parameter_description.rst 47.2 KB
 Josh Borrow committed Jan 22, 2019 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 .. Parameter Description Matthieu Schaller, 21st October 2018 .. _Parameters_basics: File format and basic information --------------------------------- The parameter file uses a format similar to the YAML format _ but reduced to only the elements required for the SWIFT parameters. Options are given by a name followed by a column and the value of the parameter: .. code:: YAML  Peter W. Draper committed Feb 04, 2019 16  ICs: santa_barbara.hdf5  Josh Borrow committed Jan 22, 2019 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48  dt_max: 1.5 shift: [2., 4., 5.] Comments can be inserted anywhere and start with a hash: .. code:: YAML # Description of the physics viscosity_alpha: 2.0 dt_max: 1.5 # seconds A typical SWIFT parameter file is split into multiple sections that may or may not be present depending on the different configuration options. The sections start with a label and can contain any number of parameters: .. code:: YAML Cosmology: # Planck13 Omega_m: 0.307 Omega_lambda: 0.693 Omega_b: 0.0455 h: 0.6777 a_begin: 0.0078125 # z = 127 The options can be integer values, floating point numbers, characters or strings. If SWIFT expects a number and string is given, an error will be raised. The code can also read an array of values: .. code:: YAML shift: [2., 4., 5.]  Peter W. Draper committed Feb 04, 2019 49   Josh Borrow committed Jan 22, 2019 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 Some options in the parameter file are optional and when not provided, SWIFT will run with the default value. However, if a compulsory parameter is missing an error will be raised at start-up. Finally, SWIFT outputs two YAML files at the start of a run. The first one used_parameters.yml contains all the parameters that were used for this run, **including all the optional parameters left unspecified with their default values**. This file can be used to start an exact copy of the run. The second file, unused_parameters.yml contains all the values that were not read from the parameter file. This can be used to simplify the parameter file or check that nothing important was ignored (for instance because the code is not configured to use some options). The rest of this page describes all the SWIFT parameters, split by section. A list of all the possible parameters is kept in the file examples/parameter_examples.yml. .. _Parameters_units: Internal Unit System -------------------- The InternalUnitSystem section describes the units used internally by the code. This is the system of units in which all the equations are solved. All physical constants are converted to this system and if the ICs use a different system (see the snapshots' ref:ICs_units_label section of the documentation) the particle quantities will be converted when read in. The system of units is described using the value of the 5 basic units of any system with respect to the CGS system. Instead of using a unit of time we use a unit of velocity as this is more intuitive. Users hence need to provide: * a unit of length: UnitLength_in_cgs, * a unit of mass: UnitMass_in_cgs, * a unit of velocity UnitVelocity_in_cgs, * a unit of electric current UnitCurrent_in_cgs, * a unit of temperature UnitTemp_in_cgs. All these need to be expressed with respect to their cgs counter-part (i.e. :math:cm, :math:g, :math:cm/s, :math:A and :math:K). Recall that there are no h-factors in any of SWIFT's quantities; we, for instance, use :math:cm and not :math:cm/h. For instance to use the commonly adopted system of 10^10 Msun as a unit for mass, mega-parsec as a unit of length and km/s as a unit of speed, we would use: .. code:: YAML # Common unit system for cosmo sims InternalUnitSystem: UnitMass_in_cgs: 1.98848e43 # 10^10 M_sun in grams UnitLength_in_cgs: 3.08567758e24 # 1 Mpc in centimeters UnitVelocity_in_cgs: 1e5 # 1 km/s in centimeters per second UnitCurrent_in_cgs: 1 # 1 Ampere  Peter W. Draper committed Feb 04, 2019 107 108  UnitTemp_in_cgs: 1 # 1 Kelvin  Josh Borrow committed Jan 22, 2019 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 Note that there are currently no variables in any of the SWIFT physics schemes that make use of the unit of electric current. There is also no incentive to use anything else than Kelvin but that makes the whole system consistent with any possible unit system. If one is interested in using the more humorous FFF unit system _ one would use .. code:: YAML # FFF unit system InternalUnitSystem: UnitMass_in_cgs: 40823.3133 # 1 Firkin (fir) in grams UnitLength_in_cgs: 20116.8 # 1 Furlong (fur) in cm UnitVelocity_in_cgs: 0.01663095 # 1 Furlong (fur) per Fortnight (ftn) in cm/s UnitCurrent_in_cgs: 1 # 1 Ampere  Peter W. Draper committed Feb 04, 2019 125  UnitTemp_in_cgs: 1 # 1 Kelvin  Josh Borrow committed Jan 22, 2019 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173  The value of the physical constants in this system is left as an exercise for the reader [#f1]_. .. _Parameters_cosmology: Cosmology --------- When running a cosmological simulation, the section Cosmology sets the values of the cosmological model. The expanded :math:\Lambda\rm{CDM} parameters governing the background evolution of the Universe need to be specified here. These are: * The reduced Hubble constant: :math:h: h, * The matter density parameter :math:\Omega_m: Omega_m, * The cosmological constant density parameter :math:\Omega_\Lambda: Omega_lambda, * The baryon density parameter :math:\Omega_b: Omega_b, * The radiation density parameter :math:\Omega_r: Omega_r. The last parameter can be omitted and will default to :math:\Omega_r = 0. Note that SWIFT will verify on start-up that the matter content of the initial conditions matches the cosmology specified in this section. This section also specifies the start and end of the simulation expressed in terms of scale-factors. The two parameters are: * Initial scale-factor: a_begin, * Final scale-factor: a_end. Two additional optional parameters can be used to change the equation of state of dark energy :math:w(a). We use the evolution law :math:w(a) = w_0 + w_a (1 - a). The two parameters in the YAML file are: * The :math:z=0 dark energy equation of state parameter :math:w_0: w_0 * The dark energy equation of state evolution parameter :math:w_a: w_a If unspecified these parameters default to the default :math:\Lambda\rm{CDM} values of :math:w_0 = -1 and :math:w_a = 0. For a Planck+13 cosmological model (ignoring radiation density as is commonly done) and running from :math:z=127 to :math:z=0, one would hence use the following parameters: .. code:: YAML Cosmology: a_begin: 0.0078125 # z = 127 a_end: 1.0 # z = 0  Peter W. Draper committed Feb 04, 2019 174 175 176  h: 0.6777 Omega_m: 0.307 Omega_lambda: 0.693  Matthieu Schaller committed Aug 24, 2019 177  Omega_b: 0.0482519  Josh Borrow committed Jan 22, 2019 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193  Omega_r: 0. # (Optional) w_0: -1.0 # (Optional) w_a: 0. # (Optional) When running a non-cosmological simulation (i.e. without the -c run-time flag) this section of the YAML file is entirely ignored. .. _Parameters_gravity: Gravity ------- The behaviour of the self-gravity solver can be modified by the parameters provided in the Gravity section. The theory document puts these parameters into the context of the equations being solved. We give a brief overview here.  Matthieu Schaller committed Aug 24, 2019 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 * The Plummer-equivalent co-moving softening length used for all dark matter particles :math:\epsilon_{\rm com,DM}: comoving_DM_softening, * The Plummer-equivalent co-moving softening length used for all baryon particles (gas, stars, BHs) :math:\epsilon_{\rm com,bar}: comoving_baryon_softening, * The Plummer-equivalent maximal physical softening length used for all dark matter particles :math:\epsilon_{\rm max,DM}: max_physical_DM_softening, * The Plummer-equivalent maximal physical softening length used for all baryon particles (gas, stars, BHs) :math:\epsilon_{\rm max,bar}: max_physical_baryon_softening, At any redshift :math:z, the Plummer-equivalent softening length used by the code will be :math:\epsilon=\min(\epsilon_{max}, \frac{\epsilon_{com}}{z+1}). The same calculation is performed independently for the dark matter and baryon particles. All the softening quantities are expressed in internal units. Calculations that only involve DM or baryons can leave the unused quantities out of the parameter file. For non-cosmological runs, only the physical softening lengths need to be supplied. In case of zoom simulations, the softening of the additional, more massive, background particles is specified via the parameter softening_ratio_background. Since these particles will typically have different masses to degrade the resolution away from the zoom region, the particles won't have a single softening value. Instead, we specify the fraction of the mean inter-particle separation to use. The code will then derive the softening length of each particle assuming the mean density of the Universe. That is :math:\epsilon_{\rm background} = f\sqrt[3]{\frac{m}{\Omega_m\rho_{\rm crit}}}, where :math:f is the user-defined value (typically of order 0.05). The accuracy of the gravity calculation is governed by the following two parameters:  Josh Borrow committed Jan 22, 2019 220 221 222  * The opening angle (multipole acceptance criterion) used in the FMM :math:\theta: theta, * The time-step size pre-factor :math:\eta: eta,  Peter W. Draper committed Feb 04, 2019 223   Josh Borrow committed Jan 22, 2019 224 The time-step of a given particle is given by :math:\Delta t =  Matthieu Schaller committed Sep 04, 2019 225 \sqrt{\frac{2\eta\epsilon}{|\overrightarrow{a}|}}, where  Matthieu Schaller committed Jan 29, 2019 226 :math:\overrightarrow{a} is the particle's acceleration. Power et al. (2003) _ recommend using :math:\eta=0.025.  Josh Borrow committed Jan 22, 2019 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 The last tree-related parameter is * The tree rebuild frequency: rebuild_frequency. The tree rebuild frequency is an optional parameter defaulting to :math:0.01. It is used to trigger the re-construction of the tree every time a fraction of the particles have been integrated (kicked) forward in time. Simulations using periodic boundary conditions use additional parameters for the Particle-Mesh part of the calculation. The last three are optional: * The number cells along each axis of the mesh :math:N: mesh_side_length, * The mesh smoothing scale in units of the mesh cell-size :math:a_{\rm smooth}: a_smooth (default: 1.25), * The scale above which the short-range forces are assumed to be 0 (in units of the mesh cell-size multiplied by :math:a_{\rm smooth}) :math:r_{\rm cut,max}: r_cut_max (default: 4.5), * The scale below which the short-range forces are assumed to be exactly Newtonian (in units of the mesh cell-size multiplied by :math:a_{\rm smooth}) :math:r_{\rm cut,min}: r_cut_min (default: 0.1),  Peter W. Draper committed Feb 04, 2019 247   Josh Borrow committed Jan 22, 2019 248 249 250 For most runs, the default values can be used. Only the number of cells along each axis needs to be specified. The remaining three values are best described in the context of the full set of equations in the theory documents.  Peter W. Draper committed Feb 04, 2019 251   Josh Borrow committed Jan 22, 2019 252 253 254 255 As a summary, here are the values used for the EAGLE :math:100^3~{\rm Mpc}^3 simulation: .. code:: YAML  Peter W. Draper committed Feb 04, 2019 256   Josh Borrow committed Jan 22, 2019 257 258  # Parameters for the self-gravity scheme for the EAGLE-100 box Gravity:  Matthieu Schaller committed Aug 24, 2019 259 260  eta: 0.025 theta: 0.7  Peter W. Draper committed Feb 04, 2019 261  mesh_side_length: 512  Matthieu Schaller committed Aug 24, 2019 262 263 264 265 266 267 268 269  comoving_DM_softening: 0.0026994 # 0.7 proper kpc at z=2.8. max_physical_DM_softening: 0.0007 # 0.7 proper kpc comoving_baryon_softening: 0.0026994 # 0.7 proper kpc at z=2.8. max_physical_baryon_softening: 0.0007 # 0.7 proper kpc rebuild_frequency: 0.01 # Default optional value a_smooth: 1.25 # Default optional value r_cut_max: 4.5 # Default optional value r_cut_min: 0.1 # Default optional value  Josh Borrow committed Jan 22, 2019 270 271 272  .. _Parameters_SPH:  Peter W. Draper committed Feb 04, 2019 273   Josh Borrow committed Jan 22, 2019 274 275 276 SPH ---  Josh Borrow committed Sep 02, 2019 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 The SPH section is used to set parameters that describe the SPH calculations. There are some scheme-specific values that are detailed in the :ref:hydro section. The common parameters are detailed below. In all cases, users have to specify two values: * The smoothing length in terms of mean inter-particle separation: resolution_eta * The CFL condition that enters the time-step calculation: CFL_condition These quantities are dimensionless. The first, resolution_eta, specifies how smooth the simulation should be, and is used here instead of the number of neighbours to smooth over as this also takes into account non-uniform particle distributions. A value of 1.2348 gives approximately 48 neighbours in 3D with the cubic spline kernel. More information on the choices behind these parameters can be found in Dehnen & Aly 2012 _. The second quantity, the CFL condition, specifies how accurate the time integration should be and enters as a pre-factor into the hydrodynamics time-step calculation. This factor should be strictly bounded by 0 and 1, and typically takes a value of 0.1 for SPH calculations. The next set of parameters deal with the calculation of the smoothing lengths directly and are all optional: * The (relative) tolerance to converge smoothing lengths within: h_tolerance (Default: 1e-4) * The maximal smoothing length in internal units: h_max (Default: FLT_MAX) * The minimal allowed smoothing length in terms of the gravitational softening: h_min_ratio (Default: 0.0, i.e. no minimum) * The maximal (relative) allowed change in volume over one time-step: max_volume_change (Default: 1.4) * The maximal number of iterations allowed to converge the smoothing lengths: max_ghost_iterations (Default: 30) These parameters all set the accuracy of the smoothing lengths in various ways. The first, the relative tolerance for the smoothing length, specifies the convergence criteria for the smoothing length when using the Newton-Raphson scheme. This works with the maximal number of iterations, max_ghost_iterations (so called because the smoothing length calculation occurs in the ghost task), to ensure that the values of the smoothing lengths are consistent with the local number density. We solve: .. math:: (\eta h_i)^{n_D} = n_i^{-1} with :math:h the smoothing length, :math:n_D the number of spatial dimensions, :math:\eta the value of resolution_eta, and :math:n_i the local number density. We change the value of the smoothing length, :math:h, to be consistent with the number density. The maximal smoothing length, by default, is set to FLT_MAX, and if set prevents the smoothing length from going beyond h_max (in internal units) during the run, irrespective of the above equation. The minimal smoothing length is set in terms of the gravitational softening, h_min_ratio, to prevent the smoothing length from going below this value in dense environments. This will lead to smoothing over more particles than specified by :math:\eta. The final set of parameters in this section determine the initial and minimum temperatures of the particles. * The initial temperature of all particles: initial_temperature (Default: InternalEnergy from the initial conditions) * The minimal temperature of any particle: minimal_temperature (Default: 0) * The mass fraction of hydrogen used to set the initial temperature: H_mass_fraction (Default: 0.755) * The ionization temperature (from neutral to ionized) for primordial gas, again used in this conversion: H_ionization_temperature (Default: 1e4) These parameters, if not present, are set to the default values. The initial temperature is used, along with the hydrogen mass fraction and ionization temperature, to set the initial internal energy per unit mass (or entropy per unit mass) of the particles. Throughout the run, if the temperature of a particle drops below minimal_temperature, the particle has energy added to it such that it remains at that temperature. The run is not terminated prematurely. The temperatures specified in this section are in internal units. The full section to start a typical cosmological run would be: .. code:: YAML SPH: resolution_eta: 1.2 CFL_condition: 0.1 h_tolerance: 1e-4 h_min_ratio: 0.1 initial_temperature: 273 minimal_temperature: 100 H_mass_fraction: 0.755 H_ionization_temperature: 1e4  Josh Borrow committed Jan 22, 2019 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 .. _Parameters_time_integration: Time Integration ---------------- The TimeIntegration section is used to set some general parameters related to time integration. In all cases, users have to provide a minimal and maximal time-step size: * Maximal time-step size: dt_max * Minimal time-step size: dt_min These quantities are expressed in internal units. All particles will have their time-step limited by the maximal value on top of all the other criteria that may apply to them (gravity acceleration, Courant condition, etc.). If a particle demands a time-step size smaller than the minimum, SWIFT will abort with an error message. This is a safe-guard against simulations that would never  Matthieu Schaller committed Mar 08, 2019 391 392 393 394 395 396 397 complete due to the number of steps to run being too large. Note that in cosmological runs, the meaning of these variables changes slightly. They do not correspond to differences in time but in logarithm of the scale-factor. For these runs, the simulation progresses in jumps of :math:\Delta\log(a). dt_max is then the maximally allowed change in :math:\Delta\log(a) allowed for any particle in the simulation. This behaviour mimics the variables of the smae name in the Gadget code.  Josh Borrow committed Jan 22, 2019 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412  When running a non-cosmological simulation, the user also has to provide the time of the start and the time of the end of the simulation: * Start time: time_begin * End time: time_end Both are expressed in internal units. The start time is typically set to 0 but SWIFT can handle any value here. For cosmological runs, these values are ignored and the start- and end-points of the runs are specified by the start and end scale-factors in the cosmology section of the parameter file. Additionally, when running a cosmological volume, advanced users can specify the value of the dimensionless pre-factor entering the time-step condition linked with the motion of particles with respect to the background expansion and mesh  Josh Borrow committed Sep 02, 2019 413 414 415 size. See the theory document for the exact equations. Note that we explicitly ignore the Header/Time attribute in initial conditions files, and only read the start and end times or scale factors from the parameter file.  Josh Borrow committed Jan 22, 2019 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441  * Dimensionless pre-factor of the maximal allowed displacement: max_dt_RMS_factor (default: 0.25) This value rarely needs altering. A full time-step section for a non-cosmological run would be: .. code:: YAML TimeIntegration: time_begin: 0 # Start time in internal units. time_end: 10. # End time in internal units. dt_max: 1e-2 dt_min: 1e-6 Whilst for a cosmological run, one would need: .. code:: YAML TimeIntegration: dt_max: 1e-4 dt_min: 1e-10 max_dt_RMS_factor: 0.25 # Default optional value .. _Parameters_ICs:  Peter W. Draper committed Feb 04, 2019 442   Josh Borrow committed Jan 22, 2019 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 Initial Conditions ------------------ The InitialConditions section of the parameter file contains all the options related to the initial conditions. The main two parameters are * The name of the initial conditions file: file_name, * Whether the problem uses periodic boundary conditions or not: periodic. The file path is relative to where the code is being executed. These parameters can be complemented by some optional values to drive some specific behaviour of the code. * Whether to generate gas particles from the DM particles: generate_gas_in_ics (default: 0), * Whether to activate an additional clean-up of the SPH smoothing lengths: cleanup_smoothing_lengths (default: 0) The procedure used to generate gas particles from the DM ones is outlined in the theory documents and is too long for a full description here. The cleaning of the smoothing lengths is an expensive operation but can be necessary in the cases where the initial conditions are of poor quality and the values of the smoothing lengths are far from the values they should have. When starting from initial conditions created for Gadget, some additional flags can be used to convert the values from h-full to h-free and remove the additional :math:\sqrt{a} in the velocities: * Whether to re-scale all the fields to remove powers of h from the quantities: cleanup_h_factors (default: 0), * Whether to re-scale the velocities to remove the :math:\sqrt{a} assumed by Gadget : cleanup_velocity_factors (default: 0). The h-factors are self-consistently removed according to their units and this is applied to all the quantities irrespective of particle types. The correct power of h is always calculated for each quantity. Finally, SWIFT also offers these options: * A factor to re-scale all the smoothing-lengths by a fixed amount: smoothing_length_scaling (default: 1.), * A shift to apply to all the particles: shift (default: [0.0,0.0,0.0]), * Whether to replicate the box along each axis: replicate (default: 1). The shift is expressed in internal units. The option to replicate the box is especially useful for weak-scaling tests. When set to an integer >1, the box size is multiplied by this integer along each axis and the particles are duplicated and shifted such as to create exact copies of the simulation volume. The full section to start a DM+hydro run from Gadget DM-only ICs would be: .. code:: YAML InitialConditions: file_name: my_ics.hdf5 periodic: 1  Peter W. Draper committed Feb 04, 2019 498 499 500 501  cleanup_h_factors: 1 cleanup_velocity_factors: 1 generate_gas_in_ics: 1 cleanup_smoothing_lengths: 1  Josh Borrow committed Jan 22, 2019 502 503 504  .. _Parameters_constants:  Peter W. Draper committed Feb 04, 2019 505   Josh Borrow committed Jan 22, 2019 506 507 508 509 510 511 Physical Constants ------------------ For some idealised test it can be useful to overwrite the value of some physical constants; in particular the value of the gravitational constant. SWIFT offers an optional parameter to overwrite the value of  Peter W. Draper committed Feb 04, 2019 512 :math:G_N.  Josh Borrow committed Jan 22, 2019 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561  .. code:: YAML PhysicalConstants: G: 1 Note that this set :math:G to the specified value in the internal system of units. Setting a value of 1 when using the system of units (10^10 Msun, Mpc, km/s) will mean that :math:G_N=1 in these units [#f2]_ instead of the normal value :math:G_N=43.00927. This option is only used for specific tests and debugging. This entire section of the YAML file can typically be left out. More constants may be handled in the same way in future versions. .. _Parameters_snapshots: Snapshots --------- The Snapshots section of the parameter file contains all the options related to the dump of simulation outputs in the form of HDF5 :ref:snapshots. The main parameter is the base name that will be used for all the outputs in the run: * The base name of the HDF5 snapshots: basename. This name will then be appended by an under-score and 4 digits followed by .hdf5 (e.g. base_name_1234.hdf5). The 4 digits are used to label the different outputs, starting at 0000. In the default setup the digits simply increase by one for each snapshot. However, if the optional parameter int_time_label_on is switched on, then we use 6 digits and these will the physical time of the simulation rounded to the nearest integer (e.g. base_name_001234.hdf5) [#f3]_. The time of the first snapshot is controlled by the two following options: * Time of the first snapshot (non-cosmological runs): time_first, * Scale-factor of the first snapshot (cosmological runs): scale_factor_first. One of those two parameters has to be provided depending on the type of run. In the case of non-cosmological runs, the time of the first snapshot is expressed in the internal units of time. Users also have to provide the difference in time (or scale-factor) between consecutive outputs: * Time difference between consecutive outputs: delta_time. In non-cosmological runs this is also expressed in internal units. For cosmological runs, this value is *multiplied* to obtain the scale-factor of the next snapshot. This implies that the outputs are  Matthieu Schaller committed Jul 10, 2019 562 equally spaced in :math:\log(a) (See :ref:Output_list_label to have  Josh Borrow committed Jan 22, 2019 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 snapshots not regularly spaced in time). When running the code with structure finding activated, it is often useful to have a structure catalog written at the same simulation time as the snapshots. To activate this, the following parameter can be switched on: * Run VELOCIraptor every time a snapshot is dumped: invoke_stf (default: 0). This produces catalogs using the options specified for the stand-alone VELOCIraptor outputs (see the section :ref:Parameters_structure_finding) but with a base name and output number that matches the snapshot name (e.g. stf_base_name_1234.hdf5) irrespective of the name specified in the section dedicated to VELOCIraptor. Note that the invocation of VELOCIraptor at every dump is done additionally to the stand-alone dumps that can be specified in the corresponding section of the YAML parameter file. Users can optionally specify the level of compression used by the HDF5 library using the parameter: * GZIP compression level of the HDF5 arrays: compression (default: 0). The default level of 0 implies no compression and values have to be in the range :math:[0-9]. This integer is passed to the i/o library and used for the  Peter W. Draper committed Jan 25, 2019 588 loss-less GZIP compression algorithm. Higher values imply higher compression but  Josh Borrow committed Jan 22, 2019 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 also more time spent deflating and inflating the data. Note that up until HDF5 1.10.x this option is not available when using the MPI-parallel version of the i/o routines. Finally, it is possible to specify a different system of units for the snapshots than the one that was used internally by SWIFT. The format is identical to the one described above (See the :ref:Parameters_units section) and read: * a unit of length: UnitLength_in_cgs (default: InternalUnitSystem:UnitLength_in_cgs), * a unit of mass: UnitMass_in_cgs (default: InternalUnitSystem:UnitMass_in_cgs), * a unit of velocity UnitVelocity_in_cgs (default: InternalUnitSystem:UnitVelocity_in_cgs), * a unit of electric current UnitCurrent_in_cgs (default: InternalUnitSystem:UnitCurrent_in_cgs), * a unit of temperature UnitTemp_in_cgs (default: InternalUnitSystem:UnitTemp_in_cgs). When un-specified, these all take the same value as assumed by the internal system of units. These are rarely used but can offer a practical alternative to  Peter W. Draper committed Feb 04, 2019 605 converting data in the post-processing of the simulations.  Josh Borrow committed Jan 22, 2019 606 607 608 609 610 611 612 613 614 615 616  For a standard cosmological run with structure finding activated, the full section would be: .. code:: YAML Snapshots: basename: output scale_factor_first: 0.02 # z = 49 delta_time: 1.02 invoke_stf: 1  Peter W. Draper committed Feb 04, 2019 617   Josh Borrow committed Jan 22, 2019 618 619 620 621 622 623 624 625 626 627 628 629 Showing all the parameters for a basic hydro test-case, one would have: .. code:: YAML Snapshots: basename: sedov time_first: 0.01 delta_time: 0.005 invoke_stf: 0 int_time_label_on: 0 compression: 3 UnitLength_in_cgs: 1. # Use cm in outputs  Peter W. Draper committed Jan 25, 2019 630  UnitMass_in_cgs: 1. # Use grams in outputs  Josh Borrow committed Jan 22, 2019 631 632 633 634 635 636 637 638 639 640  UnitVelocity_in_cgs: 1. # Use cm/s in outputs UnitCurrent_in_cgs: 1. # Use Ampere in outputs UnitTemp_in_cgs: 1. # Use Kelvin in outputs Some additional specific options for the snapshot outputs are described in the following pages: * :ref:Output_list_label (to have snapshots not evenly spaced in time), * :ref:Output_selection_label (to select what particle fields to write).  Matthieu Schaller committed Jun 17, 2019 641 642 643 644 645 646 647 648 .. _Parameters_fof: Friends-Of-Friends (FOF) ------------------------ The parameters are described separately on the page :ref:Fof_Parameter_Description_label within the more general :ref:Friends_Of_Friends_label description.  Josh Borrow committed Jan 22, 2019 649 650  .. _Parameters_statistics:  Peter W. Draper committed Feb 04, 2019 651   Josh Borrow committed Jan 22, 2019 652 653 654 655 656 657 658 659 660 Statistics ---------- Some additional specific options for the statistics outputs are described in the following page: * :ref:Output_list_label (to have statistics outputs not evenly spaced in time). .. _Parameters_restarts:  Peter W. Draper committed Feb 04, 2019 661   Josh Borrow committed Jan 22, 2019 662 663 664 665 666 667 Restarts -------- SWIFT can write check-pointing files and restart from them. The behaviour of this mechanism is driven by the options in the Restarts section of the YAML parameter file. All the parameters are optional but default to values that  Peter W. Draper committed Feb 04, 2019 668 ensure a reasonable behaviour.  Josh Borrow committed Jan 22, 2019 669 670 671 672 673 674  * Whether or not to enable the dump of restart files: enable (default: 1). This parameter acts a master-switch for the check-pointing capabilities. All the other options require the enable parameter to be set to 1.  Peter W. Draper committed Feb 04, 2019 675   Josh Borrow committed Jan 22, 2019 676 677 678 679 680 * Whether or not to save a copy of the previous set of check-pointing files: save (default: 1), * Whether or not to dump a set of restart file on regular exit: onexit (default: 0), * The wall-clock time in hours between two sets of restart files:  Matthieu Schaller committed Jul 06, 2019 681  delta_hours (default: 5.0).  Peter W. Draper committed Feb 04, 2019 682   Josh Borrow committed Jan 22, 2019 683 Note that there is no buffer time added to the delta_hours value. If the  Matthieu Schaller committed Jul 06, 2019 684 system's batch queue run time limit is set to 5 hours, the user must specify a  Josh Borrow committed Jan 22, 2019 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 smaller value to allow for enough time to safely dump the check-point files. * The sub-directory in which to store the restart files: subdir (default: restart), * The basename of the restart files: basename (default: swift) If the directory does not exist, SWIFT will create it. When resuming a run, SWIFT, will look for files with the name provided in the sub-directory specified here. The files themselves are named basename_000001.rst where the basename is replaced by the user-specified name and the 6-digits number corresponds to the MPI-rank. SWIFT writes one file per MPI rank. If the save option has been activated, the previous set of restart files will be named basename_000000.rst.prev. SWIFT can also be stopped by creating an empty file called stop in the  700 701 702 directory where the restart files are written (i.e. the directory speicified by the parameter subdir). This will make SWIFT dump a fresh set of restart file (irrespective of the specified delta_time between dumps) and exit  Josh Borrow committed Jan 22, 2019 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 cleanly. One parameter governs this behaviour: * Number of steps between two checks for the presence of a stop file: stop_steps (default: 100). The default value is chosen such that SWIFT does not need to poll the file-system to often, which can take a significant amount of time on distributed systems. For runs where the small time-steps take a much larger amount of time, a smaller value is recommended to allow for a finer control over when the code can be stopped. Finally, SWIFT can automatically stop after a specified amount of wall-clock time. The code can also run a command when exiting in this fashion, which can be used, for instance, to interact with the batch queue system: * Maximal wall-clock run time in hours: max_run_time (default: 24.0), * Whether or not to run a command on exit: resubmit_on_exit (default: 0), * The command to run on exit: resubmit_command (default: ./resub.sh). Note that no check is performed on the validity of the command to run. SWIFT simply calls system() with the user-specified command. To run SWIFT, dumping check-pointing files every 6 hours and running for 24 hours after which a shell command will be run, one would use: .. code:: YAML  Peter W. Draper committed Feb 04, 2019 730   Josh Borrow committed Jan 22, 2019 731  Restarts:  Peter W. Draper committed Feb 04, 2019 732  enable: 1  Josh Borrow committed Jan 22, 2019 733  save: 1 # Keep copies  Peter W. Draper committed Feb 04, 2019 734  onexit: 0  Josh Borrow committed Jan 22, 2019 735  subdir: restart # Sub-directory of the directory where SWIFT is run  Peter W. Draper committed Feb 04, 2019 736  basename: swift  Matthieu Schaller committed Jul 06, 2019 737  delta_hours: 5.0  Peter W. Draper committed Feb 04, 2019 738 739 740 741  stop_steps: 100 max_run_time: 24.0 # In hours resubmit_on_exit: 1 resubmit_command: ./resub.sh  Josh Borrow committed Jan 22, 2019 742 743 744 745 746 747  .. _Parameters_scheduler: Scheduler ---------  Peter W. Draper committed Feb 04, 2019 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 The Scheduler section contains various parameters that control how the cell tree is configured and defines some values for the related tasks. In general these should be considered as tuning parameters, both for speed and memory use. .. code:: YAML nr_queues: 0 Defines the number of task queues used. These are normally set to one per thread and should be at least that number. A number of parameters decide how the cell tree will be split into sub-cells, according to the number of particles and their expected interaction count, and the type of interaction. These are: .. code:: YAML cell_max_size: 8000000 cell_sub_size_pair_hydro: 256000000 cell_sub_size_self_hydro: 32000 cell_sub_size_pair_grav: 256000000 cell_sub_size_self_grav: 32000 cell_sub_size_pair_stars: 256000000 cell_sub_size_self_stars: 32000 cell_split_size: 400 when possible cells that exceed these constraints will be split into a further level of sub-cells. So for instance a sub-cell should not contain more than 400 particles (this number defines the scale of most N*N interactions). To control the number of self-gravity tasks we have the parameter: .. code:: YAML cell_subdepth_diff_grav: 4 which stops these from being done at the scale of the leaf cells, of which there can be a large number. In this case cells with gravity tasks must be at least 4 levels above the leaf cells (when possible).  Jacob Kegerreis committed Apr 24, 2019 789 790 791 792 793 794 795 796 797 798 799 To control the depth at which the ghost tasks are placed, there are two parameters (one for the gas, one for the stars). These specify the maximum number of particles allowed in such a task before splitting into finer ones. These parameters are: .. code:: YAML engine_max_parts_per_ghost: 1000 engine_max_sparts_per_ghost: 1000  Peter W. Draper committed Feb 04, 2019 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 Extra space is required when particles are created in the system (to the time of the next rebuild). These are controlled by: .. code:: YAML cell_extra_parts: 0 cell_extra_gparts: 0 cell_extra_sparts: 400 The number of top-level cells is controlled by the parameter: .. code:: YAML max_top_level_cells: 12 this is the number per dimension, we will have 12x12x12 cells. There must be at least 3 top-level cells per dimension. The number of top-level cells should be set so that the number of particles per cell is not too large, this is particularly important when using MPI as this defines the maximum size of cell exchange and also the size of non-local cells (these are used for cell interactions with local cells), which can have a large influence on memory use. Best advice for this is to at least scale for additional nodes. The memory used for holding the task and task-link lists needs to be pre-allocated, but cannot be pre-calculated, so we have the two parameters: .. code:: YAML tasks_per_cell: 0.0 links_per_tasks: 10 which are guesses at the mean numbers of tasks per cell and number of links per task. The tasks_per_cell value will be conservatively guessed when set to 0.0, but you will be able to save memory by setting a value. The way to get a better estimate is to run SWIFT with verbose reporting on (--verbose=1) and check for the lines that report the per cell or with MPI maximum per cell values. This number can vary as the balance between MPI ranks does, so it is probably best to leave some head room. If these are exceeded you should get an obvious error message. Finally the parameter: .. code:: YAML mpi_message_limit: 4096 Defines the size (in bytes) below which MPI communication will be sent using non-buffered calls. These should have lower latency, but how that works or is honoured is an implementation question.  Josh Borrow committed Jan 22, 2019 855 856 .. _Parameters_domain_decomposition:  Peter W. Draper committed Jan 22, 2019 857 858 859 860 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891 892 893 894 895 896 897 898 899 900 901 902 903 904 905 906 907 908 909 910 911 912 913 914 915 916 917 918 919 920 921 922 923 924 925 926 927 928 929 930 931 932 933 934 935 936 937 938 939 940 941 942 943 944 945 946 947 948 949 950 951 952 953 954 955 956 957 958 959 960 961 962 963 964 965 966 967 968 969 970 Domain Decomposition: --------------------- This section determines how the top-level cells are distributed between the ranks of an MPI run. An ideal decomposition should result in each rank having a similar amount of work to do, so that all the ranks complete at the same time. Achieving a good balance requires that SWIFT is compiled with either the ParMETIS or METIS libraries. ParMETIS is an MPI version of METIS, so is preferred for performance reasons. When we use ParMETIS/METIS the top-level cells of the volume are considered as a graph, with a cell at each vertex and edges that connect the vertices to all the neighbouring cells (so we have 26 edges connected to each vertex). Decomposing such a graph into domains is known as partitioning, so in SWIFT we refer to domain decomposition as partitioning. This graph of cells can have weights associated with the vertices and the edges. These weights are then used to guide the partitioning, seeking to balance the total weight of the vertices and minimize the weights of the edges that are cut by the domain boundaries (known as the edgecut). We can consider the edge weights as a proxy for the exchange of data between cells, so minimizing this reduces communication. The Initial Partition: ^^^^^^^^^^^^^^^^^^^^^^ When SWIFT first starts it reads the initial conditions and then does an initial distribution of the top-level cells. At this time the only information available is the cell structure and, by geometry, the particles each cell should contain. The type of partitioning attempted is controlled by the:: DomainDecomposition: initial_type: parameter. Which can have the values *memory*, *region*, *grid* or *vectorized*: * *memory* This is the default if METIS or ParMETIS is available. It performs a partition based on the memory use of all the particles in each cell, attempting to equalize the memory used by all the ranks. How successful this attempt is depends on the granularity of cells and particles and the number of ranks, clearly if most of the particles are in one cell, or a small region of the volume, balance is impossible or difficult. Having more top-level cells makes it easier to calculate a good distribution (but this comes at the cost of greater overheads). * *region* The one other METIS/ParMETIS option is "region". This attempts to assign equal numbers of cells to each rank, with the surface area of the regions minimised (so we get blobs, rather than rectangular volumes of cells). If ParMETIS and METIS are not available two other options are possible, but will give a poorer partition: * *grid* Split the cells into a number of axis aligned regions. The number of splits per axis is controlled by the:: initial_grid parameter. It takes an array of three values. The product of these values must equal the number of MPI ranks. If not set a suitable default will be used. * *vectorized* Allocate the cells on the basis of proximity to a set of seed positions. The seed positions are picked every nranks along a vectorized cell list (1D representation). This is guaranteed to give an initial partition for all cases when the number of cells is greater equal to the number of MPI ranks, so can be used if the others fail. Don't use this. If ParMETIS and METIS are not available then only an initial partition will be performed. So the balance will be compromised by the quality of the initial partition. Repartitioning: ^^^^^^^^^^^^^^^ When ParMETIS or METIS is available we can consider adjusting the balance during the run, so we can improve from the initial partition and also track changes in the run that require a different balance. The initial partition is usually not optimal as although it may have balanced the distribution of particles it has not taken account of the fact that different particles types require differing amounts of processing and we have not considered that we also need to do work requiring communication between cells. This latter point is important as we are running an MPI job, as inter-cell communication may be very expensive. There are a number of possible repartition strategies which are defined using the:: DomainDecomposition: repartition_type: parameter. The possible values for this are *none*, *fullcosts*, *edgecosts*, *memory*, *timecosts*. * *none* Rather obviously, don't repartition. You are happy to run with the initial partition. * *fullcosts* Use computation weights derived from the running tasks for the vertex and edge weights. This is the default. * *edgecosts*  Peter W. Draper committed Jan 25, 2019 971  Only use computation weights derived from the running tasks for the edge  Peter W. Draper committed Jan 22, 2019 972 973 974 975 976  weights. * *memory* Repeat the initial partition with the current particle positions  Peter W. Draper committed Jan 25, 2019 977  re-balancing the memory use.  Peter W. Draper committed Jan 22, 2019 978 979 980 981 982 983 984 985 986 987 988 989 990 991 992 993 994 995 996 997 998 999 1000  * *timecosts* Only use computation weights derived from the running tasks for the vertex weights and the expected time the particles will interact in the cells as the edge weights. Using time as the edge weight has the effect of keeping very active cells on single MPI ranks, so can reduce MPI communication. The computation weights are actually the measured times, in CPU ticks, that tasks associated with a cell take. So these automatically reflect the relative cost of the different task types (SPH, self-gravity etc.), and other factors like how well they run on the current hardware and are optimized by the compiler used, but this means that we have a constraint on how often we can consider repartitioning, namely when all (or nearly all) the tasks of the system have been invoked in a step. To control this we have the:: minfrac: 0.9 parameter. Which defines the minimum fraction of all the particles in the simulation that must have been actively updated in the last step, before repartitioning is considered. That then leaves the question of when a run is considered to be out of balance `