Collider physics has a steep entry ramp: a vocabulary nobody writes down, statistical machinery that arrives fully assembled, and reconstruction algorithms treated as black boxes. These are the explanations I wish had existed when I started — written for students joining the group, but open to anyone. No background is assumed beyond curiosity and a little calculus.
The opening installment of a five-part tour of normalizing flows: a plain-words, poke-the-pictures introduction to what a probability distribution actually is and what it means to reshape one. Read this first if the phrase “generative model” has ever gone by without landing.
The rest of the tour, in order: Part 2 · Keeping the Receipts · Part 3 · Machines That Bend · Part 4 · Make the Data Likely · Part 5 · Flows Without Steps
The complete tour — take a plain bell curve and bend it, carefully, until it matches your data. Five sections of sliders and live demos carry you from probability conservation through invertible transformations to image generation, closing with a bonus on importance sampling for numerical integration: the same trick, pointed at an integral instead of a picture.
Free, self-paced modules — most from the HEP Software Foundation Training Center, built by physicists for physicists. Each takes an afternoon to a few days. Do the first row before anything else.
What every number in a Les Houches Event file means and how to parse one that will not fit in memory: the init and event blocks, status codes and mother indices, colour flow, the IDWTUP weight conventions and what negative weights cost you, the LHEF 3 named weights and event attributes, and why every real reader is a line-by-line state machine rather than an XML parser. Comes with a dependency-free streaming Python reader, a validator, and a two-event sample file to try it on.
Predict, filter, smooth — the algorithm behind essentially every track in every collider experiment, built up one plane at a time. Covers what a track is numerically, propagation between detector layers, the filter and smoother steps, search windows for hit finding, whether you should believe your covariance matrix, and the failure modes that will bite you. Runs on a toy tracker you can follow end to end.
The Review of Particle Physics is the reference every number in a HEP paper traces back to — masses, widths, branching fractions, couplings, and the review articles that define how the field does statistics, kinematics and Monte Carlo. The 2026 edition is current; the reviews below are the ones you will open most.
pip install pdg) access to the listings, plus SQLite dumps and the mass_width table generators read.Every particle in every generator, HepMC record and analysis ntuple is identified by a PDG ID — an integer defined by the Monte Carlo Particle Numbering Scheme (Buckley, Krauss, Navas, Richardson). Learn the pattern once and you will never look up a hadron code again.
Fundamental particles get one or two digits; hadrons get up to seven. Read a hadron code from the right: nJ = 2J+1 (so 1 for pseudoscalars, 3 for vectors, 2 for spin-½ baryons, 4 for spin-3/2), then the quark content nq1nq2nq3 in decreasing order (mesons use two digits, with the heavier quark first), then nL and nr for orbital and radial excitations. The sign gives particle versus antiparticle; neutral self-conjugate states have no negative code. Codes 81–100 are reserved for generator internals (Pythia's string is 92, a cluster 91), nuclei are ten digits ±10LZZZAAAI, and SUSY partners add 1000000 or 2000000 (left/right) to the SM code.
In Python, particle from Scikit-HEP implements
the whole scheme: Particle.from_pdgid(211).name, PDGID(1000022).is_sm_lepton, and a full table of properties.
| PDG ID | Particle | Why that number |
|---|---|---|
| 1–6 | d, u, s, c, b, t | quarks by generation, down-type first |
| 11, 12 … 16 | e−, νe, μ−, νμ, τ−, ντ | leptons; note e− is +11, e+ is −11 |
| 21, 22, 23, 24, 25 | g, γ, Z, W+, H | gauge bosons and the Higgs |
| 32, 33, 34, 35–37 | Z′, Z″, W′, H0/A0/H+ | the standard BSM boson slots |
| 111, 211 | π0, π+ | light mesons, nJ=1 |
| 221, 331, 113, 213, 223 | η, η′, ρ0, ρ+, ω | vectors end in 3 |
| 321, 311, 130, 310 | K+, K0, KL, KS | the neutral kaon mass eigenstates are special-cased |
| 411, 421, 431, 443 | D+, D0, Ds+, J/ψ | charm: leading digit 4 |
| 521, 511, 531, 553 | B+, B0, Bs0, Υ | beauty: leading digit 5 |
| 2212, 2112, 3122, 3222 | p, n, Λ, Σ+ | baryons: three quarks then 2J+1=2 |
| 2224, 1114 | Δ++, Δ− | spin-3/2 baryons end in 4 |
| 1000022, 1000006, 2000006 | χ̃01, t̃1, t̃2 | SUSY: 1000000/2000000 + SM code |
| 1000010020, 1000020040 | deuteron, α | nuclei: 10LZZZAAAI |
| 91, 92, 93 | cluster, string, indep. frag. | generator-internal objects (81–100) |
| 4900xxx, 9900xxx | hidden-valley, user-defined | reserved blocks for models and private use |
The group is on ATLAS, so ATLAS entries come first and CMS equivalents follow. A CERN login tag means the page needs a CERN account or grid certificate.
Run 3 ended on 29 June 2026 and the machine is now in Long Shutdown 3; the accelerator complex restarts from 2028 and the HL-LHC is scheduled to begin in 2030.
Summer schools post their slides on Indico and leave them there; a good school's timetable is a free, current textbook on the topic. These are the ones whose material is worth mining even if you never attend. (Application deadlines for the schools you can still apply to are on the Opportunities page.)
CERN, 26 August – 4 September 2026. An advanced school for PhD students and early postdocs built around the “Physics, Experiments and Detectors” questions of the next machines, with FCC-ee and the EIC as the worked examples. Two weeks of lectures with PDFs on the timetable: BSM (S. Renner), electroweak (A. Vicini), QCD (G. Stagnitto), flavour (G. Isidori), PDFs and FCC-hh (V. Bertone, S. Williams), accelerators and EPOL (F. Asvesta, T. Raubenheimer, J. Keintzel, T. Pieloni), silicon, calorimetry, gas and RICH detectors (W. Snoeys, S. Monteil, R. Farinelli, F. Tessarotto), detector simulation and physics performance (A. Zaborowska, M. Selvaggi), machine–detector interfaces, the muon collider (D. Lucchesi), AI for physics (M. Pierini) and software (B. François). If you are joining the FCC-ee generator work, start with the electroweak, QCD and physics-performance sets.
The US LHC Users Association's practical guide for Americans relocating to Geneva — housing, registration, banking, transport, and the paperwork nobody warns you about — plus a companion page on voting in US elections from CERN. Read it the week you get your travel dates.
390 searchable entries across theory, reconstruction, statistics, object identification, software, calorimetry, trigger and DAQ, collaboration practice, beams, Monte Carlo, upgrades, and muon systems — each with a definition and a note on what actually trips people up. The reference for the moment a collaboration meeting turns into acronyms.
Found an error, or something that deserves an entry? Write to saptaparnab@smu.edu — corrections from newcomers are the most useful kind, because you are the one who noticed the gap.