High-Precision Ground Characterization of Test-Mass Magnetic Properties for the Taiji Gravitational Wave Mission via a Physics-Informed Neural Framework Preprint: APS/123-QED

Chang Liu, Qiong Deng, Huadong Li, Liwei Yang, Xiaodong Peng, Ziren Luo, Yuzhu Zhang, Chen Gao, Xiaotong Wei, Minghui Du, Zihao Xiao, Peng Xu, Bo Liang, Zhi Wang, Li-e Qiang · 2026

Read High-Precision Ground Characterization of Test-Mass Magnetic Properties for the Taiji Gravitational Wave Mission via a Physics-Informed Neural Framework Preprint: APS/123-QED by Chang Liu, Qiong Deng, Huadong Li, Liwei Yang, Xiaodong Peng, Ziren Luo, Yuzhu Zhang, Chen Gao, Xiaotong Wei, Minghui Du, Zihao Xiao, Peng Xu, Bo Liang, Zhi Wang, Li-e Qiang (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2604-23510.txt.

Introduction

Liu and colleagues measure the magnetic traits of test masses for Taiji, a planned space mission to sense gravitational waves, with a physics-guided neural method. Their system pairs a neural noise gauge with an exact physics solver, so it can spot and drop tainted test data. Checked on real test-stand noise, it meets Taiji's ground-test accuracy needs on every magnetic trait at once.

Catalog record

CREATORChang Liu, Qiong Deng, Huadong Li, Liwei Yang, Xiaodong Peng, Ziren Luo, Yuzhu Zhang, Chen Gao, Xiaotong Wei, Minghui Du, Zihao Xiao, Peng Xu, Bo Liang, Zhi Wang, Li-e Qiang
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE60339
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2604.23510

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

High-Precision Ground Characterization of Test-Mass Magnetic Properties for the Taiji Gravitational Wave Mission via a Physics-Informed Neural Framework Preprint: APS/123-QED Chang Liu, Qiong Deng, Huadong Li, Liwei Yang, Xiaodong Peng, Ziren Luo, Yuzhu Zhang, Chen Gao, Xiaotong Wei, Minghui Du, Zihao Xiao, Peng Xu, Bo Liang, Zhi Wang, Li-e Qiang Abstract Taiji is a gravitational wave detection mission in space initiated by the Chinese Academy of Sciences, which will open the millihertz window through a heliocentric triangular constellation of three drag-free spacecraft. Its ultimate sensitivity is determined partly by the residual acceleration noise of the gravitational reference sensors (GRS), within which the coupling between the test-mass and the fluctuating environmental magnetic field constitutes one of the key stray-force contributions. Following the path established by the LISA and TianQin teams, high-precision ground characterization of remanent magnetic moment m → r \vec{m}_{r} and volume susceptibility χ \chi of the test masses is a central step in the Taiji pre-launch test program. A persistent challenge for this characterization is the non-stationary, colored background noise inherent to torsion-pendulum facilities, which systematically biases classical Ordinary Least Squares (OLS) and Kalman filter (KF) estimators. We propose an AI-enhanced Differentiable Weighted Least Squares (AI-WLS) framework that fuses a dilated one-dimensional residual network, acting as a dynamic noise evaluator, with a fully differentiable analytical physical solver. This architecture preserves the exact linear mapping from the magnetic parameters to the torque response while autonomously identifying and suppressing contaminated data segments. Validated on real measured noise from the Changchun Institute of Optics, Fine Mechanics and Physics torsion-pendulum facility developed for Taiji, which achieves a torque sensitivity of order 10 − 13 N ⋅ m Hz − 1 / 2 10^{-13}\,\mathrm{N\cdot m\,Hz^{-1/2}} , the AI-WLS framework bounds the maximum absolute estimation errors at 4.46 × 10 − 10 ​ A ⋅ m 2 4.46\times 10^{-10}\,\mathrm{A\cdot m^{2}} for m → r \vec{m}_{r} and 7.8 × 10 − 8 7.8\times 10^{-8} for χ \chi , satisfying Taiji’s ground-test requirements on all these parameters simultaneously, whereas OLS and KF both fail under the realistic non-stationary background. Beyond magnetic characterization, the framework provides a general methodology for ground-based verification of stray-force coupling models of ultra-precision GRSs, supporting the upcoming Taiji pre-launch noise-model test campaign and readily transferable to similar precision-measurement tasks for LISA-like space-borne gravitational wave missions. I Introduction Since the first direct detection of a binary black hole coalescence, GW150914 1 , the LIGO-Virgo-KAGRA (LVK) network has catalogued hundreds of compact binary mergers 2 ; 3 , opening the era of gravitational-wave (GW) astronomy and providing precision tests of general relativity in the strong-field regime. These ground-based interferometers are, however, restricted to the 10 – 10^{3} Hz band due to seismic and control-loop noise, leaving the scientifically rich millihertz window ( 10^{-4} to 1 Hz) accessible only from space. This window hosts massive black hole binary coalescences, extreme mass-ratio inspirals, Galactic double white dwarf binaries, and cosmological stochastic backgrounds 20 ; 19 ; 21 ; 18 ; 9 , motivating an international effort to develop space-borne GW antennas, including the European-led LISA mission 17 ; 6 and the Chinese programs of Taiji and TianQin 15 ; 24 ; 26 . Initiated by the Chinese Academy of Sciences (CAS), the Taiji mission 15 ; 25 is one of the Chinese contributions to this effort and will deploy a heliocentric equilateral-triangular constellation of three drag-free spacecraft with 3\times 10^{6} km arm length, targeting the 0.1\,\mathrm{mHz} - 1\,\mathrm{Hz} band. Incident GWs induce picometer level relative displacements between the free-falling TMs, which are read out by inter-satellite laser interferometry 11 . At the core of these space-borne LISA-like antennas is the Gravitational Reference Sensor (GRS), which ensures that the TMs follow near-perfect geodesic trajectories. The ultimate sensitivity of the detector is governed by the residual acceleration noise of the GRS. For Taiji, the residual acceleration noise of the free-falling TM must be below 3\times 10^{-15}\,\mathrm{m\,s^{-2}\,Hz^{-1/2}} across the science band. Achieving such extreme inertial stability requires the mitigation of a long list of stray-force contributions, including Brownian gas damping, actuator and capacitive cross-couplings, electrostatic patch potentials, radiometer and radiation-pressure effects associated with thermal gradients, and residual charge accumulation from cosmic rays. Among these, the interaction between the TM’s magnetic properties and the fluctuating environmental magnetic field is an important contribution, which induces stray force as 8 \vec{F}=\iiint_{V}\left\{(\vec{m}_{r}\cdot\nabla)\vec{B}+\frac{\chi}{\mu_{0}}\left[(\vec{B}\cdot\nabla)\vec{B}\right]\right\}dx\,dy\,dz, (1) where V is the TM volume, \vec{m}_{r} its remanent magnetization, \chi its volume magnetic susceptibility, \mu_{0} the vacuum permeability, and \vec{B} the local magnetic induction. The significance of these effects has been demonstrated in missions such as the CHAMP mission, where the magnetic-torquer activation and ambient-field coupling to the STAR accelerometer were identified as non-negligible systematics in accelerometer mode, which had to be explicitly modeled and subtracted during Level-1B data processing 27 ; 13 . Consequently, the rigorous verification of magnetic coupling models and the high-precision characterization of \vec{m}_{r} and \chi are paramount. Such measurements are not only critical for the hardware R&D of the GRS to ensure material compliance but are also indispensable for post-processing data analysis, where accurate physical models allow for the subtraction of magnetic noise from the science signal. Motivated by these requirements, both the LISA and TianQin teams have pursued systematic ground-based characterization of TM magnetic properties using high-precision torsion pendulums 30 ; 7 ; 16 ; 10 ; 5 . The LISA specifications 4 require |\vec{m}_{r}|<2\times 10^{-8}\,\mathrm{A\cdot m^{2}} and \chi<3\times 10^{-6} , while TianQin adopts even tighter constraints |\vec{m}_{r}|<10^{-8}\,\mathrm{A\cdot m^{2}} and \chi<10^{-6} 24 ; 29 . In-flight measurements by LISA Pathfinder have demonstrated the practical feasibility of fabricating TMs at this level of magnetic cleanliness, reporting |\vec{m}_{r}|<2.5\times 10^{-10}\,\mathrm{A\cdot m^{2}} and \chi<3.4\times 10^{-5} 8 . Given Taiji’s comparable magnetic budget, the pre-launch verification of its TMs targets resolutions of 1\times 10^{-9}\,\mathrm{A\cdot m^{2}} for |\vec{m}_{r}| and 1\times 10^{-7} for \chi , ensuring the ground-test precision exceeds mission specifications by at least one order of magnitude to effectively mitigate the risk of measurement errors. To support the Taiji GRS R&D at this level, the Changchun Institute of Optics, Fine Mechanics and Physics (CIOMP) of the CAS has developed a dedicated weak-force torsion-pendulum facility with a torque sensitivity of order 10^{-13}\,\mathrm{N\cdot m\,Hz^{-1/2}} . Jointly with the Institute of Mechanics and the National Space Science Center of CAS, the CIOMP team is now positioned to carry out the systematic verification of GRS noise models for Taiji. At this stage, the CIOMP team has completed the commissioning of the torsion-pendulum facility and collected long-duration free-run data. These records exhibit the expected colored, non-stationary character of the experimental background, dominated by low-frequency drifts and transient glitches, which violates the stationary-Gaussian noise assumption underpinning both the OLS 22 and KF 23 estimators and can introduce systematic biases, especially given the strong non-linear cross-coupling between \vec{m}_{r} and \chi . To push the measurement precision of the TM magnetic parameters toward the level required by Taiji under such realistic conditions, we propose in this work an end-to-end AI-enhanced Differentiable Weighted Least Squares (AI-WLS) framework that couples a deep dilated residual network with a fully differentiable analytical WLS solver. The network acts as a dynamic, data-driven noise evaluator, assigning time-varying confidence weights to the measured torque sequence so that contaminated segments are automatically down-weighted, while the

Read the complete text