Cosmology has long predicted that the first generation of stars should differ strongly from those forming today. Born out of pristine gas of only hydrogen and helium, they could have reached masses between a thousand and ten thousand times that of the Sun, before collapsing after only a few million years. Such “primordial monsters” have been proposed as the seeds of the first quasars (see “Collapsing monster” image), but clear observations had until now been lacking.
An analysis of the galaxy GS 3073 using the James Webb Space Telescope (JWST) now carries an unexpectedly loud message from the first generation of stars: there is far too much nitrogen to be explained by known stellar populations. This mismatch suggests a different kind of stellar ancestor, one no longer present in our universe. It is the first indirect evidence for the long-sought primordial monsters, first proposed in the early 1960s by Fred Hoyle and William Fowler in the US, and independently by Yakov Zel’dovich and Igor Novikov in the Soviet Union, in attempts to explain the newly discovered quasars.
Black-hole powered
JWST’s near-infrared spectroscopy of GS 3073 reveals the highest nitrogen-to-oxygen ratio yet measured while surveying the universe’s first billion years. Its dense central gas contains almost as many nitrogen atoms as oxygen, while carbon and neon are comparatively modest. In addition, the galaxy has an active nucleus powered by a black hole that is already millions to hundreds of millions of times the mass of the Sun, despite the galaxy’s low metallicity.
Could a primordial monster explain GS 3073? The answer lies in how these huge stars mix and burn their fuel.
GS 3073 could offer the first chemical evidence for the largest stars the universe ever formed and to the early production of massive black holes
Simulations reveal that after an initial phase of hydrogen burning in the core, these stars ignite helium, producing large amounts of carbon and oxygen. Because the stars are so luminous and extended, their interiors are strongly convective. Hot material rises, cool material sinks and chemical elements are constantly stirred. Freshly made carbon from the helium-burning core leaks outward into a surrounding shell where hydrogen is still burning. There, a sequence of reactions known as the CNO cycle converts hydrogen into helium while steadily turning carbon into nitrogen. Over time, this process loads the outer parts of the star with nitrogen, while also moderately enhancing oxygen and neon. The heaviest elements produced in the final burning stages remain trapped in the core and never reach the surface before the star collapses.
Mass loss from such primordial stars is uncertain. Without metals, they cannot generate the strong line-driven winds familiar from massive stars today. Instead, mass may be lost through pulsations, eruptions or interactions in dense environments. But simulations allow a robust conclusion: supermassive primordial stars between roughly one thousand and ten thousand solar masses naturally produce gas with nitrogen-to-oxygen, carbon-to-oxygen and neon-to-oxygen ratios that match those measured in the dense regions of GS 3073. Stars significantly lighter or heavier than this range cannot reproduce the extreme nitrogen-to-oxygen ratio, even before carbon and neon are taken into account.
Under pressure
Radiation pressure could have supported these primordial monsters for no more than a few million years. As their cores contract and heat, photons become energetic enough to convert into electron–positron pairs, reducing the radiation pressure. For classical massive stars with masses in the range of nine to 120 times the mass of the sun, this instability leads to a thermonuclear explosion that we refer to as a supernova. By contrast, supermassive stars are so dominated by gravity due to their much larger mass that they collapse directly into black holes, without undergoing a supernova explosion.
This provides a natural path from supermassive primordial stars to the over-massive black hole now seen in GS 3073’s nucleus. In this scenario, one or a few such giants enrich the surrounding gas with nitrogen-rich material through mass loss during their lives, and leave behind black-hole seeds that later grow by accretion. If this picture is correct, GS 3073 offers the first chemical evidence for the largest stars the universe ever formed and ties them directly to the early production of massive black holes. Future JWST observations, together with next-generation ground-based telescopes, will search for more nitrogen-loud galaxies and map their chemical structures in greater detail.
On 2 July 2025, NASA’s Fermi Gamma-ray Space Telescope observed a gamma-ray burst (GRB 250702B) of a record seven hours in duration. Intriguingly, high-resolution images from the Hubble Space Telescope (HST) and the James Webb Space Telescope (JWST) revealed that the burst emerged nearly 1900 light-years from the centre of its host galaxy, near the edge of its disc. But its most unusual feature is that it was seen in X-rays a full day before any gamma rays arrived.
The high-energy transient sky is filled with a cacophony of exotic explosions produced by stellar death. Short GRBs of less than two seconds are produced by the merging of compact objects such as black holes and neutron stars. Longer GRBs are produced by the death of massive stars, with “ultralong” GRBs most often hypothesised to originate in the collapse of massive blue supergiants, as they would allow for accretion onto their central black-hole engines over a period from tens of minutes to hours.
Peculiar observations
GRB 250702B lasted for at least 25,000 seconds (7 hours), superseding the previous longest GRB 111209A by over 10,000 seconds. However, the duration alone was not enough to identify this event as a different class of GRB or as an extreme outlier. Two other observations immediately marked GRB 250702B as peculiar: the multiple gamma-ray episodes seen by Fermi and other high-energy satellites; and the soft X-rays from 0.5 to 4 keV seen by China’s Einstein Probe over a period extending a full day before gamma rays were detected.
No previous GRB is known to have been preceded by X-ray emission over such a period. Nor is it an expectation of standard GRB models, even those invoking a blue supergiant. Instead, these X-rays suggest a relativistic tidal disruption event (TDE) – the shredding of a star by a massive black hole, launching a jet that moves near the speed of light. All known relativistic TDE systems are produced by supermassive black holes weighing a million times the mass of our Sun, or more. Such black holes are found at the centre of their host galaxies, but the HST and JWST observations revealed that the transient had occurred near the edge of its host galaxy’s disc (see “Not from the nucleus” image).
This peripheral origin opens the door to a more exotic scenario involving an intermediate-mass black hole (IMBH) weighing hundreds to thousands of solar masses. IMBHs are a missing link in black-hole evolution between the stellar-mass black holes that gravitational-wave detectors frequently see merging and the supermassive black holes found at the centre of most galaxies. Alternative scenarios reduce the black-hole mass even further, and include a micro-TDE, where a star is shredded by a stellar-mass black hole, or a helium star being eaten by a stellar-mass black hole.
There is little consensus on the origin of GRB 250702B, beyond that it involved an accreting black hole
The rapid gamma-ray variability observed by Fermi and other high-energy satellites is an important clue. The time variability of relativistic jets is thought to be orders of magnitude slower than the characteristic scale set by a black hole’s Schwarzschild radius. While an intermediate-mass black hole of a few hundred solar masses is not incompatible, the observed variability is nearly 100 times faster than that seen in relativistic TDEs. By contrast, with characteristic physical scales smaller in proportion to the smaller masses of their black holes, micro-TDEs and helium-star black-hole mergers have no difficulty accommodating such short-timescale variability.
The environment of the transient also provides crucial clues into its origin. JWST spectroscopy revealed that the light from the transient and its host galaxy was emitted 8 billion years ago, when the universe was just a teenager. The galaxy is among the largest and most massive at that age in the universe, and – unusually for galaxies hosting GRBs – a massive dust lane splits its disc in half. Ongoing star formation at the transient’s location suggests a stellar-mass progenitor, as opposed to an IMBH.
Despite numerous studies, there is little consensus on the origin of GRB 250702B, beyond that it involved an accreting black hole. Its exceptional duration and early X-ray emission initially suggested a supermassive black hole, but its rapid variability and location in its host galaxy instead point to a stellar-mass black hole, with a far rarer IMBH potentially splitting the difference. Given that it is a notably rare once-every-50-years event, the wait for the next ultralong GRB may be long, but astrophysicists are optimistic that theoretical advances will disentangle the different progenitor scenarios and reveal the origin of this extraordinary transient.
George Smoot, who led the team that first measured tiny fluctuations in the cosmic microwave background (CMB) and began a revolution in cosmology, passed away in Paris on 18 September 2025.
George earned his undergraduate and doctoral degrees at the Massachusetts Institute of Technology (MIT), and then moved to Berkeley, where he held positions at Lawrence Berkeley National Laboratory (Berkeley Lab) and the Space Sciences Laboratory at the University of California, Berkeley (UC Berkeley). Though trained as a particle physicist, he switched to cosmology and developed research projects, including using differential microwave radiometers (DMRs) on U-2 spy planes to detect the dipole anisotropy of the CMB, a consequence of the motion of the Earth relative to the universe as a whole. He then devoted himself to the measurement of the CMB in detail, and this undertaking occupied him from his proposal of a satellite experiment using DMRs in 1974 to the results of the Cosmic Background Explorer (COBE) satellite in 1992. George subsequently continued research and teaching as a member of the faculty of the UC Berkeley physics department.
In 2006, the Nobel Prize committee recognised John Mather for leading a team that determined the CMB spectrum was a blackbody (arising from thermal equilibrium) to exquisite precision, and George for leading a team that detected temperature variations across the sky in the CMB at the level of one part in a hundred thousand. Those variations were signatures of the primordial density fluctuations that gave rise to galaxies, and so eventually to us. They have been called the DNA of cosmic structure and provide a remarkable window on the early universe and high-energy physics beyond our particle accelerators. The excitement caused by the COBE CMB results was dramatically expressed by Stephen Hawking, who declared them to be “the discovery of the century, if not all time.”
After the Nobel Prize, George intensified his efforts in science education and training young scientists. Indeed, on the day of the prize, George continued to teach his undergraduate introductory physics class.
George created new research institutes internationally to support young scientists. He used his prize money to found the Berkeley Center for Cosmological Physics, a joint effort between UC Berkeley and Berkeley Lab. He also started an annual Berkeley Lab summer workshop for high-school students and teachers, now in its 19th year. Later, he founded the Instituto Avanzado de Cosmología and the international Essential Cosmology for the Next Generation winter schools in Mexico, the Paris Centre for Cosmological Physics, the Institute for the Early Universe in South Korea at the world’s largest women’s university, and more. Many of the scientists trained at those institutes went on to become faculty in their home countries and internationally, and formed their own research groups.
His open online course “Gravity! From the Big Bang to Black Holes” taught nearly 100,000 students
George took special pride in the Oersted Medal awarded to him by the American Association of Physics Teachers in 2009 for “outstanding, widespread, and lasting impact” on the teaching of physics. His massive open online course “Gravity! From the Big Bang to Black Holes” with Pierre Binétruy taught nearly 100,000 students.
In his later years, George’s scientific interests spanned not only the CMB (in particular the Planck satellite), but new sensor technologies such as kinetic inductance detectors and ultrafast detectors that could open up new windows on astrophysical phenomena, gravitational waves and gravitational lensing, features in the inflationary primordial fluctuation spectrum, and dark-matter properties.
The primordial density fluctuations for which George was awarded the Nobel Prize lie at the heart of almost every aspect of cosmology. The revolution started by the COBE results led to the convergence of cosmology and particle physics, exemplified by the centrality of dark matter as a primary issue for both disciplines. George will be remembered for this, for the many students whose lives he touched and whose research he inspired, and for his advocacy of international science.
Over the past decade, many theoretical and experimental landscapes have shifted substantially. Traditional paradigms such as supersymmetry and extra dimensions – once the dominant drivers of LHC search strategies – have gradually given way to a more flexible, signature-oriented approach. The modern search programme is increasingly motivated by signals rather than full theories, providing an interesting backdrop for the return of the SEARCH conference series, which last took place in 2016. The larger and more ambitious 2025 edition attracted hundreds of participants to CERN from 20 to 24 October.
The workshop highlighted how much progress ATLAS and CMS have made in searches for long-lived particles, hidden-valley scenarios (see “Soft cloud” figure) and a host of other unconventional possibilities that now occupy centre stage. Although these ideas were once considered exotic, they have become natural extensions of models connected to cosmology, dark matter and electroweak symmetry breaking. Their experimental signatures are equally rich: displaced vertices, delayed showers, emerging jets or unusual track topologies that demand a rethinking of reconstruction strategies from the ground up.
Deep learning
The most transformative change since previous editions of SEARCH is the integration of AI-based algorithms into every layer of analysis. Deep-learning-driven b-tagging has dramatically increased sensitivity to final states involving heavy flavour, while machine learning is being embedded directly into hardware trigger systems to identify complex event features in real time. This is not technological novelty for its own sake: these tools directly expand the discovery reach of the experiments.
Novel ideas in reconstruction also stood out. Talks showcased how muon detectors can be repurposed as calorimeters to detect late-developing showers, and how tracking frameworks can be adapted to capture extremely displaced tracks that were once discarded as outliers. Such techniques illustrate a broader cultural shift: expanding the search frontier now often comes from reinterpreting detector capabilities in creative ways.
The most transformative change since previous editions of SEARCH is the integration of AI-based algorithms into every layer of analysis
Anomaly detection – the use of unsupervised or semi-supervised deep-learning models to identify data that deviate from learned patterns – was another major focus. These methods, used both offline and in level-one triggers, enable model-agnostic searches that do not rely on an explicit beyond-the-Standard-Model target. Participants noted that this is especially valuable for scenarios like quirks in dark-sector models, where realistic event-generation tools still do not exist. In these cases, anomaly detection may be the only feasible path to discovery.
The rising importance of precision was another theme threading through the discussions. The detailed understanding of detector performance achieved in recent years is unprecedented for a hadron collider. CMS’s muon calibration, which is crucial for its W-mass analysis, and ATLAS’s record-breaking jet-calibration accuracy exemplify the progress. This maturity opens the possibility that new physics could first appear as subtle deviations rather than as striking anomalies. As the era of the High-Luminosity LHC approaches, the upcoming additions of precision timing layers and advanced early-tracking capabilities will further strengthen this dimension of the search programme.
The workshop also provided a platform to explore connections between collider searches and other experimental efforts across particle physics. Strong first-order phase transitions, relevant to electroweak baryogenesis, motivated renewed interest in an additional scalar that would modify the Higgs potential. Such a particle could lie anywhere from the MeV scale up to hundreds of GeV – often below the mass ranges targeted by standard resonance searches. Alternative data-taking strategies such as data scouting and data parking offer new opportunities to probe this wide mass window systematically.
Complementarity with flavour physics at LHCb, long-lived particle searches at FASER, and precision experiments seeking electric dipole moments, axion-like particles and other ultralight states, was also highlighted. In a moment without an obvious theoretical favourite, this diversification of experimental approaches is a key strategic strength.
New directions in science are launched by new tools much more often than by new concepts
A recurring sentiment was that the LHC remains a formidable discovery machine, but the community must continue pushing its tools beyond their traditional boundaries. Many discussions at SEARCH 2025 echoed a famous remark by Freeman Dyson: “New directions in science are launched by new tools much more often than by new concepts.” The upcoming upgrades to ATLAS and CMS – precision timing, enhanced tracking earlier in the trigger chain and high-granularity readout – exemplify the kinds of new tools that can reshape the search landscape.
If SEARCH 2025 underscored the need to explore new signatures, technologies and experimental ideas, it also highlighted an equally important message: we must not lose sight of the physics questions that originally motivated the LHC programme. The hierarchy problem, the apparent fine tuning of quantum corrections to the Higgs mass that prevent it rising to the Planck scale, remains unresolved, and supersymmetry continues to offer its most compelling and robust solution by stabilising it through partner particles. With the dramatic advances in reconstruction, triggering and analysis techniques, and with the enormous increase in recorded data from Run 1 through Run 3, the time is ripe to revitalise the inclusive SUSY search programme. A comprehensive, modernised SUSY effort should be a defining element of the combined ATLAS and CMS legacy physics programme, ensuring that the field fully exploits the discovery potential of the LHC dataset accumulated so far.
The LHC’s increased collision energies have opened new territory for TeV-scale searches, but its vast datasets also provide unparalleled opportunities to thoroughly explore the electroweak scale. A new ATLAS result uses an unconventional trigger-level analysis (TLA) of the full Run 2 dataset to achieve record sensitivity to low-mass particles decaying into quarks or gluons. ATLAS employs a two-stage trigger system, with a fast hardware-based first-level trigger selecting about 100 kHz of events from the 40 MHz bunch-crossing rate, followed by a software high-level trigger (HLT) that performs detailed event reconstruction and further reduces the accepted event rate by about two orders of magnitude. By recording a much reduced event format at the trigger level, TLA preserves a substantially larger fraction of events than would normally be output by the HLT.
New particles that decay with a two-jet final state feature in many Standard Model (SM) extensions. For example, the properties of “dark mediators” that couple to both quarks and dark matter could explain the present abundance of dark matter by controlling how much of it remains after falling out of equilibrium with normal matter in the early universe. At the LHC, the coupling of dark mediators to quarks would enable both production and decay into quark–antiquark pairs. This should appear as resonances in the dijet mass distribution.
Searching for dijet resonances at low mass is challenging. Dijet production from strong interactions is one of the LHC’s most abundant signatures. Beyond requiring a precise understanding of these enormous backgrounds and the detector response, the low-mass dijet rate far exceeds what ATLAS can record. Only the most energetic dijet events can be kept, limiting conventional dijet searches to masses above approximately 1 TeV.
To access the low-mass region, ATLAS used TLA to record multi-jet events throughout Run 2. By dropping the raw detector data from the readout, these TLA events were ~200 times smaller than standard events while retaining all high-level jet and calorimeter-based variables reconstructed in real-time by the HLT.
The size reduction allowed ATLAS to record TLA events at rates of up to 27 kHz – compared to an average 1.2 kHz for the full detector readout. This rate was achieved in conjunction with the additional trigger bandwidth allocated to TLA at the end of LHC fills and a more efficient use of this bandwidth for dijet events. In Run 2, this was aided by ATLAS’s L1Topo trigger processor, which applies simple topological selections – such as angular correlations between jets – already at first level. The new result uses 1 billion dijet events, or up to 75 times the data sample available to the equivalent conventional search, achieving unprecedented statistical precision.
The new result achieves record sensitivity to low-mass particles decaying into quarks or gluons
This enormous dataset demands excellent control of systematic uncertainties. ATLAS developed a dedicated multi-step calibration for trigger-level jets, achieving a jet energy scale precision of 1 to 4%, comparable to calibrations using full detector readout. The overwhelming SM background was modelled using a data-driven fitting technique, reaching a relative precision better than 1 part in 104.
The search has found the dijet invariant-mass distribution to be consistent with the background expectation. The analysis provides numerical results that can be used to constrain any of the numerous models of dijet resonances, as well as explicit constraints on a specific dark mediator model used as a common benchmark for many ATLAS and CMS searches. The result sets ATLAS’s most stringent exclusion limits to date on the potential coupling of such a mediator to quarks, across a broad range of mediator masses reaching as low as 375 GeV (see figure 1).
The dijet TLA during Run 2 has established a foundation for an expanded trigger-level physics programme. In Run 3, trigger-level jets incorporate tracking information, allowing flavour tagging and improving jet energy resolution and robustness against pile-up. ATLAS also records trigger-level photons and uses them in combination with partial detector readout at full granularity. These and other advances in TLA should enable future ATLAS searches to probe a wider variety of signatures at the electroweak scale.
Millions of asteroids orbit the Sun. Smaller fragments often brush the Earth’s atmosphere to light up the sky as meteors. Once every few centuries, a meteoroid has sufficient size to cause regional damage, most recently the Chelyabinsk explosion that injured thousands of people in 2013, and the Tunguska event that flattened thousands of square kilometres of Siberian forest in 1908. Asteroid impacts with global consequences are vastly rarer, especially compared to the frequency with which they appear in the movies. But popular portrayals do carry a grain of truth: in case of an impending collision with Earth, nuclear deflection would be a last-resort option, with fragmentation posing the principal risk. The most important uncertainty in such a mission would be the materials properties of the asteroid – a question recently studied at CERN’s Super Proton Synchrotron (SPS), where experiments revealed that some asteroid materials may be stronger under extreme energy deposition than current models assume.
Planetary defence
“Planetary defence represents a scientific challenge,” says Karl-Georg Schlesinger, co-founder of OuSoCo, a start-up developing advanced material-response models used to benchmark large-scale nuclear deflection simulations. “The world must be able to execute a nuclear deflection mission with high confidence, yet cannot conduct a real-world test in advance. This places extraordinary demands on material and physics data.”
Accelerator facilities play a key role in understanding how asteroid material behaves under extreme conditions, providing controlled environments where impact-relevant pressures and shock conditions can be reproduced. To probe the material response directly, the team conducted experiments at CERN’s HiRadMat facility in 2024 and 2025, as a part of the Fireball collaboration with the University of Oxford. A sample of the Campo del Cielo meteorite, a metal-rich iron-nickel body, was exposed to 27 successive short, intense pulses of the 440 GeV SPS proton beam, reproducing impact-relevant shock conditions that cannot be achieved with conventional laboratory techniques.
“The material became stronger, exhibiting an increase in yield strength, and displayed a self-stabilising damping behaviour,” explains Melanie Bochmann, co-founder and co-team lead alongside Schlesinger. “Our experiments indicate that – at least for metal-rich asteroid material – a larger device than previously thought can be used without catastrophically breaking the asteroid. This keeps open an emergency option for situations involving very large objects or very short warning times, where non-nuclear methods are insufficient and where current models might assume fragmentation would limit the usable device size.”
Throughout the experiments at the SPS, the team monitored each pulse using laser Doppler vibrometry alongside temperature sensors, capturing in real time how the meteorite softened, flexed and then unexpectedly re-strengthened without breaking. This represents the first experimental evidence that metal-rich asteroid material may behave far more robustly under extreme, sudden energy loading than predicted.
The experiments could also provide valuable insights into planetary formation processes
After the SPS campaign, initial post-irradiation measurements were performed at CERN. These revealed that magnesium inclusions had been activated to produce sodium-22, a radioactive isotope that decays to produce a positron, allowing diagnostics similar to those used in medical imaging. Following these initial measurements, the irradiated meteorite has been transferred to the ISIS Neutron and Muon Source at the Rutherford Appleton Laboratory in the UK, where neutron diffraction and positron annihilation lifetime spectroscopy measurements are planned.
“These analyses are intended to examine changes in the meteorite’s internal structure caused by the irradiation and to confirm, at a microscopic level, the increase in material strength by a factor of 2.5 indicated by the experimental results,” explains Bochmann.
Complementary information can be gathered by space missions. Since NASA’s NEAR Shoemaker spacecraft successfully landed on asteroid Eros in 2001, two Japanese missions and a further US mission have visited asteroids, collecting samples and providing evidence that some asteroids are loosely bound rocky aggregates. In the next mission, NASA and ESA plan to study Apophis, an asteroid several hundreds of metres in size in each dimension that will safely pass closer to Earth than many satellites in geosynchronous orbit on 13 April 2029 – a close encounter expected only once every few thousand years.
The missions will observe how Apophis is twisted, stretched and squeezed by Earth’s gravity, providing a rare opportunity to observe asteroid-scale material response under natural tidal stresses. Bochmann and Schlesinger’s team now plan to study asteroids with a similar rocky composition.
Real-time data
“In our first experimental campaign, we focused on a metal-rich asteroid material because its more homogeneous structure is easier to control and model, and it met all the safety requirements of the experimental facility,” they explain. “This allowed us to collect, for the first time, non-destructive, real-time data on how such material responds to high-energy deposition.”
“As a next step, we plan to study more complex and rocky asteroid materials. One example is a class of meteorites called pallasites, which consist of a metal matrix similar to the meteorite material we have already studied, with up to centimetre-sized magnesium-rich crystals embedded inside. Because these objects are thought to originate from the core–mantle boundary of early planetesimals, such experiments could also provide valuable insights into planetary formation processes.”
Rohini Madhusudan Godbole, one of India’s most influential particle physicists, passed away in her hometown of Pune on 25 October 2024.
Rohini was born on 12 November 1952 to Madhusudan and Malati Godbole. Theirs was a cultured and highly educated family, and she grew up in an atmosphere of intellectual freedom and progressive ideas. Educated at the best schools and colleges in Pune, she joined the Indian Institute of Technology at Bombay, from which she graduated in 1972. She then moved to Stony Brook, where she completed her PhD in particle physics with Jack Smith in 1979. Returning to India, she worked temporarily at the Tata Institute of Fundamental Research before joining the faculty at the University of Bombay (now Mumbai). There she remained until 1997, when she moved to the Centre for High Energy Physics at the Indian Institute of Science at Bangalore (now Bengaluru). She worked there for the rest of her life, continuing after her formal retirement as an emeritus professor. It was only a few months before the end that she moved back to her hometown, to be with her family in her last days.
Rohini was a prolific researcher. She will probably be best remembered pioneering the development, with Manuel Drees, of photon structure functions for use with photon beams at future colliders, but her contributions spanned vacuum polarisation, Higgs physics, top-quark physics with polarised beams, and beyond the Standard Model physics, especially low-energy supersymmetry. She authored a well-known textbook on the latter subject with Probir Roy and Drees.
Rohini was indefatigable in promoting the cause of women in science
Rohini’s broad understanding and warm character combined to make her the best-known face of elementary particle physics from India. She worked tirelessly to promote high-energy physics inside India, organising schools and workshops, and often represented the country in international forums, such as to monitor India’s participation in the LHC and other large international collaborative experiments. Rohini was a dedicated teacher and mentor to a long series of graduate students and postdocs, and a universal elder sister or aunt for the entire community of younger particle physicists in India.
No description of Rohini can be complete without mentioning her indefatigable efforts to promote the cause of women in science. Having herself faced gender discrimination in her younger days, she was determined to ensure that young women scientists received proper opportunities and recognition. She authored two books highlighting the work of Indian women scientists, thereby setting up role models to inspire the younger generation. Even more than these books, however, her own presence and encouragement left a mark on two generations of particle physicists, in India and abroad.
Rohini’s signal contributions were recognised by many awards and distinctions. The government of India awarded her the coveted Padma Shri in 2019, and the government of France awarded her the Ordre National du Mérite in 2021, mentioning her important role in furthering scientific collaboration between India and France. But her true memorial lies in the unique place she holds in the hearts of thousands of students, collaborators, friends and acquaintances. She was an extraordinary person who carved out a niche all by herself, with her scientific talents, her indefatigable energy, her universal amiability and her indomitable will. Her loss is sorely felt.
To begin, could you explain what the muon’s magnetic moment is, and why it should be anomalous?
Particles react to magnetic fields like tiny bar magnets, depending on their mass, electric charge and spin – a sort of intrinsic angular momentum lacking a true classical analogue. These properties combine into the magnetic moment, along with a quantum-mechanical g-factor which sets the strength of the response. Dirac computed g to be precisely two for electrons, with a formula that applies equally to the other, then-unknown, leptons. We call any deviation from this value anomalous. The name stuck because the first measurements differed from Dirac’s prediction, which initially was not understood. The anomalous piece is a natural probe of new physics, as it arises entirely from quantum fluctuations that may involve as-yet unseen new particles.
What ingredients from the Standard Model go into computing g–2?
Everything. All sectors, all particles, all Standard Model (SM) forces contribute. The dominant and best quantified contributions are due to QED, having been computed through fifth order in the fine structure constant α. We are talking about two independent calculations of more than 12,000 Feynman diagrams, accounting for more than 99.9% of the total SM prediction. Interestingly, two measurements of α disagree at more than 5σ, resulting in an uncertainty of about two parts per billion. While this discrepancy needs to be resolved, it is negligible for the muon g–2 observable. The electroweak contribution was computed at the two-loop level long ago, and updated with better measured input parameters and calculations of nonperturbative effects in quark loops. The resulting uncertainty is close to 40 times smaller than that of the g–2 experiment. Then, the overall uncertainty is determined by our knowledge of the hadronic corrections, which are by far the most difficult to constrain.
What sort of hadronic effects do you have in mind here? How are they calculated?
There are two distinct effects: hadronic vacuum polarisation (HVP) and hadronic light-by-light (HLbL). The former arises at second order in α, is the larger of the two, and the largest source of uncertainty. While interacting with an external magnetic field, the muon emits a virtual photon that can further split into a quark loop before recombining. The HLbL contribution arises at third order and is now known with sufficient precision. The challenge is that loop diagrams must be computed at all virtual energies, down to where the strong force (QCD) becomes non-perturbative and quarks hadronise. There are two ways to tackle this.
Instead of computing the hadronic bubble directly, the data-driven “dispersive” approach relates it to measurable quantities, for example the cross section for electron–positron annihilation into hadrons. About 75% of the total HVP comes from e+e–→ π+π–, so the measurement errors in this channel determine the overall uncertainty. The decays of tau leptons into hadrons can also be used as inputs. Since the process is mediated by a charged W boson, instead of a photon, it requires an isospin rotation from the charged to the neutral current. At low energies, this is another challenging non-perturbative problem. While there are phenomenological estimates of this effect, no complete theoretical calculation exists – which means that the uncertainties are not fully quantified. Differing opinions on how to assess them led to controversy over the inclusion of tau decays in the SM prediction of g–2. An alternative to data-driven methods is lattice QCD, which allows for ab initio calculations of the hadronic corrections.
What does “ab initio” mean, in this context?
It means that there are no simplifying assumptions in the QCD calculation. The approximations used in the lattice formulation of QCD come with adjustable parameters and can be described by effective field theories of QCD. For example, we discretise space and time: the distance separating nearest-neighbour points is given by the lattice spacing and the effective field theory guides the approach of the lattice theory to the continuum limit, enabling controlled extrapolations. To evaluate path integrals using Monte Carlo methods, which themselves introduce statistical errors, we also rotate to imaginary time. While not affecting the HVP, this limits the quantities we can compute.
How do you ensure that the lattice predictions are unbiased?
Good question! Lattice calculations are complicated, and it is therefore important to have several results from independent groups for consolidating averages. An important cultural shift in the community is that numerical analyses are now routinely blinded to avoid confirmation bias, making agreements more meaningful. This shifts the focus from central values to systematic errors. For our 2025 White Paper (WP25), the main lattice inputs for HVP were obtained from blinded analyses.
How did you construct the SM prediction for your 2025 White Paper?
To summarise how the SM prediction in WP25 was obtained, sufficiently precise lattice results for HVP arrived just in time. Since measurements of the e+e–→ π+π– channel are presently in disagreement with each other, the 2025 prediction solely relied on the lattice average for the HVP. In contrast, the 2020 White Paper (WP20) prediction employed the data-driven method, as the lattice-QCD results were not precise enough to weigh in.
With the experiment’s expected precision jump, it seemed vital for theory to follow suit
While the theory error in WP25 is larger than in WP20, it is a realistic assessment of present uncertainties, which we know how to improve. I stress that the combination of the SM theory error being four times larger than the experimental one and the remaining puzzles, particularly on the data-driven side, means that the question “Does the SM account for the experimental value of the muon’s anomalous magnetic moment?” has not yet been satisfactorily answered. Given the high level of activity, this will, however, happen soon.
Where are the tensions between lattice QCD, data-driven predictions and experimental measurements?
All g–2 experiments are beautifully consistent, and the lattice-based WP25 prediction differs from them by less than one standard deviation. At present, we don’t know if the data-driven method agrees with lattice QCD due to the differences in the e+e–→ π+π– measurements. In particular, the 2023 CMD-3 results from the Budker Institute of Nuclear Physics are compatible with lattice results, but disagree with CMD-2, KLOE, BaBar, BESIII and SND, which formed the basis for WP20. All the experimental collaborations are now working on new analyses. BaBar is expected to release a new e+e–→ π+π– result soon, and others, including Belle II, will follow. There is also ongoing work on radiative corrections and Monte Carlo generators, both of which are important in solving this puzzle. Once the dust settles, we will see whether the new data-driven evaluation agrees with the lattice average and the g–2 experiment. Either way, this may yield profound insights.
How did the Muon g–2 Theory Initiative come into being?
The first spark came when I received a visiting appointment from Fermilab, offering resources to organise meetings and workshops. At the time, my collaborators and I were gearing up to calculate the HVP in lattice QCD, and the Fermilab g–2 experiment was about to start. With the experiment’s expected precision jump, it seemed vital for theory to follow suit by bringing together communities working on different approaches to the SM contributions, with the goal of pooling our knowledge, reducing theoretical uncertainties and providing reliable predictions.
As Fermilab received my idea positively, I contacted the RBC collaboration and Christoph Lehner joined me with great enthusiasm to shape the effort. We recruited leaders in the experimental and theoretical communities to our Steering Committee. Its role is to coordinate efforts, organise workshops to bring the community together and provide the structure to map out scientific directions and decide on the next steps.
What were the main challenges you faced in coordinating such a complex collaboration?
With so many authors and such high stakes, disagreements naturally arise. In WP20, a consensus was emerging around the data-driven method. The challenge was to come up with a realistic and conservative error estimate, given the up to 3σ tensions between different data sets, including the two most precise measurements of e+e–→ π+π– at the time.
As we were finalising our WP20, the picture was unsettled by a new lattice calculation from the Budapest–Marseille–Wuppertal (BMW) collaboration, consistent with earlier lattice results but far more precise. While the value was famously in tension with data-driven methods, the preprint also presented a calculation of the “intermediate window” contribution to the HVP– about 30% of the total – which disagreed with a published RBC/UKQCD result and with data-driven evaluations (CERN Courier March/April 2025 p21). Since BMW was still updating their results and the paper wasn’t yet published, we described the result but excluded it from our SM prediction. Later, in 2023, further complications came from the CMD-3 measurement.
Consolidation between lattice results was first observed for the intermediate window contribution, in 2022 and 2023. This, in turn, revealed a tension with the corresponding data-driven evaluations. Results for the difficult-to-compute long-distance contributions arrived in late fall 2024, yielding consolidated lattice averages for the total HVP, where we had to sort out a few subtleties. This was intense – a lot of work in very little time.
On the data-driven side, we faced the aforementioned tensions between the e+e–→ π+π– cross-section measurements. In light of these discrepancies, consensus was reached that we would not attempt a new data-driven average of HVP for WP25, leaving it for the next White Paper. Real conflict arose on the assessment of the quality of the uncertainty estimates for HVP contributions from tau decays and on whether to include them.
And how did you navigate these disagreements?
When the discussions around the assessment of tau-decay uncertainties stopped to converge, we proposed a conflict resolution procedure using the Steering Committee (SC) as the arbitration body, which all authors signed. If a conflict is brought to the SC for resolution, SC members first engage all parties involved to seek resolution. If none is found, the SC makes a recommendation and, if appropriate, the differing scientific viewpoints may be reflected in the document, followed by the recommendation. In the end, just having a conflict-resolution process in place was really helpful. While the SC negotiated a couple of presentation issues, the major disagreements were resolved without triggering the process.
The goal of WP25 was to wrap up a prediction before the announcement of the final Fermilab g–2 measurement. Adopting an internal conflict-resolution process was essential in getting our result out just in time, six days before the deadline.
Lattice QCD has really come of age
What other observables can benefit from advances in lattice QCD?
There are many, and their number is growing – lattice QCD has really come of age. Lattice QCD has been used for years to provide precise predictions of the hadronic parameters needed to describe weak processes, such as decay constants and form factors. A classic example, relevant to the LHC experiments, is the rare decay Bs→ μ+μ–, where, thanks to lattice QCD calculations of the Bs-meson decay constant, the SM prediction is more precise than current experimental measurements. While precision continues to improve with refined methods, the lattice community is broadening the scope with new theoretical frameworks and improved computational methods, enabling calculations once out of reach – such as the (smeared) R-ratio, inclusive decay rates and PDFs.
Some have argued that the good agreement between lattice–QCD and the final measurement of Fermilab’s muon g–2 experiment means that the g–2 anomaly has now been solved. However, this dramatically oversimplifies the situation: the magnetic moment of the muon remains an intriguing puzzle.
The extraordinary precision of 127 parts per billion (ppb) achieved at Fermilab deserves to be matched by an equally impressive theoretical prediction. At 530 ppb, theory is currently the limiting factor in any comparison. This is the longer-term goal that the Muon g–2 Theory Initiative is now working towards, with inputs from all possible sources (see “How I learnt to stop worrying and love QCD predictions“). In the near future, it will not be possible to reach this precision with lattice QCD alone. Other approaches are needed to make a competitive Standard Model prediction.
Tensions remain
Essentially, all of the uncertainty in g–2 arises from the hadronic vacuum polarisation (HVP) – a quantum correction whereby a radiated virtual photon briefly transforms into a hadronic state before being reabsorbed. Historically, HVP has been evaluated by applying a dispersion relation to cross sections for hadron production in electron–positron collisions, but this method was displaced by lattice–QCD calculations in the theory initiative’s most recent white paper. The lattice community must be congratulated for the level of agreement that has been reached between groups working independently (CERN Courier July/August 2025 p7). By contrast, data-driven predictions are at present inconsistent across the experiments in the low-energy region; even if results from the CMD-3 experiment are excluded as an outlier, tensions remain, suggesting that some systematic errors may not have been completely addressed (CERN Courier March/April 2025 p21). Could a novel experimental technique help resolve the confusion?
The MUonE collaboration proposes a completely independent approach based on a new experimental method. In MUonE, we will determine the running of the electromagnetic coupling, a fundamental quantity that is driven by the same kinds of quantum fluctuations as muon g–2. We will extract it from a precise measurement of the differential cross section for elastic scattering of muons from electrons as a function of the momentum transferred.
MUonE is a relatively inexpensive experiment that we can set up in the existing M2 beamline in CERN’s North Area, already home to the AMBER and NA64-µ experiments. Three years of running, within the conditions of M2 parameters and the performance of the MUonE detector, would reach a statistical precision of approximately 180 ppb with a comparable level of systematic uncertainty.
MUonE will take advantage of silicon sensors that are already being developed for the CMS tracker upgrade. From the results, we will be able to use a dispersion relation to extract HVP’s contribution to g–2. Perhaps more importantly, however, as our method directly measures a function that is part of the lattice calculation, we can directly verify that method. The big challenge will be to keep the systematic uncertainties in the measurement small enough. However, MUonE does not suffer from the intrinsic problem that existing data-driven techniques have, which is that they must numerically integrate over the sharp peaks of hadron production by low-energy resonances. In contrast, the function derived from the space-like process that it will measure is smooth and well-behaved.
Piecing the puzzle
CERN was the origin of the first brilliant muon g–2 measurements starting back in the 1950s (CERN Courier September/October 2024 p53), and now the laboratory has an opportunity to put another important piece into the g–2 puzzle through the MUonE project. Another component of great importance in this domain will be the new g-2/EDM experiment planned for J-PARC, which will also be performed in completely different conditions, and therefore with very different systematics to the Fermilab experiment.
Despite decades of searches, experiments have yet to find evidence for a new particle that could account for dark matter on its own. This has strengthened interest in richer “dark-sector” scenarios featuring multiple new states and interactions, potentially analogous to those of the Standard Model (SM). The CMS collaboration targeted one of the most distinctive possible signatures of a dark strong force in proton–proton collisions: a dense, nearly isotropic cloud of low-momentum particles known as a soft unclustered energy pattern (SUEP).
Searches in the LHC proton–proton collision data for events with many low-momentum particles are plagued by overwhelming backgrounds from pileup and soft QCD interactions. The CMS collaboration has recently overcome this challenge by using large-radius clusters of charged particle tracks and relying on quantities that characterise the expected isotropy of SUEP decays.
The 125 GeV Higgs boson serves in many theoretical models as a natural mediator between the SM and a hidden sector, and current experimental constraints still leave room for exotic decays. Motivated by this possibility, CMS focused on Higgs-boson production in association with a vector (W or Z) boson that decays into leptons. While these modes account for < 1% of Higgs bosons produced at the LHC, the leptons provide significant handles for triggering and background suppression.
Rather than relying on SM simulations, which face modelling and statistical challenges for such soft interactions, the background was extrapolated from events with low isotropy or relatively few charged-particle tracks per cluster, using a method that accounts for small correlations between the quantities used in the extrapolation. To validate the approach, an orthogonal sample of events with a high-momentum photon was studied, taking advantage of the Higgs boson’s minuscule coupling to photons and the similarity of background processes in W/Z + jet and photon + jet events that could mimic a SUEP signal.
The data in the search region, consisting of events with a W or Z boson candidate and many isotropically distributed charged particles, was found to be consistent with the SM expectation. Stringent limits were placed on the branching ratio of the 125 GeV Higgs boson decaying to a SUEP shower for a wide range of parameters (see figure 1).
This analysis complements a previous CMS search that primarily targeted much heavier mediators produced via gluon fusion, improving limits on the H → SUEP branching ratio by two orders of magnitude. It additionally provides model-agnostic limits and detailed reinterpretation recipes, maximising the usability of this data for testing alternative theoretical frameworks.
SUEP signatures are not unique to the benchmark scenarios under scrutiny. They naturally emerge in hidden-valley models, where mediators connect the SM to a new, otherwise isolated sector. If the hidden states interact through a “dark QCD”, proton–proton collisions would trigger a crowded cascade of dark partons rather than the familiar collimated showers.
Crucially, unlike in ordinary QCD – where the coupling quickly weakens at energies above confinement – the dark coupling could remain large well beyond its typically low confinement scale. This sustained strong coupling would drive frequent interactions and efficiently redistribute momentum, producing an almost isotropic radiation pattern. As the system cooled, it would then hadronise into numerous soft dark hadrons whose decays back to SM particles would retain this softness and isotropy – yielding the characteristic SUEP probed by CMS.
To provide the best experiences, we use technologies like cookies to store and/or access device information. Consenting to these technologies will allow us to process data such as browsing behavior or unique IDs on this site. Not consenting or withdrawing consent, may adversely affect certain features and functions.
Functional
Always active
The technical storage or access is strictly necessary for the legitimate purpose of enabling the use of a specific service explicitly requested by the subscriber or user, or for the sole purpose of carrying out the transmission of a communication over an electronic communications network.
Preferences
The technical storage or access is necessary for the legitimate purpose of storing preferences that are not requested by the subscriber or user.
Statistics
The technical storage or access that is used exclusively for statistical purposes.The technical storage or access that is used exclusively for anonymous statistical purposes. Without a subpoena, voluntary compliance on the part of your Internet Service Provider, or additional records from a third party, information stored or retrieved for this purpose alone cannot usually be used to identify you.
Marketing
The technical storage or access is required to create user profiles to send advertising, or to track the user on a website or across several websites for similar marketing purposes.
