Configuration Settings#
The configuration has eight sections:
Of these sections, Core gives the basic settings and core assumptions within Lightning (e.g., cosmology). The next five, Stellar Emission, Dust Attenuation, Dust Emission, X-ray Emission, and AGN Emission, make up the model selection sections. Lastly, Fitting Algorithm gives the choice of fitting algorithm and its corresponding hyper-parameters, and Post-processing gives the choice of how to post-process the resulting data.
Note
If you are unsure on what model to select for fitting your SED(s), see our guide on Selecting a Model. The same goes for the fitting algorithm. See our guide on Selecting an Algorithm to help you decide on a fitting algorithm or if you are having issues with your fits.
Core#
OUTPUT_FILENAME
string scalarThe name (without the file extension suffix) to give to the FITS file containing the output post-processed data. (See the Results and Post-Processing for details on the contents of the file.)
Note
A UTC timestamp can be automatically included in the filename so that you can have a unique
filename for multiple repeat runs to prevent accidentally overwriting old runs. This is
done by including a single %
character in the filename where you want the timestamp
to appear. For example, OUTPUT_FILENAME = 'test_%
will result in the output file with
a name like test_2023-01-26T20-05-31Z.fits.gz
PRINT_PROGRESS
flagA flag that indicates if the progress of Lightning should be printed to the terminal. This progress includes the current elapsed time, completed processes, and expected time remaining.
MAX_CPUS
int scalarThe maximum number of CPUs on the machine to utilize. If this value exceeds the actual number of CPUs on the machine, then all CPUs will be used. This setting allows for Lightning to fit SEDs in parallel, where one SED is fit per CPU (i.e., a number of SEDs equal to
MAX_CPUS
will be fit simultaneously in batches until all SEDs have been fit).ENERGY_BALANCE
flagA flag indicating if energy balance should be assumed in the SED fits. Energy balance is the assumption that the total integrated IR luminosity of the dust emission is equal to the total absorbed stellar (and, if set, AGN) emission.
Note
This is a key assumption in most SED fitting codes as it attempts to preserve conservation of energy. See our guide on Selecting a Model if you are unsure if you want energy balance in your model.
MODEL_UNC
int, float, or double scalarThe fractional model uncertainty to use in all filters when computing \(\chi^2\) during the SED fitting. This form of uncertainty accounts for systematic effects in the models and is computed as
\[\sigma_{{\rm mod},\ i}^2 = \big({\tt MODEL\_UNC} * L_{\nu,\ i}^{\rm mod} \big)^2,\]where \(\sigma_{{\rm mod},\ i}\) is the model uncertainty of filter \(i\), and \(L_{\nu,\ i}^{\rm mod}\) is the model luminosity of filter \(i\). The total uncertainty used in the \(\chi^2\) calculation is then given as
\[\sigma_{{\rm total},\ i}^2 = \sigma_{{\rm obs},\ i}^2 + \sigma_{{\rm mod},\ i}^2,\]where \(\sigma_{{\rm total},\ i}\) is the total uncertainty of filter \(i\), and \(\sigma_{{\rm obs},\ i}\) is the observed uncertainty of filter \(i\) as given in the input.
Note
It is common in the literature to assume a fractional model uncertainty of 5-10%, regardless of SED fitting code. Therefore, we recommend using a fractional model uncertainty of 5% when fitting any SED for the first time.
Cosmology#
The next five settings are the cosmology parameters to use in the SED fitting. These parameters determine the assumed cosmology, which set the age of the universe and the distance to objects if their distance was specified by redshift.
H0
int, float, or double scalarThe Hubble constant, \(H_0\) \([{\rm km\ s^{-1}\ Mpc^{-1}}]\).
OMEGA_M
int, float, or double scalarThe matter density normalized to the closure density, \(\Omega_m\).
LAMBDA0
int, float, or double scalarThe cosmological constant normalized to the closure density, \(\Lambda_0\).
Q0
int, float, or double scalarThe deceleration parameter, \(q_0\).
K
int, float, or double scalarThe curvature constant normalized to the closure density, \(k\).
Stellar Emission#
SSP
string scalarThe simple stellar population (SSP) models to use for the stellar population. The only SSP models currently available in Lightning are the PEGASE models. These models are selected by setting
SSP
to'PEGASE'
. To fit the SEDs without any stellar emission, setSSP
to'NONE'
.
Note
If no stellar emission model is chosen, all stellar emission model settings below can be skipped.
IMF
string scalarThe initial mass function (IMF) to use in the SSP models. The only IMF currently available in Lightning is that from Kroupa (2001). This IMF is selected by setting
IMF
to'KROUPA01'
.ZMETAL
float or double scalarThe metallicity to use in the SSP models in terms of Z, normalized to the solar metallicity. The current available metallicities in Lightning are 0.001, 0.004, 0.008, 0.01, 0.02, 0.05, and 0.1 in terms of \(Z\).
Note
Lightning currently assumes the chosen metallicity is constant for at all ages, and does not allow for metallicity evolution. To minimize any systematic effects caused by ignoring metallicity evolution, we recommend selecting a metallicity closest to current average metallicity of your input SEDs.
EMISSION_LINES
flagA flag indicating if nebular emission lines should be included in the SSP models.
NEBULAR_EXTINCTION
flagA flag indicating if nebular extinction should be applied to the SSP models.
Note
Detail on the analytical modeling for nebular extinction and emission in the
PEGASE
models can be found in Section 2.4 of Fioc & Rocca-Volmerange (1997)SFH
string scalarThe type of star formation history (SFH) to assume when fitting the SEDs. The only SFH type currently available in Lightning is the binned or “non-parametric” SFH. This SFH assumes a piece-wise constant SFH, where the SFR is a constant value within a set of age bins. This SFH type is selected by setting
SFH
to'NON-PARAMETRIC'
.STEPS_BOUNDS
int, float, or double array(Nsteps+1)The age bin (or step) boundaries to use in the “non-parametric” SFH in units of \({\rm yr}\). Values must be in ascending order.
Note
If an age bin contains ages older than the universe at an input SED’s redshift, the age bin upper bound will be automatically adjusted to the age of the universe at that redshift. If an entire age bin is older than universe at that redshift, then the entire age bin will be omitted and the next younger bin will be adjusted accordingly.
DTIME_SF
int, float, or double scalarThe time step used for interpolating the SSP models into the age bins in units of \({\rm yr}\).
Warning
We do not recommend changing this value from its default. The only case in which it should be changed is if you specified age bins with differences less than the default value. However, in that case, your age bins are likely too small.
PSI
structureThe free parameter \(\psi_i\), the SFR for of the SFH age bin \(i\) in \(M_\odot\ {\rm yr}^{-1}\). This structure contains the priors to assume for each \(\psi_i\). Values of \(\psi_i\) are limited to being non-negative numbers.
Note
Check out the Prior Examples for details on what a prior structure contains and various examples.
Dust Attenuation#
ATTEN_CURVE
string scalarThe assumed attenuation curve to apply to the stellar and/or AGN models. There are three attenuation curve options currently available in Lightning. They are the Calzetti et al. (2000) attenuation curve, modified Calzetti et al. (2000) attenuation curve, and Doore et al. (2021) attenuation curve. The modified Calzetti curve can include a variable slope as described in Noll et al. (2009), an optional 2175 Angstrom bump feature specified in Kriek & Conroy (2013), and birth cloud attenuation as described in Eufrasio et al. (2017). The Doore et al (2021) attenuation curve is based on the Tuffs et al. (2004) attenuation curves as updated by Popescu et al. (2011). These attenuation curves are selected by setting
ATTEN_CURVE
to'CALZETTI00'
,'CALZETTI_MOD'
, or'DOORE21'
, respectively.Note
Attenuation of AGN can only use the
'CALZETTI00'
or'CALZETTI_MOD'
attenuation curves. Compatibility of the AGN models with the'DOORE21'
curve is currently not supported.
Calzetti+00#
TAUV
structureThe free parameter \(\tau_V\), the V-band optical depth used for normalization in the Calzetti et al. (2000) attenuation curve. This structure contains the prior to assume for \(\tau_V\). Values of \(\tau_V\) are limited to being non-negative numbers.
Modified Calzetti+00#
TAUV_DIFF
structureThe free parameter \(\tau_V^{\rm diff}\), the V-band optical depth of diffuse dust used for normalization in the Calzetti et al. (2000) attenuation curve. This structure contains the prior to assume for \(\tau_V^{\rm diff}\). Values of \(\tau_V^{\rm diff}\) are limited to being non-negative numbers.
DELTA
structureThe free parameter \(\delta\), the power law value used to create a variable attenuation curve slope as described in Noll et al. (2009). This structure contains the prior to assume for \(\delta\). Values of \(\delta\) can be any real numbers. A value of
0
indicates the same slope as the original Calzetti et al. (2000) attenuation curve.TAUV_BC
structureThe free parameter \(\tau_V^{\rm BC}\), the V-band optical depth of the birth cloud component as described in Eufrasio et al. (2017). This structure contains the prior to assume for \(\tau_V^{\rm BC}\). Values of \(\tau_V^{\rm BC}\) are limited to being non-negative numbers. A value of
0
indicates no birth cloud attenuation.UV_BUMP
flagA flag indicating if a 2175 Angstrom UV bump feature as specified in Kriek & Conroy (2013) should be added to the attenuation curve.
Doore+21#
TAUB_F
structureThe free parameter \(\tau_B^{f}\), the face-on optical depth in the B-band. This structure contains the prior to assume for \(\tau_B^{f}\). Values of \(\tau_B^{f}\) are limited to being between
0
and8
.F_CLUMP
structureThe free parameter \(F\), the birth cloud clumpiness factor. This structure contains the prior to assume for \(F\). Values of \(F\) are limited to being between
0
and0.61
.COSI
structureThe free parameter \(\cos i\), the inclination of the galactic disk in terms of \(\cos i\). This structure contains the prior to assume for \(\cos i\). Values of \(\cos i\) are limited to being between
0
and1
.B_TO_D
structureThe free parameter \(B/D\), the bulge-to-disk ratio. This structure contains the prior to assume for \(B/D\). Values of \(B/D\) are limited to being non-negative numbers.
ROLD0_AGES
int, float, or double array(Nsteps)The binary parameter \(r^{0,\ {\rm old}}\), that designates each SFH age bin as part of the young or old population. A value of
0
for the corresponding age bin considers it to be part of the young population, and a value of1
considers it to be part of the old populations (see section 4.3 of Doore et al. 2021 for more details). The number of elements must be one less than the number of elements inSTEPS_BOUNDS
.Note
We recommend setting age bins that contain ages \(< 500\ {\rm Myr}\) to be part the young population as they can contain significant UV emission. If you choose to set age bins with ages \(< 500\ {\rm Myr}\) to the old population, the SFR may be underestimated due to under-attenuation of the UV-emitting population.
Dust Emission#
DUST_MODEL
string scalarThe dust emission model to use. The only dust emission model currently available in Lightning is the Draine & Li (2007) (DL07) model. This model is selected by setting
DUST_MODEL
to'DL07'
. To fit the SEDs without any dust emission, setDUST_MODEL
to'NONE'
.
Note
If no dust emission model is chosen, all dust emission model settings below can be skipped.
DL07#
UMIN
structureThe free parameter \(U_{\rm min}\), the minimum radiation field intensity of the diffuse ISM radiation field from the heated dust. This structure contains the prior to assume for \(U_{\rm min}\). Values of \(U_{\rm min}\) are limited to being between
0.1
and25
.UMAX
structureThe free parameter \(U_{\rm max}\), the maximum radiation field intensity of the power-law distribution of heating starlight intensities. This structure contains the prior to assume for \(U_{\rm max}\). Values of \(U_{\rm max}`\) are limited to being between
1e3
and3e5
.Note
The parameter range of \(U_{\rm max}\) is slightly less than the quoted full range of the DL07 models (\(10^6\)). This slightly limited range originates from the format of the publicly available data. The publicly available \(\delta\)-functions of \(U\), from which \(U_{\rm max}`\) can be calculated for any given \(\alpha\), have a maximum value of \(3 \times 10^5\). However, rather than extrapolating these \(\delta\)-functions to \(U = 10^6\), we limit \(U_{\rm max}`\) to the largest available value.
ALPHA
structureThe free parameter \(\alpha\), the exponent of the power-law distribution of heating starlight intensities between \(U_{\rm min}\) and \(U_{\rm max}\). This structure contains the prior to assume for \(\alpha\). Values of \(\alpha\) are limited to being between
-10
and4
.GAMMA
structureThe free parameter \(\gamma\), the fraction of the dust mass exposed to the power-law distribution of radiation field intensities. This structure contains the prior to assume for \(\gamma\). Values of \(\gamma\) are limited to being between
0
and1
.QPAH
structureThe free parameter \(q_{\rm PAH}\), the fraction of the total grain mass corresponding to PAHs containing less than 1000 carbon atoms (PAH index). This structure contains the prior to assume for \(q_{\rm PAH}\). Values of \(q_{\rm PAH}\) are limited to being between
4.7e-3
and4.58e-2
.LTIR
structureThe free parameter \(L_{\rm TIR}\), the total integrated IR luminosity in \(L_\odot\). This structure contains the prior to assume for \(L_{\rm TIR}\). Values of \(L_{\rm TIR}\) are limited to being non-negative numbers.
Note
LTIR
is only a free parameter ifENERGY_BALANCE
not is set. IfENERGY_BALANCE
is set thenLTIR
is determined instead by the absorbed the stellar (and, if set, AGN) emission.
X-ray Emission#
XRAY_EMISSION
flagA flag indicating if an X-ray emission model will be used. This always includes stellar X-ray emission, but can optionally include AGN X-ray emission (see below). The stellar X-ray emission is normalized according to the \(L_X/M\) parametrizations with stellar age from Gilbertson et al. (2022).
Note
If no X-ray emission model is used, all X-ray emission model settings below can be skipped.
XRAY_UNIT
string scalarThe form (or unit type) of X-ray data within the input catalog. Currently, there are two types of X-ray data that can be input into Lightning. These are instrumental counts or fluxes (in \({\rm erg\ cm^{-2}\ s^{-1}}\)), which are selected by setting
XRAY_UNIT
to'COUNTS'
or'FLUX'
, respectively. See the discussion on Input Formats for more details on how to format the different X-ray data types.Note
If set to
'FLUX'
, theXRAY_UNC
setting below is ignored. Uncertainties on the X-ray flux must always be provided in the input catalog.XRAY_UNC
string scalarThe type of uncertainties to assume for the X-ray counts. In Lightning, the contribution to \(\chi^2\) from the X-ray model is calculated as
\[\chi^2_X = \sum_i \frac{(n^{\rm obs}_i - n^{\rm mod}_i)^2}{\sigma_{n,\ i}^2},\]where \(n^{\rm obs}_i\) is the number of net (background-subtracted) counts in energy bin \(i\), \(n^{\rm mod}_i\) is the number of model counts, and \(\sigma_{n,\ i}\) is the uncertainty on the observed counts. There are three types of X-ray count uncertainties currently available in Lightning. They are the square root of the counts, the upper uncertainty from the Gehrels (1986) approximation, and user input uncertainties. For the square root of the counts, \(\sigma_{n,\ i}\) is assumed to be \(\sqrt{n^{\rm obs}_i}\). This is most appropriate for cases where the number of counts is large enough that the errors are approximately Gaussian. For the Gehrels (1986) approximation, \(\sigma_{n,\ i}\) is assumed to be
\[1 + \sqrt{0.75 + n^{\rm obs}_i}.\]This is more appropriate for data in the low-count regime. Finally, for the user input uncertainties, Lightning searches each X-ray spectral file for a column labeled
NET_COUNTS_UNC
and adopts this as the uncertainty on the net counts. These uncertainties types are selected by settingXRAY_UNC
to'SQRT'
,'GEHRELS'
, or'USER'
, respectively.XRAY_ABS_MODEL
string scalarThe X-ray absorption model to apply to the X-ray emission. There are three X-ray absorption models currently available in Lightning. They are the “tbabs” absorption model with the default Wilms et al. (2000) abundances and the Sherpa “atten” model from Rumph et al. (1994). These X-ray absorption models are selected by setting
XRAY_ABS_MODEL
to'TBABS-WILM'
or'ATTEN'
, respectively.NH
structureThe free parameter \(N_H\), the intrinsic HI column density along the line of sight in \(10^{20}\ {\rm cm}^{-2}\). This structure contains the prior to assume for \(N_H\). Values of \(N_H\) are limited to being between
1e-4
and1e5
.Note
While the value of \(N_H\) is allowed to be larger than \(10^{24}\ {\rm cm^{-2}}\), we caution that our emission models are not suitable for the Compton-thick case.
XRAY_AGN_MODEL
string scalarThe AGN X-ray emission model to use. There are two AGN X-ray emission models currently available in Lightning, the qsosed models from Kubota & Done (2018) and a power law model with a high-energy exponential cut-off. The power law model has a fixed photon index of \(\Gamma = 1.8\) and an exponential cut-off at 300 \({\rm keV}\). This power law model is tied to the 2500 Angstrom emission using the relationship from Lusso & Risaliti (2017). These models are selected by setting
XRAY_AGN_MODEL
to'QSOSED'
and'PLAW'
, respectively. To fit the SEDs without any AGN X-ray emission models, setXRAY_AGN_MODEL
to'NONE'
.
Note
If the 'QSOSED'
AGN X-ray emission model is not chosen, its corresponding settings
below can be skipped.
QSOSED#
AGN_MASS
structureThe free parameter \(M_{\rm AGN}\), the supermassive black hole mass in \(M_\odot\). This structure contains the prior to assume for \(M_{\rm AGN}\). Values of \(M_{\rm AGN}\) are limited to being between
1e5
and1e10
.AGN_LOGMDOT
structureThe free parameter \(\log(\dot m)\), the \(\log_{10}\) of \(\dot m\), the supermassive black hole accretion rate normalized by the Eddington rate. This structure contains the prior to assume for \(\log(\dot m)\). Values of \(\log(\dot m)\) are limited to being between
-1.5
and0.3
.
AGN Emission#
AGN_MODEL
string scalarThe UV-to-IR AGN emission model to use. The only AGN emission model currently available in Lightning is the SKIRTOR model from Stalevski et al. (2016). This model is selected by setting
AGN_MODEL
to'SKIRTOR'
. To fit the SEDs without any AGN emission, setAGN_MODEL
to'NONE'
.
Note
If no AGN emission model is chosen, all AGN emission model settings below can be skipped.
SKIRTOR#
LOG_L_AGN
structureThe free parameter \(\log(L_{\rm AGN})\), the total integrated luminosity of AGN model in \(\log_{10}(L_\odot)\), which is used for normalization. This structure contains the prior to assume for \(\log(L_{\rm AGN})\). Values of \(\log(L_{\rm AGN})\) are limited to being between
0
and20
.Note
\(\log(L_{\rm AGN})\) will not be a free parameter if fitting using a
'QSOSED'
X-ray AGN model. Instead the normalization of UV-to-IR AGN model is tied to the rest-frame 2500 Angstrom monochromatic luminosity of the qsosed model.TAU97
structureThe free parameter \(\tau_{9.7}\), the edge-on optical depth of AGN dust torus at 9.7 \(\mu \rm m\). This structure contains the prior to assume for \(\tau_{9.7}\). Values of \(\tau_{9.7}\) are limited to being between
3
and11
.AGN_COSI
structureThe free parameter \(\cos i_{\rm AGN}\), the inclination of the AGN disk in terms of \(\cos i\). This structure contains the prior to assume for \(\cos i_{\rm AGN}\). Values of \(\cos i_{\rm AGN}\) are limited to being between
0
and1
.
Fitting Algorithm#
METHOD
string scalarThe fitting algorithm used to fit the SED(s). Lightning currently has three fitting algorithms that can be used: an adaptive MCMC, an affine-invariant MCMC, and a Levenberg–Marquardt algorithm. The adaptive MCMC algorithm is Algorithm 4 from Andrieu & Thoms (2008), the affine-invariant MCMC algorithm is the algorithm from Goodman & Weare (2010), and the Levenberg–Marquardt algorithm is Craig Markwardt’s MPFIT implementation. These fitting algorithms are selected by setting
METHOD
to'MCMC-ADAPTIVE'
,'MCMC-AFFINE'
, or'MPFIT'
, respectively.
Note
See our guide on Selecting an Algorithm for more details on each algorithm and their corresponding hyper-parameters below. Additionally, the guide can help you decide on the best algorithm to fit your research needs.
MCMC#
NTRIALS
int, float, or double scalarThe number of MCMC trials to run for each parallel walker/chain.
NPARALLEL
int, float, or double scalarThe number of parallel walkers/chains.
Note
If using the affine-invariant algorithm,
NPARALLEL
must be greater than the number of free parameters plus one and ideally at least twice the number of free parameters for optimal sampling.C_STEP
int, float, or double scalarWhen calculating the autocorrelation time (\(\tau\)) of the MCMC chain, this value defines how many trials of the chain are used to calculate \(\tau\), where we integrate \(\tau\) to the smallest index \(M\) such that \(M > C_{\rm step} \tau\).
TOLERANCE
int, float, or double scalarWhen calculating the autocorrelation time (\(\tau\)) of the MCMC chain, this value defines how many multiples of \(\tau\) the length of the chain should be for us to believe the estimated value of \(\tau\).
Note
We recommend using the default values for both C_STEP
and TOLERANCE
. More details on these parameters
can be found in the emcee Autocorrelation Analysis documentation.
BETA_EXPONENT
float or double scalarThe factor controlling how fast the adaptiveness of the adaptive MCMC algorithm vanishes. Larger values stop the adaptiveness in fewer trials.
Note
This is a setting only for the adaptive MCMC algorithm.
AFFINE_A
int, float, or double scalarThe move scaling constant defining the maximum and minimum step size of the affine-invariant stretch move.
Note
This is a setting only for the affine-invariant MCMC algorithm.
MPFIT#
NSOLVERS
int, float, or double scalarThe number of times to solve for the best fit SED using different starting locations in parameters space.
FTOL
float or double scalarThe relative error desired in the sum of squares. Termination of the MPFIT algorithm occurs when both the actual and predicted relative reductions in the sum of squares are at most
FTOL
.GTOL
float or double scalarThe orthogonality desired between the function vector and the columns of the Jacobian matrix. Termination of the MPFIT algorithm occurs when the cosine of the angle between function vector and any column of the Jacobian matrix is at most
GTOL
in absolute value.XTOL
float or double scalarThe relative error desired in the approximate solution. Termination of the MPFIT algorithm occurs when the relative error between two consecutive iterates is at most
XTOL
.MAXITER
int, float, or double scalarThe maximum number of MPFIT iterations to perform.
Post-processing#
KEEP_INTERMEDIATE_OUTPUT
flagA flag indicating that the intermediate
.sav
files produced by the fitting algorithm should not be deleted.Note
This is useful if needing to inspect the original fits before post-processing. Typically this will not be necessary, but if you are having trouble getting quality fits, inspecting the original fits can help determine the issue.
MCMC Post-processing#
The next four settings are the MCMC post-processing settings. These are only used if fitting with an MCMC algorithm, and they determine the how the MCMC chains are handled during post-processing for conversion to the posterior distributions.
BURN_IN
int, float, or double scalarThe number of initial MCMC trials to discard as the burn-in phase. If set to
0
, then the number will be chosen automatically from the autocorrelation time as\[{\tt BURN\_IN} = {\rm ceiling}(2\ {\rm max}(\tau)),\]where \(\tau\) is the autocorrelation time and
ceiling
is the ceiling function that rounds values up to the nearest integer.Note
We highly recommend specifying a value rather than using the automatic calculation when using the adaptive MCMC algorithm as the chains can vary widely in the number of autocorrelation times needed for burn-in.
THIN_FACTOR
int, float, or double scalarThe factor to thin the MCMC chain after removing the burn-in trials. Thinning of an MCMC chain is common practice, and it helps reduce the correlation between trials in the chain. To clarify what a value of
THIN_FACTOR
means, here are a few examples. A value of10
will only keep every 10th trial in the chain, and a value of1
will keep every trial (i.e., no thinning). Finally, if set to0
, then the value will be chosen automatically from the autocorrelation time as\[{\tt THIN\_FACTOR} = {\rm ceiling}(0.5\tau),\]where \(\tau\) is the autocorrelation time and
ceiling
is the ceiling function that rounds values up to the nearest integer.Note
We recommend specifying a
THIN_FACTOR
of4
and0
(the automatic calculation) when using the adaptive and affine-invariant MCMC algorithms, respectively. The reason for the4
value with the adaptive MCMC algorithm is that unique elements within the chains are minimally correlated. However, by design of the algorithm, a new unique element is only accepted into the chain every four or so trials. Therefore, by thinning by a factor of four, each element in the final chain will typically be unique.FINAL_CHAIN_LENGTH
int, float, or double scalarThe number of MCMC trials to include for the final distributions as taken from the truncated, thinned, and if necessary, merged chains. In other words,
FINAL_CHAIN_LENGTH
specifies then number of samples to include in the posterior distributions. To get the posterior distributions, the raw chains output by the MCMC algorithm have their burn-in discarded and are thinned and merged if necessary. Then, a number of samples equal toFINAL_CHAIN_LENGTH
will be taken from the end of the remaining chain to serve as the posterior distribution.Note
We recommend specifying a nice round value for
FINAL_CHAIN_LENGTH
such as250
,500
,1000
,2000
, etc. Larger values will increase the fine detail of the posterior distribution at the cost of increased post-processed file size.HIGH_RES_MODEL_FRACTION
int, float, or double scalarThe fraction of samples from
FINAL_CHAIN_LENGTH
, sorted by quality of fit, from which to generate high resolution models. If set to0
, then only the best fit high resolution model will be generated. This setting dictates how many high resolution models per SED will be in the post-processed file, which are useful for plotting purposes. Having a value ofHIGH_RES_MODEL_FRACTION
greater than zero would allow for having pointwise uncertainties on the best fit high resolution model, which can show, for example, areas of the model which are not well constrained by the data. It is important to stress that this setting gives the fraction of the best-fitting models in the posterior distribution for which high-resolution SEDs will be computed and saved. As an example, settingHIGH_RES_MODEL_FRACTION
to0.68
will return the high resolution models for the best 68% of fits in the posterior distribution.Warning
Including more than the best fit high resolution model can cause the file size of the post-processed file to balloon dramatically. Be careful when increasing this value above
0
. Doing so will increase the file size by at leastFINAL_CHAIN_LENGTH
*HIGH_RES_MODEL_FRACTION
*8
kB per SED per model component.AFFINE_STRANDED_DEVIATION
int, float, or double scalarThe number of standard deviations a walker must be below the median acceptance fraction of the ensemble to be considered a stranded walker. (See the Affine-Invariant MCMC description for more details on stranded walkers.)