Games played under the laws of quantum mechanics dissipate less energy than their classical equivalents. This is the finding of researchers at Singapore’s Nanyang Technological University (NTU), who worked with colleagues in the UK, Austria and the US to apply the mathematics of game theory to quantum information. The researchers also found that for more complex game strategies, the quantum-classical energy difference can increase without bound, raising the possibility of a “quantum advantage” in energy dissipation.
Game theory is the field of mathematics that aims to formally understand the payoff or gains that a person or other entity (usually called an agent) will get from following a certain strategy. Concepts from game theory are often applied to studies of quantum information, especially when trying to understand whether agents who can use the laws of quantum physics can achieve a better payoff in the game.
In the latest work, which is published in Physical Review Letters, Jayne Thompson, Mile Gu and colleagues approached the problem from a different direction. Rather than focusing on differences in payoffs, they asked how much energy must be dissipated to achieve identical payoffs for games played under the laws of classical versus quantum physics. In doing so, they were guided by Landau’s principle, an important concept in thermodynamics and information theory that states that there is a minimum energy cost to erasing a piece of information.
This Landau minimum is known to hold for both classical and quantum systems. However, in practice systems will spend more than the minimum energy erasing memory to make space for new information, and this energy will be dissipated as heat. What the NTU team showed is that this extra heat dissipation can be reduced in the quantum system compared to the classical one.
Planning for future contingencies
To understand why, consider that when a classical agent creates a strategy, it must plan for all possible future contingencies. This means it stores possibilities that never occur, wasting resources. Thompson explains this with a simple analogy. Suppose you are packing to go on a day out. Because you are not sure what the weather is going to be, you must pack items to cover all possible weather outcomes. If it’s sunny, you’d like sunglasses. If it rains, you’ll need your umbrella. But if you only end up using one of these items, you’ll have wasted space in your bag.
“It turns out that the same principle applies to information,” explains Thompson. “Depending on future outcomes, some stored information may turn out to be unnecessary – yet an agent must still maintain it to stay ready for any contingency.”
For a classical system, this can be a very wasteful process. Quantum systems, however, can use superposition to store past information more efficiently. When systems in a quantum superposition are measured, they probabilistically reveal an outcome associated with only one of the states in the superposition. Hence, while superposition can be used to store both pasts, upon measurement all excess information is automatically erased “almost as if they had never stored this information at all,” Thompson explains.
The upshot is that because information erasure has close ties to energy dissipation, this gives quantum systems an energetic advantage. “This is a fantastic result focusing on the physical aspect that many other approaches neglect,” says Vlatko Vedral, a physicist at the University of Oxford, UK who was not involved in the research.
Implications of the research
Gu and Thompson say their result could have implications for the large language models (LLMs) behind popular AI tools such as ChatGPT, as it suggests there might be theoretical advantages, from an energy consumption point of view, in using quantum computers to run them.
Another, more foundational question they hope to understand regarding LLMs is the inherent asymmetry in their behaviour. “It is likely a lot more difficult for an LLM to write a book from back cover to front cover, as opposed to in the more conventional temporal order,” Thompson notes. When considered from an information-theoretic point of view, the two tasks are equivalent, making this asymmetry somewhat surprising.
In Thompson and Gu’s view, taking waste into consideration could shed light on this asymmetry. “It is likely we have to waste more information to go in one direction over the other,” Thompson says, “and we have some tools here which could be used to analyse this”.
For Vedral, the result also has philosophical implications. If quantum agents are more optimal, he says, it is “surely is telling us that the most coherent picture of the universe is the one where the agents are also quantum and not just the underlying processes that they observe”.
Complex systems model real-world behaviour that is dynamic and often unpredictable. They are challenging to simulate because of nonlinearity, where small changes in conditions can lead to disproportionately large effects, many interacting variables, which make computational modelling cumbersome, and randomness, where outcomes are probabilistic. Machine learning is a powerful tool for understanding complex systems. It can be used to find hidden relationships in high-dimensional data and predict the future state of a system based on previous data.
This research develops a novel machine learning approach for complex systems that allows the user to extract important information about collective variables in the system, referred to as inherent structural variables. The researchers used a type of machine learning tool called an autoencoder to examine snapshots of how atoms are arranged in a system at any moment (called instantaneous atomic configurations). Then, they matched each snapshot to a more stable version of that structure (an inherent structure), which represents the system’s underlying shape or pattern. The inherent structural variables enable the analysis of structural transitions and the computation of high-resolution free-energy landscapes. These are detailed maps that show how a system’s energy changes as its structure or configuration changes, helping researchers understand stability, transitions, and dynamics in complex systems.
The model is versatile, and the authors demonstrate how it can be applied to metal nanoclusters and protein structures. In the case of Au147 nanoclusters (well-organised structures made up of 147 gold atoms), the inherent structural variables reveal three main types of stable structures that the gold nanocluster can adopt. These are called fcc (face-centred cubic), Dh (decahedral), and Ih (icosahedral). These structures represent different stable states that a nanocluster can switch between, and on the high-resolution free-energy landscape, they appear as valleys. Moving from one valley to another isn’t easy, there are narrow paths or barriers between them, known as kinetic bottlenecks.
The researchers validated their machine learning model using Markov state models, which are mathematical tools that help analyse how a system moves between different states over time, and electron microscopy, which images atomic structures and can confirm that the predicted structures exist in the gold nanoclusters. The approach also captures non-equilibrium melting and freezing processes, offering insights into polymorph selection and metastable states. Scalability is demonstrated up to Au309 clusters.
The generality of the method is further demonstrated by applying it to the bradykinin peptide, a completely different type of system, identifying distinct structural motifs and transitions. Applying the method to a biological molecule provides further evidence that the machine learning approach is a flexible, powerful technique for studying many kinds of complex systems. This work contributes to machine learning strategies, as well as experimental and theoretical studies of complex systems, with potential applications across liquids, glasses, colloids, and biomolecules.
The Standard Model of particle physics is a very well-tested theory that describes the fundamental particles and their interactions. However, it does have several key limitations. For example, it doesn’t account for dark matter or why neutrinos have masses.
One of the main aims of experimental particle physics at the moment is therefore to search for signs of new physical phenomena beyond the Standard Model.
Finding something new like this would point us towards a better theoretical model of particle physics: one that can explain things that the Standard Model isn’t able to.
These searches often involve looking for rare or unexpected signals in high-energy particle collisions such as those at CERN’s Large Hadron Collider (LHC).
In a new paper published by the CMS collaboration, a new analysis method was used to search for new particles produced by proton-proton collisions at the at the LHC.
These particles would decay into two jets, but with unusual internal structure not typical of known particles like quarks or gluons.
The researchers used advanced machine learning techniques to identify jets with different substructures, applying various anomaly detection methods to maximise sensitivity to unknown signals.
Unlike traditional strategies, anomaly detection methods allow the AI models to identify anomalous patterns in the data without being provided specific simulated examples, giving them increased sensitivity to a wider range of potential new particles.
This time, they didn’t find any significant deviations from expected background values. Although no new particles were found, the results enabled the team to put several new theoretical models to the test for the first time. They were also able to set upper bounds on the production rates of several hypothetical particles.
Most importantly, the study demonstrates that machine learning can significantly enhance the sensitivity of searches for new physics, offering a powerful tool for future discoveries at the LHC.
The top Democrat on the House Science Committee is demanding that NASA halt efforts to close facilities at the Goddard Space Flight Center, arguing it jeopardizes key agency missions.
Learn more about the surprisingly well-preserved fossilized dinosaur egg uncovered in South America and what it could tell us about dinosaur nesting and bird evolution.
A new shock-and-awe movie might sharpen public concern about American vulnerability to nuclear attack and defense against it. The fictional ”House of Dynamite” shows United States leaders trying to respond to a single bolt-from-the-blue strike. Missile defense fails. The movie could spur real-world interest in better protection. In ”House of Dynamite”, an intercontinental ballistic missile launched from the […]
One of the largest antennas in NASA’s Deep Space Network was damaged in September and may be out of service for an extended period, further straining the system.
New Mexico-based mPower Technology has started automated, high-volume production of its space-grade solar modules in New York, the company announced Nov. 11 as it ramps up to meet demand from satellite constellations.
Classical clocks have to obey the second law of thermodynamics: the higher their precision, the more entropy they produce. For a while, it seemed like quantum clocks might beat this system, at least in theory. This is because although quantum fluctuations produce no entropy, if you can count those fluctuations as clock “ticks”, you can make a clock with nonzero precision. Now, however, a collaboration of researchers across Europe has pinned down where the entropy-precision trade-off balances out: it’s in the measurement process. As project leader Natalia Ares observes, “There’s no such thing as a free lunch.”
The clock the team used to demonstrate this principle consists of a pair of quantum dots coupled by a thin tunnelling barrier. In this double quantum dot system, a “tick” occurs whenever an electron tunnels from one side of the system to the other, through both dots. Applying a bias voltage gives ticks a preferred direction.
This might not seem like the most obvious kind of clock. Indeed, as an actual timekeeping device, collaboration member Florian Meier describes it as “quite bad”. However, Ares points out that although the tunnelling process is random (stochastic), the period between ticks does have a mean and a standard deviation. Hence, given enough ticks, the number of ticks recorded will tell you something about how much time has passed.
In any case, Meier adds, they were not setting out to build the most accurate clock. Instead, they wanted to build a playground to explore basic principles of energy dissipation and clock precision, and for that, it works really well. “The really cool thing I like about what they did was that with that particular setup, you can really pinpoint the entropy dissipation of the measurement somehow in this quantum dot,” says Meier, a PhD student at the Technical University in Vienna, Austria. “I think that’s really unique in the field.”
Calculating the entropy
To measure the entropy of each quantum tick, the researchers measured the voltage drop (and associated heat dissipation) for each electron tunnelling through the double quantum dot. Vivek Wadhia, a DPhil student in Ares’s lab at the University of Oxford, UK who performed many of the measurements, points out that this single unit of charge does not equate to very much entropy. However, measuring the entropy of the tunnelling electron was not the full story.
Timekeeping: Vivek Wadhia working on the clock used in the experiment. (Courtesy: Wadhia et al./APS 2025)
Because the ticks of the quantum clock were, in effect, continuously monitored by the environment, the coherence time for each quantum tunnelling event was very short. However, because the time on this clock could not be observed directly by humans – unlike, say, the hands of a mechanical clock – the researchers needed another way to measure and record each tick.
For this, they turned to the electronics they were using in the lab and compared the power in versus the power out on a macroscopic scale. “That’s the cost of our measurement, right?” says Wadhia, adding that this cost includes both the measuring and recording of each tick. He stresses that they were not trying to find the most thermodynamically efficient measurement technique: “The idea was to understand how the readout compares to the clockwork.”
This classical entropy associated with measuring and recording each tick turns out to be nine orders of magnitude larger than the quantum entropy of a tick – more than enough for the system to operate as a clock with some level of precision. “The interesting thing is that such simple systems sometimes reveal how you can maybe improve precision at a very low cost thermodynamically,” Meier says.
As a next step, Ares plans to explore different arrangements of quantum dots, using Meier’s previous theoretical work to improve the clock’s precision. “We know that, for example, clocks in nature are not that energy intensive,” Ares tells Physics World. “So clearly, for biology, it is possible to run a lot of processes with stochastic clocks.”
Rocket Lab has delayed the first launch of its reusable Neutron rocket to 2026, saying it wants to maximize the chances that the flight will be a success.
When you look back at the early days of computing, some familiar names pop up, including John von Neumann, Nicholas Metropolis and Richard Feynman. But they were not lonely pioneers – they were part of a much larger group, using mechanical and then electronic computers to do calculations that had never been possible before.
These people, many of whom were women, were the first scientific programmers and computational scientists. Skilled in the complicated operation of early computing devices, they often had degrees in maths or science, and were an integral part of research efforts. And yet, their fundamental contributions are mostly forgotten.
This was in part because of their gender – it was an age when sexism was rife, and it was standard for women to be fired from their job after getting married. However, there is another important factor that is often overlooked, even in today’s scientific community – people in technical roles are often underappreciated and underacknowledged, even though they are the ones who make research possible.
Human and mechanical computers
Originally, a “computer” was a human being who did calculations by hand or with the help of a mechanical calculator. It is thought that the world’s first computational lab was set up in 1937 at Columbia University. But it wasn’t until the Second World War that the demand for computation really exploded; with the need for artillery calculations, new technologies and code breaking.
Human computers The term “computer” originally referred to people who performed calculations by hand. Here, Kay McNulty, Alyse Snyder and Sis Stump operate the differential analyser in the basement of the Moore School of Electrical Engineering, University of Pennsylvania, circa 1942–1945. (Courtesy: US government)
In the US, the development of the atomic bomb during the Manhattan Project (established in 1943) required huge computational efforts, so it wasn’t long before the New Mexico site had a hand-computing group. Called the T-5 group of the Theoretical Division, it initially consisted of about 20 people. Most were women, including the spouses of other scientific staff. Among them was Mary Frankel, a mathematician married to physicist Stan Frankel; mathematician Augusta “Mici” Teller who was married to Edward Teller, the “father of the hydrogen bomb”; and Jean Bacher, the wife of physicist Robert Bacher.
As the war continued, the T-5 group expanded to include civilian recruits from the nearby towns and members of the Women’s Army Corps. Its staff worked around the clock, using printed mathematical tables and desk calculators in four-hour shifts – but that was not enough to keep up with the computational needs for bomb development. In the early spring of 1944, IBM punch-card machines were brought in to supplement the human power. They became so effective that the machines were soon being used for all large calculations, 24 hours a day, in three shifts.
The computational group continued to grow, and among the new recruits were Naomi Livesay and Eleonor Ewing. Livesay held an advanced degree in mathematics and had done a course in operating and programming IBM electric calculating machines, making her an ideal candidate for the T-5 division. She in turn recruited Ewing, a fellow mathematician who was a former colleague. The two young women supervised the running of the IBM machines around the clock.
The frantic pace of the T-5 group continued until the end of the war in September 1945. The development of the atomic bomb required an immense computational effort, which was made possible through hand and punch-card calculations.
Electronic computers
Shortly after the war ended, the first fully electronic, general-purpose computer – the Electronic Numerical Integrator and Computer (ENIAC) – became operational at the University of Pennsylvania, following two years of development. The project had been led by physicist John Mauchly and electrical engineer J Presper Eckert. The machine was operated and coded by six women – mathematicians Betty Jean Jennings (later Bartik); Kathleen, or Kay, McNulty (later Mauchly, then Antonelli); Frances Bilas (Spence); Marlyn Wescoff (Meltzer) and Ruth Lichterman (Teitelbaum); as well as Betty Snyder (Holberton) who had studied journalism.
World first The ENIAC was the first programmable, electronic, general-purpose digital computer. It was built at the US Army’s Ballistic Research Laboratory in 1945, then moved to the University of Pennsylvania in 1946. Its initial team of six coders and operators were all women, including Betty Jean Jennings (later Bartik – left of photo) and Frances Bilas (later Spence – right of photo). They are shown preparing the computer for Demonstration Day in February 1946. (Courtesy: US Army/ ARL Technical Library)
Polymath John von Neumann also got involved when looking for more computing power for projects at the new Los Alamos Laboratory, established in New Mexico in 1947. In fact, although originally designed to solve ballistic trajectory problems, the first problem to be run on the ENIAC was “the Los Alamos problem” – a thermonuclear feasibility calculation for Teller’s group studying the H-bomb.
Like in the Manhattan Project, several husband-and-wife teams worked on the ENIAC, the most famous being von Neumann and his wife Klara Dán, and mathematicians Adele and Herman Goldstine. Dán von Neumann in particular worked closely with Nicholas Metropolis, who alongside mathematician Stanislaw Ulam had coined the term Monte Carlo to describe numerical methods based on random sampling. Indeed, between 1948 and 1949 Dán von Neumann and Metropolis ran the first series of Monte Carlo simulations on an electronic computer.
Work began on a new machine at Los Alamos in 1948 – the Mathematical Analyzer Numerical Integrator and Automatic Computer (MANIAC) – which ran its first large-scale hydrodynamic calculation in March 1952. Many of its users were physicists, and its operators and coders included mathematicians Mary Tsingou (later Tsingou-Menzel), Marjorie Jones (Devaney) and Elaine Felix (Alei); plus Verna Ellingson (later Gardiner) and Lois Cook (Leurgans).
Early algorithms
The Los Alamos scientists tried all sorts of problems on the MANIAC, including a chess-playing program – the first documented case of a machine defeating a human at the game. However, two of these projects stand out because they had profound implications on computational science.
In 1953 the Tellers, together with Metropolis and physicists Arianna and Marshall Rosenbluth, published the seminal article “Equation of state calculations by fast computing machines” (J. Chem. Phys.21 1087). The work introduced the ideas behind the “Metropolis (later renamed Metropolis–Hastings) algorithm”, which is a Monte Carlo method that is based on the concept of “importance sampling”. (While Metropolis was involved in the development of Monte Carlo methods, it appears that he did not contribute directly to the article, but provided access to the MANIAC nightshift.) This is the progenitor of the Markov Chain Monte Carlo methods, which are widely used today throughout science and engineering.
Marshall later recalled how the research came about when he and Arianna had proposed using the MANIAC to study how solids melt (AIP Conf. Proc. 690 22).
A mind for chess Paul Stein (left) and Nicholas Metropolis play “Los Alamos” chess against the MANIAC. “Los Alamos” chess was a simplified version of the game, with the bishops removed to reduce the MANIAC’s processing time between moves. The computer still needed about 20 minutes between moves. The MANIAC became the first computer to beat a human opponent at chess in 1956. (Courtesy: US government / Los Alamos National Laboratory)
Edward Teller meanwhile had the idea of using statistical mechanics and taking ensemble averages instead of following detailed kinematics for each individual disk, and Mici helped with programming during the initial stages. However, the Rosenbluths did most of the work, with Arianna translating and programming the concepts into an algorithm.
The 1953 article is remarkable, not only because it led to the Metropolis algorithm, but also as one of the earliest examples of using a digital computer to simulate a physical system. The main innovation of this work was in developing “importance sampling”. Instead of sampling from random configurations, it samples with a bias toward physically important configurations which contribute more towards the integral.
Furthermore, the article also introduced another computational trick, known as “periodic boundary conditions” (PBCs): a set of conditions which are often used to approximate an infinitely large system by using a small part known as a “unit cell”. Both importance sampling and PBCs went on to become workhorse methods in computational physics.
In the summer of 1953, physicist Enrico Fermi, Ulam, Tsingou and physicist John Pasta also made a significant breakthrough using the MANIAC. They ran a “numerical experiment” as part of a series meant to illustrate possible uses of electronic computers in studying various physical phenomena.
The team modelled a 1D chain of oscillators with a small nonlinearity to see if it would behave as hypothesized, reaching an equilibrium with the energy redistributed equally across the modes (doi.org/10.2172/4376203). However, their work showed that this was not guaranteed for small perturbations – a non-trivial and non-intuitive observation that would not have been apparent without the simulations. It is the first example of a physics discovery made not by theoretical or experimental means, but through a computational approach. It would later lead to the discovery of solitons and integrable models, the development of chaos theory, and a deeper understanding of ergodic limits.
Although the paper says the work was done by all four scientists, Tsingou’s role was forgotten, and the results became known as the Fermi–Pasta–Ulam problem. It was not until 2008, when French physicist Thierry Dauxois advocated for giving her credit in a Physics Today article, that Tsingou’s contribution was properly acknowledged. Today the finding is called the Fermi–Pasta–Ulam–Tsingou problem.
The year 1953 also saw IBM’s first commercial, fully electronic computer – an IBM 701 – arrive at Los Alamos. Soon the theoretical division had two of these machines, which, alongside the MANIAC, gave the scientists unprecedented computing power. Among those to take advantage of the new devices were Martha Evans (whom very little is known about) and theoretical physicist Francis Harlow, who began to tackle the largely unexplored subject of computational fluid dynamics.
The idea was to use a mesh of cells through which the fluid, represented as particles, would move. This computational method made it possible to solve complex hydrodynamics problems (involving large distortions and compressions of the fluid) in 2D and 3D. Indeed, the method proved so effective that it became a standard tool in plasma physics where it has been applied to every conceivable topic from astrophysical plasmas to fusion energy.
The resulting internal Los Alamos report – The Particle-in-cell Method for Hydrodynamic Calculations, published in 1955 – showed Evans as first author and acknowledged eight people (including Evans) for the machine calculations. However, while Harlow is remembered as one of the pioneers of computational fluid dynamics, Evans was forgotten.
A clear-cut division of labour?
In an age where women had very limited access to the frontlines of research, the computational war effort brought many female researchers and technical staff in. As their contributions come more into the light, it becomes clearer that their role was not a simple clerical one.
Skilled role Operating the ENIAC required an analytical mind as well as technical skills. (Top) Irwin Goldstein setting the switches on one of the ENIAC’s function tables at the Moore School of Electrical Engineering in 1946. (Middle) Gloria Gordon (later Bolotsky – crouching) and Ester Gerston (standing) wiring the right side of the ENIAC with a new program, c. 1946. (Bottom) Glenn A Beck changing a tube on the ENIAC. Replacing a bad tube meant checking among the ENIAC’s 19,000 possibilities. (Courtesy: US Army / Harold Breaux; US Army / ARL Technical Library; US Army)
There is a view that the coders’ work was “the vital link between the physicist’s concepts (about which the coders more often than not didn’t have a clue) and their translation into a set of instructions that the computer was able to perform, in a language about which, more often than not, the physicists didn’t have a clue either”, as physicists Giovanni Battimelli and Giovanni Ciccotti wrote in 2018 (Eur. Phys. J. H43 303). But the examples we have seen show that some of the coders had a solid grasp of the physics, and some of the physicists had a good understanding of the machine operation. Rather than a skilled–non-skilled/men–women separation, the division of labour was blurred. Indeed, it was more of an effective collaboration between physicists, mathematicians and engineers.
Even in the early days of the T-5 division before electronic computers existed, Livesay and Ewing, for example, attended maths lectures from von Neumann, and introduced him to punch-card operations. As has been documented in books including Their Day in the Sun by Ruth Howes and Caroline Herzenberg, they also took part in the weekly colloquia held by J Robert Oppenheimer and other project leaders. This shows they should not be dismissed as mere human calculators and machine operators who supposedly “didn’t have a clue” about physics.
Verna Ellingson (Gardiner) is another forgotten coder who worked at Los Alamos. While little information about her can be found, she appears as the last author on a 1955 paper (Science122 465) written with Metropolis and physicist Joseph Hoffman – “Study of tumor cell populations by Monte Carlo methods”. The next year she was first author of “On certain sequences of integers defined by sieves” with mathematical physicist Roger Lazarus, Metropolis and Ulam (Mathematics Magazine29 117). She also worked with physicist George Gamow on attempts to discover the code for DNA selection of amino acids, which just shows the breadth of projects she was involved in.
Evans not only worked with Harlow but took part in a 1959 conference on self-organizing systems, where she queried AI pioneer Frank Rosenblatt on his ideas about human and machine learning. Her attendance at such a meeting, in an age when women were not common attendees, implies we should not view her as “just a coder”.
With their many and wide-ranging contributions, it is more than likely that Evans, Gardiner, Tsingou and many others were full-fledged researchers, and were perhaps even the first computational scientists. “These women were doing work that modern computational physicists in the [Los Alamos] lab’s XCP [Weapons Computational Physics] Division do,” says Nicholas Lewis, a historian at Los Alamos. “They needed a deep understanding of both the physics being studied, and of how to map the problem to the particular architecture of the machine being used.”
An evolving identity
What’s in a name Marjory Jones (later Devaney), a mathematician, shown in 1952 punching a program onto paper tape to be loaded into the MANIAC. The name of this role evolved to programmer during the 1950s. (Courtesy: US government / Los Alamos National Laboratory)
In the 1950s there was no computational physics or computer science, therefore it’s unsurprising that the practitioners of these disciplines went by different names, and their identity has evolved over the decades since.
1930s–1940s
Originally a “computer” was a person doing calculations by hand or with the help of a mechanical calculator.
Late 1940s – early 1950s
A “coder” was a person who translated mathematical concepts into a set of instructions in machine language. John von Neumann and Herman Goldstine distinguished between “coding” and “planning”, with the former being the lower-level work of turning flow diagrams into machine language (and doing the physical configuration) while the latter did the mathematical analysis of the problem.
Meanwhile, an “operator” would physically handle the computer (replacing punch cards, doing the rewiring, etc). In the late-1940s coders were also operators.
As historians note in the book ENIAC in Action this was an age where “It was hard to devise the mathematical treatment without a good knowledge of the processes of mechanical computation…It was also hard to operate the ENIAC without understanding something about the mathematical task it was undertaking.”
For the ENIAC a “programmer” was not a person but “a unit combining different sequences in a coherent computation”. The term would later shift and eventually overlap with the meaning of coder as a person’s job.
1960s
Computer scientist Margaret Hamilton, who led the development of the on-board flight software for NASA’s Apollo program, coined the term “software engineering” to distinguish the practice of designing, developing, testing and maintaining software from the engineering tasks associated with the hardware.
1980s – early 2000s
Using the term “programmer” for someone who coded computers peaked in popularity in the 1980s, but by the 2000s was replaced in favour of other job titles such as various flavours of “developer” or “software architect”.
Early 2010s
A “research software engineer” is a person who combines professional software engineering expertise with an intimate understanding of scientific research.
Overlooked then, overlooked now
Credited or not, these pioneering women and their contributions have been mostly forgotten, and only in recent decades have their roles come to light again. But why were they obscured by history in the first place?
Secrecy and sexism seem to be the main factors at play. For example, Livesay was not allowed to pursue a PhD in mathematics because she was a woman, and in the cases of the many married couples, the team contributions were attributed exclusively to the husband. The existence of the Manhattan Project was publicly announced in 1945, but documents that contain certain nuclear-weapons-related information remain classified today. Because these are likely to remain secret, we will never know the full extent of these pioneers’ contributions.
But another often overlooked reason is the widespread underappreciation of the key role of computational scientists and research software engineers, a term that was only coined just over a decade ago. Even today, these non-traditional research roles end up being undervalued. A 2022 survey by the UK Software Sustainability Institute, for example, showed that only 59% of research software engineers were named as authors, with barely a quarter (24%) mentioned in the acknowledgements or main text, while a sixth (16%) were not mentioned at all.
The separation between those who understand the physics and those who write the code, understand and operate the hardware goes back to the early days of computing (see box above), but it wasn’t entirely accurate even then. People who implement complex scientific computations are not just coders or skilled operators of supercomputers, but truly multidisciplinary scientists who have a deep understanding of the scientific problems, mathematics, computational methods and hardware.
Such people – whatever their gender – play a key role in advancing science and yet remain the unsung heroes of the discoveries their work enables. Perhaps what this story of the forgotten pioneers of computational physics tells us is that some views rooted in the 1950s are still influencing us today. It’s high time we moved on.
Gravity might be able to quantum-entangle particles even if the gravitational field itself is classical. That is the conclusion of a new study by Joseph Aziz and Richard Howl at Royal Holloway University of London. This challenges a popular view that such entanglement would necessarily imply that gravity must be quantized. This could be important in the ongoing attempt to develop a theory of quantum gravity that unites quantum mechanics with Einstein’s general theory of relativity.
“When you try to quantize the gravitational interaction in exactly the same way we tried to mathematically quantize the other forces, you end up with mathematically inconsistent results – you end up with infinities in your calculations that you can’t do anything about,” Howl tells Physics World.
“With the other interactions, we quantized them assuming they live within an independent background of classical space and time,” Howl explains. “But with quantum gravity, arguably you cannot do this [because] gravity describes space−time itself rather than something within space−time.”
Quantum entanglement occurs when two particles share linked quantum states even when separated. While it has become a powerful probe of the gravitational field, the central question is whether gravity can mediate entanglement only if it is itself quantum in nature.
General treatment
“It has generally been considered that the gravitational interaction can only entangle matter if the gravitational field is quantum,” Howl says. “We have argued that you could treat the gravitational interaction as more general than just the mediation of the gravitational field such that even if the field is classical, you could in principle entangle matter.”
Quantum field theory postulates that entanglement between masses arises through the exchange of virtual gravitons. These are hypothetical, transient quantum excitations of the gravitational field. Aziz and Howl propose that even if the field remains classical, virtual-matter processes can still generate entanglement indirectly. These processes, he says, “will persist even when the gravitational field is considered classical and could in principle allow for entanglement”.
The idea of probing the quantum nature of gravity through entanglement goes back to a suggestion by Richard Feynman in the 1950s. He envisioned placing a tiny mass in a superposition of two locations and checking whether its gravitational field was also superposed. Though elegant, the idea seemed untestable at the time.
“Recently, two proposals showed that one way you could test that the field is in a superposition (and thus quantum) is by putting two masses in a quantum superposition of two locations and seeing if they become entangled through the gravitational interaction,” says Howl. “This also seemed to be much more feasible than Feynman’s original idea.” Such experiments might use levitated diamonds, metallic spheres, or cold atoms – systems where both position and gravitational effects can be precisely controlled.
Aziz and Howl’s work, however, considers whether such entanglement could arise even if gravity is not quantum. They find that certain classical-gravity processes can in principle entangle particles, though the predicted effects are extremely small.
“These classical-gravity entangling effects are likely to be very small in near-future experiments,” Howl says. “This though is actually a good thing: it means that if we see entanglement…we can be confident that this means that gravity is quantized.”
The paper has drawn a strong response from some leading figures in the field, including Marletto at the University of Oxford, who co-developed the original idea of using gravitationally induced entanglement as a test of quantum gravity.
“The phenomenon of gravitationally induced entanglement … is a game changer in the search for quantum gravity, as it provides a way to detect quantum effects in the gravitational field indirectly, with laboratory-scale equipment,” she says. Detecting it would, she adds, “constitute the first experimental confirmation that gravity is quantum, and the first experimental refutation of Einstein’s relativity as an adequate theory of gravity”.
However, Marletto disputes Aziz and Howl’s interpretation. “No classical theory of gravity can mediate entanglement via local means, contrary to what the study purports to show,” she says. “What the study actually shows is that a classical theory with direct, non-local interactions between the quantum probes can get them entangled.” In her view, that mechanism “is not new and has been known for a long time”.
Despite the controversy, Howl and Marletto agree that experiments capable of detecting gravitationally induced entanglement would be transformative. “We see our work as strengthening the case for these proposed experiments,” Howl says. Marletto concurs that “detecting gravitationally induced entanglement will be a major milestone … and I hope and expect it will happen within the next decade.”
Howl hopes the work will encourage further discussion about quantum gravity. “It may also lead to more work on what other ways you could argue that classical gravity can lead to entanglement,” he says.
Discover how tracing space dust in Arctic seafloor sediment can help us understand how ice coverage changed over millennia, without the need for satellites.
November 10, 2025 – Washington, D.C.—The Commercial Space Federation (CSF) is pleased to welcome Seagate Space, Michael Baker International, and Lanteris Space Systems as new Associate Members. These new members […]
Connexion
