# Validation Checks run in three levels: 1. **Kernels** (`cpp/tests`): random numbers (Philox known answers), tables, apertures, facet grid and polynomial surfaces against brute force, obscurers, sources, transmission, analytic optics, model loading and error messages. 2. **Compilation** (`python/tests`): parameter records to optical models: Davies-Cotton geometry, seeded mirror errors, focus offsets, camera body, dual-mirror surfaces, telescope structure, units, store and reload. 3. **Equivalence with sim_telarray** (below): the same simtools applications on the same simulation models, with both backends. ![Deviation of ray tracing from sim_telarray](../_static/figures/validation_psf_area.png) ## How to repeat the comparison In a simtools environment with sim_telarray and obdeect ray tracing installed, for each telescope and backend: ```sh simtools-validate-optics --site North --telescope LSTN-01 --model_version 6.0.2 \ --zenith_angle 20 --source_distance 10 --max_offset 4 --offset_step 1 \ --offset_directions N,E --ray_tracing_backend obdeect \ --instrument_seed 11 --simulation_seed 11 --output_path out/LSTN-01/obdeect ``` Repeat with `--ray_tracing_backend sim_telarray`, then compare the `results/ray_tracing_*.ecsv` tables (D80, effective area, effective focal length per offset). Incident angles: `simtools-derive-incident-angle` with both backends. Each backend draws 20 mirror-error realisations of 100,000 photons per offset (2 million photons), so mirror errors and photon statistics limit the agreement; other seeds show the spread. ## Results (simulation models 6.0.2, zenith 20 deg, star at 10 km, 0-4 deg) Ray tracing at commit 7394e73 (after the performance rounds), simtools at commit fcd73f6033 (`--instrument_seed 11 --simulation_seed 11`). Largest deviation from sim_telarray over nine offsets: | Telescope | D80 | Effective area | Effective focal length | | --- | --- | --- | --- | | LSTN-01 | 0.6% | 0.10% | 0.01% | | MSTN-01 (NectarCam) | 1.6% | 0.07% | 0.05% (realisation spread) | | MSTS-01 (FlashCam) | 1.6% | 0.07% | 0.05% (realisation spread) | | SSTS-01 | 0.14% | 0.28% | < 0.01% | | SCTS-01 (one-pixel camera) | 0.15% | 0.29% | < 0.01% | | LSTS-01 | 0.6% | 0.10% | 0.01% | | LSTN-design | 0.5% | 0.10% | 0.01% | | LSTS-design | 0.6% | 0.10% | 0.01% | | MSTx-FlashCam | 1.6% | 0.07% | 0.05% (realisation spread) | | MSTx-NectarCam | 1.6% | 0.07% | 0.05% (realisation spread) | | SSTS-design | 0.14% | 0.28% | < 0.01% | | SCTS-design (one-pixel camera) | 0.12% | 0.22% | < 0.01% | - Single-mirror D80 deviations come from the random mirror errors, which each backend draws itself. Another seed set (simtools option `--obdeect_seed`) gave 0.3-1.1%. Between the two seed sets, sim_telarray's D80 changes by up to 1.3% (MSTN-01) and ray tracing's by up to 1.0%. Telescopes of one type share optics and seeds, hence the same largest deviations. - Ray tracing 0.3.0 and 7394e73, same simtools and seeds: LSTN-01 and MSTS-01 identical (D80 and focal length to 1e-15); SSTS-01 and SCTS-design within 3e-9 (D80) and 8e-11 (focal length), the [rounding-level changes](#rounding-level-changes); effective areas identical. - SCT rows compare with sim_telarray photon lists of a one-pixel camera (the container's sim_telarray build allows at most 2368 pixels; scripts `val/sct_simtel.sh`, `val/sct_design_simtel.sh`). - Star at 10,000 km (LSTN-01, 0-2 deg): D80 within 0.15%, area within 0.13%. - Per mirror (MSTN-01, on axis, 86 mirrors): fractions of photons reflected by each mirror agree within statistics (pulls: mean 0.05, standard deviation 0.96), including mirrors partly shadowed by the camera. - Single-mirror mode (29 MSTN-01 mirrors): mean D80 1.503 cm for both; spread 0.183 cm (sim_telarray) and 0.160 cm (ray tracing). - Incident angles (SSTS-01, 1 deg, 10^6 photons, per photon): mean angles on the focal surface 46.656/46.658 deg, primary 9.899/9.898 deg, secondary 33.184/33.183 deg; in bins of the primary radius they agree within 0.05 deg. LSTN-01 focal surface: 15.81/15.80 deg. - Frames: image centroids of off-axis stars agree in sign and size (LSTN-01, 2 deg: -102.42 and -102.40 cm) without any frame conversion. ### PSF applications LSTN-01, measured PSF curves of the simtools test resources: | Application | sim_telarray | obdeect ray tracing | | --- | --- | --- | | `validate-cumulative-psf` (6.0.2, 20 deg, 7394e73) | D80 3.052 cm, area 365.7 m², 46 s | D80 3.050 cm, area 365.8 m², 3 s | | `derive-psf-parameters` (6.0.2, 20 deg) | RMSD 0.0131, D80 3.053 cm | RMSD 0.0127, D80 3.059 cm | | `derive-mirror-rnda` (7.0.0, test mode, 5 mirrors) | fitted D80 within 0.9-3.8% of measured, 117 s | within 0.8-4.9%, 19 s | | `derive-mirror-rnda` (7.0.0, full statistics) | first 20 mirrors: within 4.2% (median 1.7%), 848 s | all 198 mirrors: within 5% (median 2.4%), 241 s | The `derive-psf-parameters` and `derive-mirror-rnda` rows are from an earlier ray-tracing version and were not rerun; the PSF they use is checked above. - `derive-psf-parameters`: both backends start below the RMSD threshold, so the optimisation stops at once. Its D80-versus-offset plot (46 offsets in four directions, 2 million photons each) needed more than 2 GB with either backend, because simtools kept every photon image (fixed in simtools#2664). At full statistics: 2120 s with sim_telarray (RMSD 0.0132 to 0.0101 in one step, peak 0.3 GB) and 115 s with ray tracing (0.0127 to 0.0107, peak 1.45 GB). In `--test` mode the application takes 11 s (sim_telarray) and 6 s (ray tracing). - `derive-mirror-rnda`: with 5000 photons per evaluation the fits differ by their noise. With a fixed `--simulation_seed` every evaluation reuses the same photons, so the ray-tracing objective is smooth. On the 20 mirrors fitted by both, the fitted random reflection angle of ray tracing is 2.8% below sim_telarray's on average, with a spread of 6.6% between mirrors (fit noise). These sim_telarray fits used `--max_workers 1`: parallel mirror jobs shared one model directory and rewrote its files (fixed in simtools#2666). The comparison found and fixed one convention: sim_telarray measures the distance of a point source from the dish vertex, not from the telescope frame origin (0.02% plate-scale difference for the SCT at 10 km). ## Telescope structure (beyond sim_telarray) `telescope_transmission` is sim_telarray's fit of the ratio of the effective mirror area of a detailed ray tracing (with masts) to its own (camera body only), versus off-axis angle. The same ratio from ray tracing, with the same photons traced with and without the structure (geometry only, mean of four azimuths), for MSTx-FlashCam (`telescope_structure` 1.0.0 on simulation-models branch `simulation-model-masts`): | Off axis | 0 deg | 1 deg | 2 deg | 3 deg | 4 deg | | --- | --- | --- | --- | --- | --- | | ray tracing, structure | 0.8945 | 0.8917 | 0.8897 | 0.8866 | 0.8855 | | `telescope_transmission` 2.0.0 | 0.8980 | 0.8967 | 0.8939 | 0.8898 | 0.8846 | `simtools-validate-optics` for MSTS-01 (production 7.0.0, both backends) agrees in effective area within 0.4% on axis and 1.2% up to 4 deg. Ray tracing shows the azimuthal dependence (up to 2% between N and E) that the radial parameterisation cannot. The structure also widens D80 by 1.4% on axis (same photons with and without structure), since it shades the dish unevenly. ## Secondary baffles `simtools-validate-optics` (simulation models 7.0.0, zenith 20 deg, star at 10 km, offsets towards N, seeds 11), with both backends, once as modelled and once with the baffle removed (`secondary_mirror_baffle` R1 = 0; sim_telarray keyword `secondary_baffle`). The container's sim_telarray build allows at most 2368 pixels, so its SCT runs use a one-pixel camera. Scripts: `val/baffle/run_baffle.sh` and `with_changes.py` (obdeect-validation). ![Effective area with and without the baffle](../_static/figures/baffle_area.png) Effective area with the baffle divided by that without it: | Off axis | SSTS-design sim_telarray | ray tracing | SCTS-design sim_telarray | ray tracing | | --- | --- | --- | --- | --- | | 0 deg | 0.9983 | 0.9982 | 0.9898 | 0.9903 | | 1 deg | 0.9974 | 0.9974 | 0.9807 | 0.9808 | | 2 deg | 0.9970 | 0.9969 | 0.9553 | 0.9567 | | 3 deg | 0.9968 | 0.9967 | 0.9465 | 0.9471 | | 4 deg | 0.9965 | 0.9966 | 0.9450 | 0.9447 | Both codes lose the same light on the baffles: within 0.01% for the SST and 0.14% for the SCT, where the baffle costs up to 5.5%. The SST baffle costs little because most light meeting it would also meet the secondary shadow plate (see [shadowing](shadowing.md)). Effective areas agree within 0.36% (ray tracing higher, with and without baffle). ## Calibration light (Phase 2) `simtools-simulate-flasher --run_mode full_simulation` and `simtools-simulate-illuminator`, both backends, simulation models 7.0.0: - sim_telarray backend: `ff-1m` or `xyzls` make the light; sim_telarray traces it through its optics and simulates the camera. - Ray-tracing backend: ray tracing makes and traces the light to the camera entrance; simtools writes the photons as photon bunches; sim_telarray simulates the camera with optics bypassed (`Bypass_Optics=1` single mirror, `Bypass_Optics=2` dual mirror). Each case runs with both backends and two seed sets (`--instrument_seed 11` and `12`, `--simulation_seed 111` and `112`). The two seed sets of one backend give the run-to-run scatter each comparison is judged against. Compared: photoelectrons per event, image centre and width, time spread of the photoelectrons, and per pixel the summed images scaled to the same total, (b - a) / sqrt(a + b), whose spread is 1 for Poisson fluctuations only ("shape pull"). Flashers: 100 pulses; illuminators: 10 pulses of 10^9 photons. | Case | p.e./event sim_telarray | p.e./event ray tracing | ray tracing / sim_telarray | centre shift [mm] | width ratio | time rms [ns] sim_telarray / ray tracing | shape pull, cross / same backend | wall time [s] sim_telarray / ray tracing | | --- | --- | --- | --- | --- | --- | --- | --- | --- | | Flasher LSFN-design, LSTN-01 | 665228 | 665272 | 1.0001 ± 0.0001 | 0.1 | 1.0000 | 0.60 / 0.60 | 1.01 / 12.2 | 512 / 397 | | Flasher LSFS-design, LSTS-01 | 669651 | 666603 | 0.9954 ± 0.0001 (a) | 0.0 | 1.0001 | 0.60 / 0.60 | 1.00 / 12.3 | 508 / 385 | | Flasher MSFx-FlashCam, MSTS-04 | 38923 | 38914 | 0.9998 ± 0.0005 | 0.2 | 0.9999 | 0.72 / 0.72 | 0.99 / 1.93 | 76 / 59 | | Flasher MSFx-NectarCam, MSTN-07 | 18639 | 18598 | 0.9978 ± 0.0007 (a) | 0.2 | 1.0007 | 1.07 / 1.07 | 1.00 / 1.45 | 43 / 35 | | ILLN-01 to MSTN-07 (415 m, elevation 10 deg) | 7911 | 7884 | 0.9967 ± 0.0036 | 0.4 | 1.0046 | 0.73 / 0.75 | 7.8 / 7.8 (b) | 135 / 5 | | ILLN-02 to LSTN-01 (797 m) | 5706 | 5668 | 0.9934 ± 0.0035 | 1.9 | 1.0002 | 0.08 / 0.08 | 1.39 / 1.66 | 15 / 6 | | LSTN-01, source 2.0 deg off axis at 600 m (camera rotated) | 5838 | 5826 | 0.9981 ± 0.0040 | 1.6 | 1.0004 | 0.10 / 0.10 | 1.42 / 1.68 | 21 / 6 | | ILLS-03 to MSTS-07 (638 m, elevation 3.6 deg) | 3340 | 3330 | 0.9969 ± 0.0054 | 0.7 | 1.026 | 1.35 / 0.71 (c) | 7.8 / 7.7 (b) | 67 / 6 | | ILLS-03 to LSTS-04 (929 m, elevation 2.3 deg) | 3750 | 3773 | 1.0062 ± 0.0062 | 8.7 (c) | 1.0016 | 10.5 / 0.07 (c) | 1.51 / 1.63 | 16 / 5 | | ILLS-03 to SSTS-17 (79 m), 300 nm | 6903 | 12051 | 1.7458 ± 0.0048 (d) | 0.2 | 1.0002 | 0.08 / 0.01 (c) | 1.03 / 1.33 | 658 / 10 | | ILLS-03 to SSTS-17, 400 nm | 22702 | 24840 | 1.0942 ± 0.0018 (d) | 0.1 | 1.0004 | 0.08 / 0.01 (c) | 0.99 / 1.54 | 647 / 9 | | SSTS-17, source 1.7 deg off axis at 400 m | 251 | 452 | 1.80 ± 0.03 (d) | 0.1 | 1.005 | 0.02 / 0.00 | 0.91 / 0.84 | 504 / 6 | - Flashers and pixels: with the same seeds, ray tracing and sim_telarray agree per pixel at the Poisson level (shape pull 0.97-1.01) in all four flasher cases. Other seeds change the pixel efficiencies of the camera simulation, which dominates the run-to-run scatter (1.4-12). - (a) The camera is at `focal_length + focus_offset` from the dish. simtools placed `ff-1m` at `focal_length`, which gave 0.3-0.5% more light (predicted (2800 / 2806.55)^2 = 0.9953 for the LSTs, measured 0.9954). simtools now places the flasher at the camera (simtools#2667); the LSTN-01 and MSTS-04 rows are rerun with it and agree within 0.02%. - (b) Random mirror errors: each realisation shapes the defocused image of a source a few hundred metres away. Ray tracing and sim_telarray draw different realisations, as do two seeds of one backend, so all pairs differ by the same 7-8 times Poisson. - (c) `xyzls` hands the light over at the horizontal plane through the telescope's fiducial sphere and applies CORSIKA's refraction and delay model, made for light from above. For nearly horizontal light the bunch times deviate from the straight flight from the source: by 0.02 ns rms at 10 deg elevation, 1.6 ns rms (up to -4.5 ns) at 3.6 deg and 11 ns rms (up to -35 ns) at 2.3 deg, for bunches handed over 100-1000 m in front of the telescope. sim_telarray's photoelectron times inherit this; at 2.3 deg also its image centre (8.7 mm, 64 arcsec; refraction over 929 m bends the light by about 3 arcsec) and its readout charge. Ray tracing's times follow the geometry; the SST optics are isochronous to 0.01 ns. - (d) sim_telarray reflectivity for 2-D (wavelength, incidence angle) tables: it applies R(lambda, 0) R(lambda, theta) / R_max(lambda) instead of R(lambda, theta) (see [Optics](optics.md)). The SST mirrors have such tables. Predicted ratios 1.7449 (300 nm) and 1.0943 (400 nm); measured 1.7458 +- 0.0048 and 1.0942 +- 0.0018. Image centres and shapes agree. For Cherenkov light this bug makes sim_telarray's SSTs about 9% too dim at 400 nm. - Photon bunch file: the bunches of an `ff-1m` file, written again by simtools, give identical sim_telarray output (photoelectrons of every pixel). The scripts are in the validation workspace (`val/phase2`: `cases.sh`, `run_case.sh`, `compare.py`, `summarize.py`, `xyzls_times.py`). ## Air-shower photons (Phase 3) CORSIKA + sim_telarray against CORSIKA + ray tracing + sim_telarray, simulation models 7.0.0, 1 TeV gammas, 50 showers with 10 core reuses per point. Both chains read the same CORSIKA IACT file. In the ray-tracing chain the simtools shower route traces the photon bunches to the camera, and sim_telarray simulates the camera with optics bypassed (`Bypass_Optics=1`). Native sim_telarray runs use three seeds; their mean is the reference. | Telescope | Zenith [deg] | p.e. sim_telarray | p.e. ray tracing | ray tracing / sim_telarray | | --- | --- | --- | --- | --- | | LSTS-01 | 20 | 1678606 | 1677207 | 0.9992 | | LSTS-01 | 40 | 1196284 | 1197360 | 1.0009 | | MSTS-04 | 40 | 406894 | 406417 | 0.9988 | | LSTN-01 | 40 | 1061542 | 1062029 | 1.0005 | | MSTN-07 | 40 | 264967 | 265766 | 1.0030 | | LSTS-01 | 60 | 390242 | 390761 | 1.0013 | | MSTS-04 | 60 | 134511 | 135149 | 1.0047 | | LSTN-01 | 60 | 325837 | 325495 | 0.9990 | | MSTN-07 | 60 | 95095 | 95157 | 1.0007 | - All points agree within 0.5% and within 1.5 sigma (native seed spread plus Poisson). - Atmospheric transmission is split between the two codes. With bypassed optics, sim_telarray takes the airmass from the camera-frame direction of each bunch (near the telescope axis), not from its sky direction. Without a correction this gave 7% too many photoelectrons at 40 deg and 33-39% at 60 deg. The route therefore scales each photon weight by the ratio of the native transmission to the one sim_telarray applies, using sim_telarray's own formula. Ray tracing returns the telescope transmission in its weights (`normalise_transmission=False`), so the total equals the native chain. A sim_telarray option to switch off its absorption with bypassed optics would remove this workaround. - Images agree as well as two native seeds agree with each other: pixel correlation within 0.004 of the seed-to-seed value; image centre and width within 0.9 mm (seed-to-seed spread 3-12 mm). - Photoelectron arrival times: the per-event mean time difference is within 0.06 ns; the Kolmogorov-Smirnov p value is at least 0.002 everywhere. The scripts and full results are in the validation workspace (`phase3`: `results/RESULTS.md`, `results/RESULTS-R19.md`, `results/RESULTS-R19-pixels.md`). ## Single-core speed Both codes pinned to one core of the validation container (aarch64 Linux VM, Apple silicon host; ray-tracing Release build), simulation models 7.0.0, the same telescope models and photon numbers. Per-photon costs are the slope between 2*10^5 and 2*10^6 photons (best of three runs), which removes start-up: sim_telarray 0.01-0.09 s; ray-tracing model compilation 0.17-0.28 s. Ray tracing at commit 7394e73; best of two such runs (LSTN-01: four) on a host that was partly busy, which also slowed sim_telarray by up to 7% against earlier runs. Star light (zenith 20 deg, star on axis at 10 km, photons sampled on the entrance disk), microseconds per photon: | Telescope | sim_telarray as simtools runs it (imaging list) | sim_telarray, no imaging list | ray tracing sampling + tracing | ray tracing tracing only | | --- | --- | --- | --- | --- | | LSTN-01 | 1.86 | 0.33 | 0.34 | 0.26 | | MSTN-07 | 1.68 | 0.31 | 0.31 | 0.24 | | MSTS-04 (with `telescope_structure`) | 1.69 | 0.31 | 0.44 | 0.36 | | SSTS-17 | 1.29 | 0.30 | 0.58 | 0.50 | - Per core, ray tracing samples and traces single-mirror telescopes as fast as sim_telarray without its imaging list. Applications gain from writing no photon lists (sim_telarray spends 80% of its time on them) and from using every core. The earlier factors of 6-18 (Phase 1, whole applications, wall time) combine both; they are not per-core speed. Dual-mirror and structure cases are still slower per core than sim_telarray, which traces no structure and loses a photon that meets a gap between segments. - The telescope structure (203 elements) is searched through a grid of the structure in the horizontal plane for steep rays and a bounding-box hierarchy for the others; its cost fell from 0.79 to 0.39 us per photon. - Aspheres (SST, SCT): a ray that crosses the surface at most once ends its search at the first root or when the surface is missed; the segment under a hit point is found from a grid; a Newton step evaluates the surface once (SSTS-17: 1.28 to 0.61 us per photon). Facet mirrors use tighter grid cells and a plane test before the sphere test (LSTN-01: 0.33 to 0.25 us). All results are bit-identical to the earlier ones. - Aspheres evaluated in r^2 and a better first Newton guess (rounding-level changes, below): SSTS-17 tracing 0.61 to 0.50 us per photon. Single-mirror telescopes are unchanged (interleaved runs of LSTN-01 against the previous build agree within the 5% timing noise). - Segment lookup: a cell of the segment grid that lies wholly inside one segment, or in none, answers without testing a segment (proven with a margin of 1e-9 m over the cell and the mirror's height range there; other points take the full test). 88% of SSTS-17 lookups are answered so, 27 to 10 ns per lookup; tracing 287 to 279 ns per photon (Apple M3, one thread). Results are bit-identical (replay of 6*10^6 photons per model). A plain loop over the few obscurers of a leg instead of the bounding-box hierarchy gained 1% and was not kept. Calibration light, the stage before sim_telarray's camera simulation (the same for both): | Case | LightEmission | ray tracing making and tracing | writing photon bunches | ratio | | --- | --- | --- | --- | --- | | Flasher MSFx-FlashCam on MSTS-04, 10^8 photons, 4.0*10^7 at the camera | `ff-1m` 58.6 s | 23.8 s | 22.1 s | 1.3 | | Illuminator 600 m above LSTN-01, 10^10 photons, 8.3*10^5 at the camera | `xyzls` 14.7 s | 0.8 s | 0.5 s | 11.8 | `xyzls` follows every emitted photon; ray tracing makes only the photons crossing the telescope entrance (see [Light sources](light-sources.md#flashers-and-illuminators)), so its cost scales with those. Scripts: `val/phase2/bench_star.py`, `bench_light.py`. ## Rounding-level changes These changes alter results in the last bits of floating-point arithmetic. Each was measured against the state at the start of the round (commit 13daac1, reference worktree `ref3`) with a replay of 6 * 10^6 photons per model, the same seeds and fixed tier G limits: at most 5 flips per 10^6 photons (a flip is a photon whose status or any surface id changes); per field, |difference| at most 1e-8 m for position and path, 1e-9 for direction, 1e-7 ns for time and 1e-9 (relative) for weight. - **Aspheres evaluated in r^2** (SST, SCT): the sag and slope polynomials take r^2 instead of r. SSTS-17: max position difference 5.8e-15 m, 0 flips. SCTS-design: max position 1.0e-10 m, 0 flips. - **Order of the first guesses for aspheres**: when the surface can have only one intersection with the ray (the single-root rule), Newton starts from the paraboloid guess inside the aperture interval instead of the one clamped to its edge. A fitted quadric as first guess gave no further gain. Secondary Newton steps for SSTS-17 fell from 4.54 to 2.95 per photon. SSTS-17: max position 2.8e-10 m, direction 8.2e-11, 0 flips. SCTS-design: max position 1.1e-10 m. - **Estrin evaluation of the polynomials**: no speed gain, reverted. - **Facet mirror scatter in closed form**: no speed gain, reverted. The check found that on curved facets the facet tangent vectors are not exactly perpendicular to the hit normal (up to 1.9e-2 rad). This property is not changed by the round. - **Telescope transmission**: the power of the outer term is skipped when it is 1; a 4096-interval table of the transmission was tried. The outer power is 1 in all models, and the table fails the 1e-8 validation for every model with a transmission (max relative table error 1.9e-7 MSTN-07, 6.7e-6 SCTS-design, 6.5e-4 SSTS-17), so no model uses a table. The replay is bit-identical and the change gave no speed gain: reverted. - **`OBDEECT_RAY_TRACING_USE_FMA`** (CMake option, default OFF): see [Installation](../user-guide/installation.md#from-source). Total change of the round (final state against 13daac1, tier G, 6 * 10^6 photons per model): | Model | Flips | Max position (m) | Max direction | Max path (m) | Max time (ns) | Max weight (rel.) | | --- | --- | --- | --- | --- | --- | --- | | LSTN-01 | 0 | 0 | 0 | 0 | 0 | 0 | | MSTN-07 | 0 | 0 | 0 | 0 | 0 | 0 | | MSTS-04 | 0 | 0 | 0 | 0 | 0 | 0 | | SSTS-17 | 0 | 2.8e-10 | 8.2e-11 | 5.5e-10 | 1.8e-9 | 2.3e-12 | | SCTS-design | 0 | 1.1e-10 | 1.1e-11 | 3.4e-10 | 1.1e-9 | 0 | Optics check (container `obdeect-val`, simtools with model version 6.0.2, seeds 11 and 11): optics validation of SSTS-01 and MSTS-01 and the cumulative PSF of LSTN-01, once with the reference build and once with the final build. Largest relative difference of D80, effective area and centroid per offset: SSTS-01 3.7e-9 (centroid), MSTS-01 0, LSTN-01 0. The gate of 1e-4 passes. The effective focal length is not deterministic in simtools and is not gated (SSTS-01: 7.9e-11).