Science Report 015

Changing the World with Data Science 03

Tracing the Cosmic History Back to the Big Bang

Modern cosmology tells us it’s been 13.8 billion years since the universe emerged. Following the Big Bang, the hot and dense universe kept cooling down through continuous expansion. About 380,000 years later, the universe became transparent to radiation, enabling us to directly observe the universe. Approximately 100 million years after the Big Bang, the first generation of stars were born, and they illuminated the universe. Cosmology, the study of the origin and evolution of the universe, has risen from serendipitous discoveries that people made while gazing into space and absorbing all the information they could see, said Prof. Naoki Yoshida of Kavli Institute for the Physics and Mathematics of the Universe at the University of Tokyo. “Cosmology, from its beginning, has always been a frontier of data science,” he said. Researchers today are trying to build on this cosmology-data science connection by taking advantage of the ability to collect literally astronomical amounts of observational data. Will the technology help elucidate how the universe came into existence? Three experts involved in such transdisciplinary research efforts will answer the question and explain what is happening on the cutting edge of cosmology and data science.

Twitter Facebook Google+ Feed

Ask the Experts: Prof. Naoki Yoshida (Kavli Institute for the Physics and Mathematics of the Universe at the University of Tokyo)

Dr. Naoki Yoshida is a cosmologist and astrophysicist working to solve mysteries surrounding dark matter and black holes. He has served as specially appointed professor at Kavli Institute for Physics and Mathematics of the Universe at the University of Tokyo and as professor of astrophysics at the University of Tokyo since 2014 and 2012 respectively. He received his bachelor of science in aerospace engineering and doctorate degree in science from the University of Munich and has worked at Harvard University’s Department of Astronomy and at Nagoya University in Japan.

Ask the Experts: Prof. Shiro Ikeda (Institute of Statistical Mathematics)

Dr. Shiro Ikeda is a senior research scientist at the Institute of Statistical Mathematics. His research focuses include separation of acoustic signals with the use of “independent component analysis,” a multivariate statistical analysis technique, as well as the development of processing and analysis techniques for noisy data. He also works to find out methods to improve astronomical and physical measurement through sparse modelling based on the premise that estimated target signals contain many zeros. He received his doctorate degree in engineering from the University of Tokyo in 1996.

Ask the Experts: Dr. Mikio Morii (Institute of Statistical Mathematics)

Dr. Mikio Morii serves as specially appointed professor at the Institute of Statistical Mathematics. He specializes in observational astronomy.

Key to Finding Out How the Universe Appeared

Prof. Naoki Yoshida is involved in a 5-year space observation project to use an advanced camera newly installed in the Subaru telescope to collect and analyze imaging data, which is expected to total as much as 1 petabyte in data size.

“A single image typically has several thousand galaxies captured in it. They may look like a cloud of fog at first, but the resolution is so high that you can zoom in far enough to identify individual stars. You can use the camera to capture what’s happening in a vast area of space,” Prof. Yoshida said. “People often think the universe is a calm place and looks pretty much the same at all times. But when you compare an image you just took from the one from the day before, you will know just how much has changed in 24 hours. In our daily analysis of the data collected with the Subaru telescope, we normally find evidence for explosions of about 100 stars a night,” he said. “You can, of course, zoom into each supernova and see the details with your own eyes, but there are just too many of them to check on manually. So, we use the computer to gauge differential values between images to understand the changes occurring in the universe.”

What’s behind Prof. Yoshida’s focus on supernova explosions is Saul Perlmutter, a 2011 Nobel Prize winner work. In 1998, the U.C. Berkeley astrophysicist used his analysis of Type Ia supernovae in the distant universe to prove that the universe continues to expand today, and that the rate of expansion is accelerating. Prof. Yoshida’s research centers around Type Ia supernovae, which he said make up about a half of supernovae observed through the Subaru telescope.

“Artificial intelligence (AI) now has remarkable image analysis capabilities and can distinguish Type Ia supernovae from other types. Prof. Shiro Ikeda of the Institute of Statistical Mathematics has been instrumental in helping us develop a tool for identifying Type Ia,” Prof. Yoshida said.

The question is how accurately AI can distinguish Type Ia from others, according to Prof. Shiro Ikeda of the Institute of Mathematical Statistics.

“Getting AI to accurately identify which supernovae are most likely Type Ia is our challenge,” Prof. Ikeda said.

Type Ia supernovae identified through analysis. They are compiled in data base along with the information on their luminary levels and transformations.

Pursuing High Accuracy in Identifying Type Ia

Type Ia supernovae are born when white dwarf stars with about 1.4 solar masses explode, and they are almost equally luminous, according to Dr. Mikio Morii of the Institute of Statistical Mathematics (ISM).

“You can estimate Type Ia supernovae’s distances based on how bright they appear to the naked eyes. You can also calculate the speeds at which they are moving away based on spectral observations,” Dr. Morii said. “Based on the distances and speeds of supernovae, you can estimate how fast the universe is expanding. Accelerating expansion means the distant universe is expanding faster than originally thought.”

Accurately identifying Type Ia supernovae isn’t easy, however.

“In fact, 99 percent of what appear to be supernovae turn out to be something else,” Prof. Ikeda said. “We compare two images taken at different times to find supernovae, but sometimes the differences aren’t as clear. We select ‘potential Type Ia’ from an image, and that typically makes up less than 1 percent of the changes we observe from the images. We would then use a telescope to continuously observe those stars that we think may be Type Ia. After collecting enough spectral data on them, we put it together with the stars’ locations and speeds to come up with a complete profile on each of them” Prof. Ikeda said. “After making great strides with accuracy, we are now able to identify 10 to 20 potential supernovae a night.”

Prof. Ikeda’s expertise is statistical science, not cosmology.

“When you have observation data that contains noise, you need to look past the noise to understand what the true data is. Doing that is a statistical scientist’s job,” Prof. Ikeda said. “We call it signal processing when the data is comprised of physical signals. It often requires multiple steps using various signal processing methods to make the actual data emerge. I do this processing with astronomical data, as well, including optical light data collected with instruments like the Subaru telescope and data captured with radio telescopes. The field of signal processing is active and growing, as the advancement of measuring instruments provides us with more opportunities to collect new kinds of data,” Prof. Ikeda said.

The Institute of Statistical Mathematics shares its building in Tachikawa City, Tokyo, with the National Institute of Polar Research.

Mapping the Distribution of Dark Matter

Prof. Yoshida’s 5-year observation project also focuses on gravitational lens.

“We find a number of galaxies with distorted shapes all around space. They are not physically distorted but just look distorted because of gravity that bends light-rays. The degrees of shape distortion reveal heavy objects have large quantities of some sort of invisible matter around them,” Prof. Yoshida said.

This “some sort of invisible matter” is what is also known as dark matter. It is believed that the universe is made up of 4 percent chemical elements and 22 percent dark matter, with the remaining 74 percent being dark energy.

“We are investigating large areas of space to understand three dimensional distributions of dark matter, which is, in essence, a map of the universe,” Prof. Yoshida said.

Scientists around the world have been studying the large-scale structure of the universe since the 1980s. But, “Just figuring out dark matter distributions based on gravitational calculations can help further theoretical models,” Prof. Yoshida said. “We want to look at every distorted galaxy, including very subtly distorted ones, and compute the degrees of distortion. We can then begin to think about the nature of dark matter that makes the galaxy look the way it does through gravitational lensing. That provides basis for developing new theories and models. But, in order to map dark matter’s distributions, you need to start with observations and trace them back to causes, which is a very difficult thing to do. This is another area in which we pull together the ISM’s resources to support the research,” he said.

The arcs shown around the bright stars are the examples of “distortion.” Strong gravity bends straight light-rays to give distant galaxies distorted appearances.

The Universe Doesn’t Show You Its True Self

Separately from the 5-year observation project, Prof. Yoshida is also working on computer simulations to show how the universe began and how stars and galaxies are formed.

“The universe is not nice enough to show you the true picture of how its components, such as galaxies, are distributed. So, we do our best to estimate the distribution based on observations. But you can create an accurate distribution map of matter through computer simulations,” Prof. Yoshida said.

“Our goal is to shed light on the evolution of the universe. We will use whatever it takes to get to the goal, including AI and super-computers,” he said.

Prof. Yoshida believes that it’s possible to understand what’s happening around the universe using the knowledge and experiences we have on Earth.

“Saying, ‘Space is nothing like here,’ won’t get you anywhere,” he said. “Contrary to what some people might think, verified facts from our experiences on Earth provide the foundation for understanding the universe. If you mobilize utilize all the knowledge we have and still don’t understand something, then we can identify what’s missing. Once we’ve exhausted the process and realize there’s a critical piece missing that we don’t have, that’s when a discovery happens.”

Prof. Naoki Yosida at his lab at the University of Tokyo’s Hongo campus

Interviewer: Rue Ikeya
Photographs: Yuji Iijima unless noted otherwise
Released on: October 10, 2018 (The Japanese version released on March 12, 2018)

Inter-University Research Institute Corporation Research Organization of Information and Systems
Hulic Kamiyacho 2F 4-3-13 Toranomon Minatoku Tokyo 105-0001, Japan
TEL:+81-3-6402-6200 FAX:+81-3-3431-3070 E-mail:
Copyright © Inter-University Research Institute Corporation Research Organization of Information and Systems (ROIS). All rights reserved.