Comsol -leaderboard other pages
Home

Topics

Reset your password

Please enter the e-mail address you used to register to reset your password

Registration complete

Thank you for registering
If you'd like to change your details at any time, please visit My account

Close

Welcome to the dark web

Cosmological filaments form the backbone of the cosmic web, the vast, interconnected network that defines the universe on the largest scales. Stretching across tens to hundreds of millions of light-years, they link galaxies and galaxy clusters along the pathways where matter assembles under gravity. They may also hold the key to one of the deepest questions in modern physics: the nature of dark matter.

For astrophysicists, filaments first drew attention as a potential reservoir of missing baryons. Big Bang nucleo­synthesis and precision measurements of the cosmic microwave background agree on how much ordinary matter the universe should hold, but the census of stars, galaxies and hot gas comes up short. The leading explanation is a warm, diffuse gas permeating cosmic filaments, too faint to detect in any single observation but increasingly accessible through statistical techniques at X-ray and radio wavelengths.

More recently, dark-matter hunters have begun to recognise the potential of filaments as probes of new physics. Filaments are not only vast but overwhelmingly dark-matter-dominated, with lower astrophysical backgrounds than traditional search targets such as the galactic centre. New simulations are pinning down their dark-matter density profiles with enough precision to make quantitative predictions, and recent theoretical work has opened detection channels that could turn these structures into laboratories for physics beyond the Standard Model.

Dark matter and the cosmic web

Our scientific understanding of stars and the structures they inhabit has grown remarkably over the past century. We now know that galaxies are vast collections of stars, and clusters are collections of galaxies. These immense systems do not float randomly; they are woven into an intricate “cosmic web” resembling that of a spider. Gravity shapes this web and governs the motion of the celestial bodies within it. Yet many observations defy expectations. Galaxies rotate too quickly, clusters bend light too strongly, and the cosmic web holds together with more gravitational pull than visible matter would allow. Something unseen must be at work. A new, invisible “dark matter” component must dominate the mass of the universe.

Dark-matter candidates in the spotlight

Dark matter accounts for roughly 85% of the matter in the cosmos, and about 27% of its content once dark energy is included, yet its nature remains unknown. Several well-motivated candidates have emerged, each predicting distinct signatures that indirect searches, including those targeting cosmic filaments, could probe. Weakly interacting massive particles, sterile neutrinos, primordial black holes and axions are among the most prominent.

Weakly interacting massive particles These hypothetical particles naturally arise in several extensions of the Standard Model and possess two defining features: they are massive, and they interact only through gravity and the weak force.

Sterile neutrinos Unlike the three known active neutrino species, they do not interact through the weak force. Their existence is motivated by extensions of the Standard Model that aim to explain both neutrino masses and the matter–antimatter asymmetry of the universe.

Primordial black holes Unlike stellar black holes, which form from collapsing stars, primordial black holes are hypothetical relics of the early universe, born from the collapse of exceptionally dense regions of matter moments after the Big Bang.

Axions Originally proposed to solve the strong CP problem, axions are hypothetical particles whose production mechanism can account for the observed dark-matter abundance, elegantly linking two of modern physics’ greatest mysteries.

Dark matter accounts for roughly 85% of the matter in the cosmos and dictates how cosmic structures form and evolve. Yet, despite decades of international effort and extraordinary experimental ingenuity, its nature remains a puzzle. The Standard Model of particle physics, describing all known fundamental particles, can’t account for the observational effects of dark matter. In response, theorists have proposed a wide range of models that include dark-matter candidates (see “Dark-matter candidates in the spotlight” panel). A well-motivated dark-matter theory, one that truly excites theorists, typically meets three criteria. First, it accounts for the observed cosmic abundance of dark matter. Second, it yields clear, testable predictions. And third, it resolves multiple open questions in fundamental physics.

Rich landscape

While the theoretical landscape is rich, testing it requires identifying cosmic environments where dark matter’s signatures might be detectable. One of the most powerful strategies is indirect detection – the search for faint cosmic messengers produced when dark matter annihilates, decays or interacts with ordinary matter. These signatures may appear as electromagnetic waves, neutrinos or charged cosmic rays. Observing these messengers requires high sensitivity and careful modelling of both the dark-matter signal and the astrophysical backgrounds. Progress, therefore, depends on close collaboration between particle physicists, astrophysicists and cosmologists, integrating theoretical predictions with multi-messenger observations.

Choosing optimal targets is crucial for indirect dark-matter searches. Traditional efforts have focused on the galactic centre and on dwarf satellite galaxies of the Milky Way. The galactic centre is expected to host the highest dark-matter density, but it also contains intense and complex astrophysical backgrounds, which is why the origin of a long-debated gamma-ray excess observed by Fermi-LAT remains uncertain (see “Gamma-ray excess” figure). Dwarf galaxies, by contrast, are dark-matter-dominated and relatively free of astrophysical emission. However, their stellar populations are orders of magnitude smaller than that of the Milky Way. This limits the available kinematic tracers – observables whose spatial distribution correlates with the underlying matter density field – and leads to sizable uncertainties in the predicted signals.

Unconventional environments

Recently, unconventional but promising probes have gained attention, such as cosmological filaments. Filaments are a natural outcome of anisotropic gravitational collapse in an expanding universe. Matter can collapse under gravity in some directions while still expanding in others, producing elongated structures that are bound across their width but continue to grow along their length. Not all cosmic filaments are alike. Some lie within galaxy clusters, linking individual galaxies over relatively short distances. Others extend far beyond cluster boundaries, forming vast inter-cluster bridges that connect galaxy clusters and even superclusters across tens and hundreds of megaparsecs. The longer the filament, the thinner and more diffuse it tends to be. This reflects the way gravity draws matter out of underdense regions and funnels it into elongated bridges between massive nodes.

Gamma-ray excess
Gamma-ray excess Since its launch in 2008, the Large Area Telescope of the Fermi Gamma-ray Space Telescope (Fermi-LAT) has observed an excess of gamma rays from the galactic centre. A map of 1.0–3.16 GeV gamma rays is shown unprocessed on the left, and with known sources removed on the right. Credit: T Linden

Together, galaxy clusters and the diffuse filaments that connect them form the cosmic web and make up most of the baryonic matter. Yet the very properties that make filaments so fundamental to cosmic structure also make them extraordinarily difficult to observe. Their emission is faint, diffuse and easily overwhelmed by brighter astrophysical sources, posing a major challenge for direct detection across the electromagnetic spectrum.

To overcome this limitation, astronomers have turned to a statistical technique known as “image stacking”. In stacking analyses, many observations of similar systems are superimposed. Any emission associated with filaments then adds coherently, while random noise and unrelated astrophysical signals average away. The result is a significant enhancement in sensitivity, allowing extremely weak, extended emission to emerge that otherwise would remain invisible.

A potent technique

The power of this approach relies on numbers: the larger the sample that can be stacked, the stronger and more reliable the resulting signal. Image stacking is therefore a potent but data-hungry technique, one that becomes increasingly effective as modern surveys deliver ever-larger datasets. This requirement poses a particular challenge for filaments, whose precise locations are generally unknown. Since cosmological filaments connect massive structures, a natural strategy is to use galaxy clusters as signposts: by stacking observations of regions between pairs of clusters, the faint emission from the filamentary bridges that link them can be statistically enhanced.

Cluster catalogues have expanded dramatically over the past decade. Today, surveys based on optical imaging, weak gravitational lensing and the Sunyaev–Zel’dovich effect, in which scattering with high-energy electrons distorts the cosmic microwave background, collectively identify tens of thousands of clusters across the sky. While progress is remarkable, it may still fall short of what is needed to robustly detect the extremely faint emission expected from typical filaments. This limitation motivates the search for alternative tracers. A reliable proxy for galaxy clusters available in far greater numbers, potentially in the millions, would enormously increase the statistical power of stacking analyses.

Stacked maps
Stacked maps From left to right, the panels show radio maps from the GLEAM survey at 154, 118 and 88 MHz, radio data from OVRO–LWA at 73 MHz and X-ray emission from ROSAT in the 0.1–2.4 keV band. The x and y axes are the normalised pair separation unit and the colour bars indicate brightness temperature (in kelvin) for the radio maps and count rate for the ROSAT data. The two bright dots in all five panels represent the pair of stacked luminous red galaxies, connected by the stacked filamentary emission. Credit: T Vernstrom et al. 2021 MNRAS 505 4178

Particularly effective proxies for galaxy clusters are luminous red galaxies (LRGs). These massive, early-type galaxies have been observed and catalogued for decades and are known to be excellent tracers of the large-scale structure of the universe. LRGs typically reside in, or near, the centres of galaxy clusters, making them reliable signposts of the densest regions of the cosmic web. Pairs of LRGs that are close to one another in the sky and in physical distance can therefore be used as proxies for nearby cluster pairs. Statistically, such pairs are likely to be connected by inter-cluster bridges or filaments, even if the filaments themselves cannot be directly identified in individual observations.

By applying this stacking technique to pairs of LRGs drawn from the Sloan Digital Sky Survey, whose catalogues contain millions of such galaxies, together with radio maps from the GLEAM and OVRO-LWA surveys, researchers have identified an intriguing anomaly. The radio emission associated with stacked filaments (see “Stacked maps” figure) exceeds theoretical predictions for diffuse filamentary gas by more than an order of magnitude.

Simulations, observations and theory
Simulations, observations and theory Graph above: radial density profiles in cosmological filaments for dark matter (black), gas (green), stars (orange) and black holes (pink), derived using the EAGLE simulations. Solid lines indicate the mean density profiles inside filaments, dotted lines the corresponding “β-model fits”, and dashed lines mark the mean density of each component in the full simulation volume. Table above: observed and expected dark-matter-induced radio emission at four frequencies, assuming the benchmark scenario of 30 GeV dark matter decaying into electron–positron pairs and producing synchrotron radiation in filamentary magnetic fields. Credit: E Pinetti et al. 2025 arXiv:2504.08025

One possible interpretation is that this excess arises from secondary radiation produced by dark matter (see “Simulations, observations and theory” figure). In this scenario, weakly interacting massive particles with masses of a few GeV decay into electrons, which then spiral through filament magnetic fields and emit synchrotron radiation at radio wavelengths. For the magnetic field strengths inferred in the stacking analysis, the amplitude of the observed signal is consistent with that expected from a dark-matter flux of this kind.

As with other anomalies, this interpretation remains debated. A more conventional explanation attributes the emission to astrophysical particle acceleration in strong accretion shocks, generated as matter falls into filaments and galaxy clusters. While shocks can in principle produce radio synchrotron emission, reproducing the observed excess appears to require acceleration efficiencies higher than those typically assumed in simulations. Significant uncertainties persist in filament properties, such as their magnetic field strengths and shock characteristics, which complicate the modelling of expected signals and remain an active area of research.

Cosmic filaments may also open a window onto more exotic dark-matter scenarios. Recent work has shown that if heavy dark matter decays into gravitons – the hypothetical quantum carriers of the gravitational interaction – these can convert into photons via the Gertsenshtein effect (see “Graviton-to-photon” figure), closely analogous to the Primakoff conversion of axions, as they propagate through the large-scale magnetic fields threading filaments. This process generates an irreducible extragalactic gamma-ray background, allowing such scenarios to be constrained with Fermi-LAT data and offering promising sensitivity for future gamma-ray observatories.

A bright future for the dark universe

For millennia, humanity has been inspired by the starry sky. Philosophers, poets and scientists alike have gazed upward, their minds filled with questions, joy and awe. Dante, one of Italy’s greatest poets, expressed this enduring fascination in the closing line of Inferno in The Divine Comedy:

“E quindi uscimmo a riveder le stelle”

“And thence we came forth to see again the stars”

Graviton-to-photon
Graviton-to-photon Illustration of a novel dark-matter search strategy: dark matter decays into gravitons, which propagate through magnetised cosmic filaments and convert into photons (left panel). This process generates a diffuse photon signal (right panel, for 250 nGauss magnetic fields). Different coloured curves correspond to different dark-matter masses, compared with the Fermi-LAT isotropic gamma-ray background (grey band) and the projected sensitivity of the Advanced Particle-astrophysics Telescope (APT, dashed). Credit: D I Dunsky et al. 2025 arXiv:2503.19019

Centuries after Dante, the sky continues to guard many of its secrets. However, we are now entering a golden era for indirect dark-matter searches. Future facilities, most notably the Square Kilometre Array (SKA), currently under construction in South Africa and Australia, will deliver unprecedented sensitivity to the diffuse structures of the cosmic web, and may soon be capable of directly imaging large filaments, characterising their properties and turning these vast structures into powerful probes of physics beyond the Standard Model.

These observational advances are being matched by progress on the theoretical front. Cosmological simulations are reaching new levels of realism, while the growing use of machine-learning and artificial-intelligence techniques is beginning to transform how filamentary structures are identified, modelled and interpreted. These developments promise a far more precise characterisation of filament properties, sharpening their role as laboratories for fundamental physics. The cosmic web may not keep its secrets much longer.

The age of the blockbuster-scale radiopharmaceutical

While radioactivity and radionuclides have been used in medicine for decades, radiopharma­ceutical drugs have only recently reached the rank of a pharma blockbuster, with more than one billion US dollars in annual sales. The clearest example is lutetium-177-based therapy, which has moved into routine use for prostate and neuroendocrine cancers, is being investigated as a first-line treatment and generates revenues previously unseen in nuclear medicine (see “How radiopharma­ceuticals work” panel).

On the production side, accelerators are a key source of innovation beyond the long-established nuclear reactors. Scaling up these technologies, however, remains challenging. No single laboratory can combine megawatt beams, advanced target engineering and full radiological infrastructure, and the products literally decay on the shelf.

How radiopharmaceuticals work

Modern radiopharmaceuticals combine a radioactive isotope with a biologically active molecule that targets specific cells in the body. The compound is typically injected into the bloodstream, where the biological component guides it to a tumour by binding to a particular protein expressed on cancer cells. Once attached, the radioactive isotope delivers a highly localised dose of radiation that damages or destroys the targeted cells while largely sparing surrounding tissue.

The type of radiation emitted determines how the compound is used. Gamma rays and photons from positron annihilation can escape the body and be detected externally for diagnostic imaging, while Auger electrons, beta particles or alpha particles deposit their energy over very short distances, making them effective for therapy. Increasingly, the same isotope–molecule combination can be used both to image disease and to treat it – an approach known as theranostics.

Many of the technologies now limiting medical radionuclide supply – high-power targets, isotope separation, beam reliability – were originally developed for nuclear and high-energy physics (HEP). These fields will remain essential not because they created today’s tools, but because future medical radionuclide production pushes those tools into regimes that only nuclear physics and HEP routinely explore.

Physicists must address four bottlenecks to meet the growing demand for radiopharmaceuticals, each of which cuts differently across the European accelerator landscape (see “The European landscape” figure). Overcoming these bottlenecks would stabilise the supply of existing treatments and open the door to R&D for entirely new diagnostic and therapeutic isotopes. In that sense, the infrastructure choices made today will shape what kinds of cancer treatments are possible a decade from now.

1. Targets that cannot survive megawatt beams

Large-scale accelerator facilities push beams to extreme power densities in order to generate secondary particles or rare isotopes, placing materials under intense thermal and radiation stress. This regime is familiar from spallation neutron sources, neutrino-production targets and radioactive-ion beam facilities, where target integrity and remote handling are central design challenges rather than secondary considerations. As medical radionuclide production scales up, it increasingly operates in this same regime, where target survivability under sustained continuous-wave irradiation becomes the primary limiting factor. Beyond a certain point, targets degrade, deform or fail faster than they can be replaced, turning material endurance, target design and monitoring – rather than accelerator capability – into the dominant constraint on production.

The European landscape
The European landscape Accelerator facilities producing medical radionuclides across Europe, plotted by year of first beam against beam power. A dotted halo flags on-site mass separation, essential for the most demanding therapeutic isotopes. Three projects with announced power but no first-beam date appear on the right. Credit: T Stora

In practice, this challenge unfolds in two forms, depending on whether a single- or a two-stage target is used. In the two-stage configuration, an intense primary beam is first converted into neutrons or photons, and those secondary particles then irradiate production targets from which the desired radionuclides are generated.

At the very front end of the supply chain, these spallation targets and beam converters – which transform protons into neutrons or electrons into photons – are among the most demanding components in the entire supply chain, because their performance and lifetime cannot be decoupled from beam power. A prominent example is the European Spallation Source (ESS) in Lund, Sweden, which will soon operate the world’s most intense long-pulse neutron beams, where a rotating, helium-cooled tungsten target enables sustained high-power operation (see “Under the beam” image). The pulsed regime introduces additional challenges from shock waves and material fatigue, which must be addressed to ensure safe operations. Likewise, the IFMIF-DONES facility in Granada will combine a 5 MW deuteron beam with a fast-flowing liquid lithium target to generate intense fast-neutron fluxes. While their primary missions lie in fundamental, material and multidisciplinary research, the extreme beam powers and neutron fluxes of these facilities also make them potentially capable of supporting radionuclide production as a secondary, but societally important, application.

Similarly, single-stage targets used for radionuclide production must withstand extreme heat loads and power densities, with geometries that depend on the type and energy of the incident particles. Active cooling, using either gases or liquids, is essential to remove heat, and careful optimisation of both beam delivery and target design can translate directly into higher radionuclide production yields.

Under the beam
Under the beam The ESS target wheel. Blocks of tungsten fitted into the radial channels absorb the 5 MW proton beam and release neutrons via spallation. Continuous rotation and helium cooling prevent any single point from melting under the load. Credit: ESS

A good example is the development of high-power bismuth targets, in either liquid or solid form, such as those proposed at RIKEN in Japan, which can handle alpha beams of around 10 kW and enable the production of At-211 in batches approaching 100 GBq – far beyond the gigabecquerel-scale batches typical of present-day facilities. RIKEN is Japan’s largest comprehensive research institute, operating major accelerator and nuclear-physics facilities with a long-standing programme in medical radio­nuclide production. In Europe, comparable facilities are under construction, such as SPIRAL2 in Caen, France, or at SCK CEN in Mol, Belgium, where ISOL@MYRRHA will deliver a 100 MeV proton beam of up to 500 µA onto a target dedicated to the production of medical and research radionuclides (see “Sorting the haul” figure).

Additional challenges arise for highly radioactive targets such as Ra-226, a long-lived precursor used in the production of the alpha-emitting medical isotopes Ac-225 or Ra-224, or for targets designed for high-power operation and mass separation, which must remain structurally and chemically stable at temperatures approaching 2500 °C.

Even when targets survive extreme beam power, increased production immediately creates a second constraint: the radionuclides they produce are only useful if they can be isolated in the required purity – a challenge that defines the next bottleneck.

2. Radionuclides that cannot be separated

When radioactive products are chemically indistinguishable from unwanted isotopes, separation must rely on their mass rather than on their chemical properties. Without such separation, many promising medical radionuclides cannot be used at all, regardless of how efficiently they are produced, because they remain too dilute or come with long-lived impurities. This challenge is familiar from isotope production for nuclear and particle physics, where ion sources, mass separators and laser techniques are used to extract rare species from intense backgrounds.

A dedicated collaborative infrastructure was established at CERN with the creation of CERN-MEDICIS, which received support from the CERN & Society Foundation and pioneers isotope mass separation for medical applications. Commissioned in 2017, MEDICIS brings together expertise in target production, purification, radiopharmaceutical development and clinical use, linking large-scale physics infrastructure directly to biomedical research.

Building on techniques originally developed at ISOLDE, the MEDICIS programme has adapted isotope mass separation specifically for medical needs. This has enabled access to radionuclides that are otherwise unavailable, or available only at insufficient purity, achieving high molar activities essential for theranostic and thera­peutic applications.

Sorting the haul
Sorting the haul The layout of the future ISOL@MYRRHA facility at SCK CEN in Mol, Belgium, with the beam line running from right to left. A 100 MeV proton beam strikes a thick target to produce rare radioactive nuclei. These are extracted as singly charged ions, selectively excited by the ISOL@MYRRHA Resonant Ionization Laser Source (IMRILS), cooled and bunched in a radiofrequency quadrupole cooler-buncher (RFQ), and mass-separated to deliver pure radioactive ion beams (RIBs) for research and medical applications. Credit: SCK CEN

In contrast to radiochemical separation, which typically achieves efficiencies above 95%, mass-separation efficiencies vary widely, ranging from a few percent to around 70%, depending on the isotope and production route. As a result, scaling up production by mass separation is not a matter of incremental optimisation: gains in yield must be traded against beam power, target lifetime and radiation handling, placing practical limits on what facilities can deliver for many isotopes.

Significant progress has nevertheless been achieved through advances in target design, compact ion sources and resonant laser ionisation, with efficiencies above 50% now considered high performance for these physics-driven separation processes. Operating at higher beam intensities further opens the possibility of separating radionuclides directly from reactor or cyclotron targets, but doing so requires accelerator conditions and radiation handling capabilities that are only available at large-scale research infrastructures. Because mass separation is both technically demanding and intrinsically inefficient, access to purified radionuclides cannot be scaled locally, making coordination across large research infrastructures unavoidable.

In Europe, this need for coordination has been addressed through a networked approach rather than a single flagship facility. The PRISMAP programme (H2020 grant #101008571), and its follow-up PRISMAP+, bring together accelerators, reactors, isotope separation facilities and biomedical hubs across national borders, allowing researchers to access purified radionuclides that no single site could reliably supply on its own (see “Special delivery” figure). By pooling infrastructure, expertise and scheduling through competitive calls, PRISMAP has lowered the barrier for biomedical researchers to work with non-conventional radionuclides, while preserving the efficiency and safety constraints imposed by large-scale physics infrastructure. It is becoming the European medical radionuclides programme that biomedical research lacked.

The same mismatch between research-scale tools and medical-scale demand reappears in accelerator design itself – the focus of the next bottleneck.

3. Machines optimised for experiments, not production

Accelerators developed for fundamental research are typically optimised for peak performance, flexibility and discovery-driven operation. Medical radionuclide production, by contrast, demands continuous, predictable delivery, exposing a growing mismatch between machines designed for experiments and those required for routine isotope production. In practice, the requirements of medical radionuclide production change dramatically between exploratory research and routine clinical supply, shifting the emphasis from flexibility and peak performance towards reliability, uptime and predictable delivery.

Scaling requires not only new radionuclides, but integrated infrastructures

Improving multi-user operation and reducing maintenance downtime can already deliver substantial gains in effective output, even without higher beam power. Intense beams of light ions and electrons are available today using cyclotrons and linear accelerators, while new concepts are being developed to better match medical needs, including synchrotrons for alpha particles and high-power electron sources for photon-based production routes.

Even when suitable accelerators exist, however, delivering medically usable radionuclides depends on meeting regulatory, dosimetric and logistical constraints that lie beyond the machine itself.

4. Production without regulatory viability

Unlike research isotopes, radiopharmaceuticals must meet strict regulatory, dosimetric and logistical requirements. Even small uncertainties in nuclear data, impurity levels or processing routes can prevent a radionuclide from reaching patients, regardless of its therapeutic promise. At this stage, scale is limited not by beam power or yield, but by whether a radionuclide can be licensed, transported and used safely in the clinic.

The case of lutetium-177 (Lu-177) illustrates both the challenge and the opportunity. Its success has been central to the emergence of modern radiotherapeutic drugs, with compounds targeting prostate and neuroendocrine cancers now used routinely in clinical practice. This success rests not only on biological targeting, but on a carefully controlled production chain that meets pharmaceutical standards.

Special delivery
Special delivery Flows of medical radionuclides within the PRISMAP network. Production facilities ship purified isotopes to biomedical hubs across Europe and on to users as far as the US. The diversity of species and routes reflects the fact that no single site can supply biomedical research with every radionuclide it needs. Credit: D De Biasio and T Stora

The mode of production of Lu-177 – either by direct neutron capture on Lu-176 or via an indirect route through Yb-176 – highlights the complexity of aligning nuclear physics, infrastructure and regulation. Choices made upstream determine radionuclidic purity, waste streams and processing requirements downstream, all of which feed directly into licensing and clinical acceptance. Lu-177’s physical and radiological properties, including its suitability for both imaging and therapy, and a half-life of about 6.64 days that is compatible with existing medical logistics, have helped it integrate into an established supply-chain organisation. This combination has enabled treatments to reach large patient populations and commercial scale.

Meeting these requirements demands substantial investment beyond irradiation itself. Facilities must support radiochemical processing, quality control, dosimetry and specialised logistics, often in shielded hot-cell environments. Recent upgrades to such infrastructure have shown that production capacity can increase significantly once these downstream constraints are addressed. At the same time, improved nuclear and biological data have revealed that even modest discrepancies in decay properties or radiation dose can have major consequences for treatment planning, licensing and waste management.

Programmes such as PRISMAP+ have begun to address this bottleneck by providing access to novel treatment radionuclides, including beta emitters that extend established therapies such as Lu-177. By generating data on production quality and radiochemical behaviour early in development, these programmes help determine which radionuclides can realistically progress from research to routine clinical use. Crucially, regulatory constraints feed back into accelerator choice, target design and separation strategy: decisions taken at the level of beam energy, target material and purification method determine whether a radionuclide can ever meet clinical purity, waste and licensing requirements.

Taken together, these constraints show why regulatory viability is itself a bottleneck. Scaling radiopharmaceuticals requires not only new radionuclides, but integrated infrastructures in which production, processing, regulation and clinical deployment are addressed together. As radiopharmaceuticals move further into mainstream oncology, success will depend on sustained collaboration between large-scale research infrastructures, regulators, clinicians and industrial partners – none of which can solve the problem alone.

• The author dedicates this article to the memory of Mark Rayner, who shaped its structure and much of its prose.

Neutrinos on the clock

Born of one kind, a neutrino can die another. Its three flavours, electron, muon and tau, do not correspond to states of definite mass, but to quantum superpositions of three distinct masses. As neutrinos propagate, the mixture reshuffles and the flavour at arrival can differ from the one at production. None of this is predicted by the Standard Model, making the observation of neutrino oscillations one of the clearest signals of physics beyond it.

Neutrino oscillations provide a unique probe of new physics, acting as an interferometer that is sensitive to neutrino mass differences down to the sub-eV level. Precise measurements at the next-generation accelerator-based oscillation experiments, Hyper-Kamiokande in Japan and DUNE in the US, are poised to answer several critical questions. What is the ordering of the three neutrino masses, given their two measured mass-squared differences? Do neutrinos and antineutrinos oscillate differently? Are there additional, as yet undetected, neutrino states? These long-baseline neutrino facilities, in which a beam of neutrinos is sent to a detector hundreds of kilometres away, will produce much larger datasets than current-generation experiments. However, they suffer from a fundamental limitation: we do not know the precise energy or intensity of the neutrino beams when they set out. Reaching ultimate precision on neutrino-oscillation parameters is therefore no longer a matter of statistics, but one of messy nuclear-physics questions related to the details of weak-interaction cross sections and proton-induced hadron production.

In focus
In focus A magnetic horn installed in the beamline that fed CERN’s Gargamelle bubble chamber, where neutral currents were discovered in 1973. Invented by Simon van der Meer, the device uses a pulsed current to focus charged pions into a beam. The pions decay in flight to produce neutrinos. Credit: CERN

Modern neutrino beams use the famous “magnetic horn” design, developed by Simon van der Meer at CERN in 1961 (see “In focus” image). Protons strike a target to produce pions, which the horn focuses into a volume for them to decay to neutrinos and leptons. The trouble is, the resulting neutrino beam covers a wide range of energies (about 0.5 GeV and 2.5 GeV for Hyper-K and DUNE, respectively), with a shape and intensity that depend on the details of tough-to-model proton–nucleus collisions. To make matters worse, the broadband neutrino flux forces the neutrino energy to be estimated from the products of neutrino–nucleus interactions, which are notoriously difficult to model accurately. Together, these challenges form a barrier to ultimate precision in neutrino-oscillation measurements.

There is, in principle, a way around both problems: neutrino tagging. Proposed by Bruno Pontecorvo in 1979, this technique associates a measurement of the four momenta of the pion and muon in a π+ → μ+νμ decay with a measurement of a neutrino in a downstream detector. Four-momentum conservation then fixes the neutrino kinematics event-by-event. As a result, the neutrino energy is known for each interaction and the flux is perfectly constrained, nullifying the key challenges for neutrino-oscillation experiments and producing well-controlled muon, pion and kaon beams as byproducts. The idea was first attempted in the 1990s at the dedicated Tagged Neutrino Facility (TNF) at the Serpukhov accelerator in Protvino (see “Dream, deferred” image). TNF recorded two candidate events in its brief pilot run, before the dissolution of the Soviet Union brought the work to a halt.

Dream, deferred
Dream, deferred The neutrino detector of the Tagged Neutrino Facility at the Serpukhov accelerator in Protvino, Russia. Built in the early 1990s, the facility was the first to attempt neutrino tagging and collected a handful of candidate events in a short pilot run. The dissolution of the Soviet Union ended the programme before it could mature. Credit: CERN Courier Nov 1993 p28

The downside with neutrino tagging is one of scale: for every neutrino seen in a massive, 100-tonne detector close to the beam, there are about 1013 pion decays. To collect a reasonable 105 neutrino interactions per year, one would therefore need to identify at least 1011 individual muons per second and successfully identify the minute fraction of them that are associated with observed neutrinos. Such a measurement demands beamline detectors with timing resolutions of 10 to 100 ps and a neutrino detector with sub-ns timing resolution, with the beamline detectors operating in a high-radiation environment.

Promising performance

These challenges proved too much for the 1990s, and the idea lay dormant for three decades after the closure of TNF. However, a revolution in detector and electronics technology has since changed what is possible (see “The fast-timing revolution” panel) – beginning with the NA62 experiment at CERN.

The fast-timing revolution

R&D for the high-luminosity phase of the LHC are expected to push fast-timing sensors beyond the performance of NA62’s GigaTracker. With bunches crossing every 25 ns and up to 200 proton collisions in each, the ATLAS and CMS upgrades require timing resolutions below 50 ps to disentangle overlapping vertices. In a silicon detector, incoming particles release small electric charges, with internal electric fields then steering them toward electrodes to be collected and measured. Faster timing demands more charge, a shorter drift for the signal to form quickly and smaller collection electrodes for sharper and stronger pulses. In conventional planar sensors, these requirements are often in conflict with one another. Since the collected charge is proportional to the sensor thickness, which also sets the drift distance, thinner sensors give faster signals but fewer carriers. Potential solutions span a wide range of architectures. Low-gain avalanche detectors (LGADs), for instance, combine thin, 50 μm sensors with a gain layer to amplify charge, compensating for the lower number of carriers produced in a thin substrate while preserving a short drift. They achieve resolutions of 20 to 50 ps and will equip the new ATLAS and CMS timing layers.

Many roads to fast timing

For the upgrade of the LHCb vertex locator, which must resist radiation levels of 1016 to 1017 1 MeV neutron equivalents per cm2 (neq/cm2), a gain layer would erode too quickly. Three-dimensional sensors sidestep the problem geometrically. Their electrodes run along the sides of each pixel rather than on the top and bottom surfaces, so charges move sideways over distances below 50 μm while the sensor remains thick enough to generate large signals. Initially developed for radiation-hard pixel detectors for ATLAS and CMS, these sensors were later redesigned for timing and have reached resolutions as good as 10 ps. The remaining challenge lies in the electronics. Readout chips must match the sensors’ speed and radiation tolerance, and recent prototypes in 28 nm CMOS have achieved 30 ps resolution over areas of a few mm2. Large-area designs are currently underway.

Depleted monolithic active pixel sensors integrate the sensor and readout electronics on the same chip. With resolutions of 10 to 200 ps and pixels smaller than 100 × 100 μm2, they are much cheaper than three-dimensional sensors and LGADs, and therefore better suited to instrumenting large surfaces and achieving a lower material budget. Some R&D initiatives are also exploring the use of Cherenkov radiation to detect charged particles, a process much faster than ionisation in silicon. In these detectors, the prompt Cherenkov light is converted into photoelectrons and amplified in a thin gaseous detector, producing signals with time resolutions of a few tens of picoseconds. While highly promising, extending this approach to finely segmented detectors operating at very high rates remains an open problem. All these new tracking technologies offer promising perspectives beyond high-energy physics, for example in real-time monitoring of the proton and ion beams used in cancer therapy.

In order to study the very rare kaon decay K+ → π+νν, the NA62 collaboration faced a similar timing challenge in the late 2010s. At the time, pixel detectors, mainly developed for experiments at the LHC, were only recording the position of particles, their time being given by the proton bunch crossing (50–25 ns). New R&D was started to face the challenge of integrating timing capabilities into every pixel (of which there are more than one thousand per cm2). Within a few years, the “TDCPix” chip was designed (see “Pixel timekeeper” image), achieving a hit time resolution of 130 ps and starting a new field of 4D tracking (measuring particle trajectories in space and time). The price to pay for this performance was a significant increase in the power density absorbed by the pixel, exceeding 2W/cm2. Absorbing this power required developing an innovative cooling technology: a 200 μm-thin silicon plate integrating a dense microfluidic cooling circuit. These two innovations led to the GigaTracker beam spectrometer (see “Fourth dimension” figure) – the first 4D tracking detector in high-energy physics, which has been in operation since 2015.

Pixel timekeeper
Pixel timekeeper A single TDCPix readout chip is connected to its silicon sensor and wire-bonded to a readout card. Each chip contains 1800 pixels of 300 × 300 μm2, with a built-in time-to-digital converter that timestamps particle hits with a resolution of 130 ps. Ten chips tile each GigaTracker station. Credit: NA62 collaboration

With the GigaTracker in hand, the NA62 collaboration achieved the main goal of measuring the K → πνν decay, and was able to put neutrino tagging to the test. The facility’s high-intensity kaon beam also serves as a neutrino source, since the kaons predominantly decay as K+ → μ+ν. Due to the intensity of the neutrino beam and its mean energy of 40 GeV, a non-negligible number of neutrinos interact in the experiment’s electromagnetic calorimeter, a 20-tonne volume of liquid krypton. The analysis of data collected in 2022 revealed one neutrino interaction candidate that could be matched to a detected parent decay. The neutrino’s energy was estimated to be 52 GeV, with a record relative precision of 0.3%. For reference, with a few exceptions (such as pion and kaon decays at rest), neutrino energies from conventional neutrino beams are known with an uncertainty of at least 10%, and not event-by-event.

Fourth dimension
Fourth dimension One of the stations from NA62’s GigaTracker, the first 4D tracking detector for high-energy physics. Designed to track around 109 beam particles per second, the detector has operated in the NA62 beamline since 2015. Reflected in its surface is Zuzana Kučerová, GigaTracker’s deputy project leader. Credit: V Moncorgé/CPPM/CNRS Images

NA62’s proof-of-concept for neutrino tagging, combined with the broader advance of fast-timing detectors, have together made it possible to revisit the original TNF idea. Developed in parallel with the first tagged-neutrino analysis by NA62, the NuTag collaboration investigated the conditions under which a tagged beam would enable measurements inaccessible with conventional neutrino beams. These efforts have led to the proposed nuSCOPE facility at CERN, which emerged within CERN’s Physics Beyond Colliders study group by combining the tagged-beam concept from NuTag with the slow-extraction-driven monitored neutrino beam pioneered by the ENUBET collaboration. Rather than using a pulsed magnetic horn, the ENUBET setup relies on slow extraction from the SPS and lines the decay tunnel with particle detectors that identify the charged leptons produced with neutrinos, constraining the flux at the percent level.

Legacy measurements

The idea of nuSCOPE echoes that of TNF (see “Beam to neutrino” figure). The first step is to direct a slow-extracted proton beam from the SPS onto a target to produce secondary pions and kaons that are then momentum-selected, using a series of dipoles and quadrupoles, to form an 8.5 GeV meson beam with a narrow momentum range. The mesons then traverse a set of ultra-fast detectors before decaying to predominantly muons and neutrinos. The muons reach a second set of fast detectors, whilst a few neutrinos interact in a dedicated detector 25 metres downstream. With sufficient timing and spatial resolution, each neutrino interaction can be associated with a measured individual meson decay. For the first time, the energy of the incoming neutrino would be known at the sub-percent level on an event-by-event basis. Measurements of neutrino cross sections, currently the dominant source of systematic uncertainty projected for DUNE and Hyper-Kamiokande, may then reach an accuracy of about 1%. Such datasets could serve as reference, “legacy” measurements for neutrino physics for decades to come.

Beam to neutrino
Beam to neutrino The layout of the proposed nuSCOPE facility at CERN. SPS protons produce a secondary beam of kaons and pions, which are momentum-selected and tracked before entering a 40 m decay tunnel. There, they may decay into a muon and a neutrino. The muon is reconstructed in a downstream spectrometer, while the neutrino interacts in a dedicated detector further on. Credit: nuSCOPE collaboration

The implications extend well beyond standard oscillation physics. Short-baseline oscillations induced by sterile neutrinos, for instance, could produce rapid patterns that would get washed out by energy smearing in conventional beams. The facility would also deliver intense, well-characterised muon and pion beams, opening additional avenues for rare process searches and precision measurements. Looking further ahead, one can even imagine how such techniques might reshape future long-baseline experiments. Depending on what DUNE, Hyper-Kamiokande and the reactor-based JUNO experiment in China will discover, the next leap in precision may not come just from higher intensities, but from beams whose properties are known with exquisite accuracy.

Directing a decade

What would you say were CERN’s scientific and technological highlights of the past 10 years?

I would start by highlighting the excellent performance of the accelerator complex, the experiments and the computing infrastructure, all of which have gone well beyond forecasts. Thanks to these achievements, and to the increasing sophistication of the data analyses, the Laboratory’s scientific output has far surpassed what one could have anticipated, both in breadth and depth. The diversity of results across all our facilities is impressive, and physics sensitivities have exceeded even the most optimistic expectations. Notably, some studies originally foreseen for the High-Luminosity LHC (HL-LHC) phase are already being carried out today.

Secondly, the approval of the HL-LHC by the CERN Council in June 2016 marked a major milestone for the future of the LHC programme, and paved the way for substantial progress in upgrading both the accelerator complex and the experiments. As a first step, the successful Long Shutdown 2, carried out during the challenging COVID-19 period, enabled significant upgrades to the injectors and the completion of the Phase-1 upgrades of the four experiments.

Following the establishment of the Physics Beyond Colliders (PBC) Study Group in 2016, we have strengthened the Laboratory’s programme of physics complementary to the LHC, notably with the approval of new experiments and projects such as FASER, SND@LHC and a high-intensity beam-dump facility at the North Area. The Neutrino Platform has remained a focal point for the neutrino community in Europe and beyond, enabling key detector R&D, technology demonstration and prototyping activities, as well as the construction of two large cryostats for the DUNE experiment at the Long-Baseline Neutrino Facility in the United States.

The extraordinary work towards the Future Circular Collider (FCC) over the past 10 years, culminating in the highly successful Feasibility Study, laid the foundations for a brilliant long-term future for CERN and the field, and served as a key input to the ongoing update of the European Strategy for Particle Physics. In December 2025, the European Strategy Group recommended the FCC-ee as the next flagship facility at CERN, a pivotal step for the future of particle physics. Chapeau to the entire CERN community for these extraordinary achievements!

You were CERN’s first female Director-General. When you first took on the role, did you find the focus on your gender a bit frustrating?

I cannot say that I found it frustrating. What surprised me was that people would tell me I was a role model. I did not consider myself a model of anything, so I was somewhat embarrassed by that description. But then I told myself that, if the fact that a woman held the position of CERN Director-General could encourage young women to pursue a career in science, I was happy to play that role.

Did your experience as Director-General differ from what you expected?

From the outset, it was an extraordinary experience, far broader than I had expected. I had to deal with a myriad of matters: not only scientific and technical ones, but also finances and human resources, environment and sustainability, relationships with governments, the public and the media, as well as the daily management of a laboratory as complex as CERN.

The extraordinary work towards the FCC has laid the foundations for a brilliant long-term future for CERN

Fortunately, I was surrounded by an excellent management team of directors, department heads and project leaders, and could count on the support, competence, enthusiasm and dedication of the entire CERN community. I am very grateful to the CERN Member States for the unique opportunity to serve as Director-General for two terms, and for their sustained support and trust throughout the 10 years.

There is one thing I would never have imagined I would need to do: raise substantial funding from the private sector. As a physicist, that was entirely outside my experience…

Despite that, you announced one billion US dollars of private funding for FCC shortly before the end of your tenure. How did you approach donations?

When I started my first term as CERN Director-General, I had no experience whatsoever with donations. I had not raised a single penny in my entire life!

I remember that, in the context of the Science Gateway, I considered hiring a professional fundraiser at some point. But then I realised that the salaries of these people were astronomical, so I abandoned the idea.

Over the years, I learned a lot from experience. I learned that fundraising is very much a matter of personal connections, and I was lucky enough to have some very good ones before becoming CERN Director-General. I also learned that donors love bold, ambitious projects like the FCC, projects that enable major progress for humanity. Finally, donors engage if they trust the institution, in this case CERN, but also if they trust the person. I think they trusted me.

What made you decide to pursue private donations, and how did the policy framework for accepting them come about?

Both the Science Gateway and FCC are extremely challenging projects in their respective areas. I quickly realised that projects of this kind can only be achieved with exceptional levels – and therefore sources – of funding.

The Science Gateway was a dream I had from the beginning of my first term, in 2016. At the time, I realised we could host 150,000 visitors annually, compared with 300,000 requests. That seemed a shame to me. The limitation came from the number of visitor areas, which led to the idea of a new building – a dedicated space that could expand our offer to the public. CERN’s budget provides only limited funding for education, communication and outreach initiatives, and the money raised annually through the CERN & Society Foundation was far from enough to cover a 100-million-Swiss-franc project. So we had to undertake a dedicated fundraising campaign.

The Science Gateway was a dream I had from the beginning of my first term

Then we obtained a 48-million-US-dollar donation from the “Eric and Wendy Schmidt Fund for Strategic Innovation” for the “Next Generation Triggers” project, which primarily supports the development of AI-based algorithms for the high-level triggers of ATLAS and CMS at the HL-LHC. It was the first donation ever made to the CERN budget for a scientific project, and prompted us to develop a policy for this type of contribution. The policy was approved by the CERN Council in December 2024. It provides a robust set of principles and boundaries to harness the potential benefits from private funding while safeguarding the integrity and independence of the CERN scientific programme. I had already begun discussions with potential donors regarding the FCC, and the policy provided a framework for those efforts.

The FCC is an unprecedented project, and it has always been clear to me that the two traditional sources of funding for CERN projects, contributions from Member and Non-Member States, would not suffice. We consequently decided to explore two avenues that are innovative for our field: the European Commission and private donors. Concerning the former, the FCC was the first on a list of potential “Moonshot” projects in the draft Multiannual Financial Framework for the 2028–2034 period, along with a substantial dedicated budget line. As for the latter, we received pledges totalling one billion US dollars from philanthropists in the United States and Europe. I am deeply grateful to them for their generosity, vision and commitment to fundamental research.

I would like to emphasise that the main source of CERN funding has been, is, and must remain the regular contributions from the Member and Associate Member States to the CERN annual budget. Private donations are extremely valuable, but they can never replace the long-term funding stability that these contributions provide, which has been one of the key reasons for CERN’s success over the decades.

How did you handle periods of crisis?

The key was teamwork, one of CERN’s strongest assets at all levels. As Director-General, I always drew on people’s strengths, and I was fortunate enough to work with extremely talented and dedicated collaborators across the entire Organisation. I should also mention that the knowledge of CERN and the experience I accumulated over 30 years as a member of staff proved particularly helpful during those challenging periods.

So we faced crises, from the COVID-19 pandemic to the consequences of the Russian invasion of Ukraine, and the high inflation and energy prices of 2023, in true CERN style: by working together. In the decision-making phase, different experts brought their perspectives, concerns and solutions to the table. In the implementation of measures, the relevant services, departments and units deployed their technical expertise. The approach was always collective. Of course, as Director-General, I had to take the final decision, and sometimes that was tough, especially in the context of budget cuts. But those decisions were always well-informed, grounded in collective thinking and shared expertise.

The serious crises we have experienced over the past 10 years have clearly highlighted the strength and resilience of CERN as an institution and of its community.

Do you think science can still be a force for peace?

Absolutely, because science is universal and unifying. Universal, because it is based on objective facts, the laws of nature, and not on opinions. An apple falls in the same way, whether it falls in Isaac Newton’s garden in 17th-century England, or today in Switzerland, China or the United States. Unifying, because the thirst for knowledge and the desire to understand how things work are intrinsic to humanity. Thus, science has no passport, belief or gender, and can help connect people in our fractured world.

There is no way to address today’s global challenges without science

In this context, the role of CERN is emblematic. It was founded in 1954, amid the ruins of the Second World War, at a turning point in history, with the dual aim of restoring the continent’s scientific excellence and promoting peaceful collaboration among Europe’s countries and peoples through science. Today, CERN is not only a world-class scientific facility and the world leader in high-energy particle physics: it is a value system, embracing and promoting knowledge, innovation, training and education, collaboration across borders, inclusion, diversity and open science.

Institutions like CERN show what humanity can achieve when we set aside our disputes and work together for the common good. They give us hope for a better world and are more relevant today than ever.

Does CERN have a responsibility to guide the world in new technologies such as AI?

I do not think that CERN’s role is to lead the world in the development of new technologies. Our primary mission is fundamental research, and we develop advanced technologies, in collaboration with our partner institutes in the Member States and beyond, insofar as they are necessary to achieve our scientific objectives.

Machine learning and other AI techniques have been used at CERN for many years, and have proven to be valuable tools in a wide range of applications, from accelerator operation to increasing the sensitivity of physics analyses. However, I do not consider it part of our mission to conduct research in AI as a goal in itself. CERN will, of course, need to develop those aspects of AI that are specific to our field when suitable solutions do not already exist. But we should neither reinvent existing solutions nor transform ourselves into a laboratory dedicated to developing all interesting technologies for their own sake.

What are you planning to do next?

I plan to return to active research, at least for part of my time. I have always loved all areas of physics, but the Higgs boson is particularly intriguing and very close to my heart. We have learned a great deal over the past 14 years, yet this key particle remains quite mysterious. There is a rich, compelling and exciting programme of studies ahead at the HL-LHC and, potentially, at the FCC-ee.

I also plan to continue promoting science in other contexts, including as a member of the Board of Trustees of the World Economic Forum. I believe there is no way to address today’s global challenges, from health to climate change, without science.

The FCC, half a century on

The community has spoken: the electron–positron Future Circular Collider (FCC-ee) is the preferred next flagship project for CERN. As an initiator of the concept of a circular Higgs factory in 2011, I was elated by this outcome. But it also made me wonder: why did it take so long? To answer this, we need to travel back 50 years.

1976 was an eventful year for particle physics. The open charm was discovered at SPEAR, while the J/ψ earned Burton Richter and Samuel Ting their Nobel Prize. The same year, Richter authored both the first yellow report on a large e+e– (LEP) colliding ring and the first paper proposing a linear e+e– collider. Gargamelle’s measurement of the ratio of Z-induced over W-induced neutrino interactions had allowed the Standard Model (SM) – by then a familiar name – to predict the masses of the W and Z bosons. And Carlo Rubbia, synergy wizard, proposed to undercut the e+e– aficionados by converting the SPS into a proton–antiproton collider. The W and Z bosons were duly discovered in 1983.

In 1987, following the La Thuile Workshop on Physics at Future Accelerators and before LEP had even been completed, Rubbia called a general meeting in CERN’s main auditorium to discuss the future beyond LEP. Two contenders stood out: a 5 TeV e+e– linear collider in a new 30 km tunnel (CLIC), or a 20 TeV pp collider (LHC) with the advantage of fitting in the already financed and nearly finished LEP tunnel. The relative physics merits of the two machines were compared on supersymmetry and Higgs compositeness. The LHC was chosen, and CLIC became a priority R&D programme.

We had a whale of a time at LEP, establishing that light neutrinos are exactly three, measuring the Z mass to six digits and predicting the top-quark mass through radiative corrections a few months before its Tevatron discovery.

By 1996, the LHC was approved and a small group at CERN was considering what could follow it. The listed options were a high-energy future LHC (FLHC), CLIC, and a 4 TeV muon collider. A first 0.5–1 TeV linear collider (LC) was assumed to be done elsewhere. In that context, a circular e+e– machine was mentioned only as a top factory add-on to the FLHC programme, with a design extrapolated from LEP and a performance well below the LC. The prevailing assumption was that the LHC would detect the Higgs boson and supersymmetry, if either existed.

A breakthrough came in 1999, when the asymmetric B factories PEP-II and KEKB, with separated e+ and e– rings and continuous top-up injection, demonstrated luminosities orders of magnitude higher than LEP. Meanwhile, LEP Higgs-hunted fiercely until the end of 2000, setting a lower mass limit at 114.5 GeV, while precision measurements set an upper mass limit of about 180 GeV.

The hunt is on

Come the summer of 2011, the LHC experiments were taking data at 7 TeV. The hunt for the Higgs and the supersymmetric particles was on, and the first limits already constrained the Higgs mass below that of a W pair. That clarified the required centre-of-mass energy of a Higgs factory: a relatively low-energy e+e– collider would do. Locating it in the LEP/LHC tunnel was an obvious possibility, and had already been discussed in the corridors of the EPS-HEP conference in Grenoble that July. Five months later, applying the B-factory design principles, a Higgs factory fitting in the LHC tunnel was evaluated. “LEP3” offered luminosity significantly higher than the linear collider, and the advantage of running several experiments simultaneously. On the downside, its maximum energy fell short of the top-pair-production threshold… and the LHC tunnel was already occupied.

None of this was a straight line. It took 15 years for the physics to make the case on its own terms

Presented with this evaluation, some members of the CERN Scientific Policy Committee suggested that an e+e– Higgs factory more than triple the size of LEP would make a great initial step towards a higher-energy version of the LHC, which was already under consideration. Inserting the Higgs factory in a 100 km tunnel did magic. With its large bending radius, the machine reached the top-pair threshold while covering a wide range of energies and luminosities. It delivered large statistics at the Z pole, with 6 × 1012 visible Z decays, and transverse polarisation for exquisite energy calibration up to the W-pair threshold. Those were key ingredients in achieving a vertiginous potential for statistical and systematic improvement, by up to a factor of 500 or more over the old LEP precision measurements. The Z run, it was later realised, would turn FCC-ee into a flavour factory for b and tau precision studies, and also enable unique searches for feebly coupled particles in the 5–80 GeV mass range.

What lies beyond

When the Higgs came, it did so at 125 GeV – too high for most incarnations of supersymmetry, too low for theories of a composite Higgs, and consistent with an unchanged SM up to very high energies. It raised the question of what lies behind the SM and at which energies, making it essential for the future of particle physics to include an extensive programme of precision measurements, in the hope of detecting deviations from the SM that could guide the next steps. The Higgs itself was also of obvious interest, and a lepton collider the natural way to study it.

These arguments were summarised in two contributions to the 2013 update of the European Strategy and led to the recommendation of a costed design study of FCC-hh and FCC-ee. The 2020 update continued it as a feasibility study, which led in turn to the 2025 recommendation.

None of this was a straight line. It took 15 years for the physics to make the case on its own terms, and the conversation had been on, in one form or another, for 35 years before that. The Higgs mass, the absence of supersymmetry and the precision reach all pointed to a machine in the best CERN tradition, where the most is made of the resources through a strong synergy between infrastructure planning and physics opportunities.

Antimatter hits the road

Take a bunch of antiprotons. To stop them annihilating, seal them inside a near-perfect vacuum, suspended in the bore of a superconducting magnet, and superimpose an electric field. Load them onto a truck, and drive off. On 24 March 2026, the BASE collaboration sent 92 antiprotons on a test loop around CERN’s Meyrin site, achieving the first controlled and reversible transport of antimatter. The trip is the culmination of years of work to move antimatter precision measurements out of CERN’s noisy Antimatter Factory, where BASE operates (CERN Courier January/February 2025 p6).

The collaboration’s main target is CPT symmetry. Charge conjugation (C), parity inversion (P) and time reversal (T), taken together, are expected to leave physics invariant. Matter and antimatter must therefore have identical masses and magnetic moments of equal magnitude, with charges of opposite sign. BASE tests CPT directly on protons and antiprotons, confined in electromagnetic traps, by measuring their cyclotron and spin-flip frequencies. “At low energies, measurements usually use only matter systems, on the assumption that antimatter behaves the same way without testing it,” says Christian Smorra, leader of the collaboration’s transportable-trap project BASE–STEP. “Antiprotons are the only stable antibaryons that can be produced and trapped at low energies, enabling precise frequency measurements.”

Noisy fields

So far, BASE results on proton and anti­proton charge-to-mass ratios agree to 16 parts per trillion, and their magnetic moments to 1.5 parts per billion. The measured frequencies, however, scale linearly with the strong magnetic field that confines the particles, so any noise in it directly affects the result. The magnetic environment of CERN’s Antimatter Factory now limits how far precision can be pushed.

One natural solution is to move the antiprotons elsewhere. To survive the journey, they must remain in an extreme vacuum, below 10–14 mbar, since contact with a single gas molecule means annihilation. “At those pressures, the only way to test the vacuum is to trap antiprotons and see how long they survive,” explains Smorra. “We had no way to predict how much the pressure would rise in the room-temperature parts of the system during transport, so we had to rely on calculations and build the best possible setup to limit the gas flow into the trap.”

Calculations suggest that the trap could hold antiprotons for more than a year

The result is BASE-STEP: a one-tonne portable electromagnetic trap with up to four hours autonomous operation and a persistent superconducting magnet cooled by liquid helium. At those temperatures, the inner walls of the trap freeze out most gas molecules on contact, preserving the vacuum. Three further measures handle what the walls cannot manage alone. A 500 mm-long differential pumping section thins residual gas in the warm part of the system, a specialised valve – now in its third generation – seals the cold interior, and a dedicated pump captures any stray hydrogen.

The first injection of antiprotons, in December 2025, lasted three days. A subsequent run kept them trapped for more than a month, with cumulative experimental lifetimes now reaching two and a half. “Calculations suggest the trap could hold antiprotons for more than a year,” says Smorra. “The limit is set by gas slowly accumulating on the cold trap surfaces. Once a single layer has built up, they can no longer trap new molecules, and the vacuum starts to deteriorate.”

The 24 March test demonstrated that BASE-STEP could survive vibration and acceleration without losing its load. The team has now requested a low-magnetic-noise space at CERN to establish methods for transferring antiprotons between the transportable trap and a receiver experiment. Further afield, BASE–HHU at Heinrich Heine University Düsseldorf is being built to receive antiprotons from BASE-STEP and perform precision measurements. “We expect to need only one or two trips per measurement,” says Smorra. The Düsseldorf transport will take around 10 hours, longer than the trap can run on its own. A generator on the truck will power a cryocooler to keep the magnet superconducting throughout.

Beyond CPT tests, the approach may prove useful to other precision searches. “We are also studying exotic interactions of antiprotons,” comments Smorra, “such as antiproton–axion coupling or collision rates of millicharged particles with trapped antiprotons.”

Two channels for the top–antitop excess

The top quark was never meant to bind. And yet a year ago, CMS reported an excess of top-quark–antiquark pairs in dilepton events near the production threshold, consistent with the fleeting formation of a top-quark–antiquark quasi-bound state: toponium. ATLAS confirmed the effect just a few months later, rejecting a pure perturbative QCD interpretation at 7.7σ (CERN Courier September/October 2025 p9). CMS has now extended the case to an independent decay process.

The analysis, presented at this year’s Rencontres de Moriond, looks at events in which one top decays into a charged lepton, a neutrino and a bottom quark, and the other into jets. In 138 fb–1 of Run 2 data at 13 TeV, the enhancement exceeds the pure-QCD prediction by more than five standard deviations, with an excess cross section of 5.1 ± 0.9 pb.

“Establishing a signal in both channels was very important,” says Regina Demina, who leads the University of Rochester CMS group. “The lepton + jets channel has higher statistics, thanks to the larger hadronic branching ratio of the W boson, and a single neutrino makes the kinematics easier to reconstruct. The systematic uncertainties differ from those in the dilepton channel.”

The charm and bottom quarks live long enough to bind tightly with their antiparticles, and the resulting mesons appear as sharp, narrow peaks in the cross section. The top quark, by contrast, decays too quickly, with a width comparable to the binding energy that would hold a top-quark–antiquark system together. Any such state would manifest as a broad threshold enhancement, smeared over the smooth QCD continuum.

“The formation of bound states of charm or bottom quarks is a well-established effect, which allowed theorists to refine our understanding of the QCD binding potential,” says Yu-Heng Yu, a graduate student at the University of Rochester who worked on the analysis. “Yet it came as a surprise that, given the very short lifetime of top quarks, such a quasi-bound state still manages to form in a small fraction of events.”

The lepton + jets channel demanded two methodological adjustments. The first replaces the invariant mass of the top-quark–antiquark pair, whose resolution is limited near threshold, with their relative velocity as the discriminating observable. “If they form a bound state, the relative velocity should be much smaller than when they are produced independently,” says Otto Hindrichs, also at Rochester. The second concerns the parity-sensitive observables that distinguish a pseudoscalar from a scalar interpretation of the bump. “These variables require a reconstruction method that identifies the down-type jet from the hadronic W decay,” explains Hindrichs. “To achieve this, we developed a machine-learning technique that improves the correct identification of the top-quark decay products.”

Some puzzles remain. The 5.1 pb cross section sits below the 8.8 pb measured in dilepton events, and the non-relativistic QCD reference of about 6.4 pb. “We do observe somewhat different signal strengths in the lepton + jets and dilepton channels, and we are actively investigating this difference,” says Yu.

“With the current sensitivity, interpretations beyond the Standard Model cannot be excluded,” Hindrichs adds. “A pseudoscalar heavy Higgs decaying into top-quark pairs would interfere strongly with the continuum, creating a characteristic peak-dip structure in the invariant tt mass. With enough statistics, this feature could be used to differentiate it from a quasi-bound state.”

The top-quark–antiquark threshold enhancement in e+e– collisions was analysed by Fadin and Khoze in 1987, and extended to hadron colliders by Fadin, Khoze and Sjöstrand in 1990, before the 1995 discovery of the top quark at Fermilab. The standard assumption was that any signal would have to wait for a next-generation e+e– collider reaching the threshold, which would provide the cleanest measurement of the top-quark mass. “Even with Run 3 data, we will not be able to resolve the structure of the threshold region itself, because the resolution on both the invariant mass and the relative velocity is too limited,” says Demina. “But the increased statistics should allow us to probe the spin-parity content of the bump, testing whether it carries scalar or vector contributions alongside the pseudoscalar one.”

Breakthrough honours g–2

The 2026 Breakthrough Prize in Fundamental Physics recognised the multi-decade programme to measure, with ever-increasing precision, the muon’s anomalous magnetic moment (“g-2”). Announced in Los Angeles on 18 April, the $3 million award is shared among the living co-authors of the key publications from the muon g–2 collaborations at CERN, Brookhaven National Laboratory and Fermilab. Five further Breakthrough prizes recognised work in theoretical physics, dark-matter searches and cosmology.

As a charged particle with spin, the muon behaves like a tiny magnet whose strength is set by a dimensionless factor close to, but not exactly, two. The deviation, known as the anomalous magnetic moment, encodes virtual loop corrections from all sectors of the Standard Model (SM), and comparing it with theoretical predictions is among the most stringent tests of the theory.

Following initial measurements at Columbia University in 1957, the story began at CERN in 1959 with a small magnet borrowed from the University of Liverpool and Leon Lederman’s idea to test quantum electrodynamics using the muon. The idea was to place muons in a uniform external magnetic field and observe their spin precession frequency, which depends on the strength of the field and the muon’s magnetic moment. By 1962, a dedicated 6 m magnet at the Synchrocyclotron had enabled the CERN team to pin down the anomalous magnetic moment with a precision of 0.4%. Two storage-ring experiments at the Proton Synchrotron followed. The third reached a precision of 7.3 parts per million by 1979, and pulled hadronic effects into view for the first time.

Brookhaven’s E821 experiment took over at the Alternating Gradient Synchrotron, reaching 540 parts per billion in its final 2006 report. The measurement stood 2.2–2.7σ above the SM evaluations of the day. In the summer of 2013, the experiment’s 14 m-diameter superconducting storage ring travelled by road and barge from Long Island to Batavia, where Fermilab’s more intense and pure muon beam awaited.

The story of the g-2 began at CERN in 1959

The final Fermilab measurement, announced in June 2025, reached a precision of 127 parts per billion: 30,000 times better than the first g–2 results (CERN Courier July/August 2025 p7). The theory side has moved as sharply. By August 2023, the discrepancy with respect to the 2020 prediction of the Muon g–2 Theory Initiative, an international consortium tasked with delivering a consensus SM value, had reached 5.1σ. Its 2025 update, which drops data-driven inputs to the hadronic vacuum polarisation in favour of a lattice-QCD consensus, sits within roughly 1σ of the measured value. The shift between the two predictions is itself about 3σ, reflecting an unresolved tension (CERN Courier January/February 2026 p41).

The 2026 Special Breakthrough Prize in Fundamental Physics went to David Gross (KITP, UC Santa Barbara) for a lifetime of contributions to theoretical physics. In 1973, Gross and his graduate student Frank Wilczek at Princeton, and independently David Politzer at Harvard, found that the strong nuclear force becomes weaker as quarks approach one another, a property known as asymptotic freedom. The three shared the 2004 Nobel Prize in Physics.

The inaugural Vera Rubin New Frontiers Prize went to Carolina Figueiredo (Princeton University). With Nima Arkani-Hamed and collaborators, she showed that the scattering amplitudes of three apparently unrelated theories, describing gluons, pions and a simplified scalar toy model, are generated by a single function, related by a simple shift of the kinematics. The result emerges naturally from a geometric formulation known as surfaceology.

Among the New Horizons in Physics Prize recipients, Benjamin Safdi (UC Berkeley) was recognised for his contributions to axion searches. Clay Córdova (University of Chicago), Thomas Dumitrescu (UCLA), Shu-Heng Shao (MIT) and Yifan Wang (New York University) shared a New Horizons in Physics Prize for the development of generalised symmetries in quantum field theory, with applications ranging from condensed-matter physics to string theory. A third New Horizons in Physics Prize recognised Dillon Brout (Boston University), J Colin Hill (Columbia University), Mathew Madhavacheril (University of Pennsylvania), Maria Vincenzi (University of Oxford), Daniel Scolnic (Duke University) and W L Kimmy Wu (Caltech) for analyses of cosmic microwave background data and Type Ia supernova samples, delivering tight constraints on the expansion and composition of the universe.

Final collisions for the RHIC

On 6 February 2026, beams of oxygen ions circulated through the Relativistic Heavy Ion Collider (RHIC) at Brookhaven National Laboratory for the last time. A leading facility in the study of hadronic matter and the strong force since 2000, RHIC now hands its tunnel and many of its components to its successor, the Electron–Ion Collider (EIC).

“Experiencing the challenges of first trying to get beams to circulate during commissioning in the fall of 1999, one could not have dreamed how far the performance of this machine would come,” said Wolfram Fischer, chair of Brookhaven’s Collider-Accelerator Department. “We’ve pushed well beyond the original design in terms of the number of collisions we can produce, the energy range of those collisions, the variety of ions we’ve collided, and our ability to align the spins of protons and maintain a high degree of this alignment or polarisation.”

RHIC was conceived above all to study the quark–gluon plasma (QGP). In QGP, quarks and gluons, normally confined inside protons and neutrons, roam free under extreme temperature and density. The early universe is thought to have existed in this state for a fraction of a second after the Big Bang, before cooling into the ordinary matter around us.

Theorists had expected this primordial soup to behave as a weakly coupled gas of quarks and gluons. Gold–ion collision data from RHIC’s four original detectors, BRAHMS, PHENIX, PHOBOS and STAR, found instead a strongly coupled liquid. By 2005 the collaborations had concluded that they were producing one of the lowest-viscosity substances ever observed, a nearly “perfect” liquid. Later runs traced how this extreme state of matter swirls, flows and cools, and revealed that even small collision systems can briefly form tiny droplets, overturning earlier ideas about how QGP forms.

RHIC transformed nuclear physics by demonstrating the remarkable consequences of ‘boiling the vacuum’

“RHIC transformed nuclear physics by demonstrating the remarkable consequences of ‘boiling the vacuum,’” said theorist Raju Venugopalan, paraphrasing T D Lee’s description of matter governed by quantum chromodynamics.

Beyond QGP, STAR and PHENIX measurements in polarised proton–proton collisions established that gluons carry a significant share of the proton’s spin. In the final run, sPHENIX, the faster successor to PHENIX, became the first detector to record a continuous streaming dataset from RHIC’s spin-polarised proton collisions – thus eliminating the need for triggers.

The final run also gave a sense of the scale of modern physics data: sPHENIX alone recorded more than 200 petabytes of raw data, more than every previous RHIC dataset combined, including 40 billion gold–ion collision events. Analysis of RHIC data will continue for at least another decade. Much of RHIC’s infrastructure will then live on in the EIC, including its ion sources, pre-accelerator chain and one of its superconducting storage rings. A new electron ring will share the tunnel, crossing the ion beam at points where polarised electrons and ions will collide. The EIC will enable precision measurements that reveal how quarks and gluons are organised within protons or atomic nuclei, helping physicists to understand how mass, spin, and nuclear structure emerge from the strong force.

The kaon stays on script

Wired for rarity
Wired for rarity Photomultiplier readout on the MUV3 detector at NA62, designed to study the K+ → π+νν̅ decay against an overwhelming background. Credit: P Traczyk, CERN

Less than one in 10 billion positively-charged kaons decay into a pion and a neutrino–antineutrino pair. The NA62 experiment has now measured the rate of this rare process with an uncertainty 40% smaller than its previous result and a central value closer to the Standard Model (SM) prediction (CERN Courier November/December 2024 p11).

“The K+ → π+νν decay is a golden mode of flavour physics,” says NA62 spokesperson Giuseppe Ruggiero. “It is highly suppressed in the SM, but its branching ratio can be predicted to better than 10% precision. The decay is also highly sensitive to new physics, with many models predicting dramatic changes to the branching ratio. Such modifications may come from indirect effects of new physics at or above the 100 TeV scale.”

The scarcity of the decay called for a kaon factory. At NA62, a high-intensity proton beam from the Super Proton Synchrotron strikes a beryllium target, producing around 500 million secondary particles per second. About 6% are positively charged kaons. From that flux, the experiment must isolate the signal against backgrounds many orders of magnitude larger. The first 5σ observation, on data collected through 2022, was reported in 2024. The branching ratio came out at (13.0+3.3–3.0) × 10–11, consistent within 1.7σ with the SM prediction of around 8 × 10–11, despite a central value about 50% higher. Two years of additional data have now doubled the signal sample, and the central value has come down to (9.6+1.9–1.8) × 10–11, reaching a sub-20% precision.

Two new machine-learning techniques drove the increase in precision. “Reconstructing beam particles in the harsh environment of up to a gigahertz of incoming particles is challenging,” says Joel Swallow of CERN, lead data analyst of the study. “To tackle this, we deployed a transformer encoder to pick out a kaon as it enters the experiment. Meanwhile, a combined convolutional and feed-forward neural network was developed for pion identification, which effectively uses images of the energy deposits in the calorimeters to more efficiently and accurately identify pions.”

Two new machine-learning techniques drove the increase in precision

“Had the central value stayed where it was, the precision of the new measurement would have been sensitive to a 3σ excess,” says Ruggiero. “If there had been an excess that large, this measurement was perfectly positioned to find it. Evidently, nature is a bit more subtle.”

The new result tightens constraints on beyond-SM scenarios that would have predicted larger branching ratios, including those involving leptoquarks or heavy Z′ bosons. Still, the dominant uncertainty remains statistical, and additional data from 2025 and 2026 will improve the precision further.

The neutral counterpart, KL → π0νν , has yet to be observed. The current upper limit on its branching ratio, set by Japan’s KOTO experiment at J-PARC, sits two orders of magnitude above the SM prediction. “Measuring both the charged and neutral modes is important,” says Ruggiero. “Together, they enable a fully independent reconstruction of the unitarity triangle from kaon decays alone. Even if, in the end, the charged mode is consistent with the SM, it does not rule out significant enhancements from new physics to the neutral mode.” The proposed KOTO-II, at J-PARC, is targeting a measurement of KL → π0νν  in the 2030s.

Copyright © 2026 by CERN
bright-rec iop pub iop-science physcis connect