FROM OBSERVATIONS OF TYPE IA SUPERNOVAE IN THE LATE 1990S, IT WAS FOUND THAT THE EXPANSION OF THE UNIVERSE WAS ACCELERATING. ONE OF THE EXPLANATION FOR THE ACCELERATED DEVELOPMENT IS DARK ENERGY. THE COSMOLOGICAL CONSTANT () IS THE SIMPLEST DARK ENERGY MODEL AND IS SUPPORTED BY MANY CURRENT OBSERVATIONS. HOWEVER, THERE ARE SEVERAL PROBLEMS INHERENT IN THE COSMOLOGICAL CONSTANT, NAMELY THE PROBLEM OF NE-TUNING AND COINCIDENCE. TO OVERCOME THIS PROBLEM, ONE OF THE PROPOSED SOLUTIONS IS DYNAMICAL DARK ENERGY (DDE), WHICH IS DARK ENERGY WHICH HAS A CHANGING ENERGY DENSITY AS A FUNCTION OF TIME. TO STUDY THE EVOLUTION OF DDE, WE NEED OBSERVATIONAL DATA SPANNING LARGE REDSHIFTS (Z). THE SN IA PANTHEON DATASET, WHICH IS THE LARGEST TYPE IA SUPERNOVA DATASET, CURRENTLY ONLY REACHES Z = 2:3. IN THIS FINAL PROJECT, THE PANTHEON DATA SET WILL BE REGRESSED TO Z = 4 BY USING A DEEP LEARNING METHOD TO PREDICT THE DATA TO THE REDSHIFT AREA THAT WE STILL CANNOT OBSERVE. THREE TYPES OF RECURRENT NEURAL NETWORK (RNN), NAMELY LSTM, GRU AND SIMPLE RNN WERE USED TO REGRESS THE SN IA PANTHEON DATA SET. THE RNN NETWORK THAT CAN REGRESS PANTHEON DATA WITH THE BEST PERFORMANCE IS THE LSTM CELL. THE DATA FROM THE DEEP LEARNING REGRESSION IS THEN USED IN TTING THE PARAMETERS OF THE CDM MODEL AND THREE DDE MODELS (CPL, JBP, AND BA) TO STUDY THE CHARACTERISTICS OF DDE AND THEIR COMPARISON WITH THE CDM MODEL. USING REAL PANTHEON DATA, THE PARAMETER VALUES W0 AND WA ARE OBTAINED WHICH ARE CONSISTENT WITH THE CDM MODEL (W0 ????1 AND WA 0). BY USING THE COMBINED REAL PANTHEON + DL LSTM DATA, WE GET A DIERENT BEST-T VALUE FOR THE WA PARAMETER COMPARED TO THE CDM MODEL, WHICH RESULTS IN A DIERENT EVOLUTION AT HIGH REDSHIFT. HOWEVER, THE HIGH UNCERTAINTY IN THE VALUE OF WA STILL ALLOWS THE DATA TO BE CONSISTENT WITH THE CDM MODEL. THE TTING RESULTS ALSO ALLOW THE DDE EQUATION OF STATE PARAMETERS TO BE IN THE REGION OF THE QUINTESSENCE AND PHANTOM ELD DARK ENERGY MODELS. IF THE DDE PARAMETERIZATION MODEL IS COMPARED WITH THE CDM MODEL BY CALCULATING THE BIC VALUE, THE CDM MODEL IS THE MOST SUITABLE MODEL, BOTH WITH REAL PANTHEON DATA AND REAL PANTHEON + DL LSTM DATA. KEY WORDS: DYNAMIC DARK ENERGY, DEEP LEARNING, TYPE IA SUPERNOVA.

• Main Author: Mounthella, Ricko
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/57188
• Institution: Institut Teknologi Bandung

From observations of type Ia supernovae in the late 1990s, it was found that the expansion of the universe was accelerating. One of the explanation for the accelerated development is dark energy. The cosmological constant () is the simplest dark energy model and is supported by many current observations. However, there are several problems inherent in the cosmological constant, namely the problem of ne-tuning and coincidence. To overcome this problem, one of the proposed solutions is dynamical dark energy (DDE), which is dark energy which has a changing energy density as a function of time. To study the evolution of DDE, we need observational data spanning large redshifts (z). The SN Ia Pantheon dataset, which is the largest type Ia supernova dataset, currently only reaches z = 2:3. In this Final Project, the Pantheon data set will be regressed to z = 4 by using a deep learning method to predict the data to the redshift area that we still cannot observe. Three types of recurrent neural network (RNN), namely LSTM, GRU and simple RNN were used to regress the SN Ia Pantheon data set. The RNN network that can regress Pantheon data with the best performance is the LSTM cell. The data from the deep learning regression is then used in tting the parameters of the CDM model and three DDE models (CPL, JBP, and BA) to study the characteristics of DDE and their comparison with the CDM model. Using real Pantheon data, the parameter values w0 and wa are obtained which are consistent with the CDM model (w0 ????1 and wa 0). By using the combined real Pantheon + DL LSTM data, we get a dierent best-t value for the wa parameter compared to the CDM model, which results in a dierent evolution at high redshift. However, the high uncertainty in the value of wa still allows the data to be consistent with the CDM model. The tting results also allow the DDE equation of state parameters to be in the region of the quintessence and phantom eld dark energy models. If the DDE parameterization model is compared with the CDM model by calculating the BIC value, the CDM model is the most suitable model, both with real Pantheon data and real Pantheon + DL LSTM data.