Thermal energy systems power everything from power generation to manufacturing. Yet accurate prediction under nonlinear, coupled, and geometry-heavy conditions remains stubbornly hard. Conventional simulation strategies may be accurate only within narrow ranges and often struggle when system configurations change or when measurements are sparse and noisy.

The review, published in ENGINEERING Energy, provides a structured taxonomy for applying PINN-DT to industrial thermal challenges. Authored by Sadegh Ataee and Mehran Ameri from Shahid Bahonar University of Kerman, the work synthesizes how physics-informed learning can be combined with digital-physical synchronization rather than treating AI as a standalone black box.

A central contribution is tackling ill-posed thermal problems. PINN-DT can learn solutions for strongly nonlinear heat and flow behaviors that conventional computational methods find inaccessible, improving stability when data is limited.

Another breakthrough is interpretability. By embedding fundamental constraints—such as energy conservation and fluid-dynamical relationships—directly into the training loss, the models remain physically consistent even when observations are incomplete.

Crucially for industry, the review connects predictive modeling to control. When PINN-DT is paired with Model Predictive Control (MPC), the digital twin can forecast future states, enforce operational constraints, and return optimized control signals in real time.

The most notable gap the authors address is exergy. They propose a novel physics-informed loss function derived from exergy analysis, combining the first and second laws of thermodynamics. This exergy-informed formulation is designed to improve predictive accuracy and reduce mismatch between learned dynamics and thermodynamic reality.

The framework is also presented as scalable across sectors, including supercritical CO₂ Brayton cycles, smart power grids, food processing refrigeration, and dynamic HVAC control for GPU-centric data centers.

“The development of robust physics-informed machine learning frameworks fundamentally depends on embedding appropriate physical principles through carefully designed constraint terms,” the authors emphasize, highlighting that loss-function design is not a detail—it is the engine of reliability.

With exergy-guided constraints and MPC-ready digital twins, the review sketches a roadmap for Industry 4.0 systems that minimize energy consumption while maximizing output—delivering viral, near-real-time optimization rather than slow, offline prediction.

Subject of Research: Physics-informed neural network-based digital twins for thermal energy systems (solvability and loss function design)
Article Title: Physics-informed neural network-based digital twins for thermal energy systems: A review of solvability and loss function design
News Publication Date: 10-Jun-2026
Web References: https://doi.org/10.1007/s11708-026-1049-1
References: Ataee, S., Ameri, M. Physics-informed neural network-based digital twins for thermal energy systems: A review of solvability and loss function design. ENG. Energy 20, 10491 (2026).
Image Credits: Sadegh Ataee & Mehran Ameri.

Keywords

Energy, digital twins, physics-informed neural networks, PINN-DT, thermal energy systems, exergy, model predictive control, MPC, thermodynamics

In a groundbreaking convergence of artificial intelligence and fundamental physics, researchers are harnessing the power of neural networks to tackle one of the most perplexing mysteries in modern cosmology: the Hubble tension. This persistent discrepancy in the measured rate of the universe’s expansion, a puzzle that has baffled cosmologists for years, is now being approached with novel computational tools that promise to shed new light on the very fabric of reality. The Hubble tension arises from two primary methods of measuring the universe’s expansion speed. One method relies on observations of the cosmic microwave background (CMB), the faint afterglow of the Big Bang, suggesting a slower expansion rate. The other employs standard candles like Type Ia supernovae in the local universe, indicating a faster rate. This fundamental disagreement hints at either subtle errors in our measurements or, more excitingly, the potential for new, undiscovered physics governing the cosmos.

The latest scientific frontier in this pursuit involves the application of Physics-Informed Neural Networks (PINNs), a sophisticated type of artificial intelligence that can simultaneously learn from data and adhere to the fundamental laws of physics. This innovative approach has been employed to analyze a complex model of dark energy known as Tsallis Holographic Dark Energy, while also accounting for the presence of neutrinos, elusive subatomic particles that play an subtle but important role in the universe’s evolution. By integrating physical principles directly into the learning process of the neural network, PINNs can avoid generating unphysical solutions and provide more robust and interpretable results, offering a powerful new lens through which to examine the universe’s expansion history and the enigmatic dark energy driving it.

The team behind this research has focused on a specific theoretical framework that attempts to explain the behavior of dark energy, which is responsible for the accelerating expansion of the universe. This framework, known as Tsallis Holographic Dark Energy, draws inspiration from concepts in statistical mechanics and gravity, suggesting that dark energy’s properties are linked to the way information is encoded on the boundary of our observable universe. This holographic principle, inspired by black hole thermodynamics, proposes that the complexity of the universe can be described by a lower-dimensional boundary. By exploring this theoretical avenue, the researchers are seeking to discover a dark energy model that can reconcile the conflicting measurements of the Hubble constant.

The inclusion of neutrinos in this cosmological model is another critical aspect of the investigation. While neutrinos are notoriously difficult to detect due to their weak interactions, they possess mass and contribute to the overall energy density of the universe. Their presence, however small, can subtly influence the expansion rate and the formation of cosmic structures. For a long time, neutrinos were considered massless, but experimental evidence has confirmed their mass, albeit tiny. Incorporating this crucial component into cosmological models is essential for achieving a comprehensive understanding of the universe’s dynamics, and their impact on the Hubble tension is a subject of intense scrutiny.

The methodology of using PINNs represents a significant leap forward in computational cosmology. Traditional neural networks are trained solely on data, which can sometimes lead them to overlook fundamental physical constraints or generate results that defy established scientific principles. PINNs, however, are designed with built-in knowledge of physical equations, such as Einstein’s field equations which govern gravity and spacetime. This “physics-informed” aspect guides the learning process, ensuring that the model’s predictions are not only consistent with observational data but also physically plausible, thereby increasing confidence in the findings and enabling a more profound exploration of cosmic phenomena.

By feeding their PINN with observational data that reflects the universe’s expansion history, including information about galaxies, supernovae, and the cosmic microwave background, the researchers are training the neural network to identify the parameters of the Tsallis Holographic Dark Energy model that best fit the observed universe. The AI essentially learns to navigate a complex landscape of theoretical possibilities, guided by physical laws, to pinpoint the most likely scenario that explains the cosmic expansion as we see it. This data-driven yet physics-constrained approach allows for a more efficient and accurate exploration of parameter spaces previously considered intractable for traditional analytical methods.

The results of this analysis have the potential to offer a compelling solution to the Hubble tension by suggesting a specific set of parameters for the Tsallis Holographic Dark Energy model that can bridge the gap between the early and late universe measurements of the expansion rate. If the PINN-derived parameters for this dark energy model, in conjunction with the effects of neutrinos, can successfully reconcile the conflicting Hubble constant values, it would represent a major triumph for theoretical cosmology and a significant step towards a unified understanding of our universe. Such a reconciliation could signal that the current models of dark energy and particle physics are indeed on the right track, or perhaps point towards subtle modifications needed to fit observations.

One of the most exciting implications of this work is its potential to reveal new physics. The Hubble tension might not be a simple measurement error but a genuine signal of something profound and unexpected about the universe. This could include the existence of new fundamental forces, exotic forms of matter or energy, or even modifications to Einstein’s theory of general relativity at cosmological scales. The accuracy and predictive power of the PINN, as it aligns observational data with theoretical frameworks, will be crucial in discerning whether the tension points to a known phenomenon acting in a new way or to entirely novel physics that will reshape our cosmic worldview.

The research presented here exemplifies the accelerating synergy between machine learning and fundamental science. As our datasets grow larger and our theoretical models become more intricate, AI tools like PINNs are becoming indispensable for making sense of the universe’s complexities. They enable scientists to explore vast parameter spaces, identify subtle correlations, and test intricate hypotheses that would be otherwise computationally prohibitive or even impossible to tackle. This interdisciplinary approach not only accelerates discovery but also opens up new avenues of inquiry, fostering a more dynamic and interconnected scientific landscape.

The Tsallis Holographic Dark Energy model, with its quantum information theoretical underpinnings, offers an intriguing candidate for explaining the observed cosmic acceleration. Its formulation draws on the idea that the universe’s gravitational dynamics might be related to holographic principles where the information content of a volume is encoded on its boundary. This concept, originating from black hole physics, suggests a deep connection between gravity, quantum mechanics, and thermodynamics. Applying this to dark energy allows for a dynamic and evolving nature of this mysterious component, which could naturally account for the changing expansion rate of the universe over cosmic epochs.

The crucial role of neutrinos in this context cannot be overstated. While often treated as bystanders in cosmological evolution, their collective mass and interaction potential can subtly influence the expansion rate. The inclusion of their contribution, especially when considering different neutrino mass hierarchies and interaction cross-sections, adds another layer of complexity to the cosmological model. The ability of the PINN to simultaneously constrain the parameters of both the dark energy model and the neutrino properties in a way that resolves the Hubble tension would be a significant achievement, demonstrating a profound understanding of the interconnectedness of cosmic constituents.

The potential impact of this research extends far beyond solving a single cosmological puzzle. A successful resolution of the Hubble tension could have profound implications for our understanding of fundamental physics, potentially leading to new theories of gravity, particle physics, and the very nature of dark energy. It could also pave the way for future observational programs and theoretical investigations, guiding cosmologists in their quest to unravel the remaining mysteries of the universe, such as the nature of dark matter and the origin of inflation. The implications could be as far-reaching as the universe itself.

The path forward involves rigorous testing and validation of the PINN-derived results. Scientists will need to compare these findings with independent observational datasets and explore alternative theoretical frameworks to build confidence in the proposed solution. Further refinement of the PINN architecture and training methodologies will also be crucial to enhance its accuracy and robustness. Nevertheless, this pioneering work offers a tantalizing glimpse into a future where artificial intelligence plays an increasingly central role in unlocking the universe’s deepest secrets, transforming our perception of cosmic evolution and our place within it.

Ultimately, the quest for a unified understanding of the universe is a testament to human curiosity and ingenuity. The Hubble tension, once a daunting obstacle, now stands as an invitation to explore new frontiers in physics and computation. As AI continues to evolve, its application in cosmology promises to accelerate our progress, bringing us closer to answering some of the most fundamental questions about our existence, the origins of the cosmos, and its ultimate fate. This research represents a pivotal moment, showcasing the power of intelligent algorithms to tackle the grandest scientific challenges.

The sophisticated nature of the Tsallis Holographic Dark Energy model, coupled with the intricate dynamics of neutrinos, creates a complex theoretical landscape that is ideally suited for analysis by advanced machine learning techniques. The neural network, acting as an intelligent agent, is tasked with navigating this complexity to find a set of physical parameters that can simultaneously satisfy the observed cosmic evolution and resolve the tension between early and late universe measurements of the Hubble constant. This is not simply curve fitting; it is a deep interrogation of physical reality guided by computational power.

The successful application of Physics-Informed Neural Networks in this context signifies more than just a technological advancement; it marks a paradigm shift in how cosmological research is conducted. By embedding physical laws into the learning process of artificial intelligence, scientists are creating tools that are not only data-efficient but also inherently grounded in our understanding of the universe. This fusion of data-driven discovery and physics-based reasoning is likely to become increasingly prevalent in scientific exploration, leading to more robust, efficient, and insightful scientific breakthroughs across diverse fields.

Subject of Research: The investigation of the Hubble tension, a significant discrepancy in the measured rate of the universe’s expansion, by analyzing the Tsallis Holographic Dark Energy model in the presence of neutrinos using Physics-Informed Neural Networks.

Article Title: Towards a machine learning solution for hubble tension: Physics-Informed Neural Network (PINN) analysis of Tsallis Holographic Dark Energy in presence of neutrinos.

Article References:

Yarahmadi, M., Salehi, A. Towards a machine learning solution for hubble tension: Physics-Informed Neural Network (PINN) analysis of Tsallis Holographic Dark Energy in presence of neutrinos.
Eur. Phys. J. C 85, 1301 (2025). https://doi.org/10.1140/epjc/s10052-025-14993-2

Image Credits: AI Generated

DOI: https://doi.org/10.1140/epjc/s10052-025-14993-2

Keywords: Hubble Tension, Dark Energy, Tsallis Holographic Dark Energy, Physics-Informed Neural Networks, PINN, Neutrinos, Cosmology, Machine Learning, Cosmic Expansion, Artificial Intelligence.

In a groundbreaking leap for astrophysics, a team of ingenious researchers is harnessing the power of artificial intelligence, specifically physics-informed neural networks, to probe the enigmatic interiors of neutron stars. These celestial behemoths, remnants of colossal stellar explosions, are among the most extreme and fascinating objects in the universe, boasting densities so immense that a mere teaspoon of their material would weigh more than Mount Everest. Until now, our understanding of their inner workings has been largely theoretical, shrouded in the mystery of conditions far beyond terrestrial experimentation. However, this pioneering work, published in the prestigious European Physical Journal C, promises to revolutionize our ability to listen to the faint whispers emanating from these cosmic giants, a field known as asteroseismology. By pushing the boundaries of computational physics and machine learning, scientists are developing tools that can decipher the complex vibrational patterns of neutron stars, revealing crucial details about their composition, structure, and the fundamental laws of physics that govern them. This advancement is not merely an incremental step; it represents a paradigm shift in how we explore and comprehend the universe’s most extreme environments, potentially leading to discoveries that will reshape our very understanding of matter and gravity. The complexity of these objects has long daunted physicists, offering a tantalizing yet elusive frontier for scientific inquiry. The advent of sophisticated AI techniques now provides a powerful key to unlock these cosmic puzzles, translating the subtle gravitational and electromagnetic signals into comprehensible insights about the heart of these dense stellar corpses, offering a glimpse into physics far stranger than our everyday reality.

The very essence of neutron stars makes them extraordinarily challenging to study. Formed when massive stars exhaust their nuclear fuel and collapse under their own gravity, they are compressed to densities that defy human comprehension. Protons and electrons are squeezed together to form neutrons, creating a state of matter unlike anything found on Earth. Their interiors are thought to be layered, with a solid crust, a fluid outer core, and a potentially exotic, superfluid inner core, possibly containing hyperons or even deconfined quark matter. The extreme conditions create a unique laboratory for testing theories of nuclear physics and general relativity. Traditional observational methods, while invaluable, often provide only macroscopic clues about these stars. We can measure their spin rates, their magnetic field strengths, and sometimes detect the gravitational waves they emit during mergers, but peering into their core has remained a formidable task. This is where the innovative approach of physics-informed neural networks enters the arena, offering a novel way to extrapolate from the observable to the unobservable, bridging the gap between theoretical models and concrete data with unprecedented precision and speed, thus moving beyond the limitations of traditional observational astrophysics.

Physics-informed neural networks (PINNs) are a special class of artificial intelligence designed to incorporate physical laws directly into their learning process. Unlike standard neural networks that learn from data alone, PINNs are trained using both observational data and the governing differential equations that describe the physical system being studied. This integration ensures that the network’s predictions are not only consistent with the data but also physically plausible, providing a robust and reliable framework for scientific inquiry. In the context of neutron stars, these PINNs are being trained to assimilate the physics of fluid dynamics, nuclear equations of state, and general relativity, allowing them to model the complex seismic behavior of these objects. By embedding the fundamental laws of physics into the neural network’s architecture and cost function, researchers can significantly enhance the accuracy and interpretability of the results, ensuring that the AI’s insights are grounded in established scientific principles while exploring uncharted territories of knowledge at an accelerated pace.

The concept of asteroseismology, traditionally applied to stars like our Sun, involves studying the oscillations or natural vibrations of a star. These oscillations are like giant sound waves that travel through the star’s interior, subtly altering its brightness. By analyzing the frequencies and patterns of these oscillations, astronomers can infer information about the star’s internal structure, temperature, density, and composition. Applying this technique to neutron stars presents a unique set of challenges and opportunities. The seismic modes of neutron stars are much more complex than those of ordinary stars, influenced by factors such as strong magnetic fields, superfluidity, and the extreme equation of state of matter at such high densities. This complexity, however, also means that their seismic signatures hold a wealth of information about these exotic conditions, offering a pathway to directly probe the fundamental physics at play within them, making neutron star asteroseismology a particularly exciting and sensitive probe of the universe’s most extreme physics.

The research by Tseneklidou, Torres-Forné, and Cerdá-Durán represents a significant advancement in applying PINNs to neutron star asteroseismology. They are developing computational frameworks that can efficiently simulate the seismic behavior of neutron stars and then use these simulations to train neural networks. The goal is to create AI models that can take observable data, such as potential future gravitational wave signals or electromagnetic emissions, and accurately predict the seismic modes of a neutron star. This would enable scientists to constrain the properties of neutron stars with unprecedented precision, offering direct insights into the fundamental forces and particles that govern their existence, thereby unlocking secrets long hidden within their dense cores. The sheer complexity of the physics involved necessitates powerful computational tools, and PINNs are proving to be exceptionally well-suited for this monumental task, transforming theoretical possibilities into empirical realities for astrophysical exploration.

One of the key challenges in studying neutron stars is the lack of direct observational probes of their interior. While we can observe their surface phenomena, inferring the properties of matter at densities millions of times greater than atomic nuclei is inherently difficult. The equation of state, which describes how the pressure of matter changes with density, is particularly important. Different theoretical models for the equation of state predict vastly different internal structures and seismic behaviors for neutron stars. By accurately measuring the seismic frequencies of a neutron star, scientists could differentiate between these competing models and gain a deeper understanding of the strong nuclear force and the behavior of matter under extreme pressure. This is where the predictive power of the trained PINNs becomes invaluable, acting as sophisticated interpreters of cosmic vibrations.

The training process for these physics-informed neural networks is a complex endeavor. It involves vast datasets generated from sophisticated numerical simulations of neutron star oscillations. These simulations, often performed on high-performance computing clusters, generate the reference data that the neural network learns from. However, the speed and efficiency of these simulations can be limiting, especially when exploring a wide range of possible neutron star parameters. PINNs aim to overcome this bottleneck by learning the underlying physics from these simulations and then being able to predict seismic behavior for new scenarios much faster. Furthermore, the integration of physical laws directly into the network’s architecture means that even with limited data, the network can make more reliable and physically consistent predictions, accelerating the discovery process significantly and opening up new avenues for research.

The potential implications of this research extend far beyond simply understanding neutron stars. The physics governing neutron stars touches upon fundamental questions in particle physics and cosmology. For instance, the equation of state of dense matter is intimately linked to the behavior of quarks and gluons, the fundamental constituents of protons and neutrons. Discoveries about neutron star interiors could provide crucial empirical evidence for theories of quantum chromodynamics (QCD) in the high-density regime, which are difficult to test experimentally. Moreover, neutron stars play a vital role in the evolution of galaxies, and their mergers are thought to be a significant source of heavy elements. A deeper understanding of their properties could therefore shed light on the origin of the elements in the universe.

The development of these AI-driven asteroseismology tools is also crucial for the next generation of gravitational wave observatories. Events like the merger of two neutron stars produce powerful gravitational waves that carry information about the colliding objects. Future observatories like the Einstein Telescope and LISA will be much more sensitive, enabling us to detect a wealth of such events. The ability to quickly and accurately analyze the seismic signatures imprinted on these gravitational waves will be essential for extracting the maximum scientific information from these observations, transforming raw data into profound insights about the universe’s most violent events and the exotic matter that comprises these fascinating astral bodies. This synergy between AI and gravitational wave astronomy heralds a new era of discovery.

The concept of “viral” in the context of scientific news often refers to findings that capture the public imagination due to their profound implications, their inherent wonder, or their revolutionary nature. This research, by delving into the heart of phenomena as extreme as neutron stars, and employing cutting-edge AI as its analytical engine, possesses precisely these qualities. It offers a glimpse into a realm of physics that challenges our everyday intuition, a realm where the very fabric of matter behaves in ways that are both alien and awe-inspiring. The idea of “listening” to these cosmic objects, deciphering their hidden symphonies through the intelligence of machines, carries a narrative power that resonates widely, sparking curiosity and wonder about the universe’s deepest mysteries and the transformative potential of human ingenuity.

The integration of physics into AI is not just a technical detail; it is a philosophical shift in how we approach scientific discovery. It signifies a move away from purely data-driven or purely theory-driven approaches towards a more holistic paradigm. By embedding physical laws, researchers are guiding the AI’s learning process, ensuring that its conclusions are not only statistically significant but also physically meaningful. This symbiotic relationship between AI and fundamental physics accelerates the pace of discovery, allowing scientists to explore hypotheses and scenarios that would be computationally prohibitive or conceptually challenging with traditional methods, thus paving the way for unprecedented breakthroughs in our understanding of the cosmos and the fundamental forces that shape it.

The path forward for neutron star asteroseismology using PINNs is rich with promise. Researchers will continue to refine the accuracy and efficiency of their models, incorporating more complex physical phenomena such as superfluidity and magnetic field effects. The ultimate goal is to develop tools that can, in real-time, analyze observational data and provide precise constraints on neutron star properties, potentially leading to the discovery of new states of matter or even new fundamental physics. This will require a collaborative effort between theoretical physicists, computational scientists, and observational astronomers, all working together to unravel the secrets held within these cosmic laboratories, pushing the frontiers of human knowledge ever outward and deepening our appreciation for the universe’s incredible complexity.

The visual representation accompanying this research, while perhaps not a direct observation of the neutron star itself, likely serves to illustrate the complex computational models or the abstract concepts being explored. In the realm of theoretical astrophysics and computational physics, visualizations are crucial for conveying intricate ideas and the outputs of sophisticated simulations. Such images, whether generated by AI or traditional rendering software, help bridge the gap between complex mathematical descriptions and a more intuitive understanding for a broader audience, making the abstract tangible and fostering a deeper engagement with the scientific endeavor. This particular image, devoid of direct observational context, likely serves as a metaphorical representation of the AI’s analytical journey or the intricate data structures it processes, contributing to the narrative’s visual appeal and conceptual depth.

In conclusion, the application of physics-informed neural networks to neutron star asteroseismology marks a watershed moment in astrophysics. It represents a powerful fusion of cutting-edge artificial intelligence and profound physical inquiry, offering a novel and potent tool for exploring the universe’s most extreme objects. As these AI models become more sophisticated, we can anticipate a cascade of discoveries that will illuminate the enigmatic interiors of neutron stars, deepen our comprehension of fundamental physics, and continue to expand the boundaries of human knowledge, transforming our perception of the cosmos and our place within it through insightful analysis and revolutionary technological application.

Subject of Research: Asteroseismology of neutron stars using physics-informed neural networks.

Article Title: Towards asteroseismology of neutron stars with physics-informed neural networks.

Article References:

Tseneklidou, D., Torres-Forné, A. & Cerdá-Durán, P. Towards asteroseismology of neutron stars with physics-informed neural networks.
Eur. Phys. J. C 85, 1218 (2025). https://doi.org/10.1140/epjc/s10052-025-14942-z

Image Credits: AI Generated

DOI: 10.1140/epjc/s10052-025-14942-z

Keywords: Neutron stars, asteroseismology, physics-informed neural networks, artificial intelligence, astrophysics, equation of state, dense matter, gravitational waves

Differential equations form the backbone of modern physics, encoding the fundamental laws governing dynamic systems from fluid flows to the curvature of spacetime. Among these, stiff differential equations—characterized by vastly different time scales or sensitive parameters—have notoriously resisted effective numerical treatment. Their stiffness results in computational instability and inefficiency that conventional solvers fail to overcome, particularly when these equations arise in inverse problems where the goal is to deduce unknown physical parameters or laws from empirical data. The ICCUB team’s breakthrough lies precisely in addressing these limitations by marrying machine learning’s flexibility with the rigorous constraints of physical law.

Physics-informed neural networks have emerged recently as promising tools capable of embedding physical laws into the architecture of artificial intelligence. Unlike classical neural networks that rely solely on data, PINNs incorporate differential equations directly into the loss function guiding training, ensuring solutions adhere to physical principles. This innovative approach enables PINNs not only to predict system behavior but also to infer unknown parameters by effectively solving inverse problems. However, their application to stiff equations remains complicated by difficulties in training stability and the risk of overfitting specific problem instances.

To overcome these hurdles, the ICCUB researchers introduced a twofold strategy advancing the capabilities of PINNs. The first innovation, Multi-Head (MH) training, allows the neural network to simultaneously learn a generalized solution space applicable to an entire family of differential equations rather than being confined to a single instance. This broader learning scope significantly enhances the model’s adaptability, enabling it to interpolate between different physical scenarios and maintain robustness when confronted with new challenges.

Complementing this, the team devised a method termed Unimodular Regularization (UR), inspired by mathematical concepts from differential geometry and general relativity. UR stabilizes the neural network’s learning trajectory by enforcing geometric constraints on the solution space, preventing pathological behaviors and improving generalization. Essentially, UR acts as a refined regulator that guides the PINN towards physically consistent and stable solutions, even under the most demanding computational conditions.

The efficacy of these combined techniques was demonstrated across three increasingly complex and physically meaningful systems. Beginning with the flame equation, which models reactive flows in combustion processes, and progressing to the Van der Pol oscillator, a classical nonlinear system exhibiting limit cycles and stiffness, the approach consistently produced accurate and stable solutions. Most notably, the framework tackled the Einstein Field Equations within a holographic setup—an audacious test since these equations underpin general relativity and are famously challenging to solve or invert. Here, the researchers succeeded in recovering unknown physical functions from synthetic observational data, achieving what was previously considered near-impossible.

Pedro Tarancón-Álvarez highlights the transformative potential of these developments, noting the recent surge in PINN popularity owes much to improvements in training efficiency and network design. He emphasizes the capability of PINNs to address inverse problems efficiently, offering a paradigm shift beyond traditional numerical solvers that often rely on brute-force methods or simplifying approximations.

Pablo Tejerina-Pérez elaborates on the complexity of inverse problems, comparing them to puzzles missing critical pieces that define unique and physically meaningful solutions. Traditional approaches struggle with guessing or approximating these pieces, frequently leading to ambiguous or incorrect outcomes. In contrast, the enhanced PINN system intelligently infers these missing elements by seamlessly integrating physical law constraints and advanced learning schemes, thus navigating complex solution landscapes with unprecedented precision.

This research not only propels the science of computational physics forward but also opens avenues for practical applications across diverse domains such as climate modeling, materials science, and astrophysics, where stiff differential equations abound and complex inverse problems are routine. By weaving deep learning with fundamental physics and sophisticated mathematical frameworks, the ICCUB team’s approach exemplifies the growing synergy between AI and traditional scientific disciplines.

The study was conducted in collaboration with esteemed colleagues Raúl Jiménez (ICREA-ICCUB) and Pavlos Protopapas from Harvard University. Funding support came from the Spanish Ministry of Science and Innovation and the Maria de Maeztu Excellence Programme, underscoring the high national and international recognition of the impact and potential of this research.

As machine learning continues to intertwine more deeply with fundamental science, frameworks such as this may become indispensable tools for scientists, augmenting human intuition and computational power alike. The enhanced PINNs forged by the Barcelona group provide a promising blueprint for future explorations into the unknown landscapes of physics, potentially accelerating discoveries and fostering deeper understanding of the universe’s most intricate phenomena.

Subject of Research: Not applicable

Article Title: Efficient PINNs via multi-head unimodular regularization of the solutions space

News Publication Date: 15-Aug-2025

Web References:
https://www.nature.com/articles/s42005-025-02248-1
http://dx.doi.org/10.1038/s42005-025-02248-1

Image Credits: Communications Physics

Keywords

/Physical sciences/ Physics