Compatibility with HS3

In parallel to development of the model description format, compatibility with HEP Statistics Serialization Standard (HS3) is investigated.

• HS3 repository
• Serialization of hadronic models, discussion #34
• Cascade models in HS3, discussion #36

Support in ROOT, what is missing

• Class: RooAmplitudeModelPDF
  • Implementation of math
  • Exporter: RooAmplitudeModelPDF::Streamer
  • Importer: RooAmplitudeModelPDF::Factory
• All lineshapes: BreitWigner, BlattWeisskopf

Example of HS3 file

• HS3_base.json give description of basic interfaces of the model variables and parameters consistently with HS3. It sets default value of variable, and describes a validation point in HS3 consistent way.
• HS3_extended.json prototypes
  • adding jacobian (prior) for phase space generation,
  • replacing phase space variables

Parameters of the model

The model parameters have to be listed in the parameter block of distributions.

"distributions": [
  {
    "name": "my_model_for_reaction_intensity",
    "type": "hadronic_cross_section_unpolarized_dist",
    "model_description": {
    },
    "variables": [
    ],
    "parameters": ["L_1520_mass"]
  }
]

The default values of the parameters are not mandatory, however, they can be listed as one of the element of the parameter_points array.

"parameter_points": [
  {
    "name": "default_values_of_my_model",
    "parameters": [
      {
        "name": "L_1520_mass",
        "value": 1.52
      }
    ]
  }
]

The validation block has to include

"parameter_points": [
  {
    "name": "validation_point",
    "parameters": [
      {
        "name": "L_1520_mass",
        "value": 4.6
      },
      {
        "name": "L_1520_mass",
        "value": 4.6
      },
      {}
    ]
  },
  {}
]

Sampling from the model

To sample from the model using the kinematic variables defined in variable section, a few sections has to be added.

"likelihoods": [
  {
    "name": "my_nll",
    "pdfs": ["my_model_for_reaction_intensity"],
    "datasets": []
  }
]

Likelihood section clarifies how the log likelihood is computed from the pdfs .

"analyses": [
  {
    "name": "my_ana",
    "parameter_domain": "default",
    "prior": "my_prior_dist"
  }
]

Analyses specifies how kinematic variables are distributed within the phase space, using priors (here, "prior": "my_prior_dist") to influence their sampling distributions.

The priors are given in the list of distributions.

"distributions": [
  {},
  {
    "name": "my_prior_dist",
    "type": "product_dist",
    "elements": ["m12_prior"]
  },
  {
    "name": "m12_prior",
    "type": "phase_space_density_1d",
    "variable": "m_12"
  }
]

Here we define my_prior_dist as a product distribution, supposedly incorporating all non-trivial dimensions of the phase-space density (the mass dimensions).