We meet Dr Francesca Mangani at TU Wien's Getreidemarkt campus in the Institute of Fluid Mechanics and Heat Transfer, where she works as a senior scientist on turbulent multiphase flows. Among bookshelves and aeroplane models, we discuss her simulations of drop distributions and supercomputing calculations, as well as the importance of fundamental physics in engineering studies. Holding her aeronautical engineering degree from Milan, she begins by telling us how she discovered her passion for fluid mechanics' fundamental science: every lift, drag, and thrust force keeping aeroplanes aloft hinges on the invisible behaviour of surrounding fluids.
Drops and bubbles
“I’m working on drops and bubbles. If you have water in air, that's drops in gas. If you have air in water, that's bubbles. The goal is to study how these drops or bubbles behave in turbulence. So, while their surface tension holds them spherical and always fights to restore that spherical shape, turbulence disturbs the surface and tries to break them apart. We're studying these processes, these so-called turbulent multiphase flows, to create reliable models about break-up phenomena, collision, or coalescence of drops."
Dr Francesca Mangani explains to us how her research field is divided into experimental and computational departments, with her working on the computational side using mathematical models of fluid dynamics – essentially sets of equations – to approximately predict flow behaviours. In order to describe the seemingly chaotic behaviour of drops in a turbulent flow, her team uses specialised software extracting critical quantities of interest, such as: how many drops exist at each size and how that distribution evolves over time, etc.
Studying these fundamental interactions, they find certain scaling laws that allow them to simplify complex phenomena related to drop dynamics, which in turn find application in climate and ocean-atmosphere modelling. Francesca points to breaking waves as an example, where spilling and plunging increase the fluid’s interface by up to 30 per cent and, in turn, speed up the exchange of heat, mass, and especially CO₂ between the ocean and the atmosphere.
Industrial engineering cases typically feature other complex geometries – like flow around aeroplanes, bird flight, or turbomachinery producing electricity from fluids. Here, high-accuracy simulations must be simplified in such setups due to high computational costs.
Scaling down phenomena
“So basically, you can simulate a physical phenomenon by solving a set of equations, discretised with appropriate numerical schemes. You use a code, written for a specific set of fundamental problems, which does all these operations, and you let this run on a computer. If the goal is to understand a physical phenomenon in the best way, the methodology used should be the most accurate possible, at a feasible computational cost. Indeed, we must often reduce the size of the physical problem considered. We focus on a small domain, just a box, maybe simulating a wave which is breaking, but with dimensions of only 10 centimetres to 10 centimetres. To have a bigger wave, you would need to have more computing power. We can set slightly different conditions, activate or deactivate some effects; for example, we can study just the flow with or without heat transfer or including mass transfer, etc.”
Francesca Mangani walks us through one such contained "box" from her simulations – a compact channel where hot drops shed heat into cold turbulent fluid (see graphic above). Even here, fully resolving the turbulence demands five hefty data files per timestep: one each for the three velocity components sweeping across the entire domain, plus separate fields for temperature and phase tracking that pinpoint every drop's position in the chaos. These transient calculations churn through timesteps until equilibrium sets in, when heat transfer ceases and the system stabilizes. File sizes explode with the number of grid points used in the simulation but getting close to accuracy affords refining the grid enough to capture every tiny vortex and structure in the chaotic behaviour. Tackling complex geometries, models must therefore be simplified to manage these computing costs and time, which requires a perpetual balance between simulation accuracy, geometric realism, and weighing of all relevant physical parameters.
Sharing code instead of big data
“Data storage is still a bottleneck for us, because big datasets require many hard disks or more advanced storing systems. Our simulations run on high performance computers (HPC), and computing hours are not for free. That’s why we constantly follow developments in computer science and try to keep our code as fast and efficient as possible. Running on GPUs helps us a lot by parallelising calculations across multiple units – so one simulation now takes one day instead of one week. This allows us to detect mistakes earlier and test more parameters in less time. It is impossible to share these volumes of raw data output from our simulations, we rather share code by making it open source.”
Dr Mangani explains that simulation outputs reach several terabytes per project – resolving fine velocity and temperature structures demands not only vast storage but also massive processing power. While her research team secures CPU/GPU hours through competitive applications to the Austrian Scientific Computing (ASC) infrastructure and the Italian HPC infrastructure CINECA, a lot of researchers lack access to supercomputing clusters. Raw data sharing would overwhelm networks and prove useless to peers unable to analyse it, so files remain on local hard disks while derived statistics reach specialised platforms like Japan's THTLAB (thtlab.jp).
Instead, Francesca Mangani's computational team embraces FAIR principles through Flow36 and MHIT36 – in-house codes refined over years by former colleagues and now open-source on GitHub. These tools let others recalculate results using shared code, detailed metadata, and community-standard documentation, rather than drowning them in data. The programs solve the Navier-Stokes equations via specific numerical schemes but stick to high-fidelity simulations in small domains, as realistic setups exceed even modern hardware limits.
Duality of fundamentals and applications
“So, storage and computational power are always things that could be improved, and you have to learn about these things because they are not exactly fluid mechanics. At some point, you realise that you are supposed to know a lot of things about operating systems, as well as how to manage and maintain clusters and working station machines. But there must always be a connection between the fundamentals of the subject and the application. That's why we still need to teach this subject in a fundamental way, doing equations – probably as our predecessors were doing, and prepare the engineers of the future with this perspective. I would like them to know the theory and then give the example or real application. And I like this duality - the fundamentals and the practical view.”
Though storage and computational power will continue to improve in the future, Dr Mangone emphasises that these advancements also require researchers capable of maintaining the systems, structuring private data management clusters, and justifying applications for HPC hours – skills sometimes closer to computer science than fluid mechanics. Current infrastructure spans private clusters, HPC access, and local storage, though Dr Mangani seeks advances in capacity and system design. This means PhD candidates need general computer science knowledge, learned through hands-on trial and error and related to high-performance computing, as cutting costs remains a constant goal.
Francesca Mangani concludes our interview by reiterating her original passion for what she calls "pure science": as fluid mechanics increasingly relies on supercomputing to handle the calculations and simulations, future scientists must still grasp the fundamental equations of fluid mechanics. She therefore actively promotes this principle in her teaching, ultimately hoping to bridge fundamental science and engineering applications so that future generations value deep theoretical insight for real-world challenges as much as practical utility.
Contact
Francesca Mangani
Research Unit of Fluid Mechanics
TU Wien
francesca.mangani@tuwien.ac.at
Center for Research Data Management
TU Wien
research.data@tuwien.ac.at