Comptes Rendus Mécanique (CRM)

Abderrahmane Ben Yagoub

Ridha Ziani

Bearing fault prognosis based on Cyclical Remaining Useful Life (CRUL)

Hans Ringström

Cosmology, the big bang and the BKL conjecture

Lan-Hsuan Huang

A survey on the positive mass theorem for asymptotically flat initial data sets

Gaëtan Boissonneau

Andréa Tommasi

Fabrice  Barou

Marco Antonio Lopez-Sanchez

Maurine Montagnat

Dynamic recrystallization and mechanical behavior of Mg alloy AZ31: Constraints from tensile tests with in-situ EBSD analysis

Frédéric Wrobel

An introduction to natural radiations and their detrimental effects on electronic devices

Gergely Molnár

Aurélien Doitrand

Rafael Estevez

Anthony  Gravouil

A review of characteristic lengths in the coupled criterion framework and advanced fracture models

Alice Peyraut

Martin Genet

Finite strain formulation of the discrete equilibrium gap principle: application to direct parameter estimation from large full-fields measurements

Said Derbane

Mouloud Mansouri

Salah Messast

Aboulghit  El Malki Alaoui

On the behavior of a granular soil deposit subjected to horizontal vibration. A discrete element modeling

Nicolas Bertier

Denis Gueyffier

Eric Quémerais

Comptes Rendus
Mécanique
Nicolas Bertier, Denis Gueyffier and Eric Quémerais
Introduction: from pioneering works to nowadays cutting-edge multiphysics
simulations
Volume 353 (2025), p. 1157-1159
Online since: 5 November 2025
https://doi.org/10.5802/crmeca.326
This article is licensed under the
Creative Commons Attribution 4.0 International License.
http://creativecommons.org/licenses/by/4.0/
C EN T R E
MER S ENN E
The Comptes Rendus. Mécanique are a member of the
Mersenne Center for open scientific publishing
www.centre-mersenne.org — e-ISSN : 1873-7234
Comptes Rendus. Mécanique
2025, Vol. 353, p. 1157-1159
https://doi.org/10.5802/crmeca.326
Foreword / Avant-propos
Introduction: from pioneering works to
nowadays cutting-edge multiphysics
simulations
Introduction : des travaux pionniers aux simulations
multiphysiques de pointe actuelles
Nicolas Bertier a, Denis Gueyffier b and Eric Quémerais c
a DMPE, ONERA, Université de Toulouse, 31000 Toulouse, France
b DSG, ONERA, Université Paris-Saclay, 91120 Palaiseau, France
c DMPE, ONERA, Université Paris-Saclay, 92320 Châtillon, France
E-mails: nicolas.bertier@onera.fr, denis.gueyffier@onera.fr, eric.quemerais@onera.fr
This issue explores the shift in the field of numerical simulation towards developing increas-
ingly complex software platforms and frameworks. Among the reasons which led to this trend is
the need to simulate complex industrial systems (e.g. airplanes, rockets, power plant...), systems
in extreme conditions involving a large number of physics (e.g. atmospheric reentry, plasma as-
sisted ignition, nuclear fusion...) or the need to create simulation software leveraging the poten-
tial and power of next generation supercomputers. We have gathered in this issue scientific arti-
cles showing the rich interplay between the sciences and techniques of numerical simulation,
high-performance computing and software engineering necessary to build simulation frame-
works for solving the multiphysics problems which are still intractable today. Advanced appli-
cations achieved with these recent tools are also presented in fields such as aerodynamics, en-
ergy, and in domains requiring the resolution of coupled multiphysical problems.
Before introducing the issue’s papers, we describe, starting from the early days of numerical
simulation, how the scientific community has made use of increasing computer power along with
advanced simulation software to solve more and more complex problems. New algorithms and
software architecture developed in the meantime has fueled the creation of such cutting-edge
simulation software.
Although many mathematical foundations already existed, modern numerical simulation is
often said to begin with the pioneering work of John Von Neumann at Los Alamos during the
Manhattan Project (from 1942 to 1946). The objective was to simulate nuclear explosions, leading
to numerous foundational works in numerical fluid mechanics, particularly on compressible
fluid dynamics. The wartime effort also led to the invention of early computers in several
countries (Z3 in Germany, Colossus in England, Harvard Mark I in USA...) and under the
impulsion of Von Neumann, the US Army’s ENIAC became the first computer with memory
ISSN (electronic): 1873-7234 https://comptes-rendus.academie-sciences.fr/mecanique/
1158 Nicolas Bertier, Denis Gueyffier and Eric Quémerais
storage capacity. At that time, models were extremely simplified, often associated with one-
dimensional approaches, and programs were written in machine languages (binary, octal, or
decimal) using punched cards as physical support.
The 1950s saw the establishment of several foundational elements of numerical simulation,
including the development of finite element and finite volume methods, the emergence of com-
puters with more advanced memory storage (IBM 701 in 1952 with a Von Neumann’s architec-
ture), and the advent of the Fortran language (Formula Translation, 1957). Several specialized
calculation codes (in thermal, structural, and fluid mechanics, acoustics...) were developed in
the 1960s–80s, preceding the rise of multiphysical numerical simulation as we know it today and
the development of commercial software. In parallel, computing power increased considerably,
roughly following Moore’s Law (“The number of transistors in an integrated circuit doubles ap-
proximately every 18 to 24 months at constant cost”). The late 70s and early 80s saw the rise of the
Cray family of supercomputers (Cray 1, Cray X-MP, Cray 2...) using vector processors and shared
memory. The release of the first Silicon Graphics SGI station (SGI IRIS 1000) in 1984 enabled sig-
nificant advancements in visualization and post-processing of simulation results. Researchers
started tackling three-dimensional problems using these computers.
The 2000s marked the entry into the era of distributed memory clusters which enabled mul-
tiphysics calculations (notably using MPI protocol) on an increasingly large number of cores.
This era led to better exploitation of processor chip architecture (memory levels and vectorized
floating-point operation registers). But, during this period, the trend of increasing processor
speeds was replaced with increasing the number of processor cores. With the development of
GPU, today’s supercomputers are hybrid clusters with multicore processors and GPU accelera-
tors. All these scientific and technical advances have led to a complexification of numerical tool
development, profoundly modifying the way numerical simulation tools are designed. While, un-
til the 1990s, each numerical simulation software could still be developed by one person, teams
with specialized profiles (experts of physical modeling, numerical method specialists, computer
scientists, HPC gurus...) were gradually formed. As application complexity and tool development
increased, it became increasingly difficult for a single team to possess the critical mass and ex-
pertise required to develop a tool with all the needed functionalities. This led to the development
of more elaborate simulation platforms with modular design, where each software component
can be developed by a different, specialized team. Each component can benefit from algorithms,
programming techniques, and languages best adapted to both usage and the developing team’s
culture. The complexity is partially shifted to inter-component communication, raising coupling
and data model problems. New architectures simplify code maintenance and addition of new
functionalities. In 2025, we are at a true turning point in the discipline of numerical simulation.
As several articles in this special issue attest, great maturity has been achieved in multiphysics
platform development and in related coupling challenges.
In this issue, de Chaisemartin et al. present “ArcNum”, an Arcane-based numerical framework,
and its application to porous media flow simulation. In a different field, Antoine et al. focus on
aeroelasticity with a strong emphasis on coupling challenges. Simon et al. discuss high-order
adaptive multistep coupling algorithms, with an application to partial differential equations. In
addition, a particularly important challenge for coupling is to provide efficient geometric tools
on parallel hardware architectures. This aspect is highlighted by Andrieu et al., with a work on
dynamic load-balanced point location algorithm for data mapping. Finally, considering the full
complexity of most advanced applications while limiting CPU cost requires the use of advanced
features such as dynamic mesh adaptation. Lugrin et al.’s work on a canonical dual mode
scramjet isolator is a fine illustration of this aspect.
We can now seriously consider the advent of genuine “digital twins”, using current tools. How-
ever, since 2012 and the deep learning revolution, artificial intelligence and its applications have
Nicolas Bertier, Denis Gueyffier and Eric Quémerais 1159
experienced exponential growth. Its incursion into numerical simulation is likely to revolutionize
the discipline. We believe that machine learning will become pervasive in the field of numerical
simulation, as can be attested by the rise of state of the art machine learning models like Aurora,
GraphCast and GenCast in the field of weather prediction, challenging classical atmosphere sim-
ulation models. Substantial advances can be imagined in modeling (turbulence, combustion,
liquid atomization...), in augmenting and replacing space and time discretization methods or in
the way code is generated (using copilot-like assistants, on-the-fly documentation, automatic
verification and optimization...).
Another consequence of the massive training of large language models and multimodal gener-
ative AI is the exploding demand for GPUs together with the multiplication of chips vendors. The
rise of GPUs makes it essential for numerical simulation tools to adopt versatile architectures
and to enable efficient use on both CPUs and GPUs. This has been performed elegantly using
code generation among other techniques as shown in the the paper of Lienhardt et al., “SONICS:
design and workflow of a new CFD software”.
Two important trends in the field of numerical simulation are not addressed in this issue. The
first one is cloud computing, as we believe it will greatly democratize access to HPC resources
(e.g., small businesses performing high-fidelity simulations without supercomputer infrastruc-
ture) and improve user experience. The second one is the development of quantum computers.
In the long term, they will likely become the backbone for solving numerical simulation prob-
lems, but this entails a completely different approach to numerical methods for solving PDEs
than the ones which have persisted since the first CPUs, making it a significant challenge. These
two trends will certainly be the subject of intense scientific and technical efforts in the commu-
nity, and may become the subject of a future issue of Comptes Rendus de l’Académie des Sciences.


Introduction: from pioneering works to nowadays cutting-edge multiphysics simulations

Aliénor Rivière

Stéphane Perrard

Bubble breakup probability in turbulent flows

Not a member yet