Gravity Probe B data analysis: III. Estimation tools and analysis results

Abstract

This paper provides detailed descriptions of the numerical estimation algorithms used to fit physics-based models to the data from the Gravity Probe B spacecraft, as well as the scientific results of the experiment, and the statistical and systematic uncertainties. The first paper in this series of three data analysis papers derives the mathematical expressions for the signals to be analyzed, and the second paper deals with science data acquisition and their preparation for the relativistic drift rate estimation. The data from each of the four gyroscopes are partitioned into six segments, each spanning several weeks to several months. These segments are first analyzed individually to check the validity of the mathematical models and the accuracy of the estimation routine by examining the consistency of the relativistic drift rate estimates from each of these 24 gyro-segments. Then, the drift rate estimates and uncertainties are calculated for each individual gyroscope and for the four gyroscopes combined. These results are presented and compared with each other and with the prediction of general relativity.

Gravity Probe B data analysis: II. Science data and their handling prior to the final analysis

“…Moreover, a very complex HF SQUID signal analysis (TFM) had to be developed and carried out, as explained in [2,9], here in section 5.4, and in [3], section 2. Nevertheless, the knowledge and skills gained during preflight preparations for data treatment were very valuable for this post-flight work.…”

“…Let us explain the goals (1)- (3). Preliminary data cleaning (1) was mostly outlier removal; typically, data points more than 3 standard deviations away from the average were considered outliers and removed from the data set.…”

“…This integration procedure provided orbital data at scattered moments of time, set by accuracy requirements (the computational error was taken to be not greater than 1 m in position, 10 m s 3 1 − − in velocity). To make the data available for any given time instant, a third-order spline approximation was made.…”

“…These L2 data were also successfully used at the initial stages of the relativistic drift estimation using the 2-floor estimator (section 5.4). However, the scale and complexity of the influence of patch effect torques and polhode damping on the science signal made evident the need to create the 2-second filter [3] and use it to get a satisfactory estimate of the relativistic effects. It also became clear that the quality of the SQUID and pointing signal preparation had to be better than that in the L2 DB.…”

“…We obtained the results of the GP-B experiment, i.e., the estimates of the relativistic drift rates, using the elaborate and sophisticated ' 2-sec filter' described in detail in our paper [3]. The success of this approach depended strongly on careful and meticulous preparation of all major input science signals, i.e., LF SQUID signal, telescope signal, roll phase signal, etc The goal was a serious improvement in the quality of the science data relative to the quality of the L2 data set delivered by preprocessing.…”