Cover page

”Kakka Vezhambal” is another name for the Malabar Grey Hornbill (Ocyceros griseus), a bird endemic to

the Western Ghats . Commonly found in Kerala’s evergreen and moist deciduous forests, it is known for its

loud, cackling calls that are considered a characteristic sound of the region’s wilderness . Recognizable by its

brownish-grey plumage and yellow-orange bill, it primarily feeds on fruits like ﬁgs and plays a vital role in the

forest ecosystem as a seed disperser . Photograph taken by Geetha Paul from airis4D campus.

Managing Editor Chief Editor Editorial Board Correspondence

Ninan Sajeeth Philip Abraham Mulamoottil K Babu Joseph The Chief Editor

Ajit K Kembhavi airis4D

Geetha Paul Thelliyoor - 689544

Arun Kumar Aniyan India

Sindhu G

Journal Publisher Details

Publisher : airis4D, Thelliyoor 689544, India

Website : www.airis4d.com

Email : nsp@airis4d.com

Phone : +919497552476

Editorial

by Fr Dr Abraham Mulamoottil

airis4D, Vol.4, No.3, 2026

www.airis4d.com

We are delighted to announce that airis4D Journal

will feature a new monthly column titled Vijnanam—the

Sanskrit word for knowledge—beginning in the May

2026 issue. The column will be authored by Prof.

K. Babu Joseph, former Vice-Chancellor of Cochin

University of Science and Technology (CUSAT) and

currently an advisor to this journal. A distinguished

physicist and accomplished writer, Prof. Joseph has

published extensively on popular science, philosophy,

and poetry in both English and Malayalam. Vijnanam

will primarily explore themes in science, though it will

occasionally venture into other disciplines, oﬀering

readers a rich and thoughtful perspective shaped by his

multifaceted intellectual pursuits. We warmly welcome

Prof. Babu Joseph to this regular feature and look

forward to his valuable contributions.

This edition starts with ”The Power of Small

Visual Language Models,” by Dr. Arun Aniyan,

who argues that the future of artiﬁcial intelligence

lies not in massive, resource-intensive systems but

in their compact, eﬃcient counterparts known as

Small Visual Language Models (SVLMs). Aniyan

explains that SVLMs achieve their eﬃciency through a

signiﬁcantly reduced parameter count and specialized

training techniques like knowledge distillation, where a

smaller ”student” model learns from a larger ”teacher”

model’s nuanced understanding. This streamlined

design enables faster response times, lower energy

consumption, reduced costs, and most importantly,

allows these models to run directly on edge devices

such as smartphones and smart cameras. By enabling

this on-device intelligence, SVLMs enhance data

privacy while powering practical applications across

accessibility tools, manufacturing quality control, retail

management, and contextual smart home systems,

ultimately representing what Aniyan describes as a

crucial step toward embedding ubiquitous, personal AI

into everyday technology.

In ”Introduction to Neuromorphic Computing-

Part II,” author Dr. Blesson George explores how

brain-inspired computing oﬀers an energy-eﬃcient

alternative to traditional systems by mimicking the

neural and synaptic structures of the human brain.

Aniyan explains that, unlike conventional computers,

which suﬀer from the Von Neumann bottleneck and high

power consumption, neuromorphic systems use event-

driven Spiking Neural Networks (SNNs) and specialized

chips like IBM’s TrueNorth and Intel’s Loihi to

perform parallel, real-time processing with remarkable

eﬃciency. The article highlights how these systems

integrate memory and computation to enable low-power

applications in robotics, edge AI, and adaptive sensing,

while also acknowledging ongoing challenges such as

the lack of standard programming frameworks and

the diﬃculty of training SNNs. Ultimately, Aniyan

positions neuromorphic computing as a promising

paradigm shift toward intelligent systems that can

learn continuously and operate with minimal energy

consumption.

In ”Plasma Physics- Magnetosphere &

Ionosphere,” Abishek P S examines how Earth’s

magnetic ﬁeld creates a dynamic plasma environment

that shields the planet while continuously exchanging

energy and particles with the upper atmosphere. The

article explains that the magnetosphere, formed by

Earth’s internal dynamo and shaped by the solar wind,

traps charged particles in radiation belts and funnels

them toward the poles to create auroras, while the

ionosphere acts as both a source of plasma—supplying

ions like oxygen and hydrogen upward—and a sink

where precipitating particles deposit energy. Abishek

details the coupling mechanisms that link these regions,

including ﬁeld-aligned currents that act as electrical

pathways, Alfv

en waves that transfer energy along

magnetic ﬁeld lines, and the pressure cooker eﬀect

that drives heavy ions into space during geomagnetic

storms. The author emphasizes that this interconnected

system serves as a natural plasma laboratory where

scientists can study universal processes like turbulence,

instabilities, and nonlinear coupling through direct

satellite measurements, oﬀering insights that extend

from space weather forecasting to understanding

astrophysical phenomena across the universe.

Aromal P provides in ”X-ray Astronomy: Theory”

a comprehensive overview of the primary cosmic

sources that emit X-rays and the physical mechanisms

responsible for their radiation. The article explains

that X-ray emission across the universe arises from

extreme environments where matter is heated to millions

of degrees or particles are accelerated to relativistic

speeds. Stellar coronae, like our Sun’s outer atmosphere,

produce X-rays through magnetic reconnection and

thermal bremsstrahlung, while supernova remnants

generate both thermal emission from shock-heated gas

and non-thermal synchrotron radiation from pulsar-

accelerated electrons. X-ray binaries, the brightest

sources in our galaxy, involve neutron stars or black

holes accreting matter from companion stars, producing

X-rays through disk viscosity, surface impact, and

inverse Compton scattering. The author also describes

isolated neutron stars cooling from formation or

galactic nuclei where supermassive black holes generate

X-rays via coronal inverse Compton scattering, and

galaxy clusters where the intracluster medium emits

thermal bremsstrahlung from gravitational heating.

Aromal, whose research focuses on thermonuclear X-ray

bursts on neutron stars, presents these diverse sources

as a testament to the universal processes of accretion,

magnetic reconnection, and shock heating that make

X-ray astronomy a powerful window into the most

energetic phenomena in the cosmos.

In ”Giant Molecular Clouds: Structure, Evolution,

and Their Role in Star Formation”, Sindhu G provides

a comprehensive overview of the complex, multi-

scale processes that transform cold interstellar gas

into nuclear-burning stars. The article explains

that star formation occurs almost exclusively within

giant molecular clouds, where dense cores become

gravitationally unstable when thermal pressure,

turbulence, and magnetic ﬁelds can no longer support

them against collapse—a condition quantiﬁed by the

Jeans criterion. As collapse proceeds, conservation

of angular momentum leads to the formation of

circumstellar accretion disks that regulate material ﬂow

onto the central protostar, while magneto-centrifugal

forces launch bipolar jets and outﬂows that remove

excess angular momentum and provide feedback into

the surrounding medium. Sindhu describes how

protostars evolve through embedded phases, powered

by gravitational contraction rather than fusion, before

emerging as pre-main-sequence objects like T Tauri

stars that gradually contract until hydrogen ignition

marks their arrival on the zero-age main sequence.

The author emphasizes that formation timescales vary

dramatically with mass—from millions of years for

Sun-like stars to less than a hundred thousand years

for massive stars—and highlights how modern infrared

and sub-millimetre observations from facilities like

JWST continue to reﬁne our understanding of these

fundamental processes that shape galaxies and set the

stage for planet formation.

Linn Abraham explores in ”Deep Learning: A

Review,” the ttransition from Mchine Learning to Deep

Learning. The article explains that while traditional

interpretability tools like Integrated Gradients were

designed for CNN architectures, transformer models

oﬀer a more natural approach by leveraging their

built-in attention mechanisms. Abraham demonstrates

how monkey patching—dynamically modifying code

at runtime—allows researchers to replace standard

attention blocks with patched versions without

retraining the model, while PyTorch hooks enable the

capture of attention matrices and their gradients during

iii

forward and backward passes. By registering hooks

before and after the softmax operation in attention

layers, the AGCAM method stores attention values

and computes gradients through backpropagation, then

combines these across layers and normalizes them

with sigmoid activation to produce class activation

maps that highlight which image regions inﬂuenced the

model’s decisions. This approach provides an intuitive

visualization tool for understanding ViT behavior, with

applications ranging from model debugging to building

trust in AI systems used in ﬁelds like medical imaging

or, as in the author’s own work, solar ﬂare prediction

and galaxy classiﬁcation.

Aengela Grace Jacob explores in ”Kombucha:

’The Symbiotic Elixir’ – A Comprehensive View

on its Bioprocessing,” the microbial science behind

the popular fermented tea beverage, revealing how

a simple mixture of sweetened tea transforms into a

complex probiotic drink through the action of a SCOBY

(Symbiotic Culture of Bacteria and Yeast). The article

explains that this rubbery, cellulose-based disc serves

as the living engine of fermentation, housing a diverse

community of Acetic Acid Bacteria that produce acetic

acid and build the SCOBY’s physical structure, yeasts

like Saccharomyces that convert sugar to ethanol and

carbon dioxide, and Lactic Acid Bacteria that contribute

sour ﬂavors and probiotic properties. Jacob details

the two-phase fermentation process—an initial aerobic

stage where oxygen enables acid production, followed

by anaerobic bottle conditioning that creates natural

carbonation—and describes how enzymes like invertase

break down sucrose into fermentable sugars. The

author emphasizes kombucha’s scientiﬁcally-backed

beneﬁts for gut health, including its ability to restore

balanced gut ﬂora through probiotic diversity, suppress

harmful pathogens with organic acids, and increase the

bioavailability of anti-inﬂammatory tea polyphenols,

positioning this ancient elixir as a modern testament to

the power of microbial symbiosis in promoting human

wellness.

In ”The Genomic Landscape of Major Depressive

Disorder,” Geetha Paul explores how modern

genetic research has fundamentally transformed our

understanding of depression from a single-gene disorder

to a complex polygenic trait inﬂuenced by thousands of

common genetic variations called single nucleotide

polymorphisms (SNPs). The article explains that

while individual SNPs exert only negligible eﬀects

on depression risk, their cumulative impact—measured

through Polygenic Risk Scores (PRS)—can identify

individuals with up to ﬁve times greater susceptibility to

MDD before clinical symptoms ever manifest, enabling

preemptive monitoring and personalized interventions.

Paul highlights how SNP analysis helps resolve the

diagnostic heterogeneity of depression by revealing

distinct biological subtypes, with some patients

carrying variants concentrated in serotonin pathways

while others show genetic burdens in inﬂammatory

genes like IL-6. The author also examines critical

pharmacogenomic applications, detailing how speciﬁc

polymorphisms such as the BDNF Val66Met variant,

which impairs neural plasticity, and the SLC6A4 5-

HTTLPR polymorphism, which inﬂuences serotonin

transporter expression, can predict antidepressant

response and guide treatment decisions. This molecular

approach, Paul argues, moves psychiatry beyond the

traditional one-size-ﬁts-all model toward a more precise,

stratiﬁed framework where genetic biomarkers bridge

the gap between subjective symptom reporting and

objective biological data.

Finally, Dr. Ajay Vibhute presents in ”From

Telescope to Data Product” a comprehensive exploration

of how computational methods have evolved from

manual calculations and mechanical aids to become

an indispensable instrument in modern astronomy,

fundamentally transforming how scientists observe,

analyze, and understand the cosmos. The article

traces the historical progression from logarithmic

tables and punched-card mainframes like the IBM 704

to sophisticated software pipelines that now process

terabytes of data from surveys such as Hipparcos and the

Sloan Digital Sky Survey, highlighting that computation

functions not merely as a tool but as a virtual scientiﬁc

instrument that actively shapes raw measurements

into meaningful discoveries through algorithms for

image deconvolution, source extraction, and statistical

modeling. Dr. Vibhute explains that astronomical

data presents unique challenges—massive volumes,

inherent noise from multiple sources, instrumental

response functions like point-spread functions, and

multi-dimensional structure across spatial, temporal,

and spectral domains—requiring specialized techniques

and careful pipeline design that balances algorithmic

sophistication with computational feasibility. The

author emphasizes that astronomical computing

encompasses the entire knowledge lifecycle from data

acquisition and calibration through reduction, analysis,

simulation of phenomena like galaxy formation,

advanced visualization, and petabyte-scale archiving,

all while remaining constrained by physical instrument

limitations and the stochastic nature of photons. As data

volumes continue to grow, Dr. Vibhute concludes that

robust, transparent computational methods will become

increasingly crucial, cementing computation’s role as

the essential bridge connecting raw observations to our

fundamental understanding of the universe.

News Desk

Vijnanam by Prof Babu Joseph, former VC, CUSAT

airis4D has the pleasure of announcing a monthly column, titled Vijnanam, which in Sanskrit, means

knowledge. It will be handled by Prof. K. Babu Joseph, formerly of CUSAT, and currently, an advisor to this

journal. He is a physicist and a writer of popular science, philosophy and poetry in English and Malayalam.The

series will be focussed on science, but occasionally, address other disciplines as well. The ﬁrst contribution will

appear in the May 2026 issue.

Contents

Editorial ii

I Artiﬁcial Intelligence and Machine Learning 1

1 The Power of Small Visual Language Models 2

1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2

1.2 What is a Visual Language Model (VLM)? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2

1.3 The ’Small’ in Small Visual Language Models (SVLMs) . . . . . . . . . . . . . . . . . . . . . . 2

1.4 Eﬃciency and Speed . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3

1.5 How SVLMs Work: The Training Secret . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3

1.6 Why Small Models are the Future . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5

1.7 Real-World Applications of Small Visual Language Models . . . . . . . . . . . . . . . . . . . . . 6

1.8 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6

2 Introduction to Neuromorphic Computing- Part II 7

2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7

2.2 Basic Idea of Neuromorphic Computing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7

2.3 Why Neuromorphic Computing is Important . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7

2.4 Spiking Neural Networks and Simple Neuron Model . . . . . . . . . . . . . . . . . . . . . . . . . 8

2.5 Neuromorphic Chips . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8

2.6 Diﬀerence from Conventional Computing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8

2.7 Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8

2.8 Challenges . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9

2.9 Future Scope . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9

2.10 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9

3 Deep Learning: A Review 10

3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10

3.2 What is Machine Learning? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10

3.3 Classical Machine Learning Techniques . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11

3.4 Deep Learning . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12

II Astronomy and Astrophysics 14

1 Plasma Physics- Magnetosphere & Ionosphere 15

1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15

1.2 The Magnetosphere as a Plasma Environment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15

1.3 The Ionosphere as a Plasma Source and Sink . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16

1.4 Coupling Mechanisms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16

1.5 Energy Flow . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17

1.6 Magnetosphere & Ionosphere in Research Perspective . . . . . . . . . . . . . . . . . . . . . . . . 18

CONTENTS

2 X-ray Astronomy: Theory 20

2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20

2.2 X-ray Binaries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20

3 Giant Molecular Clouds: Structure, Evolution, and Their Role in Star Formation 23

3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23

3.2 Chemical Composition and Physical Characteristics . . . . . . . . . . . . . . . . . . . . . . . . . 23

3.3 Formation of Giant Molecular Clouds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24

3.4 Internal Structure of GMCs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24

3.5 Cloud Dynamics: Turbulence and Magnetic Fields . . . . . . . . . . . . . . . . . . . . . . . . . . 24

3.6 Star Formation in Giant Molecular Clouds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25

3.7 Lifetimes and Evolution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25

3.8 Observational Techniques . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25

3.9 Role in Galactic Evolution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25

3.10 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25

III Biosciences 27

1 Kombucha: “The Symbiotic Elixir”

A Comprehensive View on its Bioprocessing 28

1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28

1.2 SCOBY: The Heart of the Brew . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28

1.3 Microbial Contents: The Living Force . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28

1.4 The Fermentation Mechanism . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29

1.5 Major Beneﬁts on Gut Health . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30

1.6 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30

2 The Genomic Landscape of Major Depressive Disorder 31

2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31

2.2 Quantifying Genetic Liability via Polygenic Risk Scores (PRS) . . . . . . . . . . . . . . . . . . . 31

2.3 Resolving Diagnostic Heterogeneity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32

2.4 Pharmacogenomics and Treatment Response . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32

IV Computer Programming 35

1 From Telescope to Data Product 36

1.1 From Telescope to Data Product . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36

viii

Part I

Artiﬁcial Intelligence and Machine Learning

The Power of Small Visual Language Models

by Arun Aniyan

airis4D, Vol.4, No.3, 2026

www.airis4d.com

1.1 Introduction

Artiﬁcial Intelligence (AI) is rapidly transforming

every sector of the world, and at the forefront

of this evolution are models with the remarkable

capability to process and understand both textual

and visual information. These sophisticated systems

are broadly categorized as Visual Language Models

(VLMs). While the public imagination is often

captured by immense, resource-intensive models that

boast staggering performance metrics, a quieter, yet

profoundly signiﬁcant, technological shift is underway.

This shift centers on their leaner, more eﬃcient cousins:

Small Visual Language Models (SVLMs).

The rise of SVLMs represents a critical inﬂection

point in the AI landscape. They are designed to achieve

high performance with a fraction of the parameters,

computational power, and memory footprint required

by their massive predecessors. This focus on eﬃciency

makes them ideal for deployment in environments where

resources are constrained, such as mobile devices, edge

computing systems, and integrated hardware solutions.

This detailed exploration will delve into the core

of the AI revolution driven by these compact systems.

We will meticulously deﬁne the technical architecture

and operational mechanisms of Small Visual Language

Models, scrutinize the compelling reasons behind their

growing importance—ranging from cost reduction and

energy eﬃciency to improved privacy and real-time

inference—and analyze the transformative impact they

are having on the design and functionality of everyday

technology.

1.2 What is a Visual Language Model

(VLM)?

To understand a small VLM, we ﬁrst need to grasp

the concept of a regular VLM.

Imagine a machine that can look at a picture and

simultaneously read a description of it, then use that

combined understanding to answer questions or generate

new text. This is the essence of a Visual Language

Model. It is a type of AI that has been trained on a

massive dataset of images paired with descriptive text

(like captions, articles, or transcripts).

VLMs have two main ”brains” working together:

A Vision Encoder: This part processes the

image, recognizing objects, shapes, colors, and spatial

relationships. It turns the visual information into a

numerical code the computer can understand.

A Language Decoder: This part handles the text,

understanding grammar, meaning, and context. It’s

often a powerful language model, similar to the ones

used in smart assistants or translation tools.

The magic happens when these two parts learn

to communicate, creating a uniﬁed understanding of

”what is seen” and ”what is said.”

1.3 The ’Small’ in Small Visual

Language Models (SVLMs)

When we refer to Small Visual Language Models

(SVLMs), the descriptor ”small” is fundamentally tied

to the model’s size and its computational requirements.

Unlike their larger, often monolithic counterparts,

1.4 Eﬃciency and Speed

SVLMs are characterized by a signiﬁcantly reduced

number of parameters. This decreased parameter count

directly translates into a smaller memory footprint,

making these models substantially more accessible for

deployment on edge devices, mobile platforms, and

resource-constrained environments where large-scale

GPU clusters are unavailable or impractical.

The minimized computational requirements are

a critical advantage. This involves lower demands for

processing power (FLOPs) during both the training

and, crucially, the inference phases. A smaller model

is faster to train, reducing the associated energy

consumption and monetary cost, democratizing the

development of multimodal AI. Furthermore, their

lower inference latency and energy expenditure allow

for real-time processing of visual and textual data locally,

unlocking new applications in areas like augmented

reality, robotics, and instantaneous on-device image

analysis without needing constant cloud connectivity.

Therefore, ”small” is not a compromise on capability

but a strategic design choice focused on eﬃciency,

accessibility, and real-world deployability.

1.3.1 Model Size (Parameters)

Every AI model, regardless of its speciﬁc

application—whether it’s a sophisticated Large

Language Model (LLM) or a specialized Small Visual

Language Model (SVLM)—is fundamentally built upon

an enormous number of adjustable settings referred

to as parameters. These parameters are the core,

tangible components that deﬁne the model’s structure

and capabilities.

Think of parameters as the accumulated

knowledge, memories, and intricate connections that

reside within the AI’s artiﬁcial brain. During the

training process, the AI is fed vast amounts of data

(text, images, code, etc.). As it processes this data,

it constantly adjusts the values of these billions of

parameters. These adjustments are the process through

which the model learns to identify patterns, understand

context, generate coherent text, or recognize objects in

an image.

Essentially, the greater the number of parameters,

the more potential capacity the model has to learn

complex relationships, store nuanced information, and

perform diverse tasks. It is the precise conﬁguration

of these parameters—the delicate balance of their

numerical values—that allows an AI to successfully

execute its designed function.

SVLMs contain signiﬁcantly fewer parameters

than their massive counterparts. This reduction is

intentional and oﬀers profound advantages.

1.4 Eﬃciency and Speed

Smaller models require less computing power,

which leads to several practical beneﬁts:

Faster Response Times (Latency): Because

there are fewer calculations to perform, an SVLM

can process an image and generate a response

much quicker. This is crucial for real-time

applications like autonomous driving or instant

camera translations.

Reduced Energy Consumption: Less

processing power translates directly to less energy

usage. This is vital for sustainability and for

running AI on battery-powered devices.

Lower Cost: Training and deploying smaller models is

vastly cheaper than deploying large models, making the

technology accessible to a wider range of businesses

and developers.

1.5

How SVLMs Work: The Training

Secret

SVLMs don’t just shrink the large models; they are

often trained using specialized techniques to maximize

their performance despite their size. One of the most

eﬀective and widely adopted strategies for creating

powerful, yet resource-eﬃcient, Small Visual Language

Models (SVLMs) is a technique called knowledge

distillation. This process is a revolutionary training

paradigm that fundamentally shifts how smaller models

acquire complex capabilities, moving away from brute-

force training toward targeted, eﬃcient knowledge

transfer.

1.5 How SVLMs Work: The Training Secret

Model Type Typical Parameter Count Computational Needs Primary Use Case

VLM Billions (e.g., 50B+)

Requires massive,

specialized data centers

Cutting-edge research,

complex creative tasks

SVLM

Millions to low billions

(e.g., 1B to 10B)

Can run on modern

mobile devices

or standard computers

Edge computing,

speciﬁc enterprise

applications

Table 1.1: Table showing comparison between VLMs and SVLMs

1.6 Why Small Models are the Future

Table 1.1 shows the comparison of a typical VLM

vs SVLM.

Knowledge distillation leverages a two-model

system, often conceptualized using a ’teacher’ and

’student’ analogy:

1. The ’Teacher’ Model (VLM):

This is typically a massive, state-of-the-art

Large Visual Language Model (LVLM).

It is characterized by billions of parameters,

trained on enormous and diverse datasets

of raw images and text pairs over extensive

computational resources.

The teacher model’s ’knowledge’ isn’t

just its ﬁnal answer, but the nuanced

logic, intermediate representations, and soft

probability distributions it generates across

various tasks (e.g., classifying an object as

”90% cat, 10% dog” rather than just ”cat”).

This depth of understanding forms the core

curriculum for the student.

2. The ’Student’ Model (SVLM ):

This is the small, target SVLM—the model

designed for deployment on edge devices,

mobile applications, or in environments

with strict computational and memory

constraints.

It possesses a signiﬁcantly smaller

architecture with far fewer parameters,

which inherently makes it faster and more

memory-eﬃcient during inference.

The Distillation Process (Eﬃcient Knowledge

Transfer):

Instead of the traditional, laborious training

approach where the SVLM is trained from

scratch on vast, raw datasets, the SVLM

is trained under the direct tutelage of the

larger teacher model.

The core goal of distillation is to train

the student to mimic the decisions,

sophisticated internal representations,

and complex knowledge encoded within

the larger teacher model. The student is

primarily guided by the teacher’s outputs

(the ’distilled’ knowledge) rather than just

the raw labels of the original data.

This is accomplished by utilizing a

specialized distillation loss function. This

function penalizes the student model when

its predictions deviate from the teacher

model’s soft targets (i.e., its predicted

probability distributions) for a given input.

By aligning its internal workings with

the teacher’s nuanced outputs, the student

eﬀectively learns the ’wisdom’ of the

master.

It learns the shortcuts, the most important

conceptual features, and the eﬃcient

decision-making pathways that the teacher

model already took immense computational

eﬀort to ﬁgure out.

This highly optimized and focused training approach

allows the SVLM to retain a surprising, and often near-

equivalent, amount of the teacher model’s high-level

capability and performance across a range of visual-

language tasks. Critically, it achieves this without

the burden of the teacher’s massive size, enabling the

deployment of highly capable AI in resource-limited

settings and accelerating the pace of the AI revolution

on edge devices.

1.6

Why Small Models are the Future

For most real-world applications, large models are

overkill. SVLMs are perfectly suited for deployment

outside of huge data centers—a concept known as Edge

Computing.

1.6.1 On-Device Intelligence

This is perhaps the biggest advantage. An SVLM

can be entirely installed and run on a local device (the

”edge”), such as:

Smartphones: Enabling immediate, private

image analysis and captioning without needing

to send data to the cloud.

Smart Cameras: Allowing a security camera to

identify a package delivery in real-time, even if

the internet connection is slow or unavailable.

1.8 Conclusion

Wearable Devices: Providing instant information, like

identifying a foreign plant or translating a menu sign

through smart glasses.

1.6.2 Enhanced Data Privacy

When the AI model runs directly on your device,

your images and data never have to leave it to be

processed. This ”on-device processing” signiﬁcantly

enhances user privacy and is becoming a major selling

point for consumer electronics.

1.7 Real-World Applications of Small

Visual Language Models

The eﬃciency and local nature of SVLMs open

the door to countless practical uses:

1.7.1 Accessibility and Assistance

SVLMs are powerful tools for accessibility. For

instance, a small model running on a person’s phone

could:

Describe Environments: For people with visual

impairments, an SVLM can process a live camera

feed and verbally describe the scene: ”You are

standing next to a park bench; a dog is running

toward a red ball.”

Read Complex Documents: Instantly translate

and summarize documents like utility bills

or complex medication instructions by simply

pointing a camera at them.

1.7.2 Manufacturing and Quality Control

In a factory setting, SVLMs can be deployed on a

cheap, rugged computer to monitor production lines:

Defect Detection: A camera feed can be analyzed

in real-time to spot tiny imperfections in a product, such

as a scratch on a screen or a misaligned component.

Since the model is small, it can make decisions instantly

without causing bottlenecks.

1.7.3 Retail and Inventory Management

SVLMs are streamlining retail operations:

Stock Monitoring: Cameras in a grocery store

can use an SVLM to continually check shelves,

instantly recognizing when a product is low and

notifying staﬀ for restocking.

Self-Checkout Veriﬁcation: Ensuring

customers are scanning the correct items

by visually conﬁrming the item against its

description.

1.7.4 Smart Home Devices

Future smart home hubs will rely on SVLMs for

complex, local intelligence:

Contextual Understanding: Not just recognizing

a person, but understanding context. For example,

recognizing that a child has spilled a cup of milk and

generating an alert for assistance.

1.8 Conclusion

Small Visual Language Models represent a crucial

step toward ubiquitous AI. They address the practical

limitations of massive models—cost, speed, and

privacy—by providing highly capable intelligence in a

compact, eﬃcient package. As AI continues to evolve,

it’s not always the biggest models that will win, but

the smart, swift, and highly accessible SVLMs that

will embed intelligence into the very fabric of our

everyday devices, making the technology truly personal

and powerful.

About the Author

Dr.Arun Aniyan is leading the R&D for

Artiﬁcial intelligence at DeepAlert Ltd,UK. He comes

from an academic background and has experience

in designing machine learning products for diﬀerent

domains. His major interest is knowledge representation

and computer vision.

Introduction to Neuromorphic Computing-

Part II

by Blesson George

airis4D, Vol.4, No.3, 2026

www.airis4d.com

2.1 Introduction

Traditional computing systems have enabled

remarkable technological progress. However, modern

applications such as artiﬁcial intelligence, robotics,

and real-time sensing require systems that are not

only powerful but also energy eﬃcient and adaptive.

Conventional computers consume large amounts of

energy when running complex AI models, which

motivates researchers to explore alternative computing

paradigms.

The human brain performs learning, perception,

and decision-making using very little power.

Neuromorphic computing attempts to imitate these

biological principles to create more eﬃcient intelligent

systems.

2.2 Basic Idea of Neuromorphic

Computing

The human brain consists of neurons connected

through synapses. Neurons communicate using short

electrical pulses called spikes. Learning occurs when

the strength of synaptic connections changes over time.

Neuromorphic computing tries to reproduce

this mechanism using electronic circuits. Artiﬁcial

neurons process signals, and artiﬁcial synapses store

connection strengths. Unlike traditional computers that

continuously process data, neuromorphic systems are

event-driven and become active only when input events

occur.

This approach enables highly parallel computation

and improved energy eﬃciency.

2.3

Why Neuromorphic Computing is

Important

The growing interest in neuromorphic computing

comes from several limitations of conventional

computing.

2.3.1 Energy Consumption

Modern AI systems require large computational

resources and high power usage. In contrast, the human

brain operates at approximately 20 W while performing

complex tasks. Neuromorphic systems aim to achieve

similar eﬃciency through spike-based computation.

2.3.2 Von Neumann Bottleneck

In conventional computers, memory and

processing units are physically separated. Constant

data transfer between them creates delays and energy

loss. Neuromorphic architectures combine memory

and computation, reducing this bottleneck.

2.3.3 Real-Time Intelligence

Applications such as robotics and edge devices

require fast responses and local decision-making.

Neuromorphic systems support real-time processing

due to their parallel and event-driven operation.

2.6 Diﬀerence from Conventional Computing

2.4 Spiking Neural Networks and

Simple Neuron Model

Spiking Neural Networks (SNNs) are commonly

used in neuromorphic computing because they closely

resemble biological neural systems.

A simple model used to describe neuron behavior

is the Leaky Integrate-and-Fire (LIF) model.

dV (t)

= −V (t) +RI(t) (2.1)

Here,

V (t) is the membrane potential,

I(t) is the input current,

R represents membrane resistance,

τ is the membrane time constant.

When the membrane potential reaches a threshold

value

, the neuron generates a spike and the potential

is reset. This model captures the essential ﬁring

behavior of biological neurons and forms the foundation

of many neuromorphic systems.

2.5 Neuromorphic Chips

Neuromorphic chips are specialized processors

designed to implement brain-inspired computation

directly in hardware. Instead of executing instructions

sequentially, these chips contain large numbers of

artiﬁcial neurons and synapses that operate in parallel.

2.5.1 Working Principle

The basic operation of a neuromorphic chip

includes:

1. Receiving input signals,

2. Integrating signals within neurons,

3. Generating spikes when thresholds are reached,

4. Transmitting spikes to connected neurons,

5. Updating synaptic strengths during learning.

Because computation occurs only when spikes

are present, these chips consume much less energy

compared to conventional processors.

2.5.2 Examples of Neuromorphic Chips

IBM TrueNorth is one of the earliest large-scale

neuromorphic chips, containing around one million

artiﬁcial neurons. It demonstrates extremely low power

consumption and large-scale parallelism.

Intel Loihi is another important neuromorphic

processor that supports on-chip learning. It allows

adaptive behavior and real-time learning, making it

useful for research in robotics and intelligent systems.

2.6 Diﬀerence from Conventional

Computing

Neuromorphic computing diﬀers fundamentally

from traditional computing methods.

Traditional computers use clock-driven operation,

while neuromorphic systems are event-driven.

Memory and processing are separate in

conventional systems but integrated in

neuromorphic architectures.

Conventional AI simulates neural behavior

in software, whereas neuromorphic systems

implement it directly in hardware.

Neuromorphic systems provide low-power and

parallel processing.

These diﬀerences make neuromorphic computing

attractive for future intelligent systems.

2.7 Applications

Neuromorphic computing has potential

applications in several areas:

Robotics and autonomous systems

Speech and image recognition

Edge AI and Internet of Things (IoT)

Brain–machine interfaces

Adaptive sensing and control systems

In many of these applications, fast response and

low energy consumption are critical requirements.

2.10 Conclusion

2.8 Challenges

Despite its potential, neuromorphic computing

still faces challenges:

Lack of standard programming frameworks

Diﬃculty in training spiking neural networks

Hardware design complexity

Limited large-scale commercial adoption

Research continues to address these challenges

and improve usability.

2.9 Future Scope

Neuromorphic chips may play an important

role in next-generation artiﬁcial intelligence. Future

systems could combine conventional AI methods with

neuromorphic hardware to achieve both accuracy and

eﬃciency.

As research progresses, neuromorphic computing

may enable intelligent systems that learn continuously,

adapt to their environment, and operate with extremely

low power.

2.10 Conclusion

Neuromorphic computing represents a shift toward

brain-inspired computation. By using spike-based

processing and highly parallel architectures, it oﬀers an

energy-eﬃcient alternative to conventional computing.

Neuromorphic chips demonstrate how these ideas can

be implemented in hardware, opening new possibilities

for robotics, AI, and intelligent devices. Although still

in the early stages of development, this ﬁeld has strong

potential to shape the future of computing.

About the Author

Dr. Blesson George presently serves as

an Assistant Professor of Physics at CMS College

Kottayam, Kerala. His research pursuits encompass

the development of machine learning algorithms, along

with the utilization of machine learning techniques

across diverse domains.

Deep Learning: A Review

by Linn Abraham

airis4D, Vol.4, No.3, 2026

www.airis4d.com

3.1 Introduction

In this series we attempt a review of deep

learning. We start with deﬁning machine learning

and distinguishing it from existing methods. Then

we learn about the early techniques in machine and

about a special kind of ensemble learning technique

as a case study. We follow this by learning about one

of the most important algorithms in classical machine

learning called the Perceptron. We end the review by

discussing how Multi-Layer Perceptrons have led to

the primitive forms of today’s modern deep learning

architectures.

3.2 What is Machine Learning?

We usually use the term Machine Learning in the

context of referring to a class of computer algorithms

or programs. If so, we need to ask ourselves the

following question — What diﬀerentiates machine

learning algorithms or programs from those that existed

before. There exists a plethora of mathematical and

statistical techniques from before the time Machine

Learning became popular. Take for example the familiar

least-squares ﬁtting method. To answer this, I use two

well-known quotes in the domain of machine learning

or artiﬁcial intelligence.

“Machine learning is the subﬁeld of computer

science that gives computers the ability to learn

without being explicitly programmed.” — Arthur

Samuel.

“A computer program is said to learn from

experience E with respect to some class of tasks

T and performance measure P, if its performance

at tasks in T, as measured by P, improves with

experience E”. — Tom M. Mitchell.

Combining both the ideas, we see that in

comparison to algorithms which try to solve a problem

using ﬁrst principles or domain knowledge, machine

learning algorithms rely more on the data (or instances)

that we have may already have access to.

Another key diﬀerences is the idea of

generalization to unseen data. This is where one ﬁrst

encounters the idea of splitting data into the three sets,

namely training, validation and testing.

In machine learning, the objective is not merely to

minimize error on the observed dataset, but to minimize

error on unseen data drawn from the same underlying

distribution. Formally, while classical ﬁtting procedures

often minimize the empirical risk

R(f) =

∑

i=1

ℓ



f(x

), y



machine learning is fundamentally concerned with

minimizing the expected risk

R(f) = E

(x,y)∼D



ℓ



f(x), y



where

denotes the (unknown) data-generating

distribution.

Because

is inaccessible in practice, machine

learning algorithms must rely on ﬁnite samples

to approximate this objective. This concern with

generalization leads naturally to one of the deﬁning

methodological practices in machine learning: splitting

available data into separate subsets, commonly referred

to as training, validation, and testing sets. The training

set is used to ﬁt model parameters, the validation set to

tune hyperparameters or select among models, and the

3.3 Classical Machine Learning Techniques

test set to provide an unbiased estimate of generalization

performance.

In this sense, machine learning can be viewed as

an intersection of computer science, statistics, and

optimization, with a distinctive focus on scalable

algorithms, empirical evaluation, and predictive

accuracy. The ﬁeld is less concerned with

discovering closed-form solutions derived from domain

assumptions, and more concerned with building systems

that adapt as data grows and improves performance

through experience.

This perspective provides the foundation for the

techniques discussed in the remainder of this chapter,

from early “classical” machine learning methods to

modern deep learning models.

3.3 Classical Machine Learning

Techniques

Early machine learning research focused on a

class of methods that are now commonly referred to

as classical or shallow machine learning techniques.

These methods typically rely on explicitly designed

feature representations and relatively simple model

architectures, in contrast to modern deep learning

approaches that learn hierarchical representations

automatically from raw data.

Some of the most inﬂuential classical machine

learning techniques include support vector machines,

decision trees, naive Bayes classiﬁers, and ensemble

methods such as boosting and bagging. Despite

their diversity, these techniques share a common goal:

learning a function that generalizes well from observed

examples to unseen data.

3.3.1 Gradient Boosting and XGBoost

Gradient Boosting generalizes the boosting

framework underlying AdaBoost by framing the

learning process as an optimization problem in function

space. Rather than explicitly adjusting instance weights,

Gradient Boosting sequentially ﬁts weak learners to the

negative gradient of a speciﬁed loss function, allowing

greater ﬂexibility in loss design and regularization

compared to AdaBoost. While AdaBoost is commonly

implemented using decision stumps, Gradient Boosting

typically employs shallow decision trees with controlled

depth, which can capture feature interactions.

Extensions such as XGBoost further enhance

Gradient Boosting through an optimized and scalable

implementation. XGBoost incorporates second-order

gradient information, sparsity-aware tree construction,

and approximate split-ﬁnding strategies, enabling

eﬃcient training on large and sparse datasets.

3.3.2 The Perceptron

The perceptron occupies a central place in the

history of machine learning, serving both as one of

the earliest learning algorithms and as the conceptual

foundation for modern neural networks.Introduced in

the late 1950s, the perceptron was among the ﬁrst

models capable of learning directly from data through

an iterative update rule, rather than relying on ﬁxed,

hand-crafted decision rules.

At its core, the perceptron is a binary linear

classiﬁer. Given an input vector

x ∈ R

, the model

computes a weighted sum of the inputs and applies a

threshold to produce an output:

y = sign(w

⊺

x + b),

where

w ∈ R

is a vector of learnable weights and

b ∈ R

is a bias term. This formulation makes explicit

the perceptron’s role as a linear decision function: the

model partitions the input space using a hyperplane

deﬁned by the parameters w and b.

What distinguishes the perceptron from earlier

linear models is not its functional form, but its learning

rule. Given a labeled training example

, y

)

with

∈ {−1, +1}

, the perceptron updates its parameters

only when it makes a mistake. The update rule can be

written as

w ← w + η y

, b ← b +η y

where

η > 0

is the learning rate. This simple rule

embodies the idea of learning from experience: the

model adjusts its parameters incrementally in response

to errors, improving performance over time.

Nevertheless, the perceptron remains critically

REFERENCES

important as a conceptual building block. By

stacking multiple perceptrons and introducing nonlinear

activation functions, one obtains multi-layer perceptrons

(MLPs), which are capable of representing highly

complex functions. In this sense, modern deep learning

models can be viewed as extensions of the basic

perceptron idea, augmented with depth, nonlinearity,

and large-scale optimization techniques.

The perceptron thus serves as a bridge between

classical machine learning and deep learning. It retains

the interpretability and simplicity of early models

while introducing the core mechanisms—parameterized

functions and data-driven learning—that underpin

contemporary neural network architectures.

3.4 Deep Learning

A direct extension of the perceptron is the multi-

layer perceptron (MLP), which consists of multiple

layers of neurons arranged in a feedforward architecture.

Each layer applies an aﬃne transformation followed by a

nonlinear activation function. For an MLP with

layers,

the forward computation can be written recursively as

(0)

= x

and

(l)

= σ W

(l)

(l−1)

(l)

, l = 1, . . . , L,

where

(l)

and

(l)

denote the weights and biases of

layer

, and

σ(⋅)

is a nonlinear activation function such

as the sigmoid, hyperbolic tangent, or rectiﬁed linear

unit.

The introduction of nonlinear activation functions

is essential. In the absence of nonlinearity, the

composition of multiple linear layers would reduce

to a single linear transformation, oﬀering no additional

expressive power beyond that of a single-layer model.

Nonlinearity enables deep networks to represent

complex functions and capture intricate structure in

data.

Training deep neural networks involves adjusting

their parameters to minimize a loss function that

quantiﬁes the discrepancy between predictions and

ground-truth labels. Given a dataset

{(x

, y

)}

i=1

and

a model

parameterized by

, the learning objective

is typically written as

min

∑

i=1

ℓ (f

), y

where

ℓ(⋅, ⋅)

denotes an appropriate loss function, such

as mean squared error for regression or cross-entropy

loss for classiﬁcation.

The optimization of this objective is made feasible

by the backpropagation algorithm, which eﬃciently

computes gradients of the loss with respect to all

model parameters using repeated applications of the

chain rule. These gradients are then used by gradient-

based optimization methods, such as stochastic gradient

descent and its variants, to iteratively update the

parameters.

A deﬁning characteristic of deep learning is its

ability to perform representation learning. Rather

than relying on manually engineered features, deep

models automatically learn representations that become

increasingly abstract across successive layers. Lower

layers typically capture simple patterns, while higher

layers encode more complex and task-speciﬁc features.

Depth plays a central role in this process. From a

theoretical perspective, deep architectures can represent

certain classes of functions more eﬃciently than shallow

ones. From a practical standpoint, depth allows models

to reuse and recombine intermediate representations,

improving both expressiveness and generalization.

Despite their success, deep learning models

introduce new challenges. They often require large

datasets, substantial computational resources, and

careful regularization to avoid overﬁtting. Techniques

such as dropout, weight decay, and batch normalization

are commonly employed to stabilize training and

improve generalization performance.

References

[1]

I. Goodfellow, Y. Bengio, and A. Courville, Deep

Learning. MIT Press, Cambridge, MA, 2016.

[2]

C. M. Bishop, Pattern Recognition and Machine

Learning. Springer, New York, NY, 2006.

REFERENCES

[3]

S. Marsland, Machine Learning: An Algorithmic

Perspective, 2nd ed. Chapman and Hall/CRC,

Boca Raton, FL, 2014.

About the Author

Linn Abraham is a researcher in Physics,

specializing in A.I. applications to astronomy. He is

currently involved in the development of CNN based

Computer Vision tools for prediction of solar ﬂares

from images of the Sun, morphological classiﬁcations

of galaxies from optical images surveys and radio

galaxy source extraction from radio observations.

Part II

Astronomy and Astrophysics

Plasma Physics- Magnetosphere & Ionosphere

by Abishek P S

airis4D, Vol.4, No.3, 2026

www.airis4d.com

1.1 Introduction

The magnetosphere is a vast, invisible region

surrounding Earth where its magnetic ﬁeld dominates

and interacts with charged particles from the Sun.

It is generated by the planet’s internal dynamo, the

movement of molten, electrically conducting material

in the outer core of the planet which produces a magnetic

ﬁeld resembling that of a bar magnet near the surface.

However, this ﬁeld is distorted farther out by the

solar wind, a continuous stream of charged particles

emitted by the Sun. On the side facing the Sun, the

magnetosphere is compressed, while on the opposite

side it stretches into a long tail called the magnetotail.

Its boundaries include the bow shock, where the solar

wind ﬁrst encounters resistance, and the magnetopause,

where the pressure of the solar wind balances Earth’s

magnetic ﬁeld [1].

This protective bubble plays a crucial role in

shielding Earth from harmful solar and cosmic radiation,

preventing atmospheric erosion, and making the

planet habitable. It also traps charged particles in

radiation belts and funnels some toward the poles,

producing auroras. The magnetosphere is highly

dynamic, constantly reshaped by solar activity, and

during strong solar storms it can be compressed,

leading to geomagnetic disturbances that aﬀect satellites,

power grids, and communications. Other planets

also have magnetospheres, though their strength and

structure vary[2]. Jupiter’s magnetosphere is the largest

in the solar system, while Mars largely lacks one,

contributing to its thin atmosphere. In a broader cosmic

perspective, magnetospheres are fundamental structures

that govern how celestial bodies interact with their space

environment, acting as shields and mediators between

planetary systems and the energetic universe around

them.

1.2 The Magnetosphere as a Plasma

Environment

The magnetosphere, when viewed as a plasma

environment, is far more than just a magnetic bubble

around Earth. It is a dynamic system ﬁlled with

charged particles from both the solar wind and Earth’s

ionosphere. These plasma populations interact with the

magnetic ﬁeld in complex ways, giving rise to processes

that drive space weather and inﬂuence conditions on

Earth. One important plasma population comes from

the solar wind. Protons and electrons from the Sun

continuously stream toward Earth, and some of them

penetrate the magnetosphere through a process called

magnetic reconnection at the magnetopause. This

entry point allows solar particles to mix with Earth’s

magnetic environment, fuelling storms and disturbances.

Alongside these, heavy ions such as oxygen (O

) and

helium (He

) originating from Earth’s ionosphere are

lifted upward by electric ﬁelds and waves. Once

energized, they contribute to the overall plasma density

and dynamics of the magnetosphere, especially during

geomagnetic storms[3,4].

Magnetic reconnection is a key process in this

environment. It occurs when oppositely directed

magnetic ﬁeld lines break and reconnect, releasing

stored magnetic energy and converting it into particle

kinetic energy. This drives large-scale convection of

1.4 Coupling Mechanisms

plasma throughout the magnetosphere and triggers

substorms, which are sudden disturbances that can

intensify auroras and disrupt communication systems.

Reconnection is often described as the “engine” of

magnetospheric dynamics because it powers so many

of the system’s changes[3].

Alfv

en waves are another crucial mechanism.

These are oscillations that travel along magnetic

ﬁeld lines, carrying energy and ﬁeld-aligned currents

between the magnetosphere and ionosphere. They act

as conduits for energy transfer, enabling the coupling

of Earth’s upper atmosphere with space. Through these

waves, disturbances in the magnetosphere can directly

inﬂuence ionospheric currents and auroral activity,

making them a vital link in the chain of magnetospheric

processes[5].

Ring current is a large-scale plasma structure

formed by energized ions circulating around Earth.

During geomagnetic storms, these ions build up

and create a current that encircles the planet. The

ring current contributes signiﬁcantly to geomagnetic

disturbances by altering Earth’s magnetic ﬁeld, which

can weaken the overall ﬁeld strength temporarily[6].

This weakening is one of the hallmarks of geomagnetic

storms and has practical consequences, such as aﬀecting

satellite operations and navigation systems. Magnetic

reconnection, Alfv

en waves, and ring current formation

are central to its behaviour, making the magnetosphere

a fascinating and complex plasma laboratory in space.

1.3 The Ionosphere as a Plasma

Source and Sink

The ionosphere is a dense plasma layer of Earth’s

upper atmosphere, created when solar ultraviolet (UV)

and X-ray radiation ionizes neutral atoms and molecules.

This region acts both as a source of plasma for the

magnetosphere and as a sink where magnetospheric

particles deposit their energy. Its dual role makes it a

critical part of the coupled Earth–space environment.

As a plasma source, the ionosphere provides

heavy ions such as oxygen (O

), which are accelerated

upward by electric ﬁelds, wave–particle interactions,

and heating processes. These ions escape into the

magnetosphere, where they contribute to plasma

populations in the plasma sheet and ring current.

During geomagnetic storms, this outﬂow intensiﬁes,

and O

ions become a dominant component, altering the

dynamics of the magnetosphere. In addition to heavy

ions, lighter ions like hydrogen (H

) and helium (He

)

continuously ﬂow outward in what is known as the

polar wind. This steady stream of light ions represents

a constant supply of plasma to the magnetosphere, even

under quiet conditions.

As a plasma sink, the ionosphere absorbs energy

and particles from the magnetosphere. Precipitating

electrons and ions collide with neutral atoms in the

upper atmosphere, producing auroras, the shimmering

lights seen near polar regions. These collisions also

enhance ionization in localized regions, modifying

the conductivity of the ionosphere and inﬂuencing

currents that ﬂow between the magnetosphere and

Earth. This feedback loop ensures that the ionosphere

is not only a supplier of plasma but also a recipient of

magnetospheric energy[7].

Plasma transport processes highlight the

ionosphere’s dynamic role. The polar wind represents a

continuous, low-level outﬂow of light ions, maintaining

a background plasma supply. In contrast, storm-time

outﬂows are episodic and powerful, with enhanced

ﬂuxes dominating the plasma sheet and ring

current during geomagnetic storms. These storm-

driven outﬂows signiﬁcantly reshape the magnetosphere,

fuelling geomagnetic disturbances and contributing to

the weakening of Earth’s magnetic ﬁeld during storms.

1.4 Coupling Mechanisms

Field-aligned currents (FACs) are one of the

most important coupling mechanisms between the

magnetosphere and ionosphere. These electric currents

ﬂow directly along magnetic ﬁeld lines, linking the

dynamics of the magnetosphere to the conductivity of

the ionosphere. When solar wind and magnetospheric

processes drive changes in the magnetic ﬁeld, FACs

carry those changes down into the ionosphere, where

they inﬂuence ionospheric currents and heating[8]. In

1.6 Magnetosphere & Ionosphere in Research Perspective

this way, FACs act as the “wires” of the system, ensuring

that energy and momentum are transferred between

space and Earth’s upper atmosphere.

Auroral acceleration regions represent another

key coupling process. In these regions, Alfv

enic

turbulence and parallel electric ﬁelds accelerate

electrons downward along magnetic ﬁeld lines. As

these electrons collide with neutral atoms in the upper

atmosphere, they produce the bright auroral arcs that

we see near the poles[5]. The acceleration process is

not uniform; instead, it creates structured, shimmering

displays that reﬂect the turbulent nature of the plasma

environment. These regions are therefore critical in

transforming magnetospheric energy into visible auroral

phenomena, while also enhancing ionization in the

ionosphere.

The pressure cooker eﬀect is a more subtle but

equally important mechanism. Downward currents

compress plasma at low altitudes, creating localized

regions of high pressure. This compression energizes

heavy ions such as O

, driving them upward into

the magnetosphere. In eﬀect, the ionosphere acts

like a plasma reservoir that can be “pumped” by

magnetospheric currents. During geomagnetic storms,

this eﬀect becomes especially pronounced, with large

ﬂuxes of O

ions escaping upward to dominate the

plasma sheet and ring current[6]. This upward transport

of ions alters the composition and dynamics of the

magnetosphere, feeding back into the larger system.

Together, these coupling mechanisms illustrate

how the magnetosphere and ionosphere are not isolated

systems but deeply interconnected. FACs provide the

electrical pathways, auroral acceleration regions convert

energy into light and ionization, and the pressure

cooker eﬀect drives plasma upward. Each process

contributes to the continuous exchange of energy and

particles, making Earth’s near-space environment a

highly dynamic and interactive plasma system.

1.5 Energy Flow

Solar wind energy enters Earth’s magnetosphere

primarily through magnetic reconnection at the dayside

magnetopause. This process allows the solar wind’s

magnetic ﬁeld to merge with Earth’s, opening pathways

for energy and plasma to ﬂow into the magnetosphere.

Once inside, this energy does not remain localized

but is redistributed through several interconnected

mechanisms that shape the dynamics of Earth’s near-

space environment[2].

One major redistribution process is

magnetospheric convection, which refers to the

large-scale circulation of plasma throughout the

magnetosphere. Driven by reconnection, plasma ﬂows

from the dayside magnetopause into the magnetotail

and then returns toward Earth[1]. This circulation

pattern transports energy and particles across vast

distances, fuelling substorms and setting the stage for

auroral activity. Convection ensures that solar wind

energy is spread throughout the magnetosphere rather

than concentrated in one region.

Wave–particle interactions provide another

pathway for energy redistribution. Waves such as Alfv

waves and whistler-mode waves propagate through the

magnetosphere, transferring energy to charged particles.

Alfv

en waves, in particular, carry ﬁeld-aligned currents

that couple the magnetosphere to the ionosphere,

enabling energy to ﬂow downward. Whistler-mode

waves can scatter energetic electrons, redistributing

their energy and sometimes precipitating them into

the atmosphere[5]. These interactions are essential

for regulating particle populations and maintaining the

dynamic balance of the magnetosphere.

Particle precipitation represents a direct way in

which magnetospheric energy is deposited into Earth’s

atmosphere. Electrons accelerated along magnetic

ﬁeld lines collide with neutral atoms in the upper

atmosphere, producing auroras. This process not

only creates the spectacular light displays near the

poles but also enhances ionization in the ionosphere,

altering its conductivity and inﬂuencing global current

systems. Through precipitation, magnetospheric

energy is transformed into both visual phenomena and

atmospheric changes.

1.6 Magnetosphere & Ionosphere in Research Perspective

1.6 Magnetosphere & Ionosphere in

Research Perspective

Plasma physicists regard the magnetosphere–

ionosphere system as a natural laboratory for studying

universal plasma processes, because it exhibits many of

the same behaviours seen in astrophysical environments

across the universe. One of the most important aspects

is turbulence. In the magnetosphere, plasma ﬂows

and magnetic reconnection events generate turbulent

structures, where energy cascades from large scales

down to smaller ones. This turbulence is similar to

what occurs in solar corona plasmas, astrophysical jets,

and even interstellar medium turbulence, making near-

Earth space an accessible place to study these otherwise

remote phenomena.

Instabilities are another key feature. The

magnetosphere is full of plasma instabilities, such as

the Kelvin–Helmholtz instability at the magnetopause,

which develops when solar wind ﬂows shear against

Earth’s magnetic boundary[9]. Other instabilities, like

those in the plasma sheet, can trigger substorms and

auroral activity. These processes mirror instabilities

found in fusion devices and astrophysical disks, oﬀering

scientists a chance to test theories of how plasmas

behave when disturbed.

Nonlinear coupling is also central to this

system. The magnetosphere and ionosphere are tightly

linked through ﬁeld-aligned currents, wave–particle

interactions, and feedback loops. Small changes in one

region can lead to large-scale responses in the other,

demonstrating nonlinear dynamics. This coupling is a

microcosm of the complex interactions seen in larger

astrophysical systems, such as star–planet interactions

or accretion ﬂows around black holes.

What makes near-Earth space especially valuable

is the ability to obtain in situ measurements. Satellites

and ground-based instruments can directly measure

plasma density, temperature, particle ﬂuxes, and wave

spectra. These observations provide real-time data

that allow scientists to test and reﬁne theories of

plasma circulation, turbulence, and energy transfer.

Unlike distant astrophysical plasmas, which can only

be observed indirectly through light or radiation, the

magnetosphere–ionosphere system oﬀers a hands-on

testbed for plasma physics.

In essence, the magnetosphere–ionosphere system

demonstrates turbulence, instabilities, and nonlinear

coupling in ways that are universal to plasma

environments across the cosmos. By studying it with

direct measurements, plasma physicists gain insights

that apply not only to Earth’s space weather but also to

the broader universe, from solar ﬂares to galactic jets.

References

[1] Borovsky, J. E., & Valdivia, J. A. (2018). “The

Earth’s magnetosphere: A systems science overview

and assessment”. Surveys in geophysics, 39(5), 817-

859.

[2] Cahill, L. J. (1965). “The magnetosphere”.

Scientiﬁc American, 212(3), 58-71.

[3] Cowley, S. W. H. (1991). “The plasma

environment of the Earth”. Contemporary physics,

32(4), 235-250..

[4] Pfaﬀ Jr, R. F. (2012). “The near-Earth plasma

environment”. Space science reviews, 168(1), 23-112.

[5] Chen, L., & Zonca, F. (2016). “Physics

of Alfv

en waves and energetic particles in burning

plasmas”. Reviews of Modern Physics, 88(1), 015008.

[6] Daglis, I. A. (2006). “Ring current dynamics”.

Space science reviews, 124(1), 183-202.

[7] Chappell, C. R., Moore, T. E., & Waite Jr, J.

H. (1987). “The ionosphere as a fully adequate source

of plasma for the Earth’s magnetosphere”. Journal of

Geophysical Research: Space Physics, 92(A6), 5896-

5910.

[8] L¨uhr, H., & Kervalishvili, G. (2021). “Field-

aligned currents in the magnetosphere–ionosphere”.

Magnetospheres in the solar system, 193-205.

[9] Hasegawa, A. (1971). “Plasma instabilities

in the magnetosphere”. Reviews of Geophysics, 9(3),

703-772.

1.6 Magnetosphere & Ionosphere in Research Perspective

About the Author

Abishek P S is a Research Scholar in

the Department of Physics, Bharata Mata College

(Autonomous) Thrikkakara, kochi. He pursues

research in the ﬁeld of Theoretical Plasma physics.

His works mainly focus on the Nonlinear Wave

Phenomenons in Space and Astrophysical Plasmas.

X-ray Astronomy: Theory

by Aromal P

airis4D, Vol.4, No.3, 2026

www.airis4d.com

2.1 Introduction

In our previous article, we discussed the various

sources of cosmic X-rays and the mechanisms that

produce them. We will continue with a series exploring

these diﬀerent sources in detail. As the author

specializes in X-ray binaries, we will begin with an

in-depth look at their various classes before moving on

to other topics.

As the name suggests, X-ray binaries are double-

star systems that emit electromagnetic waves in the

X-ray region. To put this in perspective, consider our

own Sun. It is an average main-sequence star whose

thermal emission peaks in the visible range, with a

surface temperature of around 5,500 K. Producing

thermal X-rays, however, requires temperatures of

millions of Kelvin! It feels unbearably hot in summer,

even though we are 8.3 light-minutes away from the

Sun. So, imagine the intensity of an environment

with temperatures reaching millions or even billions of

Kelvin! X-ray binaries are home to these incredibly

energetic phenomena.

What is the driving force behind this? Surprisingly,

it is the weakest of all fundamental forces—gravity. This

is the remarkable story of how gravity, the universe’s

most delicate force, gives rise to some of its most

powerful and energetic displays.

2.2 X-ray Binaries

Gravity achieves this spectacular heating through

a mechanism called accretion, the continuous ﬂow of

matter from a companion star into the deep gravitational

well of a dense, compact object. Just as water releases

energy as it plunges down a waterfall, the gas in these

systems converts its enormous gravitational potential

energy into intense heat and radiation as it falls.

To understand the math behind this sheer power, we

can look at the fundamental formula for the luminosity

(or power output), L, generated by this accretion:

L = GM

MR. (2.1)

In this equation, G represents the gravitational constant,

M is the mass of the compact object,

is the mass

accretion rate (how much matter falls per second), and

R is the radius of the compact object. Notice that

the radius R is in the denominator. Because compact

objects are incredibly small, the denominator is minimal,

making the resulting X-ray luminosity very large.

We can also approximate this power using Albert

Einstein’s famous equation for rest-mass energy,

E =

. The luminosity can be expressed as a fraction

of the total energy the falling matter possesses:

L =

, where

represents the eﬃciency of the energy

conversion

(η = GMRc

)

. For nuclear fusion the

exact process that makes our Sun shine this eﬃciency

is a mere 0.001 to 0.01 (that is only

0.1%

the mass is converted into energy). In contrast, the

eﬃciency

for matter accreting onto a neutron star

is roughly

0.1

(that is

10%

), and for a black hole, it

can reach an astonishing

0.42

(

42%

). This proves

that gravitational accretion onto compact objects is the

most eﬃcient mechanism known in the universe for

converting matter into energy, vastly outperforming

nuclear reactions. Just a handful of matter can produce

an X-ray luminosity 10,000 times greater than the total

2.2 X-ray Binaries

energy output of the Sun at all wavelengths.

What is at the core of these extreme systems

that generate such a strong gravitational pull? The

engines behind them are compact objects, which are the

collapsed remnants of former stars, the ghost of a dead

star. These remnants appear when a star dies. There

are generally three forms of these objects, depending

on the original mass of the dying star. It’s important to

outline these forms to provide a complete discussion of

the topic.

2.2.1 White Dwarfs

When an average star, similar to our Sun, exhausts

its nuclear fuel, its core can no longer generate the

outward pressure needed to support its own weight.

The core collapses until it is halted by the quantum

resistance of crowded electrons, a phenomenon known

as degenerate electron pressure. The resulting object is

a white dwarf. A white dwarf packs roughly the mass

of the Sun (

∼ 1 M⊙

) into a sphere about the size of

the Earth, which has a radius of approximately 10,000

kilometres. White dwarfs are strictly limited in weight

and cannot exceed

1.44 M ⊙

, a boundary known as the

Chandrasekhar limit. Systems with accreting white

dwarfs are called cataclysmic variables. Because white

dwarfs have a relatively large radius compared to other

compact objects, their gravitational wells are shallower,

making their accretion process much less luminous in

X-rays.

2.2.2 Neutron Stars

When a much more massive star (greater than

10 M⊙

) reaches the end of its life, its core collapses

so violently that even electron pressure cannot stop the

crush of gravity. In a fraction of a second, electrons and

protons are squeezed together to form a ﬂuid of neutrons.

This triggers a catastrophic supernova explosion that

blows away the outer star, leaving behind a neutron

star. A neutron star is a bizarre state of matter; it

compresses a mass equivalent to the Sun’s into a sphere

with a radius of just 10 to 15 kilometers, roughly the

size of a small city. Its density is equal to that of

an atomic nucleus, so dense that a chunk the size of

a table spoon of matter would weigh more than the

mass of earth. Because its radius R is so tiny, the

accretion formula

L = GM

MR

dictates that in falling

matter reaches immense velocities, heating up to tens

of millions of degrees and radiating copious amounts

of X-rays. Typical neutron stars have a mass of around

1.4 M ⊙, with theoretical maximum limits near 3 M ⊙.

2.2.3 Black Holes

For the most massive stars in the universe(

25 M ⊙

), the core collapse is so profound that even the

rigid pressure of compressed neutrons cannot halt the

crushing force of gravity. The core collapses entirely

upon itself, creating a black hole. A black hole is

characterized by an event horizon—a boundary beyond

which the gravitational pull is so intense that nothing,

not even light, can escape. While black holes do not

have a hard physical surface like a neutron star, we

can still observe them as brilliant X-ray sources. The

intense X-ray emission we see actually comes from the

superheated accretion disk of matter swirling frantically

just outside the event horizon before it crosses the

boundary. Black holes are also of diﬀerent sizes, and

here we are discussing about stellar-mass black holes

in X-ray binaries which typically have masses ranging

from greater than

3 M ⊙

up to about

15 M ⊙

.. Do not

get confused with the supermassive black holes that

used to be found in the center of the galaxies we will

discuss about then in the upcoming articles.

As we have discussed about the compact object

its a good time to discuss about the diﬀerent classes

of X-ray binaries usually called as XRBs. Based on

the mass of the companion star we can classify the

X-ray binaries into two as Low-mass X-ray binaries and

High-mass X-ray binaries.

2.2.4 Low-mass X-ray Binaries

In an LMXB, the companion star supplying the

fuel is an older, late-type star that is generally less

massive than our own Sun, or even a tiny degenerate

dwarf. Because these companion stars are much older

and have long lifespans, LMXBs are typically found

residing in regions populated by old stars, such as the

2.2 X-ray Binaries

central Galactic Bulge of the Milky Way and the dense,

crowded cores of globular clusters.

The mass transfer mechanism in an LMXB relies

on the evolutionary expansion of this small companion

star. As the star ages, it pulls outward until it

completely ﬁlls its gravitational boundary, a pear-shaped

equipotential surface known as a Roche lobe. Once it

ﬁlls this lobe, the star’s outer gaseous layers physically

spill over a gravitational saddle point between the two

stars—called the inner Lagrangian point (

) and fall

toward the compact object. Because the two stars

are orbiting each other, this transferred matter carries

signiﬁcant orbital angular momentum. Therefore, it

cannot plummet straight down onto the compact object.

Instead, it spirals inward to form a ﬂat, superheated,

rapidly spinning structure called an accretion disk.

Internal friction and viscosity within this disk heat

the gas to millions of degrees, producing the bright

X-rays we observe before the matter makes its ﬁnal

plunge.

2.2.5 High-mass X-ray Binaries

In these systems, the companion star is a massive,

young, early-type star (such as an O or B spectral type)

that can weigh 10 times the mass of the Sun or more.

Because these massive stars burn through their nuclear

fuel rapidly and live very short, furious lives, HMXBs

are typically found in the spiral arms of our galaxy, close

to the stellar nurseries where they recently formed.

Unlike the older, smaller stars in LMXBs that

gently spill their mass over a gravitational boundary,

these massive hot stars naturally blow oﬀ a highly

energetic stellar wind of gas similar to the solar storms

observed in sun. This wind is driven outward by the

intense pressure of the star’s own ultraviolet radiation,

and it carries oﬀ a huge amount of material in all

directions. As the compact object orbits through this

dense, outﬂowing hurricane of gas, its intense gravity

acts like a cosmic vacuum cleaner. It captures a fraction

of the passing gas in a process mathematically described

as Bondi-Hoyle accretion. This wind-fed material falls

onto the compact object, generating a powerful shock

front and emitting X-rays as it crashes down. In some

speciﬁc HMXBs, especially those containing rapidly

spinning ”Be” stars, the compact object travels on a

highly elliptical orbit and periodically plunges through

a dense equatorial disk of ejected material surrounding

the massive star, causing bright, repeating transient

outbursts of X-ray energy.

Is accretion the only process responsible for

producing X-rays in X-ray binary systems? Is there

something more? To answer this question, we need to

discuss these systems in detail, which we will continue

in the upcoming article.

Reference:

Frederick D. Seward, Philip A. Charles Exploring

the X-ray Universe

Keith A. Arnaud, Randall K. Smith, and Aneta

Siemignowska Handbook of X-ray Astronomy

H. Schatz a and K. E. Rehm X-ray binaries

About the Author

Aromal P is a research scholar in

Department of Astronomy, Astrophysics and Space

Engineering (DAASE) in Indian Institute of

Technology Indore. His research mainly focuses on

studies of Thermonuclear X-ray Bursts on Neutron star

surface and its interaction with the Accretion disk and

Corona.

Giant Molecular Clouds: Structure,

Evolution, and Their Role in Star Formation

by Sindhu G

airis4D, Vol.4, No.3, 2026

www.airis4d.com

3.1 Introduction

The space between stars in galaxies is not empty

but ﬁlled with gas, dust, cosmic rays, and magnetic

ﬁelds, collectively known as the interstellar medium

(ISM). Among its various components, molecular

clouds are the densest and coldest regions. Giant

Molecular Clouds (GMCs) are the largest of these

structures and serve as stellar nurseries where new stars

and planetary systems are formed.

In spiral galaxies such as the Milky Way, GMCs

are primarily located along spiral arms. These regions

experience enhanced gas compression, which promotes

the formation of molecular hydrogen. GMCs typically

contain tens of thousands to several million times the

mass of the Sun and extend over distances ranging from

a few tens to a few hundred parsecs. Their extremely

low temperatures, generally between 10 and 20 Kelvin,

create ideal conditions for gravitational collapse and

star formation.

3.2 Chemical Composition and

Physical Characteristics

3.2.1 Molecular Content

The dominant component of GMCs is molecular

hydrogen, which constitutes roughly three-quarters of

their total mass. Helium accounts for most of the

remaining mass, while heavier elements and dust grains

contribute a small but important fraction.

Although molecular hydrogen is the most abundant

molecule, it is diﬃcult to observe directly because it

does not emit radiation eﬃciently at low temperatures.

Astronomers therefore use carbon monoxide as a

tracer molecule. Emission from carbon monoxide

allows researchers to estimate the amount of molecular

hydrogen present within a cloud.

Dust grains play an essential role in molecular

cloud chemistry. They shield molecules from

destructive ultraviolet radiation and provide surfaces on

which molecular hydrogen can form. In addition, dust

grains emit infrared radiation, which helps regulate the

thermal balance of the cloud.

3.2.2 Temperature and Density

GMCs are extremely cold compared to most

astrophysical environments. Their temperatures

typically range from 10 to 20 Kelvin. At such low

temperatures, thermal motion of particles is weak,

allowing gravity to inﬂuence the cloud’s evolution more

eﬀectively.

The average particle density in a GMC is relatively

modest compared to terrestrial standards, but it is

signiﬁcantly higher than in the surrounding ISM. Within

these clouds, denser regions known as clumps and cores

can reach much higher densities. These dense cores are

the direct birthplaces of stars.

3.4 Internal Structure of GMCs

3.2.3 Mass and Size

GMCs are among the largest coherent structures

in galaxies. Their masses range from ten thousand to

several million solar masses. Despite their enormous

mass, they remain gravitationally delicate structures

due to their low temperatures and complex internal

dynamics.

Observations show that many GMCs share similar

surface densities, suggesting that common physical

processes govern their formation and evolution.

3.3 Formation of Giant Molecular

Clouds

The formation of GMCs is closely linked to large-

scale galactic processes.

3.3.1 Spiral Density Waves

In spiral galaxies, large-scale density waves move

through the galactic disk. As gas enters these regions,

it becomes compressed. This compression increases

density, allowing atomic hydrogen to transform into

molecular hydrogen. Over time, these compressed

regions grow into giant molecular clouds.

3.3.2 Gravitational Instabilities

If the gas density in a galactic disk becomes

suﬃciently high, gravitational forces can cause it

to fragment into large molecular complexes. These

instabilities help convert diﬀuse interstellar gas into

dense molecular clouds.

3.3.3 Stellar Feedback and Cloud Collisions

Massive stars inﬂuence their surroundings through

stellar winds and supernova explosions. These energetic

events can sweep up surrounding gas into expanding

shells. As these shells cool and accumulate material,

they may fragment and form new molecular clouds.

Additionally, collisions between smaller clouds can

compress gas and contribute to GMC formation.

3.4 Internal Structure of GMCs

Modern observations reveal that GMCs are far

from uniform. Instead, they exhibit a highly complex

and hierarchical structure.

3.4.1 Filaments

Filamentary structures are commonly observed

within molecular clouds. These elongated structures

often extend for several parsecs and contain chains

of dense cores. Filaments appear to be fundamental

building blocks of molecular clouds and play a crucial

role in star formation.

3.4.2 Clumps and Cores

Within ﬁlaments, material gathers into clumps and

cores. Clumps are intermediate-scale dense regions

that may form clusters of stars. Cores are smaller and

denser regions that typically give rise to individual stars

or small multiple systems.

The hierarchical organization of ﬁlaments, clumps,

and cores reﬂects the combined inﬂuence of turbulence

and gravity within the cloud.

3.5

Cloud Dynamics: Turbulence and

Magnetic Fields

3.5.1 Turbulence

Observations show that gas motions within GMCs

are highly turbulent and often supersonic. Turbulence

serves two important roles. On large scales, it

can support the cloud against complete gravitational

collapse. On smaller scales, it creates localized density

enhancements that promote star formation.

However, turbulence dissipates energy over time

and must be continuously replenished. Large-scale

galactic motions and feedback from young stars are

likely sources of this energy.

3.5.2 Magnetic Fields

Magnetic ﬁelds are present throughout the

interstellar medium and permeate GMCs. These ﬁelds

3.8 Observational Techniques

can provide additional support against gravitational

collapse. In some cases, magnetic forces inﬂuence the

orientation of ﬁlaments and regulate how material ﬂows

within the cloud.

The balance between gravity, turbulence, and

magnetic ﬁelds determines the eﬃciency and rate of

star formation.

3.6 Star Formation in Giant

Molecular Clouds

Star formation begins within dense cores when

gravity overcomes internal support mechanisms. As

a core collapses, material accumulates at the center,

forming a protostar. Surrounding material may form

an accretion disk, from which planets can eventually

emerge.

Most stars form in clusters rather than in isolation.

Massive GMCs can produce entire stellar associations

containing hundreds or thousands of stars.

Despite their large mass, GMCs convert only a

small fraction of their gas into stars. Radiation, stellar

winds, and eventual supernova explosions from massive

stars disperse the surrounding gas, limiting further star

formation.

3.7 Lifetimes and Evolution

GMCs are not permanent structures. Their

lifetimes are typically estimated to be between 10 and

30 million years. During this time, they evolve through

several stages:

1. Formation from diﬀuse interstellar gas.

Growth and fragmentation into ﬁlaments and

cores.

3. Active star formation.

4. Dispersal by stellar feedback.

The dispersal phase returns enriched material to the

ISM, contributing to the next generation of molecular

clouds.

3.8 Observational Techniques

Because GMCs are cold and dust-rich, they are

primarily studied at radio, infrared, and submillimeter

wavelengths.

Radio observations of carbon monoxide trace the

distribution of molecular gas. Infrared observations

reveal embedded young stellar objects and warm dust

emission. Submillimeter telescopes provide high-

resolution maps of dense cores and ﬁlaments.

Advances in observational facilities have

signiﬁcantly improved our understanding of molecular

cloud structure and star formation processes.

3.9 Role in Galactic Evolution

GMCs regulate the star formation rate within

galaxies. The continuous cycle of cloud formation, star

birth, and cloud dispersal drives chemical enrichment

and inﬂuences galactic structure.

Massive stars formed within GMCs eventually

explode as supernovae, injecting energy and heavy

elements into the ISM. This feedback process shapes

the future evolution of galaxies.

3.10 Conclusion

Giant Molecular Clouds are fundamental

components of galaxies and the principal sites of star

formation. Their low temperatures, large masses, and

complex internal structures create ideal environments

for gravitational collapse and stellar birth. Although

signiﬁcant progress has been made in understanding

their formation and evolution, ongoing research

continues to reﬁne our knowledge of these remarkable

cosmic structures.

Understanding GMCs provides essential insight

into the origin of stars, planetary systems, and the

broader evolution of galaxies.

References:

MOLECULAR CLOUDS

Protostars and Planets VI

3.10 Conclusion

Theory of Star Formation

Molecular Clouds in the Milky Way

Star formation in molecular clouds: observation

and theory

Star Formation in the Milky Way and Nearby

Galaxies

Formation of Molecular Clouds and Global

Conditions for Star Formation

About the Author

Sindhu G is a research scholar in the

Department of Physics at St. Thomas College,

Kozhencherry. She is doing research in Astronomy

& Astrophysics, with her work primarily focusing

on the classiﬁcation of variable stars using diﬀerent

machine learning algorithms. She is also involved

in period prediction for various types of variable

stars—especially eclipsing binaries—and in the study

of optical counterparts of X-ray binaries.

Part III

Biosciences

Kombucha: “The Symbiotic Elixir”

A Comprehensive View on its Bioprocessing

by Aengela Grace Jacob

airis4D, Vol.4, No.3, 2026

www.airis4d.com

1.1 Introduction

Kombucha, often referred to as the immortal health

elixir in ancient Chinese texts, is a fermented tea

beverage that has exploded in popularity across the

globe. Far more than just a trendy drink, Kombucha

is a complex, living ecosystem captured in a bottle. In

small-scale and home brewing, kombucha is typically

made in glass jars topped with fabric. Black or green

tea leaves are steeped in hot water with sugar, then

removed. When the sweetened tea has cooled, it is

mixed with a bit of kombucha from a previous batch

to make the liquid more acidic. A gelatinous mat of

symbiotic culture of bacteria and yeast (SCOBY) is

then added, and the brew is covered with a tight-weave

fabric or paper coﬀee ﬁlter and left to ferment at room

temperature for 7–30 days. This article delves into

the microbial composition, the remarkable symbiotic

culture that creates it, the intricate fermentation process,

and its powerful impact on human gut health.

1.2 SCOBY: The Heart of the Brew

The core of Kombucha is the SCOBY (Symbiotic

Culture of Bacteria and Yeast). Visually, it is a thick,

rubbery, beige disc. While often called a Kombucha

mushroom, it is actually a dense structure of cellulose

ﬁbers created by bacteria. The living components of

SCOBY can vary widely but generally include strains

of Saccharomyces cerevisiae and other yeasts, as well

as a number of bacteria, including Gluconacetobacter

Figure 1: Home made kombucha

Image Courtesy:https://www.healthygreenkitchen.com/how- to-make-kombucha-at-home/

xylinus. Fresh or dehydrated SCOBY can be bought

from suppliers, or a “mother” can be taken from a

previous batch of kombucha. In the fermentation

process, the alcohols produced by the yeasts are

converted by the bacteria into organic acids. The ﬁnal

kombucha product contains vitamin C, vitamins B6

and B12, thiamin, acetic acid, and lactic acid, as well

as small amounts of sugar and ethanol. The SCOBY

acts as the living engine that transforms sweetened tea

into an elixir, serving as both a fermenting agent and a

protective seal for the liquid.

1.3 Microbial Contents: The Living

Force

A single bottle of Kombucha is teeming with

billions of microorganisms. While the exact

1.4 The Fermentation Mechanism

Figure 2: SCOBY

Image Courtesy:https://escarpmentlabs.com/blogs/resources/the-road-to-scoby

Figure 3: SCOBY GROWTH

Image Courtesy: https://www.reddit.com/r/Kombucha/comments/glo9ez/scoby growth experimen

t varying steep times/

Figure 4: Magniﬁcation of Scoby at 400X by Benjamin

Wolfe

Image Courtesy:https://microbialfoods.org/science-digested-microbial- diversity-kombucha/

microbiome varies, the core players include:

Acetic Acid Bacteria (AAB): These produce the

deﬁning acetic acid (vinegar) and build the SCOBY’s

physical structure.

Yeast: These convert sugar into ethanol and carbon

dioxide (CO

), providing the f uel for the bacteria.

Lactic Acid Bacteria (LAB): These contribute

to the sour ﬂavor and enhance the drink’s probiotic

properties.

1.4 The Fermentation Mechanism

The creation of Kombucha is a synchronized

chemical dance consisting of two primary phases:

Phase 1:Primary Fermentation (Aerobic)

In an open container, yeast consumes sugar to

produce ethanol. Simultaneously, the Acetic Acid

Bacteria consume that ethanol and oxygen from the air

to produce acetic acid and gluconic acid. This stage

creates the tart ﬂavor proﬁle.

Phase 2: Secondary Fermentation (Anaerobic)

The liquid is bottled and sealed. Without oxygen,

the yeast continues to ferment the remaining sugar,

but the resulting

is trapped, creating natural

carbonation.

The fermentation mechanism operates through

a complex, synergistic interaction between various

microorganisms.

1.6 Conclusion

Figure 5: Fermentation Mechanism

: Image Courtesy:https://www.researchgate.net/figure/Types-of-fermentation-in- Kombucha-Creat

ed-with- BioRendercom fig1 365463218

Initial Hydrolysis (Yeast): Yeast (such as

Saccharomyces) produces the enzyme invertase,

which breaks down sucrose into glucose and

fructose. Secondly, alcoholic fermentation when

Yeast metabolizes the glucose and fructose to produce

ethanol and carbon dioxide (

), contributing

to the beverage’s carbonation. Thirdly, Acidic

fermentation Bacteria like Acetic acid bacteria (AAB,

e.g., Acetobacter and Gluconobacter) and Lactic Acid

Bacteria (LAB) consume the ethanol and sugars to

produce acetic acid, lactic acid, and gluconic acid.

Cellulose formation occurs when the AAB uses the

glucose to synthesize a thick cellulose bioﬁlm, which

is the SCOBY pellicle, acting as a barrier and enabling

oxygen exchange.

1.5 Major Beneﬁts on Gut Health

The most scientiﬁcally backed beneﬁts of

Kombucha revolve around the gastrointestinal system

which includes probiotic diversity as it restores natural

balance to gut ﬂora, aiding digestion and immune

function. The probiotics (such as Lactobacillus

and Biﬁdobacterium) can help restore balance to the

microbiome by suppressing the growth of harmful

bacteria. Fermented foods like kombucha are

recommended to help re-establish healthy gut bacteria

after a course of antibiotics disrupts the natural ﬂora.

Studies suggest that regular kombucha consumption

may modulate the gut microbiota in ways that

improve metabolic health, particularly in individuals

with obesity, by encouraging beneﬁcial bacteria like

Akkermansiaceae. Organic acids like acetic and

gluconic acids inhibit harmful pathogens and act as

postbiotics to support the gut lining. It provides an

antioxidant boost through fermentation as it increases

the bioavailability of tea polyphenols, which combat

inﬂammation in the digestive tract.

1.6 Conclusion

Kombucha is a testament to the power of microbial

symbiosis. Through the collaborative eﬀort of bacteria

and yeast, a simple pot of sweet tea is transformed

into a potent tool for human wellness. By providing a

rich source of probiotics and antioxidants, it remains a

timeless path toward internal balance.

REFERENCES

https://www.sciencedirect.com/science/article/

pii/S2772753X22000144

https://thegoodbug.com/blogs/news/kombucha

-fermentation-benefits-process

https://www.sciencedirect.com/science/article/

pii/S2772753X23002423

https://pmc.ncbi.nlm.nih.gov/articles/PMC965

8962/

About the Author

Aengela Grace Jacob is Final Year Student

of Bsc Biotechnology , Chemistry ( BSc BtC ) Dual

Major at Christ University Central campus, Bangalore

The Genomic Landscape of Major Depressive

Disorder

by Geetha Paul

airis4D, Vol.4, No.3, 2026

www.airis4d.com

2.1 Introduction

Major Depressive Disorder (MDD) is a globally

prevalent psychiatric condition characterised by

persistent low mood, anhedonia, and cognitive

dysfunction. Despite decades of clinical study, the

underlying aetiology of MDD remained largely elusive

until the advent of the genomic era. Traditional views

often sought a single depression gene, a silver bullet that

could explain the condition’s onset. However, modern

genetic research, particularly large-scale Genome-

Wide Association Studies (GWAS), has fundamentally

dismantled this notion. We now understand MDD

as a polygenic trait, meaning its genetic liability

is distributed across thousands of common genetic

variations, primarily single-nucleotide polymorphisms

(SNPs).

A SNP represents a single letter change in the

DNA sequence (e.g., a Cytosine replacing a Thymine).

Individually, these SNPs exert a negligible eﬀect on an

individual’s risk; when aggregated, they form a complex

genetic architecture that accounts for roughly 10–15%

of the total variance in depression liability (SNP-

based heritability). This polygenic nature explains the

high degree of heterogeneity seen in clinics: because

diﬀerent combinations of thousands of SNPs can lead

to a diagnosis, two patients may share the same label

but possess entirely diﬀerent biological drivers.

The role of SNPs in MDD detection is not

about a binary yes/no diagnosis but about quantifying

susceptibility and resolving the disorder’s cryptic

biological subtypes. Researchers utilise Polygenic

Risk Scores (PRS) as a standardised quantitative

metric. This score aggregates the weighted eﬀects

of thousands of single-nucleotide polymorphisms

(SNPs) discovered via large-scale Genome-Wide

Association Studies (GWAS). By calculating this

cumulative genetic burden, scientists can eﬀectively

stratify populations by risk level. This process identiﬁes

individuals with high latent vulnerability to Major

Depressive Disorder (MDD), even before clinical

symptoms manifest. This molecular approach bridges

the gap between subjective symptom reporting and

objective biological data, paving the way for a more

precise, stratiﬁed approach to mental health that

moves away from a one-size-ﬁts-all model toward

personalised psychiatry.

2.2 Quantifying Genetic Liability via

Polygenic Risk Scores (PRS)

Since no single gene causes depression, risk

detection relies on the cumulative signal from thousands

of SNPs. The Polygenic Risk Score (PRS) is the

primary tool used to measure this. Recent studies have

shown that individuals in the top 2.5% of the PRS

distribution have a signiﬁcantly higher risk (up to 5

times greater) of developing MDD than those with

average scores. PRS can identify high-risk individuals

before the ﬁrst clinical episode, enabling preemptive

monitoring and resilience-building interventions.

2.4 Pharmacogenomics and Treatment Response

2.3 Resolving Diagnostic

Heterogeneity

One of the most signiﬁcant roles of SNPs is

identifying cryptic subtypes of depression. While the

DSM-5 lists symptoms, SNPs reveal the underlying

biological pathways. Some patients carry SNPs

concentrated in serotonin/dopamine pathways (e.g.,

SLC6A4, DRD2), while others show a high burden

of SNPs in inﬂammatory genes (IL-6, CRP). Research

shows that MDD shares a p-factor (a general genetic

liability) with other disorders like Bipolar Disorder and

Schizophrenia, which SNPs help untangle.

2.4 Pharmacogenomics and

Treatment Response

SNPs are critical in detecting how a patient

will respond to medication, a ﬁeld known as

pharmacogenomics. Variants in the CYP2D6 and

CYP2C19 genes determine how quickly the liver

metabolises antidepressants. Detection of Ultra-rapid

or Poor metabolize r SNPs prevents toxicity and

treatment failure. Variations in the BDNF (Brain-

Derived Neurotrophic Factor) gene can predict the

success of SSRIs, as they inﬂuence the brain’s ability

to physically adapt and heal during treatment.

Example 1: The BDNF Val66Met Polymorphism

(rs6265)

The Brain-Derived Neurotrophic Factor

(BDNF) gene encodes a protein that acts like fertilizer

for the brain, helping neurons grow and form new

connections (synapses).

The BDNF transcript comprises one of eight 5’

untranslated exons (exon I-VIII) and the common 3’

protein coding exon IX; B: Intracellular signalling after

TrkB activation. Following BDNF binding, TrkB

dimerisation and its phosphorylation at intracellular

tyrosine residues occur. Then, the activated TrkB

stimulates three main signalling pathways: (1) mitogen-

activated protein kinase/extracellular signal-regulated

kinase (MAPK/ERK); (2) phosphatidylinositol 3-

kinase (PI3K); and (3) phospholipase C

(PLC

)

Figure 1: Brain-derived neurotrophic factor (BDNF)

gene and stimulated intracellular signalling cascades

after activation of tropomyosin-related kinase (Trk)B.

Image courtesy:https://tse3.mm.bing.net/th/id/OIP.Kk1ntIO-iez3fzmc93LTdAHaHS?pid=ImgDet

&w=204&h=200&c=7&dpr=1.3&o=7&rm=3

pathways. The MAPK pathway, in which MAPK/ERK

kinase (MEK) is involved, plays a role in neuronal

diﬀerentiation and outgrowth. PI3K signalling

promotes neuronal survival via Ras or GRB-associated

binder 1 (Gab1). Following PLC

activation, inositol-

1,4,5-trisphosphate (IP3) and diacylglycerol (DAG)

are both produced. DAG activates protein kinase

C (PKC), which is important for regulating synaptic

plasticity. Meanwhile, IP3 increases intracellular

Ca 2+ concentration via IP3 receptors on the

endoplasmic reticulum (ER), resulting in activation

of Ca 2+ /calmodulin (CaM)-dependent protein kinase,

including CaMKII, CaMKK, and CaMKI. These

MAPK/ERK, PI3K, and PLC

pathways can regulate

gene transcription.

In the SNP, a single nucleotide switch causes the

amino acid Valine (Val) to be replaced by Methionine

(Met) at position 66. The Met variant impairs BDNF

secretion. Research shows that individuals with the

Met allele often have a smaller hippocampus, the brain

region responsible for emotion regulation and memory.

Clinical Impact:

Detection of this SNP helps clinicians understand

why some patients suﬀer from treatment-resistant

depression. Because their brains have lower plasticity,

2.4 Pharmacogenomics and Treatment Response

Figure 2: Schematic view of how the SLC6A4 and

5-HTTLPR genes eﬀectiveness in diﬀerent parts of the

brain.

Image courtesy:https://medcraveonline.com/IJMBOA/assessment-of-genetic-mutations- by-the-g

enersquos-5- httlpr-and-slc6a4-and-scc6a4- in-the-human-depression.html

standard antidepressants may take longer to work or

require supplemental therapies like Exercise or TMS

(Transcranial Magnetic Stimulation), which naturally

boost BDNF levels.

Example 2: The SLC6A4 5-HTTLPR

Polymorphism

The SLC6A4 5-HTTLPR Polymorphism is perhaps

the most famous example of a cryptic genetic marker in

psychiatry. It involves the Serotonin Transporter

Gene, which is the primary target for Selective

Serotonin Reuptake Inhibitors (SSRI) antidepressants

(like Prozac or Lexapro). Serotonin is the happiness

hormone.

The SLC6A4 gene encodes the serotonin

transporter, and its 5-HTTLPR polymorphism

signiﬁcantly inﬂuences serotonin reuptake, impacting

mood regulation and responses to antidepressants.

The Variation: This is a functional polymorphism

in the gene’s promoter region, often categorised into

Short (S) and Long (L) alleles.

The Mechanism: The S allele results in lower

expression of the serotonin transporter. This makes the

brain’s emotional centre (the amygdala) hyper-reactive

to stress.

Clinical Impact

Stress Sensitivity: People with the S allele are

statistically more likely to develop MDD following a

stressful life event (Gene-Environment Interaction).

SSRI Response: Studies suggest that individuals

with the L/L genotype generally respond better and

Figure 3:

SLC6A4-5-HTTLPR. The SLC6A4 gene encodes

the protein responsible for serotonin reuptake. The

5-HTTLPR polymorphism in the promoter region

dictates expression levels: the short (S) allele reduces

transcription, leading to lower transporter density on

the presynaptic membrane compared to the long (L)

allele. This mechanism is the primary target for SSRI

antidepressants.

Image courtesy:https://genesight.com/white-papers/get-to-know-a-gene- slc6a4/

faster to SSRIs than those with the S allele, who may

experience more side eﬀects

References

https://genesight.com/white-papers/get-to-kno

w-a-gene-slc6a4/

https://medcraveonline.com/IJMBOA/assessme

nt-of-genetic-mutations-by-the-genersquos-5-httlpr-a

nd-slc6a4-and-scc6a4-in-the-human-depression.html

https://tse3.mm.bing.net/th/id/OIP.Kk1ntIO-iez

3fzmc93LTdAHaHS?pid=ImgDet&w=204&h=200

&c=7&dpr=1.3&o=7&rm=3

Howard, D. M., et al. (2019). Genome-wide

meta-analysis of depression identiﬁes 102 independent

variants and highlights the prefrontal cortex. Nature

Neuroscience. A foundational paper establishing the

polygenic architecture of depression.

McIntosh, A. M., et al. (2024). Genome-wide

study of major depression in 685,808 diverse individuals

identiﬁes 697 independent associations. medRxiv.

2.4 Pharmacogenomics and Treatment Response

This massive meta-analysis highlights the polygenic

nature of MDD and identiﬁes hundreds of novel SNP

associations.

Mullins, N., et al. (2021). Genome-wide

association study of more than 40,000 bipolar disorder

cases provides new insights into the underlying biology.

Nature Genetics. (Used for transdiagnostic SNP

comparison).

Wray, N. R., et al. (2018). Genome-wide

association analyses identify 44 independent loci

associated with major depressive disorder. Nature

Genetics. This study conﬁrms that MDD risk is driven

by the additive eﬀects of common SNPs.

Zajac, G. J., et al. (2025). Inﬂuence and role of

polygenic risk score in the development of 32 complex

diseases. Journal of Global Health. Provides data on

how PRS improves clinical risk prediction for early-

onset cases of MDD.

About the Author

Geetha Paul is one of the directors of

airis4D. She leads the Biosciences Division. Her

research interests extends from Cell & Molecular

Biology to Environmental Sciences, Odonatology, and

Aquatic Biology.

Part IV

Computer Programming

From Telescope to Data Product

by Ajay Vibhute

airis4D, Vol.4, No.3, 2026

www.airis4d.com

1.1 From Telescope to Data Product

1.1.1 Raw Measurements vs Scientiﬁc Data

The journey from a telescope’s detector to a usable

scientiﬁc data product begins with raw measurements,

which are simply numerical values—counts or

voltages—recorded by the instrument. These values

reﬂect the detector’s response to incoming radiation

rather than direct physical properties of celestial

objects. A CCD pixel, for example, measures

accumulated charge generated by photons, but that

signal also includes contributions from background

sky emission, electronic oﬀsets, thermal noise, and

occasional cosmic ray events. Raw measurements

are further shaped by the instrument itself. The

telescope’s point-spread function spreads light from

a single source over multiple pixels, and variations

in pixel sensitivity introduce spatial inconsistencies.

Detector nonlinearities and electronic eﬀects can distort

signals, particularly at high ﬂux levels. As a result, the

raw image is not a direct representation of the sky, but a

convolution of the true scene with instrumental response,

embedded within noise, ﬁgure 1. Because of this,

raw data are not immediately interpretable in physical

terms. Transforming them into scientiﬁcally meaningful

data requires computational corrections such as bias

subtraction, ﬂat-ﬁelding, background modeling, and

calibration into standardized physical units. These steps

convert instrument-dependent signals into quantities

that can be reliably compared and scientiﬁcally analyzed,

ﬁgure 2.

Figure 1: Recorded Raw image

Figure 2: Reconstructed sky image

1.1 From Telescope to Data Product

1.1.2

Data Acquisition and Instrument Eﬀects

Data acquisition in astronomy is far from a passive

recording process. Each measurement is shaped

by the design of the telescope, the properties of

the detectors, and the observing environment. The

optical conﬁguration determines angular resolution

and light-gathering power, while alignment errors,

mirror imperfections, or thermal expansion can subtly

alter image quality. In ground-based observations,

atmospheric turbulence blurs incoming light, producing

time-varying distortions commonly referred to as

seeing. Atmospheric extinction and sky background

further modify the signal before it even reaches the

detector. In space-based instruments, although the

atmosphere is absent, other challenges arise, including

cosmic ray impacts, thermal ﬂuctuations, and small

pointing instabilities of the spacecraft. Detectors

themselves introduce additional complexities. Readout

noise aﬀects faint signals, pixel-to-pixel sensitivity

variations create spatial non-uniformities, and charge

diﬀusion can blur sharp features. Bright sources

may saturate detector elements, leading to nonlinear

responses or data loss in high-intensity regions. Even

the timing of exposures and the stability of electronics

inﬂuence the ﬁnal measurement. Understanding these

inﬂuences is essential because they fundamentally

shape the raw data. Computational corrections applied

later—such as deblurring, background subtraction,

and calibration—depend on accurate models of these

instrumental and environmental eﬀects. In this way, data

acquisition and computational processing are tightly

interconnected stages of a single measurement process.

1.1.3 Calibration as a Computational Task

Calibration transforms raw measurements into

quantitative, scientiﬁcally meaningful data by

systematically correcting for instrumental and

environmental eﬀects. At its most basic level,

this includes procedures such as bias subtraction to

remove electronic oﬀsets, dark current subtraction to

account for thermally generated charge, and ﬂat-ﬁelding

to correct for pixel-to-pixel sensitivity variations.

These steps ensure that the recorded signal more

accurately reﬂects incoming radiation rather than

detector imperfections. In imaging, calibration may

also involve modeling and subtracting background sky

emission, while in spectroscopy it includes wavelength

calibration using known spectral lines. Beyond these

corrections, calibration extends to photometric and

spectral transformations, converting detector counts

into standardized physical units such as ﬂux density,

magnitude, or calibrated wavelength. This often

requires reference observations of standard stars or

laboratory calibration sources, enabling measurements

from diﬀerent instruments or observing runs to be

directly compared. Astrometric calibration further

aligns images with celestial coordinate systems,

allowing precise positional measurements. In modern

astronomy, calibration is primarily a computational

process, embedded within automated pipelines that

apply corrections consistently across thousands or

millions of observations. Because calibration deﬁnes

the quantitative scale of the data, even small inaccuracies

can introduce systematic biases that propagate into

derived parameters such as distances, masses, or

luminosities. For this reason, calibration procedures

must be rigorously validated, regularly updated, and

carefully documented to ensure the reliability and

reproducibility of scientiﬁc results.

1.1.4 Standard Data Products in Astronomy

Once calibrated, data can be transformed into

standardized products suitable for analysis, distribution,

and long-term archiving. These products include

processed images, calibrated spectra, time-series

light curves, and structured catalogs of detected

sources. Each product is typically accompanied by

extensive metadata describing observing conditions,

calibration parameters, uncertainty estimates, and

quality ﬂags. This contextual information is essential,

as it allows researchers to assess reliability, propagate

uncertainties, and reproduce results. Standardization

is critical because modern astronomy is inherently

comparative and cumulative. Datasets from diﬀerent

instruments, observatories, or observing epochs must

be interoperable to enable cross-matching, multi-

1.1 From Telescope to Data Product

wavelength analysis, and large-scale statistical studies.

Uniform data formats, coordinate systems, and

calibration conventions make it possible to combine

observations across surveys and to perform automated

analyses on millions or billions of sources. In large-scale

projects such as the Sloan Digital Sky Survey (SDSS)

or Gaia, standardized data products form the backbone

of scientiﬁc research, supporting investigations into

stellar populations, galactic structure, dark matter

distribution, and cosmology. Computational pipelines

are central to the production of these products. From

ingesting raw detector outputs to performing source

detection, photometric measurement, and catalog

assembly, pipelines ensure consistency and scalability.

They also embed quality-control procedures that ﬂag

anomalies and track processing provenance. In this

way, standardized data products represent not just

processed observations, but carefully curated outputs of

an integrated computational system designed to support

reliable scientiﬁc discovery.

1.1.5 Sources of Systematic Error

Even after calibration and processing, systematic

errors can aﬀect data quality and interpretation

in subtle but signiﬁcant ways. Unlike random

noise, which tends to average out over many

measurements, systematic eﬀects introduce consistent

biases that can shift results in a particular direction.

These errors may stem from instrumental limitations

such as optical aberrations, imperfect ﬂat-ﬁeld

corrections, detector nonlinearities, or long-term drift

in sensitivity. Environmental factors—including

variations in atmospheric transparency, scattered

light, thermal instability, or imperfect background

subtraction—can also leave residual signatures in the

data. In addition, the data reduction algorithms

themselves can introduce biases. Assumptions

about noise distributions, source shapes, background

levels, or model parametrizations may not hold

uniformly across all observations. For example, an

incorrect model of the point-spread function can

distort photometric measurements, while oversimpliﬁed

background modeling can artiﬁcially enhance or

suppress faint sources. In large surveys, even small

systematic biases can accumulate across millions

of objects, leading to measurable distortions in

statistical analyses. Systematic errors are particularly

challenging because they can mimic or obscure

genuine astrophysical signals, sometimes producing

apparent trends or correlations that are not physically

real. Detecting and mitigating these eﬀects requires

rigorous validation procedures, cross-comparisons with

independent instruments or surveys, simulation-based

testing, and careful uncertainty modeling. In modern

astronomical computing, controlling systematic error

is often as important as increasing statistical precision,

and it remains one of the central challenges in producing

reliable scientiﬁc results.

1.1.6 Concluding Remarks

The transformation from telescope to data product

illustrates the central role of computation in modern

astronomy. Every stage—from raw measurement to

calibrated image or structured catalog—depends on

algorithms that correct instrumental eﬀects, model

uncertainties, and extract meaningful signals from

complex data. Scientiﬁc results therefore reﬂect

not only the performance of the telescope, but also

the assumptions and design of the computational

pipeline. Understanding both the physical origins of

measurements and the methods used to process them is

essential for producing reliable and reproducible science.

In this sense, modern data products are true co-creations

of hardware and software, where computational

processes are as fundamental to discovery as the

instruments that collect the light.

About the Author

Dr. Ajay Vibhute is currently working

at the National Radio Astronomy Observatory in

the USA. His research interests mainly involve

astronomical imaging techniques, transient detection,

machine learning, and computing using heterogeneous,

accelerated computer architectures.

About airis4D

Artiﬁcial Intelligence Research and Intelligent Systems (airis4D) is an AI and Bio-sciences Research Centre.

The Centre aims to create new knowledge in the ﬁeld of Space Science, Astronomy, Robotics, Agri Science,

Industry, and Biodiversity to bring Progress and Plenitude to the People and the Planet.

Vision

Humanity is in the 4th Industrial Revolution era, which operates on a cyber-physical production system. Cutting-

edge research and development in science and technology to create new knowledge and skills become the key to

the new world economy. Most of the resources for this goal can be harnessed by integrating biological systems

with intelligent computing systems oﬀered by AI. The future survival of humans, animals, and the ecosystem

depends on how eﬃciently the realities and resources are responsibly used for abundance and wellness. Artiﬁcial

intelligence Research and Intelligent Systems pursue this vision and look for the best actions that ensure an

abundant environment and ecosystem for the planet and the people.

Mission Statement

The 4D in airis4D represents the mission to Dream, Design, Develop, and Deploy Knowledge with the ﬁre of

commitment and dedication towards humanity and the ecosystem.

Dream

To promote the unlimited human potential to dream the impossible.

Design

To nurture the human capacity to articulate a dream and logically realise it.

Develop

To assist the talents to materialise a design into a product, a service, a knowledge that beneﬁts the community

and the planet.

Deploy

To realise and educate humanity that a knowledge that is not deployed makes no diﬀerence by its absence.

Campus

Situated in a lush green village campus in Thelliyoor, Kerala, India, airis4D was established under the auspicious

of SEED Foundation (Susthiratha, Environment, Education Development Foundation) a not-for-proﬁt company

for promoting Education, Research. Engineering, Biology, Development, etc.

The whole campus is powered by Solar power and has a rain harvesting facility to provide suﬃcient water supply

for up to three months of drought. The computing facility in the campus is accessible from anywhere through a

dedicated optical ﬁbre internet connectivity 24×7.

There is a freshwater stream that originates from the nearby hills and ﬂows through the middle of the campus.

The campus is a noted habitat for the biodiversity of tropical Fauna and Flora. airis4D carry out periodic and

systematic water quality and species diversity surveys in the region to ensure its richness. It is our pride that the

site has consistently been environment-friendly and rich in biodiversity. airis4D is also growing fruit plants that

can feed birds and provide water bodies to survive the drought.