6.2.7.2. UsineFit@FreePars¶
Enable fit/nuisance params for minimization runs, whose syntax is name:X,Y,[min,max],init,sigma
name
is the name of the parameter (see table below)
X
= FIT, FIXED, or NUISANCE:
- FIT: use the parameter name as a fit parameter. If the minimizer allows it, for instance Minuit, sets initial value to
init
, initial error tosigma
, and lower and upper values of the parameter limited to[min,max]
(use[-,-]
, or[-,max]
, or[min,-]
for no lower/upper limits).- FIXED: use the parameter name as a fixed parameter (if the minimizer allows it), set to
init
. The parameterssigma
and[min,max]
are unused.- NUISANCE: assumes a Gaussian distributed parameter \({\cal N}(\mu =\)
init
, \(\sigma =\)sigma
), which contributes to the global \(\chi^2\) as\[\chi^2_{\rm nuisance} = (\frac{{\rm val}-{\rm init}}{\sigma})^2\]where val is the value tested in the fit. Not that we strongly penalize values outside
[min,max]
(they are excluded in practice).
Y
= LIN or LOG: enables to sample the parameter as name or \(\log_{10}\) (name).
Warning
All @ParNames parameters can be in principle declared as fit parameters. However, a subtlety arises because of the possibility to have universal vs per CR source parameters: for universal parameters, USINE directly uses the parameter name, otherwise it builds NCR parameters X_Y with X=parameter name and Y=CR name (e.g., for q
, generates q_1H
, q_2H
, q_3HE
… which can all be left as free parameters). In addition, modulation parameters can be declared as free/nuisance parameters for all data periods used in FitTOAData (see below), and data error uncertainties can also be added as nuisance parameters: in both cases, the name of the parameter is built internally from the experiment name (removing special characters ([*%@)]...
by _
). Run ./bin/usine_run -m1 inputs/your.init.file
(with your.init.file you initialisation file) to list what parameters can be left free for your initialisation file.
Group@subgroup@parameter | Description |
---|---|
UsineFit@FreePars@CRs (M=1) | Half-life of all unstable CRs are enabled as nuisance parameters, e.g. HalfLifeBETA_10Be:NUISANCE,LIN,[1.3,1.5],1.387,0.012 (use EC instead of BETA if EC-unstable nucleus). |
UsineFit@FreePars@DataErr (M=1) | If @FitTOAData@ErrType parameter set to kERRCOV for CR quantity (see next subgroup), its systematics are enabled as centred nuisance parameters, e.g., - If a systematics not set as nuisance parameter, it is accounted for in covariance matrix (see Section 5.5);
- if systematics set as nuisance parameter, amounts to a model bias: for \(N_s\) types of systematics set as nuisance for a quantity Q , the model calculation read
\({\rm model}^Q(R_k) = {\rm model}^Q_{\rm true}(R_k) \times {\rm bias}^Q(R_k) =\)
\({\rm model}^Q_{\rm true}(R_k) \times \Pi_{l=0\dots n} (1+\nu^Q_l \times \sigma^Q_l(R_k))\),
with \(\nu^Q_l\) the nuisance parameter (centred on 0, variance 1) and \(\sigma^Q_l(R_k)\) the relative error for the calculated point \(R_k\) (read from the diagonal of the covariance matrix for simplicity). In that case, the systematics is not in \(\chi^2_{\rm cov}\), but as a standard \(\chi^2_{\rm nuis}\) (see above).
|
UsineFit@FreePars@Geometry (M=1) | Any parameter in Model…@Geometry@ParNames. |
UsineFit@FreePars@ISM (M=1) | Any parameter in Model…@ISM@ParNames. |
UsineFit@FreePars@Modulation (M=1) | Run ./bin/usine with option -m1 to get enabled names for data and modulation levels, e.g. phi_AMS02_201105201605_:NUISANCE,LIN,[0.3,,1.1],0.73,0.2 ( Force-Field approx.). |
Run@FitFreePars@SrcSteadyState (M=1) | Any source parameter, e.g. q_4HE:FIT,LOG,[-5.,-3],-4.,0.1 or alpha:NUISANCE,LIN,[1.7.,2.5],2.3,0.1 |
Run@FitFreePars@Transport (M=1) | Any parameter in Model…@Transport@ParNames, e.g. delta:FIT,LIN,[0.2,0.9],0.6,0.02 , or Va:FIT,LIN,[1.,120],70.,0.1 , etc. |
UsineFit@FreePars@XSection (M=1) | XS parameters format are PREFIX_REACTION (see examples in - ‘PREFIX’ enabled:
- for standard bias of XS:
- Norm → normalisation parameter (all E)
- EAxisScale → \(\sigma(E_{k/n}^{\rm new})=\sigma(E_{k/n} \times {\rm EAxisScale})\)
- EnhancePowHE → (
TUXSections::XS_EnhancementHE )^(EnhancePowHE)- EThresh and SlopeLE → below threshold \(\sigma(E_{k/n}) = (E_{k/n}/{\rm EThresh})^{\rm SlopeLE} \times \sigma(E_{k/n})\), with \({\rm EThresh}=5~{\rm GeV/n}\) if not set as nuisance.
- for linear combination of XS (\(\sigma_{\rm LC} = \sum_i C_i \times \sigma_i\)):
- LCInelBar94, LCInelTri99, LCInelWeb03, LCInelWel97 for inelastic XS,
- LCProdGal17, LCProdSou01, LCProdWeb03, LCProdWKS98 for production XS.
- enabled ‘REACTION’ keywords
- Total inelastic XS:
-
12C+H → specific reaction-
ANY+H → any CR on given target-
12C+ANY → 12C on all ISM targets- Production XS:
-
12C+H->10Be → specific reaction-
ANY+H->10Be → any CRs-
12C+ANY->10Be → any target-
12C+H->ANY → any fragment- Tertiary XS (tot. and diff. similarly biased):
-
1H-BAR+H->1H-BAR → specific reaction-
1H-BAR+ANY->1H-BAR → any target-
ALL → if not linear combination of XS, applies to all XS (inel, prod, and tertiary); otherwise, can be applied for all linear combination of inelastic and/or production XS applies to all XS (inel, prod, and tertiary)!Note that it is not allowed to have the same prefix used for overlapping reactions (e.g., 12C+ANY has overlapping reactions with 12C+H).
|