Life, computers, and the universe.

Three systems that have held my attention for two decades, and the ideas that connect them.

UniverseCosmology, simulation, gravitational waves
ComputersNumerical methods, HPC, data science
LifeInformation, intelligence, entropy

In the early 2000s I began my Ph.D. in cosmology, the study of the origin and evolution of the universe. I found that the cosmos is vast in time and space, and that a human lifetime is far too short to watch it change. I also found that many of the equations are too complex to solve analytically. In both cases computers were invaluable. They let us simulate clusters of galaxies and stars, and solve equations such as Einstein's General Relativity numerically. Alongside theory, observation and experiment, numerical simulation is now a central way to advance science.

Life takes a different turn. From Schrödinger to Freeman Dyson, physicists have been drawn to living systems. Life as we know it has one system for managing energy (mostly proteins built from amino acids) and another for managing information (DNA, built from nucleic acids). The chance of both arising together is very low, but it rises if the two abilities developed independently at different times. Life also runs against the simple picture of the second law of thermodynamics, since disorder does not simply grow with time inside it. Wherever information is at stake, physicists and computer scientists take an interest.

At the meeting point of life, information, intelligence and consciousness there is a great deal of active research, mostly centred on human beings. Building a theory from a single example may not be fruitful. Powered flight only took off once people studied the principles of aviation rather than copying birds. Life, intelligence and consciousness may be the same: the goal should be principles that cover far more cases than the ones we know.

That brings up Artificial General Intelligence, intelligence superior to ours in every respect. It is a hard target when we still cannot quantify overall human intelligence. Humans, for instance, are far better learners than AI systems in terms of how much training they need.

For a semi-popular article about the universe, read the English version or the Hindi version. If you have questions, or just want to say hi, get in touch.

About

My old IUCAA webpage is still online but no longer maintained: visit it here.

I was born and raised in a small village in Rudraprayag district, Uttarakhand. I completed my Master's in Physics (specialising in Electronics & Communication) at HNB Garhwal University, Srinagar Garhwal, in 1998. I then taught physics to classes XI and XII at SGRR Public School and earned a Diploma in Education in 2000. After clearing the national JEST and CSIR-UGC (NET) tests, I moved to Harish-Chandra Research Institute, Allahabad (now Prayagraj), in 2001 to pursue a Ph.D. in Astrophysics.

2001–2008 · HRI, Allahabad. Ph.D. in Astrophysics under Prof. J. S. Bagla.
2008 · NCRA-TIFR. First postdoc with Prof. Jayaram Chengalur, working on computational tasks and GMRT data-processing pipelines.
2010–2018 · IUCAA. Postdoc with Prof. Tarun Souradeep, later Principal Investigator of a project funded by the Department of Science & Technology. Member of the LIGO Scientific Collaboration from 2012 to 2018, and helped set up a 200 TF LIGO Data Grid cluster at IUCAA.
2018 · Accelere (now Embold) Technologies. Data scientist.
2020 · DISYS India, Chennai. Data scientist.

At IUCAA I became deeply involved in Big Data, high-performance and grid computing. Our work with LIGO was well appreciated, and we were part of the team behind the first detection of a black hole merger in 2015. I have co-authored over 100 publications with LIGO. Some press coverage:

Research

My full publication list and a longer research summary are available as PDFs.

For almost two decades I worked across computational, observational and theoretical astrophysics, and published over 100 papers in peer-reviewed journals. My areas were cosmology (large-scale structure, the cosmic microwave background, the early universe), radio astronomy (transient detection, H21 mapping, software design and data analysis) and gravitational waves. In gravitational waves my role was mostly setting up and managing high-performance grid computing systems and deploying data-analysis pipelines, alongside some scientific contributions.

Large-scale structure of the universe

In the Big Bang model the universe expanded and cooled from a singularity about 13.7 billion years ago. Redshifted light from distant galaxies, the cosmic microwave background, the abundances of light elements and the distribution of galaxies all support it. The energy budget is dominated by dark matter, which does not radiate, and dark energy, which accelerates the expansion.

The early universe had no galaxies, only small fluctuations that gravity amplified into the structures we see. This growth is nonlinear and cannot be modelled analytically, so most progress comes from cosmological N-body simulations, which evolve huge numbers of particles in an expanding background. My Ph.D. work covered mode coupling between scales and finite-volume effects in these simulations.

Role of substructure

In cold dark matter models, small scales collapse first. I asked whether small-scale perturbations affect the collapse of larger structures beyond what shows up in the power spectrum.

  • Bagla, Prasad & Ray, 2005, MNRAS, 360, 194 (astro-ph/0408429). Gravitational collapse in an expanding background and the role of substructure I: Planar collapse.
  • Bagla & Prasad, 2008, MNRAS (arXiv:0802.2796). Substructure II: Excess power at small scales and its effect on the collapse of structures at larger scales.

Finite-volume effects in N-body simulations

Simulations model a finite box and ignore fluctuations larger than it. I studied how measures of gravitational clustering depend on box size.

  • Bagla & Prasad, 2006, MNRAS, 370, 993 (astro-ph/0601320). Effects of the size of cosmological N-body simulations on physical quantities I: Mass function.
  • Prasad, 2007, J. Astrophys. Astron. 28, 117 (astro-ph/0702557). II: Halo formation and destruction rate.
  • Bagla, Prasad & Khandai, 2009, MNRAS, 395, 2 (arXiv:0804.1197). III: Skewness.

Artificial intelligence and machine learning

There have been four industrial revolutions: mechanisation (18th century), electricity (19th), computers and the Internet (20th), and now, in the 21st, AI, machine learning and data science. Data is the new oil. For technical detail, see my Medium, GitHub and LinkedIn pages.

What is intelligence?

This is as hard as asking what life is. The working definition of life is "life as we know it on Earth", a definition drawn from the one example we have. Intelligence is similar: we mostly have human intelligence to go on, so we define it relative to that. Other kinds may exist that we do not know about.

AI rests on the premise that human-like intelligence can be built from machines, as Alan Turing proposed around 1950. Language is the simplest and most common way to show intelligence, and any intelligence probably has language at its centre. Turing suggested that a machine able to answer questions like a human could be called intelligent. ChatGPT is an example.

A truly intelligent system needs more than language. Russell and Norvig list:

  • Language (natural language processing)
  • Knowledge representation
  • Learning from experience (machine learning)
  • Sensing and manipulating the environment (robotics)

Machine learning is software that learns from data without being explicitly programmed and improves at tasks such as understanding speech, recognising images or making predictions. Siri, Alexa and ChatGPT are built on it. Many different algorithms exist for different tasks, and I write about them on Medium.

Downloads

I could not read the original downloads page, so this list only holds the documents linked from other pages of the site.

Resources

Tutorials, primers and articles I have found useful.

Artificial intelligence

Computer science

Statistics and mathematics

Data science and machine learning

Deep learning

Natural language processing

Computer vision

Time series

Quantum computing

Software engineering

Miscellaneous

Learning

Reading is a hobby of mine, across science, society, technology and philosophy. My ten all-time favourites:

  1. Jared Diamond, Guns, Germs and Steel
  2. Jared Diamond, Collapse
  3. Thomas L. Friedman, The World Is Flat
  4. Richard Dawkins, The Selfish Gene
  5. V. S. Ramachandran, The Tell-Tale Brain
  6. Siddhartha Mukherjee, The Gene: An Intimate History
  7. Yuval Noah Harari, Sapiens
  8. Max Tegmark, Life 3.0
  9. Ray Dalio, Principles
  10. Daniel Kahneman, Thinking, Fast and Slow

I also teach, train and mentor, and my learning content is spread across platforms. Some of it sits behind a paywall; open-source material is a good alternative.

Structured courses

Articles on Medium

Video lectures

Contact

The best way to reach me is through my social pages.