Research interests

I research Multi-Phase Flows in the context of geophysical fluid dynamics (the oceans and atmosphere) and industrial applications like aircraft and reactors. I do basically theory and computer simulations. I am passionate about turbulent dynamics of air-sea interactions, bubbly flows, cloud condensation, gas exchange, general circulation, upper ocean dynamics, internal waves and shock-boundary layer interactions. My studies revolve around science of boiling, bubble dynamics, cavitation, drops, interfaces and turbulence. Some of the research areas that I am continuously learning and updating my knowledge are mentioned below (click them to see more information).

Air-Sea interactions and climate change

A bubble in the ocean may seem insignificant. Still, let's consider all the bubbles in all the oceans. They drive our planet's climate. Bubbles in the ocean are significant for several reasons, ranging from mixing the ocean's upper layer to scavenging biological matter, which means they can also impact the state of the ocean's surface where they are present. Bubbles serve as an essential mechanism by which air is dissolved in the ocean. The wave breaking and ultimate bubble bursting at the surface can cause droplets and aerosols to be ejected into the atmosphere. What dynamical processes affect the bubble creation, transport and bursting at the air-sea interfaces to influence overall gas transfer across the air-sea interfaces? Is that due to the atmosphere wind, wave-breaking events, turbulence at a surface level due to Langmuir circulations, or thermal convection? What is the role played by rain and precipitation on ocean sprays: will that knock down the wave-breaking events to decrease the gas exchange, or will that enhance via satellite drop production? How do we quantify the complex state of turbulence and its energy transfer mechanism when the upper surface layers involve multi-phase phenomena? Many quantities transferred across the sea surface are associated with dynamical processes taking place at much smaller scales than the dominant surface waves (driven by the winds and planetary rotation). We aim to advance the understanding of gas transfer theory across the interfaces.

Chaotic transport of species

Transport of scalar quantities (e.g. chemical species, nutrients, heat) in deterministic flows is key to a wide range of phenomena and processes in industry and Nature. This encompasses length scales ranging from microns to hundreds of kilometres, and includes systems as diverse as viscous flows in the processing industry, micro-fluidic flows in labs-on-a-chip and porous media, large-scale geophysical and environmental flows, physiological and biological flows and even continuum descriptions of granular flows. An essential contribution to the net transport of a scalar quantity is its advection by fluid motion. The Lagrangian perspective (arguably) is the most natural way to investigate advection and leans on the fact that fluid trajectories are organized into coherent structures that geometrically determine the advective transport properties. This notion enables systematic investigation of fundamental transport phenomena, and the application of scientific insights thus gained for purposes ranging from transport studies on environmental flows to the design and optimization of industrial processing equipment.

Cloud condensation on aircraft

The phenomenon of condensation is well-known for cloud formation in the atmosphere, liquid droplets that form on car windshields, or shock collars that appear around fighter planes. Suppose an aircraft approaches transonic velocity or flow over any of its convex parts (like intakes, canopy etc.) and reaches supersonic speed. In that case, a rapid temperature and pressure decrease, thus condensation. The variation in temperature due to the perturbation in airflow is called Prandtl-Glauert singularity and is particularly interested in boundary layer dynamics. The flow perturbation causes the particular cloud shape associated with the singularity. At that point, the airflow can reach supersonic speed and generate a shock wave (that appears when the fluid decelerates and the temperature suddenly rises). Moreover, the shock due to the quick jump from a low-pressure-low temperature-supersonic airflow zone to a high-pressure-high temperature-subsonic speed zone generates an acoustic bang. Aircraft safety ultimately depends on cloud formation physics.


Cavitation is typically a multi-scale problem since it involves local transfers at interfaces, bubble interactions, all scales of turbulence, and large-scale organized instabilities. It is thus a complex and exciting phenomenon, which is harmful in large scale applications involving high speed flows, such as hydraulic machinery, injectors or naval propellers, but it is also very beneficial for chemical processes, medical and pharmaceutical applications, especially in configurations of micro and nano bubbles generated by acoustic waves or laser. At large scale, for example in marine propellers and pumps, cavitation results in turbulent, compressible unsteady flows, in which the local proportion of gas varies between 0 and 100%. These properties – quite unusual in fluid mechanics - make the understanding of their structure and dynamics a challenging task. Cavitation induces bizarre effects such as performance decrease, instabilities and erosion after a long operating time, because of pressure waves due to the bubble collapses. In pumps, instabilities perturb the rotor equilibrium and lead to significant vibrations, which have been found responsible for the failure of the Japanese launcher H-II in 1999. Study of cavitation in pumps, and especially in space launcher turbopumps used by ISRO (Indian Space Agency), CNES (French Space Agency) and SNECMA (manufacturer of the Ariane V engines) are fascinating topics for further exploration. An in-depth theoretical study of the governing equations for the cavitating flows has been conducted.

Flow transition in buoyant plumes

A thermal plume is a layer (or blob) of warm fluid rising in a relatively cool fluid due to buoyancy. In nature, they are widely visible as exhaust from hot chimneys, volcanic eruptions, cloud formations and more. Usually, they are created on hot spots (or sources) and rise upwards. As they move away from the source, their lateral size (width) gets widened due to entrainment of ambient fluid. The important questions in this regard are `how does local velocity and temperature changes as we move away from the source?', `Is the observed flow stable with time or undergoes any transition and turbulence?', or `how much heat flux will be taken out from the source?'.

Machine learning in turbulent flows

Hundred years of research on turbulence show that the stochastic approach is a powerful concept and gave many interesting facts to improve our understanding of turbulence. Nevertheless, it could not answer some of the questions such as, `why a flow undergoes the transition to turbulence?', 'which dominant structures (or building blocks of flow) are responsible for the observed chaotic field? 'how do they interact and evolve with time from the smooth deterministic field to a chaotic state?', 'Are there any multiple states of turbulence instead of a commonly assumed universal concept?'. Though dynamical system theory partially gives a perspective and direction via finding exact solutions to NS equations (at moderate Reynolds numbers) such as equilibrium (steady state), non-equilibrium or periodic orbits, significant Reynolds number limits are beyond to handle. For example, establishing the criterion for detecting the turbulent/nonturbulent interface is challenging in turbulence research. As a new technique, machine learning provides powerful tools to extract information from data that can generate knowledge about the underlying fluid mechanics. To ensure a turbulent/nonturbulent interface as an independent quantity of the reference frame, we consider invariants of tensors appearing in the transport equations of velocity fluctuations, strain rates, and vortices as input features to identify the flow state. Compared with the traditional detection method, the detector trained by machine learning captures simultaneously and objectively the different fundamental properties of turbulent flows, including instability, vortex stretching, and three-dimensionality, providing an accurate and novel definition of turbulence.

Nucleate boiling and critical heat flux limit

Liquid-vapour phase transitions such as boiling and condensation are omnipresent. They can be widely seen in the kitchen to make coffee, in process industries to do distillations, and in thermal power plants to generate vapour that drives turbines. Though numerous applications are involved, understanding the process's fluid dynamics is only at a fundamental level. So I attempted to understand boiling phenomena at a fundamental level by combining fluid mechanics with heat transfer. In boiling, liquid vaporizes by absorbing latent heat from its surroundings. For this to happen, the temperature in the liquid has to exceed a specific temperature called `boiling temperature', at which the vapour pressure of the liquid exceeds the ambient pressure from the surroundings. If we decrease the local ambient pressure, the temperature at which boiling occurs also decreases. The formed vapour and the surrounding heated liquid are carried away from its initial location due to buoyancy and transport large amounts of heat and mass. Many empirical correlations exist on heat transport in boiling phenomena, but a physical understanding of these phenomena remains inconclusive. How to understand the observed heat transport and the flow dynamics based on a suitable model with the proper physics of boiling? Can we increase critical heat flux limits in boiling?

Ocean circulation- a big voyage

The ocean circulation is the result of a balance between wind forcing and air-sea heat and freshwater fluxes at planetary scales and dissipation at centimeter scales. A full theory of the circulation must therefore include a discussion of the processes that take the energy from the forcing to the dissipation scales, also known as the turbulent cascade. Can we develop a theory with a combination of observations and fluid dynamics arguments. How to construct a theory by taking care of lee waves, submesoscale eddies and small and large scales of motions? One way is to develop theoretical model of the large-scale ocean circulation where the effect of turbulence is represented with simple scaling laws.

Particle-laden flows

Particle-laden turbulent flow is a complex problem that plays an essential role in various industrial and environmental applications, including pneumatic mixing, thermal processes in manufacturing, heat transfer in power stations, and sea spray in hurricanes. Due to these problems' fundamental importance and practical interests, turbulent particle-laden flows have received considerable attention in experimental and numerical studies in recent years. Under the influence of turbulence, particles exhibit a variety of phenomena, such as dispersion, concentration, deposition, and re-suspension. On the other hand, due to the interactions between turbulence structure and dispersed particles, turbulence characteristics of momentum and heat transport can be modified by the presence of particles. Our group has been researching the momentum and heat transfer in particle-laden turbulent channel flows. We are also investigating the role sea sprays play in air-sea interactions. We have developed a coupled level-set, and volume-of-fluid method to simulate the air and water flows in wave breaking. A Lagrangian particle tracking method based on a point-force approximation is used to simulate the motions of sea spray. We have simulated spray generation and transport in the wind over breaking waves.

In addition to point particle simulation, we also focus on fully resolved particle simulation. In this simulation, the flow field is resolved by direct numerical simulation. At the same time, the particle motion is calculated by the Newton-Euler equations, with the particle-turbulence interaction solved using the immersed boundary method. Particle-resolved simulation can provide deep insights into particle-turbulence interaction physics to help improve the point particle model. We have performed a particle-resolved simulation to study turbulence modulation by sediment transport. This study uses the discrete element method to resolve particle-particle and particle-wall collisions. The particles can modify the statistical properties and structure of turbulence. We are currently investigating how particles influence turbulence properties and structures.

Planetary atmospheric circulations

Mars, Venus, the Earth and Titan in the solar system are the nearest planetary bodies where substantial evidence of atmospheric flows are found. The nature of atmospheric circulation on these bodies significantly differ from each other [2]. The dynamic meteorology of these planets depend on the body rotation, solar heat information, the obliquity, the gravity, emitting temperatures, the overall stratification, terrestrial content, gases and many more. Some of these atmospheres will be as multifaceted as the Earth's or a combination of the complexity. To quote in the words of greatest Meteologist Hide (1969): `the central scientific problem concerning global atmospheric circulation is arguably that of predicting from the laws of classical physics that [an] atmosphere is necessarily organised as it is'. In other words, for the chosen laws of science (fluid dynamics, radiative transfer, thermodynamics) with a set of control parameters and suitable boundary conditions, can we quantitatively predict the atmospheric circulation and the consequent state of its climate? Can our powerful computers help in understanding the atmospheric circulations under a variety of conditions? Can we help the General Circulation Models (GCMs) by providing better estimations of energy dissipation rates through direct numerical simulations on spherical convection? Can we provide reasons for the hydrodynamic and thermal instabilities observed in atmosphere?

Laminar or Turbulent flows?

Ever since the publication of the pioneering work of Osborne Reynolds in 1883 defining the concept of laminar and turbulent flows, the subject has remained a central theme in fluid mechanics due to its fundamental importance to the subject and its relevance to engineering applications and the natural world. We know, for example, that the flow in a pipe is likely to be laminar when the Reynolds number is below around 2300 and turbulent when the Re is higher. One way in this direction is to understand the bypass transition ( the transient flow following a rapid increase of flow rate from an initially turbulent flow is characterized by a time-developing laminar boundary layer followed by laminar-turbulent transition even though the initial flow is turbulent), which develops from an established turbulent wall shear flow.

Landslides in oceans

Giant landslides and sediment avalanches on the seafloor are a demonstrated hazard to seafloor infrastructure (e.g. internet cables and oil pipelines) as well as being the key mechanism by which terrestrial sediment is transported thousands of kilo-metres before ultimate burial in the deep sea. Our understanding of these landslides and avalanches, from how seafloor slopes fail to how the flows evolve is limited because we know little about the material properties (i.e. the rheology) of the sands, silts and clays that make up the seafloor in the deep sea. Understanding these properties will lead to a better understanding of where and why landslides and avalanches occur, how such flows evolve and therefore enable better modelling capabilities. This will ultimately inform, where to locate and how to protect seafloor infrastructure; how such flows interact with seafloor habitats and how sediment is transported in our oceans.

Turbulence as a non-equilibrium phenomenon

In dissipative systems at the stationary state, injected energy balances dissipation. Generally, the dynamics of such systems might be near the thermal equilibrium or away from it. If the system is at equilibrium, the fluctuation-dissipation theorem gives a relation between the amplitude of fluctuation and the dissipation rate, e.g., the Einstein-Smoluchoski relation well describes the motion of polymers in a thermal bath. If the system is away from the equilibrium, significant deviations from the average behaviour are observed, and no general relations are available. Recently, Fluctuation Theorems (FT) has been introduced in statistical physics to generalize the relations between the fluctuations and the dissipation of non-equilibrium systems. There are various forms of FT, and the important three are: 1) Crooks theorem, 2) Jarzynski inequality, and 3) Steady state-transient state fluctuation theorem. For a system away from the equilibrium, FTs give a relation of asymmetry involved in doing work (or time-averaged entropy production) with time. This indicates that with time the probability of seeing negative work in a time-averaged process decreases exponentially. With the help of these theorems, the second law of thermodynamics and Green-Kubo relations can be generalized. The derivation of these theorems requires only conditions such as ergodicity and time reversibility, but not on the work PDF (need not be a Gaussian). FT has been checked and thriving in various systems such as granular gases, colloid science, mechanical oscillations and liquid crystals with reasonable accuracy where the microscopic reversibility is not satisfied. We know that in fluid dynamics, large fluctuations can occur at varying scales of motion. So, are these FTs helpful in exploring fluctuations in fluid turbulence? I am presently working on this to test the validity of FTs.

Shock-Turbulence interactions in boundary layers

The accurate numerical solutions of shock wave and turbulent boundary layer interaction remain one of the challenging problems in computational aerodynamics. Understanding the interactions will help control fluctuations in high pressure and heat transfer loads and delay flow separation. The complex mechanism requires detailed investigation for safer and fuel-efficient aeroplane designs. The shock interactions with the boundary layers can occur for internal and external surfaces and are different. The adverse pressure gradient imposed by shock and shock propagation through multilayered viscous and inviscid layers can dramatically change the drag. Large-eddy simulations based on hybrid schemes will be the direction in which unsteady mechanisms can be investigated. Instead of introducing artificial low-frequency modes that could affect the interaction, a method based on a digital-filter approach is to be looked further for mean and turbulent quantities. The shock reflections and the role of recirculation bubbles over coherent vortices to be studied. The effect of micro-ramps, bleed holes and flaps in the context of shock-boundary layer interactions leads to further insight of the problem.

Wall bounded turbulence

Wall bounded flows such as boundary layers, pipe flows, and channel flows are very relevant to many technological applications, e.g., oil transport through pipes, air drag over vehicles, flow past turbine blades etc. Though the study of wall-bounded flows started more than a hundred years back, identifying a correct physical mechanism that generates turbulence and its sustainment remains ambiguous. What makes these flows challenging to understand compared to an isotropic homogeneous flow? The presence of a mean velocity gradient creates anisotropy in the flow, making it difficult to understand. Briefly, I will describe how turbulence is generated in these flows. Energy from the mean flow is taken out by Reynolds stresses and generates turbulence near the walls (buffer layers). The log layers transport the generated energy continuously to the outer layers. The outer flow interacts with the wall layer and influences turbulence production. Many exciting queries such as `how are Reynolds stresses created?', 'How is energy transported through log-layer?', `do any organized structures (hair-pin vortices and streaks) exist? If so, what is their role?', `how do outer layers interact with inner layers?', `is log-law behaviour universal?', or `how do fluctuations adjust themselves to get a momentum balance?', become meaningful. Though some of the questions are partly explored in the earlier literature, one needs high Reynolds number data to validate them. So I am planning to do numerical simulations to explore them further.

Extreme events prediction using Machine learning techniques

`The history of numerical weather prediction (NWP) and that of machine learning (ML) differ substantially. First, manual NWP was attempted by Lewis Fry Richardson in Britain in 1922, and early computer-aided weather forecasts were produced in 1950. These were soon followed by operational weather forecasts in Sweden, the USA and Japan. While there has been steady progress in the development of NWP, the history of ML has been more disruptive: the first neural network (NN) was proposed in 1943 by McCulloch & Pitts. The field expanded until the early 1960s, when the existing algorithms proved inefficient and unstable. The invention of backpropagation in 1970 led to a second wave of ML applications as it became possible to build more extensive NNs and train them to recognize nonlinear relationships in data. Even though the development of ML algorithms continued, the enthusiasm about them soon dwindled again, because they rarely showed significant performance gain, and computing resources were not sufficient to solve larger problems. Furthermore, big amounts of labelled data which are mandatory for most data-driven ML approaches were hard to come by (note that this was before the advent of the world wide web). Three significant developments around 2010 started the third wave of artificial intelligence, which continues to the present: computing capabilities were vastly expanded due to massive parallel processing in graphical processing units (GPUs), convolutional neural networks (CNN) allowed much more efficient analysis of massive (image) datasets, and large benchmark datasets were made available on the internet. Highly complex NNs with greater than 106 parameters enabled a breakthrough in image recognition, soon followed by remarkable success stories in speech recognition, gaming, and video analysis and prediction. Today’s NNs are often deep networks with greater than 10 layers, and the research field which develops such NNs and the associated methods for training and validation is called deep learning (DL). The weather and climate research community is increasingly aware of modern DL technologies and tries to adopt them to solve specific data analysis, numerical modelling and post-processing problems in the context of NWP. However, the weather prediction is nontrivial task due to complex state of extreme events. How to predict such extreme events through ML?'