Running SIMPLE¶

This page covers the settings of a run. Field inputs are in Field sources and coordinates, outputs in Outputs.

Command line and threads¶

OMP_NUM_THREADS=16 simple.x [config_file]

simple.x reads simple.in or the given file, resolves relative input paths against the working directory and writes all results there. Markers are distributed over OpenMP threads. On multi-socket nodes, OMP_PROC_BIND=spread OMP_PLACES=cores keeps threads on separate cores. Field setup before tracing is partly serial; the log reports the time of each phase. GPU runs: GPU tracing; cluster jobs: Clusters.

Time and markers¶

Key	Default	Meaning
`trace_time`	`1d-1`	tracing time in s; markers not lost by then count as confined
`ntestpart`	1024	number of markers \(N\); a loss fraction \(f\) has standard error \(\sqrt{f(1-f)/N}\)
`ntimstep`	10000	output times (macrosteps) up to `trace_time`, rows of `confined_fraction.dat`
`macrostep_time_grid`	`'linear'`	`'log'`: logarithmic spacing (last/first interval 10), resolves early losses
`npoiper2`	256	steps per path length \(2\pi R_0\) at speed \(v_0\)
`contr_pp`	`-1d0`	markers with \(\kappa \le\) `contr_pp` (deeply passing) count as confined untraced (exit code 3); `-1d10` traces all
`notrace_passing`	0	1: all passing markers count as confined
`deterministic`	`.false.`	fixed seed `ran_seed` (default 42) for sampling and collisions

Confidence intervals: Plotting results. Fewer steps: Numerical settings. \(\kappa\) (trapping parameter): Guiding-centre model.

Species and energy¶

n_e (charge number, default 2), n_d (mass number, default 4d0) and facE_al (default 1d0, energy \(E_0 = 3.5\,\text{MeV}/\) facE_al) set the species: 100 keV protons are n_e = 1, n_d = 1d0, facE_al = 35d0 (collisionless only, see below).

Start positions¶

`startmode`	Start positions
1 (default)	Random points on one field line on the surface `sbeg(1)` (default 0.5), started at `thetabeg`, `phibeg` and followed for `nper` field periods; volume-weighted. `grid_density` in (0, 1) places \((\text{int}(1/\text{grid\_density} - 1))^2\) markers on a \((\vartheta, \varphi)\) grid instead and overrides `ntestpart`.
2	Read `start.dat`: \((s, \vartheta, \varphi, v/v_0, \lambda)\) per line, reference coordinates.
3	Read an ANTS-format file (`start_ants.dat` with `special_ants_file = .true.`, else `start.dat`).
4	Random batch of `batch_size` lines of `start.dat`.
5	Uniform random \(s\) in the range set by `num_surf` (0: \([0,1]\), 1: \([0,\) `sbeg(1)` \(]\), 2: \([\) `sbeg(1)`, `sbeg(2)` \(]\)), clamped to \(s \ge 0.01\); random angles.
6	Read `start.dat` with positions in integration coordinates.
7	D-T fusion births from `&profiles`: Fusion-weighted alpha births.

Modes 1, 5 and 7 write start.dat; generate_start_only = .true. stops after that. Generated markers have \(v/v_0 = 1\) and pitch \(\lambda = v_\parallel/v\) uniform in \([-1, 1]\). Angles are in radians and are not wrapped.

Integrators¶

`integmode`	Method	Order	Notes
1 (default)	symplectic Euler, explicit-implicit	1	cheapest per step
2	symplectic Euler, implicit-explicit	1	adjoint of 1
3	implicit midpoint	2	symmetric
4–7	Gauss-Legendre, 1–4 stages	2, 4, 6, 8	one Newton solve for all stages
15	Lobatto IIIA-IIIB, 3 stages	4	partitioned
0	adaptive Runge-Kutta 4(5)	adaptive	not symplectic; any `integ_coords`
-1	as 0	adaptive	skips loading `field_input`

integmode > 0 needs canonical coordinates (integ_coords not 1 or 5). Implicit steps are solved by Newton iteration to relative tolerance relerr (default 1d-13), which also sets the Runge-Kutta tolerance.

symplectic_newton_warning_mode = .true. (default) accepts bounded Newton corrections at the iteration limit and retries a failed step with smaller symplectic steps, then with Runge-Kutta. Markers that still fail get a numerical exit code (101–105), loss time NaN and count as unresolved, not lost. .false. ends the marker at the first failure. Tolerances: Numerical settings.

Collisions¶

swcoll = .true. adds Coulomb collisions with electrons and two ion species (am1, am2, Z1, Z2; default D and T). Density and temperature profiles \(f(s)\) come from &profiles, either 'power_series' \(f = f_0 \sum_{n=1}^{10} c_n s^{n-1}\) or 'two_power' \(f = f_0 (1 - s^{p_1})^{p_2}\):

&config
swcoll = .true.
am1 = 2d0, am2 = 3d0, Z1 = 1d0, Z2 = 1d0     ! D and T
/
&profiles
profile_type = 'two_power'
Te_scale  = 1.5d4, Te_p1  = 2d0, Te_p2  = 1d0    ! eV
Ti1_scale = 1.5d4, Ti1_p1 = 2d0, Ti1_p2 = 1d0
Ti2_scale = 1.5d4, Ti2_p1 = 2d0, Ti2_p2 = 1d0
ni1_scale = 5d19,  ni1_p1 = 2d0, ni1_p2 = 0.5d0  ! m^-3
ni2_scale = 5d19,  ni2_p1 = 2d0, ni2_p2 = 0.5d0
/

Limits that change results:

The operator uses alpha mass and charge. n_d must be 4 (other values stop with v0_coll != v0) and n_e is ignored; facE_al sets the energy.
Collisions cannot be combined with classification (tcut > 0, class_plot, fast_class), full orbits or GPU tracing.
trace_time must cover the slowing-down time for complete heat loads. The marker energy at loss is \((v/v_0)^2 E_0\), from column 4 of the end state.

Collision operator: Collisions.

Restart and batches¶

restart = .true. reads times_lost.dat and orbit_exit_code.dat; markers with exit code 0, 1, 2 or 4 keep their result, the others are traced again. batch_size smaller than ntestpart traces a random subset of the start positions, whose indices go to batch.dat (reuse_batch = .true. reads them back).