Chima McGruder - Software Products

THIS PAGE IS STILL UNDER DEVELOPMENT

what should be added for each software package

software goals and capabilities:

where published:

Other collaborators:

Primary Developer

1) Simulated Annealing for detecting volcanism on Io

This package was developed at the Space Telescope Science Institute (STScI) to study the potential of using a non-redundant mask (NRM) on the James Webb Space Telescope (JWST) to observe volcanoes on Jupiter’s moon, Io. In particular, I used our physical understanding of the moon’s tidal heating, our mechanical understanding of JWST’s instrument Near InfraRed Imager and Slitless Spectrograph (NIRISS), and Fourier optics to create synthetic data of JWST images of Io. Once these synthetic images were created, I tested how accurately we’d be able to obtain the position of volcanic eruptions, the flux of those eruptions, and the surface brightness of Io itself. The algorithm used optimization algorithms such as simulated annealing and gradient descent, and MCMC to explore the parameter space. The bulk of my coding was with Python but included IDL. Anand Sivaramakrishnan, Alexandra Greenbaum, and Deepashri Thatte helped me on this project. This work was included in the publication Observing Outer Planet Satellites (Except Titan) with the James Webb Space Telescope.

2) Plotting and Calculation codes for Orbital dynamics

The packages developed here were designed to help understand the dynamics of a binary star system. The main data analyzed were Doppler Shift data and photometric observations.

Doppler shift observations measure how fast a star is moving radially from the observer. Measuring this for a given system over time, and incorporating physics allow for the dynamical properties of the system to be determined. For this, we incorporated a two-dimensional cross-correlation algorithm to reduce the data.

Monitoring how the light flux of a binary system periodically changes over time (photometric monitoring) provides more dynamical insights into the system. This package used MCMC to explore the parameter space of feasible system characteristics based on the Doppler and photometric data. Through this analysis, the most accurate and precise measurements of the system were obtained. These parameters were then compared against physical models to determine the most likely evolutionary track this system underwent. Finally, we developed packages to analyze historical data dating back to 1961 to obtain a more accurate understanding of the apsidal motion and ephemeris of the system.

My advisor, Guillermo Torres, assisted in the package development and this work was published in The Astrophysical Journal in 2017.

Astrophysics and PhD context:

During my PhD, most of my work was to understand exoplanet atmospheres. To do so, our main method of observing was transmission spectroscopy. This is when we spectroscopically observe the light of a star hosting a planet as that planet passes in front of the host star. We then compare the spectroscopy of the star right before and right after the planet passes in front of the star with the spectroscopy while the planet is in front of the star. That comparison allows us to deduce properties of the upper atmosphere of the planet, based on which wavelengths are attenuated by the planetary atmosphere.

3) SyntheticSpectrum

I developed a Python routine to create synthetic transmission spectral data of planetary atmospheres. This data included white-light noise (shot noise), chromatic (wavelength-dependent), and non-chromatic systematic noise. The systematic noise was created by randomly drawing from a Gaussian Process (GP) distribution, where the GP was constructed to be correlated to auxiliary parameters measured in a typical observation (e.g. airmass, fwhm, sky flux, and telescope pointing). 500 of these synthetic data were then tested against two detrending techniques: PCA+GP (see Yan et. al. 2020) and CMC+Polynomial regression (see # 7 or TransSpecDetrend). These tests were directly used in my McGruder et. al. 2022 publication to determine the most optimal detrending techniques for specific data sets. I was the sole author of this package, which can be found on GitHub.

4) Second-order light correction

A spectrograph differentiates all the wavelengths of white light via diffraction. That diffraction occurs multiple times, each time with reduced intensity, called 'orders'. To observe exoplanet atmospheres, we observe minute changes in the host stars' stellar spectra due to the exoplanets’ atmosphere. Specifically, we used the Inamori Magellan Areal Camera and Spectrograph (IMACS) mounted on the Baade Magellan telescope to observe the stellar spectra.

For our observations, we only used the first-order diffracted light and initially thought the higher-order light was diffracted off our detectors. We soon discovered that this was not the case and we needed to use a light-blocking filter to prevent second-order light contamination. However, to use data that was obtained before we included the blocking filter, I had to model out the light contamination. To do this, I created a script that modeled the light contamination. This script (see on GitHub):

1) First uses iSpec to model the normalized continuum of the spectra observed. This is needed so the general spectroscopic light curve can be obtained without spectroscopic features found in specific observations. So things like changing spectroscopic lines don’t change the correction.

2) Then, the normalized continuum of the older unfiltered spectra is convolved with the throughput profile of the filter that was used on newer observations. This produces the spectral profile of the observation if the filter had been added, but with any second-order light still present.

3) Given the only difference between the newer observations *with* the filter and the older observation convolved with the filter (step 2) is the second-order light, I used the ratio of the two continua as a correction continuum for the second-order light.

I then applied this correction to all data observed without the blocking filter. The results were a part of the McGruder et. al. 2022 publication. I worked on this script alone, and it can be found on GitHub.

https://upload.wikimedia.org/wikipedia/commons/0/08/Magellan-Telescopes-at-LCO-2014-04-19.jpg

5) Out of Transit Stellar Contamination Correction

When conducting transmission spectroscopy for atmospheric detection, there are many sources of contamination, including the Earth's atmosphere and instrumental systematics, which have been alluded to in software products 4 and 5. Another big source of uncertainty is the host star's variability. This is a problem because when doing transmission spectroscopy, there has to be the assumption, at some level, that the host star is relatively constant. There are many ways to address this issue, such as adding host star variability in the models of transmission spectral data or obtaining supporting data such as photometric and or high energy monitoring data of the host star. However, in the models, there is a degeneracy between the observed data being caused by the planet or the host star, and the monitoring data requires longer-term monitoring and a coordinated effort from multiple observatories to be useful.

Thus, I built a package that tests the effectiveness of using the standard spectroscopic data taken before and after the planet passes in front of the host star. The spectroscopic data before and after the planet transits is called "out-of-transit data." Understanding how well we can use this out-of-transit (OOT) data helps us determine the best observing, data-detrending, and atmospheric modeling strategies.

To thoroughly understand the limits of the OOT data, I built a grid of synthetic stellar spectra at a variety of resolutions and stellar properties (surface temperature, mass, radius, stellar composition). I built this grid by downloading high-resolution stellar spectra models developed by T.-O. Husser et. al. (see paper and library links), called PHOENIX models. Though the PHOENIX models had a wide range of stellar spectra with different stellar properties, I still need to fill in some of the missing parameter space because the PHOENIX models were too sparse. To do this, I would interpolate the stellar spectra based on the spectra of models with parameters near my desired spectra (see ModelAtmosphere.py). Next, I had to add the effects of stellar variability, which is caused by passing hot and cold spots over the stellar surface. This can be modeled by replacing a small piece of the star with a different stellar spectrum. For passing cold spots, I would add a stellar spectrum that has the same composition but a cooler temperature. Conversely, I added a hotter stellar spectrum covering a fraction of the stellar surface to simulate passing hot spots. How I applied this can be seen in ModelAtmosphere.py, CombinedStellAtmo(). I also wanted to consider how well we can use the OOT light with different instruments and observing conditions. Thus, I tested the effects of various SNR levels, specific instrumental throughputs, tellurics (for ground-based observations), and injected systematics such as shot noise and cosmic rays (see MakeSynSpec.py). I also used iSpec (see AnalyzeDat.py) to degrade the resolution from the high-resolution PHOENIX models to that expected from different telescopes such as JWST, HST, and some ground-based telescopes. Finally, to determine how effective the OOT data can be used based on the specific observing conditions, telescopes used, targets observed, etc. I tested a chi^2, simulated annealing, and two nested sampling methods to observe how well the stellar parameters (including variability) were obtained.

The results were a part of the Rackham et. al. 2023 publication, and I worked on this script alone. Most of the specific packages pertaining to this project can be found on GitHub, titled KnowThyStar. A significant portion of this work was done on the Smithsonian High-Performance Computer Cluster (Hydra) because of the heavy usage of data and a large number of parallel processing done to conduct all of the tests quickly and efficiently. Thus, the GitHub version is incomplete, as I no longer have access to Hydra.

Background: astrobiology.com - "Characterisation Of Stellar Activity Of M dwarfs. I. Long-timescale variability In A Large Sample And Detection Of New Cycles"

6) Bootstrap test

During my PhD, I applied for HST time to observe the atmosphere of a few specific planets, which I identified to be an ideal test bed for cloud and aerosol formation rates. See McGruder et. al. 2023a for further explanation about this planetary sample. I used a Bootstrap test to support my claim that even with a few planets, we can still identify significant trends. I did this by creating synthetic data assuming a linear, quadratic, and sigmoidal correlation of cloud formation and metallicity. Then, I used those functional dependencies on metallicity to bootstrap 10000 instances of 6 planets with a range of metallicities dictated by the literature values, but a variance in the metallicity generated from a Gaussian. The standard deviation of the randomly drawn metallicity was given by the expected precision found in our observation. Next, I fit a line to each of the iterations and recorded their probability values (p-values). Doing this, I found that correlations can easily be found, even if the inherent correlation function wasn't linear, and with random noise on the measurements. I built this software product on my own, and it can be found on GitHub.

7) Modeling_RRLyrae_Signals (github: same name)

As part of the work in McGruder et. al. 2023a , I needed to more precisely determine the stellar and planetary properties of the planets in my sample. Most of the analysis for this was done with Juliet and CERES (see packages 6 and 7 of Collaborative work). However, the HATS-29b system was a special case. For all my systems, I would use data from the Transiting Exoplanet Survey Satellite (TESS) to more precisely determine the orbital parameters of the planets.

Given that the TESS observations have a large field of view, a lot of other stars can be in the data of the star one is interested in. Normally, this is not a problem because TESS detects planets by detecting variations in light due to a planet passing in front of its host star. Since, stars' variation is marginal compared to the signal caused by a transiting planet, the overall effect of background stars doesn't hinder the measurements of the passing planet. However, for HATS-29b, there was an RR Lyrae star in the background. RR Lyrae stars have strong periodic variations of light. The variation of the RR Lyrae star had a frequency and amplitude much higher than that of the planet transit. Thus, the features of the planet transit could not be characterized until the RR Lyrae background signal was modeled out. This code does just that.

I first had to isolate the RR-Lyrae signal by removing the signal of the transiting planet, using predicted transit times for the paper that discovered HATS-29b. Then, I applied a Lomb-Scargle periodogram analysis to this out-of-transit data. I used astropy's LombScargle function to produce the periodogram and found an RR-Lyrae period of 0.631 days. Next, I phase folded the data to the 0.631 d period, binned the data by 100 points, and smoothed the binned data with a scipy signal.savgol_filter. This was my model of the RR-Lyrae variation. I then subtracted this RR-Lyrae light curve model from the TESS data, and reduced the corrected TESS data with the same procedures as the other TESS observations (see packages 6 and 7 of Collaborative work). I also checked that I modeled out the RR-Lyrae signal correctly, by comparing the best-fit transit with the corrected TESS data against a fit with the discovery data (which had no background star contamination). Both observations agree. This code for this analysis can be found on GitHub, and I was the only contributor to the package.

8) CMC correction (github: TransSpecDetrend)

9) AI_Material_Properties (github: MaterialsWork)

10) Additional visualization codes (basically most of the plots in my papers)

Collaborating Developer or Modifications to Public Packages

1) TEPSPEC (github: same name, private)

2) Exomoons (radiative transfer and exoplanets course)

3) exoretrievals (github: same name, private)

4) Platon (github: platon_ACCESS)

5) GPTransmissionSpectra (github: same name, private)

6) Activity indicators and Stellar Parameters (github: HARPSN_activity, twin-planets, and ceres. Also Similar 7 paper)

7) JointParameterEstimation and Juliet (modifications for photometry and my targets) (github under same names)

background: https://medium.com/mlwithdev/the-role-of-machine-learning-in-enhancing-climate-change-models-339ee4dfb642

Advisor for Student Software Projects

1) ffdnet_pytorch_2024 (github: same name)

2) Orbit_Engine (github: same name)

3) Mixed Linear integer programming (github: private repository)

4) Using reinforcement learning to optimize long-haul transportation (github: private repository)

Background: NYU Public Programs talk series -"Multimodal Machine Learning and Climate Change Adaptation" - publicprograms.nyuad.nyu.edu

Page updated

Google Sites

Report abuse