer, intent(in) | tile_size, | ||
| type(resamp_bin_type), dimension(:,:,:), allocatable | bins, | ||
| integer, intent(in) | cce_sync_sendrcv | ||
| ) |
This routine initializes the resampling module and sets some global parameters such as v-grid size, etc.
| [in] | grid | XGC mesh data structure, type(grid_type) |
| [in] | inode1 | Index of first vertex in the mesh domain, integer |
| [in] | inode2 | Index of last vertex in the mesh domain, integer |
| [in] | dsmu | v_perp/mu resolution of XGC's total-f velocity grid, integer |
| [in] | dvp | v_para resolution of XGC's total-f v-grid, integer |
| [in] | nmu | Number of v_perp/mu bins in total-f v-grid, integer |
| [in] | nvp | Half the number of v_para bins in total-f v-grid, integer |
| [in] | vp_max | v_para cutoff of total-f v-grid, real(8) |
| [in] | mu_max | v_perp/mu cutoff of total-f v-grid, real(8) |
| [in] | tile_size | Resampling combines tile_size x tile_size bins of the total-f v-grid to one macro bin, integer |
| [in,out] | bins | Resampling phase space bins, type(resamp_bin_type) |
| subroutine resamp_module::restart_resample | ( | ) |
Move phase space density from the 5D grid over to the particles for restarting simulatin with a different grid.
| subroutine resamp_module::set_cce_buffer_and_overlap_id | ( | integer, intent(in) | id1, |
| integer, intent(in) | id2, | ||
| integer, intent(in) | inode1, | ||
| integer, intent(in) | inode2 | ||
| ) |
| subroutine resamp_module::set_cce_buffer_id | ( | integer, intent(in) | id, |
| integer, intent(in) | inode1, | ||
| integer, intent(in) | inode2 | ||
| ) |
| subroutine resamp_module::setup_fullf | ( | type(ptl_type), intent(in) | ptli, |
| integer, intent(in) | spec_type, | ||
| real (8), intent(out) | den, | ||
| real (8), intent(out) | flow, | ||
| real (8), intent(out) | temp_ev, | ||
| real (8), dimension(3), intent(out) | bvec, | ||
| real (8), intent(out) | B, | ||
| real (8), intent(out) | psi | ||
| ) |
Evaluate the background density, temperature, flow, magnetic field and psi at the input particle's position.
| [in] | ptli | Input particle data, type(ptl_type) |
| in[ | spec_type Species type of input particle, integer | |
| [out] | den | Background density, real(8) |
| [out] | flow | Background toroidal flow velocity, real(8) |
| [out] | temp_ev | Background temperature, real(8) |
| [out] | bvec | Magnetic field vector B, real(8) |
| [out] | B | Magnetic field |B_0|, real(8) |
| [out] | psi | Poloidal magnetic flux, real(8) |
| subroutine resamp_module::setup_opt | ( | integer, intent(in) | nconstraint, |
| integer, intent(in) | part_num, | ||
| real (8), dimension(nconstraint,part_num), intent(in) | Cmat, | ||
| real (8), dimension(nconstraint), intent(inout) | constraint, | ||
| real (8), dimension(part_num), intent(inout) | weights, | ||
| integer, intent(in) | n_eq_bin, | ||
| integer, intent(in) | n_ineq_bin, | ||
| integer, dimension(resamp_nongridmom), intent(in) | eq_bin, | ||
| integer, dimension(resamp_nongridmom), intent(in) | ineq_bin, | ||
| logical, intent(in) | grid_ineq_on, | ||
| real (8), intent(in) | ineq_tol, | ||
| logical, intent(in) | positivity, | ||
| integer, intent(out) | ierr | ||
| ) |
This routine sets up the input for the QP optimization that computes the delta-f weights of the new particles. There are two types of constraints, bin and grid constraints. Bin constraints correspond to moments of the distribution function evaluated with nearest neighbor interpolation on the configuration space grid. Grid constraints correspond to moments evaluated with linear (barycentric) interpolation on the conf. space grid. Bin constraints can be enforced as equality and inequality constraints: use idx_constraint and idx_constraint2 to choose which constraint is an equality and which is an inequality constraint. Grid constraints are enforced as inequality constraints. Since grid constraints make the optimization problem much harder due to their large number, using them is optional (grid_ineq_on) A positivity constraint for the weights can be activated with the variable "positivity". This is used for full-f optimization.
| [in] | nconstraint | Number of constraints, integer |
| [in] | part_num | Number of particles, integer |
| [in] | Cmat | Constraint matrix, N_constraint x N_ptl, real(8) |
| [in] | constraint | Values of the constraints, real(8) |
| [in,out] | weights | Delta-f weights w1, N_ptl, real(8) |
| [in] | grid_ineq_on | Whether to conserve the grid charge or not, logical |
| [in] | ineq_tol | Error tolerance for the inequality constraints, real(8) |
| [in] | positivity | Whether to require weight>0, logical |
| [in,out] | ierr | Error flag, integer |
| subroutine resamp_module::update_cmat | ( | type(grid_type), intent(in) | grid, |
| type(psn_type), intent(in) | psn, | ||
| integer, intent(in) | nconstraint, | ||
| integer, intent(in) | nptl, | ||
| integer, intent(in) | vert, | ||
| integer, dimension(1:vert), intent(in) | patch, | ||
| real (8), dimension(3), intent(in) | bary, | ||
| integer, intent(in) | tri, | ||
| real (8), intent(in) | psi, | ||
| real (8), intent(in) | B, | ||
| real (8), dimension(3), intent(in) | bvec, | ||
| integer, intent(in) | spec, | ||
| integer, intent(in) | pindex, | ||
| type(ptl_type), intent(in) | part, | ||
| real (8), dimension(nconstraint,nptl), intent(out) | Cmat, | ||
| integer, intent(in) | op_mode | ||
| ) |
This subroutine sets up the matrix C (constraint matrix) with the coefficients needed to compute the conserved quantities charge, parallel momentum, magnetic moment, parallel and perp. kinetic energy and the grid charge on the vertices connected to the bin. Those quantities are then obtained by multiplying the matrix C with the delta-f (–> w1) weight vector from the right. This matrix forms the basis for the QP optimization program that finds the delta-f weights for the resampled particle set.
| [in] | grid | XGC grid data structure, type(grid_type) |
| [in] | psn | XGC E-field/potential data, type(psn_type) |
| [in] | nconstraint | Number of equality+inequality constraints, integer |
| [in] | nptl | Number of particles in bin, integer |
| [in] | vert | Number of vertices connected to this bin (patch size), integer |
| [in] | patch | Indices of the vertices connected to this bin, integer |
| [in] | bary | Barycentric coordinates of the particle, real(8) |
| [in] | tri | Triangle in which the particle is, integer |
| [in] | psi | Pol. flux at particle position, real(8) |
| [in] | B | Local magnetic field |B|, real(8) |
| [in] | bvec | Magnetic field vector B, real(8) |
| [in] | spec | Species index, integer |
| [in] | pindex | Position of the particle in the particle set, integer |
| [in] | part | Data of one particle, type(ptl_type) |
| [in,out] | Cmat | The matrix C from the (see documentation), real(8) |
| [in] | op_mode | Switch for optimizing full-f weights w0 (0)or delta-f weights (1), integer |
| integer resamp_module::cce_buffer_first_node |
| integer resamp_module::cce_buffer_id |
Index of currently resampled buffer.
| integer resamp_module::cce_buffer_last_node |
node interval of current buffer
| integer resamp_module::cce_grid_nnode |
Number of node on the grid.
| integer resamp_module::cce_inode1 |
| integer resamp_module::cce_inode2 |
| integer, parameter resamp_module::cce_nbconstraint =5 |
| logical resamp_module::cce_skip_nodes |
| real (kind=8), dimension(:,:,:,:,:,:), allocatable resamp_module::fbins |
|
private |
Number of downsampled bins.
|
private |
Number of upsampled bins.
|
private |
threshold for auto downsample
|
private |
Reference density for normalization.
| logical resamp_module::resamp_distribute_evenly_subbins =.false. |
Whether to fill/remove evenly in 1x1 velocity cells in the bin when resamp_tile_size > 1 and upsampling/downsampling.
|
private |
Number of bins for which downsampling failed.
|
private |
v_perp/mu spacing of original v-grid (=f0_dsmu)
|
private |
v_para spacing of original v-grid (=f0_dvp)
|
private |
Number of empty sampling bins.
|
private |
Number of bins for which re-/up-/down-sampling failed.
| logical resamp_module::resamp_fill_empty =.false. |
Whether to fill empty bins.
| integer resamp_module::resamp_fill_empty_subbin_target =1 |
If resamp_fill_empty_subbins=.true., target of sub bins.
| logical resamp_module::resamp_fill_empty_subbins =.false. |
Whether to fill all empty 1x1 velocity cells in the bin when resamp_tile_size > 1 and there are already particles in the bin.
|
private |
Number of full-f failed bins.
| logical resamp_module::resamp_fullf_on =.false. |
Whether to resample the full-f weights in addition to delta-f weights.
| logical resamp_module::resamp_grid_ineq_on =.false. |
Switch for using inequality constraints for the grid charge for resampling.
| real (8) resamp_module::resamp_highv_max = 10D0 |
energy cutoff of the high velocity bins v_para>f0_vp_max and v_perp>f0_smu_max
| real (8) resamp_module::resamp_highv_max_ratio = 4D0 |
Downsampling threshold for high-velocity bins.
| real (8) resamp_module::resamp_ineq_tol = 1D-7 |
Threshold for relative error in the inequality constraints in the QP optimization.
| real (8) resamp_module::resamp_ineq_tol_max = 1D-3 |
Maximal threshold for relative error in inequality constraints for retried bins.
|
private |
Index of the first node of local patch of the configuration space mesh (=f0_inode1)
|
private |
Index of the last node of local patch of the configuration space mesh (=f0_inode2)
|
private |
Inverse of the number of macro bins in v-space.
|
private |
Inverse phase-space volume element.
| real (8) resamp_module::resamp_max_ratio = 2.0D0 |
max ratio of (# of ptl)/(target # of ptl) in bin for auto-downsample
| integer resamp_module::resamp_max_target = 4 |
Overrides the number of constraints in determining the target # of ptl of a bin.
| real (8) resamp_module::resamp_min_ratio = 0.2D0 |
min ratio of (# of ptl)/(target # of ptl) in bin for auto-upsample
|
private |
v_perp/mu cutoff of v-grid (=f0_mu_max)
|
private |
Number of sampling bins in v_perp/mu direction.
|
private |
Number of cells on original vperp/mu grid (=f0_nmu)
| character (len=200) resamp_module::resamp_node_file |
File containing the vertex positions of the mesh for which to resample.
|
private |
Number of bin moments to be conserved.
| integer resamp_module::resamp_nphi_new =1 |
Number of poloidal planes in simulation with new mesh.
|
private |
Number of cells on original v_para grid (=f0_nvp)
|
private |
Counter for number of particles that are not on the v-grid.
|
private |
Counter for number of particles that are on the v-grid.
| real (kind=8), dimension(:,:), allocatable resamp_module::resamp_patch_rzdims |
R-Z boundaries of the Voronoi-triangle patch.
| integer, dimension(:), allocatable resamp_module::resamp_patch_size |
total number of patch vertices in Voronoi cell
| integer, dimension(:,:), allocatable resamp_module::resamp_patches |
!< node numbers of vertices contained in triangle patch of voronoi cell
|
private |
Sum of phase space volume of particles added in empty bins.
|
private |
Sum of phase space volume of all particles on a node.
| integer resamp_module::resamp_rate = 2 |
timesteps between subsequent resamples (multiplied by sml_f_source_period
|
private |
Number of bins for which variance resampling failed.
|
private |
Number of bins that were picked for re-/up-/down-sampling.
| logical resamp_module::resamp_restart = .false. |
Perform resampling before dumping the final restart file.
|
private |
| logical resamp_module::resamp_restart_read =.false. |
Whether to read a restart file written from a simulation with different grid.
|
private |
Number of bins retried after failure.
|
private |
Number of bins failed again after retry.
| logical resamp_module::resamp_retry = .false. |
Retry QP optimization for failed bins with relaxed inequality constraints.
|
private |
Upsampling ratio for restarts with finer grid.
|
private |
Factor used in the context of tapering off the marker density at high velocity.
|
private |
Reference temperature for normalization.
| integer resamp_module::resamp_tile_size = 2 |
Bin size on the velocity space grid in cells (not vertices) (input parameter)
| integer resamp_module::resamp_tile_size_now = 2 |
Used to override resamp_tile_size.
|
private |
Number of bins for which upsampling failed.
| real (8) resamp_module::resamp_var = 0.02D0 |
threshold for relative standard deviation in bin for auto-resample
|
private |
Number of variance resampled bins.
|
private |
Factor used in the context of tapering off the marker density at high velocity.
|
private |
v_para cutoff of v-grid (=f0_vp_max)
|
private |
Number of sampling bins in v_para direction.
|
private |
Number of particles in wwrong bins.
1.8.5
