Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

On the computational feasibility of Bayesian end-to-end analysis of LiteBIRD simulations within Cosmoglobe

R. Aurvik, M. Galloway, E. Gjerløw, U. Fuskeland, A. Basyrov, M. Bortolami, M. Brilenkov, P. Campeti, H. K. Eriksen, L. T. Hergt, D. Herman, M. Monelli, L. Pagano, G. Puglisi, N. Raffuzzi, N. -O. Stutzer, R. M. Sullivan, H. Thommesen, D. J. Watts, I. K. Wehus, D. Adak, E. Allys, A. Anand, J. Aumont, C. Baccigalupi, M. Ballardini, A. J. Banday, R. B. Barreiro, N. Bartolo, S. Basak, M. Bersanelli, A. Besnard, T. Brinckmann, E. Calabrese, E. Carinos, F. J. Casas, K. Cheung, M. Citran, L. Clermont, F. Columbro, G. Coppi, A. Coppolecchia, P. Dal Bo, P. de Bernardis, E. de la Hoz, M. De Lucia, S. Della Torre, P. Diego-Palazuelos, T. Essinger-Hileman, C. Franceschet, G. Galloni, M. Gerbino, M. Gervasi, R. T. Génova-Santos, T. Ghigna, S. Giardiello, C. Gimeno-Amo, A. Gruppuso, M. Hazumi, S. Henrot-Versillé, K. Kohri, L. Lamagna, T. Lari, M. Lattanzi, C. Leloup, F. Levrier, A. I. Lonappan, M. López-Caniego, G. Luzzi, J. Macias-Perez, B. Maffei, E. Martínez-González, S. Masi, S. Matarrese, T. Matsumura, S. Micheli, L. Montier, G. Morgante, L. Mousset, R. Nagata, A. Novelli, I. Obata, A. Occhiuzzi, A. Paiella, D. Paoletti, G. Pascual-Cisneros, F. Piacentini, M. Pinchera, G. Polenta, L. Porcelli, M. Remazeilles, A. Ritacco, A. Rizzieri, M. Ruiz-Granda, J. Sanghavi, V. Sauvage, M. Shiraishi, S. L. Stever, Y. Takase, K. Tassis, L. Terenzi, M. Tomasi, M. Tristram, L. Vacher, B. van Tent, P. Vielva, G. Weymann-Despres, E. J. Wollack, M. Zannoni, Y. Zhou

TL;DR

The paper investigates the computational feasibility of performing an end-to-end Bayesian analysis of LiteBIRD data within the Cosmoglobe/Commander3 framework by analyzing simulated TOD for a detector subset. It extrapolates from a one-year, reduced-data set to the full three-year mission, estimating data volumes ($ ext{uncompressed}$ ≈ $238$ TB; $ ext{compressed}$ ≈ $70$ TB) and a per-Gibbs-sample cost of about $3{,}000$ CPU-hours for the full dataset. Using an ideal instrument model with only $1/f$ noise, the study demonstrates that such an analysis is within the capabilities of current and near-future HPC resources, though wall times would be lengthy (years). The work provides concrete scalability estimates, identifies the dominant computational steps (correlated-noise sampling and TOD processing), and outlines steps to incorporate additional systematics (e.g., half-wave plate non-idealities, beam/non-idealities) and to develop a massively parallel Commander4 for practical production runs.

Abstract

We assess the computational feasibility of end-to-end Bayesian analysis of the JAXA-led LiteBIRD experiment by analysing simulated time ordered data (TOD) for a subset of detectors through the Cosmoglobe and Commander3 framework. The data volume for the simulated TOD is 1.55 TB, or 470 GB after Huffman compression. From this we estimate a total data volume of 238 TB for the full three year mission, or 70 TB after Huffman compression. We further estimate the running time for one Gibbs sample, from TOD to cosmological parameters, to be approximately 3000 CPU hours. The current simulations are based on an ideal instrument model, only including correlated 1/f noise. Future work will consider realistic systematics with full end-to-end error propagation. We conclude that these requirements are well within capabilities of future high-performance computing systems.

On the computational feasibility of Bayesian end-to-end analysis of LiteBIRD simulations within Cosmoglobe

TL;DR

The paper investigates the computational feasibility of performing an end-to-end Bayesian analysis of LiteBIRD data within the Cosmoglobe/Commander3 framework by analyzing simulated TOD for a detector subset. It extrapolates from a one-year, reduced-data set to the full three-year mission, estimating data volumes ( ≈ TB; ≈ TB) and a per-Gibbs-sample cost of about CPU-hours for the full dataset. Using an ideal instrument model with only noise, the study demonstrates that such an analysis is within the capabilities of current and near-future HPC resources, though wall times would be lengthy (years). The work provides concrete scalability estimates, identifies the dominant computational steps (correlated-noise sampling and TOD processing), and outlines steps to incorporate additional systematics (e.g., half-wave plate non-idealities, beam/non-idealities) and to develop a massively parallel Commander4 for practical production runs.

Abstract

We assess the computational feasibility of end-to-end Bayesian analysis of the JAXA-led LiteBIRD experiment by analysing simulated time ordered data (TOD) for a subset of detectors through the Cosmoglobe and Commander3 framework. The data volume for the simulated TOD is 1.55 TB, or 470 GB after Huffman compression. From this we estimate a total data volume of 238 TB for the full three year mission, or 70 TB after Huffman compression. We further estimate the running time for one Gibbs sample, from TOD to cosmological parameters, to be approximately 3000 CPU hours. The current simulations are based on an ideal instrument model, only including correlated 1/f noise. Future work will consider realistic systematics with full end-to-end error propagation. We conclude that these requirements are well within capabilities of future high-performance computing systems.

Paper Structure

This paper contains 15 sections, 6 equations, 4 figures, 2 tables.

Table of Contents

  1. Introduction
  2. The LiteBIRD experiment
  3. Algorithm
  4. Data model
  5. Bayesian analysis and Gibbs sampling
  6. Simulations
  7. Galactic foregrounds
  8. Extra Galactic sources
  9. LiteBIRD Instrument Model and TOD simulation pipeline
  10. Final data selection and compression
  11. Results
  12. Time-ordered data
  13. Frequency maps
  14. Computational costs
  15. Summary and conclusions

Figures (4)

  • Figure 1: Time ordered data for 1 hour of samples for the 40GHz channel. The plot includes the total signal (CMB, noise and the foreground emissions) for the simulated TOD (blue). Also plotted is the total signal with the scaled down noise levels that compensate for reducing the number of detectors from 48 to 4 (orange). Units are in millikelvin.
  • Figure 2: Time ordered data for the same three hours worth of samples for one detector for the LFT 40GHz channel (top), the MFT 119GHz channel (middle) and the HFT 402GHz channel (bottom). The plots include the CMB (orange), white and correlated noise (blue) and the foregrounds (green). The peaks correspond to scanning the Galaxy plane, where synchrotron dominates the lowest frequency, the middle frequency has the lowest level of foregrounds, and the highest frequency channel foregrounds are dominated by thermal dust.
  • Figure 3: Time ordered data for 40min for the 40GHz channel (blue), the MFT 119GHz channel (orange), and the 402GHz channel (green) including CMB, foregrounds, white and correlated noise, and the dipole. The dipole signal causes the large sinusoidal pattern and the other components are responsible for the thickness of the lines. The spikes of the 402GHz channel are foreground signals. The dipole in the LFT channel has the opposite sign compared to the MFT and HFT, which is caused by the telescope pointing in nearly opposite sky direction.
  • Figure 4: Frequency maps of a single Gibbs sample for three frequency bands for 40GHz (top), 119GHz (middle) and 402GHz (bottom), represented by Stokes $I$ (left column), $Q$ (middle) and $U$ (right column) parameters. The lowest frequency can be seen to be dominated by synchrotron while the highest one is dominated by thermal dust. The middle frequency has the lowest sky signal. All units are in microkelvin.