2019 IRIS-HEP AS workshop: Particles and decays

Eduardo Rodrigues, Henry Schreiner
University of Cincinnati , Princeton University
Particles and decays
in the Scikit-HEP project
IRIS-HEP AS workshop, 19th June 2019
PDG particle data
and identification codes
Parse decay files, describe
and convert particle decays
between digital
representations

IRIS-HEP AS Workshop, 19th June 2019 2/22Henry Schreiner and Eduardo Rodrigues
The grand picture – the Scikit-HEP project
InteroperabilityReproducibility
Collaboration
 Create an ecosystem for particle physics data analysis in Python
 Initiative to improve the interoperability between HEP tools
and the scientific ecosystem in Python
- Expand the typical toolkitset for particle physicists
- Set common APIs and definitions to ease “cross-talk”
 Initiative to build a community of developers and users
- Community-driven and community-oriented project
 Effort to improve discoverability of relevant tools
Sustainability

Who uses (some of) Scikit-HEP ?

PDG particle data
and identification codes

Particle package – motivation
 The Particle Data Group (PDG) provides a downloadable table of particle masses, widths, charges
and Monte Carlo particle ID numbers (PDG IDs)
- Most recent file here
 It also provided an experimental file with extended information (spin, quark content, P and C parities, etc.)
until 2008 only, see here (not widely known!)
 But anyone wanting to use these data, the only readily available, has to parse the file programmatically
 Why not make a Python package to deal with all these data, for everyone?
 The C++ HepPID and HepPDT libraries provide functions for processing particle ID codes
in the standard particle (aka PDG) numbering scheme
 Different event generators have their separate set of particle IDs: Pythia, EvtGen, etc.
 Again, why not make a package providing all functionality/conversions, Python-ically, for everyone?

Particle package – overview
 Pythonic interface to
PDG particle data table
and
particle identification codes
 With extra goodies
 2 separate submodules
 Comprehensive documentation (docstrings)
 Continuous Integration (CI): extensive tests for excellent test coverage
- In packages such as these, tests should target both the code itself but also the physics it deals with!
 We use Azure DevOps
- Seamlessly test on Linux, macOS and Windows

Particle package – PDG IDs module overview
 Process and query PDG IDs, and more – no look-up table needed
- Current version of package reflects the latest version of the HepPID & HepPDT utility functions defined in
the C++ HepPID and HepPDT versions 3.04.01
- It contains more functionality than that available in the C++ code … and minor fixes too
 Definition of a PDGID class, PDG ID literals,
and set of standalone HepPID functions to query PDG IDs
(is_meson, has_bottom, j_spin, charge, etc.)
- All PDGID class functions are available standalone
 PDG ID queries also available on the command line
 PDGID class
- Wrapper class for PDG IDs
- Behaves like an int, with extra goodies
- Large spectrum of properties and methods,
i.e. the functions defined in the HepPID and HepPDT
C++ libraries, with a Pythonic interface, and yet more
- To print them all:

Particle package – particle module overview
 Simple and natural API to deal with the PDG particle data table, with powerful look-up and search utilities
 Definition of a Particle class and particle name literals
- Typical queries should be, and are, 1-liners
 Advanced usage: ability to specify or build a particle data table, conversion tools
 Particle / PDG ID searches available on the command line too

Particle package – data files
 All data files stored under particle/data/
 PDG particle data files
- Original PDG data files, which are in a fixed-width format
- Code uses “digested forms” of the PDG files, stored as CSV, for optimised querying
- Latest PDG data used by default (2019 at present)
- Advanced usage: user can load older PDG table, load a “user table” with new particles, append to default table
 Other data files
- CSV file for mapping of PDG IDs to particle LaTeX names
…
…

Particle package – particle look-up
 Particle class
- Standard look-up via from_pdgid(…)
- Various other from_X(…)
methods exist
 - Large spectrum of
properties and methods:
- Get particle properties
- Deal with underlying particle table
- Powerful search engine …
 Particle literals
- Easily recognizable names for manipulations,
e.g. in plots

Particle package – powerful particle search
 Particle.find(…) – search a single match (exception raised if multiple particles match the search specifications)
 Particle.findall(…) – search a list of candidates
 Powerful search methods
that can query any particle property!
 One-line queries
 E.g., trivially find all pseudoscalar charm mesons:

Particle package – future directions & developments
 Addition of particle IDs and names relevant to other MC programs
- (Yes, not consistent across programs!)
- Useful IDs such as those used in PYTHIA, Geant and EvtGen
 Bring in other communities where Particle is / can be relevant
- Ongoing discussions with astroparticle physics community
- Particle IDs used in EPOS, CORSIKA, DpmJet, QGSJet, Sybill, UrQMD, …
 Ongoing discussions with PDG group
- Provide the right tool
- Can we provide more?
- Stay tuned …

Parse decay files, describe and
convert particle decays
between digital representations

DecayLanguage package – motivation and overview
Motivation
 Ability to describe decay-tree-like structures
 Provide a translation of decay amplitude models from AmpGen to GooFit
- Idea is to generalise this to other decay descriptions
 Any experiment uses event generators which, among many things, need to describe particle decay chains
 Programs such as EvtGen rely on so-called .dec decay files
 Many experiments need decay data files
 Why not make a Python package to deal with decay files, for everyone?
Overview
 Tools to parse decay files and programmatically manipulate them, query, display information
- Descriptions and parsing built atop the Lark parser
 Tools to translate decay amplitude models from AmpGen to GooFit, and manipulate them
Library and set of applications for fitting and
generating
multi-body particle decays using the isobar model

DecayLanguage package – decay files
“Master file” DECAY.DEC
 Gigantic file defining decay modes for all relevant particles,
including decay model specifications
 LHCb example:
~ 450 particle decays, thousands of decay modes,
over 11k lines in total
User .dec files
 Needed to produce specific MC samples
 Typically contain a single decay chain
(except if defining inclusive samples)
Define dm 0.507e12
...
Alias B0sig B0
Alias anti-B0sig anti-B0
ChargeConj B0sig anti-B0sig
...
Decay pi0
0.988228297 gamma gamma PHSP;
0.011738247 e+ e- gamma PI0_DALITZ;
0.000033392 e+ e+ e- e- PHSP;
0.000000065 e+ e- PHSP;
Enddecay
...
CDecay tau+
...
# Decay file for [B_c+ -> (B_s0 -> K+ K-) pi+]cc
Alias B_c+sig B_c+
Alias B_c-sig B_c-
ChargeConj B_c+sig B_c-sig
Alias MyB_s0 B_s0
Alias Myanti-B_s0 anti-B_s0
ChargeConj MyB_s0 Myanti-B_s0
Decay B_c+sig
1.000 MyB_s0 pi+ PHOTOS PHSP;
Enddecay
CDecay B_c-sig
Decay MyB_s0
1.000 K+ K- SSD_CP 20.e12 0.1 1.0 0.04 9.6 -0.8 8.4 -0.6;
Enddecay
CDecay Myanti-B_s0

DecayLanguage package – decay file parsing and display
 Parsing should be simple
- Expert users can configure parser choice and settings, etc.
 Parsing should be (reasonably) fast
- Example of LHCb’s master DECAY.DEC file:
Over 11k lines in total, ~ 450 particle decays, ~60 charge-conjugate decays created on-the-fly (‘CDecay’ statements),
thousands of decay modes
 After parsing, many queries are possible
 One can also visualise decay chains … 

Decay chain – simplest view with no sub-decays shown
Decay D*+
0.6770 D0 pi+ VSS;
0.3070 D+ pi0 VSS;
0.0160 D+ gamma VSP_PWAVE;
Enddecay
Decay D*-
0.6770 anti-D0 pi- VSS;
0.3070 D- pi0 VSS;
0.0160 D- gamma VSP_PWAVE;
Enddecay
Decay D0
1.0 K- pi+ PHSP;
Enddecay
Decay D+
1.0 K- pi+ pi+ pi0 PHSP;
Enddecay
Decay pi0
0.988228297 gamma gamma PHSP;
0.011738247 e+ e- gamma PI0_DALITZ;
0.000033392 e+ e+ e- e- PHSP;
0.000000065 e+ e- PHSP;
Enddecay
(Considered by itself, this file in in fact incomplete,
as there are no instructions on how to decay the anti-D0 and the D-.
Good enough for illustration purposes, though.)

Decay chain – p0 decays displayed

Decay chain – full view

DecayLanguage package – conversion of decay models / representations
 Decay chains
- A universal modelling of decay chains would profit many use cases,
e.g. description of components for amplitude analyses
 Present code understands AmpGen syntax and can generate code for the GooFit fitter
 Note:
makes use of the Particle package

DecayLanguage package – future directions & developments
Decay files
 Streamline and enhance the .dec parser
- Ex.: syntax such as
p.find_decay_chains(final_state=[‘K+’, ‘K-’, ‘pi+’, ‘pi-’], extra_particles=[‘pi0’])
could be a neat/trivial way to query the master DECAY.DEC and
“find all decay chains leading to either ‘K+ K- pi+ pi-’ or ‘K+ K- pi+ pi- pi0’”
 Provide a universal description and visualisation of decay trees (a lot done on this in the last week …)
- We already have customers interested, e.g. visualisation of decays in pyhepmc
Decay models / representations
 Implement more backend formats: GooFit in Python, etc.
 Longer term – implement decay logic inside model descriptions
- Provide a reference for other packages

Interested ? Want to try it ?
Particle
 GitHub: https://github.com/scikit-hep/particle/
 Releases: PyPI
 Kindly recognise software work – cite us:
DecayLanguage
 GitHub: https://github.com/scikit-hep/decaylanguage
 Releases: PyPI
Scikit-HEP project
 GitHub: https://github.com/scikit-hep/
 Website: http://scikit-hep.org/
 Get in touch: http://scikit-hep.org/get-in-touch.html
See interactive demos:
- Particle
- DecayLanguage

1/4
AmpGen is a library for tting and
generating multibody particle decays.
Takes a decay chain representation and
converts into fast, low-level code.
Fairly agnostic about the types of particles in
the initial / nal state (pseudoscalars, NEW
fermions, photons ...)
No hard limit on number of particles in the
nal state, practically, more than about six
bodies is a dicult.
Classic three-body (pseudoscalar) decays are
a two-body problem (A Dalitz plot).
Four-body decays are a ve-dimensional
problem =⇒ only unbinned ts are practical.
Speed is critical as already have O(106)
candidates in many channels, and expecting
10 → 50× these yields by 2030.
For a typical four-body amplitude, can
perform ∼ 106evaluations/s/core on consumer
laptop.
Part of the GooFit organisation on GitHub.
High Level Code
Low Level Code
Decay chain representation
Decay descriptors:

2/4
More on the user interface
User species couplings between dierent particles, either partially or as full decay chains, for
example:
D0{K*(892)bar0,rho(770)0}
Couples a D0 meson to a pair of vectors, then if you add the couplings:
K*(892)bar0{K-,pi+}
K*(892)bar0{K0S0,pi0}
K*(892)bar0{K0L0,pi0}
rho(770){pi+,pi-}
You can generate the amplitude coupling the D0 meson to the various nal states.
You can add additional orbital couplings as
D0[P]{K*(892)bar0,rho(770)0}
D0[D]{K*(892)bar0,rho(770)0}
Which will be coherently summed to give the total amplitude, mod-squared to give the
dierential rate:
P(ψ)dψ ∝
i
giAi(ψ)
2
dψ (1)

3/4
Developments this year
Big development is the inclusion of spin-half
particles in the initial/nal state.
Have to deal with additional spin-indicies in
the probability / need to incoherently sum the
dierent spin congurations.
In general, this leads to a rate equation that
looks like (for an initial state polarised as ˆρ
P (ψ, ˆρ) dψ =
ll
glgl
imm
(δmm +σmm ˆρ)Al
mi(ψ) Al
m i(ψ)
∗
(2)
A tricky thing to calculate (and normalise)
eciently!
Toy examples
Λb → Λpp with intermediate spin-3/4
contributions:
2.5 3 3.5 4 4.5
mΛp [GeV/c2
]
0
0.5
1
1.5
2
2.5
3
3.5
4
4.5
×103
Entries/(0.03GeV/c2
)
2 2.5 3 3.5 4 4.5
mpp [GeV/c2
]
0
2
4
6
8
10
12
14
16
18
20
22
24
×103
Entries/(0.03GeV/c2
)
B → pppp
10
20
30
40
50
10 15
sppp
8
10
12
14
16
18
sppp

4/4
Future developments
With some minor modications, can generate Cuda kernels that can be integrated into GooFit
Evaluate amplitude D0 → K∗
2 (1430)0 [K−π+] π0 on 8.4 million events with 1 CPU core (i7-8550U
CPU @ 1.80GHz) vs 1 GPU (GeForce GTX 1050M)
GPU version is about a factor of two faster than the OpenMP version (i.e. using 8 cores rather
than 1 for the CPU version). Further tuning of the code generator has given another factor of
2 → 3 compared to what is shown here.
1 2 3
sK−π+ GeV2
/c4
0.5
1
1.5
2
2.5
3
sK−π0GeV2
/c4
CPU ∼ 1300ms
1 2 3
sK−π+ GeV2
/c4
0.5
1
1.5
2
2.5
3
sK−π0GeV2
/c4
GPU ∼ 89ms

Additional notes
Cincinnati
• Future work in GooFit: line shapes/spinfactor system, 𝐾 𝑠 𝜋𝜋, support AmpGen baryonic
decays - may not be done in a year
• Daniel Vieira starts in October
Princeton - Henry
• ROOT Conda-Forge
• Some further work on fitting (GooFit and others)
1/1Henry Schreiner Other projects June 19, 2019

 PDGID literals
- Provide (PDGID class) aliases
for the most common particles,
with easily recognisable names
 All is consistent. Ex.:
Particle package – PDG identification code literals
 Literals: handy way to manipulate things with human-readable names

DecayLanguage package – Lark parser grammar for decay files
start : _NEWLINE? (line _NEWLINE)+ (End _NEWLINE)?
?line : define | pythia_def | alias | chargeconj | commands | decay | cdecay | setlspw
pythia_def : PythiaBothParam LABEL : LABEL = (LABEL | SIGNED_NUMBER)
setlspw : SetLineshapePW label label label value
cdecay : CDecay label
define : Define label value
alias : Alias label label
chargeconj : ChargeConj label label
?commands : global_photos
global_photos : boolean_photos
boolean_photos : yesPhotos - yes
| noPhotos - no
decay : Decay particle _NEWLINE decayline+ Enddecay
decayline : value particle* photos? model _NEWLINE // There is always a ; here
value : SIGNED_NUMBER
photos : PHOTOS
label : LABEL
particle : LABEL // Add full particle parsing here
model : MODEL_NAME model_options?
model_options : (value | LABEL)+
%import common.WS_INLINE
%import common.SIGNED_NUMBER
// New lines filter our comments too, and multiple new lines
_NEWLINE: ( /r?n[t ]*/ | COMMENT )+
MODEL_NAME.2 : BaryonPCR|BTO3PI_CP|BTOSLLALI|BTOSLLBALL|BTOXSGAMMA|BTOXSLL| ...
LABEL : /[a-zA-Z0-9/-+*_()']+/
COMMENT : /[;#][^n]*/
// We should ignore comments
%ignore COMMENT
// Disregard spaces in text
%ignore WS_INLINE
 Decay file parser
grammar:
decfile.lark !
 This file is enough
to parse and
understand
decay files

2019 IRIS-HEP AS workshop: Particles and decays

More Related Content

What's hot

Similar to 2019 IRIS-HEP AS workshop: Particles and decays

More from Henry Schreiner

Recently uploaded

2019 IRIS-HEP AS workshop: Particles and decays