[mesa-users] Code snippets to restart a from_file search in astero module
Warrick Ball
wball at astro.physik.uni-goettingen.de
Wed Apr 20 04:25:12 EDT 2016
Hi all,
Some time ago, I coded up a feature that I've found useful, so I thought I
should share it. The astero module includes various optimization routines
as well as an option to try the parameter values listed in a file. This
is selected by
search_type = 'from_file'
in inlist_astero_search_controls. The other methods have options to
restart (i.e. resume from a previous run) but from_file doesn't. So,
because MESA is so easy to understand and modify, I added one! I have my
own use case (using external programs to choose parameters) but the
obvious use is if the run terminates for some reason. e.g. if you,
like me, have hard upper limits on how long single cluster jobs can run.
Anyway, my implementation (which is very basic and not thoroughly tested)
works like this. Referring to r8118, I added the following snippets,
which are included in the attached files.
First, a default option, in star/astero/defaults/astero_search.defaults
at line 387:
restart_from_file_from_file = .false.
Second, to read this option, in star/astero/src/astero_data.f at line 194:
logical :: restart_from_file_from_file
Finally, to use this flag to skip ahead in the output file, in
star/astero/src/run_star_extras_astero.f, I first added an integer to
count how many lines to add. It's declared at line 264:
integer :: iounit, num_to_read, from_file_skip_number
and initialized to 0 at line 271:
from_file_skip_number = 0
Then, at line 274, I read the existing samples if necessary:
if (restart_from_file_from_file) then
call read_samples_from_file(from_file_output_filename, ierr)
if (ierr /= 0) return
from_file_skip_number = sample_number
sample_number = 0
write(*,2) 'from_file_skip_number', from_file_skip_number
end if
and at line 346, skip them:
if (sample_number < from_file_skip_number) then
sample_number = sample_number+1
cycle
end if
With these additions, if you ask to restart from file, the from_file
method will preserve your existing results. Note that there is no
checking that the results actually correspond to your input file! It's
basically just reading the existing data and checking how many lines to
skip in the input file. This possibly causes a mismatch if MESA fails to
find a matching chi^2 for a particular run but I simply have a script
that, when the MESA exits completely, reads the output file and writes the
input parameters to the input file so that they match.
Cheers,
Warrick
------------
Warrick Ball
Postdoc, Institut für Astrophysik Göttingen
wball at astro.physik.uni-goettingen.de
+49 (0) 551 39 5069
-------------- next part --------------
A non-text attachment was scrubbed...
Name: run_star_extras_astero.f
Type: text/x-fortran
Size: 49603 bytes
Desc:
URL: <https://lists.mesastar.org/pipermail/mesa-users/attachments/20160420/997dc5dc/attachment.bin>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: astero_data.f
Type: text/x-fortran
Size: 80606 bytes
Desc:
URL: <https://lists.mesastar.org/pipermail/mesa-users/attachments/20160420/997dc5dc/attachment-0001.bin>
-------------- next part --------------
! astero_search_controls.defaults
! the overall chi^2 is a combination of spectroscopic and seismic
! chi2 = chi2_seismo*chi2_seismo_fraction
! + chi2_spectro*(1 - chi2_seismo_fraction)
chi2_seismo_fraction = 0.667d0
! chi2_spectro is an evenly weighted combination of terms
! only need to give values for target and sigma if include in chi^2
include_Teff_in_chi2_spectro = .true.
Teff_target = -1
Teff_sigma = -1
include_logg_in_chi2_spectro = .true.
logg_target = -1
logg_sigma = -1
include_logL_in_chi2_spectro = .false.
logL_target = -1
logL_sigma = -1
include_FeH_in_chi2_spectro = .true. ! [Fe/H]
FeH_target = -1
FeH_sigma = -1
! we calculate model [Fe/H] as = log10((Z/X)/Z_div_X_solar)
! using model surface average values for Z and X.
Z_div_X_solar = 0.02293d0
! NOTE: for plotting with pgstar loggTe window -- logg vs. Teff
! you need to set the following pgstar controls
! loggTe_target_logg
! loggTe_target_logg_sigma
! loggTe_target_Teff
! loggTe_target_Teff_sigma
! same if you are plotting HR window -- lg_L vs. lg_Teff
! you need to set the following pgstar controls
! HR_target_logL
! HR_target_logL_sigma
! HR_target_Teff
! HR_target_Teff_sigma
! optional extra chi2 terms -- e.g., can be used for solar calibration
! for simplicity, we lump these in with the chi2_spectro terms.
! only need to give values for target and sigma if include in chi^2
include_logR_in_chi2_spectro = .false.
logR_target = 0
logR_sigma = 1d-4
include_surface_Z_div_X_in_chi2_spectro = .false.
surface_Z_div_X_target = 2.292d-2 ! GS98 value
!surface_Z_div_X_target = 1.81d-2 ! Asplund 09 value
surface_Z_div_X_sigma = 1d-3
! reminder: surface_Z_div_X = surface_Z/surface_X
! and FeH = log10(surface_Z_div_X/Z_div_X_solar)
! where Z_div_X_solar is a parameter specified in this inlist
include_surface_He_in_chi2_spectro = .false.
surface_He_target = 0.2485d0 ! Bahcall, Serenelli, Basu, 2005
surface_He_sigma = 0.0034
include_age_in_chi2_spectro = .false.
age_target = 4.5695d9 ! (see Bahcall, Serenelli, and Basu, 2006)
age_sigma = 0.0065d9
! note: when include_age_in_chi2_spectro
! in your &astero_search_controls inlist
! set min_age_for_chi2 and max_age_for_chi2, and
! set eval_chi2_at_target_age_only = .false.
! in your &controls inlist
! set max_years_for_timestep
! but do not set max_age or num_adjusted_dt_steps_before_max_age.
! instead use the following:
num_smaller_steps_before_age_target = 50 ! only used if > 0
dt_for_smaller_steps_before_age_target = 0.0065d8 ! 1/10 age_sigma
! this should be << age_sigma
include_Rcz_in_chi2_spectro = .false. ! radius of base of convective zone
Rcz_target = 0.713d0 ! Bahcall, Serenelli, Basu, 2005
Rcz_sigma = 1d-3
include_csound_rms_in_chi2_spectro = .false. ! check sound profile
csound_rms_target = 0
csound_rms_sigma = 2d-6
report_csound_rms = .false.
! if true, then calculate and report even if not included in chi2
! to include user-defined variables in chi^2 spectro
! set the "my_var" variable in your extras_check_model routine.
! use astero_data, only: my_var1, my_var2, my_var3
! my_var1 = ......
include_my_var1_in_chi2_spectro = .false.
my_var1_target = 0
my_var1_sigma = 0
my_var1_name = 'my_var1' ! change this to whatever you want
include_my_var2_in_chi2_spectro = .false.
my_var2_target = 0
my_var2_sigma = 0
my_var2_name = 'my_var2' ! change this to whatever you want
include_my_var3_in_chi2_spectro = .false.
my_var3_target = 0
my_var3_sigma = 0
my_var3_name = 'my_var3' ! change this to whatever you want
! chi2_seismo is a weighted combination of
! frequencies, ratios of frequencies, delta_nu, and nu_max
! you specify the weighting of the terms in chi2_seismo by setting the following
chi2_seismo_delta_nu_fraction = 0d0
! if > 0 then delta_nu and delta_nu_sigma must be set (see below)
chi2_seismo_nu_max_fraction = 0d0
! if > 0 then nu_max and nu_max_sigma must be set (see below)
chi2_seismo_r_010_fraction = 0d0
! if > 0, then include r_010 frequency ratios
chi2_seismo_r_02_fraction = 0d0
! if > 0, then include r_02 frequency ratios
! fraction for frequencies = 1 - (frac_r_010_ratios + frac_r_02_ratios + frac_delta_nu + frac_nu_max)
! naturally the fractions must all be between 0 and 1, else get error message.
trace_chi2_seismo_delta_nu_info = .false. ! if true, output info to terminal
trace_chi2_seismo_nu_max_info = .false. ! if true, output info to terminal
trace_chi2_seismo_ratios_info = .false. ! if true, output info to terminal
trace_chi2_seismo_frequencies_info = .false. ! if true, output info to terminal
trace_chi2_spectro_info = .false. ! if true, output info to terminal
! seismic data (frequencies in microHz)
nu_max = -1
! nu_max is needed when
! chi2_seismo_nu_max_fraction > 0 or correction_factor > 0 (see below)
nu_max_sigma = -1
! nu_max_sigma is needed when chi2_seismo_nu_max_fraction > 0
delta_nu = -1
delta_nu_sigma = -1
! if delta_nu in the inlist is > 0, then the code uses values for both
! delta_nu and delta_nu_sigma from the inlist.
! if delta_nu from inlist <= 0, then code estimates it
! by linear fit to the observed radial frequencies and orders, l0_obs and l0_n_obs.
! along with calculating delta_nu, if delta_nu_sigma from the inlist is <= 0, then
! the code also sets it by using the radial data.
! note that by setting delta_nu_sigma to a positive value and delta_nu to <= 0,
! you can have the code get delta_nu from the given l0_obs and l0_n_obs,
! while still using the delta_nu_sigma from the inlist.
! observed l=0 modes -- in order of increasing frequency
nl0 = 0 ! number of observed l=0 modes
l0_obs(:) = 0 ! frequencies. set l0_obs(1), l0_obs(2) .... l0_obs(nl0)
l0_obs_sigma(:) = 0 ! l0_obs_sigma(i) is uncertainty for l0_obs(i), for i=1,nl0
l0_n_obs(:) = -1 ! l0_n_obs(i) is order for l0_obs(i), for i=1,nl0
! need to give these if code is to calculate delta_nu and delta_nu_sigma
! if you provide delta_nu, then you don't need to set these.
! observed l=1 modes -- in order of increasing frequency
nl1 = 0 ! number of observed l=1 modes
l1_obs(:) = 0 ! frequencies. set l1_obs(1), l1_obs(2) .... l1_obs(nl1)
l1_obs_sigma(:) = 0 ! l1_obs_sigma(i) is uncertainty for l1_obs(i), for i=1,nl1
! observed l=2 modes -- in order of increasing frequency
nl2 = 0 ! number of observed l=2 modes
l2_obs(:) = 0 ! frequencies. set l2_obs(1), l2_obs(2) .... l2_obs(nl2)
l2_obs_sigma(:) = 0 ! l2_obs_sigma(i) is uncertainty for l2_obs(i), for i=1,nl2
! observed l=3 modes -- in order of increasing frequency
nl3 = 0 ! number of observed l=3 modes
l3_obs(:) = 0 ! frequencies. set l3_obs(1), l3_obs(2) .... l3_obs(nl3)
l3_obs_sigma(:) = 0 ! l3_obs_sigma(i) is uncertainty for l3_obs(i), for i=1,nl3
! search controls
eval_chi2_at_target_age_only = .false.
! set this true if you want chi2 only for a specific age and no others,
! in addition,
! set max_age
! set max_years_for_timestep
! set num_adjusted_dt_steps_before_max_age
! use these if you only want to evaluate chi2 for a given range of ages
min_age_for_chi2 = -1 ! (years) only use if > 0
max_age_for_chi2 = -1 ! (years) only use if > 0
search_type = 'use_first_values'
! this option means no search
! just do a single run using "first" values for mass, alpha, Y, ...
!search_type = 'scan_grid'
! eval chi^2 for each parameter combination
! that is within min to max with steps given by "delta"
! for a first rough scan, consider setting chi2_seismo_delta_nu_fraction = 1
! that skips the relatively costly calculations of frequencies
! and simply uses delta_nu along with spectro information.
! then you might follow up with medium resolution scans
! in smaller regions around candidates from the rough scan
! with chi2_seismo_delta_nu_fraction = 0 to include frequencies
! output goes to the following file when search_type = 'scan_grid'
scan_grid_output_filename = 'scan_grid_results.data'
restart_scan_grid_from_file = .false.
! if true, then reads the scan_grid_output file
! and continues from where that stopped.
! there are 3 types of optimization provided: simplex, newuoa, and bobyqa
!search_type = 'simplex'
! search for minimal chi^2 model using Nelder-Mead simplex algorithm
! Nelder, J. A. and Mead, R.
! "A Simplex Method for Function Minimization."
! Comput. J. 7, 308-313, 1965.
! there are versions of this in Numerical Recipes under the name "amoeba",
! in Matlab under the name "fminsearch", and in Mathematica as an option for "NMminimize".
! our version has lots of bells and whistles and is, of course, superior to the others. ;)
!search_type = 'newuoa'
! M.J.D. Powell, "Developments of NEWUOA for unconstrained minimization without derivatives",
! Department of Applied Mathematics and Theoretical Physics, Cambridge, England, report NA05, 2007.
! search for minimal chi^2 model using Powell's NEWUOA algorithm
! unconstrained minimization without derivatives
! by quadratic polynomial approximation.
!search_type = 'bobyqa'
! M.J.D. Powell, "The BOBYQA algorithm for bound constrained optimization without derivatives",
! Department of Applied Mathematics and Theoretical Physics, Cambridge, England, report NA06, 2009.
! search for minimal chi^2 model using Powell's BOBYQA algorithm
! "BOBYQA" = Bounded Optimization BY Quadratic Approximation
! any location within the bounds is available for consideration.
! the two methods from Powell use quadratic interpolation,
! either unconstrained (NEWUOA) or bounded (BOBYQA).
! the Nelder-Mead SIMPLEX method doesn't do interpolation;
! instead it simply compares values and moves toward lower chi^2's
! and away from higher ones. in general, you can expect the
! Powell methods to be faster to converge than simplex if the chi^2
! terrain is not too "bumpy" (bumps confuse the interpolation).
! Since the simplex scheme doesn't do interpolation, bumps
! don't cause it trouble, so it may be more robust.
! if you are just getting started, go with simplex at first.
! try the interpolation methods when you have a very good
! candidate and want to look near it for even better results.
!search_type = 'simplex'
! terminate if reach max allowed iterations or function calls
simplex_itermax = 1000 ! each iteration revises the simplex
simplex_fcn_calls_max = 10000 ! may use several function calls per iteration
! each "function call" is a stellar evolution to give a chi^2
! terminate if simplex gets "small" enough
simplex_x_atol = 1d-10 ! tolerance for absolute differences
simplex_x_rtol = 1d-10 ! tolerance for relative differences
! if you want the details, here's the snippet of code.
! simplex(i,j) is value of i'th parameter for point j
! l is the index of the best point. there are n parameters and n+1 points.
! term_val_x = 0
! do j=1,n+1 ! check each point
! if (j == l) cycle ! l is the best point; so skip it
! do i=1,n ! check each coordinate of point j vs point l
! term1 = abs(simplex(i,j)-simplex(i,l)) / &
! (x_atol + x_rtol*max(abs(simplex(i,j)), abs(simplex(i,l))))
! if (term1 > term_val_x) term_val_x = term1
! end do
! end do
! if (term_val_x <= 1d0) exit ! converged
! terminate if best point has a chi^2 below simplex_chi2_tol
simplex_chi2_tol = 1d-10 ! tolerance for chi^2
! options
simplex_centroid_weight_power = 0d0
! each iteration starts by doing a reflection
! of the worst point through the centroid of the others.
! centroid points are weighted by (1/chi^2)**power
! power = 0 gives the standard unweighted centroid.
! power = 1 shifts the reflection toward the better points.
! in some (many? most?) cases, this shift improves convergence rate.
simplex_enforce_bounds = .false.
! if true, then points outside the bounds will be rejected without evaluation.
! if false, then bounds will only be used in creating the initial simplex
! and for adaptive random search.
simplex_adaptive_random_search = .false.
! this flag controls what is done if the standard options of reflect or contract
! fail to produce a improvement in the simplex.
! if true, test uniform random samples within the bounds until find a better point.
! if false, "shrink" the simplex toward the best point.
simplex_output_filename = 'simplex_results.data'
restart_simplex_from_file = .false.
! if true, then reads the output file (simplex_output_filename)
! and continues from where that stopped
! using the best n+1 results as the initial simplex
! (where n is the number of parameters)
! NOTE: this restores the best simplex, but you may still
! see it rerun recent cases if they were not good enough to be
! included in the simplex. we don't restore the information
! about those failed attempts, so we need to rerun them.
simplex_seed = 1074698122 ! seed for random number generation
!search_type = 'newuoa'
! output goes to the following file when search_type = 'newuoa'
newuoa_output_filename = 'newuoa_results.data'
! search controls for bobyqa
! see mesa/num/public/num_newuoa for details
newuoa_rhoend = 1d-6
! this is the tolerance that determines relative accuracy of final values
! i.e., stops search when results are changing by less than this
!search_type = 'bobyqa'
! output goes to the following file when search_type = 'bobyqa'
bobyqa_output_filename = 'bobyqa_results.data'
! search controls for bobyqa
! see mesa/num/public/num_bobyqa for details
bobyqa_rhoend = 1d-6
! this is the tolerance that determines relative accuracy of final values
! i.e., stops search when results are changing by less than this
!search_type = 'from_file'
! this uses a file to provide a series of parameter combinations.
! the input file for the parameters is the following:
filename_for_parameters = 'undefined'
max_num_from_file = -1 ! if > 0, then stop after doing this many lines from file.
! for each line in the file after the 1st (which holds column names)
! for each parameter with vary_<Param> = .true.,
! set value of first_<Parama> to value from file.
! then run that set of parameters as if search_type = 'use_first_values'
! save the best chi^2 info for that set of parameters,
! then go to the next line in the file.
! stops when finishes the file or reaches "max_num_from_file"
! you need to say which columns in the file hold the various parameters.
! for example, if you file starts like the following:
! chi2 mass alpha init_Y init_FeH init_f_ov
! 654 0.81543178 1.35000000 1.76000000 0.27000000 0.21000000 0.01000000
! then set the column numbers like this
file_column_for_mass = 3
file_column_for_alpha = 4
file_column_for_Y = 5
file_column_for_FeH = 6
file_column_for_f_ov = 7
! note that if you are not varying one of the parameters, f_ov e.g.,
! then you don't need to set the file_col for that parameter.
! output goes to the following when search_type = 'from_file'
from_file_output_filename = 'from_file_results.data'
restart_from_file_from_file = .false.
! status of Y -- parameter or function of Z
Y_depends_on_Z = .false.
! if false, then Y is a parameter like FeH.
! you should set vary_Y, first_Y, min_Y, and max_Y.
! if true, then Y depends on Z as follows: Y = Y0 + dYdZ*Z
! in this case, set vary_Y = .false.
! first_Y, min_Y, and max_Y are unused.
Y0 = 0.248d0
dYdZ = 1.4d0
! set these flags to specify which parameters will vary during searches
vary_FeH = .false. ! FeH = [Fe/H] = log10((Z/X)/Z_div_X_solar)
vary_Y = .false. ! must be false if Y_depends_on_Z is true
vary_mass = .false. ! initial mass
vary_alpha = .false.
vary_f_ov = .false.
! >>>> NOTE: if "vary" is false, the "first" value is used for all runs.
! so you must set the first value even when you have vary = .false.
! use the following as the first values in searches
first_FeH = 0
first_Y = 0
first_mass = 0
first_alpha = 0
first_f_ov = 0
! lower bounds for searches
min_FeH = 0
min_Y = 0
min_mass = 0
min_alpha = 0
min_f_ov = 0
! upper bounds for searches
max_FeH = 0
max_Y = 0
max_mass = 0
max_alpha = 0
max_f_ov = 0
! search_type = 'scan_grid' uses this grid spacing
! note: grid spacing does not apply to other searches
delta_FeH = 0
delta_Y = 0
delta_mass = 0
delta_alpha = 0
delta_f_ov = 0
! overshoot_f0 is changed along with overshoot_f
! f0_ov = f0_ov_div_f_ov * f_ov
f0_ov_div_f_ov = -1 ! this must be set to a positive value
! calculating mode frequencies is a relatively costly process,
! so we don't want to do it for models that are not good candidates.
! i.e., we want to filter out the bad candidates using the following
! less expensive tests whenever possible.
! NOTE: if none of the models in a run pass these tests,
! then you will not get any total chi2 result for that run.
! in some situations that might not matter,
! but if you are eliminating too many candidates in this way,
! the search routines might not be getting enough valid results to work properly.
! So watch what you are doing! If your search or scan is getting lots of
! runs that fail to give chi^2 results, you'll need to adjust the limits.
! don't consider models that aren't old enough
min_age_limit = 1d6
! don't consider models with L_nuc/L less than this limit
Lnuc_div_L_limit = 0.95 ! this rules out pre-zams models
! don't consider models with chi2_spectroscopic above this limit
chi2_spectroscopic_limit = 1000
! don't consider models with chi2_delta_nu above this limit
chi2_delta_nu_limit = 1000
! we calculate radial modes only if pass the previous checks
! calculating nonradial modes is much more expensive than radial ones.
! so we skip the nonradial calculation if the radial results are poor.
! don't consider models with chi2_radial above this limit
chi2_radial_limit = 100
! only calculate full chi^2 if pass all these limit checks
! adjust max timestep depending on how close to target
! NOTE: if you set the timestep limits too large you run the risk of missing good chi^2 cases.
! but if they are very small, you will spend a lot of runtime calculating lots of frequencies
! for lots of models. There is no standard set of best values for this.
! The choice will depend on the stage of evolution and how fast things are changing
! in the general region of the models with good chi2 values.
! There is no alternative to trying things and tuning the controls for your problem.
! these are just default values -- you will undoubtedly need to adjust them for your problem.
max_yrs_dt_when_cold = 1d8 ! when fail Lnuc/L, chi2_spectro, or ch2_delta_nu
max_yrs_dt_when_warm = 1d7 ! when pass previous but fail chi2_radial; < max_yrs_dt_when_cold
max_yrs_dt_when_hot = 1d6 ! when pass chi2_radial; < max_yrs_dt_when_warm
chi2_limit_for_small_timesteps = 50
max_yrs_dt_chi2_small_limit = 3d5 ! < max_yrs_dt_when_hot
chi2_limit_for_smaller_timesteps = 20 ! < chi2_limit_for_small_timesteps
max_yrs_dt_chi2_smaller_limit = 1d5 ! < max_yrs_dt_chi2_small_limit
chi2_limit_for_smallest_timesteps = 10 ! < chi2_limit_for_smaller_timesteps
max_yrs_dt_chi2_smallest_limit = 5d4 ! < max_yrs_dt_chi2_smaller_limit
! we need a way to decide when to stop an evolution run.
! the following limits are used for this.
! NOTE: we don't want to stop too soon, so these limits
! are only tested for models that are okay for the Lnuc_div_L_limit.
! logg_limit = logg_target + logg_sigma*sigmas_coeff_for_logg_limit
! logL_limit = logL_target + logL_sigma*sigmas_coeff_for_logL_limit
! Teff_limit = Teff_target + Teff_sigma*sigmas_coeff_for_Teff_limit
! logR_limit = logR_target + logR_sigma*sigmas_coeff_for_logR_limit
! surface_Z_div_X_limit = surface_Z_div_X_target +
! surface_Z_div_X_sigma*sigmas_coeff_for_surface_Z_div_X_limit
! surface_He_limit = surface_He_target + &
! surface_He_sigma*sigmas_coeff_for_surface_He_limit
! Rcz_limit = Rcz_target + Rcz_sigma*sigmas_coeff_for_Rcz_limit
! csound_rms_limit = csound_rms_target + &
! csound_rms_sigma*sigmas_coeff_for_csound_rms_limit
! delta_nu_limit = delta_nu + &
! delta_nu_sigma*sigmas_coeff_for_delta_nu_limit
! only use limits with sigma_coeff /= 0
! if the sigma_coeff is > 0, then stop when value is > limit
! if the sigma_coeff is < 0, then stop when value is < limit
! so use positive sigma_coeff for values that are increasing (such as logL)
! and negative ones for values that are decreasing (logg, Teff, delta_nu)
sigmas_coeff_for_logg_limit = -5 ! disable by setting to 0
sigmas_coeff_for_logL_limit = 5 ! disable by setting to 0
sigmas_coeff_for_Teff_limit = -5 ! disable by setting to 0
sigmas_coeff_for_logR_limit = 0 ! disable by setting to 0
sigmas_coeff_for_surface_Z_div_X_limit = 0 ! disable by setting to 0
sigmas_coeff_for_surface_He_limit = 0 ! disable by setting to 0
sigmas_coeff_for_Rcz_limit = 0 ! disable by setting to 0
sigmas_coeff_for_csound_rms_limit = 0 ! disable by setting to 0
sigmas_coeff_for_delta_nu_limit = 0 ! -5 ! disable by setting to 0
sigmas_coeff_for_csound_rms_limit = 0 ! disable by setting to 0
sigmas_coeff_for_my_var1_limit = 0 ! disable by setting to 0
sigmas_coeff_for_my_var2_limit = 0 ! disable by setting to 0
sigmas_coeff_for_my_var3_limit = 0 ! disable by setting to 0
! you can stop the run if chi^2 is rising.
! here is a relative limit
chi2_relative_increase_limit = 2.0
limit_num_chi2_too_big = 20
! if limit_num_chi2_too_big consequtive chi2s
! are > chi2_relative_increase_limit times the best chi2 for the run,
! then stop the run.
! and here is an absolute limit
chi2_search_limit1 = 3.0
chi2_search_limit2 = 4.0
! if best chi2 for the run is <= chi2_search_limit1,
! then stop the run if chi2 >= chi2_search_limit2.
! if you are doing a search or scanning a grid, you can use previous results
! as a guide for when to stop a run
min_num_samples_for_avg = 2 ! want at least this many samples to form averages
max_num_samples_for_avg = 10 ! use this many of the best chi^2 samples for averages
! these use results for the best chi^2 model of the previous best samples
avg_age_sigma_limit = 10 ! stop if age > avg age + this limit times sigma of avg age
avg_model_number_sigma_limit = 10 ! ditto for model number
! surface corrections
correction_scheme = 'kjeldsen' ! options are
! 'kjeldsen' Correction of Kjeldsen et al. (2008)
! 'cubic' Cubic correction of Ball & Gizon (2014, eqn 3)
! 'combined' Combined correction of Ball & Gizon (2014, eq 4)
! '' no corrections
correction_factor = 0
! use this fraction of the correction; set to 0 to skip doing corrections.
correction_b = 4.90d0
save_next_best_at_higher_frequency = .true.
save_next_best_at_lower_frequency = .true.
! note: to set nu_max_sun or delta_nu_sun, see star/defaults/controls.defaults
! if you'd like to experiment with your own correction scheme,
! you can use the other_astero_freq_corr "hook" in mesa/star.
! output controls
write_best_model_data_for_each_sample = .true.
num_digits = 4 ! number of digits in sample number (with leading 0's)
sample_results_prefix = 'outputs/sample_'
! note that you can include a directory in the prefix if desired
sample_results_postfix = '.data'
model_num_digits = 4 ! number of digits in model number (with leading 0's)
write_fgong_for_each_model = .false.
fgong_prefix = 'fgong_'
! note that you can include a directory in the prefix if desired
fgong_postfix = '.data'
write_fgong_for_best_model = .false.
best_model_fgong_filename = ''
write_gyre_for_each_model = .false.
gyre_prefix = 'gyre_'
! note that you can include a directory in the prefix if desired
gyre_postfix = '.data'
max_num_gyre_points = -1 ! only used if > 1
write_gyre_for_best_model = .false.
best_model_gyre_filename = ''
write_profile_for_best_model = .false.
best_model_profile_filename = ''
save_model_for_best_model = .false.
best_model_save_model_filename = ''
save_info_for_last_model = .false. ! if true, treat final model as "best"
last_model_save_info_filename = '' ! and save info about final model to this file.
shell_script_for_each_sample = '' ! executed after at end of sample run
shell_script_num_string_char = '#' ! replace by num string for sample
! Do whatever you like in the script. e.g.,
! 'cp best.mod outputs/sample#_best.mod; cp LOGS/history.data outputs/sample#_history.data'
! miscellaneous
! save info about next best matches
save_next_best_at_higher_frequency = .false.
save_next_best_at_lower_frequency = .false.
! trace limits
trace_limits = .false.
! if true, write info to terminal about status relative to various limits
! such as Teff_limit = Teff_target + Teff_sigma*sigmas_coeff_for_Teff_limit
! run will stop when Teff < Teff_limit.
! trace will write out values of Teff and Teff_limit
! same for other limits such as for logg, logL, delta_nu, etc.
! save all control settings to file
save_controls = .false. ! dumps &astero_search_controls controls to file
save_controls_filename = '' ! if empty, uses a default name
! composition control
Y_frac_he3 = 1d-4 ! = xhe3/(xhe3 + xhe4); Y = xhe3 + xhe4
! save an eigenfunction
save_mode_model_number = 0
save_mode_filename = ''
el_to_save = 0
order_to_save = 0
! options for input model to pulsation codes
add_atmosphere = .false.
! if true, then star adds atmosphere before passing model to adipls
! the atmosphere model is determined by the mesa/star control which_atm_option
! it should either be one of the T(tau) integration options or Paczynski_grey,
! scheme inspired by B. Paczynski, 1969, Acta Astr., vol. 19, 1.
! which takes into account radiation dilution when tau < 2/3,
keep_surface_point = .false.
! if true, keep k=1 point of model.
add_center_point = .true.
! if true, add point at r=0
! oscillation analysis
oscillation_code = 'adipls' ! or 'gyre' <<< lowercase
trace_time_in_oscillation_code = .false.
! gyre controls
gyre_input_file = 'gyre.in'
gyre_non_ad = .false.
! comments from Rich on setting gyre controls
! I suggest setting freq_min to 0.9*MINVAL(l0_obs),
! and freq_max to 1.1*MAXVAL(l0_obs)
! (similarly for the other l values).
! freq_units should be 'UHZ',
! and set grid_type to 'LINEAR'.
! For n_freq, I suggest either setting it to 10*(freq_max - freq_min)/dfreq,
! where dfreq is the estimated frequency spacing; or, set it to 10*nl0.
! The factor 10 is arbitrary, but seems to be a good safety factor.
! adipls controls
do_redistribute_mesh = .true.
! note: number of zones for redistribute is set in redistrb.c.pruned.in
! if you set this false, then the mesh from star is used directly.
! if you set this true, then astero calls adipls redistb before doing
! the frequency analysis.
! iscan for adipls = the following factor times the given number of observed modes
iscan_factor_l0 = 15
iscan_factor_l1 = 15
iscan_factor_l2 = 15
iscan_factor_l3 = 15
! frequency range is set from observed frequencies times these factors
nu_lower_factor = 0.8
nu_upper_factor = 1.2
! adipls looks for frequencies in a given range and with a given "density" of coverage
! for example,
! for l=0, the adipls freq search range is nu_lower_factor*l0_obs(1) to nu_upper_factor*l0_obs(nl0)
! and it uses iscan = iscan_factor_l0*nl0 to determine how fine the scan is over the range.
! similar for l=1 and l=2
! misc adipls parameters for experts
adipls_irotkr = 0
adipls_nprtkr = 0
adipls_igm1kr = 0
adipls_npgmkr = 0
! include other inlists
read_extra_astero_search_inlist1 = .false.
extra_astero_search_inlist1_name = 'undefined'
! if read_extra_astero_search_inlist1 is true, then read this namelist file
read_extra_astero_search_inlist2 = .false.
extra_astero_search_inlist2_name = 'undefined'
! if read_extra_astero_search_inlist2 is true, then read this namelist file
read_extra_astero_search_inlist3 = .false.
extra_astero_search_inlist3_name = 'undefined'
! if read_extra_astero_search_inlist3 is true, then read this namelist file
read_extra_astero_search_inlist4 = .false.
extra_astero_search_inlist4_name = 'undefined'
! if read_extra_astero_search_inlist4 is true, then read this namelist file
read_extra_astero_search_inlist5 = .false.
extra_astero_search_inlist5_name = 'undefined'
! if read_extra_astero_search_inlist5 is true, then read this namelist file
More information about the Mesa-users
mailing list