Delta nu related to pure sound speed of star
Numax related to surface gravity
Constrained by temperature, luminosity
Hard to get age w/o using star models
Construct simulation of star to the point of same properties
Age of model - compare to age of models stars
TESS input catalogue created as an analogue to Kepler
Include ones in the Kepler input catalogue.
Temperature and Radius has been estimated through photometry.
GAIA records movements and motion of stars
Ability to measure distances and parallaxes of billions of stars in galaxy
Radius and Temperature tabulated in TESS catalog.
Nice data comes from Kepler
TESS has many gaps and weird data.
Red giants pulate at lower frequencies
Depends a lot on the type of star.
Depending on the variability, oscillations can tell
Difficult to come up with prediction that works for all types of star
Narrow down to a particular type of star
Stars at different points exhibit different pulsations
Especially in the instability strip - excited by a very specific mechanism - range of temperature
As the sun gets old, it becomes a red giant and eventually puff up. Oscillation frequencies get to variable values. When instrument gets large, frequency becomes very low. Structure is also different from Sun
Kepler great for solar like stars - long time series.
ML time-series struggle for high frequency signals
Depending on variability type, might vary in timescale
Vary super quickly
Neural network to generalize
Struggle with the fast trends
Start working with red giants on frequency domain
The lower the numax, the larger the amplitude of oscillations in light curve
Lightkurve library is good
Oscillation frequency shift for sun
Some stars may be too quiet or too noisy.
Solve using Kepler’s laws, compare to asteroseismology
Equal pay - accurate, very precise
Depends on the quality of data.
Look at smaller variability range of stars. Instead of entire periodogram, individual peaks around numax.
Dominant feature is numax, teaching network how to infer temperature from numax
If we want to learn detailed stuff, we need to do feature engineering - extract specific features initially and then feed into a neural network.
Use individual peaks -
Mass, radius, surface gravity won’t be completely new
What is new - focus on effective temperature and metallicity
Temperature: photometry, spectroscopy
Metalicity:
Weak dependencies on deltanu and numax. No machine learning, only plotted trend of numax vs. metallicity.
Haven’t used individual modes
Precision comparable to measurements taken from other photometry…
Solar like Short cadence data
Red Giant normal cadence data
Red giant spectra look very clean, but then it becomes much more complicated as it rise up the giant branch..
Not all the peaks correspond to the same thing
Different peaks identify with different waves inside the star. Peaks distributed differently
Know which peaks correlate to which kind of waves.
If main sequence, interior stars’ insides don’t show this level of complexity.
Start learning about main sequence stars first - then read red giants.
Received: Main sequence stars, red giant stars list, paper for gentle learning curve of observed things in asteroseismology
Difficulty - irregular timestamps
Another thought - classification of variables in time domain with irregular timestamp data
Classification of stellar variability is increasingly important in the future.
RNN struggle with long sequences
Most RNN researches use
CNNs are maybe better/applicable to temporal data.
Deep neural networks
Astronomy - have to find the right tool for the job. Start simple, multilayer random forest, classification, then NN.