21cmlstm: A Fast Memory-based Emulator of the Global 21 cm Signal with Unprecedented Accuracy
Year: 2024
Type: article
Abstract: Neural network (NN) emulators of the global 21 cm signal need an emulation error much less than the observational noise in order to be used to perform unbiased Bayesian parameter inference. To this end, we introduce 21cmLSTM—a long short-term memory (LSTM) NN emulator of the global 21 cm signal that leverages the intrinsic correlation between frequency channels to achieve exceptional accuracy compared to previous emulators, which are all feedforward, fully connected NNs. LSTM NNs are a type of recurrent NN designed to capture long-term dependencies in sequential data. When trained and tested on the same simulated set of global 21 cm signals as the best previous emulators, 21cmLSTM has an average relative rms error of 0.22%—equivalently 0.39 mK—and comparably fast evaluation time. We perform seven-dimensional Bayesian parameter estimation analyses using 21cmLSTM to fit global 21 cm signal mock data with different adopted observational noise levels, σ 21. The posterior 1σ rms error is ≈three times less than σ 21 for each fit and consistently decreases for tighter noise levels, showing that 21cmLSTM can sufficiently exploit even very optimistic measurements of the global 21 cm signal. We have made the emulator, code, and data sets publicly available so that 21cmLSTM can be independently tested and used to retrain and constrain other 21 cm models.
Source: The Astrophysical Journal
Language: English
Subfield: Astronomy and Astrophysics
Field: Physics and Astronomy
Domain: Physical Sciences
Open Access status: gold
Estimated APC paid: $4,499
Funders: National Aeronautics and Space Administration, Oak Ridge Associated Universities, California Institute of Technology, Nuclear Safety and Security Commission, Jet Propulsion Laboratory, Marshall Space Flight Center, University of Colorado Boulder, Solar System Exploration Research Virtual Institute
Awards: G1005378446, MULTIPLE YEAR AWARD WITH INCREMENTAL FUNDING FOR YEARS ONE TO FIVE. THIS PROPOSAL IS ENTITLED "NETWORK FOR EXPLORATION AND SPACE SCIENCE NESS." THIS IS AWARDED IN RESPONSE TO COOPERATIVE AGREEMENT NOTICE CAN NO. NNH16ZDA009C NASA SOLAR SYSTEM EXPLORATION RESEARCH VIRTUAL INSTITUTE SSERVI CYCLE THREE. THE PRINCIPAL INVESTIGATOR IS DR. JACK O. BURNS WITH THE UNIVERSITY OF COLORADO UCO. PROJECT SUMMARY: NESS WILL IMPLEMENT CROSS-DISCIPLINARY PARTNERSHIPS TO ADVANCE SCIENTIFIC DISCOVERY AND HUMAN EXPLORATION OF TARGET DESTINATIONS AND WILL CONDUCT RESEARCH IN ASTROPHYSICS AND HELIOPHYSICS THAT IS UNIQUELY ENABLED BY HUMAN AND ROBOTIC EXPLORATION OF TARGET BODIES. OUR RESEARCH IS DERIVED FROM THREE DECADAL SURVEYS AND FROM THE NASA ASTROPHYSICS AND GLOBAL EXPLORATION ROADMAPS. NESS RESEARCH INCLUDES: TELEROBOTIC OPERATION OF SURFACE SCIENCE EXPLORATION ASSETS AT THE MOON AND MARS THE NATURE OF THE MOON'S CORE AN IMPROVED SELENODETIC COORDINATE SYSTEM PRECISION TESTS OF GRAVITATION ENERGETIC SOLAR PARTICLE EVENTS IN INTERPLANETARY SPACE AND EXTRASOLAR SYSTEMS DUST ENVIRONMENTS OF SPACE AND TARGET BODIES DETECTION OF THE UNIVERSE'S FIRST GALAXIES AND HABITABLE EXOPLANETS. FOUR KEY PROJECTS WILL BE PROPOSED. EXPLORATION SCIENCE NESS WILL MATURE A CRITICAL TECHNOLOGY OF SCIENCE AND EXPLORATION MISSIONS FROM THE ORBITS OF TARGET BODIES: SURFACE TELEROBOTICS. WE WILL DEVELOP HUMAN-ROBOTIC SCIENCE ASSOCIATED WITH ORION/HABITAT MISSIONS IN CIS-LUNAR ORBITS. OUR TEAM WILL BUILD TOOLS FOR THE CREW TO TELEOPERATE ROVERS ON THE LUNAR FARSIDE THAT WILL FEED-FORWARD TO THE FIRST CREWED MARS MISSIONS. WE WILL INVESTIGATE THE SYSTEMS NEEDED ON ORION/HABITAT TO SUPPORT TELEROBOTICS ANALYZE LUNAR LOW FREQUENCY RADIO TELESCOPE DEPLOYMENT ALONG WITH A LUNAR SAMPLE RETURN AND DEVELOP CREW TRAINING STRATEGIES. LLR CURRENT ALTERNATE THEORIES OF GRAVITY PREDICT DEVIATIONS FROM GENERAL RELATIVITY GR AT A LEVEL THAT IS WITHIN THE GRASP OF THE NEXT GENERATION OF LUNAR LASER RANGING LLR INSTRUMENTS. NESS WILL IMPROVE BY FACTORS GREATER THAN 100 THE SIGNAL PHOTOELECTRON ACCURACY OF LLR TO SET NEW CONSTRAINTS ON GR AND THE LUNAR CORE. WE WILL ALSO MEASURE/MODEL DUST ON LLR OPTICAL SURFACES AND SIGNIFICANTLY IMPROVE THE SELENODETIC COORDINATE SYSTEM FOR EXPLORATION AND SCIENCE MISSIONS. HELIOPHYSICS AND SPACE PHYSICS_NESS HELIOPHYSICS RESEARCH DIRECTLY IMPACTS HUMAN EXPLORATION VIA OUR STUDIES OF SOLAR PARTICLE RADIATION AND INTERPLANETARY DUST. WE WILL ADVANCE THE DESIGN FOR A LUNAR-BASED LOW FREQUENCY LESS THAN 10 MHZ RADIO IMAGER UNIQUELY CAPABLE OF IMAGING THE TRAJECTORY OF PARTICLE ACCELERATION IN SOLAR ERUPTIONS AND MEASURING INTERPLANETARY NANODUST. FROM COSMIC DAWN TO HABITABLE PLANETS NESS ADDRESSES KEY QUESTIONS FROM THE ASTROPHYSICS DECADAL SURVEY: "WHEN AND HOW DID THE FIRST GALAXIES FORM?" AND "CAN WE FIND ANOTHER PLANET LIKE EARTH ORBITING A NEARBY STAR?" OUR TEAM WILL USE NEW MODELING TOOLS AND TECHNOLOGIES TO INVESTIGATE THE FORMATION OF THE FIRST GALAXIES USING THE NEUTRAL HYDROGEN 21-CM SIGNAL RED-SHIFTED TO 2-3 M. THE MOON'S FARSIDE IS UNIQUELY SHIELDED FROM RADIO FREQUENCY INTERFERENCE AND FREE FROM IONOSPHERIC EFFECTS TO ENABLE SUCH OBSERVATIONS. A FARSIDE OBSERVATORY IS IMPORTANT FOR EXPLORATION TO INVESTIGATE THE ENVIRONMENTAL AND TECHNICAL CHALLENGES TO DEPLOYING/OPERATING A LARGE-AREA SCIENTIFIC STATION ON THE MOON. WE WILL ADVANCE 21-CM COSMOLOGY THEORY THE DESIGN OF A LOW RADIO FREQUENCY ARRAY WIDE-BAND RECEIVERS AND NEW CALIBRATION TECHNIQUES. NESS ALSO WILL ADVANCE THE STUDY OF THE CHARACTERISTICS OF THE HABITABLE EXOPLANETS BY EMPLOYING POLARIMETRY ON LARGE UV_OPTICAL TELESCOPES CONSTRUCTED IN CIS-LUNAR SPACE AND BY USING A FARSIDE RADIO ARRAY TO DETECT MAGNETOSPHERES FOR EARTH-LIKE PLANETS., G2676722403, G303888369, G3648211085, G540462668, G7002947457, G830101876, G8626377472
OpenAlex ID: W4404868853