LiteBIRD: Official Simulation Framework

• The official LiteBIRD simulations are mission-scale synthetic data products generated by the modular Python-based LiteBIRD Simulation Framework (LBS) to model the data-acquisition chain of three flight instruments.
• The framework paper details LBS’s architecture, ensuring reproducibility, accurate metadata tracking, and high-performance computation for end-to-end simulation studies.
• The first official release, using a reduced subset of detectors and one year of observations, lays the groundwork for advanced Bayesian inference and systematic error analyses in CMB polarization research.

Official LiteBIRD simulations are the collaboration’s mission-scale synthetic data products and the software infrastructure used to generate them for instrument design, validation, end-to-end data-production studies, and downstream scientific analysis. In the published LiteBIRD literature, this official simulation effort is centered on the LiteBIRD Simulation Framework (LBS), a Python package that models the data acquisition chain of the three flight instruments—LFT, MFT, and HFT—and on the first official end-to-end release built with that framework. The simulations are tied to LiteBIRD’s core objective of measuring primordial CMB $B$-mode polarization from the Sun–Earth L2 point, with a target uncertainty on the tensor-to-scalar ratio of $\delta r \sim 0.001$ (Tomasi et al., 7 Jul 2025, Bortolami et al., 8 Jul 2025).

1. Official scope and mission role

The official LiteBIRD simulation program has two distinct but closely connected components. The first is the framework paper, which introduces and documents LBS itself as the collaboration’s official simulation software framework. The second is the first release paper, which presents the first official end-to-end simulation products generated with a new pipeline developed using that framework (Tomasi et al., 7 Jul 2025, Bortolami et al., 8 Jul 2025).

This distinction is operationally important. The framework paper is not the mission-performance paper; it defines the reusable simulation machinery, its architecture, and the guarantees of accuracy, reproducibility, and provenance tracking. The first-release paper is not a complete mission simulation; it is explicitly described as a pathfinder release using one year of observing time and approximately one-third of the nominal focal plane (Tomasi et al., 7 Jul 2025, Bortolami et al., 8 Jul 2025).

A common misconception is that the official simulations are equivalent to a low-level hardware model of the spacecraft. The framework paper is explicit that the simulation is focused on scientific samples and timelines rather than low-level hardware physics: it does not attempt to simulate mirror or strut thermal mechanics, or detailed optical hardware responses, but instead the essential data-acquisition chain that turns sky radiation into time-ordered detector outputs. A second misconception is that LiteBIRD’s official simulations are a single monolithic pipeline. LBS is designed to be modular, so that nominal end-to-end runs, HWP systematics studies, and planet-observation studies can all be assembled from common blocks (Tomasi et al., 7 Jul 2025).

2. LiteBIRD Simulation Framework

LBS is organized as a Python package with a central Simulation object that coordinates modules and keeps track of metadata, inputs, and outputs. A typical run loads instrument information from the LiteBIRD Instrument Model database, or IMo; constructs Observation objects containing time-ordered data matrices; attaches detector attributes; computes pointings; injects sky signals and noise; performs map-making; and writes outputs to disk (Tomasi et al., 7 Jul 2025).

The framework models all three LiteBIRD instruments—Low-Frequency Telescope, Mid-Frequency Telescope, and High-Frequency Telescope—and is intended to simulate their outputs over the nominal three-year mission. Its sky model is map-based. CMB and foreground maps can be generated through the Mbs module via PySM3 and then scanned into detector timelines. The supported signal components explicitly include the CMB itself, the Solar dipole, the Kinematic dipole, and diffuse Galactic foregrounds such as synchrotron and free-free emission. Instrumental noise includes white noise and correlated $1/f$ noise, and the framework also supports systematic effects such as pointing wobbling and Half-Wave Plate effects (Tomasi et al., 7 Jul 2025).

The spacecraft motion is modeled through the nominal survey strategy: spinning around the spin axis, precessing around the Sun–Earth direction, and orbiting in the Sun–Earth L2 environment. Pointings are generated from quaternions sampled at lower cadence and interpolated to detector sample times, allowing the telescope sky direction and orientation to be computed for each detector sample. The time-domain data model is correspondingly explicit. In the framework paper, the detector noise power spectrum is written as

$$P(f) = \sigma^2 \left(1 + \left(\frac{f_k}{f + f_\text{min}}\right)^\alpha\right),$$

where $\sigma^2$ is the white-noise power, $f_k$ is the knee frequency, $\alpha$ is the spectral slope, and $f_\text{min}$ regularizes the $f=0$ singularity. The dipole signal is modeled from the relative velocity of the Solar System with respect to the CMB through

$$T(\vec\beta, \hat n) = \frac{T_0}{\gamma \bigl(1 - \vec\beta \cdot \hat n\bigr)},$$

with $\delta r \sim 0.001$0 the CMB rest-frame temperature, $\delta r \sim 0.001$1, and $\delta r \sim 0.001$2 the line-of-sight direction (Tomasi et al., 7 Jul 2025).

Version $\delta r \sim 0.001$3 models the HWP as ideal, but the framework is already designed to accommodate more realistic non-ideal HWPs and beam-related systematics. That forward-compatible design is part of the reason LBS functions არა only as a production code for a fixed release, but as the collaboration’s general simulation substrate (Tomasi et al., 7 Jul 2025).

3. End-to-end pipeline and first official release

The first official LiteBIRD release is an end-to-end simulation pipeline constructed on top of LBS and intended to emulate the LiteBIRD observation process as realistically as possible. In the release paper, the baseline mission configuration is summarized as 4508 detectors sampling at 19.1 Hz, surveying the full sky for three years across 15 frequency bands from 34 to 448 GHz, with an effective polarization sensitivity of $\delta r \sim 0.001$4 and an angular resolution of 31 arcmin at 140 GHz. The scan strategy uses a boresight offset of $\delta r \sim 0.001$5, a precession angle of $\delta r \sim 0.001$6, and a spin rate of 0.05 rpm, yielding full-sky coverage in about six months (Bortolami et al., 8 Jul 2025).

The end-to-end pipeline comprises loading the instrument model and scan strategy, computing detector pointing and polarization angles, scanning beam-convolved sky maps into time-ordered data, adding white noise and $\delta r \sim 0.001$7 noise, adding the CMB dipole for the first realization, binning the time stream back into sky maps with a simple map-maker, and producing noise covariance matrices and validation products. The input sky includes CMB realizations generated from Planck 2018 cosmological parameters, Galactic foregrounds from PySM/MBS, extragalactic components from WebSky, radio source catalogs, and beam convolution using the official LiteBIRD beam specifications. The foreground model includes dust, synchrotron, anomalous microwave emission, free-free, CO lines, tSZ, kSZ, CIB, lensing convergence, and radio sources (Bortolami et al., 8 Jul 2025).

The first release is intentionally reduced relative to the nominal mission. It uses one year of observing time and 1678 detectors selected to preserve focal-plane symmetry; channels with at most 48 detectors are simulated in full. The delivered products are 500 full-sky simulated maps at HEALPix $\delta r \sim 0.001$8, together with one year of time-ordered data for roughly one-third of LiteBIRD’s detectors, saved only for the first simulation realization. The TOD set includes CMB TOD, foreground TOD, noise-only TODs, dipole TOD, pointing information, and noise covariance matrices, and occupies about 35 TB of storage (Bortolami et al., 8 Jul 2025).

The release also establishes specific noise scenarios. Two $\delta r \sim 0.001$9 cases are simulated: 30 mHz as the realistic case and 100 mHz as the pessimistic case. A prominent validation result is that the ideal HWP strongly suppresses $1/f$0 noise in polarization, so that polarization maps remain close to white-noise behavior even when temperature shows low-frequency excess. The paper is explicit that this first release is a baseline, reproducible collaboration product and a foundation for later releases with the full focal plane, the full three-year mission, and more realistic systematics such as gain drifts, downtime, cosmic rays, beam systematics, and HWP systematics (Bortolami et al., 8 Jul 2025).

4. Data model, parallelism, and reproducibility

One of the defining characteristics of the official LiteBIRD simulation program is the degree to which software architecture is tied to data volume, parallel execution, and provenance. In LBS, a TOD is stored as a two-dimensional matrix with one detector per row and time samples as columns. Large simulations rely on MPI decomposition by detector, by time, or by both. The user specifies this through n_blocks_det and n_blocks_time, and the layout can be changed after allocation with Observation.set_n_blocks(). Pointing information is stored in a parallel structure with the same distribution scheme, and the framework can retain multiple TOD-like matrices simultaneously so that signal components such as sky signal and noise remain separated throughout the simulation (Tomasi et al., 7 Jul 2025).

Provenance is handled through IMo. Inputs are versioned and accessed by path, and LBS records the database requests made during a run so that outputs can be traced back to the exact instrument description used. This is paired with explicit run reporting. For every execution, the framework generates an automatic Markdown or HTML report containing the run date, LBS version, Git commit hash, and even git diff when there are uncommitted changes. Parameter files are encouraged in TOML format and are copied into the output directory (Tomasi et al., 7 Jul 2025).

The reproducibility model is correspondingly strict. The random seed must be set explicitly unless nondeterministic behavior is chosen by setting random_seed=None. In parallel runs, each MPI process receives a deterministic independent RNG stream derived from the shared seed. The framework states that reproducibility holds when the same script is run with the same seed and the same number of MPI processes. Validation spans unit tests, integration tests, MPI consistency tests, and end-to-end tests, and the code contains many assertions and explicit error messages intended to catch inconsistent input or misuse (Tomasi et al., 7 Jul 2025).

These choices make the official simulations not merely a collection of synthetic maps, but a controlled computational environment. For collaboration-scale cosmology, that distinction is substantial: the output products are intended to be reproducible artifacts with recoverable configuration state, rather than ad hoc mock data (Tomasi et al., 7 Jul 2025).

5. Downstream scientific use and computational scaling

The official LiteBIRD simulations are already being used as inputs to downstream Bayesian analysis. A concrete example is the feasibility study of end-to-end Bayesian analysis within Cosmoglobe and Commander3, which ingests official LiteBIRD TOD generated by the collaboration’s simulation pipeline and evaluates whether future global Gibbs sampling from TOD to cosmological parameters is computationally viable (Aurvik et al., 7 Jul 2025).

In that study, the simulated TOD are produced using LBS and IMo, with separate streams written to disk for tod_cmb, tod_fg, tod_dip, tod_wn, tod_wn_1f_30mHz, and tod_wn_1f_100mHz. The input sky includes CMB, Galactic synchrotron, thermal dust, free-free, AME, CO emission, CIB, tSZ, kSZ, radio sources, and lensing convergence, using the PySM d1s1a1f1co1 model for Galactic foregrounds and WebSky-based extragalactic inputs. The working analysis uses a reduced subset of 88 detectors and one year of observations, with noise per detector scaled down to preserve effective band sensitivity (Aurvik et al., 7 Jul 2025).

The computational results are among the clearest quantitative measures of what official LiteBIRD simulations imply at full mission scale. For the reduced one-year, 88-detector set, the uncompressed TOD volume is 1.55 TB, reduced to 470 GB after Huffman compression, with a total memory requirement of 700 GB. Extrapolated to the full three-year mission with 4508 detectors, the study estimates 238 TB uncompressed and 70 TB compressed, the latter described as a conservative upper bound because the current data are stored as floating-point numbers rather than ADC-quantized integers. The estimated runtime is approximately 25.4 CPU hours per Gibbs sample for the reduced case and about 3000 CPU hours per Gibbs sample for the full mission (Aurvik et al., 7 Jul 2025).

The same paper is careful to state that this remains an idealized feasibility study. The current simulations assume perfect gain, perfectly known bandpasses, perfect Gaussian beams, no beam asymmetries, and no ADC quantization effects, with only white noise and correlated $1/f$1 noise included in the analysis. Future work is explicitly reserved for more realistic systematics and full end-to-end error propagation. Even so, the study concludes that the official LiteBIRD simulations are already suitable for large-scale Bayesian time-domain inference on future HPC systems (Aurvik et al., 7 Jul 2025).

6. Related simulation programs, specialized studies, and design feedback

The official LiteBIRD simulations sit within a broader mission-specific simulation ecosystem. Some of these studies use LBS or litebird_sim directly, while others use adjacent software such as TOAST, Falcons.jl, or dedicated optical and cosmic-ray simulators. Together they define much of the systematics and design context into which the official framework is being extended.

A scan-optimization study treats the LiteBIRD scanning strategy as a multi-parameter optimization problem in $1/f$2 space and introduces Falcons.jl, a fast spacecraft scanning simulator. That work recommends a configuration essentially centered on $1/f$3, while also emphasizing that a continuously rotating HWP suppresses many systematic effects but does not eliminate everything (Takase et al., 2024). Instrument-specific optical simulations supply another input line. For MHFT, system-level simulations are intended to produce polarized sky beams, sidelobe pickup patterns, detector-position-dependent beam response, and inputs for more realistic TOD sky simulations (Lamagna et al., 2021). For LFT, the modified crossed Dragone baseline is validated with optical simulations yielding Strehl ratios $1/f$4 at 161 GHz across the field, Mueller QU/UQ cross-polarization response $1/f$5 dB at 34 GHz, and direct and diffuse triple-reflection sidelobes $1/f$6 dB when realistic absorber reflectivity is included analytically (Matsuda et al., 21 May 2025).

Systematics studies increasingly interface directly with the official software. A study of HWP differential optical load and TES nonlinearity uses the official litebird_sim framework and derives the requirement $1/f$7 for MHFT channels under the allocated systematic budget $1/f$8 (Micheli et al., 2024). A reflective-HWP misalignment study uses LBS to inject a wedge-like effect into TOD and finds that the contamination is lensing-like rather than primordial-tensor-like, with a baseline two-detector tolerance $1/f$9 arcmin (Stellati et al., 3 Sep 2025). Gain-calibration requirements derived from LiteBIRD simulated maps and blind component separation produce per-channel tolerances from about $$P(f) = \sigma^2 \left(1 + \left(\frac{f_k}{f + f_\text{min}}\right)^\alpha\right),$$0 to $$P(f) = \sigma^2 \left(1 + \left(\frac{f_k}{f + f_\text{min}}\right)^\alpha\right),$$1, with a factor $$P(f) = \sigma^2 \left(1 + \left(\frac{f_k}{f + f_\text{min}}\right)^\alpha\right),$$2 tightening required in the most pessimistic high-complexity sky and realistic MC-NILC case (Carralot et al., 2024). Earlier TOAST-based LiteBIRD-like simulations showed that template fitting can reduce calibration and beam-induced leakage by about two orders of magnitude at the power-spectrum level (Puglisi et al., 2021).

Telemetry and radiation effects have also been studied through simulation. On-board compression simulations of TES bolometer TOD showed that the mission’s required lossless compression ratio of about 0.50 is achievable, with differentiation-based compression or cosine-fitting-based compression plus Rice coding reaching about 12 bits per sample after processing (Tominaga et al., 2022). Cosmic-ray simulations for LiteBIRD wafers and TES detectors, using mission scan strategy and map-making, found initial $$P(f) = \sigma^2 \left(1 + \left(\frac{f_k}{f + f_\text{min}}\right)^\alpha\right),$$3-mode contamination estimates at the level of $$P(f) = \sigma^2 \left(1 + \left(\frac{f_k}{f + f_\text{min}}\right)^\alpha\right),$$4 in a one-year observation with 12 detectors under a specific differential-mode noise assumption (Tominaga et al., 2021), while a separate end-to-end simulator for LFT subsets reported CR-induced NEP of order $$P(f) = \sigma^2 \left(1 + \left(\frac{f_k}{f + f_\text{min}}\right)^\alpha\right),$$5 and flat $$P(f) = \sigma^2 \left(1 + \left(\frac{f_k}{f + f_\text{min}}\right)^\alpha\right),$$6-mode contamination that scales down with detector count and can be mitigated by favorable thermal coupling and deglitching (Stever et al., 2021).

This broader literature suggests a layered simulation strategy rather than a single code path. The official framework and release establish the common collaboration baseline; specialized studies then interrogate specific elements—scanning, optics, compression, cosmic rays, gain, HWP behavior, and Bayesian analysis—that can either feed into later official releases or delimit the assumptions under which those releases are interpreted (Tomasi et al., 7 Jul 2025, Bortolami et al., 8 Jul 2025).