Repository navigation
Study 50 CMS Raw Data From The LHC
In March 2011 the Large Hadron Collider at CERN smashed protons together, and the CMS detector wrote down everything it saw. CERN made those recordings public, byte for byte. Study 50 lets anyone with a Mac open them and watch the collisions β as a documentary in seven chapters, as a hologram you can turn in your hands, computed in exact integers on your own machine.
Press release: Study 50 β CMS's 2011 collision data, turning in your hands on a Mac
- Download the Affine IDE from affine.earth/download β version 0.2.6.1 or later.
- Open it and choose Study 50 β the LHC's own data in the studies list.
- Watch. The IDE reads the first file straight from CERN and the story begins on its own.
Take the controls whenever you like: β βΆ step one collision, β file / file βΆ move through the corpus, LIVE follows the stream again, Turn / Slide / Zoom move the hologram (drag, shift-drag, scroll). There is nothing else to install, no account, no GPU, and nothing is kept on your disk except the integers the study finds.
The canvas is CMS itself, drawn as a hologram: the pixel layers at the centre, the silicon strip tracker, the crystal calorimeter (ECAL), the hadron calorimeter (HCAL), the 3.8-tesla magnet and the muon chambers seven metres out. Two gold bunches of protons fly in along the beam and meet in the middle. Then the detector answers, from the inside out:
- cyan points β every pixel that registered charge in that collision, placed where its chip sits inside CMS;
- emerald paths β the tracks of charged particles: three hits, one on each pixel layer, that line up the way a particle curling out through the magnet lines up; drawn on, dimmer, along the curve their own bend implies;
- the glow of each part of the detector β how much it had to say about this collision, in bytes.
The canvas switches between three views of the same collision β from the side, and these two, which complete the picture in space and time:
- Down the beam β the view physicists read collisions in: looking straight along the beam pipe at the three pixel rings, every hit around the beam spot, and each track as the curve the magnet bends it into. A switch moves between the pixel layers close up and the whole silicon tracker.
- The LHC clock β one turn of the LHC as a ring of 3,564 bunch slots, each drawn with how many collisions it has carried so far; the bunch trains stand out as spokes. This collision's slot is marked in gold; when its event record and its pixel boards disagree on the slot, the boards' slot is marked too, three slots away.
| down the beam | the LHC clock |
|---|---|
 |
 |
Beside the canvas, two forms name every element on screen β its colour, what it is, and its value for this collision β and walk through the projection step by step as it plays, the step being drawn highlighted.
The chat beside the canvas tells the story chapter by chapter, with the numbers for the collision on screen.
1 Β· The bytes. In 2011 the LHC collided protons at 7 TeV, 3.5 TeV in each beam. CMS recorded run 160957 on 21 March 2011, and CERN publishes those recordings exactly as the detector wrote them: Open Data record 35, /MinimumBias/Run2011A-v1/RAW, DOI 10.7483/OPENDATA.CMS.I8HN.DF32, free to use (CC0) β 71 files, 248,789,344,017 bytes, 1,913,190 recorded events. These are the LHC's ordinary collisions β the record calls them soft-QCD events, kept for the presence of low-energy particles, not for anything rare. Affine fetches each file from CERN in 16 MiB pieces, keeps none of it, and checks every byte against the checksum CERN publishes.
2 Β· The container. The file is ROOT, the container format of particle physics: a chain of compressed records. One record holds one collision's complete readout; a second stream holds each collision's own event record β run, luminosity section, event number, time, LHC orbit and bunch slot. Affine opens it with its own decompressor β no ROOT library, no CMS software β and checks every compressed block against its own checksum.
3 Β· The crossing. The LHC's beams circulate in opposite directions; each turn of the ring has 3,564 bunch slots, and where a slot holds a bunch from each beam, the bunches cross at the centre of CMS. The first collision of the first file carries the timestamp 2011-03-21 03:09:05.121034 UTC in its event record, on orbit 175,639,113, in bunch slot 2,705. Its event record and its pixel boards' own headers carry the slot independently, and they agree. Affine places every collision on the clock axis C4 = (run, luminosity section, orbit, bunch slot) β four integers, as recorded.
4 Β· The detector answers. A collision is read out by up to 1,024 boards β pixel, strip tracker, preshower, ECAL, HCAL, the muon chambers and the trigger. Every board opens with the same header and closes with a trailer that states its own length. Affine checked every one: across the first file's 16,915 collisions, 10,588,790 board readouts (626 boards in every collision), none broken β one ECAL readout flagged its own checksum, and was counted.
5 Β· The innermost layer. The pixel tracker sits 4.4, 7.3 and 10.2 cm from the beam, with two disks at each end: 15,865 chip entries in CMS's cabling table, each chip a grid of 80 Γ 52 pixels. A pixel reports only when the charge in it crosses a threshold. Affine places each hit where its chip sits β using CMS's own published cabling table and chip-to-module map β and bonds neighbouring hits (same chip, row and column each within one) into clusters. Then it finds the paths: three clusters whose angle around the beam and distance along it both grow in step with the radius. Their lines meet the beam at one point β where along the beam that collision happened, typically a few centimetres from the centre of CMS.
6 Β· Every crossing. Over the whole stream two kinds of crossing appear. Across the first five files, 138 of the 3,564 bunch slots carry 93,110 of the 120,742 collisions, and there a crossing lights up 2,850 pixels on average; at every other slot, 60. The busy slots are where collisions happened; at the others the detector sees almost nothing. That is the LHC's bunch pattern, read straight out of the detector. And one quiet fact that only the whole stream reveals: in about 3 % of crossings the event record's bunch slot is exactly 3 lower than the slot in the pixel boards' headers β in 3,854 of 120,742 collisions over five files, always 3, never any other amount.
7 Β· The seal. Every file ends in a seal β a SHA-256 over the census Affine writes β and a second seal over its index of collisions. Anyone who downloads the IDE and runs Study 50 reads the same bytes from CERN and must arrive at the same seals. The answer does not depend on whose computer runs it.
Affine is an exact-integer substrate. In this study that means:
- The data is the detector's own. Nothing between you and CERN's bytes but integer arithmetic: no physics model, no CMS software, no ROOT library, no floating-point number anywhere on the data's path.
- Every byte is pinned. Each file is checked against the Adler-32 checksum CERN publishes, then sealed with SHA-256.
- Every refusal has a name. A hit at an address a chip does not have, a board whose length disagrees with itself, a record that does not close β each is counted and named, never smoothed away.
- Nothing is stored. The study streams each 2β3 GB file, keeps an index of where every collision's bytes live in CERN's file, and can fetch any one of them again with a single request.
- What the raw data does not carry, it let the data decide. Which way each pixel module faces is not in the raw bytes (CMS keeps it in its geometry description). Affine tried each possibility and kept the one under which the most tracks agree on a single collision point β checked on collisions it had not used to choose (165 β 402).
| file | collisions | pixel hits | clusters | census seal | index seal |
|---|---|---|---|---|---|
7AEB2818β¦ |
16,915 | 36,088,470 | 9,374,091 | fb007f85c22e10e6β¦ |
dacdd4a526f3187c⦠|
02FD18E5β¦ |
25,734 | 57,360,147 | 14,872,039 | 7f77ad5b24f2f18b⦠|
a9ea27fa73e25e4b⦠|
0E402934β¦ |
25,769 | 57,038,442 | 14,793,073 | 12e5f812b19965fe⦠|
40f20f5ca86dab1e⦠|
107A864F⦠|
26,809 | 60,589,436 | 15,780,658 | 4c9bbe2129857e34β¦ |
089577a5f1e7338d⦠|
00CD8D3F⦠|
25,515 | 56,033,401 | 14,519,333 | 275fe293447d70d7β¦ |
15271a1c9823efcd⦠|
Every event record in these five files joined its collision (120,742 of 120,742). The table grows as the study reads on.
The emerald paths show where charged particles went. Turning each path's bend into a momentum, and momenta and calorimeter energies into the masses of what was made, needs two more published ingredients: the detector's alignment and its calibration, both kept in CERN's conditions database for this run (global tag FT_53_LV5_AN1). That is the next tier of this study β and the same exact-integer law will carry it.
| claim | grade | where it is shown |
|---|---|---|
| every byte matches CERN's published checksum | MEASURED | the journal of each file; the Adler-32 check refuses otherwise |
| census and index seals reproduce on any machine | MEASURED (one machine so far) | the seals above; the law runs the same integers everywhere |
| 10,588,790 board readouts in file 1 (626 per collision), none broken; one ECAL CRC flag | MEASURED | file 1's census |
| first collision's event record: stamped 2011-03-21 03:09:05.121034 UTC, orbit 175,639,113, slot 2,705 | MEASURED | the event record, decoded byte by byte; also a self-test vector |
| record slot = pixel slot β 3 in 3,854 of 120,742 collisions, never another offset | MEASURED | the slot-offset count in each file's tally |
| 138 busy slots hold 93,110 of 120,742 collisions; 2,850 vs 60 pixel hits per collision | MEASURED | the per-slot counts in each file's tally |
| hits placed in CMS by the published cabling and layout | DERIVED (barrel to ~1 mm, endcap ~1 cm) | the geometry law; module orientation chosen on held-out collisions |
| tracks, and where along the beam each collision happened | DERIVED | aligned triplets; 402 of 1,815 held-out triplets meet within 3 mm of their collision's point |
| momenta and masses | NOT YET | needs alignment and calibration β the next tier |
- CMS Collaboration, CMS RAW data, /MinimumBias/Run2011A-v1/RAW, CERN Open Data Portal, record 35, DOI 10.7483/OPENDATA.CMS.I8HN.DF32 (CC0).
- CMSSW_5_3_X sources (Apache 2.0):
FEDNumbering.h,fed_header.h,fed_trailer.h,PixelDataFormatter.cc,FrameConversion.cc, and the pixel cabling tablepixelToLNK.ascii. - CMS Collaboration, The CMS experiment at the CERN LHC, JINST 3 S08004 (2008) β the detector's layout and dimensions.